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 |
|---|---|---|---|---|---|
Membrane fusion is the critical step for infectious cell penetration by enveloped viruses . We have previously used single-virion measurements of fusion kinetics to study the molecular mechanism of influenza-virus envelope fusion . Published data on fusion inhibition by antibodies to the 'stem' of influenza virus hemagglutinin ( HA ) now allow us to incorporate into simulations the provision that some HAs are inactive . We find that more than half of the HAs are unproductive even for virions with no bound antibodies , but that the overall mechanism is extremely robust . Determining the fraction of competent HAs allows us to determine their rates of target-membrane engagement . Comparison of simulations with data from H3N2 and H1N1 viruses reveals three independent functional variables of HA-mediated membrane fusion closely linked to neutralization susceptibility . Evidence for compensatory changes in the evolved mechanism sets the stage for studies aiming to define the molecular constraints on HA evolvability .
Membrane fusion is the mechanism for directed interchange of contents among intracellular compartments . Carrier vesicles fuse with target organelles , secretory vesicles fuse with the plasma membrane , mitochondria fuse with each other . Enveloped viruses fuse with a cellular membrane to deposit their genomic contents into the cytosol . Lipid bilayer fusion is a favorable process but with a high kinetic barrier ( Chernomordik and Kozlov , 2003 ) . Each of the examples of fusion just cited requires a protein catalyst . The SNARE complexes catalyze vesicle fusion ( Brunger , 2005 ) ; mitofusins catalyze mitochondrial membrane fusion ( Chan 2012 ) ; viral fusion proteins catalyze the fusion step essential for infectious cell entry ( White et al . , 2008 , Harrison 2008 , 2015 ) . The influenza hemagglutin ( HA ) is the best studied and most thoroughly characterized of the viral fusion proteins . Crystal structures determined in the 1980s and 1990s captured the fusion endpoints and showed that extensive structural rearrangements , triggered during entry by the low pH of an endosome , are part of the catalytic mechanism ( Wilson et al . , 1981 , Skehel et al . , 1982 , Bullough et al . 1994 , Chen et al . , 1998 , 1999 ) . Models for the fusion process then ‘interpolated’ intermediate states between these endpoints , supported by indirect evidence for specific features of these intermediates ( Figure 1 ) ( Daniels et al . , 1985 , Godley et al . , 1992 , Carr and Kim , 1993 , Harrison 2008 , 2015 ) . 10 . 7554/eLife . 11009 . 003Figure 1 . Productive and non-productive HA refolding , and membrane fusion by cooperative action of multiple , stochastically triggered HAs . ( A ) Proton binding increases the relative time HA spends in the ‘open’ conformation allowing fusion peptides to project toward the target membrane . HA1 is shown in green and HA2 in magenta ( fusion peptides ) , gray ( N-terminal ‘half’ ) and blue ( C-terminal ‘half’ ) . Right-hand arrow: Productive HA refolding proceeds through an extended-intermediate state with fusion peptides inserted in the target membrane ( Ivanovic et al . , 2013 ) . We illustrate a possibility that membrane-engaged HAs might represent an ensemble of folded-back conformations; the corresponding distance between the two membranes might fluctuate around a different value depending on how many HAs are cooperating . Left-hand arrow: Non-productive HA-refolding event occurs if HA assumes the low-pH form without target membrane engagement , resulting in loss of that HA as a potential fusion participant . ( B ) Individual-HA triggering and membrane insertion occur at random within a larger virion area that contacts the target membrane ( ~50 HAs shown in green are contained within this interface for a small , spherical influenza virion [Ivanovic et al . , 2013] ) . Fusion ensues once a sufficient number of HAs – as needed to overcome the resistance of membranes to bending and apposition – are pulling jointly on the same membrane region ( Ivanovic et al . , 2013 ) . 3D coordinates ( PDB ID ) used for displayed HA cartoons: the pre-fusion HA ( 2HMG ) , inactivated HA ( 1QU1 ) ; depicted intermediates are derived from a subset of either or both sets of coordinates ( 2HMG and/or 1QU1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 003 Single-molecule techniques applied to studies of influenza virus fusion have yielded more direct information about the HA molecular transitions that facilitate it ( Floyd et al . , 2008 , Imai et al . , 2006 , Ivanovic et al . , 2012 , Ivanovic et al . , 2013 , Otterstrom and van Oijen , 2013 , Otterstrom et al . , 2014 , Wessels et al . , 2007 ) . The following picture emerged from experiments we described in 2013 , in which we combined single-virion fusion observations with structure-guided mutation of HA ( Figure 1 ) ( Ivanovic et al . , 2013 ) . Trimeric HA ‘spikes’ densely cover the surface of an influenza virus particle . The contact zone between virus and target membrane ( a supported lipid bilayer in the case of our experiments ) contains between 50 and 150 HA trimers—a number that may be even larger for filamentous virions . When the pH drops below a critical threshold , individual HAs within the contact zone adopt an ‘extended state’ , in which the fusion peptide at the N-terminus of HA2 engages the target membrane , while the C-terminal transmembrane anchor remains embedded in the viral membrane . Note that the ‘extended state’ might represent an ensemble of folded-back conformations ( Figure 1A ) . The probability of this stochastic event increases with proton concentration over the range at which groups on the protein titrate . A single HA trimer in the extended conformation cannot then fold back to its most stable , postfusion conformation , because of elastic resistance from the two membranes . Only when several neighboring HAs have extended and engaged can their joint action pull the two membranes together ( Figure 1B ) . When the critical number of extended neighbors is present , foldback is cooperative and progression toward fusion is fast . These observations led us to propose that the cooperativity of foldback comes simply from the mutual insertion of the cooperating HAs in both fusing membranes and that the number of HAs required is a function of the free energy released from individual HA fold-back events . When the total free energy is enough to overcome the ‘hydration-force’ barrier to merger ( Rand and Parsegian , 1984 ) , fusion can ensue . We called this a ‘tug-of-war’ mechanism— ( N-1 ) trimers are not enough , but adding one more immediately precipitates a change , just as adding a critical extra team member will promptly snap a rope pulled against a fixed force . The team members need not touch each other as long as all are pulling on the same rope . An alternative model for cooperative action of fusion proteins comes from structural observations on alphavirus membrane fusion proteins , which suggest that a ring of five envelope-protein trimers might work as a single-unit fusion assembly ( Gibbons et al . , 2004 ) . This picture is a particular instance of mechanisms that require a defined , lateral interaction between participating proteins . The probability of assembling a group of HA neighbors inserted into the target membrane depends on the fraction of active HAs . Some positions in the contact zone may be occupied by uncleaved HA0 , which cannot undergo the fusion-inducing conformational change ( Chen et al . , 1998 ) , and others , by the viral neuraminidase , NA ( although NA appears to cluster on one side of the budded particle: Harris et al . , 2006 , Calder et al . , 2010 , Wasilewski et al . , 2012 ) . Moreover , the fusion peptides of some HAs that do undergo the low-pH induced conformation change might fail to insert into the target membrane ( Figure 1A ) . Exposure of unattached virions to low pH leads to inactivation , with the fusion peptides of rearranged HAs inserted back into the viral membrane , providing an experimental demonstration that non-productive conformational changes can indeed occur ( Weber et al . , 1994 , Wharton et al . , 1995 ) . Simulations we used to derive kinetic parameters from single-virion fusion data can include estimates of inactive sites and unproductive events , and we show below the usefulness of this extension ( Figure 2 ) . 10 . 7554/eLife . 11009 . 004Figure 2 . The functional variables of influenza membrane fusion modeled in this work . We modeled the kinetics and the extent of membrane fusion with the following parameters: ( A ) the number of HAs in contact with the target membrane ( patch size , PS ) , ( B ) the rate ( ksim ) of stochastic HA triggering , ( C ) the required number ( Nh ) of cooperating HA neighbors during fold-back ( see Figure 2—figure supplement 1 for the complete definition of six-mers ( Nh = 6 ) in the simulation ) , and ( D ) the frequency of inactive ( left ) or unproductive ( middle ) HAs , combined in the common parameter fnp ( right ) as described in Materials and methods . Illustrations represent sample contact patches at the times of hemifusion except in panel B ( left and middle ) , where they represent earlier time points . We compare the effects of various functional variables by either showing the ratios of mean hemifusion delays ( ksim-independent values ) ( A , C and D ) , or by directly showing mean hemifusion delays for two ksim values , and PS = 121 , Nh = 3 and fnp = 0 ( B ) . Our fusion model predicts that smaller patch size , lower ksim , higher Nh , or higher fnp , will each increase hemifusion delay , and , with the exception of ksim , will also , under certain conditions , reduce the theoretical fusion yield ( see Figure 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 00410 . 7554/eLife . 11009 . 005Figure 2—figure supplement 1 . Definition of six-mers ( Nh = 6 ) in the simulation . Since multiple HAs must fold back cooperatively , possible six-mers have distinct properties: two groups of three HAs that can join around either two adjacent or one wider target-membrane deformation ( six-mers 1 and 2 ) ; HAs that all surround the same central target-membrane deformation ( six-mer 3 ) . Because HAs must fold away from the membrane deformation ( see illustration in Figure 1 , the productive path ) , six-mer 4 will tend to act as two isolated groups of three HAs or , more likely , as a tetramer with two nearby HAs that are not close enough to join their efforts in pulling on the same target-membrane area . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 005 Addition of neutralizing antibodies can create additional inactive HAs . Otterstrom et al . ( 2014 ) recently used the single-virion assay together with fluorescently tagged IgGs or Fabs to study the occupancy required to achieve complete inhibition of viral fusion . They found that occupancies short of 100% were sufficient to reduce the yield of fusion to threshold . They concluded that these observations were consistent with the model we had proposed ( Ivanovic et al , 2013 ) and that bound antibodies need simply to disrupt the network of potential neighbors rather than saturate the viral surface . In the work we report here , we have used computer simulations to extend the analysis of fusogenic molecular events at the virus-target membrane interface ( Figure 2 ) and compared the results with published single-virion experiments , including the recent studies of Otterstrom et al . ( 2014 ) . The extension includes an explicit parameter for the fraction ( fnp ) of 'non-participating surface elements' ( those HAs that fail to engage and stochastically inactivate , those that have bound antibodies , those that are HA0 , and those sites in the model that might be occupied by NA ) ( Figure 2D ) . This analysis yields new conclusions concerning the course of viral fusion . We identify three independent functional variables of HA-mediated membrane fusion and find that virions from H3 and H1 influenza subtypes differ in at least two and possibly all three respects , and offer evidence for compensatory features of the evolved mechanism . The results illustrate the relative degrees of freedom available to influenza virus as it evolves in response to external pressures , whether from inhibitors , host immunity , or adaptation to replication in a new host species .
We simulated stochastic HA triggering within the ‘contact patch’ between virus particle and target membrane , for patch sizes ( PS ) of 121 and 55 HA trimers ( Figure 3 and Figure 3—figure supplement 1 ) , using the algorithm previously described ( Ivanovic et al . , 2013 and Materials and methods ) . We included a range for the fractions of non-participating sites ( fnp – HA0 , NA , non-productively refolded HA1:HA2 ) ( Figure 3A ) and allowed simulations to proceed to completion , i . e . until all the virions with potential to hemifuse had done so , or , until all HAs in the contact patch had extended and become either target-membrane engaged or inactivated ( the highest value of fnp we included yielded ~2% hemifusion ) . We defined the time of hemifusion as the moment at which the Nhth HA trimer joins a preexisting cluster of ( Nh-1 ) HAs and determined , as functions of fnp , both the yield of hemifusion ( percent of virions that hemifused ) ( Figure 3B ) and the distribution of times from pH drop to hemifusion ( Figure 3C–E ) . We ran the simulations for values of Nh between 3 and 6 . We previously concluded that Nh = 2 yields data that do not agree with experiment results for H3 influenza ( X31 and Udorn ) ( Ivanovic et al . , 2013 ) , and we provide here additional results to justify exclusion of this value in further analysis ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 11009 . 006Figure 3 . Effects of fnp on hemifusion yield and kinetics for Nh = 3–6 ( PS = 121 ) . ( A ) Illustration of simulated contact patches . ( B ) Hemifusion yield as a function of fnp . ( C ) Mean hemifusion-delay times normalized to fnp = 0 . ( D ) Parameter N derived from fitting hemifusion delay distributions with the gamma probability distribution . Errors are 95% confidence intervals for the fit-derived values . ( E ) Parameter k derived from fitting hemifusion delay distributions with the gamma probability distribution expressed as ratio with ksim . By normalizing mean hemifusion-delay times and kgamma , we obtained general trends , independent of the ksim value used in simulations . Plotted results are derived from simulations that yielded 1000–3000 hemifusion events . Blue shaded regions are estimates for the range of fnp values consistent with Ngamma values derived from experiment . The corresponding results for PS = 55 are shown in Figure 3—figure supplement 1 . Refer to Figure 3—figure supplement 2 for the simulation results for Nh = 2 and both patch sizes . Refer to Figure 3—figure supplement 3 for Ngamma values derived from our previously published experimental datasets ( Ivanovic et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 00610 . 7554/eLife . 11009 . 007Figure 3—figure supplement 1 . Effects of fnp on hemifusion yield and kinetics for Nh = 3–6 ( PS = 55 ) . Refer to the main Figure 3 legend , which shows results of an analogous set of simulations using PS = 121 instead of PS = 55 . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 00710 . 7554/eLife . 11009 . 008Figure 3—figure supplement 2 . Effects of fnp on hemifusion yield and kinetics for Nh = 2 . Simulation results for both PS = 55 and PS = 121 are shown . Hemifusion yield ( top row ) , mean hemifusion-delay times normalized to fnp = 0 ( second row ) , parameter N derived from fitting hemifusion delay distributions with the gamma probability distribution ( third row ) , and parameter k derived from fitting hemifusion delay distributions with the gamma probability distribution expressed as ratio with ksim ( bottom row ) . The simulation results for Ngamma when Nh = 2 are inconsistent with previous experiments that routinely report values of 3 or higher ( please refer to Figure 3—figure supplement 3 ) . This comparison further validates our original conclusion that Nh > 2 , from experimental data that appeared to rule it out ( Ivanovic et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 00810 . 7554/eLife . 11009 . 009Figure 3—figure supplement 3 . Ngamma for pH-drop-to-hemifusion frequency distributions from previously published experiment data ( Ivanovic et al . , 2013 ) . Udorn and X31-G4S are previously published values . X31 and Udorn-S4G are newly determined values from those published datasets . ( A ) Ngamma values with 95% confidence intervals ( error bars ) for the fit-derived values . ( B ) Frequency distributions for the newly derived values fitted with the gamma distribution . 95% confidence intervals for the fit-derived values are shown as a range of values in parentheses . Samples that resulted in larger Ngamma values ( Ngamma>4 ) also yielded poorer fits . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 009 The dependence of hemifusion yield and delay time on fnp as Nh varied over a reasonable range led us to conclude that the data in Otterstrom et al . ( 2014 ) could yield new information about these parameters ( see what follows and the next results section , The gamma-distribution approximation ) . The simulations showed that the yield of hemifusion is relatively insensitive to the presence of inactive HAs for Nh between 3 and 6 ( Figure 3B ) . For Nh = 3 , more than 70% ( fnp = 0 . 7 ) of the sites on a virion surface must be unproductive or inactive in order to detect any reduction in fusion yield; for Nh = 6 , we saw reduced yield whenever more than 50% of the sites lacked the potential to participate . The simulations also yielded relatively large increases in mean lag time to hemifusion for the tested range of fnp values ( Figure 3C ) . For Nh = 3 , we found a tenfold , and for Nh = 6 , a fivefold increase in mean time to hemifusion . In contrast to our simulation results , Otterstrom et al . ( 2014 ) observed sudden decreases in hemifusion yield for even the small numbers of bound antibodies or Fabs , and at most about a two-to-threefold increase in hemifusion lag times until complete inhibition of hemifusion . This difference could not be explained by a smaller patch size ( Figure 3—figure supplement 1 ) and suggested to us that even for virions with no bound antibodies , a significant portion of surface sites lacked the potential to participate in fusion ( i . e . the experiment was sampling from the right-hand portion of an entire theoretical inhibition curve ) . This qualitative conclusion is independent of the actual value of Nh or fnp . For all values of Nh , the mean hemifusion-lag times had the same overall dependence on fnp . As fnp increased , a phase of relatively shallow dependence of the lag time gave way to a much stronger rate of increase , at about the same fraction at which the overall yield of hemifusion began to decline ( compare Figure 3B and C ) . For fnp values at which more than half of the simulated virions no longer yielded hemifusion , the lag time dependence reached a plateau . Otterstrom et al . ( 2014 ) indeed observed a plateau in mean hemifusion lag times as a function of increasing antibody or Fab concentration , thus offering experimental support for the prediction derived from the proposed mechanism of fusion ( Ivanovic et al . , 2013 ) . Plateau occurs when additional reduction in the fraction of participating HAs is more likely to result in complete inhibition of hemifusion rather than further increase in the lag time . Indeed , for Fab concentrations in the plateau region for hemifusion delay , Otterstrom et al . ( 2014 ) found a continuing decrease in hemifusion yield as Fab concentrations increased . The result is intuitively reasonable . A high fraction of non-participating sites in a contact patch corresponds to a high probability that any particular HA will fail to engage the target membrane , either because it cannot change conformation ( unprocessed HA0 or inhibitor bound HA1:HA2 ) or because it has irreversibly inactivated ( Figure 3A ) . When this probability becomes high enough , it becomes almost impossible to achieve Nh membrane-engaged neighbors within a contact patch of fixed size ( consider , for example , the number of ways one can fit Nh = 6 active HA neighbors within the contact patches illustrated in Figure 3A for different fnp values ) . The gamma probability distribution represents the kinetics of a process in which N rate-limiting events of ( uniform ) rate constant k occur in sequence . The first single-virion fusion experiments took N from this representation as an estimate of the number of HAs required for hemifusion ( Floyd et al . , 2008 ) . Subsequent comparison with simulation showed that the estimate is inaccurate when 100% of the virion surface can participate ( Ivanovic et al . , 2013 ) . Dependence of k on mutations that affect the docking of the fusion peptide in the pre-fusion trimer led to the conclusion that the rate-limiting step in the fusogenic conformational change is fusion-peptide exposure ( Ivanovic et al . , 2013 ) . To explore the effects of fnp on the derived values of N and k , we fitted hemifusion-delay distributions from our simulations with gamma distributions ( designating the parameters Ngamma and kgamma ) ( Figure 3D and E ) . We confirmed our previous conclusion that Ngamma is an overestimate when all HAs in the contact patch are active ( Figure 3D ) . We further found that simulation-derived Ngamma approached the experimental values from previous studies of H3 viruses at high fnp and Nh = 3–5 . Except for a few specific data points , experimental values for Ngamma are between 2 and 4 ( see Materials and methods for summaries of previously published Ngamma values and Figure 3—figure supplement 3 for a subset of our own experimental data [Ivanovic et al . , 2013] ) . Thus , considered in the context of our current simulations ( Figure 3D ) , the relatively low experimental Ngamma values support and generalize ( beyond the experimental results of Otterstrom et al . ( 2014 ) the interpretation that even in the absence of targeted inhibition , a substantial portion of the sites on the virion surface lacks the potential to participate in fusion . We further conclude that contrary to previous contentions ( by us and others ) , Ngamma alone does not distinguish among 3 , 4 and 5 as the number of cooperating HA-neighbors because at high fnp , the theoretical Ngamma values all closely match the experimental observations . On the other hand , the experimental values do rule out 6 , for which the simulation derived Ngamma was greater than 4 , even for the highest fnp values . Furthermore , in the simulations , kgamma derived from hemifusion-delay distributions was larger than the value for the rate constant ( ksim ) corresponding to the probability used in the computation , but it approached this value at high fnp ( Figure 3E , plateau regions yield kgamma/ksim between 1 . 5 and 2 ) . In a large contact patch with a high fraction of participating HAs ( low fnp values ) , there are many ways to achieve Nh neighbors ( Figure 3A ) ; as fnp increases , that redundancy decreases , and kgamma becomes a better approximation to ksim . kgamma does not reach the value of ksim even at the highest fnp values , at which a majority of the virions that can hemifuse have only one way to reach hemifusion because they have only a single patch of Nh active neighbors within a larger contact area containing mostly inactive or non-productively refolded HAs . Thus , to determine the rate constant for membrane engagement by individual HAs , one needs to determine the fraction of non-participating sites . We have examined as follows the relative contributions to non-participating sites from NA , HA0 and non-productive HA1:HA2 refolding . The clustered localization of NA on a virion and its surface occupancy of 10-15% ( Harris et al . , 2006 , Calder et al . , 2010 , Wasilewski et al . , 2012 ) lead us to expect NA to make only a very small contribution . In Figure 4 and Figure 4—figure supplement 1 , we show that the virions used in our previous experiments ( Ivanovic et al . , 2013 ) had fully processed HA and that the HAs had full potential to assume the low-pH induced conformation . We thus conclude that non-productive HA refolding is the major component of non-participating sites in our previous experiments . Given similar predictions for fnp values based on experimental Ngamma values from the preceding paragraph , this conclusion might well extend to other single-virion experiments of influenza membrane fusion ( Floyd et al . , 2008 , Otterstrom et al . , 2014 ) , although we cannot formally conclude that here . For simplicity , however , in the subsequent set of analyses , we refer to non-participating sites in the absence of targeted HA inhibition as unproductive HAs , and their frequency on the virion surface as fun . 10 . 7554/eLife . 11009 . 010Figure 4 . Complete processing of virion-associated HAs and complete conformational change at low pH . We show WT UdornHA-Udorn and X31HA-Udorn virions used in our previous single-virion fusion experiments ( Ivanovic et al . , 2013 ) . SDS-PAGE and western blot of virions probed with HA1-specific antibody that detects both HA0 and HA1 alone . ( A ) Recombinant X31 HA0 and HA1:HA2 are included as a reference . The various HA forms appear to show varying levels of glycosylation resulting in different gel migration patterns . A trace amount of unprocessed HA0 is apparent in only one of two X31HA-Udorn preparations ( lane 6 , band location marked with an arrow ) . ( B , C ) Virions were incubated in either neutral or pH5 . 2 buffer for indicated times at 37°C . ( B ) Virions were either loaded directly onto the gel or treated with trypsin prior to loading . Resistance to trypsin digestion of virion-HA incubated in neutral buffer is a control for pre-fusion HA integrity . HA1tr is the trypsin-resistant fragment of HA1 ( C ) Virions were immunoprecipitated with LC89 antibody ( specific for the low-pH form of HA2 [Wharton et al . , 1995] ) , and the entire bead-associated fraction ( P ) and the supernatant ( S ) were loaded onto separate lanes of the gel . Ab refers to the band corresponding to the heavy chain of the antibody used for immunoprecipitation , detected with the secondary antibody used in the western blot . Complete HA conversion to trypsin-sensitive form or to a form that can be immunoprecipitated with LC89 antibody is apparent by 1 min for Udorn HA and by 60 min for X31 HA . The conversion kinetics for X31-HA are disproportionately slower than its fusion kinetics ( Ivanovic et al . , 2013 ) ; see the Discussion for consideration of the consequences of these observations for the fusion mechanism . An analogous set of results for the second UdornHA-Udorn and X31HA-Udorn clones are shown in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 01010 . 7554/eLife . 11009 . 011Figure 4—figure supplement 1 . Complete processing of virion-associated HAs and complete conformational change at low pH . See main Figure 4 , panels ( B , C ) legend , showing an analogous experiment performed with a different set of UdornHA-Udorn and X31HA-Udorn clones . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 011 Otterstrom et al . ( 2014 ) studied inhibition of hemifusion by Fabs and IgGs of HA stem-directed antibodies . They determined that for H3N2 X31 virions , an average of 261 bound Fabs gave half-maximal hemifusion inhibition and that 493 Fabs inhibited hemifusion completely . ( We consider only their Fab data here , to avoid potential complications from divalent binding of IgGs . ) We simulated inhibition , taking 375 as the number of HAs per virion ( 1125 Fab sites ) ( see Materials and methods ) ( Figure 5A ) . We assumed random HA occupancy and postulated that a single bound Fab prevents the fusion transition of a trimer . We varied fun values and looked for fractions that gave 50% hemifusion-yield inhibition for 261 bound Fabs and near complete inhibition for 493 bound Fabs . For Nh = 3 and Nh = 4 , we obtained essentially unique answers for fun ( Figure 5B and C ) : 0 . 65 with Nh = 3 , and 0 . 4 with Nh = 4 . With Nh = 5 , no condition was consistent with the measured values ( Figure 5D ) . This treatment of the inhibition data has thus allowed us to determine possible pairs of values for the number of neighboring HAs required for hemifusion and the fraction of unproductive HAs . For somewhat reduced fun values , the data are also consistent with a smaller patch size ( see Figure 5—figure supplement 1 ) . This result makes intuitive sense because conceptually , a smaller patch size is like a larger patch size with more non-participating sites . 10 . 7554/eLife . 11009 . 012Figure 5 . Hemifusion yield as a function of the fraction of unproductive HAs ( fun ) for virions with no bound antibody and for those with 261 or 493 bound Fabs ( PS = 55 ) . ( A ) Illustrations of simulated contact patches . The frequency of Fab-bound HAs ( fFab ) and fun were combined in the parameter fnp as described in Materials and methods . ( B–D ) The results for Nh = 3 ( B ) , Nh = 4 ( C ) , and Nh = 5 ( D ) were derived from simulations that yielded 1000-3000 hemifusion events . Non-zero fun values ( boxed out regions in ( B ) and ( C ) are required to explain the experimentally observed number of Fabs required for half-maximal ( 261 ) and maximal ( 493 ) inhibition of H3N2 X31 influenza virus hemifusion ( Otterstrom et al . , 2014 ) . Experimental data are inconsistent with Nh = 5 . The corresponding results for PS = 55 are shown in Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 01210 . 7554/eLife . 11009 . 013Figure 5—figure supplement 1 . Hemifusion yield as a function of the fraction of unproductive HAs ( fun ) for virions with no bound antibody and for those with 261 or 493 bound Fabs ( PS = 121 ) . See main Figure 5 legend which shows results of an analogous set of simulations using PS = 121 instead of PS = 55 . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 013 The following more complete analysis of the data in Otterstrom et al . ( 2014 ) favors the interpretation that for X31 H3 HA , three HA neighbors cooperate during fold-back . To facilitate comparison with the reported data , we derived from simulations values for the yield of hemifusion , for the geometric mean of hemifusion-delay times , and for Ngamma and kgamma , as functions of the number of Fabs bound per virion ( Figure 6 and Figure 6—figure supplement 1 ) . We carried out these simulations for the permitted Nh:fun pairs ( obtained from the data in Figure 5 and Figure 5—figure supplement 1 ) as we increased fFab across the reported range . We adjusted ksim so that the geometric mean of the hemifusion delay times in the absence of any bound Fabs was ~30 sec , the value reported for H3N2 X31 virions under the conditions of the measurements in Otterstrom et al . ( 2014 ) . For either patch size , this procedure yielded values for ksim of 0 . 02 and 0 . 017 sec-1 for Nh = 3 and Nh = 4 , respectively ( Figure 6 and Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 11009 . 014Figure 6 . Effects of Fab binding on hemifusion yield and kinetics for given pairs of Nh and fun ( PS = 121 ) . ( A ) Illustrations of simulated contact patches at the time of hemifusion for several fnp values ( fun was kept constant while fFab was increased ) . ( B–E ) Comparison of simulation-derived results ( 1000–3000 hemifusion events ) for hemifusion yield ( B ) , hemifusion delay ( geometric mean ) ( C ) , Ngamma ( D ) and kgamma ( E ) with experimental data for H3N2 X31 influenza from Otterstrom et al . ( 2014 ) ( black triangles ) . Experimental hemifusion yield data in ( B ) ( their Figure 2C ) were scaled so that the highest measured hemifusion yield value became 100% ( i . e . each data point was multiplied by 4/3 ) . The corresponding results for PS = 55 are shown in Figure 6—figure supplement 1 . For simulations testing the effect of sample size on variability in Ngamma , see Figure 6—figure supplement 2 . For a further test of the robustness of the conclusions derived from this figure , see Figure 6—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 01410 . 7554/eLife . 11009 . 015Figure 6—figure supplement 1 . Effects of Fab binding on hemifusion yield and kinetics for given pairs of Nh and fun ( PS = 55 ) . See main Figure 6 legend , which shows results of an analogous set of simulations using PS = 121 instead of PS = 55 . For the adjusted values of fun , PS = 55 data are indistinguishable from PS = 121 data . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 01510 . 7554/eLife . 11009 . 016Figure 6—figure supplement 2 . Effect of sample size on variability in Ngamma . We derived Ngamma values for 6 small datasets and a single large data set for both PS = 55 and PS = 121 and for each pair of Nh and fun identified for H3 X31 and H1 PR8 HAs in Figure 5 and 7 , respectively . ( A ) Plots showing derived Ngamma values for small datasets ( black symbols , n~100 ) and large datesets ( red pluses , n~2000 ) . Circled results are derived using the functional parameters ( Nh and fun ) that best agree with Otterstrom et al . ( 2014 ) data . ( B ) Individual histograms for PS = 55 data that went into plots shown in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 01610 . 7554/eLife . 11009 . 017Figure 6—figure supplement 3 . Effects on our conclusions of potential error in the measurement of the number of Fabs needed for 50% hemifusion inhibition ( #Fab1/2hemi ) for H3N2 X31 influenza virions . For the simulation results shown , we included several #Fab1/2hemi values covering the entire 95% confidence interval for this measurement ( Otterstrom et al . , 2014 ) . ( A ) fun required to give half-maximal inhibition in the hemifusion yield . Each data point on this plot represents the result of a separate analysis , either shown as red bars in Figure 5 ( orange diamonds ) or an analogous result derived for each new #Fab1/2hemi value ( black diamonds ) . ( B–E ) Subsequent analyses then used the fun:Nh pairs determined in ( A ) to repeat simulations shown in Figure 6 – fun was kept constant , while fFab was increased . We expressed results from Figure 6 , panels ( B-E ) ( orange diamonds ) and from these new analogous titrations ( black diamonds ) each as a single value: ( B ) hemifusion yield for virions with no bound Fab , ( C ) increase in hemifusion-delay over the entire range of bound Fabs , ( D ) Ngamma for virions with no bound Fab , and ( E ) decrease in kgamma over the entire range of bound Fabs . We found that even the lowest value for #Fab1/2hemi did not include the possibility of 4 HAs participating in hemifusion , as the value of Ngamma ( D ) and decrease in kgamma ( E ) from simulations fit poorly ( compare to experimentally derived values shown in Figure 6 , black triangles ) . The results for the increase in hemifusion delay ( C ) and the kgamma drop ( E ) limit the acceptable range of #Fab1/2hemi for Nh = 3 to the lower half of the 95% confidence interval ( the boxed-out shaded region ) ( see Figure 6 , black triangles ) . We thus conclude that 3 stochastically triggered HA neighbors cooperate during fold-back , and that fun is between 0 . 6 and 0 . 7 ( PS = 121 ) . An analogous set of analyses for PS = 55 yielded indistinguishable main conclusions and yielded values for fun between 0 . 5 and 0 . 6 ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 017 Figure 6C shows that for Nh = 3 , the mean hemifusion delay time in the simulation increased from ~30 to ~80 sec ( a 2 . 7-fold increase ) as the number of bound Fabs increased from zero to 500 ( the latter corresponding to slightly under half occupancy ) . For Nh = 4 , the delay time with 500 bound Fabs was 100 sec ( a 3 . 6-fold increase ) . Again , the comparison is independent of patch size , as expected ( see comment above ) ( Figure 6—figure supplement 1 ) . Otterstrom et al . ( 2014; their Figure 3 ) reported a 2 . 6 ± 0 . 4-fold increase , i . e . a delay time of ~80 sec for 500 bound Fabs , in good agreement with the simulation for Nh = 3 ( to facilitate comparison with our simulations , we plotted these published experimental data onto the panels in Figure 6B–E ) . Figure 6D shows that for Nh = 3 , Ngamma was approximately equal to 3 and nearly independent of the number of bound Fabs , while for Nh = 4 , Ngamma fell from greater than 5 , for no bound Fabs , to about 3 at higher Fab occupancies . Otterstrom et al . ( 2014 ) reported Ngamma ~2 . 5 , with little dependence on Fab occupancy , again in better agreement with the Nh = 3 simulation results . We verified that the predicted 2-point drop in Ngamma would be evident despite the uncertainty in fitting Ngamma inherent in small datasets ( Figure 6—figure supplement 2 , H3N2 results ) . Furthermore , for Nh = 3 , kgamma from simulation showed a moderate ( ~threefold ) drop from ~0 . 1 to ~0 . 03 sec-1 , again in much better agreement with the shown experiment values ( Otterstrom et al . , 2014 ) than the predicted ~fivefold drop in this value for Nh = 4 ( Figure 6E ) . We further tested the robustness of the above conclusions against potential uncertainty in the measured value for the number of Fabs ( #Fab1/2hemi ) needed to achieve half-maximal hemifusion inhibition ( Figure 6—figure supplement 3 ) . We conclude that Nh = 3 gives very good agreement of simulation and experiment for several observed or derived parameters and a range of #Fab1/2hemi values . A consequence is that for H3N2 X31 virions under the experimental conditions of Otterstrom et al . ( 2014 ) , the rate constant ( ke ) for the limiting kinetic step during productive HA extension corresponds to ksim for the combination of parameters that best fits all the observations ( ~0 . 02 sec-1 ) ( see above ) . Moreover , Figure 6B shows that to fit the observed data , all virions must have the potential to fuse ( that is , the simulated yield of hemifusion in the absence of Fabs is 100% , when the simulations are run with the parameters that best fit all the observations ) . The yield of hemifusion for H3N2 X31 virions reported by Otterstrom et al . ( 2014 ) was about 60% , which thus calibrates the efficiency of the assay and the method of virion detection . The yield in our own earlier work on H3N2 X31 and Udorn particles was about 80% ( Ivanovic et al , 2013 ) . The number of bound Fabs required to inhibit fusion of H1N1 PR8 influenza virions in the experiments of Otterstrom et al . ( 2014 ) was substantially lower than for H3N2 X31 — on average , 74 Fabs for half-maximal inhibition and 248 Fabs for complete inhibition . This difference suggests either that PR8 viruses require more HAs for hemifusion or that non-productive conformational changes are more likely ( or both ) . ( Virion size was the same for the H3 and H1 strains , so patch-size difference is not the reason for their differential neutralization susceptibility . ) Following the same procedure as above for H3 HA ( Figure 5 ) , we could find , for each value of Nh between 3 and 6 , a single value for fun that gave both 50% fusion-yield inhibition for 74 bound Fabs and near-complete inhibition for 248 bound Fabs ( Figure 7 ) . As expected , for somewhat reduced fun values , the data are also consistent with a smaller patch size ( see Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 11009 . 018Figure 7 . Hemifusion yield as a function of fun for virions with no bound antibody or those with 74 or 248 bound Fabs ( PS = 121 ) . ( A ) Illustrations of simulated contact patches . ( B–E ) The results for Nh = 3 ( B ) , Nh = 4 ( C ) , Nh = 5 ( D ) , and Nh = 6 ( E ) were derived from simulations that yielded 1000 to 3000 hemifusion events . Non-zero fun values ( boxed-out regions ) are required to explain the experimentally derived number of Fabs required for half-maximal ( 74 ) and maximal ( 248 ) inhibition of H1N1 PR8 influenza virus hemifusion ( Otterstrom et al . , 2014 ) . For different fun values , data are consistent with Nh = 3–6 . The corresponding results for PS = 55 are shown in Figure 7—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 01810 . 7554/eLife . 11009 . 019Figure 7—figure supplement 1 . Hemifusion yield as a function of fun for virions with no bound antibody or those with 74 or 248 bound Fabs ( PS = 55 ) . See main Figure 7 legend , which shows results of an analogous set of simulations using PS = 121 instead of PS = 55 . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 019 We proceeded to distinguish among the potential pairs of values for Nh and fun as we did with the H3N2 X31 data ( Figure 8 and Figure 8—figure supplement 1 ) . We carried out the simulations for each of the permitted Nh:fun pairs ( obtained from the data in Figure 7 and Figure 7—figure supplement 1 ) , and calculated the various experimentally observed parameters as we increased fFab until near complete hemifusion inhibition ( Figure 8A ) . We adjusted the values for ksim so that the mean hemifusion delay time in the absence of bound Fab was about 46 sec , as determined by Otterstrom et al . ( 2014 ) . For either patch size , the corresponding ksim ranged from 0 . 029–0 . 037 sec-1 for Nh from 3–6 . The simulated yield of hemifusion for no bound Fab varied from about 65–70% for Nh = 5 or 6 to less than 50% for Nh = 3 or 4 ( panel B in Figure 8 and Figure 8—figure supplement 1 ) . Otterstrom et al . ( 2014 ) reported a 45% yield for H1N1; if we calibrate based on their yield for H3N2 of 60% , for which simulation indicates 100% ( see above ) , we get a ‘corrected’ yield of 75% . Although approximate , this rescaling takes into account the experimental uncertainties that will make the observed yield lower than modeled by the simulation; for example , the program used to select virus particles will with some frequency pick non-particles ( fluorescent spots ) that will certainly fail to fuse ( at least 7-9% in our published experiments: Ivanovic et al , 2013 ) . Imperfections in the planar bilayer would prevent detection of potential fusion events from particles that might land on them ( e . g . , stick to glass exposed at a hole in the bilayer ) . Moreover , within the assumptions of the simulation , the observed yield may not be higher than simulated , and in general lower . In experiments at low Fab concentration , Otterstrom et al . ( 2014 ) reported as much as 55% fusion; with IgGs , up to 65% in individual measurements . Even without rescaling , both these values are higher than the simulated values of yield for Nh = 3 or 4 at low Fab or IgG concentration . The more complete analysis in Figure 8—figure supplement 2 rules out Nh = 3 and disfavors Nh = 4 . 10 . 7554/eLife . 11009 . 020Figure 8 . Effects of Fab binding on hemifusion yield and kinetics for given pairs of Nh and fun ( PS = 121 ) ( A ) Illustrations of simulated contact patches at the time of hemifusion for several fnp values ( fun was kept constant while fFab was increased ) . ( B–E ) Comparison of simulation-derived results ( 1000–3000 hemifusion events ) for hemifusion yield ( B ) , hemifusion delay ( geometric mean ) ( C ) , Ngamma ( D ) and kgamma ( E ) with experimental data for H1N1 PR8 influenza from Otterstrom et al . ( 2014 ) ( black pluses ) . Experimental hemifusion yield data in ( B ) ( their Figure 2C ) were scaled using the same factor as in Figure 6B ( each data point was multiplied by 4/3 ) . The corresponding results for PS = 55 are shown in Figure 8—figure supplement 1 . For a further test of the robustness of the conclusions derived from this figure , see Figure 8—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 02010 . 7554/eLife . 11009 . 021Figure 8—figure supplement 1 . Effects of Fab binding on hemifusion yield and kinetics for given pairs of Nh and fun ( PS = 55 ) . See main Figure 8 legend , which shows results of an analogous set of simulations using PS = 121 instead of PS = 55 . For the adjusted values of fun , PS = 55 data are indistinguishable from PS = 121 data . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 02110 . 7554/eLife . 11009 . 022Figure 8—figure supplement 2 . Effects on our conclusions of potential error in the measurement of the number of Fabs needed for 50% hemifusion inhibition ( #Fab1/2hemi ) for H1N1 PR8 influenza virions . For the simulation results shown , we included several #Fab1/2hemi values covering the entire 95% confidence interval for this measurement ( Otterstrom et al . , 2014 ) . ( A ) fun required to give half-maximal inhibition in the hemifusion yield for each #Fab1/2hemi value . Each data point on this plot represents the result of a separate analysis , either shown as red bars in Figure 7 ( green diamonds ) or an analogous result derived for each new #Fab1/2hemi value ( black diamonds ) . ( B ) For each fun:Nh pair determined in ( A ) , we determined the hemifusion yield for virions with no bound Fab . The highest experimentally derived values for the hemifusion yield ( 65% , black vertical dotted line ) ( Otterstrom et al . , 2014 ) and the scaled value ( 75% , green dotted line , see Results for detailed reasoning ) are shown where included in the tested range of #Fab1/2hemi values . For Nh = 3 , the exclusion of the scaled hemifusion-yield value rules out the interpretation that 3 neighbors cooperate during hemifusion of H1N1 virions . While we cannot exclude Nh = 4 as a possible interpretation , the data are more consistent with Nh = 5 or 6 . An analogous set of analyses for PS = 55 yielded indistinguishable main conclusions ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 022 Simulation results for mean hemifusion delay , Ngamma and kgamma remained relatively constant as a function of bound Fab for all Nh ( Figure 8C–E ) because the corresponding fun was such that the starting point ( no bound Fab ) landed in the corresponding ‘plateau’ regions for these values ( see Figure 3 ) . Results for mean hemifusion delay times were indistinguishable for different Nh and thus could not help discriminate among these various possibilities . Furthermore , the published data in Otterstrom et al . ( 2014 ) show relatively small ( and hence noisy ) samples for their H1N1 experiments ( their Figure S8 and re-plotted here in Figure 8 ) . As we show in Figure 6—figure supplement 2 , estimates of Ngamma from runs with only 100 particles scatter quite widely around the value used in the simulation , and the observed Ngamma is thus not a good discriminator for deciding among Nh values between 4 and 6 . We conclude that for H1N1 PR8 viruses , Nh is greater than 3 and might be higher than 4 . A more precise estimate will require larger data sets . A consequence of the somewhat larger Nh is that for H1N1 PR8 virions under the experimental conditions of Otterstrom et al . ( 2014 ) , the rate constant ( ke ) for productive extension by individual HAs is ~0 . 034–0 . 035 sec-1 , nearly twice the rate of the corresponding step for H3 X31 influenza HA ( see above ) .
The outcomes of simulations we report here and their application to analysis of newly published data on inhibition of fusion by stem-directed Fabs ( Otterstrom et al . , 2014 ) are fully consistent with the model developed in our previous papers ( Floyd et al . , 2008 , Ivanovic et al . , 2013 ) . In that model , the number of HAs needed to generate a fusion event is not fixed by the organization of some intermediate state ( e . g . , by lateral interactions within a ring of HAs ) , but rather by the relationship between the free energy needed to overcome the kinetic barrier to hemifusion and the free energy gained in the HA2 conformational transition . Variation in Nh between influenza strains supports this mechanism . The new simulations extend the earlier model by including inactive ( or inactivated ) HAs and by showing that data on Fab inhibition can help restrict the estimates for the number of HAs required to generate hemifusion and the fraction of participating HAs . Our new simulation results further expose limitations of the original analytical model that we and others used to interpret single-virion fusion kinetic data ( Floyd et al . , 2008 , Ivanovic et al . , 2013 , Otterstrom et al . , 2014 ) . The standard analytical treatment of sequential kinetics ( the gamma distribution ) falls short , because the fusion mechanism involves stochastic events across a large enough interface that one of several potential initiating events will go on to completion . Even in the context of targeted HA inhibition analyzed here , and in a particular instance when most of the virions that are fusion competent have only a single potential region with Nh active HA neighbors , the gamma distribution parameters , N and k do not reflect the underlying number of HA participants or the rate of their extension ( Figure 3 ) , because the Nh HAs can extend in any order and there are more ways for the initial event to occur than for the next . Although the gamma distribution continues to be a useful tool to correlate experiment and simulation results , an updated analytical model would be needed to capture the fusogenic molecular events at the virus target-membrane interface , as we now understand them . Moreover , while our current simulation model does well in the context of accumulating single-virion membrane fusion data , it is likely that this model also will evolve as we gain new experimental insight . The experiment , computer simulation , and mathematical modeling will continue to evolve together , because they serve as independent tests for mutual validity and reliability and because each can lead to predictions that can be tested by one of the other , complementary approaches . We showed in our previous paper that the rate of fusion-peptide release from the pocket near the three fold axis sets the rate constant for target-membrane engagement ( Ivanovic et al . , 2013 ) . This rate in turn depends on the overall stability of the pre-fusion conformation and ( at a given pH ) on the overall pK of the particular HA species . Simulations described here and comparisons with data from Otterstrom et al . ( 2014 ) identify two additional parameters that determine the overall rate of HA-mediated fusion — the number ( Nh ) of participating HA trimers required to distort the apposed membranes into a hemifusion stalk and the fraction within the contact zone of participating ( active and productively refolded ) HAs ( Figure 9 ) . We show by comparing data from an H3N2 strain and an H1N1 strain that Nh can vary from one strain to another even under the same experimental conditions . ( These differences may or may not represent subtype specific differences . ) Nh times the free energy recovered in the fold-back step from an extended intermediate to the postfusion ‘trimer of hairpins’ must exceed the kinetic barrier to hemifusion , estimated to be at least 50 kcal/mol ( Harrison , 2015 ) . It is reasonable to expect that the free-energy recovery , and hence the required Nh , will depend on the particular HA in question . 10 . 7554/eLife . 11009 . 023Figure 9 . Independent functional determinants of HA-mediated membrane fusion and their effects on the influenza virus susceptibility to neutralization . Conclusions are presented in the context of the PS = 121 contact patch . ( A ) The rate of irreversible HA extension ( ke ) and the frequency of unproductive or inactive HAs determine the rate of target membrane engagement by individual HAs . First-event delay – the average time to the first HA conversion , either productive or non-productive – is determined solely by the ke and the patch size . ( See Figure 9—figure supplement 1 for the corresponding model that uses PS = 55 ) . Stochastic HA triggering dictates that small changes in the number ( Nh ) of HAs required for fold-back have significant effects on the kinetics of fusion . Small increases in Nh significantly reduce the extent of fusion ( purple bars ) in the context of the large fun values . Compensatory differences in ke , fun and Nh between X31 H3N2 and PR8 H1N1 influenza result in similar overall rates of hemifusion ( delay of about 36 and 58 sec , respectively ) . *Note that by exchanging the ke values between the H3 and H1 functional variables ( i . e . compare results for ke = 0 . 034 sec-1 , Nh = 3 , fun = 0 . 65 and ke = 0 . 02 sec-1 , Nh = 4 or 5 , fun = 0 . 75 or 0 . 65 ) , we obtain ‘extreme’ values for hemifusion delay or ~20 and ~100 sec , respectively . ( B ) Illustration of the effects of Fab binding on fusion kinetics ( mean hemifusion delay ) and the theoretical hemifusion yield ( purple bars ) in the context of functional variables revealed for H3N2 X31 and H1N1 PR8 influenza virions . Our conclusions reveal an intricate link between the molecular features of the evolved fusion mechanism and its susceptibility to neutralization . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 02310 . 7554/eLife . 11009 . 024Figure 9—figure supplement 1 . Independent functional determinants of HA-mediated membrane fusion and their effects on the influenza virus susceptibility to neutralization . Conclusions are presented in the context of the PS = 55 contact patch . See main Figure 9 legend . DOI: http://dx . doi . org/10 . 7554/eLife . 11009 . 024 The fraction of participating HAs , which determines the probability that Nh neighboring HAs will all be active , will depend on the percent of uncleaved HA0 , the percent of inhibitor-bound ( e . g . , Fab-bound ) HA , and the probability that any particular HA will fail to engage the target membrane and instead fold back and insert its fusion peptides into the viral membrane . In addition to governing the rate of release , the fusion-peptide amino-acid sequence , which is very highly conserved ( Nobusawa et al . , 1991 , Cross et al . , 2009 ) , may influence the efficiency of target-membrane insertion . It is also plausible that continued receptor engagement by HA1 might contribute to the probability of target-membrane engagement ( Figure 1 ) . Ordering of HA conformational transitions in the context of membrane fusion may vary among strains , but some features are suggested by studies of soluble HA ectodomain ( Godley et al . , 1992 , Garcia et al . , 2015 ) . If the C-terminus of HA2 becomes disordered before the rest of the conformational changes that allow HA extension , then HA1-receptor engagement will increase the probability that fusion peptide sequences project toward the target membrane instead of inserting back into the viral membrane ( Figure 1 ) . The general approach developed in our previous papers ( Floyd et al . , 2008; Ivanovic et al . , 2013 ) has also been used to study the flavivirus fusion mechanism ( Chao et al , 2014 ) . Flaviviruses have about 25% the surface area of even the smallest influenza virions and can display at most 60 trimers ( about 15% of the number on a typical small influenza virus particle ) . A transition from dimer-clustered E-protein subunits to fusogenic trimers is a component of the mechanism not required when the fusogen is already trimeric like influenza virus HA . Nonetheless , the fusion mechanisms for the two groups of viruses are relatively similar . Trimerization of the flavivirus E-protein subunits and target-membrane engagement of their fusion loops are rate-limiting; hemifusion requires at least two adjacent trimers . Simulations show that trimerization is a bottleneck because of limited availability of competent monomers within the contact zone between virus and target membrane , so that trimer formation must await monomer activation ( e . g . , dimer dissociation ) . The basic concepts revealed by our current analyses might thus be generalizable to other viral membrane fusion systems . The constraints imposed by fitting the hemifusion yield and the hemifusion delay time as functions of the number of Fab-inactivated HAs have allowed us to determine the fraction of unproductive HAs . This determination has in turn allowed us to associate the ksim value with the rate constant ( ke ) for the limiting step during membrane engagement . Although the specific value for fun depended on patch size , the underlying rate constant did not ( compare Figure 9 and Figure 9—figure supplement 1 ) . We have previously concluded from the fusion kinetics of HA mutants that the rate-limiting step of membrane engagement is the release of the fusion peptide from its ‘prefusion’ pocket near the threefold axis of the trimeric HA ( Ivanovic et al . , 2013 ) . Reversible fluctuations at HA1:HA1 , HA1:HA2 and HA2:HA2 interfaces ( Figure 1 , ‘open state’ ) determine a ‘window of opportunity’ for fusion-peptide release and for their irreversible projection beyond the outer margins of adjacent HA1 heads . The link between ke and the rate constant for a specific molecular rearrangement should in the longer term allow us to derive direct information for individual fusion catalysts in a functionally relevant context . In our analysis of conformational changes for virion-associated HA in the absence of target membranes , we have made the unexpected observation that the rate of irreversible inactivation for X31 HA is accelerated at the target membrane interface . It took 10 min at pH5 . 2 and 37C for about half of the HAs on a virion surface to inactivate irreversibly ( Figure 4 and Figure 4—figure supplement 1 ) . The same virions hemifuse with a mean delay of ~1 min at the same pH and at a much lower temperature ( 23°C ) ( Ivanovic et al . , 2013 ) . According to our current simulation model , for fnp = 0 . 5 and Nh = 3 , at the time of hemifusion , an average of 34% of HAs at the target-membrane interface are no longer in the pre-fusion conformation ( i . e . have inserted in the target membrane or become inactivated ) . This is at least an order of magnitude greater than their rate of inactivation on free virions . Because the frequency of non-productive HA refolding is high ( at least ~50% ) , the presence of a target membrane appears to accelerate both productive and non-productive refolding . We illustrate in Figure 1 a model that could explain these observations . Receptor engagement might retain HA1 in a configuration separated from HA2 ( an ‘open-HA" conformation ) and thereby increase the time interval for fusion peptide release and irreversible HA extension . Receptor engagement might also influence the ratio of membrane insertion to HA inactivation ( see our earlier comment ) , but an overall increase in the rate of committed HA extension would in any case increase the rate at which HAs reach one or the other of those endpoints . The degree of rate increase ( with respect to inactivation of HAs on free virions ) will depend on the relationship between the lifetime of the open state and the probability of fusion-peptide release during the interval when HA1 is not in the way . Udorn HA does not exhibit the same relative increase in the rate of refolding ( Figure 4 and Figure 4—figure supplement 1 ) . After 1 minute of incubation at low pH , most of its virion-associated HAs have assumed the low-pH conformation , but the rate of Udorn hemifusion at pH 5 . 2 is only ~twofold higher than that of X-31 ( Ivanovic et al , 2013 ) . Udorn HA , with a destabilized docking of the fusion peptide , appears to have a much greater probability of fusion-peptide release during its unconstrained ( i . e . on free virions ) open-state lifetime than does X-31 HA , which requires , for comparably rapid extension , the increased open-state lifetime afforded by receptor interactions with HA1 . The Udorn fusion peptide might , however , be less efficient at inserting into the target membrane , because of the mutation of Gly to Ser at its fourth position . If so , the ratio of non-productive to productive HA transitions might be higher for Udorn than for X-31 . The proposed role for HA-receptor contacts in catalysis of membrane fusion , not just in cell attachment , should be directly testable by future single-virion membrane fusion experiments . An important consequence of this possibility is that adjustments in receptor affinity would effectively modulate not only the yield and kinetics of fusion , but also the susceptibility of the virus to neutralization ( Figure 9B ) . The rate of fusion-peptide exposure is higher for HA from PR8 H1N1 virus than for HA from X-31 H3N2 , but a greater Nh and potentially also a decreased productivity of refolding for the former strain leads to a somewhat lower overall rate of fusion ( panel A in Figure 9 and Figure 9—figure supplement 1 ) . Thus , compensatory changes appear to maintain the overall rate within an acceptable range and imply some degree of independence of the molecular mechanisms that modulate the three fusion-rate determinants . Influenza virus penetrates from low-pH endosomes , and the rate of fusion may have an optimum determined by a balance between the rate of acidification of the virion interior ( required to release viral RNPs from the matrix protein [Martin and Helenius , 1991] ) and the efficiency of penetration before the virus particle undergoes lysosomal degradation ( Ivanovic et al . , 2012 ) . Replication of influenza virus in birds , humans and pigs is constrained by different kinds of pressures on its cell-entry machinery ( stability of HA in the extracellular environment and its roles in receptor binding and membrane fusion ) ( Schrauwen and Fouchier , 2014 ) . Distinct mechanisms that independently modulate the properties of this molecular machinery might determine the potential of a given strain to adapt to replication in a new host . Similar considerations will determine the potential of HA to evolve resistance to inhibitors that target it . Higher Nh ( combined with relatively low productivity of HA refolding ) reduces the baseline yield of fusion and increases the susceptibility of the H1N1 strain used by Otterstrom et al . ( 2014 ) to a fusion inhibitor ( antibody ) ( Figure 9B ) . A recent study of HIV-1 cell entry combined experiment and simulation to show infectivity differences among HIV-1 strains that differ in the number of participating fusion proteins required for entry ( Brandenberg et al . , 2015 ) . Further studies of the range over which Nh can vary among influenza strains , even within subtypes , and molecular determinants of Nh , will be valuable for assessing levels of antibodies ( or other entry inhibitors ) required for protection . The high percentage of unproductive HAs is probably the most unexpected result of our analysis . In our own experiments , cleavage was complete , so remaining HA0 is not the reason for this observation . After release of the fusion peptide and formation of an extended intermediate ( driven , presumably , by the strong α-helical propensity of the segment between the α1 and α2 helices in HA2: Carr and Kim , 1993 ) , the relative efficiency of membrane engagement , which traps the extended intermediate , and HA2 fold-back will determine whether the HA is productive or not . Under the conditions of our experiments ( Floyd et al . , 2008 , Ivanovic et al . , 2013 ) and those of Otterstrom et al . ( 2014 ) , the two efficiencies appear to be comparable , and fusion occurs even with more than half of the HAs inactive . The relatively large proportion of non-productive conformational transitions ( fun ~0 . 65-0 . 75 ) ( Figure 9 ) lies within the region of the fusion inhibition curve in which small changes in fun will influence both yield and rate ( see Figure 3 ) . The large effect on fusion of a small number of bound antibodies ( Otterstrom et al . , 2014 ) is consistent with this prediction . A potential evolvability benefit for the virus is that a small decrease in fun will have a comparably strong effect , directly offsetting the effects of antibodies or potential fusion inhibitors . The relative insensitivity of the fusion mechanism to a high ratio of unproductive to productive HAs , and the potential for a direct contribution to the efficiency of fusion from adjustments in the fraction of non-productive events , combine to produce an extremely robust general mechanism .
We used the computer simulation algorithm we described previously ( Ivanovic et al . , 2013 ) with several modifications indicated below and annotated in the accompanying Source code ( the script – s_arrest_hemifusion_simulation_eLife2015resubmission . m , and the functions used by the script – generate_patch . m , s_randomdist . m , isaN2tuplet6AllGeos . m , and findFlippedNeighbors . m ) . In brief , we defined a circular contact patch incorporating either 121 or 55 HAs arranged in a hexagonal lattice , where each internal HA has exactly 6 HA neighbors ( Figure 2A ) . For simulations involving virions with a reduced fraction of active HAs , a defined fraction of HAs in random positions within the contact patch were flagged as inactive or unproductive ( different random positions for each analyzed virion ) ( Figure 2D ) . We assumed a single-step process for the irreversible extension of individual HAs leading either to membrane insertion ( active HA; productive path , Figure 1A ) or inactivation ( unproductive HA; non-productive path , Figure 1A ) . We first derived lag times for each HA ( both active and inactive/unproductive ) in the contact patch by random drawing from an exponentially decaying function with rate constant , ksim ( see below ) . We then sorted these times in ascending order and defined hemifusion time as the lag-time for the active HA that contributed the final , Nhth member to the previously inserted group of ( Nh-1 ) active HA neighbors . Inactive or unproductive HAs simply could not contribute to the inserted HA neighborhood . If the hemifusion event was not detected after the HA with the longest lag time was analyzed , the given ‘virion’ was flagged as ‘dead’ . The simulation process was repeated as many times ( ntotal ) as needed to yield ~1000–3000 hemifusion events ( nhemi ) for all results shown . We defined hemifusion yield as 100 ( nhemi/ntotal ) . A2 antibody hybridomas were a generous gift from Judith White , University of Virginia . LC89 antibody was a generous gift from Stephen Wharton , MRC National Institute for Research , London , UK . We previously verified that HA was completely processed to HA1:HA2 on all virions that were used in Ivanovic et al . ( 2013 ) study . We show this result here for WT virions of two different X31HA-Udorn and UdornHA-Udorn virus preparations used in that study ( each was derived from a separate plaque during initial purification ) . We further demonstrate the ability of these virion-associated HAs to convert to their low-pH form ( Figure 4 and Figure 4—figure supplement 1 ) . | Influenza ( or flu ) viruses can infect humans and other animals and can lead to life-threatening illness . To multiply , the virus particles must first enter a host cell . The final step in the entry process is the fusion of the membrane that surrounds the influenza virus with the membrane of the host cell . This event releases the core of the virus particle into the cell , where it can stimulate the cell to make more copies of the virus . To ensure that membrane fusion takes place at the right place and time , influenza virus decorates the surface of its membrane with a protein called hemagglutinin . This protein senses cues provided by the target cell and then undergoes a series of transformations that lead to membrane fusion . During this process , hemagglutinin molecules insert into the target cell membrane to bring together the viral and cellular membranes . In 2013 , a group of researchers developed a computer simulation algorithm to study the events that lead to membrane fusion . In the model , the hemagglutinin molecules on a virus particle are activated at random to insert into the cell membrane . Now , Ivanovic and Harrison – two of the researchers from the earlier work – compared the predictions of this model to experimental data from previous studies of membrane fusion by influenza virus particles . This approach shows that a substantial fraction of hemagglutinin molecules fail to contact the target-cell membrane and are permanently inactivated instead . Fusion nonetheless proceeds efficiently . Ivanovic and Harrison suggest that these inactive hemagglutinins provide an evolutionary backup store . For example , the proportion of hemagglutinins on a virus particle that insert into the cell membrane affects how fast fusion occurs and how sensitive the virus is to attack by host immune-system proteins called antibodies . Therefore , an ability to control how often hemagglutinins insert into the membrane could allow the virus to adapt to host immune responses . In the future , Ivanovic and Harrison’s findings could aid the discovery of drugs that inhibit the entry of influenza into human cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"microbiology",
"and",
"infectious",
"disease"
] | 2015 | Distinct functional determinants of influenza hemagglutinin-mediated membrane fusion |
Membrane constriction is a prerequisite for cell division . The most common membrane constriction system in prokaryotes is based on the tubulin homologue FtsZ , whose filaments in E . coli are anchored to the membrane by FtsA and enable the formation of the Z-ring and divisome . The precise architecture of the FtsZ ring has remained enigmatic . In this study , we report three-dimensional arrangements of FtsZ and FtsA filaments in C . crescentus and E . coli cells and inside constricting liposomes by means of electron cryomicroscopy and cryotomography . In vivo and in vitro , the Z-ring is composed of a small , single-layered band of filaments parallel to the membrane , creating a continuous ring through lateral filament contacts . Visualisation of the in vitro reconstituted constrictions as well as a complete tracing of the helical paths of the filaments with a molecular model favour a mechanism of FtsZ-based membrane constriction that is likely to be accompanied by filament sliding .
Membrane dynamics during cytokinesis are some of the most fundamental processes in biology , yet are poorly understood at the molecular and mechanistic level . During prokaryotic cell division the cell membrane and the cell envelope constrict , eventually leading to cell separation . In most bacteria and archaea , this is guided by a ring structure containing the bacterial tubulin homologue FtsZ protein ( Bi and Lutkenhaus , 1991; Löwe and Amos , 1998 ) , which polymerises in a GTP-dependent manner ( Mukherjee and Lutkenhaus , 1994 ) . During constriction , the FtsZ ring decreases in diameter through an unknown mechanism . The C-terminal tail of FtsZ links it to other components of the divisome , an ensemble of many proteins that facilitates essential functions during the cell division process , most importantly remodelling of the cell envelope . Components of the divisome engage in cell wall synthesis ( PBPs ) , synchronisation with chromosome dimer resolution ( FtsK ) , lipid II cell wall precursor flipping ( FtsW or MurJ ) , and many components currently have no known function ( reviews: Adams and Errington , 2009; Lutkenhaus et al . , 2012 ) . In Escherichia coli , binding of the FtsZ tail to ZipA and possibly more importantly to FtsA anchor the FtsZ ring to the membrane ( Pichoff and Lutkenhaus , 2007 ) . FtsA is a bacterial actin-like protein that forms domain-swapped , canonical actin-like protofilaments that are membrane associated through FtsA's C-terminal amphipathic helix ( Pichoff and Lutkenhaus , 2005; Szwedziak et al . , 2012; van den Ent and Löwe , 2000 ) . Several cellular regulatory processes influence the onset and progression of cell division through mechanisms that directly act on FtsZ . For example , SulA is induced during the SOS stress response and sequesters monomers , stopping FtsZ polymerisation ( Chen et al . , 2012 ) . In E . coli , both nucleoid occlusion and the oscillating , pole-protecting MinCDE system contain components that inhibit FtsZ function within the ring directly ( Bernhardt and de Boer , 2005; Dajkovic et al . , 2008 ) . Although progress has been exhilarating over that past 20 years or so , some of the most fundamental questions still remain: what happens during FtsZ ring constriction ? How are the filaments arranged in the ring ? What drives constriction ? Many different models have been proposed for the mechanism of FtsZ-based constriction ( reviewed in Erickson , 2009; Erickson et al . , 2010 ) . Essentially , three different approaches have been taken to validate the models: in vivo imaging of FtsZ constrictions using fluorescently labelled proteins . Electron cryotomography of frozen hydrated cells without labelling and , thirdly , in vitro reconstitution experiments with pure , fluorescently labelled proteins . The most recent results emanating from those studies are that the rings appear to show strong fluorescence intensity variations that may suggest that the FtsZ ring is discontinuous ( Holden et al . , 2014 ) . Equally , tomography data have been interpreted to show scattered individual FtsZ filaments , some precise distance away from the membrane ( Li et al . , 2007 ) . Reconstitution experiments with FtsZ and FtsA showed dynamic behaviour and liposome constrictions ( Osawa and Erickson , 2013; Loose and Mitchison , 2014 ) . However , obtaining detailed molecular and mechanistic information regarding the FtsZ ring , particularly the arrangement of individual filaments and subunits within a constricting Z-ring , has remained a formidable challenge . In this study , we obtained high-resolution images of the FtsZ ring in Caulobacter crescentus and Escherichia coli by means of electron cryotomography . Furthermore , we reconstituted a minimal constriction force-generating system from purified components in vitro , encapsulating Thermotoga maritima FtsA ( TmFtsA ) and FtsZ ( TmFtsZ ) in liposomes of sizes corresponding to those of a bacterial cell . We produced images and three-dimensional maps of filaments arranging themselves into ring structures around the liposome perimeters that coincided with constriction sites . The observed FtsZ ring architectures in C . crescentus and E . coli cells and in liposomes favour a mechanism of FtsZ-based cell membrane constriction that is accompanied by filament sliding , as was proposed previously ( Lan et al . , 2009 ) .
We started out by visualising division sites in an unmodified C . crescentus strain ( NA1000/CB15N ) because the thin Caulobacter cells are most suitable for electron cryotomography . When a log-phase culture was plunge-frozen and imaged , many dividing cells could be found . At the division sites , a series of dots arranged in a single line were found ( Figure 1A , top ) . Careful analysis of cellular tomograms ( Video 1 and Figure 1A , bottom ) revealed that the dots were in fact 2D projections of filamentous structures encircling the cell and likely forming a continuous ring , disrupted in the images at the top and bottom by the missing wedge of the tomography method . The filaments were at a distance of 15 ± 2 nm from the inner membrane ( Figure 1B ) , as previously reported ( Li et al . , 2007 ) . The long-standing problem of the missing wedge in electron tomography caused by our current inability to tilt the specimen much beyond 65° ( Figure 1A , bottom , white triangle , see also Figure 1—figure supplement 1 for more details on the missing wedge problem related to this study ) makes it impossible to follow features all the way around the cell's perimeter . It is important to note , however , that the protein filaments are visible and uninterrupted everywhere the missing wedge allows it , as can be gauged from the disappearance of the cell membrane and envelope ( Figure 1A , B ) . We could detect the filamentous rings in 20 out of 28 dividing cells after tomography . Of the 8 without obvious filamentous structures , the division had progressed too far in 3 and 5 were of poor quality because of the cell's orientation with respect to the tilt axis ( Figure 1—figure supplement 1 ) . Hence , the finding of complete rings is supported by the fact that perfect coincidence of any hypothetical gaps in constricting rings with the missing wedges in all of the analysed tomograms , found after all in entirely random orientations , would be remarkably implausible . 10 . 7554/eLife . 04601 . 003Figure 1 . FtsZ forms bands of filaments completely encircling C . crescentus and E . coli division sites , as visualised by electron cryotomography . ( A ) C . crescentus NA1000/CB15N division site with filaments near the inner membrane IM ( top panel , black dots highlighted by arrow , see also Video 1 ) . Bottom panel shows the same cell rotated 90° around the short axis of the cell . The Z ring ( arrow ) is continuous and only invisible where there is no image because of the missing wedge ( shaded triangle ) ( see Figure 1—figure supplement 1 for more details on the missing wedge problem ) . The cytoplasm ( beige ) , periplasm ( blue ) , and space between the OM and S layer ( cyan ) have been coloured for clarity . ( B ) More examples of continuous FtsZ rings found in C . crescentus cells . The filaments were on average 15 nm from the inner membrane . ( C ) Electron cryotomographic slice of the constriction site of a B/r H266 E . coli cell visualised perpendicular to the longitudinal axis , showing very similar FtsZ filaments when compared to C . crescentus ( Figure 1A , B ) and FtsZ ( D212A ) expressing E . coli cells ( Figure 1F ) and having roughly the same distance ( 16 nm ) to the IM . Video 2 demonstrates the likely helical nature of the arrangement of the FtsZ filaments ( see also Figure 1—figure supplement 2 ) . ( D ) Western blot showing total FtsZ levels in cells used in ( E–G ) are about 2 . 5× that of wild-type cells . ( + ) refers to un-induced , ( ++ ) was induced by 0 . 02% arabinose . EcZ is purified E . coli FtsZ protein . ( E–G ) 10-nm thick electron cryotomographic slices of E . coli cells expressing FtsZ ( D212A ) protein in a wild-type B/r H266 background . See also Figure 1—figure supplement 3 . ( E ) E . coli division site showing the cross-section of FtsZ filaments ( single row of black dots ) at the constriction site . See Video 3 . ( F ) Visualisation of the same cell along the longitudinal axis shows that FtsZ filaments are located ∼16 nm from the inner membrane ( IM ) . ( G ) Closer examination of the constriction site of another cell with higher expression level reveals FtsZ filaments form pairs , appearing as doublets of dark dots ( upper ) and orange spheres in the schematic illustration , on average 6 . 8 nm apart within the doublets ( lower ) . ( H–K ) 10-nm thick electron cryotomographic slices of E . coli cells expressing engineered protein constructs based on FtsZ ( D212A ) ( see also Figure 1—figure supplements 3 , 5 and Supplementary file 1 , Table B ) . ( H ) Extending the C-terminal linker of FtsZ by inserting a linker sequence pushes the filaments further away from the IM ( distance changed from 16 nm to a somewhat variable 16–21 nm ) . ( I ) Replacing the C-terminal FtsA-binding sequence of FtsZ with a membrane-targeting sequence ( mts ) makes FtsZ directly bind to the IM and results in FtsZ filaments closer to IM ( distance changed from 16 nm to 10 nm ) . No cell constrictions were observed with this construct . ( J ) Removing the C-terminal FtsA-binding sequence of FtsZ renders it unable to maintain a fixed distance to the IM and FtsZ filaments that were observed within the cytoplasm . ( K ) Removing the C-terminal flexible linker of FtsZ makes it prone to form multiple layers of filaments that form complete rings or helices . Tomography using this construct works better because it produces small minicells . ( L ) A closer inspection of the area marked with the black arrowhead in G shows beads along the filament as illustrated by the schematic drawing with a repeat distance of 4 nm as expected for FtsZ filaments . IM: inner membrane; OM: outer membrane; WT: wild-type; Q-rich: FtsN-derived flexible linker; mts: membrane-targeting sequence . Scale bars: 100 nm in ( A ) and ( B ) , 50 nm in ( E , F , H , I , J ) , 20 nm in ( C , G , K ) , 10 nm in ( H ) , 20 nm in ( L ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 00310 . 7554/eLife . 04601 . 004Figure 1—figure supplement 1 . The missing wedge problem in cellular electron cryotomography . Since it is impossible to tilt the sample support ( EM grids ) from −90° to +90° and because the thickness of the ice film increases at high tilt angles , electron tomograms miss significant amounts of data . ( A ) Simulation of the effects of the missing wedge . Modified from Palmer and Löwe , ( 2013 ) . A phantom image resembling a cell envelope was reconstructed for a full ±90° range and a ±60° range , the latter being typical for tilt series acquisition . ( B ) Schematic drawings explaining the angle ( blue ) between the tilt axis ( red ) and the cell axis ( black dashed line ) and the missing wedge angle ( green ) . The former can be anything between 0 and 90° , whereas the latter can be anything between 0 and 180° . Tilt series for the C . crescentus study ( Figure 1A–B ) were obtained using the ±65° range . ( C ) Examples of the effects of different orientations of cells in the microscope with respect to the tilt axis on the missing wedge . Cells that were aligned with the tilt axis produced the most complete tomograms since the cell thickness stayed constant over the angular range . High tilts of those perpendicular to the tilt axis did not provide any useful information since the effective cell thickness in the electron beam increased . Shown are projections along the long axis of the cell . It is important to note that the angle between the tilt axis and the longitudinal axis of the cell is crucial in order to obtain high quality tilt series , other factors such as cell thickness , ice thickness , and membrane invagination progression also affect the quality of the resulting tomograms significantly . Scale bar: 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 00410 . 7554/eLife . 04601 . 005Figure 1—figure supplement 2 . Electron cryotomograms of wild-type E . coli cells show filaments at the constriction sites . ( A , C ) 10-nm thick tomographic slices of two cells showing black dots near the constriction sites corresponding to cross-sections of filaments . Filaments are difficult to discern in this viewing direction because of the thick E . coli cells ( B , D ) Filaments are better visualised when viewed perpendicular to the constriction planes showing filaments near the IM . These images , together with Video 2 , suggest that FtsZ forms a closed ring with slight helicity near the constriction site . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 00510 . 7554/eLife . 04601 . 006Figure 1—figure supplement 3 . FtsZ forms bands of filaments at constriction sites in E . coli cells . ( A ) 10 nm electron cryotomographic slice of a cell expressing more FtsZ ( D212A ) protein than in Figure 1E ( corresponds to Figure 1G ) , oriented parallel to the longitudinal axis , showing one layer of dots near the constriction site , corresponding to cross-sections of FtsZ filaments that are 16 nm away from the IM . ( B ) Electron cryotomographic slice of the cell viewed perpendicular to the dashed line in ( A ) . FtsZ filaments and their relative position to the IM are illustrated with the schematic representation of the tomographic slice in ( C ) . ( D–E ) 10 nm electron cryotomographic slices of a cell with very low level expression of FtsZ ( D212A ) protein ( un-induced ) viewed parallel to the longitudinal axis in ( D ) and perpendicular to the dashed line in ( D ) , showing similar architecture of FtsZ filaments at the constriction site . Scale bars: 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 00610 . 7554/eLife . 04601 . 007Figure 1—figure supplement 4 . Engineered FtsZ proteins form filaments with altered localisation patterns in E . coli cells . ( A ) Extending the C-terminal flexible linker of FtsZ ( D212A ) makes the protein form filaments further away from the membrane with a distance to IM increased from 16 nm to 21 nm; ( B ) and ( C ) are tomographic slices of the cell viewed perpendicular to the dashed lines in ( A ) and segmentation illustrating the relative positions of FtsZ filaments and the IM; ( D ) cells expressing a membrane-binding FtsZ construct produced by fusing the E . coli MinD membrane-targeting sequence ( mts ) to the C-terminus of FtsZ produce filaments that are 10 nm away from IM; ( E ) removing the C-terminal FtsA-binding sequence of FtsZ gives filaments further away from the IM; ( F ) FtsZ without the C-terminal flexible linker tends to form multiple layers of filaments near the constriction site , and ( G ) such filaments appear to form complete rings or helices when viewed perpendicular to the plane of cell constriction . Scale bars: 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 00710 . 7554/eLife . 04601 . 008Figure 1—figure supplement 5 . Overview of FtsZ constructs used for in vivo tomography . Please also consult Supplementary file 1A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 00810 . 7554/eLife . 04601 . 009Video 1 . Tomogram of a wild-type C . crescentus cell showing tomographic slices parallel to the longitudinal axis of the cell . A single layer of dark dots corresponding to cross-sections of FtsZ filaments is clearly visible at a distance from the membrane on both sides of the septum . The missing wedge is located at top and bottom . The distance between adjacent filaments highlighted by the arrow varies along the z-direction . This corresponds to Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 009 To investigate the generality of these findings , we imaged unmodified E . coli B/r H266 cells , which we chose because of their thinness ( Woldringh , 1976 ) . Although the E . coli cells were thicker than C . crescentus , we found the wild-type filaments still discernible , most obviously when imaged along the long axis of cells ( Figure 1C and Figure 1—figure supplement 2 , especially B and D; Video 2 ) , and these were 16 nm away from the IM . Based on these observations we concluded that E . coli Z-rings were , like in C . crescentus , probably continuous and consisted of single-layered bands that are 5–10 filaments wide . 10 . 7554/eLife . 04601 . 010Video 2 . Tomogram of a wild-type E . coli cell showing the constriction site along the longitudinal axis of the cell . FtsZ filaments are visible in certain slices and are likely to be forming continuous helices indicated by its pattern when viewed along the slices . This corresponds to Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 010 In order to investigate if the filaments imaged in wild-type cells so far contained FtsZ protein , we over-expressed FtsZ ( D212A ) , a mutant protein that hydrolyses GTP much more slowly ( Redick et al . , 2005 ) in E . coli B/r H266 cells . When over-expressed to 2 . 5-fold total FtsZ ( Figure 1D ) , the protein formed a wide single layer of filaments at the division site ( Figure 1E , see also Video 3 and Figure 1—figure supplement 3 ) , very similar to the bands seen in unmodified cells , but wider and containing more filaments as would be expected because there is now more FtsZ protein in the cell and filament dynamics have been reduced because of the GTPase-reducing mutation D212A . Figure 1F ( Figure 1—figure supplement 3A–C ) provides a view rotated by 90° , showing again that the filamentous ring was located approximately 16 nm away from the inner membrane ( IM ) . The band of filaments most likely consisted of doublets of individual protofilaments , as is indicated in Figure 1G , which shows a cell with higher expression level ( Figure 1—figure supplement 3A–C ) . The filaments were on average 6 . 8 nm apart ( n = 17 , distance between centres of adjacent filaments within a doublet , Figure 1G , lower ) . It is currently not known what lateral interactions between FtsZ filaments cause this arrangement or if it is facilitated by other proteins . 10 . 7554/eLife . 04601 . 011Video 3 . Tomogram: FtsZ ( D212A ) expressed in E . coli cell forms doublet FtsZ filaments at the constriction site . The video shows tomographic slices parallel to the longitudinal axis of the cell . One single layer of dark dots corresponding to cross-sections of FtsZ filaments is clearly visible , and these dark dots tend to form pairs suggesting a doublet FtsZ filament architecture at the constriction sites formed with FtsZ and FtsZ ( D212A ) . This corresponds to Figure 1E . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 011 Because no specific label for electron cryotomography currently exists that works in E . coli , we decided to further confirm the identity of the filaments as being composed of FtsZ by systematic perturbations of the system in four ( a–d ) separate experiments with subsequent imaging by electron cryotomography ( Figure 1—figure supplement 5 , Supplementary file 1A , B ) . ( a ) Introducing extra amino acids into the flexible linker ( Buske and Levin , 2013; Gardner et al . , 2013 ) within FtsZ that separates the globular N-terminal domain of FtsZ from the small C-terminal helix that binds FtsZ's membrane anchor , FtsA ( Ma and Margolin , 1999; Szwedziak et al . , 2012 ) , increased the distance between the FtsZ ring and the IM from 16 nm to a somewhat variable 16–21 nm ( Figure 1H and Figure 1—figure supplement 4A–C ) . ( b ) Removing the C-terminal FtsA-interacting helix and replacing it with a membrane-targeting sequence ( mts ) from MinD protein ( Hu and Lutkenhaus , 2003 ) shortened the distance from the IM to 10 nm . No constrictions of the cells were observed ( Figure 1I and Figure 1—figure supplement 4D ) . Therefore , despite it having been used in earlier studies ( Osawa et al . , 2008 , 2009; Osawa and Erickson , 2011; Loose and Mitchison , 2014 ) , we agree with previous findings ( Osawa et al . , 2008 ) that this construct is non-functional in vivo and we did not include it in our subsequent in vitro investigations below . ( c ) Removing the C-terminal FtsA-interacting helix from FtsZ detached the filaments from the membrane , making them appear throughout the cytoplasm ( Figure 1J and Figure 1—figure supplement 4E ) . ( d ) Finally , removing most amino acids between the C-terminal FtsA-interacting helix and the globular body of FtsZ caused a minicell phenotype . Because of their size , minicells produce tomograms of higher quality and it was possible to determine the longitudinal subunit repeat of the filaments to be around 4 nm , very close to the expected value of 4 . 2 nm for FtsZ ( Figure 1K and L and Figure 1—figure supplement 4G ) ( Erickson et al . , 1996 ) . We conclude that the filament localisations reacted to our perturbations as expected for FtsZ and the subunit repeat was the same as for all known FtsZ protofilaments . The minicell tomogram ( Figure 1K ) is another indication that the filaments most likely encircle entire cells . Encouraged by reports that simultaneous over-expression of FtsZ and its membrane anchor FtsA led to additional division sites ( Begg et al . , 1998 ) , we imaged E . coli cells in which extra FtsZ ( D212A ) and FtsA were produced from a bicistronic expression vector , by electron cryotomography ( Figure 2A–E ) . Providing just these two proteins in excess ( twofold to fourfold total vs WT ) produced a severe phenotype with many extra constrictions visible ( Figure 2A , B ) . Since in these cells the normal FtsZ to FtsA ratio had been altered to be close to 1:1 , from normally 5 FtsZ:1 FtsA ( Rueda et al . , 2003 ) , FtsA filaments became visible at the constricting division sites ( Figure 2C–E ) . Actin-like FtsA binds to the membrane directly via its C-terminal amphipathic helix and polymerises into canonical actin-like protofilaments ( Lara et al . , 2005; Szwedziak et al . , 2012 ) . In this way , an artificially strong ‘FtsA ring’ became apparent , indicating that FtsA is located between the IM and FtsZ , being 8 nm away from both ( Figure 2D , E ) . 10 . 7554/eLife . 04601 . 012Figure 2 . Co-expression of FtsZ and FtsA in E . coli cells leads to extra septa . ( A ) A low-magnification 2D electron cryomicrograph ( transmission ) showing multiple constriction sites ( marked with black arrowheads ) along the cell . ( B–E ) 10-nm thick electron cryotomographic slices of cells co-expressing FtsZ ( D212A ) and FtsA ( bicistronic , 1:1 ) . Two layers of dots are visible at constriction sites in ( B ) and ( C ) , corresponding to FtsZ filaments and FtsA filaments , respectively , as labelled in the orthogonal view along the long axis of the cell ( D ) . FtsA filaments are almost in the middle between FtsZ filaments and the IM , at a distance of 8 nm from both FtsZ filament and IM as indicated in ( E ) . ( F ) Structured illumination microscopy images of cells expressing FtsZ ( D212A ) and FtsA , showing cell division and minicell formation , proving that the extra septa function to completion . ( G ) 10-nm thick electron cryotomographic slice of an E . coli minicell formed from cells expressing Thermotoga maritima FtsZ and FtsA proteins , with a deeply constricted area showing cross-sections of FtsZ and FtsA filaments ( black dots marked with white arrows ) . Distance between FtsZ filaments and IM is around 12 nm ( inset in G ) . The view highlights striking similarities to the in vitro reconstruction shown in Figure 3H–J & 5C . IM: inner membrane; OM: outer membrane . Scale bars: 500 nm in ( A ) , 100 nm in ( B ) , 10 nm in ( C , and also for inset in G ) , 20 nm in ( E , and also for D ) , 2 μm in ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 012 We then demonstrated that the FtsZ ( D212A ) and FtsA-over-expressing E . coli cells , we had imaged at high-resolution with electron cryotomography , were dividing and separating , despite looking quite distorted . For this , we employed structured illumination microscopy ( SIM ) on live cells ( Figure 2F ) . The cells showed a strong minicell phenotype , while performing many cell divisions randomly distributed along the cell length . This indicated to us that the extra constrictions and division sites were functional in the sense that they led to complete cell separation ( abscission ) . We concluded that the ability to produce extra constriction sites and divisions by just providing more FtsZ and FtsA indicates that these two proteins may be central components of the IM constriction force generator that is localised within the inner division apparatus ( Rico et al . , 2013 ) and that it may be possible to reconstitute membrane constriction with just the two of them . Indeed , this has recently been reported and has been imaged at low resolution using fluorescence microscopy ( Osawa et al . , 2008; Osawa and Erickson , 2013 ) , although not providing any molecular insights . So , we then used purified FtsZ and FtsA proteins for in vitro reconstitution experiments , in order to observe constriction . We avoided fluorescently tagged proteins as they have been shown to introduce artefacts ( Margolin , 2012 ) . We therefore used completely unmodified TmFtsZ and TmFtsA proteins from Thermotoga maritima , both of which are easy to obtain and handle and have crystal structures available ( Oliva et al . , 2004; van den Ent and Löwe , 2000 ) . It should be noted that extra care had to be taken in order to obtain proteins that did not have their disordered but important C-terminal tails cleaved during purification . Just to confirm that TmFtsZ and TmFtsA formed structures similar to the E . coli counterparts in vivo , we over-expressed TmFtsZ and TmFtsA in E . coli and imaged the sample by electron cryotomography ( Figure 2G ) . Minicells were formed and , at constriction sites , they contained filaments that closely resembled the filament arrangement observed here for E . coli FtsA and FtsZ over-expression in E . coli ( Figure 2C ) . The distance of TmFtsZ to the IM was shorter at 12 nm; this was expected because the linker between the very C-terminal TmFtsA-interacting helix and the body of TmFtsZ is much shorter , measuring around nine amino acids . Minicell formation might indicate that TmFtsA and TmFtsZ interacted with the E . coli cell division machinery or even supported membrane constriction on their own , but we did not investigate this further . When added onto a flat lipid monolayer , TmFtsZ and TmFtsA formed striking spirals ( Figure 3A ) . The filaments forming the spirals tended to form weak doublets and in the centre of the spirals , white material was visible that may have been lipid that had been pushed up by the spiral constricting , possibly via a sliding filament mechanism as has previously been observed for FtsZ alone by AFM ( Mingorance et al . , 2005 ) . Intriguingly , much larger dynamic chiral spirals of polar FtsA and FtsZ filaments have recently been reported on supported lipid bilayers ( Loose and Mitchison , 2014 ) , but the exact relationship with our observation is currently unclear as treadmilling and no constriction were observed on the supported bilayers . 10 . 7554/eLife . 04601 . 013Figure 3 . In vitro reconstitution of bacterial cell membrane constriction by the FtsZ ring from purified components . ( A ) Thermotoga maritima FtsA ( TmFtsA ) and Thermotoga maritima FtsZ ( TmFtsZ ) form spirals on a flat lipid monolayer , as indicated by a white dotted line . The filaments tend to appear as double strands ( doublets ) . Negative-stain electron microscopy . ( B ) Transmission electron cryomicroscopy allows resolution of the inner and outer leaflet of undisturbed liposomes ( top panel ) . When TmFtsA is added to the outside , an additional layer of density corresponding to FtsA becomes apparent ( middle panel ) . Recruitment of TmFtsZ by TmFtsA leads to the formation of two layers ( bottom panel ) . Taken together , we conclude that FtsA is sandwiched between the membrane and FtsZ filaments ( bottom panel ) . See also Figure 3—figure supplement 1 and Figure 3—figure supplement 2 . ( C–G ) Constriction sites are efficiently formed when TmFtsA and TmFtsZ are encapsulated in liposomes that have sizes comparable to bacterial cells . Five representative liposomes are shown using transmission electron cryomicroscopy ( hence are 2D projections of 3D objects ) . Importantly , constriction sites are only formed where a ring made of the two proteins is present ( black arrowheads ) and not at other sites where filaments are located . The TmFtsA and TmFtsZ layers are clearly visible ( inset H , same as boxed area ‘1’ in C; inset J , same as boxed area ‘2’ in C and inset I , which is from Figure 4 electron cryotomography data ) and the protein's organisation mirrors that present in E . coli cells ( compare with Figure 2C ) . The distance of 12 nm between TmFtsZ and the membrane ( inset J ) resembles that found in over-expressing cells ( see Figure 2G and also Figure 5C ) . ( E ) Intriguingly , liposomes are being constricted ( partially ) in the absence of added nucleotide . Scale bars: 50 nm in ( A–C ) , 25 nm for insets . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 01310 . 7554/eLife . 04601 . 014Figure 3—figure supplement 1 . TmFtsZ and TmFtsA on the outside of liposomes and in the presence of GMPCPP deform liposomes . ( A ) Low-magnification ( upper panel ) . More detailed snapshots ( lower panel ) show that the filaments are on the outside; however , they do not form rings but curved structures that are positioned in areas of negative membrane curvature that they probably induce . ( B ) Schematic representation of the curvature produced by co-polymerisation of FtsA and FtsZ , which have differing repeat distances of 5 and 4 nm , respectively . Since FtsA binds to the membrane , this arrangement will lead to negative curvature . Hence , the intrinsic , negative curvature of the FtsA:FtsZ filaments fits the curvature of the membrane on the inside . However , on the outside , the membrane curvature is positive , as is also shown in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 01410 . 7554/eLife . 04601 . 015Figure 3—figure supplement 2 . Control experiments showing that both TmFtsA and TmFtsZ form straight filaments when polymerised separately . And liposomes deform mostly after dilution . ( A ) When mixed , FtsA and FtsZ form curved filaments ( right panel ) . ( B ) TmFtsZ does not bind to liposomes on its own . Random electron cryomicroscopy images taken immediately after detergent dilution were analysed for liposome deformations . The plot in ( C ) shows the number of liposomes , out of 63 , that are perfectly round ( as per solidity quantity , defined in ( ImageJ ) ) . Clearly , liposomes become more deformed over a 30-min period after dilution . ( D ) Shows a spherical liposome without proteins added and ( E ) at time point 0 min , right after dilution . Scale bars 50 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 015 Since the FtsZ ring in vivo does not act on flat membranes , we then switched to liposomes formed from E . coli lipid extract . First , TmFtsZ and TmFtsA were added to pre-formed liposomes so that the proteins remained on the outside ( Figure 3B and Figure 3—figure supplement 1A ) . Reactions containing liposomes and proteins , as indicated , were vitrified and imaged by conventional 2D transmission electron cryomicroscopy . When no protein was added , the liposomes appeared as almost perfect circles ( spheres in projection ) and the bilayers were clearly visible as a double line , 5 nm apart ( Figure 3B , top , Figure 3—figure supplement 2D ) . The addition of TmFtsA alone led to the formation of an additional layer , probably consisting of only partly polymerised protein , and no strong deformations were observed . TmFtsZ alone did not cause deformations or generation of an additional layer ( Figure 3—figure supplement 2B ) . But when both TmFtsA and TmFtsZ were added , two extra layers , in addition to the liposome bilayer , became visible ( Figure 3B ) . Particularly in the presence of nucleotide , strong negative curvature was induced , leading to deformations of the liposomes ( Figure 3—figure supplement 1A ) . As was suggested previously , the co-polymerisation of FtsZ and FtsA will lead to bending and curvature because the subunit repeat lengths of FtsZ and FtsA are roughly 4 and 5 nm , respectively ( Figure 3—figure supplement 1B and Figure 3—figure supplement 2A ) ( Szwedziak et al . , 2012 ) . We propose that the subunit repeat mismatch causes some deformations from the outside of the liposomes , especially when FtsZ polymerisation-inducing GMPCPP is present so that long filaments are formed that will exert more mechanical force . Since TmFtsZ and TmFtsA induce negative membrane curvature , we concluded that in order to reconstitute the actions of these proteins correctly , we needed to incorporate them on the inside of liposomes . For this , CHAPS detergent-solubilised E . coli lipid extract was mixed with the proteins at high concentrations and then diluted many-fold . Lowering of the detergent concentration by dilution led to spontaneous liposome formation with most of the proteins on the inside . We demonstrated with a time-lapse experiment that liposomes did not form around pre-existing FtsZ scaffolds since we observed that most liposomes were initially perfectly spherical and that they deformed over a 30-min period , after which most of them were heavily misshapen ( Figure 3—figure supplement 2C , E ) . The liposomes were then analysed by transmission electron cryomicroscopy . Amazingly , when both TmFtsZ and TmFtsA were included on the inside of liposomes , clear constriction sites appeared and these occurred only when supported by filaments ( Figure 3C–G ) . The liposomes were around 300 nm in diameter , similar to that of a small bacterial cell . The protein filaments were arranged into three distinct structures: arcs that were mostly single filaments , presumably formed by the co-polymerisation of TmFtsZ and TmFtsA , showing the characteristic curvature caused by the repeat mismatch ( Szwedziak et al . , 2012 ) ; spirals with decreasing curvature that appeared in perfectly spherical areas of the liposomes; and rings of filaments , appearing as bands in projection , formed around liposome constriction sites , with varying diameters . It is very important to note that constrictions only appeared where there were rings of filaments . Filaments within the liposomes had very different curvatures , for example in Figure 3D , filaments seemed to go round the liposome at a very large diameter , compared to the ones in the constriction zone further to the left . When the constriction sites were imaged at higher magnification , it became possible to discern the TmFtsA and TmFtsZ filaments end-on ( Figure 3H–J ) . The FtsA filaments were again sandwiched between the liposome membrane and the FtsZ filaments , just as in the images obtained with E . coli cells ( Figure 2C , G ) . The architecture is easily explained with TmFtsA-binding to the inside of the liposome membrane via its amphipathic helix , presumably polymerising , and FtsZ polymerising on top of FtsA , binding to it via its C-terminal FtsA-binding peptide ( Pichoff and Lutkenhaus , 2005; Szwedziak et al . , 2012 ) . Nucleotide presence had some influence on the appearance of these constricted liposomes as GTP addition produced the most bilobed liposomes . Constriction itself , however , was largely nucleotide hydrolysis independent since not adding any nucleotide produced constrictions as well , and they appeared even tighter ( Figure 3E ) . Therefore , despite not being strictly required for constriction , we believe that the nucleotide only has an influence on the appearance thereof and this might be due to the fact that FtsZ forms much longer filaments with polymerisation-inducing GTP added and this would enable the filaments to span larger liposomes when the process starts . Given that TmFtsZ is a hyperthermophilic protein , significant hydrolysis of GTP to GDP is not expected . Next , we employed electron cryotomography to image the liposomes in three dimensions ( Figure 4 , Figure 4—figure supplement 1 , Videos 4–10 ) . Because the samples only contained lipid and two proteins , contrast was very high in the resulting tomograms ( Video 4 and 5 ) , making it possible to represent the volume data without segmentation , as single-threshold surfaces or as volume renderings ( Video 6–10 , Figure 4–source data 1 PyMOL session file ) . None of the figures or videos we present has been segmented , manually or automatically . Figure 4A , top shows a constricted liposome in stereo , highlighting the three distinct filament architectures in detail: arcs , spiral domes at the ‘poles’ , and the filamentous ring , pulling and constricting the membrane . Note that the liposome was only deformed where the ring was located ( and where it unfortunately touched the carbon grid at the top right ) . More examples in Figure 4A , bottom and Figure 4—figure supplement 1 show the same overall architecture , with the same mix of filament architectures ( also Videos 6–10 ) . The TmFtsAZ rings were between 30 and 90 nm in diameter , when we looked at several different liposomes , and the filaments were on average 7 . 8 nm apart ( n = 16 ) laterally ( Figure 4B , C ) as compared to 6 . 8 nm seen in E . coli cells ( within doublets ) . Because contrast was very high , the ring of filaments could be traced in most tomograms all around the inside of the liposome and this revealed that the filaments were not totally equidistant and often came into contact . This was also true for the in vivo situation in C . crescentus ( Video 1 ) where the distance between the filaments ( grey arrow ) changes around the ring . The rings were always closed , continuous without gaps , with some possibly consisting of one double filament forming a helix ( Figure 4—figure supplement 1 and Video 6 ) and others containing several shorter filaments in a helix-like arrangement ( Figure 4A stereo view and Video 10 ) . When the filaments were investigated end-on , they appeared to always be arranged close to 90° with respect to the liposome membrane tangent ( Figure 4C ) , which was also observed for the filaments in cells ( Figure 2C , G ) . All of these features are best demonstrated in Video 10 , which provides an overview of the constriction and filament architectures and should be consulted to appreciate these findings properly . 10 . 7554/eLife . 04601 . 016Figure 4 . Electron cryotomography of liposomes constricted in vitro by rings of TmFtsA and TmFtsZ . ( A ) Stereo view of a representative liposome highlighting three different structures made by the enclosed TmFtsA and TmFtsZ proteins . Note that our images derived from tomographic volume data have not been segmented , they are volume representations of the actual 3D tomographic data . Arcs ( also on the outside ) are filaments made of both FtsA and FtsZ , whose curvature is determined by the mismatch in TmFtsA and TmFtsZ polymers subunit spacing ( 5 nm vs 4 nm , see also Figure 3—figure supplement 1 & Figure 4—figure supplement 2 ) . Dome-like structures are slightly helical spirals of condensing TmFtsZ filaments attached to the membrane by TmFtsA . Importantly , only complete rings seem capable of constriction force generation . The ring might consist of overlapping filaments ( as in the stereo view and Video 10 ) or maybe a continuous helix of double filaments ( bottom panel , middle liposome with black arrowheads , see also Figure 4—figure supplement 1 and Video 6 ) . The bottom panel depicts more examples of different liposome shapes and sizes . The cross-section ( right ) shows the distribution of filaments ( red ) inside a liposome ( membrane in blue ) ( bottom right ) . Video 4 shows a complete 3D volume in grey scale . Video 5 shows a slice view at high magnification , demonstrating the excellent contrast these specimens generate , making it possible to see individual subunits and complete filament traces . Videos 6–9 show 3D views of several constricted liposomes . Figure 4—source data 1 enables 3D viewing of a liposome volume with PyMOL . ( B ) Close-up view of the FtsZ ring ( purple ) attached to the membrane ( blue ) , here shown as single-threshold surface representations ( these are not automatic or manual segmentations ) . The filaments overlap and interact laterally ( left panel ) . View along the long axis shows that the ring is a perfect closed circle ( middle panel ) . The black arrow points to where TmFtsZ and TmFtsA filaments are fully detached from each other . Individual filaments are resolved ( right panel ) . Video 10 shows a 3D walk-through the liposome , highlighting most features on the way . ( C ) Comparison of filament arrangements and geometries within the dome-like structures ( left panel ) and ring-like structures ( right panel ) . Cross-sections demonstrate that in both cases , the TmFtsAZ filaments are positioned close to perpendicular with respect to the membrane ( red symbols ) . However , the constriction force is generated only in the rings ( see Figure 5D for explanation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 01610 . 7554/eLife . 04601 . 017Figure 4–source data 1 . PyMOL ( version 1 . 7 ) session file showing volume and surface renderings of the liposome in stereo in Figure 4A , top . Note that this version of the data has been volume edited , removing some of the filaments in the surroundings of the liposome . Nothing has been changed on the surface . These representations have not been segmented ( automatically or manually ) ; they show the volume data points as present in the ( edited ) tomogram . Both surface ( threshold ) as well as volume data are available as objects ‘surf’ and ‘vol’ , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 01710 . 7554/eLife . 04601 . 018Figure 4—figure supplement 1 . Constrictions occur only at the site of filament ring formation . A stereo view of the liposome marked with the black arrowheads in Figure 4A ( bottom middle panel ) . A single helix made of filament doublets is marked with red arrow . Video 6 shows its architecture in more detail and in 3D . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 01810 . 7554/eLife . 04601 . 019Figure 4—figure supplement 2 . A mechanism explaining variable intrinsic FtsA:FtsZ filament curvature . At some stages of constriction , the ratio of FtsZ to FtsA in the ring may be higher than one . Normally , there is around five times more FtsZ in cells than FtsA , therefore only a few FtsA molecules may be sandwiched in between the IM and FtsZ filaments ( which form more easily than FtsA filaments ) , upper panel . As curvature increases , the mismatch of the FtsA ( orange ) and FtsZ ( grey ) repeats ( 5 vs 4 nm , respectively ) makes it possible to add more FtsA since the double filament ‘wants’ to bend . Full occupancy of both FtsA and FtsZ in the double filament leads to a curvature of about 60 nm . This mechanism could be another source of energy for constriction in addition to or alternative to the condensation energy gained from filament overlap ( mechanism B ) in the discussion . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 01910 . 7554/eLife . 04601 . 020Video 4 . This video shows a typical field of view from tomographic reconstruction of the in vitro reconstitution specimen . The filaments present on the water/air interface consist of TmFtsA and TmFtsZ filaments and therefore adopt a curved geometry . This corresponds to Figure 4A . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 02010 . 7554/eLife . 04601 . 021Video 5 . This video shows a volume of the liposome whose stereo view is depicted in Figure 4A . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 02110 . 7554/eLife . 04601 . 022Video 6 . This video shows a volume representation of the liposome that is depicted in Figure 4A ( bottom middle panel , black arrowheads ) and whose stereo view is shown in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 02210 . 7554/eLife . 04601 . 023Video 7 . This video shows the two remaining liposomes that are depicted in Figure 4A . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 02310 . 7554/eLife . 04601 . 024Video 8 . This video shows the two remaining liposomes that are depicted in Figure 4A . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 02410 . 7554/eLife . 04601 . 025Video 9 . This video shows a well-pronounced constriction with spirals being very prominent on lateral sides of the leading membrane edge , which eventually might lead to abscission . Not shown in any other figure . See also Figure 5D , middle for an explanation . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 02510 . 7554/eLife . 04601 . 026Video 10 . This video runs through a surface representation of the liposome whose stereo view is depicted in Figure 4A , top , with features of interest highlighted along the way . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 026 Map quality allowed us to fit the FtsZ crystal structure manually and roughly through a spline curve and arrive at a pseudo-atomic model for an FtsZ ring ( Figure 5A ) , making it possible to judge distances and dimensions relative to the crystal structures . A more detailed view using sphere representation ( Figure 5B ) shows , again , that the filaments within a ring were not exactly equidistant ( black arrows ) but came into direct contact only at certain points . Fitting the FtsA crystal structure into the map as well revealed two closely associated filaments and showed that the outline fit of the tomographic density is extremely good , although exact orientations and locations of the subunits along the filament of the molecules can only be guessed in most places given the resolution limit . It should be noted , though , that peaks appear in many places indicating the centre positions of individual FtsZ molecules ( Figure 5C , Figure 4–source data 1 , a PyMOL session file ) . 10 . 7554/eLife . 04601 . 027Figure 5 . Visualising the FtsZ ring at the molecular level . ( A ) A semi-atomic model of the FtsZ ring constricting a liposome . 294 monomers of S . aureus FtsZ have been roughly positioned using a spline-fitting approach ( PDB 3VO8 ( Matsui et al . , 2012 ) ) . This uses the same tomography data as Figure 4A . ( B ) The ring is 90 nm in diameter ( left ) and 60-nm thick ( middle ) . It consists of at least four individual filaments ( right , atoms shown as spheres ) with varying lateral interfilament distances ( right , atoms shown as spheres , black arrows ) . ( C ) FtsZ filaments are single protofilaments , but they tend to pair in doublets . A precision manual fit of the TmFtsA polymer crystal structure ( PDB 4A2B ) ( Szwedziak et al . , 2012 ) in addition to 3VO8 FtsZ polymer crystal structure was performed in a region of very good density . The fit is excellent and dimensions and distances match well with CcFtsZ , EcFtsZ , and TmFtsAZ in vivo situations ( Figure 1A , E , 2E , G ) . ( D ) Left: in the ring-like structures ( black ) , force ( red arrows ) is perpendicular to the membrane ( blue ) , leading to constriction . Middle: during constriction , the ring develops into two helical spirals , leading to forces pushing membrane inwards , and this might explain how abscission is accomplished since membranes will presumably not fuse while the protein filaments are in between ( see Figure 4A bottom right and Video 9 for an example of this in liposomes ) . Right: the domes we observed do not deform liposomes because the force generated is almost perfectly tangential to the membrane . ( E ) Constriction force generation and filament sliding . In the discussion , three different energy sources for constriction are listed: maximising filament overlap , repeat mismatch within FtsA–FtsZ copolymers ( Figure 4—figure supplement 2 ) and filament shortening and turnover due to nucleotide hydrolysis by FtsAZ . While it is currently not obvious which of these or if a combination of the three mechanisms drives constriction , it seems clear to us that constriction , at least in the liposome reconstitution experiments , requires filaments to slide past each other as is depicted in two dimensions . Since also unmodified wild-type cells ( Figure 1 ) show closed continuous rings at division sites , we would assume the same holds true in vivo . Filament sliding can also explain the spirals on lipid monolayers ( Figure 3A ) and spirals in the dome-like structures with liposomes ( Figure 4A ) . The schematic drawn is a simplification into two dimensions , of course , in vivo and in vitro FtsZ filaments overlap in the third dimension , forming single-layered bands since each filament is anchored to the membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 04601 . 027
How do FtsZ and FtsA constrict liposomes ? Is this likely to be related to the in vivo situation ? Given that the filament architectures observed here in C . crescentus and E . coli and in constricting liposomes are so similar , we would suggest that the model we propose should be valid for both in vitro and in vivo , at least at some primordial level . FtsA forms ( partial ) filaments between the membrane and FtsZ filaments , and the filaments together encircle the constriction site while forming a single-layered small band of filaments . The entire structure is slightly helical , and shorter filaments overlap to form a continuous , closed ring . By imaging completely unmodified cells , utilising recent advances in cryo-EM and acquiring tomograms of cells parallel to the tilt axis , we concluded that the FtsZ ring in cells is most likely continuous , probably made of shorter overlapping filaments . Previous analysis of C . crescentus cells by cryo-ET also showed that the FtsZ ring consists of overlapping filaments inside the inner membrane , although not all cells showed continuous rings ( Li et al . , 2007 ) . Equally , results obtained with super resolution fluorescence microscopy techniques ( Holden et al . , 2014 ) showed punctuated fluorescence , possibly indicating non-continuous rings . We think it is important to point out that fluorescence microscopy only images the labelled species and intensity fluctuations within the ring may have arisen from using non-functional GFP fusions and/or their over-expression . Or fluctuations coming from overlapping filaments may have been over-emphasised during image analysis because of very low signal-to-noise . Unfortunately , currently , cellular tomography data are too weak to be able to trace individual filaments and their ends with confidence , so we have no direct evidence from our in vivo data for the length of individual FtsZ filaments making up continuous rings . Given a continuous ring , at least in the in vitro situation , with no GTP turnover ( or no added nucleotide ) there is no filament shortening , meaning constriction requires the filaments to slide along each other as the ring decreases in diameter and the membrane deforms . That opposing forces from the filaments and the membrane surface are at work is most evident from the fact that the rings were always perfectly round , in contrast to the rest of the liposomes ( Figure 4B , middle , Video 10 ) . In this context , the filament spirals that form shallow ‘dome’ structures in some of the liposomes and on flat monolayers ( Figures 3A and 4A , top and Figure 4—figure supplement 1 ) are most revealing . They appear to be the result of sliding and condensation and show decreasing filament curvature , but do not deform the lipid membrane . This can be explained because the constriction force , which acts in the plane of the spirals , will not exert any force on the membrane since it is tangential ( Figure 5D , right ) . In contrast , if the filaments form a ring around the volume of the liposome ( in the middle , not at the poles ) , the constriction force is perpendicular to the membrane and will lead to the membrane being pulled in ( Figure 5D , left ) . Since in this model force generation is dependent on a closed ring , the system becomes self-regulating since constriction will only commence after a complete ring has formed . If the cell is too large or not enough FtsZ is available , constriction will not begin . It is important to note that only a closed , continuous ring is required , but it may consist of a number of shorter , overlapping filaments ( as it did in the liposome reconstitutions , Figure 4 ) . What drives constriction of the closed rings and filament sliding ? We propose three possible mechanisms that may even act in concert: ( a ) maximising filament overlap via sliding , ( b ) increasing repeat mismatch , and ( c ) repeated filament shortening through nucleotide turnover . A . When the overlap between the filaments that are attracted to each other increases , more and more binding energy is produced . This has been proposed before to be theoretically sufficient for the constriction process ( Lan et al . , 2009 ) ( Figure 5E ) . The lateral spacing of 6 . 5–8 nm between filaments we report here is slightly larger than the thickness of FtsZ filaments and presumably also FtsA filaments ( Matsui et al . , 2012; Szwedziak et al . , 2012 ) . Previous in vitro work reported an interfilament distance of 5 nm using FtsZ-mts; however , this was after negative staining and dehydration and no constrictions were observed , possibly due to lateral interactions being too tight ( Milam et al . , 2012 ) . In C . crescentus cells , the lateral spacing between filaments was found to be 9 . 3 nm previously ( Li et al . , 2007 ) ( Li et al . , 2007 ) and we report here distances of ∼7 . 8 ( Figure 1A ) in C . crescentus and ∼6 . 8 nm ( Figure 1G ) in E . coli . AFM using only FtsZ , but observing spiral condensation , provided an even larger distance of 12 nm ( Mingorance et al . , 2005 ) . All of these measurements are averages with large variances . One may conclude that the filaments in the FtsZ ring interact transiently and direct contact is localised to only a few small regions within the ring at a time . This could facilitate the constriction process since the filaments have to be free to slide . It was suggested previously that instead of forming many intermolecular solid bonds , which would lead to avidity and a barrier to sliding , an attractive force over a longer distance would keep the filaments apart while interacting ( Hörger et al . , 2008 ) . This is more akin to the liquid state of matter , where many transient homotypic interactions , counteracted by thermal motion , lead to a fluid situation without absolute order but still keeping the molecules together . B . The second possible driver of constriction comes from the repeat length mismatch of FtsA and FtsZ ( Szwedziak et al . , 2012 ) . Although it is evident from our data that the curvature of individual filaments can not deform liposomes significantly from the inside , the decreasing diameter of the ring accompanied by increasing membrane curvature might enable more and more FtsA to be added , until an optimum curvature of the system has been achieved ( Figure 4—figure supplement 2 ) ; this could provide additional energy and would also explain why FtsA exists at all and FtsZ is not directly attached to the membrane . C . Why does FtsZ hydrolyse GTP then ? We speculate that when constriction starts at large diameters ( 1 µm in E . coli ) , longer GTP-induced FtsZ filaments are needed to reach reliably around the cell in order to produce overlap for the ring and 'force engagement' to start the process . However , increasing overlap , developing as the constriction progresses , might lead to a kinetic barrier of sliding through avidity and the filaments would then have to be shortened . Formally this provides a third possible driver of constriction , at least for large constriction distances . Continual depolymerisation and re-polymerisation through nucleotide turnover by FtsZ and/or FtsA , as shown in vivo by FRAP ( Stricker et al . , 2002 ) , might also ensure that the ring never reaches a highly condensed state and FtsZ monomer-sequestering inhibitors such as SulA remain able to stop the process at any time ( Chen et al . , 2012 ) . However , it has been reported that the GTP hydrolysis-deficient FtsZ ( D212G ) mutant generates prominent constrictions of tubular liposomes in vitro ( Osawa and Erickson , 2011 ) and functions in cell division in E . coli ( Bi and Lutkenhaus , 1992; Trusca et al . , 1998; Osawa and Erickson , 2006 ) . Furthermore , the use of non-hydrolysable GTP analogues did not impair the formation of condensed FtsZ structures on mica ( Hörger et al . , 2008 ) . Taken together with our result that constriction of liposomes may in principle be independent of GTP hydrolysis , these data question the alternative idea of force generation by filament bending upon GTP hydrolysis as has been suggested previously ( Lu et al . , 2000; Erickson et al . , 2010; Li et al . , 2013 ) . We suggest that nucleotide binding/hydrolysis is required solely for filament growth/shrinkage as these are essential to maintain the dynamic state of the FtsZ ring in cells . Taken together , we envisage that in vitro liposome constrictions and in vivo cell division quite possibly utilise a different set of energetic drivers ( a–c ) ; for example , GTP hydrolysis was not required in vitro but clearly plays a role in vivo . And of course , it is likely that the cell wall synthesis in the periplasm , guided by the Z-ring through the divisome , provides additional force in cells . So far , wall-less bacterial L-forms have not shed light on this interdependence of cell wall synthesis and the FtsZ ring since artificial L-forms were found to divide by blebbing , most likely a mechanism that is not actively supported by cellular machinery ( Leaver et al . , 2009 ) . FtsZ-based cell division is not functioning in these L-forms and we suggest this may be because L-forms are too large for Z-rings to close , given the amounts of FtsAZ present . The described FtsZ filament arrangement might also provide a solution to the abscission problem: how do the membranes fuse at the end of division when the protein filaments are in between ? Figure 5D , middle shows how an intermediate between the rings and the domes ( as is present in the liposome constriction shown in Figure 4A , bottom right and Video 9 ) may explain abscission , since the protein ring would normally be in the way of membrane fusion/fission at the end . The change from a flat band of filaments towards the helical spirals enables inward force to be developed on each side of the constriction , with a helical spiral on each side . The spirals observed in the dome-like structures might even be remnants of such liposome abscission events , although we have no evidence for this ( Figure 5D , right ) . It remains to be seen if FtsZ is involved in final abscission since it has recently been reported that FtsZ might leave the septum earlier ( Söderström et al . , 2014 ) . It is important to mention that membrane constriction with ESCRT-III and dynamin filaments has also been suggested to involve sliding helical filaments ( Roux et al . , 2006; Guizetti et al . , 2011 ) and similar arrangements to those depicted in Figure 5D , right were predicted for the ESCRT-III system ( Fabrikant et al . , 2009 ) . Finally , our reconstitution of cell division can easily be adapted to include other cell division proteins , such as the division site selection mechanism MinCDE , nucleoid occlusion , FtsZ cross-linkers such as ZapA , and many more middle and outer divisome components .
Plasmids used in this work are listed in Supplementary File 1A . E . coli DH5α was used for cloning . Caulobacter crescentus NA1000/CB15N and E . coli B/r H266 ( Trueba and Woldringh , 1980 ) were used for cellular electron cryotomography . Caulobacter crescentus was grown overnight in PYE medium at 30°C . The overnight culture was used to inoculate 50 ml of M2G medium . The culture was grown at 30°C until the OD reached 0 . 5 . 11 µl of this culture was mixed with 1 µl of protein-A conjugated to 10 nm gold beads ( CMC , Leiden ) and applied to freshly glow-discharged 300 mesh Cu/Rh Quantifoil ( 3 . 5/1 ) grids . The grids were plunge-frozen into liquid ethane using a FEI Vitrobot ( Mark IV ) and stored in liquid nitrogen . E . coli cells ( some containing relevant plasmids , for FtsZ mutant co-expression with endogenous wild-type FtsZ , Supplementary file 1 , Tables A and B ) were grown at 30°C in M9 minimal media supplemented with 0 . 4% glycerol until log-phase . Cells were then diluted into fresh M9 media with 0 . 02% arabinose ( where needed , final concentration ) and grown for 1–2 hr for FtsZ mutant protein expression . Thermotoga maritima FtsZ ( TmFtsZ ) and FtsA ( TmFtsA ) proteins were purified and 2D monolayers were prepared as described previously ( Szwedziak et al . , 2012 ) , taking extra care and verifying by ESMS that the C-terminal tails of both proteins were intact after purification as they are prone to proteolytic cleavage . This was not obvious from gels , sometimes . 20 μl of E . coli total lipid extract ( Avanti Polar Lipids , Alabaster , AL ) chloroform solution at 10 mg/ml was dried in a glass vial ( Wheaton , Millville , NJ ) under a stream of nitrogen gas and left overnight under vacuum to remove traces of the solvent . The resulting thin lipid film was hydrated with 200 μl of TEN1007 . 5 buffer ( 50 mM Tris/HCl , 100 mM NaCl , 1 mM EDTA , 1 mM NaN3 , pH 7 . 5 ) , containing either TmFtsA at 20 µM or TmFtsZ at 60 µM or both proteins . After 10 min of incubation at room temperature , the solutions were sonicated for 1 min in a water bath sonicator and then 2 . 5 µl of sample was plunge-frozen onto Quantifoil R2/2 holey carbon grids ( Quantifoil , Germany ) using an FEI Vitrobot ( FEI Hillsboro , OR ) . Samples were stored in liquid nitrogen . 50 μl of E . coli total lipid extract chloroform solution at 10 mg/ml was dried in a glass vial under a stream of nitrogen gas and left overnight under vacuum to remove traces of the solvent . The resulting thin lipid film was hydrated with 50 μl of TEN1007 . 5 plus 20 mM CHAPS ( Anatrace , Maumee , Ohio ) and shaken vigorously at 800 rpm using a benchtop Eppendorf shaker for 2 hr . The lipid–detergent solution was sonicated for 1 min in a water bath sonicator . Subsequently , 50 μl of TmFtsZ ( 30 µM ) and TmFtsA ( 10 µM ) solutions supplemented with 0 . 5 mM MgGTP or MgGMPCPP ( Jena Bioscience , Germany ) or no nucleotide was added and left for 30 min at room temperature . Next , the mixture was gradually diluted within 10 to 20 min to 600 μl with TEN1007 . 5 or TEN1007 . 5 plus nucleotides ( both without detergent ) to trigger spontaneous liposome formation . 2 . 5 µl of the solution was mixed with 0 . 2 µl 5 nm IgG immunogold conjugate ( TAAB , UK ) and plunge-frozen onto Quantifoil R2/2 holey carbon grid using an FEI Vitrobot . 2D electron cryomicroscopy ( cryo-TEM ) images were taken on an FEI TECNAI Spirit TEM operating at 120 kV with a 2k × 2k CCD camera at a magnification of 42 k , corresponding to a pixel size of 0 . 25 nm . For electron cryotomography , samples were imaged using an FEI Polara or FEI Titan Krios TEM operating at 300 kV , equipped with a Gatan imaging filter set at zero-loss peak with a slit-width of 20 eV . A Gatan Ultrascan 4000 CCD camera binned to 2k × 2k or a 4k × 4k K2 Summit direct electron detector was used for data acquisition with SerialEM software ( Mastronarde , 2005 ) . Cells or in vitro reconstituted systems were imaged at a magnification of 41 k , corresponding to a pixel size of 5 . 8 Å ( with US4000 ) , or at a magnification of 26 k , corresponding to a pixel size of 4 . 5 Å ( for K2 ) at the specimen level . Specimens were tilted from approximately −60° to +60° ( ±65° for C . crescentus cells ) with a 1° increment . The defocus was set between 8 and 10 µm , and the total dose for each tilt series was around 120 e/Å2 for in vitro reconstitution samples and 150–200 e/Å2 for cells . Tomographic reconstructions from tilt series were calculated using RAPTOR ( Amat et al . , 2008 ) and the IMOD tomography reconstruction package , followed by SIRT reconstruction with the PRIISM software or the TOMO3D package ( Chen et al . , 1996; Kremer et al . , 1996; Agulleiro and Fernandez , 2011 ) . Measurements of distances between structures were carried out within IMOD . Videos showing liposomes were prepared with PyMOL ( DeLano , 2002 ) . E . coli B/r H266 cells with plasmid pMZ124 were grown in LB medium at 30°C . At an OD600 of 0 . 2 , FtsZ ( D212A ) and FtsA expression was induced by adding 0 . 02% arabinose . After 2 hr , cell membranes were stained with FM4-64 membrane dye and cells were mounted on an agarose pad and visualised using a Nikon N-SIM microscope in the 2D-SIM mode . FtsZ expression level in cells used for electron cryotomography experiments were examined with Western blots using rabbit anti-FtsZ primary antibodies ( Agrisera , Sweden ) and donkey anti-rabbit IgG conjugated with horseradish peroxidase ( GE Healthcare ) and detected with ECL blotting reagent . The Caulobacter crescentus tomogram shown in Figure 1A has been deposited in the EM databank with accession number EMD-2814 . The edited tomogram of TmFtsAZ constricting a liposome , as shown in Figures 4A and 5A and Video 10 , has been deposited with accession number EMD-2815 . | Cell division is the process by which new cells are made . It is therefore vital for the growth and development , and the regeneration and repair of damaged tissues . When bacterial and animal cells divide , they must constrict their membrane inwards to split a single cell into two . In most bacteria , this constriction is guided by a ring-like structure that contains filaments of a protein called FtsZ . During cell division , this structure forms around the inside edge of the cell and when it contracts , it pulls the membrane inwards and causes the cell to constrict and eventually divide . In recent years , this arrangement of FtsZ filaments has been intensively investigated , giving rise to various theories about how it is made and how it works: for example , some recent studies suggested that FtsZ does not form a continuous ring . Nevertheless , many details about the cell division process remain unknown . Szwedziak , Wang et al . have now investigated this protein ring in two species of bacteria by turning to advanced forms of microscopy to closely observe its structure and how it works . This included mapping the ring in three dimensions . Contrary to earlier reports that the FtsZ ring is discontinuous , in both a bacterium called Caulobacter crescentus and another called Escherichia coli , the ring forms a continuous shape made up of overlapping filaments . Szwedziak , Wang et al . then increased the levels of two of the ring's main components: the FtsZ protein that forms the filaments and a protein that anchors these filaments to the cell membrane . This caused the modified cells to constrict and divide at extra sites , which resulted in the formation of abnormally small cells . These findings suggest that these two ring components by themselves are able to generate both the structures and force required for cell constriction . This is supported by the fact that when they were introduced into artificial cell-like structures , these proteins spontaneously self-organised into rings and triggered constriction where they formed . Szwedziak , Wang et al . propose that constriction only starts once the FtsZ protein forms a closed ring and that the ring's overlapping filaments slide along each other to further decrease its diameter and constrict the cell . The degree of filament overlap likely also increases with constriction , requiring filaments to be shortened to maintain sliding . This shortening , along with sliding , could provide a mechanism by which to drive the constriction process . This work will be followed by even more detailed studies in order to understand the process of bacterial cell division at the atomic scale and how the cell's wall is reshaped during the process . In the long run , intricate knowledge of how a bacterial cell divides might enable the design of new classes of antibiotics targeting the molecular machinery involved . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2014 | Architecture of the ring formed by the tubulin homologue FtsZ in bacterial cell division |
Neurons develop highly stereotyped receptive fields by coordinated growth of their dendrites . Although cell surface cues play a major role in this process , few dendrite specific signals have been identified to date . We conducted an in vivo RNAi screen in Drosophila class IV dendritic arborization ( C4da ) neurons and identified the conserved Ret receptor , known to play a role in axon guidance , as an important regulator of dendrite development . The loss of Ret results in severe dendrite defects due to loss of extracellular matrix adhesion , thus impairing growth within a 2D plane . We provide evidence that Ret interacts with integrins to regulate dendrite adhesion via rac1 . In addition , Ret is required for dendrite stability and normal F-actin distribution suggesting it has an essential role in dendrite maintenance . We propose novel functions for Ret as a regulator in dendrite patterning and adhesion distinct from its role in axon guidance .
Accurate functional connectivity and sensory perception require proper development of the neuronal dendritic field , which ultimately determines the ( sensory ) input a specific neuron can receive and detect . Thus , coordinated dendrite growth and patterning is important for establishing the often complex , but highly stereotyped organization of receptive fields . Two of the organizing principles in dendrite development are self-avoidance and tiling ( Grueber and Sagasti , 2010; Jan and Jan , 2010; Zipursky and Sanes , 2010 ) . While self-avoidance describes the phenomenon of recognition and repulsion of isoneuronal dendritic branches , tiling refers to the complete yet non-redundant coverage of a receptive field by neighboring neurons of the same type . Both phenomena have been described in different systems across species including the mouse , zebrafish , medicinal leech , Caenorhabditis elegans , and Drosophila melanogaster . Dendritic patterning by self-avoidance , tiling , and other mechanisms is thought to be mediated by cell surface receptors and cell adhesion molecules ( CAMs ) , which play a pivotal role in integrating environmental and cellular cues into appropriate growth and adhesion responses . Many such receptors , prominently Robo ( Spitzweck et al . , 2010 ) and Ephrin receptors ( Egea and Klein , 2007 ) , have well understood roles in axon guidance . Although some of these axonal cues including Robo/Slit play a role in dendrite development as well ( Dimitrova et al . , 2008; Gibson et al . , 2014 ) , dendritic surface receptors and their functions are not fully characterized to date . Recent efforts have yielded some progress in this area . Down's syndrome cell adhesion molecule ( Dscam ) has been shown to regulate dendrite self-avoidance in Drosophila ( Hughes et al . , 2007; Matthews et al . , 2007; Soba et al . , 2007 ) and mouse ( Fuerst et al . , 2008 , 2009 ) . More recently , studies on protocadherins have revealed that they play an important role in dendrite self-avoidance in mammals ( Lefebvre et al . , 2012 ) . In C . elegans , sax-7/L1-CAM and menorin ( mnr-1 ) form a defined pattern in the surrounding hypodermal tissue to guide PVD sensory neuron dendrite growth via the neuronal receptor dma-1 ( Dong et al . , 2013; Salzberg et al . , 2013 ) . However , given the complexity and stereotypy of dendritic arbors within individual neuronal subtypes , it is important to search for additional signals for directing dendrite growth . The Drosophila peripheral nervous system ( PNS ) has served as an excellent model which has helped to elucidate several molecular mechanisms regulating dendrite development ( Grueber and Sagasti , 2010 ) . The larval PNS contains segmentally repeated dendritic arborization ( da ) neurons which have been classified as class I–IV according to their increasing dendritic complexity ( Grueber et al . , 2002 ) . All da neuron classes feature highly stereotyped sensory dendrite projections . Moreover , all da neurons exhibit self-avoidance behavior allowing them to develop their individual receptive fields without overlap . It has been demonstrated that all da neuron classes require Dscam for dendrite self-avoidance ( Hughes et al . , 2007; Matthews et al . , 2007; Soba et al . , 2007 ) . In addition , the atypical cadherin flamingo ( Matsubara et al . , 2011 ) and immunoglobulin super family ( IgSF ) member turtle ( Long et al . , 2009 ) might play a more restricted role in C4da neuron self-avoidance . Netrin and its receptor frazzled have also been shown to act in parallel to Dscam in class III da neurons ensuring their proper dendritic field size and location by providing an attractive growth cue which is counterbalanced by self-avoidance ( Matthews and Grueber , 2011 ) . For tiling , no surface receptor has been identified to date . However , the conserved hippo and tricornered kinases , and more recently the torc2 complex , have been implicated in C4da neuron tiling , as the loss of function of these genes results in iso- and heteroneuronal crossing of dendrites ( Emoto et al . , 2004 , 2006; Koike-Kumagai et al . , 2009 ) . Recent work has further shown that dendrite substrate adhesion plays an essential role in patterning . Da neuron dendrites are normally confined to a 2D space through interaction with the epithelial cell layer and the extracellular matrix ( ECM ) on the basal side of the epidermis ( Yamamoto et al . , 2006; Han et al . , 2012; Kim et al . , 2012 ) . 2D growth of da neuron dendrites requires integrins , as loss of the α-integrin mew ( multiple edomatous wing ) or ß-integrin mys ( myospheroid ) results in dendrites being freed from the 2D confinement due to detachment from the ECM . Thus , they can avoid dendrites by growing into the epidermis leading to 3D crossing of iso- and hetero-neuronal branches ( Han et al . , 2012; Kim et al . , 2012 ) . Integrins are therefore essential to ensure repulsion-mediated self-avoidance and tiling mechanisms , which restrict growth of dendrites competing for the same territory ( Han et al . , 2012; Kim et al . , 2012 ) . How integrins are recruited to dendrite adhesion sites and whether they cooperate with other cell surface receptors is unknown . To identify novel receptors required for generating complex , stereotypical dendritic fields , we performed an in vivo RNAi screen for cell surface molecules in C4da neurons . We identified the Drosophila homolog of Ret ( rearranged during transfection ) as a patterning receptor of C4da dendrites . Loss of Ret function in C4da neurons severely affects dendrite coverage , dynamics , growth , and adhesion . In particular , dendrite stability and 2D growth are impaired resulting in reduced dendritic field coverage and abnormal 3D dendrite crossing , respectively . These defects can be completely rescued by Ret expression in C4da neurons . We further show that Ret interaction with integrins is needed to mediate C4da dendrite-ECM adhesion , but not dendrite growth . Our data suggest that Ret together with integrins acts through the small GTPase rac1 , which is required for dendrite adhesion and 2D growth of C4da neuron dendrites as well . We thus describe a novel role for the Ret receptor in dendrite development and adhesion by direct receptor crosstalk with integrins and its downstream signals .
To identify cell surface receptors mediating dendrite development of C4da neurons we used an in vivo RNAi screening approach . We focused on functionally defined classes of proteins containing conserved extracellular domains , including IgSFs and receptor tyrosine kinases ( RTKs ) . To this end we expressed available RNAi transgenes specifically in the embryonic/larval PNS ( 21-7-Gal4 ) together with a C4da neuron specific reporter line ( ppk-CD4-tdTomato , Han et al . , 2012 ) . We also utilized a UAS-Dcr2 transgene to enhance RNAi efficiency ( Dietzl et al . , 2007 ) . Using this approach , we screened approximately 400 RNAi lines targeting IgSFs and RTKs and found that knockdown of Ret with two independent lines led to strong dendrite defects in C4da neurons ( Figure 1A ) . Knockdown of Ret resulted in abnormal C4da dendrite patterning with crossing of dendritic branches and incomplete coverage of their receptive field . We did not observe defects in other classes of da neurons ( Figure 1—figure supplement 1 and data not shown ) suggesting that Ret plays a specific role in C4da neuron dendrite morphogenesis . 10 . 7554/eLife . 05491 . 003Figure 1 . In vivo RNAi knockdown of Ret causes C4da neuron dendrite pattering defects . ( A ) RetRNAi transgenes together with Dcr2 were driven by Gal421–7 and C4da neuron morphology was visualized with a specific fluorescent reporter ( ppk-CD4-tdTomato ) . Confocal live images of control animals ( ctl ) show wildtype dendrite morphology , while expression of either of two independent Ret-RNAi transgenes led to severely disorganized dendrites with incomplete receptive field coverage . ( B ) A Gal4 insertion in the Ret genomic locus drives CD8-GFP expression in C4da neurons indicating the presence of Ret . Scale bar 50 μm . ( C–E ) Immunohistochemical analysis of Ret expression in third instar larvae of wildtype ( C and D ) and Ret deficient animals ( E ) . Overlays with GFP expressing C4da neurons ( ppk-Gal4 > CD4-tdGFP ) show that specific anti-Ret signal could be detected throughout the C4da dendritic arbor ( C ) . Resliced portions ( in Z direction ) of primary ( C′ ) and terminal ( C′′ ) C4da dendrites show that Ret strongly labels the basal side of dendrites facing the ECM as shown in the schematic drawing . ( D ) Ret was also present in distal terminal dendrites of the dorsal field of C4da neurons ( D′ and D′′ , scale bar 20 μm , 10 μm for insets ) . Residual Ret signal was detected in the C4da neuron soma in Ret mutant animals but not in dendrites ( E inset ) . Arrows indicate non-specific antibody signal present in wildtype and Ret mutant samples . Scale bar 20 μm . ( F ) Quantitative analysis of Ret immunoreactivity in C4da somata of wildtype and Ret mutant samples showing the signal over background ( ΔF/F , mean ± SD , n = 5 , p < 0 . 001 , Student's two-tailed t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 00310 . 7554/eLife . 05491 . 004Figure 1—figure supplement 1 . Class I da neurons do not display obvious morphological defects in Ret mutant animals . RetC168 homozygous third instar larvae display no obvious phenotype in class I da neuron morphology visualized by 2-21-Gal4 > CD8-GFP . Scale bar 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 00410 . 7554/eLife . 05491 . 005Figure 1—figure supplement 2 . phospho-Ret levels are reduced in Ret mutant C4da neurons . ( A and B ) Immunohistochemical analysis of phospho-Ret in third instar larvae of wildtype ( A ) and Ret mutant animals ( B ) . Specific anti-phospho-Ret signal could be detected in C4da neuron somata of control but not Ret mutant animals ( A and B inset ) . Arrows indicate non-specific antibody signal at epithelial junctions . Scale bar 25 μm . ( B ) Quantitative analysis of phospho-Ret immunoreactivity in C4da somata of wildtype and Ret mutant samples showing signal over background ( ΔF/F , mean ± SD , n = 3 , p < 0 . 05 , Student's two-tailed t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 005 Drosophila Ret encodes a highly conserved receptor tyrosine kinase ( RTK ) with neuronal expression ( Sugaya et al . , 1994; Hahn and Bishop , 2001; Kallijärvi et al . , 2012 ) . In order to validate our RNAi results , we tested the expression of Ret in C4da neurons . A Gal4 enhancer trap line inserted within the 5′-region of the Ret gene showed C4da neuron expression when driving a GFP reporter ( UAS-CD8-GFP , Figure 1B ) . We next performed immunohistochemical analysis of larval filet preparations using Ret specific antibodies . We first used a commercial phospho-Ret antibody which showed a mostly somatic granular signal only in C4da neurons ( Figure 1—figure supplement 2A ) . We tested its specificity with a mutant allele containing a P-element insertion within the 3′UTR region of the Ret gene ( RetC168 ) , which in combination with a deficiency covering the Ret locus ( Df ( Bsc312 ) ) led to a strong reduction of the Ret immunoreactivity in C4da neurons ( Figure 1—figure supplement 2A–C ) . To validate and analyze Ret expression in C4da neurons in more detail , we raised a Ret specific antibody against its intracellular domain ( see ‘Materials and methods’ ) . Interestingly , Ret was detected throughout the dendritic arbor showing almost continuous staining along major dendrites ( Figure 1C , C′ ) and a more granular pattern in terminal dendrites ( Figure 1C , C′′ ) . We detected granular Ret signal even in distal terminal dendrites suggesting that is localizing throughout the C4da dendritic arbor ( Figure 1D ) . In addition , the strongest Ret puncta consistently localized to the basal side of dendrites facing the ECM ( Figure 1C′ , C′′ ) . In Ret mutant animals , only residual immunoreactivity could be detected in C4da neuron somata but not in dendrites ( Figure 1E , E′ , F ) suggesting that the allelic combination used is hypomorphic . Together , these results reveal expression and subtype specific functions of Ret in C4da neurons . We next wanted to corroborate the Ret-RNAi induced dendrite phenotype of C4da neurons using the RetC168 allele . Compared to our RNAi results , we found very similar dendrite defects of C4da neurons in RetC168 mutant animals ( Figure 2A , B ) . These defects were strongly enhanced when we combined RetC168 with a chromosomal deficiency line ( Df ( 2L ) Bsc312 , Figure 2C ) , indicating that RetC168 is indeed a hypomorphic allele . Strikingly , Ret mutant C4da neurons displayed incomplete coverage of the receptive field , with dendritic terminals exhibiting patchy distribution , reflecting abnormalities in both shape and growth directionality . In addition , we observed severe isoneuronal dendrite crossing , resulting in extensive overlap of dendritic branches not normally observed in controls ( Figure 2B , C , arrows ) . This phenotype is reminiscent of self-avoidance defects described for Dscam ( Matthews et al . , 2007; Soba et al . , 2007; Hattori et al . , 2009 ) or tiling mutants like trc , hpo , and fry ( Emoto et al . , 2004 , 2006 ) . To understand the nature of the dendritic crossing phenotype in Ret mutant larvae , we performed high resolution confocal live imaging of C4da neurons in vivo . Detailed 3D analysis of dendrite crossing points revealed that virtually all overlapping dendrites were not within the same focal plane , but grew in different planes without directly contacting one another ( Figure 2D , D′ , D′′ ) . Despite the severe patterning and field coverage defects , Ret mutant C4da neurons did not show a significant reduction in total dendrite length ( Figure 2E ) . They did however display excessive crossings of isoneuronal dendrites outside their normal 2D growth plane ( Figure 2F ) and significantly reduced dendritic field coverage ( see Figure 5B , F ) . Taken together , our results show that Ret is required for C4da neuron dendrite patterning and field coverage , and Ret loss of function causes out of plane dendritic crossing due to abnormal 3D expansion of dendrites . 10 . 7554/eLife . 05491 . 006Figure 2 . Loss of Ret function impairs C4da neuron dendrite patterning and results in out of plane crossing of isoneuronal branches . ( A–C ) Confocal live images of C4da neurons in ( A ) wildtype , ( B ) RetC168 homozygous , and ( C ) RetC168/Df ( Bsc312 ) third instar larvae . Loss of Ret function causes abnormal dendrite patterning featuring reduced receptive field coverage and overlapping dendrites ( indicated by arrows ) . Note that the phenotype is more severe in RetC168/Df ( Bsc312 ) than in RetC168 mutant C4da neurons . Scale bar: 50 μm . ( D ) 3D reconstruction of a Ret deficient C4da neuron with magnified 3D views of regions with dendrites crossing in different growth planes ( D′ and D′′ ) . Scale bar: 25 μm . ( E ) Total dendrite length of Ret mutant C4da neurons was not significantly changed compared to wildtype . ( F ) Out of plane crossing of C4da neuron dendrites in Ret mutants was however highly elevated ( mean ± SD , n = 8; p < 0 . 001 , Student's two-tailed t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 006 Due to the out of plane crossing observed in Ret mutant C4da neurons , we suspected that dendrites were partially detached from the ECM and embedded within the epithelial cell layer . A similar phenotype has recently been described for the integrins mew and mys , which are required for dendrite-ECM adhesion in da neurons ( Han et al . , 2012; Kim et al . , 2012 ) . In order to visualize C4da neuron dendrites and the ECM , we took advantage of GFP protein traps for the ECM components viking ( vkg-GFP ) and trol ( trol-GFP ) ( Han et al . , 2012 ) . We performed two-color high resolution confocal imaging of C4da neuron dendrites together with the GFP-labeled ECM in vivo . In wildtype , the majority of dendrites were tightly interacting with the ECM as previously reported ( Figure 3A , and Han et al . , 2012 ) . In Ret mutant neurons however , large stretches of dendrites , in particular terminal arbors , were detached from the ECM and enclosed by the epidermis ( Figure 3B ) . While occasional detachment and out of plane crossing of dendrites could be observed in controls ( Figure 3A′ ) , the majority of branches were in tight proximity to the ECM ( Figure 3A′′ ) . Ret mutant C4da neurons however displayed frequent detachment and out of plane crossing of dendrites ( Figure 3B′ , B′′ ) resulting in approx . 11% ECM-detached dendrites within the dorsal field ( Figure 3C ) . Our results show that Ret is required for robust dendrite-ECM interaction in C4da neurons . 10 . 7554/eLife . 05491 . 009Figure 3 . Dendrite-ECM detachment due to Ret loss of function is genetically linked to integrins . C4da neuron dendrites along the dorsal midline ( ppk-CD4-tdTomato ) and the extracellular matrix ( trol-GFP or vkg-GFP ) were co-visualized in third instar larvae . High resolution two-color confocal z-stacks were analyzed for dendrite-ECM interaction and detached dendrite segments are indicated in magenta as indicated schematically . ( A ) In wildtype animals , very few dendrite segments were not in contact with the ECM . Magnified regions of ( A ) and cross-sections illustrate dendrite-ECM proximity with few dendrite segments not contacting the ECM ( A′ , A′′ , ECM in green , dendrites in magenta , see schematic for color code ) . ( B ) A strong increase of detached dendrites could be observed in Ret mutant animals , highlighted in the magnified sections displaying severe displacement of dendrites from the ECM ( B′ and B′′ ) . Scale bar: 25 μm . ( C ) Quantitative analysis of dendrite-ECM interaction in wildtype and Ret mutant C4da neurons revealed a strong increase in dendrite detachment in Ret deficient animals ( mean ± SD , n = 5 , p < 0 . 001 , Student's two-tailed t-test ) . ( D–E ) Genetic interaction analysis of dendrite-ECM adhesion in ( D ) RetC168 , ( E ) mewM6 , ( F ) mys1 heterozygous and ( G ) mewM6/RetC168 ( H ) mys1/Ret C168 trans-heterozygous third instar larvae . The combination of either integrin mutant with the RetC168 allele showed highly increased loss of dendrite-ECM interaction ( G and H ) compared to wildtype or the heterozygous alleles alone . Scale bar: 25 μm . ( I ) Quantitative analysis of dendrite detachment for the individual genotypes as indicated . mewM6/RetC168 and mys1/RetC168 trans-heterozygotes showed significantly impaired dendrite-ECM adhesion ( mean ± SD , p < 0 . 05 , n = 4 , Mann–Whitney U-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 009 To investigate a possible mechanistic link between dendrite-ECM adhesion defects in Ret and integrin mutant C4da neurons , we next conducted genetic interaction experiments . While animals heterozygous for Ret or integrin mutations showed no significant dendrite detachment compared to wildtype , trans-heterozygous combinations of Ret and mys or mew integrin alleles resulted in a significant increase in dendrite crossing defects ( not shown ) and a substantial loss of dendrite-ECM interaction ( Figure 3D–H ) . Similarly to Ret mutant C4da neurons , about 10% of dendrites were detached from the ECM in RetC168/mewM6 and Ret C168/mys1 trans-heterozygous animals ( Figure 3I ) . These results suggest that Ret and integrins act in the same genetic pathway to promote dendrite–matrix interactions . To further assess the link between Ret and integrins , we examined the cellular localization and interaction of Ret with Mys and Mew . We first performed immunohistochemical analysis of larval filet preparations co-expressing Ret together with Mys and Mew in C4da neurons , as integrins endogenous to C4da neurons are difficult to detect due to strong expression in the surrounding epithelial cells ( Han et al . , 2012 ) . To this end , we used a mCherry-tagged Ret transgene which showed a distribution in dendrites similar to endogenous Ret and did not cause obvious phenotypes ( Figure 4—figure supplement 1A ) . We found partial colocalization of Ret and Mew or Mys in punctate structures in C4da neuron dendrites ( Figure 4—figure supplements 2 , 3 ) . Colocalized punctae of Ret and integrins could readily be detected in both primary and high order dendritic branches of C4da neurons ( Figure 4—figure supplements 2A′ , A′′ , 3A′ , A′′ ) . Consistently , transfected S2 cells also displayed colocalization of Ret and Mys , particularly in filopodia-like structures close to the cell surface ( Figure 4—figure supplement 1B ) suggesting that Ret and integrins may form a molecular complex . We therefore performed co-immunoprecipitation experiments of Ret and Mys/Mew in S2 cells . Indeed , Ret showed strong and specific interaction with Mys , and additional co-expression of Mew robustly increased the Ret-integrin interaction indicating that Ret preferentially interacts with the mature integrin complex ( Figure 4—figure supplement 1C ) . To investigate if Ret can also localize to the surface of C4da dendrites and possibly interact with adhesion mediating integrin complexes , we used a tagged Ret transgene carrying an N-terminal pHluorin in addition to the C-terminal mCherry-tag . We performed cell surface immunostainings of Ret using an anti-GFP antibody recognizing the pHluorin-tag together with Mew immunostaining to assess its localization and co-distribution . Ret itself displayed pronounced surface labeling along the entire dendritic tree , strongly indicating that Ret is present at the cell surface of C4da neurons ( Figure 4A , B ) . Overall , the pHluorin signal showed surface Ret labeling throughout the dendritic arbor , while the intracellular mCherry-tag revealed additional punctate structures likely reflecting intracellular vesicles ( Figure 4A ) . Strikingly , Ret showed significant colocalization with Mew in dendrites ( Figure 4B ) . Intensity plots of Ret and Mew signals display a high degree of co-distribution in low and high order dendrites ( Figure 4B′ , B′′ ) . In particular , many of the high intensity peaks of Mew showed concomitant Ret signal peaks with both tags indicating that Mew and Ret can be indeed colocalized in dendrites and at the surface of C4da neurons . Quantitative colocalization analysis of Ret and integrin signals in C4da neuron somata and dendrites revealed a strong positive correlation based on Pearson coefficient analysis for overexpressed Ret/Mew and Ret/Mys ( Figure 4C ) . 10 . 7554/eLife . 05491 . 010Figure 4 . Ret localizes to the dendrite surface and co-localizes with integrins . ( A ) Cell surface immunostaining of pHluorin-Ret-mCherry expressed in C4da neurons using ppk-Gal4 . Surface exposed Ret labeled by pHluorin ( surface-stained with anti-GFP antibody ) showed even labeling of the entire dendritic tree , while the mCherry signal displayed a more granular distribution of cellular Ret . Scale bar: 30 μm . ( B ) Cell surface immunostaining of pHluorin-Ret-mCherry co-expressing mys and mew in C4da neurons using ppk-Gal4 . Cell surface and cellular Ret were partially co-distributed with Mew in dendrites , as evident from fluorescence intensity plots along ( B′ ) low order and ( B′′ ) terminal dendrites . Scale bar 20 μm . ( C ) Quantitative colocalization analysis of Ret and integrins coexpressed in C4da neurons ( ppk-Gal4 , CD8-GFP > Ret-mCherry , mys , mew ) . Pearson coefficients were calculated for C4da neuron soma and dendrite regions showing a positive correlation of Ret and integrin signals ( see ‘Materials and methods’ for details , mean ± SD , n = 5 per genotype ) . ( D and E ) Colocalization analysis of endogenous Ret and integrins overexpressed in C4da neurons ( ppk-Gal4 , CD4-tdGFP > mys , mew ) . Endogenous Ret signal was colocalized with ( D ) anti-mys or ( E ) anti-mew immunoreactivity in C4da neurons and colocalized pixels visualized in false color representations ( coloc ) . ( D′ and E′ ) Stretches of terminal dendrites resliced in Z direction showing partial colocalization of ( D′ ) Ret and mys or ( E′ ) Ret and mew along the basal dendrite facing the ECM as indicated by arrows ( see schematic drawing ) . Line intensity plots of the same dendrite portion show signal distribution of endogenous Ret and integrins together with the colocalized signals ( coloc ) and the CD4-tdGFP membrane marker . ( D′′ and E′′ ) Line intensity plots for a primary dendrite portion ( indicated in D and E ) . ( F ) Quantitative colocalization analysis of endogenous Ret and integrins expressed in C4da neurons ( ppk-Gal4 , CD4-tdGFP > mys , mew ) . Pearson coefficients calculated for C4da neuron soma and dendrites showing a positive correlation of endogenous Ret and integrin signals ( see ‘Materials and methods’ for details , mean ± SD , n = 5 per genotype ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 01010 . 7554/eLife . 05491 . 011Figure 4—figure supplement 1 . Colocalization and biochemical interaction of Ret and integrins in S2 cells . ( A ) Analysis in third instar larval fillet preparations expressing Ret-mCherry ( anti-DsRed immunostaining ) in class IV da neurons with ppk-Gal4 > CD8-GFP ( anti-GFP ) . Ret was fairly evenly distributed in low and high order dendrites and the axon ( yellow arrow ) . Scale bar 25 μm . ( B ) S2 cells were co-transfected with mys and Ret and imunostained with specific anti-mys ( green ) and anti-Ret antibodies ( magenta ) . Mys and Ret colocalization in S2 cells is particularly strong in filopodial tips ( arrows ) . Scale bar 1 μm . ( C ) S2 cells were transfected with V5-tagged Ret and flag-tagged mys with or without mew co-transfection . Mys-flag was immunoprecipitated and Ret interaction was probed by Western blotting with an anti-V5 antibody . The asterisks indicate secondary antibody cross-reactivity with the IP antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 01110 . 7554/eLife . 05491 . 012Figure 4—figure supplement 2 . Overexpressed Ret and mys colocalization in C4da neurons . ( A ) Immunohistochemical analysis of C4da neurons co-expressing Ret-mCherry with Mys/Mew using ppk-Gal4 > CD8-GFP . Ret-mCherry signal was colocalized with anti-Mys immunoreactivity in C4da neurons and colocalized pixels visualized in false color representations ( coloc ) . Ret and Mys showed colocalized punctae in dendritic arbors ( indicated by arrows ) . Scale bar 20 μm . ( A′ ) High magnification view and cross-sections ( indicated by yellow rectangle , Z cutting plane indicated by dashed lines ) shows 3D colocalization of Ret and Mys in dendrites . ( A′′ ) Line intensity plot along a dendrite portion ( indicated by dashed yellow line in A′ ) shows the signal covariance of Ret and Mys . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 01210 . 7554/eLife . 05491 . 013Figure 4—figure supplement 3 . Overexpressed Ret and mew colocalization in C4da neurons . ( A ) Immunohistochemical analysis of C4da neurons co-expressing Ret-mCherry with integrins Mys/Mew using ppk-Gal4 > CD8-GFP . Ret and Mew colocalized pixels were visualized in false color representations ( coloc ) and colocalized punctae are present in dendritic arbors ( indicated by arrows ) . Scale bar 20 μm . ( A′ ) High magnification view and cross-sections ( indicated by yellow rectangle , Z cutting plane indicated by dashed lines ) shows 3D colocalization of Ret and Mew in dendrites . ( A′′ ) Line plots of signal intensities along a dendrite portion ( indicated by dashed yellow line in A′ ) show covariance of Ret and Mew localization . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 013 Finally , we assessed colocalization of endogenous Ret with overexpressed integrins . As for Ret overexpression , endogenous Ret displayed a significant degree of co-distribution with Mys and Mew in C4da neurons ( Figure 4D , E ) . Besides pronounced colocalization in the C4da soma , patches of Ret/Mys and Ret/Mew punctae were detected throughout the dendritic arbor in high ( Figure 4D′ , E′ ) and low order branches ( Figure 4D′′ , E′′ ) . Interestingly , the majority of colocalized signals were found facing the basal dendritic surface which contacts the ECM ( Figure 4D′ , E′ , arrows and Video 1 ) suggesting that Ret and integrins are potentially interacting at adhesion sites . Overall , quantitative colocalization analysis of endogenous Ret and integrins showed positively correlated Pearson coefficients ( Figure 4F ) suggesting that a subset of Ret and integrins is associated in C4da neurons . 10 . 7554/eLife . 05491 . 014Video 1 . Colocalization of endogenous Ret and the integrin Mew in C4da neurons . The video shows a 3-dimensional flight view of Ret-Mew colocalization along the C4da dendritic arbor . Shown are CD4-tdGFP labeling the C4da neuron ( green ) , anti-Ret ( magenta ) , and anti-Mew ( cyan ) together with the colocalized signal ( grayscale ) between Ret and Mew . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 014 Our genetic and immunohistochemical data therefore support the molecular interaction between Ret and integrins in C4da dendrites . These findings further suggest that Ret and integrins are forming a common pathway required for normal dendrite growth and ECM adhesion . Although the Ret-RNAi results and its expression pattern ( see Figure 1 ) indicated a cell-autonomous function of Ret , we wanted to test its sufficiency for supporting normal C4da neuron dendrite morphology . Due to the close proximity of the Ret gene to the chromosomal centromere it was not possible to conduct MARCM analysis ( mosaic analysis with a repressible cellular marker [Lee and Luo , 1999] ) . Instead , we performed a rescue experiment by specifically expressing a mCherry-tagged Ret transgene in C4da neurons in the Ret mutant background . Indeed , driving expression of Ret in C4da neurons was sufficient to completely rescue the dendrite growth and crossing phenotype ( Figure 5B , B′ and Figure 5C , C′ ) demonstrating that Ret is cell-autonomously required for C4da dendrite patterning . 10 . 7554/eLife . 05491 . 007Figure 5 . Cell-autonomous rescue of Ret mutant dendrite defects by C4da-specific expression of Ret or integrins . ( A–D ) Maximum projections of C4da neurons visualized with ppk-Gal4 > CD4-tdGFP are shown for ( A ) wildtype and ( B–D ) Ret mutant animals . C4da neuron specific expression of ( C ) UAS-Ret-mCherry or ( D ) UAS-mys/UAS-mew in Ret mutant larvae rescues dendrite crossing defects . However , only re-expression of Ret fully restores dendritic field coverage to wildtype levels . Scale bar 100 μm . ( A′–D′ ) Magnified view of the indicated dendrite area for the different genotypes . Dendrite crossing points are indicated by arrowheads . Note that Ret-mCherry or integrin expression in Ret mutant C4da neurons strongly reduced dendrite crossing events . Scale bar 30 μm . ( E ) Quantitative analysis of out of plane dendrite crossing for the indicated genotypes . Overexpression of Ret or integrins specifically in C4da neurons of Ret mutant animals fully rescues dendrite crossing defects ( p < 0 . 001 , n = 5 ) ( F ) Dendrite coverage index ( ratio of dendrite field area and segment area ( Parrish et al . , 2009 ) ) is shown for the indicated genotypes . Defects in receptive field coverage of Ret mutant C4da neurons are rescued by C4da specific expression of the Ret-mCherry transgene , but not integrin overexpression ( mean ± SD , p < 0 . 001 , n = 5 , Mann–Whitney U-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 00710 . 7554/eLife . 05491 . 008Figure 5—figure supplement 1 . Integrins suppress Ret overexpression induced adult eye phenotypes . Integrin co-expression suppresses Ret induced photoreceptor degeneration . Overexpression of Ret with GMR-Gal4 caused a rough eye phenotype and pigmentation defects , which can be fully rescued by co-expression of integrins . Co-expression of Ret-mCherry and Mys-GFP could be readily detected by their fluorescent tag confirming the presence of both proteins in the rescued eye . Note that although Ret dependent photoreceptor phenotype was rescued , pigmentation defects observed upon Ret overexpression were not . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 008 We also asked if overexpression of the α/β-integrin complex is able to rescue dendrite-ECM interactions in Ret mutant animals . To this end , we overexpressed Mys and Mew in Ret mutant C4da neurons and assessed dendritic crossing and field coverage . Although the overexpression of integrins did not rescue all aspects of the Ret dendritic growth phenotype ( Figure 5D , D′ ) , it completely prevented dendritic out of plane crossing indicating restored interaction with the ECM ( Figure 5E ) . However , unlike Ret itself , integrin overexpression in C4da neurons did not rescue coverage defects of the dendritic field in Ret mutant animals ( Figure 5F ) . Co-overexpression of Mys and Mew alone did not cause dendrite abnormalities ( data not shown and Han et al . , 2012 ) indicating that Ret dependent growth involves both integrin dependent and integrin independent pathways . Interestingly , Ret overexpression in the compound eye has been reported to cause photoreceptor degeneration ( Read et al . , 2005 ) . We thus tested if co-expression of integrins could modulate Ret neurotoxicity in photoreceptor neurons ( Figure 5—figure supplement 1 ) . Integrin overexpression itself did not cause any eye phenotype , but was able to completely suppress Ret dependent photoreceptor degeneration . These findings show that , similarly to their interactions in C4da neurons , Ret and integrins genetically and functionally interact in the eye . Our data provide evidence that Ret and the integrins mys and mew likely operate together in a common pathway to regulate dendrite adhesion to the ECM . In order to gain insight into the underlying intracellular signals we investigated candidate pathways common to RTK and integrin signaling . Rho family GTPases , particularly rac1 , have been implicated in actin dynamics and dendrite morphogenesis of sensory neurons ( Lee et al . , 2003 ) and are a common target of both RTK and integrin signaling in many systems ( reviewed in Ivaska and Heino , 2011 ) . Using loss of function analysis , we tested whether rac1 is involved in dendrite-ECM adhesion of C4da neurons . Loss of rac1 function led to strong dendrite-ECM detachment and epithelial enclosure of dendrites highly similar to Ret and integrin mutant phenotypes ( Figure 6A–C ) . Overall , in rac1 mutant C4da neurons , 15% of dendrites in the field were detached from the ECM , an effect that was even more pronounced when the gene dosage of the other two rac-like GTPases , rac2 , and mtl was also reduced ( Figure 6C , D ) . In both cases , loss of rac1 function also led to many non-contacting crossing defects of dendritic branches ( Figure 6B , C arrows , and Figure 6E ) indicating that self-avoidance is impaired due to 3D growth of dendrites lacking rac1 function . 10 . 7554/eLife . 05491 . 015Figure 6 . Rac1 loss of function phenocopies and genetically interacts with Ret and integrins . The dorsal field of C4da neurons ( ppk-CD4-Tom ) and the extracellular matrix ( vkg-GFP ) were co-visualized in third instar larvae . High resolution two-color confocal z-stacks were analyzed for dendrite-ECM interaction and enclosed dendrite segments are indicated in magenta . Images of ( A ) Wildtype , ( B ) rac1J11 homozygous , and ( C ) rac1J11/rac1J10 , rac2Δ , mtlΔ larvae are shown displaying a strong increase of detached dendrites in the rac1 mutant animals . Scale bar 30 μm . ( D and E ) Quantitative analysis of dendrite-ECM interaction and dendrite crossing in wildtype and rac1 mutants shows a significant increase in dendrite detachment ( C ) and Z crossing points ( D ) in rac1 deficient C4da neurons . ( F–L ) Genetic interaction analysis of dendrite-ECM adhesion in ( F ) Ret , ( G ) mew , ( H ) mys , ( I ) rac1 heterozygous and ( J ) rac1/Ret ( K ) mew/rac1 ( L ) mys/rac1 trans-heterozygous third instar larvae . The combination of rac1 mutants with either Ret or integrins showed increased loss of dendrite-ECM interaction illustrated by detached portions of the dendritic tree ( in magenta ) . Scale bar 30 μm . ( M ) Quantitative analysis of dendrite detachment from the ECM for the individual genotypes as indicated ( mean ± SD , p < 0 . 05 , n = 4 per genotype , Mann–Whitney U-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 015 We next tested whether Ret and integrins are in the same genetic pathway as rac1 by analyzing heterozygous allelic combinations . Animals heterozygous for Ret , integrins , or rac1 did not show significant changes in C4da neuron dendrite-ECM adhesion compared to wildtype ( Figure 6F–I ) . When combining rac1 alleles with Ret or integrin mutant alleles however , we observed strong enhancement of dendritic loss of ECM adhesion ( Figure 6J–L ) . For both Ret/rac1 and integrin/rac1 heteroallelic combinations , 15–20% percent of dendrites lost contact with the ECM indicating strong crosstalk and regulation of Ret and integrins via rac1 . These findings strongly suggest that rac1 functions in the same pathway as Ret and integrins in regulating the attachment of dendrites to the ECM . Unlike Ret , neither integrin nor rac1 loss of function resulted in major defects in dendrite growth and coverage of C4da neurons . In addition , integrin overexpression was not able to rescue Ret dependent dendrite coverage defects ( see Figure 5 ) . We reasoned that Ret likely exerts additional functions independent of integrins/rac1 . We thus investigated the specific effects of Ret on dendrite growth and turnover . To this end , we performed time lapse analysis during late larval development at 72 hr and 96 hr after egg laying ( AEL ) . This allowed us to monitor dendrite dynamics within a 24 hr period . In wildtype C4da neurons , significant turnover of terminal dendrites could be observed ( Figure 7A ) , as reported previously ( Parrish et al . , 2007 ) . Overall , we detected dendrite growth and terminal dynamics required to maintain overall field coverage and tiling during larval growth . In Ret mutant animals however , C4da neuron dendrites already showed defects at 48 hr AEL ( not shown ) and 72 hr AEL , as evident from incomplete coverage and abnormal dendrite patterning at that stage ( Figure 7B ) . Increased dendrite crossing was also already apparent . Remarkably , analysis of dendrite turnover between 72 and 96 hr AEL revealed highly increased dynamics of dendritic growth and retraction , with major remodeling of the terminal dendrite arbors . Both growth and retraction of dendrite terminals in Ret mutant C4da neurons were increased two to threefold compared to wildtype ( Figure 7C ) . Overall , the growth/retraction ratio of dendrites was significantly reduced in Ret mutant C4da neurons due to a disproportional increase in retraction compared to growth ( Figure 7D ) . Our findings therefore suggest that Ret function is important for stabilizing growing dendrites . 10 . 7554/eLife . 05491 . 016Figure 7 . Ret mutant C4da neurons display increased dendrite dynamics and aberrant F-actin localization . ( A–B ) The dorsal dendrite field of the same C4da neurons expressing CD4-tdGFP was imaged at 72 hr and 96 hr AEL in ( A ) wildtype and ( B ) Ret mutant animals . Growth and retraction events of dendritic branches were traced and are highlighted in green ( growing dendrite ) and magenta ( retracting dendrite ) . Scale bar 25 μm . ( C ) Quantitative analysis of dendrite growth and retraction events in wildtype and Ret mutant C4da neurons showing the relative change in dendrite length normalized by the total dendrite length of the dorsal field . Note that growth and retraction of dendrites in Ret mutant neurons is strongly increased ( mean ± SD , p < 0 . 01 for growth , p < 0 . 05 for retraction , n = 4 ) . ( D ) Dendrite growth/retraction ratios of Ret deficient C4da neurons show a significant decrease compared to wildtype ( mean ± SD , p < 0 . 05 , n = 4 , Student's t-test ) . ( E ) F-actin levels in primary branches of Ret mutant C4da neurons are elevated in proximal regions compared to wildtype . Normalized LifeAct-mCherry signal along primary dendrites was plotted against increasing soma distance showing significantly different intensity profiles ( shaded area ± SD , p < 0 . 05 , n = 5 ) . ( F–G ) Dorsal fields of ddaC neurons expressing CD4-tdGFP and LifeAct-mCherry were imaged at 96–100 hr AEL in ( F ) wildtype and ( G ) Ret mutant animals . LifeAct levels normalized to GFP are shown as color coded arbitrary intensities . Ret mutant C4da neuron dendrites display increased proximal , but reduced distal F-actin levels . Abnormal accumulation of F-actin in aberrant , crossing dendrites was frequently seen in Ret mutant but not wildtype neurons ( indicated by arrows ) . Additionally , many terminal branches of Ret loss of function neurons have lower levels ( below threshold ) of F-actin than wildtype . Scale bar 30 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05491 . 016 We suspected that the observed increase of dendrite dynamics in Ret mutant C4da neurons was linked to the actin cytoskeleton , as integrins , Rac1 as well as Ret are able to directly or indirectly regulate actin ( Fukuda et al . , 2002 ) . To investigate this possibility , we used LifeAct imaging ( Riedl et al . , 2008 ) to analyze F-actin distribution in dendrites in vivo . Wildtype C4da neurons displayed fairly even distribution of F-actin along major dendrites , while terminal branches typically featured hotspots of increased F-actin levels ( Figure 7E , F ) . In contrast , F-actin levels were significantly increased and unevenly distributed in proximal regions of primary dendrites of Ret mutant neurons ( Figure 7E , G ) . The increase in proximal F-actin leveled out at more distal parts of the primary branches , where it reached similar levels in both genotypes ( Figure 7E ) . The overall F-actin distribution in major branches of Ret mutant C4da neurons was significantly different from wildtype ( p < 0 . 05 , n = 5 ) . Strikingly , aberrantly crossing terminal branches had increased levels of F-actin , while other terminals displayed either normal or very low F-actin levels compared to wildtype ( Figure 7G , arrows ) . This was also reflected in the centralization and overall uneven distribution of F-actin along dendritic arbors due to the lack of Ret function ( Figure 7E ) . These data clearly show that Ret is required for the appropriate localization of F-actin in dendrites , which likely contributes to the destabilization of growing branches . Taken together , our findings show that loss of Ret function impairs dendrite stability and increases dynamic growth and retraction of dendritic branches resulting in the failure of complete receptive field coverage . This is likely linked to the observed abnormal distribution of F-actin in the absence of Ret , which indicates a lack of appropriate growth signals and abnormal activity of the underlying signaling machinery coordinating actin dynamics .
In this study , we provide evidence that Ret is a regulator of dendrite growth and patterning of C4da neurons . Ret is a conserved receptor tyrosine kinase ( RTK ) expressed in the nervous system of vertebrates ( Pachnis et al . , 1993; Schuchardt et al . , 1994 ) and D . melanogaster ( Sugaya et al . , 1994; Hahn and Bishop , 2001 ) , and has been shown to have a number of important functions in nervous system development and maintenance: it regulates motor neuron axon guidance ( Kramer et al . , 2006 ) , dopaminergic neuron maintenance and regeneration ( Kowsky et al . , 2007; Kramer et al . , 2007 ) , and mechanoreceptor differentiation and projection to the spinal cord and medulla ( Bourane et al . , 2009; Luo et al . , 2009 ) . Ret signaling is activated by binding to glial cell line derived neurotrophic factor ( GDNF ) family ligands and their high affinity co-receptors , the GDNF family receptors ( GFRα ) ( reviewed in Runeberg-Roos and Saarma , 2007 ) . Ret also plays an important role in human development and disease as loss of function mutations of Ret lead to Hirschprung's disease displaying colonic aganglionosis due to defective enteric nervous system development ( Amiel et al . , 2008 ) . Conversely , Ret gain of function mutations are causal for autosomal dominant MEN2 ( multiple endocrine neoplasia type 2 ) type medullary thyroid carcinoma ( Lairmore et al . , 1993; Almeida and Hoff , 2012 ) . Prior to this study , Ret has not been implicated in dendrite development . Here , we show that Ret is required specifically for 2D growth of C4da neurons by regulating integrin dependent dendrite-ECM adhesion . Normally , C4da neuron dendrites are virtually always in contact with the ECM and the basal surface of the epithelium lining the larval cuticle , and thus tightly sandwiched between the two compartments ( Yamamoto et al . , 2006; Han et al . , 2012; Kim et al . , 2012 ) . In both integrin and Ret mutants , dendrite-ECM adhesion is impaired . Ret and integrins can co-localize in dendrites and thus likely form a functional complex that could induce and maintain adhesion of dendrites to the ECM . Since Ret loss of function primarily leads to detached terminal dendrite branches , it is tempting to speculate that Ret might be required to recruit integrins to sites of growing dendrites to promote ECM interaction . This is supported by the colocalization of Ret and integrins on the dendrite surface . Their cooperative interaction could thus ensure proper adhesion of growing branches and , conversely , the fidelity of self-avoidance and tiling . Our results also highlight the importance of integrating different guidance and adhesion cues to achieve precise neuronal patterning . This has so far only been studied in axon guidance in vivo ( Dudanova and Klein , 2013 ) . Interestingly , vertebrate Ret has been shown to cooperate with Ephrins to ensure high fidelity axon guidance in motor neurons by mediating attractive EphrinA reverse signaling ( Kramer et al . , 2006; Bonanomi et al . , 2012 ) . Similar mechanisms may conceivably be employed for growing dendrites , which also encounter a multitude of attractive , repulsive , and adhesive cues that have to be properly integrated . Besides pathways acting independently or in a parallel fashion , an emerging view is that receptors exhibit direct crosstalk to integrate incoming signals . So far , only parallel receptor pathways like Dscam and Netrin-Frazzled signaling in class III da neurons ( Matthews and Grueber , 2011 ) or Dscam/integrins ( Han et al . , 2011; Kim et al . , 2012 ) have been identified co-regulating dendrite morphogenesis . Our data from this study show that the Ret receptor and integrins integrate dendrite adhesion and growth by collaborative interaction of the two cell surface receptors . The molecular and genetic link between Ret and integrins suggests that in this case direct receptor crosstalk plays a major role in their function . How exactly these cell surface receptors cooperate and interact remains to be elucidated . Integrins have been shown to display extensive crosstalk with other signaling receptors , including RTKs ( Ivaska and Heino , 2011 ) . Although integrins are involved in adhesion of virtually all cell types , the underlying signaling and recruitment of integrins to sites of adhesion in vivo is complex and not completely understood . It has been suggested that integrin and growth factor receptor crosstalk can occur by concomitant signaling , collaborative activation , or direct activation of associated signaling pathways ( Ivaska and Heino , 2011 ) . For example , matrix-bound VEGF can induce complex formation between VEGFR2 and β1-integrin with concomitant targeting of β1-integrin to focal adhesions in endothelial cells ( Chen et al . , 2010 ) . Our findings of biochemical interaction and colocalization of Ret with the α/β-integrins mys and mew in C4da neuron dendrites argue in favor of direct receptor interaction and subsequent activation of a common signaling pathway . Integrins and RTKs like Ret do share some of the same intracellular signaling components . These comprise , among others , the MAPK ( mitogen-activated protein kinase ) pathway , Pi3-Kinase ( Pi3K ) , and Rho family GTPases including Rac1 ( Ivaska and Heino , 2011 ) . Previous studies provide evidence for Ret-integrin-Rac1 interplay in vitro showing that Ret can enhance integrin mediated adhesion ( Cockburn et al . , 2010 ) and induce Rac1 dependent lamellipodia formation ( Fukuda et al . , 2002 ) in cell culture models . In primary chick motor neurons , Rac1 is involved in neurite outgrowth on the integrin substrates laminin and fibronectin ( Kuhn et al . , 1998 ) . Interestingly , Rac1 has previously been shown to regulate dendrite branching in C4da neurons ( Lee et al . , 2003; Emoto et al . , 2004 ) , however a role in dendrite adhesion in vivo has not been described before . In our study , we show that Rac1 is required for dendrite-ECM adhesion similarly to what has been described for integrins and we genetically link Ret and integrin dependent adhesion with Rac1 function . In Drosophila , MAPK , Src and PI3K can be activated by constitutively active Ret overexpression in the compound eye ( Read et al . , 2005; Dar et al . , 2012 ) . Moreover , novel inhibitors of Ret signaling targeting Raf , Src , and S6-Kinase ( S6K ) prevent lethality induced by Ret over-activation in a Drosophila multiple endocrine neoplasia ( MEN2 ) model ( Dar et al . , 2012 ) . Interestingly , S6K has been shown to be involved in dendrite growth but not tiling in C4da neurons ( Koike-Kumagai et al . , 2009 ) . It remains to be shown if these pathways play a direct role in Ret function in dendrite adhesion and growth . Notwithstanding important commonalities , Ret function in C4da neurons cannot be fully explained by crosstalk with integrins and rac1 . Reduced dendritic field coverage , likely due to the observed increase in dendrite turnover , is only evident in Ret but not in integrin or rac1 mutant C4da neurons . Moreover , increasing integrin expression in a Ret mutant background did not rescue dendrite coverage defects , albeit it prevented dendrite crossing . These findings indicate that Ret has additional functions in dendritic branch growth and stability that require as yet unknown extracellular and intracellular mediators . This is also supported by the aberrant F-actin localization in neurons lacking Ret . Here , Ret dependent intracellular effectors are likely important for F-actin assembly to support directed dendrite growth and stabilization , and their localization and activity might be deregulated in the absence of Ret . Drosophila Ret is a highly conserved molecule , its cognate vertebrate ligand GDNF , however is not ( Airaksinen et al . , 2006 ) . In addition , Drosophila Ret can neither bind GDNF nor transduce GDNF signaling , although it has been shown to contain a functional tyrosine kinase domain ( Abrescia et al . , 2005 ) . In mammals , the GFRα co-receptors are essential components of GDNF/Ret signaling ( Runeberg-Roos and Saarma , 2007 ) . A Drosophila GFR-like homolog ( dGFRL ) has recently been characterized and was found to function and interact with the NCAM homolog FasII ( Kallijärvi et al . , 2012 ) . Therefore , it appears that Ret's functional interaction partners in dendrite development differ significantly from the previously described co-factors in other systems . It is interesting to speculate that a yet undiscovered Ret ligand is involved in Ret mediated dendrite growth and branch stabilization , which might have implications for mammalian Ret function as well: due to its role in the maintenance of dopaminergic neurons and motor axon growth in mouse ( Kramer et al . , 2006 , 2007 ) , adhesion related signaling via integrins could well be important during these processes . Moreover , the formation of a dorsal root ganglia derived mechanosensory neurons and their afferent and efferent fiber growth and innervation depends on Ret expression ( Bourane et al . , 2009; Luo et al . , 2009 ) . It will be interesting to investigate the functional interplay of Ret and integrins in central and peripheral target innervation and neurite maintenance in these systems , given the interdependent function of Ret and integrins in sensory dendrite growth as shown in our study . In summary , we describe a novel role for the Ret receptor in dendrite branch growth and stability in Drosophila C4da neurons . This role involves cell-autonomous effects of Ret on ECM adhesion , and F-actin localization in these neurons . Moreover , we have linked dendritic adhesion defects attributable to Ret to integrin and rac1 function featuring a novel and possibly conserved mode of action for Ret in dendrite development .
All fly stocks were maintained at 25°C and 70% rel . humidity on standard cornmeal/molasses food . The following fly stocks were used: RetC168 , RetNP7645 , mys1 , mewM6 , rac1J11 , rac1J10rac2ΔmtlΔ ( Hakeda-Suzuki et al . , 2002; Ng et al . , 2002 ) , ppk-CD4-tdTomato , ppk-Gal4 , UAS-CD4-tdGFP ( Han et al . , 2012 ) , trol-GFP , vkg-GFP , UAS-LifeAct-mCherry ( kindly provided by J Wildonger ) . RetC168 is a Piggyback transposon insertion in the 3′UTR of Ret and likely a hypomorphic allele leading to reduced Ret protein expression . Df ( 2L ) Bsc312 is a deficiency covering the genomic Ret locus . Ret cDNA was amplified from pGMR-Ret ( kind gift from Ross Cagan ) and cloned into pUAST-attB ( Groth et al . , 2004 ) . In addition , we generated Ret constructs carrying a C-terminal mCherry-tag or an additional pHluorin-tag inserted downstream of the 5′ signal peptide sequence . UAS-LifeAct-mCherry , UAS-Ret and UAS-Ret-mCherry , UAS-pHluorin-Ret-mCherry transgenes were generated by embryo injection into vasa-ΦC31;attP2 carrying flies according to standard procedures ( Groth et al . , 2004 ) . The ΦC31-integrase was outcrossed and transgenes were combined with the appropriate alleles and markers . An antibody against the Ret receptor was generated in guinea pigs by co-injecting the synthetic peptides ETKEVSPGWQAEDAV ( peptide 1 , corresponding to amino acids 1221–1235 of the intracellular C terminus ) and DIHDQATSYDQSEEEM ( peptide 2 , corresponding to amino acids 1095–1110 of the intracellular C terminus ) . The resulting antiserum was affinity purified against peptide 1 and used at a 1:1000 dilution . Peptide synthesis , antibody generation , and affinity purification were outsourced to Eurogentec . Larval filet preparation and staining was essentially performed as described ( Han et al . , 2012 ) . The following antibodies were used: rabbit anti-phospho-Ret ( 1:50 , Cell Signaling Technology , Danvers , MA ) , mouse anti-GFP ( 1:100 , Roche Diagnostics , Mannheim , Germany ) , mouse anti-mys ( 1:200 , Developmental Studies Hybridoma Bank , ( DSHB ) , Iowa City , IO ) , mouse anti-mew ( 1:50 , DSHB ) . Secondary DyLight or Alexa conjugated donkey antibodies were from Jackson ImmunoResearch ( Westgrove , PA ) and were used at 1:400–1:1000 . Larval filets of wildtype or Ret mutant animals carrying a C4da neuron specific marker ( ppk-Gal4 > CD4-tdGFP ) were prepared and immunostained as described above with rabbit anti-phospho-Ret ( 1:50 , Cell Signaling Technology ) or guinea pig anti-Ret ( 1:1000 ) antibodies . We quantified the Ret antibody signal of C4da neuron somata in confocal image stacks and calculated signal over background ratios ( ΔF/F ) by subtracting averaged background signal flanking the C4da soma region ( F ) . Statistical significance for the two genotypes was calculated using a two-tailed t-test . For cell surface staining of pHluorin-Ret-mCherry , larval filets were prepared in Ringer solution and incubated with a rabbit anti-GFP antibody ( 1:50 in Ringer solution , Life Technologies , Carlsbad , CA ) for 1 hr on ice . Filets were washed with Ringer solution ( 3 × 10 min at 4°C ) and then fixed with 4% Formaldehyde/PBS for 15 min on ice . Subsequent steps including anti-mew and secondary antibody incubation were carried out in the presence of 0 . 3% Triton X-100 as above . Third instar larval filet preparations immunostained for Ret and integrins were analyzed by confocal microscopy with high resolution using a high NA oil objective ( Zeiss LSM700 , 40×/NA1 . 3 , z step size: 300 nm ) . Confocal stacks covering the soma and part of the dorsal field of ddaC neurons ( xy dimension: 160 × 160 μm ) were obtained and deconvoluted using a blind deconvolution algorithm with an adaptive PSF ( AutoQuant , BitPlane AG , Zürich , Switzerland ) . Colocalization analysis was then performed on the deconvolved confocal stacks in 3D using the Imaris Coloc module ( BitPlane AG ) . The GFP reporter signal ( ppk-Gal4 > UAS-CD8-GFP or UAS-CD4-tdGFP ) was used to create a mask for the C4da neuron soma and dendrite signal to specifically analyze neuronal integrin and Ret signals . Ret and integrin signals were then automatically thresholded and colocalization was calculated ( Pearson coefficients , n = 5 per genotype ) . Costes randomization was performed for all samples to ensure that the calculated colocalization coefficients between Ret and integrins are non-random ( n = 100 iterations , p = 1 for all samples ) . S2 cells were grown at 25°C in Schneider's Drosophila medium ( Life Technologies ) with 10%FBS and Glutamine/Pen/Strep . For experiments , cells were seeded in 6 well plates and transfected at 50% density in an adherent state using Transfectene ( Qiagen , Venlo , Netherlands ) . For S2 cell expression , UAS constructs were co-transfected with pActin-Gal4 . The following constructs were used: pUAST-Ret-mCherry , pUAST-mys-3xflag-His , pUAST-mew . For co-immunoprecipitation experiments , cells were harvested 48 hr after transfection and lysed in 500 µl lysis buffer ( 50 mM Tris pH7 . 4 , 150 mM NaCl , 1% Triton X-100 ) for 20 min on ice . After 10 min/4°C/10 . 000×g centrifugation , supernatants were pre-incubated with mouse IgG Agarose ( Sigma–Aldrich , St . Louis , MO ) for 30 min at 4°C , and then incubated with anti-flag M2 agarose beads ( Sigma–Aldrich ) for 4 hr at 4°C . After extensive washing with lysis buffer , the samples were denatured and analyzed on Tris-Acetate gels ( Life Technologies ) and Western blotting against Ret-mCherry ( anti-DsRed , 1:1000 , BD Clontech , Mountain Viev , CA ) and mys ( anti-flag M2 , 1:10 . 000 , Sigma ) . For immunostaining , S2 cells were allowed to adhere to Concavalin A ( Sigma–Aldrich ) coated cover slips for 1 hr and subsequently fixed in 4% Paraformaldehyde/PBS solution for 10 min . After washing with PBS , cells were permeabilized with 0 . 1% Triton X-100 for 10 min and blocked with 5% donkey serum/PBS . Primary antibodies used were: rabbit anti-DsRed ( 1:500 , BD Clontech ) , mouse anti-mew ( 1:100 , DSHB ) , mouse anti-mys ( 1:100 , DSHB ) . Primary antibodies were applied over night at 4°C in 5% donkey serum/PBS , while secondary donkey antibodies with conjugated DyLight fluorophores ( Jackson ImmunoResearch ) were subsequently incubated for 1 hr at room temperature . Cells were mounted and imaged by confocal microscopy ( Leica SP5 and Olympus FV1000 ) . C4da neurons of third instar Drosophila larvae ( 72–100 hr AEL ) were imaged alive by confocal microscopy ( Leica SP5 or Olympus FV1000 ) , using ppk-CD4-tdTomato or ppk-Gal4 , UAS-CD4-tdGFP transgenes ( Han et al . , 2011 ) . Vkg-GFP or trol-GFP trap lines labeling endogenous vkg or troll proteins with GFP were used to visualize the extracellular matrix . Confocal stacks were taken to image ddaC C4da neuron dendrite fields either with 20× ( full ddaC field ) or 40× oil objectives ( dorsal ddaC field ) . For imaging dendrite-ECM interaction , two-color high resolution confocal stacks using a high NA oil objective ( Leica or Olympus , 40× , NA 1 . 3 ) were taken with a step size of 300 nm . For time lapse imaging of C4da neurons in wildtype and Ret mutant animals , ppk-Gal4 , UAS-CD4-tdGFP embryos were collected on grape agar plates for 2 hr and allowed to develop at 25°C . Two C4da neurons per animal ( abdominal segments a3 and a4 ) were imaged at 72 hr AEL , and the same neurons were imaged again 24 hr later ( 96 hr AEL ) . The imaged larvae were allowed to develop to adulthood to ensure that handling and imaging did not interfere with normal development . Dendrites of C4da neurons were traced with the Imaris Filament Tracer module ( BitPlane AG ) using deconvoluted confocal stacks . Total dendritic length was calculated from the traces and Z-crossing points were counted and manually confirmed as non-contacting by visual inspection of the z-stack . The high resolution 3D reconstruction was done in Imaris . Dendritic field coverage was calculated essentially as described ( Parrish et al . , 2009 ) by measuring the area covered by dendrites ( polygon formed by connecting all dendritic terminals of the field ) divided by the total segment area using Fiji/ImageJ ( NIH , Bethesda , MD ) . Statistical significance was calculated by comparison of all genotypes using a Mann–Whitney test ( Origin Pro , Origin Lab , Northhampton , MA ) . To quantify net dendrite growth/retraction , confocal Z-projections of the same neuron at 72 and 96 hr AEL were overlayed and adjusted for position and interstitial growth with bUnwarpJ ( Fiji , ImageJ ) . Landmarks marking all major branch points were used to get accurate overlay images . Growing ( branch is shorter at 72 hr than 96 hr AEL ) and retracting ( branch is longer at 72 hr than 96 hr AEL ) portions of dendrites were traced and growth/retraction values were normalized to the total length of the unchanged portion of the dendritic tree at 96 hr AEL . Statistical significance was calculated using a Student's two-tailed t-test . Analysis of dendrite-ECM interaction was essentially performed as described ( Han et al . , 2012 ) . Briefly , high resolution confocal stacks were deconvoluted using a blind deconvolution algorithm with an adaptive PSF ( AutoQuant , BitPlane AG ) . Dendrites were traced semi-automatically using the Imaris Filament Tracer module ( Bitplane ) and thresholded colocalization of the ECM and dendrite signals was performed . Non-contacting and contacting portions of the dendritic tree were traced and verified manually . Percentages of detached dendrites were calculated by the ratios of detached vs total dendrite length . Statistical significance was calculated by comparison of all genotypes using a Mann–Whitney test ( Origin Pro ) . The dorsal field of C4da ddaC neurons expressing LifeAct-mCherry with ppk-Gal4 > CD4-tdGFP was imaged with confocal microscopy at the third instar larval stage ( 96–100 hr AEL ) . Acquired stacks were then processed with Fiji/ImageJ ( NIH ) to normalize the LifeAct signal to GFP expression using a previously described procedure with minor adaptations ( Kardash et al . , 2011 ) . Briefly , we used maximum projections of 2-color stacks and aligned the GFP and LifeAct channels using the TurboReg plugin ( ImageJ ) . The LifeAct signal was thresholded automatically ( ImageJ ) and normalized to GFP levels using the RatioPlus plugin ( ImageJ ) . For LifeAct quantification along major dendritic arbors , normalized intensity values were obtained by ROI tracing of 2 major branches per sample using the PlotProfile function ( ImageJ , n = 5 per genotype ) . Statistical significance was then calculated by using an Allometric fitting function ( y = a × xb , Origin Pro ) for the intensity profiles of wildtype ( average R2 = 0 . 74 ) and Ret mutant ( average R2 = 0 . 79 ) samples and fit comparison of the two datasets ( p < 0 . 05 , n = 5 per genotype , Origin Pro ) . | There are hundreds of types of neurons , but all of them are variations on the same basic theme . Each neuron consists of a cell body , which contains the nucleus , and various structures that stick out from the cell body . These include a large number of short protrusions called dendrites , and a long thin cable-like structure called the axon . The dendrites receive incoming signals from the environment or neighboring neurons and transmit these signals to the cell body , which then relays them along the axon and on to the dendrites of the next neuron . As the brain develops , newly formed dendrites recognize and repel other dendrites belonging to the same neuron , thereby spreading themselves out to occupy a larger volume . This patterning process is called self-avoidance . At the same time , in order to repel each other , the dendrites must encounter each other in the first place , which means that they need to grow on a common substrate or surface . Soba et al . have now identified one of the proteins responsible for the self-avoidance process by studying the growth of dendrites on neurons in living fruit fly larvae . When the gene for a protein called the Ret receptor was deleted or inhibited , the dendrites that grew were shrunken and disorganized . High-resolution microscopy revealed that the dendrites were usually anchored to a scaffolding structure called the extracellular matrix , which ensured that they could only grow in two dimensions . However , when the gene for the Ret receptor did not work properly , the dendrites detached from this matrix and grew in three dimensions instead . Further experiments revealed that this detachment occurred because the Ret receptor was no longer interacting with a group of structural proteins called integrins . The Ret receptor plays a role in human disease and has previously been implicated in axon growth , but this is the first evidence to suggest that it also has a role in the patterning of dendrites . Given that Ret is present in vertebrates and has changed little over time , it is likely that this protein also helps to shape communication within the extensive networks of neurons that support complex cognitive functions in mammals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | The Ret receptor regulates sensory neuron dendrite growth and integrin mediated adhesion |
Aging is thought to be associated with increased molecular damage , but representative markers vary across conditions and organisms , making it difficult to assess properties of cumulative damage throughout lifespan . We used nontargeted metabolite profiling to follow age-associated trajectories of >15 , 000 metabolites in Drosophila subjected to control and lifespan-extending diets . We find that aging is associated with increased metabolite diversity and low-abundance molecules , suggesting they include cumulative damage . Remarkably , the number of detected compounds leveled-off in late-life , and this pattern associated with survivorship . Fourteen percent of metabolites showed age-associated changes , which decelerated in late-life and long-lived flies . In contrast , known metabolites changed in abundance similarly to nontargeted metabolites and transcripts , but did not increase in diversity . Targeted profiling also revealed slower metabolism and accumulation of lifespan-limiting molecules . Thus , aging is characterized by gradual metabolome remodeling , and condition- and advanced age-associated deceleration of this remodeling is linked to mortality and molecular damage .
The gradual and irreversible decline in cellular homeostasis that accompanies aging can be delayed by using several methods . For example , lifespan of various model organisms can be extended genetically via perturbation of nutrient sensing , pharmacologically by rapamycin treatment , or nutritionally by dietary restriction ( Bishop and Guarente , 2007; Fontana et al . , 2010 ) . Up-regulation of stress response and repair pathways , often linked to conditions of mild stress or hormesis ( Rattan , 2010 ) , may also promote longevity . On the other hand , the specific molecular processes that accompany the delay in the aging process are not well understood . It is often discussed that aging involves damage accumulation , therein an increase in DNA mutations , errors in protein synthesis , unwanted posttranslational modifications , metabolite by-products and many other damage forms contribute to the aging process ( Rattan , 2008 ) . The source of damage is frequently linked to metabolic activities , such as those that produce reactive oxygen species and increase oxidative modifications ( Kirkwood and Austad , 2000 ) . Damage may also be driven by global metabolic infidelity and engage nearly all biological processes involving synthesis and breakdown of cellular components ( Kirkwood and Austad , 2000; Gladyshev , 2013 ) . The exact mechanisms responsible for tolerance against molecular damage are not fully understood , but studies identified robust transcriptional remodeling that accompanies the aging process ( Lee et al . , 1999; Zou et al . , 2000; Pletcher et al . , 2002; McCarroll et al . , 2004; de Magalhaes et al . , 2009; Somel et al . , 2010 ) . How such responses are linked to cumulative damage is not known , since gene expression changes may not reflect immediate biological activities and metabolic fluxes . Another complication in such analyses is that individual damage types ( e . g . , mutations , lipofuscin accumulation , oxidative modifications ) vary across conditions and among species ( Rattan , 2008; Edman et al . , 2009; de Magalhaes , 2012; Jonker et al . , 2013 ) . The patterns of damage accumulation are also incompletely defined since many of the previous studies did not analyze samples representing very old ages ( Andziak et al . , 2006; Edman et al . , 2009; Jonker et al . , 2013 ) . Yet , insights into the properties of cumulative damage are required for comprehensive assessment of the activity-driven damage models of aging . Addressing these critical questions requires examination of damage composition and heterogeneity and the mortality-associated trends in the accumulation of numerous damage forms under both control and lifespan extending conditions . Aside from damage forms , it is unclear how cellular components generally change as a function of age , whether their patterns mimic the transcriptional responses and whether possible changes in their levels can have causal roles in the aging process . Metabolite profiling is an emerging method that aims to characterize a large number of small molecules in biological systems and identify proximal markers of biological activity ( Kristal and Shurubor , 2005; Patti et al . , 2012; Kotze et al . , 2013 ) . Recently , metabolite profiling was used to explore metabolic signatures of aging in young vs old mice ( Houtkooper et al . , 2011; Tomas-Loba et al . , 2013 ) using a set of markers , thereby confirming that aging is associated with alterations in nutrient sensing , lipid and amino acid metabolism , and redox homeostasis . Another strength of metabolite profiling lies in nontargeted profiling , which enables analysis of thousands of small molecules , albeit of unknown chemical properties . Nontargeted profiling may also be used to characterize patterns of molecular damage by following changes in metabolite diversity . For example , damaged molecular species are expected to exhibit age-associated mass shifts , which may be represented by an increase in metabolite diversity , potentially offering a larger repertoire of damage forms than analyzed previously . By following global changes in metabolome remodeling in response to aging and lifespan-extending interventions , the nontargeted profiling may also lead to the identification of small molecule regulators of the aging process . Here , we utilized nontargeted , liquid chromatography mass spectrometry ( LC-MS ) -based metabolite profiling to explore age-associated patterns in metabolite diversity and biological activity in Drosophila males maintained on diets that support different lifespan . We discover that metabolite diversity shows a robust age-associated increase . In contrast , known metabolites as well as many metabolites that are detected at all ages show bi-directional trends during aging: they fall predominantly into increasing and decreasing age-associated clusters . Interestingly , older cohorts feature deceleration in both the rise in metabolite diversity and overall cellular activity . Together , these data suggest that changes reflecting biological activity are linked to the display of metabolome diversity , which is reflected in the appearance of new small molecule species . We followed metabolites implicated in certain forms of damage and found their levels to be higher in long-living flies . These data further implicate a heterogeneous nature of damage whose effect on survivorship is condition dependent . Overall , these metabolite analyses provide critical insights into the activity-driven nature of the aging process .
We maintained fruit flies ( males ) throughout their lifespan on two dietary regimens: standard sugar and yeast diet and fully defined diet ( Mair et al . , 2005; Lee and Micchelli , 2013 ) . The defined diet , which was prepared from chemical components and mimicked dietary restriction conditions ( Bass et al . , 2007 ) , led to lifespan extension ( Figure 1A ) , similar to the well-characterized effect of dietary restriction ( Mair et al . , 2005 ) . Our design and choice for sampling age groups were also similar to the previous studies , which analyzed gene expression during Drosophila aging ( Zou et al . , 2000; Pletcher et al . , 2002 ) , except that we increased the number of samples at the end of the lifespan curve , including very old flies , which allowed us to better examine changes associated with advanced age . 10 . 7554/eLife . 02077 . 003Figure 1 . Dynamics of metabolite diversity throughout lifespan . ( A ) The number of detected nontargeted metabolites rises and then levels off as a function of cohort's age . Age-dependent changes in the number of detected metabolites ( red curve ) and intensity of total signal ( blue curve ) for nontargeted ( two left panels ) and targeted ( two right panels ) metabolites for standard ( two upper panels ) and defined ( two lower panels ) diets are shown . The lines were drawn using cubic polynomial fit function . Triangles mark data for the separately collected replicates for each age group . Significance for age-associated pattern in metabolite diversity was established using repeated measures ANOVA and was significant for nontargeted metabolites ( p<6 × 10−6 ) but not significant for targeted metabolites ( p>0 . 2 ) . The corresponding lifespan curves are shown in black in each panel , and the curves in grey ( for the other diet ) are shown for convenient comparison of survivorship on the two diets . Mean lifespan of flies on standard and defined diets was 50 . 8 and 64 . 4 days , respectively ( log-rank test [p<0 . 001] ) . ( B ) Metabolites which registered at zero in at least one sample ( 21 total samples [three associated replicates for each of the 7 age groups] ) were isolated from the dataset and , for visualization purposes , non-detected signals ( ones registering at 0 ) were changed to 1 × 10−5 and Log10 transformed with the remaining signals . Accordingly , points of non-detection in black along with the color gradient of the mass-spectrometry peak intensities for detected signals are provided on two age-supervised hierarchically clustered heatmap images . For comparison purposes , only metabolites overlapping in both dietary regimens were used . Side bracket exemplifies rises in age-associated metabolites . Side bars highlight metabolites from the lipid fraction ( yellow ) . Other metabolites are shown in black in this bar . ( C ) Histograms show overlaps in the distribution of signal intensities for all nontargeted metabolites vs targeted metabolites used to construct heat map images in B . DOI: http://dx . doi . org/10 . 7554/eLife . 02077 . 00310 . 7554/eLife . 02077 . 004Figure 1—figure supplement 1 . Correspondence in late life transition between metabolite diversity and mortality . For metabolite diversity , each triangle represents a sample at a given age , wherein a total of three replicates were used per age group . Circles correspond to the number of dead flies at each respective age . The lines were drawn using cubic polynomial fit function . DOI: http://dx . doi . org/10 . 7554/eLife . 02077 . 004 Our metabolite profiling platform detected >15 , 000 unique analytes ( nontargeted , non-redundant , de-isotoped LC-MS peaks ) , whose relative abundance was followed as a function of age . We found that the number of detected metabolites was lowest in the young flies in each dietary group and increased as a function of age ( repeated measures ANOVA , p<6 × 10−6 for each diet ) . On the other hand , the overall signal ( the sum of signals for all molecular species detected ) slightly decreased throughout lifespan ( Figure 1A ) . As metabolites were extracted from the same number of flies at each time point , the decrease in total signal may represent lower biomass in older flies , whereas the number of detected metabolites indicated increased metabolite diversity during the aging process . This pattern was observed in each replicate sample and in each diet . Interestingly , the rise in the number of detected metabolites was steady during adulthood , yet leveled off ( and even slightly decreased ) at the most advanced ages ( Figure 1A ) . This transition in the metabolite diversity corresponded to the late-life transition in the lifespan curves and mortality ( Figure 1—figure supplement 1 ) , suggesting relationship between metabolite diversity and the aging process . In contrast , analysis of metabolite diversity using a panel of 205 known metabolites did not show significant age-associated trends ( Figure 1A ) . Furthermore , the distribution of signal intensities between targeted metabolites and metabolites , which show diversity fluctuations throughout lifespan were different by few orders of magnitude ( Figure 1B , C ) , suggesting that the latter generally corresponded to molecules of low abundance and were not due to errors of detection . Also , changes in metabolite diversity could not be attributable to the presence of yeast-derived metabolites in our diet , since the chemically defined diet contained no cellular products . Lastly , the observed changes in metabolite diversity could be explained by the presence of outliving cohorts in samples near the end of the lifespan curve , as previously proposed for mortality rates ( Curtsinger et al . , 1992 ) . These data suggest a fair degree of synchronization between damage accumulation , expressed by the pattern of metabolite diversity , and the mortality pattern , including their changes in late life . Among the signals consistently detected in all aging samples ( those showing no difference in diversity ) , there may be metabolites that are passengers or drivers of the aging process , similar to the genes with age-associated differential expression ( Lee et al . , 1999; de Magalhaes et al . , 2009; Houtkooper et al . , 2011; Plank et al . , 2012; Tomas-Loba et al . , 2013 ) . Using repeated measures ANOVA and stringent statistical cut-off that estimated <0 . 2% of false positives ( ‘Materials and methods’ ) , we identified 2234 and 2216 metabolites ( ∼14% of all detected metabolites ) that qualified as age-associated in the standard and defined diets , respectively . This revealed widespread and dynamic metabolome reorganization throughout lifespan . Of the detected age-associated features , 1 , 066 were common to both dietary regimens ( Figure 2A ) . These molecules exhibited very similar trajectories of change between the two diets ( median Pearson correlation coefficient = 0 . 76 ) , in contrast to the lack of such correlation when using all metabolite pairs between the diets ( permutation test p<0 . 001 ) ( Figure 2B ) . To further characterize these age-related metabolites , we performed two types of scaling: between-group ( a metabolite was scaled across both dietary conditions ) and within-group ( a metabolite was scaled within each dietary condition ) . Principal component analysis ( PCA ) of the between-group scaled values segregated samples first by diet and then by age ( Figure 2C ) , suggesting that the magnitudes of these metabolites were affected by dietary conditions . On the other hand , PCA of the within-group scaled data revealed an age-dependent , but diet-independent segregation , with the replicate samples of both diets grouping closely together ( Figure 2D ) . Thus , these metabolites indeed followed very similar trajectories with age , after normalization of differences in their intensity levels . Furthermore , most age-related changes in metabolite intensity ( levels hereafter for simplicity ) within individual diets did not exceed 4–8-fold across lifespan ( Figure 3A , B ) , which was consistent with the differences in gene expression between Drosophila subjected to dietary restriction or stress ( Zou et al . , 2000; Pletcher et al . , 2002; Girardot et al . , 2004; Landis et al . , 2012 ) . These data show that diet- and age-influenced changes in metabolome remodeling throughout lifespan closely recapitulated transcriptional changes and progression of aging . 10 . 7554/eLife . 02077 . 005Figure 2 . Age- and diet-associated changes in metabolite levels accompany Drosophila aging . ( A ) Venn diagram of age-related features for two dietary regiments . The diagram shows that a large fraction of detected age-related features overlap between the two diets ( ‘overlapping metabolites’ ) . ( B ) Kernel Density Plot showing the distribution of Pearson correlation coefficients of the overlapping metabolites between the two diets ( color ) . The coefficients for the total metabolites ( gray ) are shown for comparison . ( C and D ) Clustering of the overlapping metabolites by Principal Component Analysis . Squares and triangles denote standard and defined diets , respectively . Each plot includes three replicates per age . ( C ) Plot with scaling expression values across diets . In this plot , individuals are separated according to their ages and diets . ( D ) PCA plot with diet-specific scaling separates individuals according to their ages only . DOI: http://dx . doi . org/10 . 7554/eLife . 02077 . 00510 . 7554/eLife . 02077 . 006Figure 3 . Distribution of fold-changes for age-related metabolites . ( A and B ) Fold-change differences within standard ( A ) and defined ( B ) diets were calculated by comparing changes in intensity from the ratio of maximum to minimum lifespan-associated values . ( C ) Inter-dietary differences are shown in two heatmap panels after their separation into twofold thresholds , which also show metabolite remodeling during aging . Heatmaps were generated as follows . Replicate values were averaged and then scaled within individual and also across the diets . The resulting matrix was then subjected to age-guided complete hierarchical clustering using hclust algorithm in R where ages were assigned to columns and individual metabolites were assigned to rows . The resulting image allows convenient visualization of clusters containing metabolites with common trajectories ( left side ) , which may also show inter-dietary differences in levels ( right side ) . Side bars were added to highlight metabolites derived from the lipid fraction and also trajectories bearing strong correlation to lifespan curves ( Pearson coefficient |r| >0 . 75 , color coded for each diet ) . Age-related trajectories were derived from trimming the distance matrix into 12 k-means clusters using rect . hclust function in R . Plots in each box represent averages of the scaled values of contributing metabolites whose number is listed in at the top of each graph . DOI: http://dx . doi . org/10 . 7554/eLife . 02077 . 006 We explored the relationship between changes in age-related metabolites and longevity . We carried out k-mean clustering to group metabolites with similar trajectories , separately for those with below twofold and above twofold inter-dietary changes ( Figure 3C ) , which showed that the diet-influenced changes were relatively underrepresented . Specifically , among the 1 , 542 age-related metabolites common to both diets , 76% showed less than twofold changes between the two diets ( Figure 4A ) , indicating that the influence of diet was noticeable , but mostly small in magnitude . These age-related changes may be linked to longevity as there was high prevalence of strong correlations ( both positive and negative ) between individual metabolite levels across lifespan and lifespan curves ( Figure 4B ) . While only a small fraction of metabolites showed more than twofold changes between the diets , they typically exhibited strong correlation with lifespan , suggesting they might represent mechanisms through which dietary modification ( e . g . , dietary restriction ) affected longevity . Next , to visualize the changes in metabolite levels during aging , we performed hierarchical clustering on both within-group and between-group scaled data for all signals commonly changing with age in both diets ( Figure 4C ) . Interestingly , a very large fraction of metabolites that decreased with age was derived from the lipid fraction , and this observation applied to both diets ( Figure 4C ) . Lipid profiles also declined with age in other animals ( Houtkooper et al . , 2011; Lapierre et al . , 2011; Singh and Cuervo , 2012 ) , suggesting that our analysis captures a general scheme of age-related changes in lipid metabolism . Among decreases , metabolites in clusters 6 and 8 declined more rapidly beginning in early , very young samples , suggesting that these signals had an earlier role , for example , during development and morphogenesis . In fact , age-related transcripts with the roles in morphogenesis experienced similar trends ( Clusters 6 , 7 , Figure 5A , B ) . Although most of the decreases in clusters 6 and 8 were significant , they were not strongly different between the diets . This may reinforce the idea of the developmental role for these lipid profiles as flies were reared on the standard medium until day 3 post eclosion , before the transfer to dietary restricted media . In contrast , most polar metabolites increased with age and their rates of change between the diets were more dynamic ( Figure 5C ) . In summary , the majority of age-related trajectories were very similar in both diets: they showed strong correlations to lifespan curves and were largely continuous as they followed trajectories observed in younger samples , while very few showed mid-life reversals ( clusters 10–12 ) ( Figure 4D ) . Some of the variations in metabolite magnitude between the diets , on the other hand , may be responsible for the observed lifespan differences . 10 . 7554/eLife . 02077 . 007Figure 4 . Common and distinct patterns of metabolite remodeling during the aging process . ( A ) Fold-change differences between common age-related metabolites from two dietary conditions ( inter-dietary differences ) . Fold change was calculated using averages of individual metabolite levels across each lifespan . ( B ) Distribution of Pearson coefficients . The Kernel-Density function in R was used to plot the distribution of all Pearson coefficients representing correlations between each of the age-related metabolites and lifespan curves for standard ( green ) and defined ( purple ) diets . Signals were split into groups that showed inter-dietary differences of under ( solid lines ) or above ( dashed lines ) twofold-change . ( C ) Signals were clustered using methods described in Figure 3C legend . Side bars were added to highlight positions of the metabolites bearing statistically significant inter-dietary differences ( Student t test , p<0 . 05 ) , metabolites meeting above twofold inter-dietary change , metabolites from the lipid fraction , and metabolites bearing strong correlation to lifespan ( Pearson coefficient |r| >0 . 75 , color coded for each diet ) . ( D ) Age-related trajectories were derived from the hierarchical tree as described in Figure 3C legend . DOI: http://dx . doi . org/10 . 7554/eLife . 02077 . 00710 . 7554/eLife . 02077 . 008Figure 5 . Age-related transcripts and metabolites follow similar trajectories and show a delayed response under lifespan-extending dietary conditions . One way repeated measures ANOVA was used to identify transcripts with age-related changes at p<0 . 0013 . A total of 1171 features showed significance in both diets . ( A ) Normalization and clustering were performed according to the procedures described for Figure 3C legend . Each box represents individual clusters trimmed from hierarchically clustered tree using hclust algorithm in R . The number of genes contributing to each cluster is provided in the bottom left corner . ( B ) ClusterProfiler ( Yu et al . , 2012 ) package in R ( Yu et al . , 2012 ) was used to test for enrichment for Biological Process ontology in Clusters 1–12 in ( A ) . Clusters 3 , 10 , and 12 did not enrich and therefore are not present . ( C ) Comparison of diet-dependent and diet-independent frequencies in gene and metabolite expression data . Frequencies of diet-dependent to diet-independent changes in gene expression and metabolites were obtained from signals provided by clusters 112 in panel A and Figure 4D , respectively . Differences that were continuous across lifespan were categorized as progressive , and those that were not as intermittent . ( D ) Average trajectories of upregulated ( solid lines ) and downregulated ( dashed lines ) signals in gene ( top panel ) and metabolite ( bottom panel ) expression datasets . For gene expression , the upregulated trajectories are averages of all signals from Clusters 1–5 shown in panel A , while all downregulated signals were derived from Clusters 6–8 . Similarly , the global increases and decreases in metabolite levels were generated by averaging signals in Clusters 2–4 and Clusters 5–8 , respectively , from Figure 4D . Plots show normalized trajectories' values , which were obtained using quadratic polynomial fit through sample replicates . Points indicate sampled ages . DOI: http://dx . doi . org/10 . 7554/eLife . 02077 . 008 If the observed global changes in metabolite levels are biologically relevant , we should expect to see similar patterns in gene expression across lifespan . For this , we analyzed a gene expression data set that sampled mated female flies raised similarly to our two dietary regimens ( Pletcher et al . , 2002 ) . By applying similar statistical methods as those used for identifying the age-associated metabolites , we identified 1172 age-associated genes ( ca . 8% of total ) common to both diets ( repeated measures ANOVA , p<0 . 0013 ) . Clustering analysis revealed that about 70% of age-related genes also featured gradual transitions throughout lifespan , both increasing ( clusters 2 through 5 ) and decreasing ( clusters 6 through 8 ) with age ( Figure 5A ) . We used Gene ontology ( GO ) ( Ashburner et al . , 2000 ) to test for enrichment of biological processes within each cluster and found that longevity in dietary restricted ( DR ) flies was associated with reduced metabolism ( cluster 2 ) , increased transport ( cluster 11 ) and altered response to pathogens , albeit early in life ( cluster 4 ) ( Figure 5B ) . Similarly to previous studies in mammals and other species , genes with increased levels in cluster 1 were enriched for stress response genes , whose expression rose at slower rate in DR flies , suggesting that improved homeostasis in these individuals may reduce the need for certain stress protection ( Lee et al . , 1999; de Magalhaes et al . , 2009; Plank et al . , 2012 ) . Another interesting observation was that all clusters , except for clusters 4 and 5 , were enriched for non-overlapping GO terms , suggesting that the age-related changes in distinct biological processes proceeded synchronously . We suggest that this should also apply to nontargeted metabolites . To compare metabolite and gene expression data further , we examined the frequency of diet-dependent vs diet-independent changes ( e . g . , trajectories that trended at different rates , levels and directions between the diets ) ( Figure 5C ) . First , for transcript changes , there were two types of diet-dependent effects: progressive , which differed at all age points ( clusters 1–3 , 8 , 10 and 11 ) and intermittent which differed at some , but not all age points ( clusters 6 , 9 and 12 ) . The diet-independent effects , however , predominated ( 68% ) and fell into clusters 4 , 5 and 7 . Strikingly , we observed similar frequencies for age-related metabolites whereby clusters 2 , 3 , 6 and 10 can be categorized as diet-independent making up to 62% of total age-related signals common to both diets . The remaining fraction represented both progressive ( clusters 4 , 5 , 7 , 8 , and 11 ) and intermittent ( clusters 1 , 9 and 12 ) effects . The diet-dependent differences in metabolite levels were more obvious after the clusters were separated into groups with below and above twofold inter-dietary difference ( Figure 3C ) . The observed gradual , dynamic metabolome remodeling throughout lifespan , including both increases and decreases in metabolite signals , implied a role of predetermined , programmatic ( but not necessarily programmed ) , genetically-defined aging ( Pletcher et al . , 2002; Somel et al . , 2010; de Magalhaes , 2012 ) . Importantly , similarities between their trajectories indicate that age-related metabolites effectively recapitulate gradual and continuous aging transitions . We propose that the vast majority of age-related metabolite trajectories are directly related to aging and may contribute to its manifestation and/or delay . We further examined coordination in transcriptome and metabolome remodeling by examining transcript and metabolite signals that changed in both dietary groups . Interestingly , these age-related changes ( both increases and decreases common to both diets ) in either gene expression or metabolite levels were delayed in long-living flies ( Figure 5D ) . Moreover , metabolome remodeling represented by these signals was decelerated with age within each dietary group ( i . e . , the signals initially showed robust changes , but then almost leveled off ) . Together with the analyses that revealed the diet-associated changes in the metabolome composition , these findings suggested that a change in lifespan involves both a delay in age-related metabolome and gene expression remodeling and a change in the set of expressed metabolites and genes at any point throughout lifespan . These data also resemble the metabolite diversity pattern ( Figure 1A ) and suggest a close relationship between activity-dependent damage accumulation and aging . Behavioral or genetic alterations of nutrient sensing pathways and metabolism lead to lifespan extension in diverse phyla ( Lin et al . , 2002; Pletcher et al . , 2002; McElwee et al . , 2007; Panowski et al . , 2007 ) , suggesting that metabolite profiling could capture metabolic alterations between aging control and long-lived flies . We addressed this by examining several metabolic pathways by screening for age-associated profiles among 205 annotated metabolites , which we recalled from nontargeted spectral features and confirmed using authentic in-house standards . Among these , 81 metabolites displayed age-associated trends under both dietary regimens ( repeated one-way ANOVA , p<0 . 05 ) and were associated with biosynthesis of amino acids , lipids , energy homeostasis , and damage production ( Figure 6 ) . As a whole , the clustering analysis by means of inter-dietary comparison of these age-associated molecules showed no prevalence for metabolite downregulation in flies from the dietary restricted regimen: there was nearly an equal frequency of diet-dependent to diet-independent effects ( Figure 7A , B ) . Examination of the molecules within individual clusters ( as they share similar age-related trajectories ) led to several important observations . First , we find that , as in the case of nontargeted metabolite profiles , decreases of targeted signals were predominantly associated with lipids ( cluster 1 , Figure 7A , B ) . Eighty percent of metabolites in this cluster were triglycerides ( TAGs ) and diacylglycerols ( DAGs ) , all of which normally rise highly during larval growth but decline thereafter ( Carvalho et al . , 2012 ) . The high developmental expression of these lipid species may explain their rapid drop during adulthood in our data , well before the first major change in survivorship curves . Such patterns are also present within decreasing clusters of nontargeted metabolites that contain a large number of metabolites from the lipid fraction ( clusters 6 , 8 , Figure 4D ) . Also , the age-associated tendencies between our standard and DR-like regimens suggest increased synthesis ( or reduced utilization ) of lipids in long-living flies and agree with higher lipid stores and starvation resistance of DR flies in previous studies ( Burger et al . , 2007; Katewa et al . , 2012 ) . In contrast , we observed a progressive age-associated increase in the cholesteryl esters in our long-living flies compared to controls ( cluster 11 ) . While accumulation of this lipid class has been inferred to negatively affect mammalian fitness ( Lawton et al . , 2008 ) , it may extend Drosophila lifespan via increased pathogen protection ( Caragata et al . , 2013 ) . Second , consistent with the effects of dietary restriction , we find that downregulated metabolites in long-lived flies ( clusters 5–8 , n = 39 ) were associated with insulin signaling , energy homeostasis , and amino acid metabolism ( Figure 7C ) . Third , a striking observation was the elevation of methionine sulfoxide , which represents damage ( Orentreich et al . , 1993; Ruan et al . , 2002; Koc et al . , 2004 ) , and taurine , a product of cysteine degradation and a component of bile acids ( Massie et al . , 1989; Brosnan and Brosnan , 2006 ) , in long-lived but not in control flies ( cluster 9 , Figure 7D ) . The levels of these molecules started to decline at the point corresponding to high mortality in both groups ( at 40 and 50 days , respectively ) , which may be due to the low levels of these metabolites in cohorts surviving to old age ( Curtsinger et al . , 1992 ) . Tryptophan is also a lifespan limiting amino acid ( De Marte and Enesco , 1986 ) and a member of that cluster showing a similar pattern . Such bell-shaped lifespan-associated curves are not surprising as degradation of tryptophan and rise of the downstream metabolite kynurenine ( also lifespan limiting metabolite [Oxenkrug et al . , 2011] expressed higher on the defined diet ) were reported in older individuals ( Moroni et al . , 1988; Frick et al . , 2004 ) . In sum , the higher levels of methionine sulfoxide , tryptophan , and kynurenine in our long-living flies are consistent with the notion that lifespan extension by dietary restricted regimen may increase damage tolerance . We suggest that long-living species may also feature higher levels of lifespan-limiting molecules and that metabolome remodeling upon dietary interventions activates compensatory mechanisms . Whether this is broadly applicable to a variety of lifespan-extending conditions and to long-lived species is a possibility worth exploring . 10 . 7554/eLife . 02077 . 009Figure 6 . Identification and metabolic pathway representation of significant age-related targeted metabolites . ( A ) Overview of molecules significantly associated with aging according to biological processes in both diets ( repeated measures ANOVA , p<0 . 05 ) . ( B ) Venn diagram showing the number of significantly changing metabolites with relation to the number of metabolites uniquely significant to standard ( solid ) or defined ( dashed ) diets . DOI: http://dx . doi . org/10 . 7554/eLife . 02077 . 00910 . 7554/eLife . 02077 . 010Figure 7 . Metabolic signatures of aging in control and dietary restricted flies . Annotated ( targeted ) metabolites were derived from raw nontargeted data and represent only signals of established chemical identities . ( A and B ) Patterns of targeted metabolites . Clustering and graphing were done identically to the procedures described for Figure 3C legends . Side bars highlight lipid species and cluster boundaries that correspond to consequently arranged plots in ( B ) . ( C ) Metabolite set enrichment analysis was performed by MetaboAnalyst 2 . 0 ( Xia and Wishart , 2011 ) . The panel overviews low expressing signals in long-living flies ( Clusters 5–8 ) . ( D ) Metabolites representing known damage and lifespan limiting factors overlayed with lifespan curves for standard ( solid ) and defined ( dotted ) diets . Taurine and kynurenine showed statistically significant inter-dietary changes across lifespan . Methionine sulfoxide differed significantly between 10 and 25 day groups ( Student t test p<0 . 50 ) . Tryptophan showed no significant inter-dietary differences at 60–63 days . Circles correspond to sampled age groups , whereby z-scored expression values are generated from averages of randomly measured replicates representing separately sampled cohorts in standard ( green ) or defined ( purple ) diets . DOI: http://dx . doi . org/10 . 7554/eLife . 02077 . 010
Aging was suggested to result from accumulation of molecular damage that leads to age-associated decrease in organism's fitness ( Orgel , 1963; Kirkwood and Austad , 2000 ) . Yet , individual damage forms , for example , damage caused by reactive oxygen species , proved to be both condition- and species-specific , challenging the relevance of damage accumulation to aging ( Blagosklonny , 2008 ) . However , aging may be associated with cumulative damage , whereby mild effects of the myriad damage forms may exhibit additive effects on organism's fitness . In this case , increased mortality may be associated with inevitable accumulation of heterogeneous damage forms , which are also condition dependent . For example , in interventions that extend lifespan ( e . g . , dietary restriction ) reduction of some damage forms may be compensated by accumulation of other forms of damage . In addition to the role of damage , the contribution of changes in the levels of purposely used metabolites remains unclear . An age-related remodeling has been described for transcriptome , epigenome , and proteome , but how these changes relate to each other and what the role of metabolome remodeling is in the progression of aging is not well understood . Our report offers insights into the nature of aging through a comprehensive assessment of changes in a large number of endogenous small molecules throughout Drosophila lifespan . Our initial observations showed that global metabolite profiling effectively recapitulated distinct stages of the aging process and revealed metabolites that strongly correlate with organismal survivorship . Specifically , analysis of nontargeted features showed a robust increase in the number of detected molecules as a function of age , followed by cessation of the increase in late life . This pattern corresponded to the transitions in the lifespan curves and was also in concert with the observed decelerated metabolome remodeling in late life , altogether supporting the idea that aging subsides at advanced ages . On the other hand , the increase in the number of detected metabolites clearly differed from general age-associated changes , and its commonalities in the two dietary conditions exclude the role of exogenous compounds . The extended metabolome composition may include by-products that affect viability and damage forms that reach and pass the detection thresholds during organism's aging ( Phoenix and de Grey , 2007 ) , but they likely represent only the tip of the iceberg of cumulative damage . Further advances in instrumentation and sensitivity of metabolite profiling should expose additional metabolites that characterize cumulative damage . We further followed age-related trajectories of 1066 individual metabolites common to short- and long-living flies and found that the trajectories of most metabolites correlated between the two dietary groups , yet varied in magnitude from initial to final concentration changes , suggesting that differences in accumulation of individual metabolites , in addition to the overall pattern , were responsible for lifespan differences . A substantial fraction of these signals showed biologically-relevant differences during aging and correlated with survivorship of flies . Metabolite trajectories were gradual and continuous starting from early adulthood , with very few signals showing mid-life reversals . These results are generally consistent with the observations from the studies on transcript and protein profiles across diverse aging populations ( Zou et al . , 2000; Pletcher et al . , 2002; Somel et al . , 2010 ) . Indeed , our analysis of age-associated gene expression involving both control and dietary restricted long-lived individuals indicates that metabolites and transcripts exhibit nearly identical frequencies of diet-dependent to diet-independent effects . Thus , metabolites , similarly to genes , may exhibit diverse lifespan-regulatory roles . Similarly to targeted metabolite profiling in mammals ( Tomas-Loba et al . , 2013 ) , the gradual transitions in thousands of nontargeted metabolites may be used to derive a signature that predicts survivorship across variable lifespans . Consistent with this idea , we observed a delay in the age-associated changes in transcripts and metabolites in the case of longer-lived flies . Thus , a change in lifespan ( here , in response to a dietary regimen , but this should also apply to longevity-related genetic manipulations and evolutionary processes that affect species lifespan ) involves both a change in the expressed transcripts and metabolites at any time throughout lifespan and a delay in age-related changes for those transcripts and metabolites that commonly change with age in both conditions ( or genotypes ) . A significant overlap between metabolites that show age-associated changes between the diets can also identify the biological , rather than the chronological , aging component . Together with the component that reflects diet-specific patterns of metabolite change , it should help define the aging process in model systems and beyond . For instance , comparison of inter-dietary age-associated trajectories using known metabolites reveals that a fraction of downregulated signals were associated with slower metabolism in long-living flies . However , we also found that the molecular species with higher levels in long-living flies represented damage forms . While the inter-dietary differences for such increases were below twofold and varied in significance across age , the data nonetheless implies that lifespan extension may be associated with damage tolerance via compensatory mechanisms . In fact , long-living flies , when compared to control , featured a large set of nontargeted metabolites that increase at higher levels . How many of these metabolites represent damage will need to be determined through structural elucidation of nontargeted signals . Overall , from the current study the aging process emerges as a gradual , dynamic metabolome remodeling that involves changes in the levels ( both increases and decreases ) in numerous cellular metabolites , delays in these changes in late life and under conditions that lead to longer lifespan , and accumulation of damage , whose condition-dependent cumulative effects may impact survivorship .
Progenies used in aging assays and metabolite profiling experiments were prepared by mating wild-type animals of Canton-S background , which were backcrossed for seven generations . Aging progeny was grown on rich media ( flystocks . bio . indiana . edu ) at regular density . Newly eclosing F1 adult males were collected for 3 days , mixed and distributed within vials containing respective food for sampling at predetermined ages . The recipes for preparation of the defined diet have been described ( Troen et al . , 2007 ) . Between 30 and 50 flies were sampled at each age , rapidly frozen in liquid nitrogen and stored at −80°C . Once all samples were collected they were immediately processed for LC-MS profiling . Three LC-MS methods were used to measure polar metabolites and lipids in whole fly homogenates . Conditions for the analysis were set using a panel of routinely analyzed 293 standards . Polar and lipid-associated species were extracted from 7 and 2 flies , respectively , in three separate replicates which were ran in randomized order . All data were acquired using an LC-MS system comprised of a Nexera X2 U-HPLC ( Shimadzu , Marlborough , MA ) and a Q Exactive hybrid quadrupole orbitrap mass spectrometer ( Thermo Fisher Scientific; Waltham , MA ) . Hydrophilic interaction liquid chromatography ( HILIC ) analyses of water soluble metabolites in the positive ionization mode were carried out by extracting 10 µl homogenate using 90 µl of 74 . 9:24 . 9:0 . 2 vol/vol/vol acetonitrile/methanol/formic acid containing stable isotope-labeled internal standards ( valine-d8 , Isotec; and phenylalanine-d8 , Cambridge Isotope Laboratories; Andover , MA ) . The samples were centrifuged ( 10 min , 9000×g , 4°C ) and the supernatants were injected directly onto a 150 × 2 mm Atlantis HILIC column ( Waters; Milford , MA ) . The column was eluted isocratically at a flow rate of 250 µl/min with 5% mobile phase A ( 10 mM ammonium formate and 0 . 1% formic acid in water ) for 1 min followed by a linear gradient to 40% mobile phase B ( acetonitrile with 0 . 1% formic acid ) over 10 min . The electrospray ionization voltage was 3 . 5 kV and data were acquired using full scan analysis over m/z 70–800 at 70 , 000 resolution and a 3 Hz data acquisition rate . Negative ion mode analyses of polar metabolites were achieved using a HILIC method under basic conditions . Briefly , 30 µl homogenate was extracted using 120 µl of 80% methanol containing inosine-15N4 , thymine-d4 , and glycocholate-d4 internal standards ( Cambridge Isotope Laboratories; Andover , MA ) . The samples were centrifuged ( 10 min , 9000×g , 4°C ) and the supernatants were injected directly onto a 150 × 2 . 0 mm Luna NH2 column ( Phenomenex; Torrance , CA ) that was eluted at a flow rate of 400 µl/min with initial conditions of 10% mobile phase A ( 20 mM ammonium acetate and 20 mM ammonium hydroxide in water ) and 90% mobile phase B ( 10 mM ammonium hydroxide in 75:25 vol/vol acetonitrile/methanol ) followed by a 10 min linear gradient to 100% mobile phase A . MS full scan data were acquired over m/z 70–800 . The ionization source voltage is −3 . 0 kV and the source temperature is 325°C . Lipids were extracted from 10 µl of homogenate using 190 µl of isopropanol containing 1-dodecanoyl-2-tridecanoyl-sn-glycero-3-phosphocholine ( Avanti Polar Lipids; Alabaster , AL ) . After centrifugation , supernatants were injected directly onto a 150 × 3 . 0 mm Prosphere HP C4 column ( Grace , Columbia , MD ) . The column was eluted isocratically with 80% mobile phase A ( 95:5:0 . 1 vol/vol/vol 10 mM ammonium acetate/methanol/acetic acid ) for 2 min followed by a linear gradient to 80% mobile-phase B ( 99 . 9:0 . 1 vol/vol methanol/acetic acid ) over 1 min , a linear gradient to 100% mobile phase B over 12 min , then 10 min at 100% mobile-phase B . Full scan MS analyses ( m/z 400–1000 ) were carried out in the positive ion mode using full scan analysis at 70 , 000 resolution and 3 Hz data acquisition rate . All raw data were processed using Progenesis CoMet software ( version 2 . 0 , NonLinear Dynamics ) for feature alignment , signal detection , and signal integration . Signal peak areas were converted into numerical intensity values and normalized to internal standards added to each sample and to total signal at each time point . The raw peak intensity values are provided as Supplementary dataset ( Avanesov et al . , 2014 ) . Age-associated signals were identified using repeated measures ANOVA , and significance was established at p<0 . 0014 ( <0 . 2% of false positives ) and False discovery rate ( FDR ) of 0 . 029 . To calculate FDR , we first established relationships between error rate and its corresponding p value ( Storey and Tibshirani , 2003 ) and used an adjusted p value at the interface that marked transition between linear and exponential rise in error rate . Then , FDR was calculated using compute . FDR function in R brainwaver package with the significance thresholds established above . Data analysis was also carried on in the R environment; information on packages used in this study can be found within respective figure legends . Metabolite enrichment analysis was performed in MetaboAnalyst 2 . 0 ( Xia and Wishart , 2011 ) . For gene expression analysis , we used a dataset that was normalized using RMA package of Bioconductor in R and kindly provided by Dr Scott Pletcher ( Pletcher et al . , 2002 ) . Age-associated gene expression signatures were identified identically to the metabolites above and corrected for multiple testing ( p<0 . 0013 , FDR = 0 . 025 ) using aforementioned computations and significance threshold selections . | Signs of aging have been observed in many different species , but the underlying mechanisms are still poorly understood . It is thought that aging is influenced by metabolism . For example , scientists have found that metabolism can lead to the accumulation of byproducts , which may cause damage to cells . Moreover , as organisms get older , the diversity of these byproducts can increase . However , it has proven difficult to measure this cumulative damage . Avanesov et al . have now tried a different approach and examined the relationship between cellular metabolism , lifespan , and cumulative damage . Male fruit flies were raised on one of two diets—a standard diet , or a restrictive diet that extends their lifespan—and a technique called metabolite profiling was then used to monitor more than 15 , 000 metabolites in both sets of flies . Avanesov et al . found that the number of metabolites increased over time , suggesting that damage or mistakes in molecular synthesis increased with age . But the number of metabolites reached a plateau in the oldest flies , even in those whose lifespans were artificially extended . This could be due to cells becoming less active as they get very old . Avanesov et al . also found that the profile of the metabolites changed in a way that was similar to the way that patterns of gene transcription changed . This suggests that there may be a link between transcription—which is the first step in the process that produces proteins in cells—and metabolism and aging . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2014 | Age- and diet-associated metabolome remodeling characterizes the aging process driven by damage accumulation |
Obesity is associated with blunted β-adrenoreceptor ( β-AR ) -mediated lipolysis and lipid oxidation in adipose tissue , but the mechanisms linking nutrient overload to catecholamine resistance are poorly understood . We report that targeted disruption of TGF-β superfamily receptor ALK7 alleviates diet-induced catecholamine resistance in adipose tissue , thereby reducing obesity in mice . Global and fat-specific Alk7 knock-out enhanced adipose β-AR expression , β-adrenergic signaling , mitochondrial biogenesis , lipid oxidation , and lipolysis under a high fat diet , leading to elevated energy expenditure , decreased fat mass , and resistance to diet-induced obesity . Conversely , activation of ALK7 reduced β-AR-mediated signaling and lipolysis cell-autonomously in both mouse and human adipocytes . Acute inhibition of ALK7 in adult mice by a chemical-genetic approach reduced diet-induced weight gain , fat accumulation , and adipocyte size , and enhanced adipocyte lipolysis and β-adrenergic signaling . We propose that ALK7 signaling contributes to diet-induced catecholamine resistance in adipose tissue , and suggest that ALK7 inhibitors may have therapeutic value in human obesity .
Agonists of β3-adrenoceptors ( β3-AR ) are effective anti-obesity agents in rodents due to their ability to stimulate lipolysis and lipid oxidation in adipose tissue ( Arch , 2011 ) . However , efforts to develop similar compounds in humans stalled when it became clear that the pharmacology of rodent and human β3-AR differed and lipolysis in human adipocytes was mainly mediated by classical β1- and β2-AR , which can also induce hypertension , tachycardia , and other undesired complications ( Arch , 2011 ) . Alternative strategies to enhance catecholamine sensitivity and β-adrenergic signaling selectively in adipose tissue may be effective in combating obesity , but remain unproven . In both humans and rodents , obesity is associated with blunted β-AR-mediated lipolysis and lipid oxidation in adipose tissue ( Reynisdottir et al . , 1994; Arner , 1999; Jocken et al . , 2008 ) , but the mechanisms linking nutrient overload to catecholamine resistance remain poorly understood . ALK7 is a receptor for a subset of ligands from the TGF-β superfamily , including Nodal , activin B , and GDF-3 ( Rydén et al . , 1996; Reissmann et al . , 2001; Tsuchida et al . , 2004; Andersson et al . , 2008 ) . ALK7 is not necessary for mouse embryogenesis ( Jörnvall et al . , 2004 ) , suggesting alternative functions in postnatal development and adult physiology . ALK7 is highly expressed in rodent and human adipose tissue ( Kang and Reddi , 1996; Andersson et al . , 2008; Carlsson et al . , 2009; Murakami et al . , 2012 ) , as well as in a few other tissues implicated in metabolic regulation , such as pancreatic islets ( Bertolino et al . , 2008 ) and the arcuate nucleus of the hypothalamus ( Sandoval-Guzmán et al . , 2012 ) . Mice lacking ALK7 show reduced fat accumulation after a high fat diet ( Andersson et al . , 2008 ) and in a polygenic model of obesity ( Yogosawa et al . , 2013 ) , but the mechanisms underlying the effects of ALK7 signaling on diet-induced obesity are not understood . Pancreatic islets from Alk7 knock-out mice show enhanced glucose-stimulated insulin secretion ( Bertolino et al . , 2008 ) , a phenotype that is also present in islets from mutant mice lacking the ALK7 ligand activin B ( Wu et al . , 2014 ) . Moreover , the arcuate nucleus of Alk7 knock-out mice shows reduced expression of Npy mRNA and lower numbers of Npy-expressing neurons compared to wild type controls ( Sandoval-Guzmán et al . , 2012 ) . It has therefore been unclear whether ALK7 affects fat accumulation cell-autonomously in adipose tissue or through other sites , and whether its effects on adult physiology are developmental or homeostatic , via acute regulation of adult cell function . In this study , we developed a conditional knock-out mouse lacking ALK7 in adipose tissue and a knock-in mouse model carrying an analogue-sensitive kinase allele ( ASKA ) of ALK7 , which can be specifically inhibited by administration of ATP competitive inhibitors . Using these animals , as well as cell culture models , we have established that ALK7 functions cell-autonomously and acutely in adult adipocytes to control energy expenditure and fat accumulation by suppressing adipocyte mitochondrial biogenesis , fatty acid oxidation , and β-AR mediated-lipolysis . Importantly , we found that ALK7 signaling negatively regulates adipocyte β-AR expression and β-adrenergic signaling during a high fat diet , providing a link between nutrient overload and catecholamine resistance in adipose tissue .
In order to dissect the cell-autonomous functions of ALK7 in specific tissues , we generated a conditional knock-out allele of the mouse Alk7 gene ( also known as Acvr1c ) with loxP sites flanking exons 5 and 6 , encoding essential regions of the ALK7 kinase domain ( Figure 1—figure supplement 1 ) . Gene deletion in adipose tissue was achieved by crossing Alk7fx mice with Ap2CRE mice ( He et al . , 2003 ) . Although Ap2CRE has also been reported to be expressed in adipose tissue macrophages ( Lee et al . , 2013 ) , Alk7 mRNA expression could only be detected in the adipocyte fraction of adipose tissue but not in the stromal-vascular fraction ( containing macrophages ) or in spleen ( Figure 1—figure supplement 2A–D ) . Expression of Alk7 mRNA was reduced by 60% in the adipose tissue of Alk7fx/fx::Ap2CRE mice , while a 98% reduction was achieved in Alk7fx/−::Ap2CRE mice ( i . e . , compound heterozygotes carrying floxed and knock-out Alk7 alleles ) ( Figure 1—figure supplement 3A , B ) . No change in Alk7 mRNA expression was observed in the pancreas or brain ( Figure 1—figure supplement 2B ) . Both lines of fat-specific Alk7 mutant mice showed significantly reduced weight gain during 12 weeks on a high fat diet compared to controls ( Figure 1A , B ) . In contrast , weight gain in Alk7fx/fx::NestinCRE mice , lacking ALK7 expression in the nervous system ( Figure 1—figure supplement 3C ) , did not differ from controls ( Figure 1C ) . Diet-induced fat accumulation , as measured by epididymal and retroperitoneal fat depot weight ( Figure 1D–F ) , magnetic resonance imaging ( MRI ) of total fat mass ( Figure 1G , H ) , and adipocyte cell size ( Figure 1I , J ) , was also significantly reduced in fat-specific Alk7 mutant mice compared to controls . In contrast , fat depots of nervous system-specific Alk7 mutant mice were not different from controls ( Figure 1K ) . In agreement with reduced diet-induced obesity , serum leptin levels were also lower after a high fat diet in both global and fat-specific Alk7 knock-out mice ( Figure 2A , B ) . However , fed serum insulin levels remained unchanged in fat-specific and brain-specific Alk7 knock-out mice ( Figure 2C , D ) , suggesting unaltered peripheral insulin sensitivity . In addition , glucose and insulin tolerance tests performed in fat-specific Alk7 mutant mice and controls indicated normal glucose and insulin responses in the mutants ( Figure 2E–H ) . Obesity has been associated with a state of inflammation in adipose tissue in which resident macrophages play important roles ( Hotamisligil , 2006; Fujisaka et al . , 2009 ) . Following 8 weeks of a high fat diet , adipose tissue of global and fat-specific Alk7 knock-out mice showed decreased expression of markers of pro-inflammatory M1 macrophages , such as TNF1α , IL-12b , and Itgax ( Figure 2I , J ) , but increased expression of Mgl2 , a marker of M2 macrophages , the major resident macrophages involved in remodeling and repair that are normally present in adipose tissue from lean mice ( Figure 2K , L ) . This profile is in agreement with reduced adipose tissue inflammation and protection against diet-induced obesity . Together , these results indicate that ALK7 functions cell-autonomously in adipose tissue to regulate fat accumulation during nutrient overload . 10 . 7554/eLife . 03245 . 003Figure 1 . Conditional deletion of ALK7 in adipose tissue attenuates weight gain and fat deposition under a high fat diet . ( A–C ) Weight gain under chow ( squares ) and a high fat diet ( HFD , triangles ) in fat-specific Alk7fx/fx::Ap2CRE ( A ) and Alk7fx/−::Ap2CRE ( B ) , and brain-specific Alk7fx/fx::NestinCRE ( C ) knock-out mice ( solid symbols ) compared to control mice ( open symbols ) . Arrows denote the first week of HFD . N = 8 mice per group in all cases , except N = 6 in ( C ) on a chow diet . ( D–F ) Weights of epididymal ( Epi ) and retroperitoneal ( Retro ) fat depots in global Alk7−/− ( D ) and fat-specific Alk7fx/fx::Ap2CRE ( E ) and Alk7fx/−::Ap2CRE ( F ) knock-out mice after 16 weeks on HFD . N = 6 mice per group . ( G and H ) Fat and lean mass assessed by magnetic resonance imaging ( MRI ) in global Alk7−/− ( G ) and fat-specific Alk7fx/−::Ap2CRE ( H ) knock-out mice after 16 weeks on HFD . N = 8 mice per group in ( G ) , N = 5 in ( H ) . ( I and J ) Adipocyte cell size in fat-specific Alk7fx/−::Ap2CRE knock-out mice and Alk7fx/− controls after chow or HFD as visualized by hematoxylin-eosin staining in tissue sections of epididymal adipose tissue ( I ) . Quantitative analysis is shown in ( J ) . Small , 400–5000 μm2; Med , 5000–10 , 000 μm2; Large , 10 , 000–60 , 000 μm2 . N = 4 mice per group ( four sections per mouse ) . ( K ) Weights of epididymal ( Epi ) and retroperitoneal ( Retro ) fat depots in brain-specific Alk7fx/fx::NestinCRE knock-out mice after 16 weeks on HFD . N = 6 mice per group . *p < 0 . 05; **p < 0 . 01; NS , non-significant ( mutant vs control ) . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 00310 . 7554/eLife . 03245 . 004Figure 1—figure supplement 1 . Generation of a conditional allele of the mouse Acvr1c gene encoding ALK7 . CRE-mediated recombination deletes exons 5 and 6 , encoding the ALK7 kinase domain , which is essential for signaling , and introduces an in-frame stop codon after exon 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 00410 . 7554/eLife . 03245 . 005Figure 1—figure supplement 2 . Alk7 expression in adipocytes , but not in adipose tissue macrophages . ( A–C ) Relative levels of Alk7 ( A ) , adipocyte marker Adiponectin ( B ) , and macrophage markers Itgax ( C ) and Mgl2 ( D ) mRNA in adipocytes , adipose tissue stromal-vascular fraction ( SVF ) , and spleen assessed by Q-PCR . In each case , results were normalized to adipocyte levels . N = 6 mice per group . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 00510 . 7554/eLife . 03245 . 006Figure 1—figure supplement 3 . Alk7 expression in conditional knock-out mice . ( A–C ) Relative levels of Alk7 mRNA expression assessed by Q-PCR in epididymal fat ( epi ) , brain , and pancreas in fat-specific Alk7fx/fx::Ap2CRE ( A ) and Alk7fx/−::Ap2CRE ( B ) , and brain-specific Alk7fx/fx::NestinCRE ( C ) conditional knock-out mice . N = 5 mice per group in ( A ) , N = 4 in ( B ) and ( C ) . **p < 0 . 01; NS , non-significant ( mutant vs control ) . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 00610 . 7554/eLife . 03245 . 007Figure 2 . Reduced serum leptin levels and adipose tissue inflammation but normal glucose and insulin metabolism in global and fat-specific Alk7 knock-out mice . ( A and B ) Serum levels of leptin in global Alk7−/− ( A ) and fat-specific Alk7fx/fx::Ap2CRE ( B ) knock-out mice after 16 weeks on a high fat diet ( HFD ) . N = 6 mice per group . ( C and D ) Serum levels of insulin in fat-specific Alk7fx/−::Ap2CRE ( C ) and brain-specific Alk7fx/fx::NestinCRE ( D ) knock-out mice after 16 weeks on HFD . N = 6 mice per group . ( E and F ) Blood glucose levels after glucose injection in 4-month-old Alk7fx/−::Ap2CRE mice and controls after chow ( E ) or a high fat diet ( F ) . Mice had been starved overnight prior to glucose injection . N = 5 mice per group . ( G ) Serum insulin levels after glucose injection in 4-month-old Alk7fx/−::Ap2CRE mice and controls kept on a chow diet . Mice had been starved overnight prior to glucose injection . N = 5 mice per group . ( H ) Blood glucose levels after insulin injection in 4-month-old Alk7fx/−::Ap2CRE mice and controls kept on a chow diet . Mice had been starved for 3 hr prior to insulin injection . N = 5 mice per group . ( I–L ) Relative mRNA expression levels of M1 macrophage markers TNF1a , Itgax , and IL12b ( I and J ) and the M2 macrophage marker Mgl2 ( K and L ) assessed by quantitative PCR in epididymal adipose tissue of global Alk7−/− ( I and K ) and fat-specific Alk7fx/fx::Ap2CRE ( J and L ) knock-out mice after 16 weeks on HFD . N = 6 mice per group . *p < 0 . 05; **p < 0 . 01; NS , non-significant ( mutant vs control ) . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 007 The reduced obesity in Alk7 knock-out mice after a high fat diet could be a result of lower calorie intake or higher energy expenditure . Both global knock-out and fat-specific Alk7 mutant mice displayed increased energy expenditure ( Figure 3A , B ) and oxygen consumption ( Figure 3C , D ) after a high fat diet compared to controls . Food intake remained unchanged in the mutant mice ( Figure 3E ) . Changes in energy expenditure in Alk7 mutant mice were not due to ‘browning’ of subcutaneous adipose tissue , as expression of brown adipose tissue ( BAT ) marker genes Ucp1 and Elovl3 was not increased in the subcutaneous fat of the mutants ( Figure 3—figure supplement 1A , B ) . Moreover , the browning effects of the β3-AR-specific agonist CL316243 were comparable in subcutaneous adipose tissue of wild type and Alk7 knock-out mice ( Figure 3—figure supplement 1C , D ) . Neither was expression of BAT markers elevated in the BAT of Alk7 mutant mice ( data not shown ) . Global and fat-specific Alk7 knock-out mice showed higher physical activity than wild type controls after a high fat diet ( Figure 3F , G ) . However , it was recently reported that changes in activity do not drive changes in energy expenditure in groups of mice below thermoneutrality ( Virtue et al . , 2012 ) . We hypothesized that increased energy expenditure in Alk7 mutant mice on a high fat diet may be due to higher basal metabolic rate , and investigated adipose tissue mitochondria biogenesis and function , which are impaired by chronic nutrient overload in rodents and humans ( Heilbronn et al . , 2007; Rong et al . , 2007; Sutherland et al . , 2008 ) . Mitochondria biogenesis , as measured by mitochondrial DNA content ( Figure 4A , B ) , citrate synthase activity ( Figure 4C , D ) , and ATP content ( Figure 4E , F ) , was significantly increased in adipose tissue of both global and fat-specific Alk7 knock-out mice on a high fat diet compared to controls . In addition , several markers of mitochondrial biogenesis and function were also significantly upregulated in the mutants , including PGC-1α , a master regulator of mitochondrial biogenesis ( Wu et al . , 1999 ) , mitochondrial uncoupling protein 3 ( UCP-3 ) , cytochrome C , and Hadhb , a key mitochondrial enzyme for β-oxidation of long chain fatty acids ( Middleton , 1994 ) ( Figure 4G , H ) . In line with these changes , fatty acid oxidation activity was enhanced in the adipose tissue of global and fat-specific Alk7 mutant mice ( Figure 4I , J ) . These results suggest that resistance to diet-induced obesity in Alk7 mutant mice is due to increased energy expenditure and enhanced mitochondrial function and lipid oxidation in adipose tissue . 10 . 7554/eLife . 03245 . 008Figure 3 . Increased energy expenditure and oxygen consumption in global and fat-specific Alk7 knock-out mice on a high fat diet . ( A and B ) Energy expenditure assessed by calorimetric measurements in global Alk7−/− ( A ) and fat-specific Alk7fx/−::Ap2CRE ( B ) knock-out mice after 16 weeks on a high fat diet ( HFD ) . N = 8 mice per group in ( A ) , N = 6 in ( B ) . ( C and D ) Oxygen consumption assessed by calorimetric measurements in global Alk7−/− ( C ) and fat-specific Alk7fx/−::Ap2CRE ( D ) knock-out mice after 16 weeks on HFD . N = 8 mice per group in ( C ) , N = 6 in ( D ) . ( E ) Daily food intake in fat-specific Alk7fx/−::Ap2CRE knock-out mice during 16 weeks on HFD . N = 6 mice per group . ( F and G ) Physical activity assessed as ambulation in global Alk7−/− ( F ) and fat-specific Alk7fx/−::Ap2CRE ( G ) knock-out mice after 16 weeks on HFD . N = 8 mice per group in ( F ) , N = 6 in ( G ) . *p < 0 . 05; **p < 0 . 01; NS , non-significant ( mutant vs control ) . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 00810 . 7554/eLife . 03245 . 009Figure 3—figure supplement 1 . Alk7 deletion does not result in enhanced 'browning' of subcutanous adipose tissue . ( A and B ) Relative mRNA expression of brown adipose tissue ( BAT ) markers Ucp1 and Elovl3 in subcutaneous adipose tissue of Alk7fx/fx::Ap2CRE fat-specific knock-out mice ( A ) , global Alk7 knock-out mice ( B ) , and corresponding controls . N = 4 mice per group in ( A ) , N = 5 in ( B ) . ( C and D ) Relative mRNA expression of brown adipose tissue ( BAT ) markers Ucp1 ( C ) and Elovl3 ( D ) in subcutaneous adipose tissue of wild type and global Alk7 knock-out mice following injection of the β3-AR-specific agonist CL316243 or vehicle as indicated . N = 4 mice per group . *p < 0 . 05; **p < 0 . 01; ***p < 0 . 001; NS , non-significant ( mutant vs control ) . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 00910 . 7554/eLife . 03245 . 010Figure 4 . Elevated adipose tissue mitochondrial biogenesis and activity in fat-specific Alk7 knock-out mice on a high fat diet . ( A–F ) Mitochondrial biogenesis assessed by measurements of mitochondrial ( mito ) DNA content ( A and B ) , citrate synthase activity ( C and D ) , and ATP content ( E and F ) in epididymal adipose tissue of global Alk7−/− ( A , C , E ) and fat-specific Alk7fx/−::Ap2CRE ( B , D , F ) knock-out mice after 16 weeks on a high fat diet ( HFD ) . N = 8 mice per group in all cases . ( G and H ) Relative mRNA expression of markers of mitochondrial biogenesis and function PGC1a , Hadhb , UCP3 , and cytochrome C assessed by quantitative PCR ( Q-PCR ) in epididymal adipose tissue of global Alk7−/− ( G ) and fat-specific Alk7fx/−::Ap2CRE ( H ) knock-out mice after 16 weeks on HFD . N = 6 mice per group in all cases . ( I and J ) Lipid oxidation in epididymal adipose tissue of global Alk7−/− ( I ) and fat-specific Alk7fx/−::Ap2CRE ( J ) knock-out mice . N = 3 mice per group in all cases . *p < 0 . 05; **p < 0 . 01; NS , non-significant ( mutant vs control ) . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 010 Activation of β-ARs by catecholamines is the major regulatory pathway of fat mobilization during starvation and exercise . Mice lacking all three types of β-ARs are massively obese on a high fat diet without an increase in food intake ( Bachman et al . , 2002 ) . Conversely , β-agonist treatment induces mitochondria biogenesis in adipose tissue and reduces fat mass ( Ghorbani et al . , 1997 ) , resembling the phenotype observed in Alk7 mutant mice . β-AR-dependent fat mobilization is severely impaired during a high fat diet , allowing fat accumulation in adipose tissue , but the underlying mechanisms of catecholamine resistance have not been clarified ( Reynisdottir et al . , 1994; Jocken et al . , 2008 ) . We investigated whether ALK7 could play a role in the catecholamine sensitivity of adipose tissue under a low caloric diet and after nutrient overload . Under a chow diet , injection of the β3-AR-specific agonist CL316243 enhanced lipolysis , as measured by free fatty acid release , significantly more in fat-specific Alk7 knock-out mice than in wild type controls ( Figure 5A ) . After 16 weeks on a high fat diet , global and fat-specific Alk7 knock-out mice also displayed elevated lipolysis compared to controls under both basal conditions and following β3-AR stimulation ( Figure 5B , C ) . On the other hand , we detected no change in basal or agonist-stimulated lipolysis in brain-specific Alk7 knock-out mice ( Figure 5D ) . As levels of epinephrine and norepinephrine were not significantly altered in the adipose tissue of Alk7 knock-out mice ( Figure 5—figure supplement 1 ) , these data suggested enhanced catecholamine sensitivity in adipose tissue lacking ALK7 . In agreement with this , the levels of Adrb1 and Adrb3 mRNAs were significantly protected in Alk7 knock-out mice after 16 weeks on a high fat diet , but were barely detectable in the adipose tissue of control mice ( Figure 5E , F ) . In addition , expression of the negative regulator of adrenergic signaling Rgs2 , an inhibitor of adenylate cyclase , was elevated after a high fat diet in the adipose tissue of wild type mice but not in Alk7 knock-out mice ( Figure 5G ) or fat-specific Alk7 knock-out mice ( Figure 5H , I ) . Adrb1 and Adrb3 mRNA levels were also significantly higher in fat-specific Alk7 knock-outs compared to controls after a high fat diet ( Figure 5H , I ) . In line with enhanced adrenergic signaling , we observed increased levels of total and phosphorylated hormone-sensitive lipase ( HSL ) , a major target of the β-adrenergic pathway in adipocytes , and phosphorylated PKA substrates in the adipose tissue of Alk7 knock-out mice after 16 weeks on a high fat diet compared to wild type controls ( Figure 5J , K ) . Moreover , norepinephrine injection induced a more pronounced increase in PKA activity in the adipose tissue of Alk7 knock-out mice than in control mice ( Figure 5L ) . Together , these data suggest that cell-autonomous ALK7 signaling in adipose tissue suppresses catecholamine sensitivity and β-adrenergic signaling under a high fat diet . 10 . 7554/eLife . 03245 . 011Figure 5 . Enhanced catecholamine sensitivity and β-adrenergic signaling in adipose tissue of global and fat-specific Alk7 knock-out mice on a high-fat diet . ( A ) Basal and CL316243 ( CL ) -stimulated serum free fatty acids ( FFA ) in fat-specific Alk7fx/−::Ap2CRE knock-out mice on a chow diet . N = 6 mice per group . ( B–D ) Basal and CL316243-stimulated lipolysis in global Alk7−/− ( B ) , fat-specific Alk7fx/fx::Ap2CRE ( C ) , and brain-specific Alk7fx/fx::NestinCRE ( D ) knock-out mice after 16 weeks on a high fat diet ( HFD ) . N = 6 mice per group . ( E–G ) Relative mRNA expression of Adrb1 ( E ) , Adrb3 ( F ) , and Rgs2 ( G ) assessed by Q-PCR in epididymal adipose tissue of global Alk7−/− knock-out mice on chow and after 1 week ( wk ) or 2 months ( mo ) on HFD . N = 6 mice per group . ( H and I ) Relative mRNA expression of Adrb1 , Adrb3 , and Rgs2 assessed by Q-PCR in epididymal adipose tissue of fat-specific Alk7fx/fx::Ap2CRE ( H ) and Alk7fx/−::Ap2CRE ( I ) knock-out mice after 16 weeks on HFD . N = 6 mice per group . ( J and K ) Levels of Adrb3 , phospho-HSL , total HSL , and phosphorylated PKA substrates assessed by Western blotting in epididymal adipose tissue of global Alk7−/− ( J ) and fat-specific Alk7fx/fx::Ap2CRE ( K ) knock-out mice after 16 weeks on HFD . The results shown are representative of four biological replicates . ( L ) Levels of phosphorylated PKA substrates were assessed by Western blotting in epididymal adipose tissue from 2-month-old global Alk7−/− knock-out mice on a chow diet following injection of norepinephrine ( NE ) or vehicle ( PBS ) . Results show the levels of phosphorylated PKA substrates quantified by image analysis after normalization to β-actin . N = 3 mice per group . *p < 0 . 05; **p < 0 . 01; NS , non-significant ( mutant vs control ) . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 01110 . 7554/eLife . 03245 . 012Figure 5—figure supplement 1 . Epinephrine and norepinephrine levels in adipose tissue of Alk7 knock-out mice after 16 weeks on a high fat diet . Levels of epinephrine ( A ) and norepinephrine ( B ) in epididymal fat of Alk7−/− mice and wild type controls following 16 weeks on a high fat diet . N = 6 mice per group . NS , non-significant . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 012 The enhanced catecholamine sensitivity of adipose tissue in Alk7 mutant mice prompted us to investigate the acute effects of ALK7 signaling on β-AR expression and adrenergic signaling in cultured adipocyte cells derived by differentiation from mouse embryonic fibroblasts ( MEFs ) . Stimulation of wild type adipocytes with the ALK7 ligand activin B reduced expression of Adrb2 and Adrb3 mRNAs as well as mRNA for the β-adrenergic target gene Hsl ( Figure 6A ) . Activin B had no effect on adipocytes derived from Alk7 knock-out MEFs ( Figure 6B ) , indicating that its effects on adipocyte gene expression were mediated by ALK7 . Activin B also downregulated expression of mRNA encoding PPARγ , a master regulator of adipogenesis ( Figure 6A ) . However , the effects of activin B on adipocyte Adrb mRNA expression are likely to be independent of PPARγ , since inhibition of PPARγ had no significant effect on Adrb3 mRNA levels and activin B could suppress Adrb3 mRNA expression even in the presence of the PPARγ inhibitor or the PPARγ agonist rosiglitazone ( Figure 6—figure supplement 1 ) . Conversely , the related ligand activin A , which does not stimulate ALK7 signaling , enhanced adipocyte PPARγ mRNA expression but had no significant effect on Adrb2 , Adrb3 , and Hsl mRNA levels in wild type adipocytes ( Figure 6C ) . Expression of Adrb2 , Adrb3 , and Hsl mRNAs was also significantly reduced in Alk7 knock-out adipocytes following ALK7 adenovirus-mediated overexpression ( Figure 6D ) , which leads to ligand-independent receptor activation , showing that ALK7 is sufficient to suppress β-AR expression and signaling in adipocytes . In addition , activin B antagonized the stimulatory effects of the β3-AR agonist on the phosphorylation of HSL , perilipin , and PKA substrates in wild type adipocytes ( Figure 6E , F ) . In line with these findings , activation of ALK7 signaling by either ligand or adenovirus-mediated overexpression suppressed β-agonist induced lipolysis in wild type adipocytes ( Figure 6G , H ) . Interestingly , both activin B and adenovirus-mediated ALK7 overexpression also suppressed β-agonist stimulated lipolysis in human adipocytes ( Figure 6I , J ) . Together , these results indicate that acute activation of ALK7 is sufficient to suppress β-adrenergic signaling in both mouse and human adipocytes . 10 . 7554/eLife . 03245 . 013Figure 6 . ALK7 signaling negatively regulates catecholamine sensitivity and β-adrenergic signaling in mouse and human adipocytes . ( A ) Relative mRNA expression of Adrb2 , Adrb3 , Hsl , and PPARg assessed by Q-PCR in adipocytes derived from wild type ( WT ) mouse embryonic fibroblasts ( MEFs ) following stimulation with activin B . N = 4 wells per condition . ( B ) Relative mRNA expression of Adrb2 , Adbr3 , PPARg , and Hsl assessed by Q-PCR in adipocytes derived from Alk7 knock-out MEFs after stimulation with activin B . N = 4 wells per condition . ( C ) Relative mRNA expression of Adrb2 , Adbr3 , PPARg , and Hsl assessed by Q-PCR in adipocytes derived from WT MEFs following stimulation with activin A . N = 4 wells per condition . ( D ) Relative mRNA expression of Alk7 , Adrb2 , Adrb3 , and Hsl assessed by Q-PCR in adipocytes derived from WT MEFs following adenovirus-mediated ALK7 overexpression . N = 4 wells per condition . ( E ) Levels of phospho-HSL and phospho-perilipin assessed by Western blotting in MEF-derived adipocytes after stimulation with β3-AR-specific agonist CL316243 ( CL ) or vehicle ( veh ) in the presence or absence of activin B ( Act B ) . Independent biological duplicates are shown . ( F ) Assay of PKA activity in lysates of MEF-derived adipocytes after stimulation with β3-AR-specific agonist CL316243 ( CL ) or vehicle ( veh ) in the presence or absence of activin B ( Act B ) . Independent biological duplicates are shown . ( G and H ) Basal and CL316243-stimulated lipolysis in MEF-derived adipocytes following stimulation with activin B ( G ) or adenovirus-mediated ALK7 overexpression ( H ) . FFA: free fatty acids . N = 4 wells per condition . ( I and J ) Basal and isoproterenol ( ISO ) -stimulated lipolysis in human adipocytes following stimulation with activin B ( I ) or adenovirus-mediated ALK7 overexpression ( J ) . N = 4 wells per condition . *p < 0 . 05; **p < 0 . 01; NS , non-significant ( activin treatment vs vehicle or Adeno-Alk7 vs Adeno-LacZ , as indicated ) . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 01310 . 7554/eLife . 03245 . 014Figure 6—figure supplement 1 . The effects of activin B on adipocyte Adrb mRNA expression are independent of PPARγ . ( A ) Relative mRNA expression of Adrb3 assessed by Q-PCR in adipocytes derived from wild type ( WT ) mouse embryonic fibroblasts ( MEFs ) following stimulation with PPARγ antagonist T0070907 in the presence and absence of activin B . N = 4 wells per condition . ( B ) Relative mRNA expression of Adrb3 assessed by Q-PCR in adipocytes derived from WT MEFs following stimulation with activin B in the presence or absence of PPARγ agonist rosiglitazone . N = 4 wells per condition . **p < 0 . 01; NS , non-significant ( activin B vs vehicle ) . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 014 Finally , we sought to determine whether acute blockade of ALK7 signaling in adult mice , by-passing possible developmental effects , could also enhance catecholamine sensitivity in adipose tissue and ameliorate diet-induced obesity . To this end , we devised a chemical-genetic approach to acutely inhibit the ALK7 kinase in adult mice using synthetic ATP analogues ( Bishop et al . , 1998 , 2001 ) . An analogue-sensitive kinase allele of Alk7 ( termed Alk7ASKA ) was engineered by mutation of two residues in the active site of the ALK7 kinase . Substitution of the ‘gatekeeper’ residue Ser270 with Gly creates an extra pocket in the ALK7 active site , which can be further expanded by mutating the adjacent residue Leu250 to Val . As assessed in transfected cells , the mutations had no effect per se on ligand-mediated signaling , but rendered the mutant receptor sensitive to inhibition by ATP competitive inhibitors , such as 2NaPP1 ( Figure 7—figure supplement 1 ) . 2NaPP1 treatment did not affect signaling by the wild type ALK7 receptor ( Figure 7—figure supplement 1 ) . Knock-in mice were generated carrying the two mutations in the Alk7 locus ( Figure 7—figure supplement 2 ) . Alk7ASKA mice developed normally , and displayed normal growth and normal fasting insulinemia and glycemia ( data not shown ) . Activin B signaling was suppressed by 1NaPP1 ( an isomer of 2NaPP1 with established in vivo stability [Savitt et al . , 2012] ) in adipocytes derived from Alk7ASKA homozygote mice but not from wild type controls ( Figure 7A , B ) . Residual activin B activity in Alk7ASKA adipocytes treated with 1NaPP1 was likely due to expression of the related receptor ALK4 , which can also respond to activin B by activating Smad3 . We subjected cohorts of 2-month-old Alk7ASKA homozygote and wild type mice to a high fat diet together with twice daily injection of 30 mg/kg 1NaPP1 or vehicle , and monitored their weight daily during a period of 2 weeks . Alk7ASKA and wild type mice gained weight at the same rate on a high fat diet , increasing by approximately 4 g by the end of the second week ( Figure 7C , D ) . There was no weight gain on a chow diet for either genotype ( Figure 7C , D ) . Injection of 1NaPP1 significantly reduced the weight gain rate in Alk7ASKA mice , which on average increased by only 1 . 6 g after 2 weeks on a high fat diet ( Figure 7C ) . 1NaPP1 had no effect on wild type mice ( Figure 7D ) . These results indicate that acute disruption of ALK7 signaling can protect adult mice from diet-induced obesity . Analogue treatment significantly reduced fat accumulation in epididymal and retroperitoneal adipose tissue depots in Alk7ASKA but not wild type mice ( Figure 7E , F ) . Adipocyte cell size was also significantly reduced in Alk7ASKA mice treated with 1NaPP1 compared to vehicle ( Figure 7G , H ) . We also observed enhanced induction of lipolysis by the β3-AR-specific agonist CL316243 in adipose tissue biopsies extracted from Alk7ASKA mice that had been treated with 1NaPP1 ( Figure 8A ) . Moreover , treatment with 1NaPP1 increased the levels of Adrb3 mRNA in epididymal and retroperitoneal adipose tissue of Alk7ASKA mice on a high fat diet ( Figure 8B ) . In line with enhanced catecholamine sensitivity , 1NaPP1 treatment attenuated the suppressive effects of a high fat diet on adrenergic signaling in adipose tissue of Alk7ASKA mice , as assessed by the levels of phosphorylated HSL and PKA substrates ( Figure 8C–E ) . Together , these data indicated that acute disruption of ALK7 signaling in adult mice can uncouple nutrient overload from catecholamine resistance in adipose tissue , resulting in reduced fat accumulation and decreased weight gain on a high fat diet . 10 . 7554/eLife . 03245 . 015Figure 7 . Acute inhibition of ALK7 signaling in adult mice through a chemical-genetic approach reduces diet-induced weight gain and fat accumulation . ( A and B ) Activin B signaling was assessed by p-Smad3 nuclear translocation in mouse embryonic fibroblast ( MEF ) -derived adipocytes from wild type ( WT ) or Alk7ASKA ( Aska ) mice in the presence and absence of ATP competitive inhibitor 1NaPP1 . Nuclear p-Smad3 ( red ) was specifically evaluated by immunohistochemistry in adipocytes , identified by BODIPY 493/503 staining ( green ) . Representative photomicrographs are shown in ( A ) . Scale bar , 50 μm . Quantitative analysis is shown in ( B ) . N = 3 independent experiments each performed in triplicate . ( C and D ) Weight gain on chow ( squares ) and a high fat diet ( HFD , triangles ) in Alk7ASKA ( C ) and wild type ( D ) mice treated with 1NaPP1 ( solid triangles ) or vehicle ( open squares and triangles ) . N = 6 mice per group in ( C ) , N = 9 in ( D ) . ( E and F ) Weights of epididymal ( Epi ) and retroperitoneal ( Retro ) fat depots in Alk7ASKA ( E ) and wild type ( F ) mice after chow ( open bars ) or 2 weeks on HFD treated with 1NaPP1 ( black bars ) or vehicle ( gray bars ) . N = 6 mice per group in ( E ) , N = 7 in ( F ) . ( G and H ) Adipocyte cell size in Alk7ASKA mice after chow or HFD as visualized by hematoxylin-eosin staining in tissue sections of epididymal adipose tissue of Alk7ASKA mice after chow or 2 weeks on HFD treated with 1NaPP1 or vehicle ( veh ) ( G ) . Quantitative analysis is shown in ( H ) . Small , 400–5000 μm2; Med , 5000–10 , 000 μm2; Large , 10 , 000–20 , 000 μm2 . N = 4 mice per group ( four sections per mouse ) . *p < 0 . 05; **p < 0 . 01; NS , non-significant ( 1NaPP1 vs vehicle ) . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 01510 . 7554/eLife . 03245 . 016Figure 7—figure supplement 1 . Validation of Alk7ASKA allele in transfected R4-2 cells . ( A and B ) Luciferase reporter assay ( A ) and Smad-2 phosphorylation ( B ) in response to activin B in R4-2 cells reconstituted with Alk7WT or Alk7ASKA constructs in the presence or absence of 1 μM 2NaPP1 analogue . N = 3 wells per condition ( A ) . Results shown in ( B ) are representative of two independent experiments . **p < 0 . 01; NS , non-significant ( ± 2NaPP1 ) . Error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 01610 . 7554/eLife . 03245 . 017Figure 7—figure supplement 2 . A chemical-genetic approach for acute inactivation of ALK7 in adult mice . Gate-keeper residues Ser270 and Leu250 in the ATP-binding pocket of the ALK7 kinase were replaced into Gly and Val , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 01710 . 7554/eLife . 03245 . 018Figure 8 . Acute inhibition of ALK7 signaling in adult mice reduces diet-induced catecholamine resistance . ( A ) Basal and CL316243 ( CL ) -stimulated lipolysis in adipose tissue biopsies extracted from Alk7ASKA mice after 2 weeks on a high fat diet ( HFD ) treated with 1NaPP1 ( solid bars ) or vehicle ( open bars ) . N = 6 mice per group . Epi , epididymal; Retro , retroperitoneal . ( B ) Relative mRNA expression levels of Adrb3 assessed by Q-PCR in epididymal ( Epi ) adipose tissue of Alk7ASKA mice after chow ( open bars ) or 2 weeks on HFD treated with 1NaPP1 ( black bars ) or vehicle ( gray bars ) . N = 6 mice per group . ( C–E ) Levels of phospho-HSL ( p-HSL ) and phosphorylated PKA ( P-PKA ) substrates assessed by Western blotting in epididymal adipose tissue of Alk7ASKA mice after chow or 2 weeks on HFD treated with 1NaPP1 or vehicle . Histograms show quantification by image analysis of P-HSL levels ( D ) and levels of phosphorylated PKA substrates ( E ) normalized to β-actin levels . N = 4 mice per group . *p < 0 . 05; **p < 0 . 01; NS , non-significant ( 1NaPP1 vs vehicle ) . All error bars show mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03245 . 018
In mice and humans , nutrient overload induces catecholamine resistance in adipose tissue , thereby suppressing lipolysis and fatty acid oxidation , and enhancing fat accumulation . This is an efficient adaptive mechanism for energy storage during times of abundant food supply , enhancing survival upon starvation . In the modern industrialized world , however , the effortless availability of foods with high caloric content , together with reduced physical activity , has resulted in an obesity pandemic . Remarkably , the mechanisms linking nutrient overload to catecholamine resistance in adipose tissue are not well understood . Clarifying these mechanisms is important , as selective enhancement of catecholamine sensitivity and β-adrenergic signaling in adipose tissue could be useful in the treatment of obesity . Obesity is thought to induce catecholamine resistance in adipose tissue by downregulating β-AR expression and interfering with downstream β-adrenergic signaling pathways , such as PKA activity ( Reynisdottir et al . , 1994; Arner , 1999; Jocken et al . , 2008 ) . Our results demonstrate that ALK7 signaling contributes to catecholamine resistance by limiting both β-AR expression and signaling in adipocytes during a high fat diet . Signaling pathways are known to be regulated by negative feedback loops that limit intracellular responses upon sustained stimulation . Adrenergic signaling in adipose tissue self-attenuates under a high fat diet through PKA-mediated upregulation of Rgs2 ( Song et al . , 2010 ) . Interestingly , a mutation in the Rgs2 promoter that increases Rgs2 expression has been shown to enhance susceptibility to metabolic syndrome in humans ( Freson et al . , 2007 ) . However , it is unclear how elevated levels of Rgs2 can be sustained in adipose tissue during persistent catecholamine resistance . In our studies , disruption of ALK7 in adipose tissue largely prevented the upregulation of Rgs2 under a high fat diet , suggesting that ALK7 signaling may contribute to the maintenance of diet-induced Rgs2 expression , possibly by cooperating with intermediate regulators , such as Crtc3 ( Song et al . , 2010 ) . Our studies in fat-specific Alk7 knock-out mice and Alk7ASKA knock-in mice demonstrate that ALK7 functions cell-autonomously and homeostatically in adult adipocytes to regulate catecholamine sensitivity in response to the diet . Adipocytes lacking ALK7 retain significant levels of β-AR expression and adrenergic signaling , allowing sustained lipid catabolism and increased energy expenditure . Although such a short-circuit would clearly be maladaptive upon starvation , it offers a therapeutic opportunity for the treatment of obesity . The fact that activin B and adenovirus-mediated ALK7 overexpression suppressed β-agonist stimulated lipolysis in human adipocytes suggests functional conservation of the ALK7 signaling pathway in human diet-induced obesity . A high fat diet and obesity increase adipocyte expression of the two main ALK7 ligands activin B and GDF-3 ( Witthuhn and Bernlohr , 2001; Sjöholm et al . , 2006; Hoggard et al . , 2009 ) , suggesting enhanced ALK7 signaling in obesity . In line with this , diet-induced obesity elevates the level of activated phospho-Smad3 in mouse adipose tissue ( Yadav et al . , 2011 ) . Mouse and human Alk7 mRNA expression is highest in adipose tissue , but undetectable in heart , liver or muscle ( Kang and Reddi , 1996; Rydén et al . , 1996; Carlsson et al . , 2009; Murakami et al . , 2012 ) . Thus , modulation of ALK7 signaling may offer an alternative approach to regulate catecholamine sensitivity and β-adrenergic signaling specifically and selectively in adipose tissue . We propose that ALK7 represents a novel link between nutrient overload and catecholamine resistance in adipose tissue , and suggest that strategies to suppress ALK7 signaling may be beneficial to combat human obesity .
Mice were housed under a 12 hr light–dark cycle , and fed a standard chow diet or a high fat diet ( HFD , 60% of calorie from fat; ResearchDiet ) . The following transgenic mouse lines were used for experiments: Alk7 knock-out ( Jörnvall et al . , 2004 ) , Ap2CRE ( He et al . , 2003 ) , NestinCRE ( Tronche et al . , 1999 ) , Alk7fx conditional knock-out ( this study ) , and Alk7ASKA knock-in ( this study ) . The Alk7fx conditional knock-out allele was generated by inserting loxP sites flanking exons 5 and 6 ( Figure 1—figure supplement 1 ) . Fat-specific knock-out lines were generated by crossing Alk7fx or Alk7fx/ to Ap2CRE . Brain-specific knock-out lines were generated by crossing Alk7fx to NestinCRE . The Alk7ASKA knock-in allele was generated by introducing L250V and S270G mutations into the Alk7 gene exons 4 and 5 , respectively ( Figure 7—figure supplement 2 ) . Targeting vectors were generated using BAC clones from the C57BL/6J RPCIB-731 BAC library and transfected into the TaconicArtemis C57BL/6N Tac ES cell line . Gene-targeted mice were generated at TaconicArtemis ( Germany ) by standard methods . Animal experiments were approved by Stockholm North Ethical Committee for Animal Research . Fat and lean mass were measured using a body composition analyzer ( EchoMRI Medical System , Houston , TX , USA ) . Indirect calorimetry , food intake , and locomotor activity were determined using a comprehensive laboratory animal monitoring system ( Columbus Instruments , Columbus , OH , USA ) as previously described ( Chibalin et al . , 2008 ) . Mice were housed individually with ad libitum access to a high fat diet and water . Mice were acclimatized to the metabolic cages for 24 hr prior to a 24 hr period of automated recordings . Oxygen consumption ( VO2 ) and CO2 production were determined by an open-circuit Oxymax . Accumulated VO2 was presented in l/kg/day and energy expenditure was reported as heat production divided by kilograms of body weight ( kcal/kg/day ) . Ambulatory locomotor activity ( XAMB ) was measured by consecutive beam breaks in adjacent beams during a 24 hr period and presented as counts/min . Alk7ASKA and wild type mice ( 8 weeks old ) kept on a chow diet were individually caged for 1 week , then switched to HFD or kept on a chow diet . 1NAPP1 was freshly formulated in PEG400 at a concentration of 12 . 5 mg/ml before use . Twice daily ( 9:30 AM and 5:30 PM , respectively ) mice on HFD received subcutaneous injections of vehicle or 1NAPP1 ( 50 mg/kg/day ) . Glucose tolerance tests were performed on animals kept on chow or a high fat diet after overnight fasting . Blood glucose was measured by tail tip bleeding using a glucometer ( Accutrend; Roche , Sweden ) at the indicated time points before and after intraperitoneal injection of 2 g/kg ( body weight ) glucose . Serum insulin was measured from tail blood with an ultrasensitive mouse insulin ELISA kit ( Mercodia , Sweden ) . For the insulin tolerance test , mice were fasted for 3 hr , then injected intraperitoneally with 1 . 5 IU/kg recombinant human insulin ( Humulin R; Lilly , Sweden ) . Glucose levels were determined at the indicated time points as above . Fresh tissue samples were fixed in 10% formalin , dehydrated , embedded in paraffin , sectioned , and stained with hematoxylin and eosin . Photographs of hematoxylin and eosin stained cross-section slides were taken with a light microscope . Areas of individual adipocytes were measured with ImageJ software ( National Institutes of Health ) and used for quantification of cell size . For in vivo lipolysis , mouse tail blood was taken from non-fasted mice for measurement of serum free fatty acids before and after intraperitoneal injection of the β3-AR-specific agonist CL316243 ( Sigma-Aldrich , Sweden ) ( 100 μg/kg body weight ) at the indicated times ( 20 min or 40 min ) . For ex vivo lipolysis , epididymal fat pieces weighing about 20 mg were incubated in Krebs–Ringer Bicarbonate Buffer ( KRBH ) containing 1% fatty acid-free BSA ( Sigma-Aldrich ) and glucose ( 2 . 5 mM ) . The samples were treated with either vehicle or CL316243 ( 1 μM ) for 2 hr at 37°C with mild shaking at 150 rpm . After incubation , glycerol release was measured using a free glycerol reagent ( Sigma-Aldrich ) , and normalized to the total amount of protein in adipose tissue samples . For in vitro lipolysis in MEF-derived or human adipocytes , differentiated adipocytes were washed with KRBH and treated with either vehicle , CL316243 ( 1 μM ) , Norepinephrine ( 1 μM ) , or isoproterenol ( 1 μM ) for 3 hr , and glycerol release was measured as before . Primary adipocytes were isolated from epididymal adipose tissue after collagenase II ( 1 mg/ml , Sigma-Aldrich ) digestion , and were incubated with KRBH containing 1% fatty acid-free BSA ( Sigma-Aldrich ) , glucose ( 2 . 5 mM ) , palmitic acid ( 100 μM ) , and 3H palmitic acid ( 0 . 5 μCi/ml ) . For normalization of counts by total protein , an aliquot of adipocytes was snap frozen without BSA and 3H palmitic acid incubation . After incubation for 5 hr at 37°C with shaking at 150 rpm , an equal volume of chloroform was added . After chloroform separation , 100 μl of the aqueous phase was further mixed with 900 μl 10% activated charcoal slurry thoroughly by shaking for 30 min . The mixture was subsequently centrifuged at maximum speed at room temperature for 10 min , and 100–200 μl supernatant was taken for assessment of 3H counts . MEFs were prepared from 13 . 5-day wild type and Alk7 mutant embryos , and differentiated into adipocytes according to a standard protocol ( Zhang et al . , 2009 ) . Briefly , MEF cells were plated at about 50–70% confluence on 0 . 1% gelatin-coated dishes , cultured in DMEM , supplemented with 10% fetal bovine serum , 1 mM sodium pyruvate , MEM Non-Essential Amino Acids Solution ( Invitrogen , Sweden ) , 0 . 5 mM 2-mercaptoethanol , and 100 U/ml penicillin and streptomycin ( MEF medium ) . MEF monolayers were allowed to grow to confluence before initiation of adipogenesis . Adipogenesis was induced by incubation in MEF medium supplemented with 10 μg/ml insulin , 0 . 5 mM IBMX , 1 μM dexamethasone , and 0 . 5 μM rosiglitazone ( adipocyte differentiation medium ) . After 2 days , the medium was changed to MEF medium supplemented with only 10 μg/ml insulin and 0 . 5 μM rosiglitazone ( adipocyte maintenance medium ) , and then changed again every 2 days until full differentiation ( i . e . , day 8–10 ) . Human preadipocytes were purchased from Lonza Biologics and differentiated into adipocytes according to the manufacturer's instructions . For activin B or activin A treatment , cells were incubated with 100 ng/ml activin B or activin A ( R&D Systems , UK ) in adipocyte maintenance medium for 48 hr before lipolysis assay or , alternatively , harvested for extraction of RNA and proteins . PPARγ agonist rosiglitazone was used at 0 . 5 μM; PPARγ antagonist T0070907 at 1 μM . Adenovirus particles were produced and amplified in HEK293 cells as previously described ( Fujii et al . , 1999 ) . Adipocytes were infected with Adeno-ALK7 or Adeno-LacZ at 2 . 0 multiplicity of infection ( MOI ) for 48 hr before lipolysis assay or extraction of RNA . For assessment of nuclear p-Smad3 , MEF-derived adipocytes were incubated for 150 min with 2 μM 1NaPP1 or vehicle ( DMSO ) then treated for 40 min with 20 ng/ml activin B , or left untreated , and fixed with 4% paraformaldehyde . Adipocytes were identified by BODIPY 493/503 staining and p-Smad3 was assessed by immunocytochemistry with antibodies from Epitomics . DAPI was used for nuclear staining . Images were obtained with a Zeiss laser confocal microscope . The intensity of nuclear p-Smad3 was determined using Zen software ( Zeiss , Germany ) in adipocytes from three random fields in three different wells per condition . The number of cells displaying nuclear p-Smad3 intensity above an arbitrary threshold was determined in each field , added up for each well , and averaged across the three wells of each condition . All values were then normalized to the number of responding cells in cultures of wild type adipocytes stimulated with activin B ( set to 100 ) . The experiment was repeated three times and the results were averaged across the three experiments . Serum triglyceride ( Infinity triglyceride reagent; Thermo DMA ) , free fatty acids ( Free Fatty Acids Half Micro Test; Roche ) , leptin ( mouse leptin ELISA; Abcam , UK ) , insulin ( Ultrasensitive mouse insulin ELISA; Mercodia ) , and tissue ATP contents ( ATPLite; PerkinElmer ) were measured with commercial kits according to the manufacturer's instructions . For quantification of mitochondria DNA , total adipose tissue DNA was extracted with a DNeasy kit ( Qiagen , Sweden ) and 1 ng was used for Q-PCR quantification of the copy number for mitochondrial encoded gene cytochrome B . This was normalized to the copy number of the nuclear gene H19 . Citrate synthase activity was measured as described ( Srere et al . , 1963 ) in tissue samples homogenized in RIPA buffer . Epinephrine and norepinephrine were measured in adipose tissue extracts by a commercial ELISA kit according to the manufacturer's instructions ( Labor Diagnostika Nord , Sweden ) . A PKA activity assay was performed with PepTag Non-Radioactive cAMP-Dependent Protein Kinase Assay ( Promega , Sweden ) . For isolation of adipocyte and stromal-vascular fractions ( SVF ) , tissues were dissected from epididymal depots of 8-week-old mice , minced with forceps in PBS , and washed three times with 0 . 1% BSA KRBH . Digestion was carried out for 90–120 min in 2% BSA KRBH containing 0 . 2 mg/ml collagenase II ( Sigma-Aldrich ) with constant agitation . Digested tissue was filtered through 250 μm nylon mesh and centrifuged for 10 min at 1000 rpm ( 200×g ) to separate floating adipocytes . The supernatant and the pellet were further centrifuged for 10 min at 1500 rpm ( 240×g ) . The resulting pellet ( SVF ) and the adipocytes were resuspended in lysis buffer and kept at −80°C . ‘Browning’ of subcutaneous adipose tissue was induced by chronic treatment with CL316243 . A number of 12–16-week-old mice were injected daily intraperitoneally with either 1 mg/kg CL316243 ( Sigma ) or PBS only ( vehicle ) for 7 days . Inguinal subcutaneous adipose tissue was isolated from these mice . Total RNA from tissues or cells was isolated with an RNAeasy kit ( Qiagen ) according to the manufacturer's instructions . Isolated RNA samples were digested with DNase prior to reverse transcription with Superscript II ( Invitrogen ) . Q-PCR of cDNA samples was performed on a ABI StepOne Plus instrument ( Applied Biosystems , Sweden ) using SYBR Green master mix ( Applied Biosystems ) and primers as indicated in Supplementary file 1 . The R4-2 cell line is a derivative of MvLu1 mink lung epithelial cells that expresses low levels of type 1 TGF-β receptors ( Andersson et al . , 2008 ) . R4-2 cells were cultured in 24-well plates in DMEM medium supplemented with 10% serum and antibiotics . They were transfected using Lipofectamine 2000 ( Invitrogen ) with expression plasmids carrying wild type or Aska variants of the rat Alk7 cDNA , the Smad3-dependent luciferase reporter plasmid CAGA-Luc ( Dennler et al . , 1998 ) , and a Renilla plasmid for normalization . At 24 hr after transfection , cells were stimulated with 25 ng/ml activin B together with 1 μM 2NaPP1 ( solubilized in DMSO ) or DMSO only ( vehicle ) . Then 24 hr after stimulation , luciferase activity was analyzed using the Dual-Luciferase Reporter Assay System ( Promega ) in a 1450 Microbeta Jet counter ( Wallac ) . For assays of p-Smad2 by Western blotting , cells were treated 48 hr after transfection with 1 μM 2NaPP1 for 3 hr prior to addition of 25 ng/ml activin B for an additional hour . Monolayers were then processed for Western blotting as described below . Protein extraction and Western blotting were performed according to standard protocols . Briefly , snap-frozen adipose tissue or cell samples were homogenized in ice-cold RIPA buffer ( 50 mM Tris–HCl pH 7 . 4 , 1% NP-40 , 0 . 25% sodium deoxycholate , 150 mM sodium chloride , 1 mM EDTA ) containing 1 mM sodium orthovanadate , 1 mM sodium fluoride , and proteinase inhibitor cocktail ( Roche ) , and centrifuged at 13 , 000×g for 15 min to collect supernatants . Supernatants ( 30 μg protein ) were used for reducing SDS PAGE and Western blotting . Primary antibodies against p-HSL , HSL , p-PKA substrates , p-Perilipin , p-Smad2 , β-actin , and α-tubulin ( Cell Signaling , Sweden ) were used at 1:2000 dilution . The levels of target proteins were quantified by the intensity of Western blot bands using ImageJ software ( National Institutes of Health ) . Statistical significance was determined by using unpaired two-tailed or one-tailed Student's t tests and one-way ANOVA . Differences were considered significant at a p value less than 0 . 05 . Quantified data are presented as mean ± SEM . | Adrenaline and noradrenaline are two hormones that trigger the burst of energy and increase in heart rate and blood pressure that are needed for the ‘fight-or-flight’ response . Both belong to a group of chemicals called catecholamines . These chemicals bind to cells carrying proteins called adrenoceptors on their surface and stimulate the breakdown of fat , which releases energy . However , when nutrients are plentiful , fat cells become resistant to catecholamines and instead store fat so it can be used for energy if food becomes scarce . In the industrialized world where food is easily and constantly accessible , this resistance can cause an unhealthy increase in body fat and result in obesity . Increasing fat metabolism by making fat cells more able to respond to catecholamines is an attractive strategy for combating obesity . Indeed , drugs that mimic the effect of catecholamines on an adrenoceptor found in mice reduce obesity caused by over-eating . However , these drugs are ineffective in humans and can cause harmful side effects to the cardiovascular system , including high blood pressure and an increased heart rate . Devising a strategy that specifically targets catecholamine resistance in fat cells is therefore desirable . A protein called ALK7 is a cell surface receptor that is predominantly found in fat cells and tissues involved in controlling the metabolism . Mice with a mutation in ALK7 that stops this protein from working properly accumulate less fat than mice with a functional version of the protein , but it is not known why . To understand ALK7's involvement in fat metabolism , Guo et al . created mice whose fat cells lack ALK7 , but whose other cells all produce ALK7 as normal . When fed a diet rich in fat , these mice are leaner than regular mice and they burn more energy . The metabolic responses seen in ALK7 mutant mice are very similar to those seen in mice treated with drugs targeting adrenoceptors , suggesting that there may be a link between ALK7 and catecholamine resistance . Indeed , Guo et al . demonstrate that fat cells lacking ALK7 have an increased sensitivity to catecholamines when the mice are on a high fat diet , which decreases the amount of fat the mice accumulate . Conversely , increasing the activity of ALK7 reduces the ability of the cells to respond to catecholamines , and they accumulate more fat . Guo et al . also generated a second line of mice carrying a mutation in ALK7 that does not affect its function , but renders it sensitive to inhibition by a custom-made chemical . When these animals were on a high-fat diet , administering the chemical made the mice leaner , suggesting that inhibiting the ALK7 receptor can prevent obesity in adult animals . Guo et al . also performed experiments in human fat cells , which showed that the ALK7 receptor works in a similar way in human cells as it does in mice . As ALK7 is largely specific for fat cells and is not known to affect the cardiovascular system , drugs that inhibit ALK7 could potentially safely suppress catecholamine resistance and reduce human obesity . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2014 | Adipocyte ALK7 links nutrient overload to catecholamine resistance in obesity |
Life expectancy has risen sharply in the last 50 years . We applied the classic Michaelis–Menten enzyme kinetics to demonstrate a novel mathematical relationship of income to childhood ( aged 0–5 years ) and adult ( aged 15–60 years ) survival . We treat income as a substrate that is catalyzed to increase survival ( from technologies that income buys ) for 180 countries from 1970 and 2007 . Michaelis–Menten kinetics permit estimates of maximal survival and , uniquely , the critical income needed to achieve half of the period-specific maximum . Maximum child and adult survival rose by about 1% per year . Critical incomes fell by half for children , but doubled for men . HIV infection and smoking account for some , but not all , of the rising critical incomes for adult survival . Altering the future cost curve for adult survival will require more widespread use of current interventions , most notably tobacco control , but also research to identify practicable low-cost drugs , diagnostics , and strategies .
The relationship between health and income has been described using a variety of empirical models , generally of the form ( Wagstaff and Van Doorslaer , 2000 ) [1]hi=f ( yi ) ; where f′>0 and f″<0 , where hi is a health indicator ( such as life expectancy or survival rates ) and yi is income , for country or unit i . Preston , originally , characterized the relationship between life expectancy ( e0 ) at birth and gross domestic product per capita ( ‘GDP’ ) as a logistic function in the form[2]e0=a1+e ( b+c×dGDP′ ) , where GDP′ is GDP linearized on a scale of 0 to 1 . Figure 1A is a reproduced plot of the original ‘Preston curve’ , with an upward shift between the years 1960 and 2000 ( Preston , 1975 ) . Although the fit of this model to data is good , the model parameters are harder to interpret . It is suggested that identifying an inflection point ( the point of diminishing returns ) could clarify the analysis of these parameters ( Rogers and Crimmins , 2011 ) , for a generalized logistic function of the form[3]e0=c1+a×e ( −b×GDP ) , where ‘c’ is the maximal life expectancy and the inflection point is at a GDP value of ‘ln ( a ) /b’ . More commonly , however , a log-linear relationship is used as a simpler alternative to the logistic function[4]e0=a+b×ln ( GDP ) . The Preston model explains the majority of the observed variance in life expectancy , but the coefficients ( ‘a’ and ‘b’ ) are not interpretable as no meaning exists for each coefficient nor is a theoretical relationship between them stipulated . The inadequacy of the Preston models to methodologically identify a single point ( such as an inflection point; Rogers and Crimmins , 2011 ) , which describes the changing curvature at lower income levels , has led to greater emphasis on the upward rise in the Preston curve , and less attention is devoted to lateral movements ( along the income axis ) . Nonparametric regressions have been used to identify a ‘hinge’ on the Preston curve ( Deaton and Case , 2009 ) ; however , we know of no known study that has explicitly quantified shifts across income levels . Here , we identify a new construct called ‘critical income’ through the novel application of a mathematical model by Michaelis and Menten ( MM ) to empirically track child and adult mortality at different incomes . The Michaelis–Menten mathematical model first described in 1913 ( Michaelis and Menten , 1913 ) was initially developed to analyze enzyme kinetics . Enzymes are biomacromolecules that act as catalysts , agents that accelerate the rate of a chemical reaction without being consumed in the process . In the absence of these enzymes , some thermodynamically favorable reactions may be kinetically hindered from occurring . For a single reaction , enzyme E binds to a substrate S to form an intermediate complex ES , which is converted into a product P and the original enzyme[5]E+S⇌ES⇀E+P . Figure 1B is a plot of the MM equation for a hypothetical reaction . The classic MM equation describes the dependence of the enzyme velocity ν on substrate concentration [S] . ν asymptotically approaches a maximum value ( νmax ) at high [S] when enzymatic sites are saturated . Km is the half saturation constant—substrate concentration at which ν = 0 . 5νmax . Km is a function of the forward and reverse reaction rate constants , where a lower Km value indicates a more efficient catalyst[6]ν=νmax[S][S]+Km . We extend the application of the Michaelis–Menten kinetics model to describe life expectancy , child , and adult survival . We use the analogy that GDP is a substrate , and health determinants and widespread applications of public health research , treatments and interventions are catalysts that increase health and survival ( Ad Hoc Committee on Health Research Relating to Future Intervention Options , 1996 ) . Indeed , infrastructure ( such as water sanitation and education systems ) , vaccinations ( leading to long-term immunity ) , and public health knowledge can be viewed as catalysts that are not consumed in entirety during the process . [7]e0=e0 , max×GDPGDP+Kinc , where for a given year , the mean life expectancy at birth ( e0 ) in a country is related to its GDP per capita ( per day , at constant 2005 international dollars , adjusted for purchasing power parity and inflation ) . The MM model is characterized by two parameters: the life expectancy of the highest income countries ( e0 , max ) , and critical income ( Kinc ) . We introduce a new parameter called ‘critical income’ as a meaningful construct that can be estimated to assess the relationship between income and mortality . This construct is defined as the level of daily income associated with the achievement of 50% of the maximum life expectancy ( i . e . , GDP per capita at e0 = 0 . 5 × e0 , max ) ; ‘maximum life expectancy’ is empiric , approximating the observed average life expectancy in high-income countries ( Rodgers , 2002 ) . While it is a biochemical convention to report the 50% mark , the critical income value is adaptable to determine higher fractional level of maximum life expectancy . In particular , two , four , and nine times the critical income yields the income required to achieve 66 . 7% , 80% , and 90% of the maximal life expectancy , respectively . The asymptotic leveling of life expectancy at high income seen in the Preston or MM curves is analogous to the saturation of enzymatic sites at high substrate concentration , thus , further increases in GDP leads to only marginal increases in life expectancy . This model establishes a systematic and empirical method for monitoring not only previously documented upward rises in the Preston curve but also potential shifts along the income axis . Whereas an upward rise can increase the maximum achievable survival for all countries , lateral shifts indicate an increase or decrease of income required to achieve this survival . These lateral shifts are particularly significant for more resource-limited countries that have yet to reach asymptotic leveling in the health and income relationship . Age-specific contributions to changes in life expectancy are quantified using child mortality rate as the probability of a child born in a specific year dying before reaching the age of 5 years , referred to as 5q0 , and the gender-specific adult mortality rate , representing the probability ( for a given year ) that an individual who has just turned 15 years will not reach the age of 60 years , referred to as 45q15 . We transform these values to survival rates , where a mortality rate of 5 per 1000 corresponds to a 99 . 5% survival rate . The corresponding MM functions for child survival from age 0 to 5 years ( 5p0 ) and for adults from the age of 15 to 60 years ( 45p15 ) , by gender , are[8]5p0=5p0max×GDPGDP+Kinc;[9]45p15=45p15max×GDPGDP+Kinc , where in a given year , 5p0max and 45p15max are the maximum survival rates in high income countries and Kinc is the critical income associated with that age-specific group and gender . Two large and widespread public health factors that have influenced mortality over the last four decades have been the HIV/AIDS pandemic and smoking , which makes more common most vascular , respiratory , and neoplastic diseases as well as tuberculosis ( Gajalakshmi et al . , 2003; Jha , 2009 ) . We test if adjustment for the marked heterogeneity of HIV prevalence ( between the ages of 15 and 49 years , as proxy for general population infection levels ) and cigarette consumption ( at ages 15 years or older ) make less efficient the relationship between GDP per capita and adult mortality—as indicated by critical income values .
Figure 2A shows the graphical similarity between the logistic , log-linear , and MM model fits of life expectancy for the year 1990 . For the years 1970 to 2007 , there was no significant difference in the coefficient of determination ( R2 ) for the logistic ( M = 0 . 666 , SD = 0 . 064 ) , log-linear ( M = 0 . 645 , SD = 0 . 073 ) , and MM regressions ( M = 0 . 622 , SD = 0 . 083; F ( 2 , 26 ) = 0 . 807 , p=0 . 458 , not statistically significant ) , indicating that the MM model is statistically as valid as the Preston log-linear and logistic models . See Table 1 for the regression coefficient for all years . 10 . 7554/eLife . 00051 . 004Figure 2 . ( A ) The graphical similarity between the logistic , log-linear , and Michaelis–Menten model fits of life expectancy for the year 1990 . ( B ) Preston curve plotted as an enzyme kinetics reaction with coefficients critical income and maximum life expectancy for the years 1970 and 2007 . DOI: http://dx . xoi . org/10 . 7554/eLife . 00051 . 00410 . 7554/eLife . 00051 . 005Table 1 . Maximum life expectancy , critical income , and regression coefficients ( 95% confidence intervals ) for all countries at 5-year intervals from 1970 to 2007DOI: http://dx . doi . org/10 . 7554/eLife . 00051 . 005YearnR2MaxLife expectancy ( LEmax , years ) 5% trimmed mean LE for high-income countriesIncome require for varying levels of LEmaxFull sample95% random sampleCritical income ( Kinc , 50% ) 66 . 70%80%90%Full sample95% random sample19701480 . 53567 . 8 ( 65 . 4–70 . 1 ) 67 . 666 . 71 . 48 ( 1 . 18–1 . 78 ) 1 . 432 . 965 . 9213 . 3219751480 . 57469 . 3 ( 67 . 2–71 . 4 ) 69 . 270 . 31 . 50 ( 1 . 22–1 . 77 ) 1 . 533 . 006 . 0013 . 5019801490 . 66871 . 3 ( 69 . 6–73 . 0 ) 71 . 171 . 81 . 51 ( 1 . 28–1 . 74 ) 1 . 463 . 026 . 0413 . 5919851520 . 71673 . 2 ( 71 . 8–74 . 8 ) 73 . 173 . 21 . 50 ( 1 . 29–1 . 70 ) 1 . 463 . 006 . 0013 . 5019901640 . 73574 . 6 ( 73 . 2–75 . 9 ) 73 . 574 . 31 . 45 ( 1 . 27–1 . 63 ) 1 . 342 . 905 . 8013 . 0519951770 . 67775 . 0 ( 73 . 6–76 . 4 ) 74 . 875 . 41 . 31 ( 1 . 13–1 . 49 ) 1 . 272 . 625 . 2411 . 7920001780 . 6475 . 2 ( 73 . 8–76 . 7 ) 75 . 376 . 21 . 27 ( 1 . 08–1 . 46 ) 1 . 272 . 545 . 0811 . 4320051770 . 53275 . 3 ( 73 . 6–76 . 9 ) 75 . 676 . 51 . 19 ( 0 . 97–1 . 41 ) 1 . 232 . 384 . 7610 . 7120071720 . 52175 . 5 ( 73 . 9–77 . 1 ) 75 . 776 . 41 . 21 ( 0 . 98–1 . 44 ) 1 . 222 . 424 . 8410 . 89Note: All model parameters were found to be significant , p<0 . 0001 . Although the fits of the models were comparable , the parameters of the logistic and log-linear models have less obvious explanatory value . For the logistic function , all inflection points were determined to be negative ( from −$2 . 69 in 1970 to −$8 . 47 in 2007 ) and are impossible income values for any country . This suggests that the logistic function is overly complex and not necessary to model the data . For the log-linear function , the model parameters did not allow for a means to intuitively track lateral movements in the Preston curve . In addition , when using annual income ( rather than daily income ) , the parameter ‘a’ was not statistically significant for any year from 1970 to 2000 . Table 2 is a comparison of all three models for the year 1990 . 10 . 7554/eLife . 00051 . 006Table 2 . Comparison of the logistic adapted Michaelis–Menten and log-linear models for the year 1990DOI: http://dx . doi . org/10 . 7554/eLife . 00051 . 006ModelFormR2ParametersLogisticLE=LEmaxa+e ( −b×GDP ) 0 . 745LEmax = 73 . 6 ( 72 . 1–75 . 1 ) a = 0 . 642 ( 0 . 546–0 . 739 ) b = 0 . 129 ( 0 . 159–0 . 100 ) Inflection point = −3 . 43Adapted Michaelis–MentenLE=LEmax×GDP ( kinc+GDP ) 0 . 735LEmax = 74 . 6 ( 73 . 2–75 . 9 ) kinc = 1 . 50 ( 1 . 29–1 . 70 ) Log-linearLE = a + b × ln ( GDP ) 0 . 731a = 44 . 1 ( 42 . 0–46 . 2 ) b = 7 . 65 ( 6 . 93–8 . 37 ) For each year , the maximum life expectancy approximated the 5% trimmed mean life expectancy observed in countries with annual incomes greater than $12 , 276 ( t ( 8 ) = −1 . 596 , p=0 . 149 , not statistically significant; Table 1 ) . In addition , we performed a sensitivity analysis on the maximum life expectancy and critical income parameters and determined that both were not sensitive to the random removal of 5% of the data ( t ( 8 ) = 1 . 07 , p=0 . 314 , not statistically significant; t ( 8 ) = 1 . 46 , p=0 . 178 , not statistically significant , respectively; Table 1 ) . An upward rise is observed for life expectancy and income from 1970 to 2007 ( Figure 2B ) . The maximal life expectancy rose from 67 . 8 years ( 95% uncertainty interval 65 . 4–70 . 1 years ) in 1970 to 75 . 5 years ( 95% uncertainty interval 73 . 9–77 . 1 years ) in 2007 . This change in maximal life expectancy represents a linear increase of 75 ( 95% uncertainty interval 48–99 years ) days per calendar year ( R2 = 0 . 875 ) over the last 40 years , and is comparable to the life expectancy increase of almost 90 days per calendar year for the 20th century ( Oeppen , 2002 ) . In addition to maximal life expectancy increase , a lower national income is associated with a higher life expectancy now than it was 40 years ago . The critical income ( in constant 2005 international dollars ) needed to achieve half of maximal overall life expectancy declined from $1 . 48 ( $1 . 18–$1 . 78 ) in 1970 to $1 . 21 ( $0 . 98–$1 . 44 ) in 2007 , equivalent in 2007 to the extreme poverty line of $1 . 25 per day; this represents an 18% decrease in critical income to gain almost 4 additional years of life expectancy . Critical income declined linearly at a rate of −$0 . 09 per decade ( −$0 . 06 to −$0 . 13 , R2 = 0 . 839 ) . Table 1 also reports the incomes required to achieve 66 . 7% , 80% , and 90% of the maximal life expectancies . Maximum survival for all age groups rose at statistically the same rate between 1970 and 2007 ( Figure 3A , Table 3 ) . Maximum child survival to the age of 5 years increased from 94 . 5% ( 92 . 7–96 . 2% ) to 98 . 0% ( 97 . 3–98 . 6% ) , an increase of 0 . 8% ( 0 . 3–1 . 3% ) per decade ( R2 = 0 . 665 ) ; maximum adult female survival at ages 15–59 years rose from 85 . 9% ( 84 . 1–87 . 7% ) to 90 . 1% ( 87 . 8–92 . 3% ) , an increase of 1 . 1% ( 0 . 5–1 . 6% ) per decade ( R2 = 0 . 762 ) ; and maximum adult male survival at ages 15–59 years rose from 77 . 7% ( 75 . 6–79 . 7% ) to 82 . 1% ( 79 . 0–85 . 0% ) , an increase of 1 . 1% ( 0 . 6–1 . 7% ) per decade ( R2 = 0 . 749 ) . 10 . 7554/eLife . 00051 . 007Figure 3 . Trends for maximum survival ( A ) and critical income ( B ) for children and adults from 1970 to 2007 . DOI: http://dx . doi . org/10 . 7554/eLife . 00051 . 00710 . 7554/eLife . 00051 . 008Table 3 . Maximum survival , critical income , and regression coefficients ( 95% confidence intervals ) from 1970 to 2005DOI: http://dx . doi . org/10 . 7554/eLife . 00051 . 008YearChildFemaleFemale ( with HIV covariate ) MaleMale ( with HIV covariate ) R2Max%Kinc , $R2Max%Kinc , $R2Max%Kinc , $HIVR2Max%Kinc , $R2Max%Kinc , $HIV19700 . 44494 . 5 ( 92 . 7–96 . 2 ) 0 . 58 ( 0 . 46–0 . 70 ) 0 . 37685 . 9 ( 84 . 1–87 . 7 ) 0 . 57 ( 0 . 43–0 . 70 ) 0 . 25377 . 7 ( 75 . 6–79 . 7 ) 0 . 54 ( 0 . 38–0 . 71 ) 19800 . 56996 . 9 ( 95 . 7–98 . 2 ) 0 . 57 ( 0 . 48–0 . 66 ) 0 . 44888 . 3 ( 86 . 7–89 . 9 ) 0 . 62 ( 0 . 49–0 . 74 ) 0 . 28679 . 5 ( 77 . 5–81 . 6 ) 0 . 65 ( 0 . 46–0 . 83 ) 19900 . 62998 . 3 ( 97 . 4–99 . 2 ) 0 . 48 ( 0 . 42–0 . 55 ) 0 . 48290 . 2 ( 88 . 7–91 . 8 ) 0 . 68 ( 0 . 55–0 . 81 ) 0 . 59590 . 4 ( 88 . 8–91 . 4 ) 0 . 54 ( 0 . 41–0 . 67 ) −1 . 8 ( −1 . 2 to −2 . 4 ) 0 . 38082 . 1 ( 80 . 2–84 . 1 ) 0 . 81 ( 0 . 62–0 . 99 ) 0 . 49881 . 8 ( 79 . 8–83 . 8 ) 0 . 60 ( 0 . 42–0 . 79 ) −2 . 2 ( −1 . 3 to −2 . 8 ) 20000 . 53097 . 9 ( 97 . 2–98 . 7 ) 0 . 30 ( 0 . 25–0 . 34 ) 0 . 37790 . 2 ( 88 . 0–92 . 4 ) 0 . 86 ( 0 . 65–1 . 06 ) 0 . 78792 . 1 ( 90 . 5–93 . 6 ) 0 . 69 ( 0 . 55–0 . 82 ) −1 . 8 ( −1 . 6 to −2 . 0 ) 0 . 28481 . 9 ( 79 . 0–84 . 8 ) 1 . 10 ( 0 . 77–1 . 43 ) 0 . 72083 . 5 ( 81 . 4–85 . 6 ) 0 . 79 ( 0 . 57–1 . 00 ) −2 . 2 ( −1 . 9 to −2 . 5 ) 20050 . 46698 . 0 ( 97 . 3–98 . 6 ) 0 . 25 ( 0 . 20–0 . 29 ) 0 . 32390 . 1 ( 87 . 8–92 . 3 ) 0 . 86 ( 0 . 63–1 . 12 ) 0 . 80392 . 4 ( 90 . 9–93 . 8 ) 0 . 68 ( 0 . 54–0 . 81 ) −2 . 2 ( −1 . 8 to −2 . 5 ) 0 . 25482 . 0 ( 79 . 0–85 . 0 ) 1 . 14 ( 0 . 81–1 . 58 ) 0 . 73984 . 2 ( 82 . 1–86 . 3 ) 0 . 82 ( 0 . 59–1 . 04 ) −2 . 6 ( −3 . 0 to −2 . 2 ) Note: All model parameters were found to be significant , p<0 . 0001 . However , critical incomes diverged dramatically for children and adults ( Figure 3B ) . From 1970 to 1980 , the critical income values for children , adult males , and adult females were statistically equivalent , between $0 . 54–$0 . 58 per day . For children , the critical income values declined gradually , however , with a large drop in 1990–1995 . Over the 40-year period from 1970 to 2010 , the critical income for child survival fell by over half from $0 . 58 ( $0 . 46–$0 . 70 ) to $0 . 24 ( $0 . 20–$0 . 28 ) . In contrast , critical income more than doubled for adult male survival from $0 . 54 ( $0 . 38–$0 . 71 ) to $1 . 20 ( $0 . 81–$1 . 58 ) and rose over 50% for adult female survival from $0 . 57 ( $0 . 43–$0 . 70 ) to $0 . 89 ( $0 . 65–$1 . 12 ) ( see Table 3 for all years and Supplementary file 1 for each country-specific critical income ) . These percentage increases were similar even if critical income was defined differently , for example , income needed to achieve 66 . 7% , 80% , or 90% of maximum survival ( not shown ) . The 1970 and 2007 survival curves for adult men crossed over at a value for $10 . 95 GDP per capita per day . There are 58 countries with a total adult male population of approximately 780 million ( or 35% of the world adult male population ) below this income value . For these countries , adult male survival was lower in 2007 than in 1970 ( Figure 4A ) . For women , the comparable 1970 and 2007 survival curves crossed at a value of $5 . 93 GDP per capita per day , corresponding to 32 countries , with approximately 150 million women ( or 7% of the world adult female population ) , where female survival in 2007 was worse than in 1970 ( Figure 4B ) . By contrast , between 1970 and 2007 , for the 58 countries with an income under $10 . 95 per capita per day in 2007 , child mortality improved by 11 . 8% , adult male survival fell by 2 . 7% , and incomes rose by 8 . 9% ( or an absolute increase of $0 . 98 ) . 10 . 7554/eLife . 00051 . 009Figure 4 . Adult male ( A ) and female ( B ) survival regression curves for the years 1970 and 2007 . Adult survival in 2007 is lower than in 1970 for countries under the income threshold of $10 . 95 ( for men ) and $5 . 93 ( for women ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00051 . 009 Particularly given the marked increase in critical incomes needed to achieve maximal adult male survival , we tested if HIV infection and deaths and smoking ( both , greater in males than in females ) explained the increasing critical income values for adult males ( Figure 5 ) . For adult males in the year 2000 , HIV prevalence ( range 0 . 06–26% aged 15–49 years ) and cigarette consumption ( range 54 . 6–3385 . 2 per year at ages 15 years or older ) were shown to influence critical income . Selecting the 86 countries with complete survival , income , HIV , and cigarette data , the critical income value was $2 . 02 ( $1 . 32–$2 . 73 ) . Adjusting for HIV and smoking prevalence to the average across countries reduced the critical income to $1 . 22 ( $0 . 76–$1 . 69 ) , meaning that HIV and smoking could explain about half of the increases in critical income . In 1970 , the critical income was $0 . 54 ( $0 . 38–$0 . 71 ) , thus , HIV infection and smoking do not explain all of the worsening of critical income observed by the year 2000 . HIV infection and smoking had no statistical impact on the maximum survival . 10 . 7554/eLife . 00051 . 010Figure 5 . Impact of smoking and HIV on critical income for adult males in 2000 . DOI: http://dx . doi . org/10 . 7554/eLife . 00051 . 010 These results are further supported by a first differences analysis ( Table 4 ) . Over the 10-year period from 1990 to 2000 , a rise of one cigarette consumed per day per person and 1% in HIV prevalence led , for adult men , to a $1 . 70 ( $0 . 96–$2 . 43 ) and $0 . 70 ( $0 . 27–$1 . 13 ) increase in critical income , respectively . For adult females , the impact of a 1% rise in HIV prevalence per capita was associated with a $0 . 40 ( $0 . 31–$0 . 49 ) increase in critical income; however , the impact of smoking was lower , with a change of $0 . 38 ( $0 . 23–$0 . 54 ) . Moreover , adding a covariate for HIV prevalence also improved the goodness of fit dramatically for all years and genders ( Table 3 ) . 10 . 7554/eLife . 00051 . 011Table 4 . First differences analysis for HIV prevalence and cigarette consumption on country-specific critical income from 1990 to 2000DOI: http://dx . doi . org/10 . 7554/eLife . 00051 . 011NR2HIV α ( $ per HIV % ) Standardized αSmoking β ( $ per cigarette per person per day ) Standardized βAdult male920 . 2400 . 70 ( 0 . 27–1 . 13 ) 0 . 3021 . 70 ( 0 . 96–2 . 43 ) 0 . 425Adult female920 . 5040 . 40 ( 0 . 31–0 . 49 ) 0 . 6550 . 38 ( 0 . 23–0 . 54 ) 0 . 366Note: All model parameters were found to be significant , p<0 . 0001 .
Our establishment of a new parameter ‘critical income’ provides novel insights into the relationship of global mortality changes with income . It builds on the well-established Preston curve functions by quantifying the rise in survival . More profoundly , our analyses reveal that while less and less income is required to improve childhood survival , the opposite is true to improve adult survival—particularly in low- and middle-income countries . Unlike for adults , child survival replicates the upward and lateral trend in the original Preston curve for life expectancy . This suggests that for low- and middle-income countries , the majority of the past gains in life expectancy have been achieved via declining child mortality . This is consistent with the UN Population Division trends given the large impact of child mortality ( compared to adult mortality ) on overall life expectancy ( United Nations , 2011 ) . The greatest decline in child critical income was achieved after 1990 , coinciding with actions following the UN's World Summit for Children . With justification , cost-effective interventions have been disproportionately devoted to child and maternal health ( Daar et al . , 2007 ) , and more recently to control of infectious diseases such as HIV/AIDS , malaria , and tuberculosis . Increasing coverage of inexpensive health interventions such as immunization , insecticide-treated nets , and case management of childhood infections could be contributing to the decline in critical income for child survival ( Jamison et al . , 2006; Mathers et al . , 2008 ) . Moreover , there might be complementary benefits of education in reducing child mortality ( Gakidou et al . , 2010 ) . Our study does not address any causal relationship between such interventions and reductions in child mortality; however , our results imply that the achievement of the UN MDG 4 ( to reduce child mortality by two-thirds from 1990 levels ) might be due to the falling levels of income needed to increase child survival . For adult survival , however , there is a reversal of fortune . While achievable adult survival rates have improved , improvements are only associated with those countries at higher income levels . The rise in critical income suggests that the marginal costs of increasing adult longevity are rising; this may explain the lower rate of decline in adult mortality in countries with low income ( Rajaratnam et al . , 2010 ) . The emergence of HIV/AIDS in the 1980s and the rise in global smoking prevalence in low- and middle-income countries ( Guindon and Boisclair , 2003 ) can explain much , but not the entire rise in critical income for adult males . For males , the rise in smoking accounts for over 40% more of the variance in critical income compared to 30% for HIV/AIDS . For adult women , the impact of cigarettes on critical income is much lower , which reflects the five times lower prevalence of smoking among women compared to men ( Guindon and Boisclair , 2003 ) . Indeed , previous studies have already highlighted the impact of smoking and HIV/AIDS on adult survival in developing countries . Even low levels ( 4% ) of HIV prevalence in rural Tanzania can increase overall adult mortality by more than 50% ( Todd et al . , 1997 ) . Our findings also show that smoking increases critical income but has no statistical impact on maximum survival . This is in line with trends in global smoking , where prevalence of smoking ( and subsequently the deaths attributed to smoking ) are rising in low- and middle-income countries but declining in high-income countries ( Jha , 2009 ) . Under the current conditions , an approximate national income per capita of $2 . 20 per day would be required in 2007 to attain the same achievable adult male survival rate with $1 . 25 per day in 1970 . Moreover , should the critical income costs for adults continue to rise ( in line with current trends ) , survival rates for low- and middle-income countries might well deteriorate into the future . In contrast , high-income countries have benefited from the rise in maximum survival among adults . This is likely due to more widespread availability of secondary treatments for chronic diseases , most notably for vascular disease , and in particular from sharp reductions in smoking ( Jha , 2009 ) . The probabilities of premature adult deaths before the age of 70 from vascular , respiratory , and neoplastic diseases are remarkably similar in low- , middle- , countries and high-income countries ( Strong et al . , 2005 ) . Low-cost , effective , and feasible interventions ( Jha et al . , 2002 ) exist against these diseases , most importantly tobacco control ( Jha , 2009 ) , but also increased reduction in hazardous alcohol intake ( particularly in former Soviet states; Zaridze et al . , 2009 ) , and low-cost drugs for secondary management of existing disease . However , these interventions are still not widely used in low-income countries ( Jha et al . , 2012 ) . Over the past few decades , research and development of new technologies ( drugs , vaccines , and policies ) have focused mostly on childhood and infectious disease , with fewer worldwide investments for adult chronic diseases ( Ad Hoc Committee on Health Research Relating to Future Intervention Options , 1996 ) . Thus , the longer-term trajectory of critical incomes for adult survival might well depend on the development of newer interventions , as well as more widespread application of interventions already proven to be cost effective ( Jamison et al . , 2006 ) . This is , as far as we know , the first application of enzyme kinetics to mortality changes . There is no a priori reason to exclude a relationship of mortality and income being comparable to biological reactions . Indeed , since its initial formulation in 1913 , the MM equation has been sufficiently adaptable to explain various levels of biological complexity from thousands of single enzyme reactions ( English et al . , 2005 ) , reversible and quasi-steady state systems ( Briggs , 1925 ) , the growth of microbial cultures on a nutrient substrate ( Monod , 1949 ) , the growth in size of a variety of different animal species ( López et al . , 2000 ) , and the ability of organisms to acclimate to changes in environmental conditions ( Bonachela et al . , 2011 ) . Like all models , ours faces certain limitations . The relationship of income to mortality is comparable using the log-linear , logistic , MM , or indeed other methods . But none of the models can directly elucidate the mechanism ( s ) that converts income into better survival . The novel insight from the MM method is that we can model income as a substrate that catalyzes further changes that more directly impact survival . National income is quite large in relation to health spending for most countries ( the median national spending on health is 5% of the GDP , and the United States at about 18% of GDP represents an outlier ) . Thus , income fulfills , partially , the MM definition of a substrate that is an input converted into health interventions and their use . We also reason that income directly enables certain technologies , immunization programs , epidemiological knowledge , education , and sanitation systems and other areas , which may themselves be interpreted as ‘catalysts’—agents that accelerate the rate of a reaction without being fully consumed in the process . Moreover , the MM method advances some understanding of mechanisms by defining critical income levels that measure the efficiency of particular countries using available income to keep up with maximal achievable survival by other countries during that narrow 5-year time period . The new variable critical income ( akin to Km ) provides a measure to observe a ‘lateral shift’ that was not possible mathematically with the log-linear or logistic models . Additionally , the obvious impacts of two large risk factors ( HIV and tobacco ) on adult survival suggest some mechanistic insights between income and survival ( and indeed the fact that male survival was more unequal from greater male smoking , and female and male survival were affected equally by the more even spread of HIV infection , further strengthens this case ) . We caution however that full explanation of the links between specific technologies catalyzed by income need more research . For example , the Monod equation ( an adaptation of the MM equation to fit the growth of bacterial cultures; Monod , 1949 ) , which was first proposed in 1949 , was only interpreted thermodynamically 50 years later ( Liu et al . , 2003 ) . Nonetheless , this equation was ( and continues to be ) used extensively in the pharmaceutical and food industries as well as in waste treatment systems . Additionally , our analyses could be considerably strengthened by examining trends in age , gender , and cause-specific mortality , but cause-specific mortality data are simply unavailable for most countries; for instance , notwithstanding global efforts to identify data for child mortality , less than 3% of all child deaths worldwide were certified according to cause of death ( Liu et al . , 2012 ) . Indeed , expanded efforts to measure causes of death is a big global priority ( Jha , 2012; Vogel , 2012 ) and the development of new physically or biologically inspired analytical models can help elucidate better understanding of global mortality trends .
Data for 180 countries at 5-year intervals from 1970 to 2007 were included in the analysis , with definitions of income as per those from the World Bank ( Supplementary file 1 ) ; the Institute for Health Metrics and Evaluation ( Institute for Health Metrics and Evaluation , 2012 ) , UN Population Division ( United Nations , 2011 ) , UNAIDS online database ( UNAIDS , 2010 ) , and Penn World Table 6 . 3 databases ( Heston et al . , 2009 ) were used for child and adult mortality , life expectancy and country population by age-groups , prevalence of HIV among adults aged 15–49 years , and GDP per capita at constant 2005 international dollars and purchasing price parity , respectively . Cigarette consumption per capita is from the American Cancer Society ( Guindon and Boisclair , 2003 ) . Availability of data for all 180 countries was not consistent across all years and all variables . Data from missing countries ( by individual year and population group and disease and risk burden ) were removed from that specific regression analysis . A list of countries is found in Supplementary file 1 . The UN Population Division database for child ( UN Inter-agency for Child Mortality Estimates , 2012 ) and adult ( United Nations , 2011 ) mortality were used to confirm results . To confirm the completeness and accuracy of the Institute for Health Metrics and Evaluation database , statistical analysis was duplicated using 5-year rolling averages from UN Inter-agency Group for Child Mortality Estimates ( UN Inter-agency for Child Mortality Estimates , 2012 ) and UN Population Division database adult mortality ( United Nations , 2011 ) , with general consistency in the trend lines ( Figure 6A , B ) . 5-year rolling averages also increased the R2 values compared to single-year regressions; for regressions using UN datasets , the average R2 rose to above 0 . 5 for all age groups ( including high HIV-prevalent countries ) —compared with lower correlation coefficients for single-year regressions using the Institute for Health Metrics and Evaluation dataset ( Table 3 ) . 10 . 7554/eLife . 00051 . 012Figure 6 . Child , adult male and adult female maximum survival ( A ) and critical income ( B ) curves from 1970 to 2005 using two different data sources ( IHME and UN Population Division ) . Maximum survival and critical income values were calculated using 5-year averages where the national income per capita and country survival rates for year i was an average for the years i to i + 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 00051 . 012
For the Michaelis–Menten adapted model and the logistic model , a nonlinear regression analysis ( using iterative parameter estimation algorithms; Greco and Hakala , 1979 ) was used to calculate the parameters . The logistic and log-linear regressions were analyzed for life expectancy to compare the goodness of fit . To study progress over time , we calculated these eo , max , 5p0max and 45p15max , and Kinc values for each age group and gender at 5-year intervals from 1970 to 2007 . Countries with missing mortality or income estimates were eliminated from the regression analysis on a year-by-year basis . We also computed estimates of critical income for each specific country , year , age group , and gender , assuming that the maximum survival was a constant across all countries . Additional analyses were completed by adding covariates for HIV prevalence in a given year ( from 1990 to 2007 ) and cigarette consumption per capita ( in the year 2000 ) . For the year 2000 , analysis was restricted to countries with complete data for both HIV prevalence and cigarette consumption ( n = 86 ) . First differences analysis was used to confirm the impact of HIV/AIDS prevalence and cigarette consumption on critical income values over the 10-year period from 1990 to 2000 , such that[10]Kinc=f ( HIV , Cig ) , [11]ΔKinc=αΔHIV+βΔCig; where α=∂Kinc∂HIV and β=∂Kinc∂Cig . We used SPSS ( version 19 ) to conduct all regression analyses . | In 1975 Samuel Preston published a classic paper that showed life expectancy was related to national income . When plotted as a graph , with national income on the horizontal axis and life expectancy on the vertical axis , the Preston curve shows that an increase in national income leads to an increase in life expectancy , with the increases in life expectancy becoming proportionally smaller as income increases . Moreover , Preston showed that innovations in healthcare ( such as vaccinations , public health education , and sanitation systems ) were increasing the maximum life expectancy that can be achieved for any given national income ( defined as GDP per capita ) : this can be seen by comparing the Preston curves from 1960 and 2000 shown in Figure 1A . Indeed , global life expectancy increased by about 25 years over the course of the 20th century , which suggests that the level of daily income needed to achieve a certain life expectancy should be falling over time . To explore this in greater detail , Hum et al . have constructed a mathematical model to investigate the relationship between health and income across different age groups and income levels . They found that most of the gains in life expectancy for low- and middle-income countries have been achieved by reducing child mortality , with gains in life expectancy for adults being restricted mostly to high-income countries . The model , which is based on the mathematical equations used to describe the kinetics of enzymatic reactions , makes it possible to estimate the improvements of health that can be made over time , and also the level of income that is needed to achieve these improvements . In particular , Hum et al . have established a new parameter , the critical income , which is the level of income needed to achieve half of the maximal health found in high-income countries for the year in question . Based on available data from over 150 countries , they found that critical incomes fell by half for children between 1970 and 2007 , but doubled for adult males during the same period . The rise in critical income for adults was due partly to the HIV epidemic and increases in smoking in low- and middle-income countries , reflecting the growing problems presented by noncommunicable diseases . Hum et al . conclude that increasing the survival among adults will require increased use of proven cost-effective interventions , most notably tobacco control , plus new research to identify low-cost drugs , diagnostics , and other public health strategies . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"(data",
"sources)",
"Methods"
] | [
"epidemiology",
"and",
"global",
"health",
"microbiology",
"and",
"infectious",
"disease"
] | 2012 | Global divergence in critical income for adult and childhood survival: analyses of mortality using Michaelis–Menten |
Proper orchestration of quiescence and activation of progenitor cells is crucial during embryonic development and adult homeostasis . We took advantage of the zebrafish sensory lateral line to define niche-progenitor interactions to understand how integration of diverse signaling pathways spatially and temporally regulates the coordination of these processes . Our previous studies demonstrated that Schwann cells play a crucial role in negatively regulating lateral line progenitor proliferation . Here we demonstrate that ErbB/Neuregulin signaling is not only required for Schwann cell migration but that it plays a continued role in postmigratory Schwann cells . ErbB expressing Schwann cells inhibit lateral line progenitor proliferation and differentiation through non-cell-autonomous inhibition of Wnt/β-catenin signaling . Subsequent activation of Fgf signaling controls sensory organ differentiation , but not progenitor proliferation . In addition to the lateral line , these findings have important implications for understanding how niche-progenitor cells segregate interactions during development , and how they may go wrong in disease states .
The cell signaling events that govern progenitor quiescence , activation and differentiation are incompletely understood , but emerging data in many tissues indicate that dynamic interactions between progenitors and specialized niche environments play key roles in regulating the properties of progenitor pools . Hence , understanding niche-progenitor interactions at the cellular level is crucial for building a general understanding of this process . The development of the zebrafish lateral line is an excellent model system to study progenitor cell regulation , as it consists of relatively few cells that are easily accessible and amenable to experimental manipulations . The sensory organs of the lateral line are called neuromasts . Neuromasts contain support and mechanosensory hair cells that detect water motion . The first set of neuromasts is laid down by a migrating primordium ( primI ) that develops from a placode just posterior to the otic vesicle . As the primordium migrates posteriorly along the trunk of the embryo it deposits five to six primary neuromasts and a chain of interneuromast cells that connects each neuromast ( Ghysen and Dambly-Chaudiere , 2007 ) . Before the placode becomes migratory , its anterior portion splits off and forms the posterior lateral line ganglion ( Northcutt and Brandle , 1995 ) . Lateral line axons closely follow the migrating primordium and eventually innervate deposited neuromasts ( Gilmour et al . , 2004 ) . In turn , neural crest-derived Schwann cells migrate along the axons which they eventually myelinate ( Gilmour et al . , 2002; Lyons et al . , 2005 ) . Thus , interneuromast cells , axons and Schwann cells are in close contact during the early stages of lateral line development ( see diagram in Figure 1A; Whitfield , 2005 ) . 10 . 7554/eLife . 01832 . 003Figure 1 . Illustration of cell types in the migrating lateral line . ( A ) As the primordium migrates it deposits neuromasts and a chain of interneuromast cells ( green cells ) . Pioneer axons ( yellow line ) of the posterior lateral line ganglion grow out with the primordium . Schwann cells ( red cells ) migrate and proliferate along axons . nrg1-3z26 mutants and pharmacological inhibition of ErbB signaling mimics the erbb phenotype . ( B–E ) Double in situ hybridization was performed to label Schwann cells with myelin basic protein ( mbp ) and neuromasts with klf4 at 5 dpf . ( B ) Control siblings with Schwann cells ( arrows ) along the lateral line nerve and normal neuromast number . nrg1-3z26 mutants mimic erbb2 and erbb3b mutants in that they lack Schwann cells along the lateral line and have increased neuromast number ( C ) . The brown cells along the midline in both sibling and nrg1-3z26 are pigment cells . ( D and E ) Double in situ hybridization for mbp and klf4 in DMSO or AG1478 treated larvae from 50 hpf . Compared to DMSO treatment ( D ) , increased neuromasts are seen in AG1478 treated larvae ( E ) . mbp expression along the midline shows that Schwann cells ( arrows ) are still present at 5 dpf when AG1478 was given at 50 hpf ( E ) , compare to DMSO treated ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 00310 . 7554/eLife . 01832 . 004Figure 1—figure supplement 1 . Mutations in the erbb signaling pathway show precocious neuromast formation by 5 dpf . Alkaline phosphatase staining of control ( A ) , erbb2 ( B ) , erbb3b ( C ) and nrg1-3z26 ( D ) zebrafish at 5 dpf . Quantification of alkaline phosphatase stained larvae shows significant increase in neuromast number in all mutants compared to control siblings ( E , Student's t-test , p≤1 . 0E−44 ) . ( F ) AG1478 induces extra neuromasts even if given after Schwann cell migration is complete . AG1478 was added at 24 , 50 , 59 , 72 , or 80 hpf and neuromasts were counted at 5 dpf . For the negative control DMSO was added at 24 hpf only . AG1478 induces a significant increase in neuromasts if given up to 59 hpf ( F , one-way ANOVA with Tukey pairwise comparison , p≤4 . 0 E−6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 00410 . 7554/eLife . 01832 . 005Figure 1—figure supplement 2 . nrg1-3z26 mutants have defects in adult pigment pattern . Control siblings at one month of age show typical stripe pattern of melanophores ( A–A′ ) . nrg1-3z26 at 1-month-old show patchy placement of melanophores in the anterior trunk with a more adult like pattern in the posterior region reminiscent of erbb3b mutants ( B–B′ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 00510 . 7554/eLife . 01832 . 006Figure 1—figure supplement 3 . nrg1-3z26 mutants lose neuromasts as they age . Control sibling Tg ( SqET20:gfp ) ( A ) or nrg1-3z26/Tg ( SqET20:gfp ) ( B ) , were imaged at 1 month of age . Neuromasts that stay along the midline can be seen in control siblings ( A , arrowhead ) . These neuromasts are lost from the more posterior region in nrg1-3z26 adult zebrafish ( B , arrowhead ) . Similarly neuromasts are also lost from the more ventral lateral line ( arrows ) , which are mostly derived from primI , in nrg1-3z26 ( B ) . ( C–D′ ) At 4 months of age the degeneration of neuromasts is even more severe . In controls at four months multiple stitches of neuromasts can be seen after DASPEI staining along the ventral line ( C ) and tail fin ( C′ ) . nrg1-3z26 have no ventral lateral line ( D ) or tail fin ( D′ ) neuromasts remaining at 4 months . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 00610 . 7554/eLife . 01832 . 007Figure 1—figure supplement 4 . ErbB inhibition after lateral line migration is complete causes a decrease in proliferation and number of lateral line Schwann cells . BrdU plus DMSO or AG1478 was given to Tg ( foxd3:gfp ) fish at 48 hpf then fixed at 6 , 14 , or 24 hr post treatment . BrdU index is decreased ( A , Student's t-test , p=0 . 0006 ) , but Schwann cell number is normal 6 hr post AG1478 treatment ( B , Student's t-test , p=0 . 16 ) . There is an average count of 12 or 13 Schwann cells in control or AG1478 treatments respectively . At 14 hr post AG1478 treatment both BrdU index ( C , Student's t-test , p=5 . 8 E−6 ) , and Schwann cell numbers are decreased ( D , Student's t-test , p=0 . 0008 ) . There is an average count of 20 or 14 Schwann cells in control or AG1478 treatments respectively . By 24 hr post treatment , BrdU index ( E , Student's t-test , p=2 . 6E−19 ) , and Schwann cell numbers ( F , Student's t-test , p=2 . 4E−28 ) , are even more decreased by AG1478 compared to DMSO . There is an average count of 14 or 6 Schwann cells in control or AG1478 treatments respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 007 The adult posterior lateral line contains many more neuromasts than the 7–8 neuromasts initially laid down by primI . These additional ‘secondary’ neuromasts originate from several sources . ( A ) A second primordium , primII , develops at 40 hr post fertilization ( hpf ) and deposits neuromasts in between the previously deposited sensory organs ( Sapede et al . , 2002; Nunez et al . , 2009 ) . ( B ) Intercalary neuromasts arise during the first 2 weeks of development by proliferation and differentiation of primI deposited interneuromast cells ( Sapede et al . , 2002; Grant et al . , 2005; Lopez-Schier and Hudspeth , 2005; Nunez et al . , 2009 ) . and ( C ) During juvenile stages neuromast stitches arise through budding from primary neuromasts ( Ledent , 2002; Wada et al . , 2013a ) . We and others have previously shown that Schwann cells play a crucial role in negatively regulating the timing of differentiation of interneuromast cells into intercalary neuromasts ( Grant et al . , 2005; Lopez-Schier and Hudspeth , 2005 ) . In zebrafish that lack Schwann cells along the lateral line , such as in mutants for sox10 and the ErbB pathway members erbb2 , erbb3b and nrg1-3 , intercalary neuromasts form precociously ( Grant et al . , 2005; Rojas-Munoz et al . , 2009; Perlin et al . , 2011 ) . As Schwann cells require axons for migration along the lateral line , neurogenin mutants that lack a posterior lateral line ganglion , also show extra neuromasts ( Lopez-Schier and Hudspeth , 2005 ) . Likewise , extra neuromasts form after posterior lateral line ganglion extirpation or Schwann cell ablation ( Grant et al . , 2005; Lopez-Schier and Hudspeth , 2005 ) . These experiments suggest that Schwann cells contribute to an inhibitory niche that keeps lateral line progenitor cells from undergoing precocious proliferation and differentiation . The signaling pathways that orchestrate intercalary neuromast formation are currently unknown . In contrast , the early development of the migrating lateral line has been extensively studied . Complex cell signaling interactions between Wnt/β-catenin , Fgf , Notch and chemokine pathways regulate proliferation , neuromast formation and migration ( Aman and Piotrowski , 2009; Ma and Raible , 2009; Chitnis et al . , 2012 ) . Wnt/β-catenin signaling in the leading region of the primordium initiates and restricts Fgf signaling to the trailing region . In turn , Fgf signaling upregulates dkk1b , a secreted Wnt/β-catenin inhibitor , that restricts Wnt/β-catenin signaling to the leading region ( Aman and Piotrowski , 2008 ) . Fgf signaling induces apical constriction in clusters of cells resulting in the morphogenesis of rosette shaped protoneuromasts ( Lecaudey et al . , 2008; Nechiporuk and Raible , 2008 ) . Fgf signaling is also required for hair cell differentiation ( Millimaki et al . , 2007; Nechiporuk and Raible , 2008 ) , and both Wnt/β-catenin and Fgf signaling are required for proliferation within the migrating primordium ( Aman et al . , 2011 ) . This study focuses on the development of intercalary neuromasts to elucidate the molecules that regulate progenitor cell proliferation and development . We characterize the signaling pathways required for precocious intercalary neuromast formation downstream of ErbB signaling . In the absence of Schwann cells , or ErbB/Neuregulin signaling , Wnt/β-catenin and Fgf signaling are increased . Wnt/β-catenin signaling is required for interneuromast proliferation while Fgf signaling is required for subsequent rosette formation and cellular differentiation . Schwann cells maintain interneuromast cells as quiescent progenitors by expressing a , as yet unidentified Wnt/β-catenin inhibitor . These findings illustrate the intricate manner in which diverse signaling pathways coordinate distinct aspects of the niche-progenitor interaction needed to maintain the proper balance and timing of this dynamic cell population .
Intercalary neuromasts arise during a 2-week period from interneuromast cells , which are initially deposited by primI as a chain of single cells in between primary neuromasts ( Grant et al . , 2005 ) . The cellular relationships within the migrating lateral line are outlined in Figure 1A . Deposited interneuromast cells are initially in close contact with Schwann cells ( Figure 1A , green and red cells respectively ) . A variety of lines of genetic evidence , including our own , demonstrates that ErbB signaling plays a fundamental role in the migration of Schwann cells that control the proliferation of interneuromast cells . Mutations in ErbB receptors ( row/erbb2 , hps/erbb3b ) cause a loss of Schwann cells along the lateral line nerve leading to precocious interneuromast proliferation and intercalary neuromast differentiation ( Figure 1—figure supplement 1A–E; Grant et al . , 2005; Lyons et al . , 2005; Rojas-Munoz et al . , 2009 ) . We recently identified a mutation in the ErbB ligand neuregulin 1-3 ( nrg1-3z26 ) that also lacks Schwann cell migration along lateral line axons ( Perlin et al . , 2011 ) , and forms supernumerary neuromasts ( Figure 1B–C ) . nrg1-3z26 mutants survive to adulthood but exhibit an adult pigment pattern and neuromast degeneration phenotype ( Figure 1—figure supplement 2 , 3 ) , similarly to erbb3b adult mutant fish ( Budi et al . , 2008; Honjo et al . , 2011 ) . Below we identified in which cell types different members of the ErbB/Neuregulin pathway are functioning to control Schwann cell migration and lateral line progenitor proliferation and differentiation . During development , signaling pathways are repeatedly employed . We therefore wanted to test if the extra neuromast phenotype is due solely to loss of Schwann cells along the lateral line , or if ErbB signaling plays an additional role in inhibiting proliferation of interneuromast cells . Therefore , ErbB signaling was inhibited with the ErbB tyrosine kinase inhibitor AG1478 ( Osherov and Levitzki , 1994 ) , before ( 24 hpf ) and after ( 48 hpf ) completion of Schwann cell migration , and neuromast number was assessed at 5 days post fertilization ( dpf ) . As expected , inhibition of ErbB signaling at 24 hpf , when Schwann cells migrate , leads to a loss of Schwann cells and the formation of extra neuromasts ( Figure 1—figure supplement 1F; Rojas-Munoz et al . , 2009 ) . Interestingly , ErbB inhibition is able to increase neuromast numbers even in the presence of Schwann cells , if supplied between 50–59 hpf ( Figure 1D–E , Figure 1—figure supplement 1F ) . The presence of Schwann cells is based on detection of myelin basic protein ( mbp ) expression ( Figure 1D–E , arrows ) . These data suggest that ErbB signaling not only regulates Schwann cell migration but also plays a continued role in post-migratory Schwann cells in inhibiting interneuromast cell proliferation . A potential caveat for that interpretation is that ErbB signaling is also required for Schwann cell proliferation ( Lyons et al . , 2005; Raphael et al . , 2011 ) , and pharmacologically lowering the number of Schwann cells could secondarily affect interneuromast proliferation . To test when Schwann cell numbers are reduced upon ErbB inhibition at 48 hpf we used the Tg ( foxd3:gfp ) zebrafish line that expresses EGFP in neural crest derived tissues including Schwann cells ( Gilmour et al . , 2002 ) . Using BrdU labeling in control and AG1478 treated Tg ( foxd3:gfp ) fish , we counted BrdU positive Tg ( foxd3:gfp ) Schwann cells at 6 , 14 or 24 hr post treatment . ErbB inhibition induces a decrease in BrdU incorporation in Schwann cells at 6 hr post treatment , however the total Schwann cell number remains unchanged ( Figure 1—figure supplement 4A–B ) . A reduction in Schwann cell proliferation continues at 14 and 24 hr post treatment , at which point it is accompanied by a decrease in total Schwann cell numbers ( Figure 1—figure supplement 4C–F ) . The finding that Schwann cell numbers are not affected at 6 hr post treatment is important , as the first molecular changes in interneuromast cells are already observed at this stage ( see below , 'Wntβ-catenin signaling activation occurs prior to Notch and Fgf activation within interneuromast cells after ErbB inhibition' ) . This suggests that ErbB signaling affects lateral line proliferation directly , rather than indirectly via the regulation of Schwann cell number . Thus , ErbB signaling has independent functions in Schwann cell migration and lateral line progenitor proliferation and differentiation . To elucidate if ErbB signaling controls progenitor proliferation cell-autonomously or non-cell-autonomously , we performed transplantation experiments between mutant and wild type embryos . Transplantation experiments revealed that erbb3b and sox10 act cell-autonomously in Schwann cells to regulate their migration and inhibit precocious interneuromast proliferation ( Grant et al . , 2005 ) . As erbb2 is ubiquitously expressed and AG1478 blocks ErbB signaling globally , we wanted to clarify in which cell type ErbB2 signaling is required . We transplanted dextran-Alexa568 labeled cells from erbb2 mutant blastomere stage donor embryos into wild type Tg ( foxd3:gfp ) host embryos and analyzed clones that gave rise to interneuromast cells . erbb2 mutant interneuromast cell clones failed to produce extra neuromasts by 4 dpf , suggesting that ErbB2 signaling is not required in interneuromast cells to prevent intercalary neuromast formation ( Figure 2A–A′ , n = 0/4 ) . On the other hand , when transplanted Tg ( foxd3:gfp ) wild type cells gave rise to Schwann cell clones in an erbb2 mutant host embryo , the extra neuromasts phenotype at 4 dpf was rescued ( Figure 2B–B“’ , n = 9/9 ) . Rescue was only achieved when Schwann cell clones extended all the way along the trunk by 48 hpf . Transplanted cells that only gave rise to interneuromast cells , or a few Schwann cells that did not reach the tail tip , failed to rescue the erbb2 mutant phenotype ( n = 0/12 ) . These transplant experiments illustrate that , similar to ErbB3b , ErbB2 is required in Schwann cells to inhibit intercalary neuromast formation . 10 . 7554/eLife . 01832 . 008Figure 2 . Transplantation and transgenic analysis demonstrates that ErbB2 is required within Schwann cells and Nrg1-3 within lateral line neurons to inhibit extra neuromast formation . ( A and A’ ) Alexa-568 dextran ( red ) labeled erbb2 mutant cells were transplanted into Tg ( foxd3:gfp ) ( green ) wild type fish . ( A ) High magnification view shows erbb2 interneuromast and mantle cells along the lateral line and around neuromasts ( arrows ) . These erbb2 interneuromast cells fail to induce extra neuromasts by 4 dpf ( A’ ) . ( B–B’’’ ) Alexa-568 dextran ( red ) /Tg ( foxd3:gfp ) ( green ) wild type cells were transplanted into erbb2 mutant host . At 4 dpf neuromast were labeled by DASPEI staining ( green ) . ( B and B’ ) On the untransplanted side there are no Schwann cells and eighteen neuromasts . ( B’’ and B’’’ ) On the transplanted side you can see complete migration of wild type Schwann cells in an otherwise erbb2 mutant fish and rescue of neuromast number ( arrows ) . ( C and D ) Dominant negative ErbB receptor expression in neural crest derived cells mimics erbb mutant phenotype . ( C ) Control Tg ( SqET20:gfp ) siblings at 4 dpf . ( D ) Tg ( SqET20:gfp ) /Tg ( sox10:DNerbb4 ) showing extra neuromasts . ( E–E’’’ ) Alexa-568 dextran ( red ) /Tg ( clndB:lyngfp ) ( green ) wild type cells were transplanted into nrg1-3z26 mutant host . ( E–E’ ) At 4 dpf the untransplanted side has nineteen neuromasts . ( E’’ and E’’’ ) On the transplanted side there are GFP labeled axons ( arrowhead ) along the entire length of the lateral line with rescue of neuromast number . All axons cannot be seen because some are obscured underneath pigment cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 008 We confirmed these results genetically by generating a transgenic zebrafish line that drives human dominant-negative ErbB4 ( DNErbB4 ) in neural crest-derived tissues , including Schwann cells , using the sox10 promoter . DNErbB4 blocks Neuregulin-induced signaling in cell culture and in vivo ( Rio et al . , 1997; Chen et al . , 2006 ) . We isolated a stable transgenic line for analysis designated Tg ( sox10:DNhsaerbb4-rfp ) , from now on called Tg ( sox10:DNerbb4 ) . As expected , in transgenic embryos Schwann cells fail to migrate along the lateral line and they develop extra neuromasts ( data not shown and Figure 2C–D ) . Transgenic fish survive to adulthood , are fertile and exhibit no other obvious phenotypes . Along with the transplantation experiments these results demonstrate that ErbB2/3b signaling is required in Schwann cells in order to non-cell-autonomously regulate neuromast number . nrg1-3 is expressed in lateral line ganglia suggesting that it is required in lateral line axons ( Perlin et al . , 2011 ) . To functionally test in which cell type Nrg1-3 is required we performed transplantation experiments between wild type and nrg1-3z26 embryos . As donors we used Tg ( cldnb:lyngfp ) embryos that express EGFP in all lateral line cells including the ganglion ( Haas and Gilmour , 2006 ) . We rescued the extra neuromast phenotype in nrg1-3z26 mutants with transplanted wild type cells that gave rise to large posterior lateral line ganglion clones ( Figure 2E–E’’’ , n = 13/13 ) . Transplanted cells that only contributed to interneuromast cells or to few lateral line ganglion neurons failed to rescue nrg1-3z26 ( n = 0/19 ) . This is consistent with prior findings that wild type posterior lateral line ganglion clones rescue Schwann cell migration in nrg1-3z26 mutant embryos ( Perlin et al . , 2011 ) . Thus , the ligand Nrg1-3 is required in axons to induce migration and proliferation of ErbB expressing Schwann cells and inhibit precocious formation of intercalary neuromasts . Combined , these experiments revealed that the quiescent niche consists of axonal , membrane bound , Nrg1-3 that signals to ErbB receptors within Schwann cells . In response , Schwann cells send a signal to interneuromast cells that inhibits their precocious differentiation into neuromasts . The following experiments were designed to identify signaling pathways that are regulated in interneuromast cells in response to Schwann cell-derived signals . The identification of interneuromast cell behaviors that are inhibited by ErbB signaling provides clues to which signaling pathways might be regulated by ErbB signaling . To identify the earliest changes in lateral line cell behavior in response to the loss of ErbB signaling we performed time-lapse analyses . We imaged interneuromast cells in Schwann cell-depleted larvae derived from crosses between Tg ( sox10:DNerbb4 ) and Tg ( SqET20:gfp ) . Tg ( SqET20:gfp ) larvae express EGFP in neuromast mantle cells and interneuromast cells ( Parinov et al . , 2004 ) . In a 40-hr time-lapse four intercalary neuromasts are formed from interneuromasts cells ( Video 1 ) . The time-lapse analyses revealed that interneuromast cell proliferation precedes clustering of interneuromast cells . In addition , interneuromast cells are highly motile and migrate into and out of the forming neuromasts . The clusters of interneuromast cells continue to proliferate and differentiate into neuromasts as evident by the mature pattern of a ring of Tg ( SqET20:gfp ) positive mantle cells that surrounds GFP-negative sensory hair cells . In contrast , control Tg ( SqET20:gfp ) larvae show no proliferation and little migration of interneuromast cells during the same time period ( Video 2 ) . In conclusion , the absence of Schwann cells , leads first to interneuromast cell proliferation , followed by an increase in migration and clustering of interneuromast cells that eventually differentiate into sensory hair and support cells . 10 . 7554/eLife . 01832 . 012Video 1 . Time-lapse recording of Tg ( sox10:DNerbb4 ) /Tg ( SqET20:gfp ) /Tg ( clndB:H2B-mcherry ) during intercalary neuromast formation . The time-lapse runs from 32 to 72 hpf . One frame was taken every 7 min . Four intercalary neuromasts form during this time . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 01210 . 7554/eLife . 01832 . 013Video 2 . Time-lapse recording of control Tg ( SqET20:gfp ) / ( Tg ( clndB:H2A-mcherry ) from approximately 48–72 hpf . One frame was taken every 7 min . No interneuromast cell proliferation is seen during this time . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 013 To be able to correlate cell behavior with gene expression changes ( see below ) , we sought to determine how many hours after ErbB inhibition interneuromast proliferation begins . We added BrdU plus DMSO or AG1478 to Tg ( SqET20:gfp ) larvae at 48 hpf , after Schwann cell migration is completed . After 14 hr of ErbB signaling inhibition there is no significant increase in BrdU incorporation in GFP-positive interneuromast cells ( Figure 3A–B , E ) . After 24 hr of treatment we detected a significant increase in BrdU labeling in interneuromast cells ( Figure 3C–E ) . Concurrent with the increase in BrdU incorporation , an increase in interneuromast cells is observed after 24 hr of AG1478 treatment ( Figure 3F ) . The increase in proliferation begins sometime between 14 and 24 hr post ErbB inhibition . 10 . 7554/eLife . 01832 . 009Figure 3 . ErbB inhibition , after Schwann cell lateral line migration is completed , induces proliferation of interneuromast cells . BrdU plus DMSO or AG1478 was given to Tg ( SqET20:gfp ) fish at 48 hpf then fixed at 14 or 24 hr post treatment . Immunohistochemistry for BrdU ( red ) and GFP ( green ) reveals no difference in BrdU incorporation within interneuromast cells between DMSO ( A ) or AG1478 ( B ) 14 hr post treatment . At 24 hr post treatment DMSO ( C ) treated fish show little BrdU incorporation while AG1478 ( D ) treated fish show increased BrdU incorporation and interneuromast cell number . Quantification of both BrdU index ( E , Student’s t-test , p=0 . 18 for 14 hr and p=4 . 5E−18 for 24 hr time point ) and interneuromast cell number ( F , Student’s t-test , p=0 . 69 for 14 hr and p=0 . 003 for 24 hr time point ) shows a significant increase with AG1478 only after 24 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 009 As ErbB signaling acts cell-autonomously in Schwann cells but proliferation occurs in interneuromast cells , we aimed to identify the signaling pathways that are activated in interneuromast cells when ErbB signaling is inhibited . The Wnt/β-catenin , Fgf and Notch signaling pathways are excellent candidates for being involved in intercalary neuromast formation as they regulate progenitor cell proliferation in several other organs , such as the CNS ( Logan and Nusse , 2004; Guillemot and Zimmer , 2011; Koch et al . , 2013 ) . In addition , these three pathways play multiple , crucial roles in the primordium of the lateral line ( reviewed in Aman and Piotrowski , 2009; Ma and Raible , 2009; Chitnis et al . , 2012 ) . Briefly , Wnt/β-catenin and Fgf signaling regulate cell proliferation , while Fgf is also required for neuromast rosette formation and hair cell differentiation ( Aman and Piotrowski , 2008; Lecaudey et al . , 2008; Nechiporuk and Raible , 2008; Aman et al . , 2011 ) . Notch signaling regulates sensory hair cell production and primordium cohesion ( Itoh and Chitnis , 2001; Matsuda and Chitnis , 2010 ) . To test if the Wnt/β-catenin , Fgf and Notch pathways are also involved in the extra neuromast phenotype we performed an in situ expression screen on 48 hpf nrg1-3z26 mutant larvae ( Figure 4 ) . The Wnt/β-catenin pathway members wnt10a , lef1 , myca , and β-catenin-2 ( ctnnb2 ) are upregulated in interneuromast cells in nrg1-3z26 larvae ( Figure 4A–H ) . lef1 and ctnnb2 are also expressed at lower levels in interneuromast cells in control animals while wnt10a and myca show no expression . The expression of wnt10a correlates with the differentiation status of intercalary neuromasts . wnt10a is expressed in proliferating interneuromast cells but is down regulated in differentiating neuromasts ( Figure 4—figure supplement 1A–D ) . The Notch receptor notch3 and the Notch target gene her4 . 1 are expressed in primary neuromasts in control animals but not in interneuromast cells ( Figure 4I , K , arrows ) . notch3 is broadly induced in nrg1-3z26 interneuromast cells ( Figure 4J ) . her4 . 1 is also induced in nrg1-3z26 but in a more discrete cluster of cells ( Figure 4L , arrowhead ) . her4 . 1 is not expressed in proliferating interneuromast cells but is induced as intercalary neuromasts mature ( Figure 4—figure supplement 1E–H ) . This suggests that Notch signaling is only active in differentiating neuromasts . Fgf signaling pathway components are also upregulated in interneuromast cells in nrg1-3z26 mutant larvae ( Figure 4N , P , R , T ) . In control embryos , fgfr1a , fgf3 , fgf10 and the Fgf target pea3 only show strong expression in primary neuromasts but not interneuromast cells ( Figure 4M , O , Q , S , arrows ) , suggesting that , similar to Notch signaling , Fgf signaling might be involved in neuromast differentiation . 10 . 7554/eLife . 01832 . 010Figure 4 . Increase in Wnt/β-catenin , Notch and Fgf signaling pathway gene expression in nrg1-3z26 mutant interneuromast cells . Control siblings and nrg1-3z26 mutants were processed for in situ hybridization at 48 hpf . A Wnt ligand , wnt10a , is not expressed in control interneuromast cells ( A ) but is increased in nrg1-3z26 ( B ) . The Wnt/β-catenin target gene lef1 is expressed in interneuromast cells in control siblings ( C ) but is greatly increased in nrg1-3z26 ( D ) . An additional Wnt/β-catenin target myca shows no expression in interneuromast cells ( E ) but strong expression in clumps of interneuromast cells in nrg1-3z26 ( F ) . ctnnb2 shows weak expression in control interneuromast cells ( G ) which is upregulated in nrg1-3z26 ( H ) . ( I ) In controls , notch3 is expressed in primary neuromasts ( arrow ) but not interneuromast cells . ( J ) notch3 is upregulated in mutant interneuromast cells . ( K ) In controls , the Notch target her4 . 1 is expressed in primary neuromasts ( arrow ) but not interneuromast cells . ( L ) In mutants , her4 . 1 is expressed in primary neuromasts ( arrow ) but is also increased in discrete clusters of cells ( arrowhead ) . In controls the Fgf pathway genes including the receptor fgfr1a ( M ) , the two ligands fgf3 ( O ) and fgf10 ( Q ) and the Fgf target gene pea3 ( S ) are all expressed in primary neuromasts ( arrow ) but not in interneuromast cells . All Fgf pathway genes , fgfr1a ( N ) , fgf3 ( P ) fgf10 ( R ) and pea3 ( T ) , retain expression in primary neuromasts ( arrow ) but are upregulated in interneuromast cells of nrg1-3z26 . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 01010 . 7554/eLife . 01832 . 011Figure 4—figure supplement 1 . wnt10a is expressed in proliferating interneuromast cells while her4 . 1 is expressed in differentiating and mature neuromasts . Lateral line cells from Tg ( sox10:DNerbb4 ) /Tg ( SqET20:gfp ) were imaged at 48 hpf . Larvae were then fixed at processed for wnt10a or her4 . 1 in situ hybridization and photographed at the same level . ( A–D ) Proliferating interneuromast cells ( arrowhead ) express higher wnt10a than more mature intercalary neuromasts ( square ) . ( E–H ) her4 . 1 is expressed in more mature intercalary neuromasts ( squares ) , but is absent from more immature proliferating interneuromast cells ( arrowhead ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 01110 . 7554/eLife . 01832 . 017Figure 4—figure supplement 2 . wnt10a , fgf3 and fgf10 expression depends on Wnt/β-catenin signaling . Control or Tg ( sox10:DNerbb4 ) siblings were treated with DMSO , IWR-1 , PD173074 or LY411575 at 32 hpf then fixed at 48 hpf for in situ hybridizations . In control siblings , inhibitor treatment has no effect on wnt10a ( A–D ) , fgf3 ( I–L ) or fgf10 ( Q–T ) expression in interneuromast cells . wnt10a is strongly induced in Tg ( sox10:DNerbb4 ) DMSO ( E ) , PD173074 ( G ) or LY411575 ( H ) treated embryos but is greatly inhibited in IWR-1 treated embryos ( F ) . Likewise , IWR-1 inhibits fgf3 ( N ) and fgf10 ( V ) expression in Tg ( sox10:DNerbb4 ) . Neither PD173074 or LY411575 effect fgf3 ( O–P ) or fgf10 expression ( W–X ) in Tg ( sox10:DNerbb4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 01710 . 7554/eLife . 01832 . 018Figure 4—figure supplement 3 . wnt10a , fgf3 and fgf10 expression are induced within interneuromast cells after Wnt/β-catenin activation . Wild type zebrafish were treated with DMSO or BIO at 48 hpf then fixed at 6 , 12 or 24 hr post treatment for in situ hybridizations . The known Wnt/β-catenin target gene lef1 is expressed in DMSO treated interneuromast cells ( A , C , E ) and is greatly increased by BIO treatment ( B , D , F ) . wnt10a is not expressed in control interneuromast cells from 6 to 24 hr post treatment ( G , I , K ) . wnt10a is induced by BIO within 6 hr post ( H ) and this increase is maintained from 12 to 24 hr post ( J , L ) . fgf3 ( M , O , Q ) and fgf10 ( S , U , W ) are not expressed within DMSO treated interneuromast cells . fgf3 is induced by 24 hr post BIO treatment ( N , P , R ) , while fgf10 is induced within 12 hr of treatment ( T , V , X ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 018 If increased Wnt/β-catenin and Fgf signaling are important for intercalary neuromast formation in Schwann cell-deficient larvae , these pathways should also be upregulated during post-embryonic intercalary neuromast formation in wild type larvae . To enable us to compare gene expression changes with neuromast formation we quantified neuromasts formed between 2–6 dpf . In wild type larvae two primordia form the majority of the posterior lateral line system ( Ghysen and Dambly-Chaudiere , 2007 ) . PrimI deposits five to six primary neuromasts between 20–40 hpf . PrimII begins migration at around 40 hpf along the same path as primI ( Sapede et al . , 2002; Nunez et al . , 2009 ) . We counted the number of neuromasts after alkaline phosphatase staining ( Figure 5D–G ) . The total number of neuromasts increases steadily from six to seven neuromasts at 2 dpf to 12 at 6 dpf ( Figure 5A ) . Of these 12 neuromasts at 6 dpf , three have been deposited by primII ( Figure 5B , E–G , asterisks ) . PrimII-derived neuromasts are always located dorsally to the primI-derived chain of interneuromast cells . The first clusters of interneuromast cells that will differentiate into intercalary neuromasts appear by 3 dpf ( Figure 5E , arrowhead ) . Typically , at least one intercalary neuromast has formed by 4 dpf , with a second formed by 6 dpf ( Figure 5C , F–G , squares ) . Therefore , any genes crucial for interneuromast proliferation and differentiation should commence expression between 2–3 dpf . 10 . 7554/eLife . 01832 . 014Figure 5 . Wnt/β-catenin and Fgf signaling target genes are expressed in interneuromast cells during wild type intercalary neuromast formation . To see when intercalary neuromasts first arise we alkaline phosphatase stained wild type zebrafish at 2 , 3 , 4 , 5 and 6 dpf . Quantification of total neuromast number shows a steady increase from 2 to 6 dpf ( A ) . Most of the increase comes from primII deposited neuromasts ( B ) . There is one intercalary neuromast by 4 dpf and two by 6 dpf ( C ) . ( D–G ) Alkaline phosphatase staining from 2–5 dpf . The images were taken so that the first primI deposited neuromast is always at the left ( without a label ) . Asterisk labels primII-derived neuromasts , the arrowhead labels interneuromast cells and squares label intercalary neuromasts . ( H–K ) lef1 in situ hybridization from 2–5 dpf . lef1 is expressed in interneuromast cells at 2 dpf ( H , arrowhead ) and is maintained in clumps of interneuromast cells from 3–5 dpf ( I–K , arrowheads ) that will likely give rise to intercalary neuromasts . The strong cluster of lef1 expression at 3 dpf is the leading edge of primII ( I , arrow ) . lef1 is not expressed in mature neuromasts . ( L–O ) beta-catenin2 ( ctnnb2 ) expression from 2–5 dpf . Similar to lef1 , ctnnb2 is expressed in interneuromast cells at 2 dpf ( L , arrowhead ) and is maintained in clumps of interneuromast cells from 3–5 dpf ( M–O , arrowhead ) . Unlike lef1 , ctnnb2 is expressed in primary neuromasts ( M–O , asterisk ) . ( P–S ) pea3 expression from 2–5 dpf . ( P ) At 2 dpf pea3 shows strong expression in primary neuromasts but not in interneuromast cells . ( Q–R ) At 3 and 4 dpf pea3 shows expression in primII ( arrow ) and primII derived neuromasts ( asterisk ) but still no interneuromast cell expression . At 5 dpf pea3 can be seen in a few cells near somite boundaries ( S , arrowhead ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 01410 . 7554/eLife . 01832 . 015Figure 5—figure supplement 1 . lef1 is expressed in clusters of interneuromast cells before they become intercalary neuromasts . Lateral line cells expressing Tg ( SqET20:gfp ) were imaged at 4 or 6 dpf . Larvae were then fixed at processed for lef1 in situ hybridization and photographed at the same level . ( A ) At 4 dpf a small group of Tg ( SqET20:gfp ) positive interneuromast cells ( arrowhead ) can be seen between two primII deposited primary neuromasts ( asterisk ) . This small cluster of interneuromast cells is positive for lef1 ( B , arrowhead ) . ( C ) At 6 dpf a larger cluster of interneuromast cells is seen forming . Again this cluster of cells is lef1 positive ( D , arrowhead ) . Interestingly , at 6 dpf one intercalary neuromast has formed ( C , square ) , but it is not positive for lef1 ( D ) . Suggesting lef1 is quickly decreased as neuromasts mature . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 015 In situ hybridization experiments for the Wnt/β-catenin targets lef1 and ctnnb2 revealed that lef1 and ctnnb2 are expressed in deposited interneuromast cells at 2 dpf ( Figure 5H , L , arrowheads ) . Unlike lef1 , ctnnb2 is also expressed in differentiating neuromasts ( Figure 5M–O , asterisk ) . lef1 and ctnnb2 are downregulated in interneuromast cells between 3–5 dpf , with the exception of forming clusters of interneuromast cells that will differentiate into intercalary neuromasts ( Figure 5I–K , M–O , arrowheads ) . To study lef1 expression at single cell resolution we photographed the lateral line in Tg ( SqEt20:gfp ) larvae at 4 or 6 dpf and then performed lef1 in situ hybridization on the same larvae ( Figure 5—figure supplement 1 ) . These experiments illustrate that lef1 is expressed only in the few interneuromast cells that begin to proliferate to form clusters ( Figure 5—figure supplement 1A–D , arrowheads ) . As lef1 expression is shut off in mature neuromasts , Wnt/β-catenin signaling is likely involved in the initiation of intercalary neuromast proliferation and not differentiation . In contrast , the Fgf target pea3 is never expressed in interneuromast cells but is strongly expressed in differentiated neuromasts from 2–5 dpf ( Figure 5P–S , asterisks ) . Discrete clusters of pea3 expressing interneuromast cells are observed at 5 dpf ( Figure 5S , arrowhead ) . The expression analyses shows that Wnt/β-catenin pathway activation coincides with interneuromast cell proliferation and cluster formation , whereas Fgf signaling is initiated later , during the differentiation phase of intercalary neuromast formation . The expression analyses during wild type intercalary neuromast formation suggest that the onset of Wnt/β-catenin expression precedes Fgf and Notch signaling . To determine the temporal dynamics of pathway activations after ErbB inhibition we blocked ErbB signaling with AG1478 at 48 hpf and fixed larvae at 6 , 12 , 24 and 36 hr post treatment followed by in situ hybridization . We focused our analysis on interneuromast cells between the first and second deposited primary neuromasts . We performed in situ hybridization with the Wnt/β-catenin target lef1 and a dgfp probe for a transgenic Wnt/β-catenin reporter line , Tg ( Tcf/Lef-miniP:dGFP ) that contains 6 copies of a consensus TCF/Lef binding site followed by destabilized EGFP ( Shimizu et al . , 2012 ) . At 6 and 12 hr post treatment , DMSO treated fish show strong expression of the Wnt/β-catenin reporter in primII ( Figure 6A–B , arrow ) , and in some interneuromast cells ( arrowhead ) . At 24 hr post DMSO the Wnt/β-catenin reporter is expressed in few interneuromast cells ( Figure 6C , arrowhead ) . By 36 hr post DMSO the Wnt/β-catenin reporter is only expressed in clusters of newly forming intercalary neuromasts and is downregulated in other interneuromast cells ( Figure 6D , arrowhead ) . This correlates with the clusters of lef1 expression seen in wild type fish at 3–5 dpf ( Figure 5I–K ) . ErbB inhibition induces a large increase of the Wnt/β-catenin reporter expression in interneuromast cells at 6 and 12 hr post treatment ( Figure 6E–F , arrowheads ) . By 24 and 36 hr post ErbB inhibition the Wnt/β-catenin reporter expression level has decreased and is only maintained in a few clumps of interneuromast cells ( Figure 6G–H , arrowhead ) . The Wnt/β-catenin reporter is not seen in primary neuromasts ( Wada et al . , 2013b ) , and is turned off in interneuromast cell clusters as they differentiate into intercalary neuromasts ( data not shown ) . The pattern and timing of lef1 expression mirrors the Wnt/β-catenin reporter expression , with ErbB inhibition inducing high expression at early stages followed by a gradual decrease ( Figure 6I–P ) . 10 . 7554/eLife . 01832 . 016Figure 6 . Wnt/β-catenin expression precedes Fgf and Notch expression after ErbB inhibition . To determine which signaling pathways are induced first we treated wild type zebrafish starting at 48 hpf with DMSO or AG1478 then fixed at 6 , 12 , 24 and 36 hr post treatment . All images were taken between the first two primary neuromasts . In DMSO treated Wnt reporter Tg ( Tcf/Lef-miniP:dGFP ) strong expression is seen in primII ( arrow ) at 6 and 12 hr post treatment but also in some interneuromast cells ( A–B , arrowhead ) . By 12 and 36 hr there are only clumps of interneuromast cells expressing the reporter ( C–D , arrowhead ) . 6 hr post AG1478 treatment there is a large increase in Wnt reporter expression specifically within interneuromast cells ( E , arrowhead ) . This AG1478 induced increased expression is maintained after 12 hr ( F ) but has started to go down by 24 hr ( G ) and is only seen in a few interneuromast cells by 36 hr ( H ) . ( I–P ) lef1 mirrors Wnt reporter expression . In DMSO , lef1 is expressed in primII ( arrow ) and few interneuromast cells at 6 ( I ) and 12 hr ( J ) . At 24 hr post DMSO lef1 is maintained in a few interneuromast cells ( K ) . At 36 hr post DMSO , lef1 is seen in clumps of cells likely corresponding to normally developing intercalary neuromasts ( L , arrowhead ) . AG1478 induces lef1 within interneuromast cells at 6 ( M ) and 12 hr ( N ) . By 24 ( O ) and 36 hr ( P ) post AG1478 treatment lef1 is decreased in interneuromast cells compared to the 6 and 12 hr time points , and is maintained in a few clumps of interneuromast cells ( arrowhead ) . ( Q–X ) pea3 shows a later induction than Wnt target genes . In controls pea3 is only seen in primII ( arrow ) or primII deposited neuromasts ( asterisk ) at all time points tested ( Q–T ) . ( U–X ) After AG1478 treatment , pea3 is maintained in deposited neuromasts ( asterisk ) and begins to be expressed in interneuromast cells only at 36 hr post treatment ( X , arrowhead ) . ( Y–FF ) her4 . 1 expression is seen before pea3 expression . ( Y–BB ) In controls , her4 . 1 is seen in primII ( arrow ) and deposited neuromasts ( asterisk ) . In AG1478 treated larvae her4 . 1 is seen in large clusters of cells starting at 24 hr post treatment and can still be seen at 36 hr ( EE–FF , arrowhead ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 016 The temporal expression analysis of the Fgf target pea3 shows that the Fgf pathway is also induced after ErbB inhibition but that the induction happens at a much later time point compared to the Wnt/β-catenin pathway . From 6 to 24 hr post treatment with DMSO or AG1478 pea3 shows no expression in interneuromast cells , even though it is strongly expressed in primII and in deposited neuromasts ( Figure 6Q–S , U–W ) . At 36 hr post ErbB inhibition pea3 appears in small clusters of interneuromast cells ( Figure 6X , arrowhead ) . No pea3 expression is seen in interneuromast cells of control treated fish at 36 hr ( Figure 6T ) . This later induction of pea3 expression after AG1478 correlates with the later induction seen during wild type intercalary neuromast formation ( Figure 5 ) . To test when the Notch pathway becomes active after ErbB inhibition we examined the expression of its target gene her4 . 1 . Similar to pea3 , her4 . 1 is only expressed in primII and deposited neuromasts but not in interneuromast cells in control embryos ( Figure 6Y–BB ) . After ErbB inhibition her4 . 1 is upregulated in large clusters of cells at 24 hr post treatment . The Notch pathway is induced many hours later than the Wnt/β-catenin pathway but before the activation of Fgf signaling ( Figure 6EE–FF , arrowhead ) . In the migrating primordium the Fgf ligands fgf3 and fgf10 are Wnt/β-catenin targets , whereas it is not known what induces wnt10a ( Aman and Piotrowski , 2008 ) . To test if wnt10a , fgf3 and fgf10 are Wnt/β-catenin targets during intercalary neuromast formation we treated control or Tg ( sox10:DNerbb4 ) larvae with pharmacological inhibitors of Wnt/β-catenin , Fgfr or Notch signaling at 32 hpf then fixed for in situ hybridization at 48 hpf . To block Wnt/β-catenin signaling we used the Axin2 stabilizing drug IWR-1 ( Chen et al . , 2009 ) . We blocked Fgfr or Notch signaling with PD173074 and the γ-secretase inhibitor LY411575 , respectively ( Mohammadi et al . , 1998; Wong et al . , 2004 ) . wnt10a , fgf3 and fgf10 expression were only inhibited by blocking Wnt/β-catenin , but not Fgfr or Notch signaling ( Figure 4—figure supplement 2 ) . To test if Wnt/β-catenin signaling induces Wnt or Fgf ligand expression we treated larvae with BIO , a pharmacological inhibitor of the Wnt/β-catenin inhibitor GSK-3 ( Meijer et al . , 2003 ) . wnt10a is induced within 6 hr of treatment and expression is maintained up to 24 hr post-treatment ( Figure 4—figure supplement 3G-L ) . fgf3 and fgf10 are also induced by BIO treatment , but induction takes longer ( Figure 4—figure supplement 3M–X ) . These experiments show that Wnt/β-catenin signaling is both necessary and sufficient for expression of wnt10a , fgf3 and fgf10 within interneuromast cells of Schwann cell deficient larvae . The expression time course analysis of Wnt/β-catenin , Fgf and Notch target genes demonstrates that the loss of ErbB signaling leads to a fast activation of the Wnt/β-catenin pathway followed by Notch and Fgf pathway activation several hours later . As Wnt/β-catenin signaling precedes and coincides with interneuromast cell proliferation , Wnt/β-catenin signaling might be required for proliferation . At the same time Wnt/β-catenin signaling is required for fgf ligand induction . Notch and Fgf signaling are only upregulated in clusters of cells suggesting that they might be playing a later role in hair cell differentiation and rosettogenesis as during early development . To test if Wnt/β-catenin signaling is necessary for extra neuromast formation we employed several methods to block Wnt/β-catenin signaling . First we analyzed larvae mutant for the Wnt/β-catenin signal transducer lef1 , as they show a decrease in primordium cell proliferation ( Gamba et al . , 2010; McGraw et al . , 2011; Valdivia et al . , 2011 ) . During development lef1 mutant larvae generate intercalary neuromasts , but not as many as control larvae ( Figure 7—figure supplement 1A ) . Lack of lef1 also partially rescued the extra neuromast phenotype induced by ErbB inhibition ( Figure 7—figure supplement 1B–F ) . It is likely that intercalary neuromast formation in lef1 mutant larvae was only incompletely inhibited due to redundancy with three additional TCF family members that are also expressed in the developing lateral line ( McGraw et al . , 2011; Valdivia et al . , 2011 ) . To pharmacologically block Wnt/β-catenin signaling we soaked the larvae in Tankyrase inhibitors . We treated 32 hpf wild type and Tg ( sox10:DNerbb4 ) larvae with IWR-1 or XAV939 ( Huang et al . , 2009 ) . Larvae were soaked in the inhibitors for 24 hr and then transferred to embryo medium . At 3 dpf we counted the number of alkaline phosphatase stained neuromasts ( Figure 7C–H ) , up to somite 14 because both inhibitors induced stalling of primI migration ( data not shown and Matsuda et al . , 2013 ) . Wnt/β-catenin inhibition induced a decrease in neuromast number in wild type siblings due to an effect on primII deposition ( Figure 7A , blue bars ) . Importantly , both Wnt/β-catenin pathway inhibitors significantly decreased extra neuromast formation compared to DMSO in Tg ( sox10:DNerbb4 ) larvae ( Figure 7A , red bars ) . 10 . 7554/eLife . 01832 . 019Figure 7 . Wnt/β-catenin signaling is required for extra neuromast formation in the absence of ErbB signaling . To block Wnt/β-catenin signaling wild type or Tg ( sox10:DNerbb4 ) fish were treated with two different inhibitors IWR-1 or XAV939 for 24 hr starting at 32 hpf . Neuromast number up to somite 14 was counted at 3 dpf ( A ) . Compared to DMSO , both IWR-1 and XAV939 significantly inhibited neuromast formation in Tg ( sox10:DNerbb4 ) ( A , red bars , One-way ANOVA with Tukey pairwise comparison , p≤0 . 05 ) . Representative images of alkaline phosphatase stained control siblings treated with DMSO ( C ) , IWR-1 ( E ) or XAV939 ( G ) or Tg ( sox10:DNerbb4 ) treated with DMSO ( D ) , IWR-1 ( F ) or XAV939 ( H ) . ( B ) Neuromast counts at 3 dpf of control , Tg ( sox10:DNerbb4 ) , Tg ( hsp70l:dkk1b ) or Tg ( sox10:DNerbb4 ) /Tg ( hsp70l:dkk1b ) after heat shock at 32 hpf . Tg ( sox10:DNerbb4 ) /Tg ( hsp70l:dkk1b ) double transgenics show a complete loss of extra neuromast formation seen in Tg ( sox10:DNerbb4 ) ( B , Student’s t-test , p=2 . 4E−16 ) . Representative images of alkaline phosphatase stained sibling ( I ) , Tg ( sox10:DNerbb4 ) ( J ) , Tg ( hsp70l:dkk1b ) ( K ) or Tg ( sox10:DNerbb4 ) /Tg ( hsp70l:dkk1b ) ( L ) at 3 dpf . The first deposited neuromast is to the left for all alkaline phosphatase images . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 01910 . 7554/eLife . 01832 . 020Figure 7—figure supplement 1 . lef1 mutants have decreased intercalary neuromast formation in the absence of ErbB signaling . Control siblings or lef1 mutants were fixed at 2 , 3 , 5 , 7 , 10 and 14 dpf and processed for alkaline phosphatase staining . Because lef1 mutants have a decrease in neuromast deposition , we counted neuromasts just up to somite 14 . At 2 and 3 dpf lef1 mutants have the same number of neuromasts as controls ( A ) . From 5 dpf onward there is a significant decrease in neuromast formation in lef1 ( A , Student's t-test , p≤0 . 001 ) . Challenging lef1 mutants with AG1478 at 48 hpf shows only a 19 . 5% increase in neuromast formation compared to DMSO treatment by 5 dpf , vs 35% increase seen in AG1478 treated control siblings ( B , Student's t-test , p=1 . 7 E−8 ) . Representative images of alkaline phosphatase stained DMSO treated sibling ( C ) or lef1 mutant ( E ) or AG1478 treated sibling ( D ) or lef1 ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 020 As a third method to inhibit Wnt/β-catenin signaling we heat shock induced Dkk1b ( Stoick-Cooper et al . , 2007 ) , a secreted antagonist of Wnt/β-catenin signaling that blocks Wnt/β-catenin dependent gene expression and cell proliferation in the lateral line primordium ( Aman and Piotrowski , 2008; Aman et al . , 2011 ) . Tg ( hsp70l:dkk1b ) zebrafish were crossed to Tg ( sox10:DNerbb4 ) and embryos were heat shocked at 32 hpf . Alkaline phosphatase staining at 3 dpf shows a clear reduction in intercalary neuromasts in Tg ( hsp70l:dkk1b ) /Tg ( sox10:DNerbb4 ) compared to Tg ( sox10:DNerbb4 ) ( Figure 7I–L ) . Quantification of neuromast number up to somite fourteen verified a complete absence of extra neuromast formation induced by dominant-negative ErbB when Wnt/β-catenin signaling is inhibited by Dkk1b overexpression ( Figure 7B ) . These different methods of blocking Wnt/β-catenin signaling all show that this pathway is necessary for intercalary neuromast formation in the absence of ErbB signaling . The lack of extra neuromast formation after Wnt/β-catenin inhibition could be due to a decrease in cell proliferation or a lack of differentiation . Because ErbB inhibition induces an increase in Wnt/β-catenin pathway expression prior to the increase in interneuromast proliferation ( Figures 3 and 6 ) , we hypothesized that the inability to induce extra neuromasts after Wnt/β-catenin inhibition is due to a lack of proliferation . We performed BrdU incorporation analyses on Tg ( hsp70l:dkk1b ) /Tg ( sox10:DNerbb4 ) /Tg ( SqET20:gfp ) larvae . Larvae were heat shocked at 32 hpf , then raised in BrdU solution until 48 hpf and fixed . Immunostaining for BrdU and GFP shows a significant increase in double labeling in Tg ( sox10:DNerbb4 ) compared to control siblings ( Figure 8A–B , E ) . Such increase in BrdU incorporation is not observed in Tg ( hsp70l:dkk1b ) /Tg ( sox10:DNerbb4 ) double transgenic larvae ( Figure 8D–E ) . Tg ( hsp70l:dkk1b ) larvae show no significant difference in the BrdU index compared to control larvae ( Figure 8C , E ) . Quantification of the number of Tg ( SqET20:gfp ) -positive interneuromast cells shows a significant decrease in Tg ( hsp70l:dkk1b ) /Tg ( sox10:DNerbb4 ) larvae compared to Tg ( sox10:DNerbb4 ) larvae ( Figure 8F ) . These experiments demonstrate that inhibition of Wnt/β-catenin prevents the formation of intercalary neuromasts in ErbB signaling deficient larvae by inhibiting proliferation of progenitor cells . 10 . 7554/eLife . 01832 . 021Figure 8 . Wnt/β-catenin signaling is required for proliferation of interneuromast cells in the absence of ErbB signaling . Control sibling , Tg ( sox10:DNerbb4 ) , Tg ( hsp70l:dkk1b ) or Tg ( sox10:DNerbb4 ) /Tg ( hsp70l:dkk1b ) , all with Tg ( SqET20:gfp ) in the background , were heat shocked at 32 hpf then placed in BrdU solution . Fish were fixed at 48 hpf and processed for BrdU and GFP immunohistochemistry ( A–D ) . Compared to control siblings ( A ) , Tg ( sox10:DNerbb4 ) ( B ) show a strong increase in BrdU incorporation into Tg ( SqET20:gfp ) positive interneuromast cells . BrdU incorporation in Tg ( sox10:DNerbb4 ) is blocked by over expression of Dkk1b ( D ) . ( E ) Quantification of BrdU index shows no difference between control and Tg ( hsp70l:dkk1b ) ( Student’s t-test , p=0 . 6 ) but a large decrease between Tg ( sox10:DNerbb4 ) and Tg ( sox10:DNerbb4 ) /Tg ( hsp70l:dkk1b ) ( Student’s t-test , p=3 . 4 E−8 ) . ( F ) Quantification of Tg ( SqET20:gfp ) cell number again shows no difference between control and Tg ( hsp70l:dkk1b ) ( Student’s t-test , p=0 . 14 ) but a large decrease between Tg ( sox10:DNerbb4 ) and Tg ( sox10:DNerbb4 ) /Tg ( hsp70l:dkk1b ) ( Student’s t-test , p=1 E−4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 021 To examine if Wnt/β-catenin signaling is sufficient to induce proliferation we treated 48 hpf larvae with BIO . In DMSO treated larvae lef1 is expressed in primII and weakly in a few interneuromast cells ( Figure 9A , arrowhead ) . BIO treatment induced expression of lef1 in neuromasts and interneuromast cells ( Figure 9B , Figure 4—figure supplement 3 ) . To study proliferation in interneuromast cells Tg ( SqET20:gfp ) larvae were treated with DMSO or BIO in the presence of BrdU . Similarly to ErbB inhibition , BIO induces an increase in proliferation and the number of interneuromast cells 24 hr post treatment ( Figure 9C–J ) . BIO treatment does not result in a reduction of Schwann cells along the lateral line ( Figure 9—figure supplement 1 ) . The BIO-induced increase in interneuromast cell proliferation did not result in extra neuromasts , due to a strong increase in cell death in interneuromast cells after prolonged BIO treatment ( data not shown ) . We also transplanted apcmcr mutant lateral line cells into wild type hosts and similarly observed that these clones did not survive more than 48 hpf ( data not shown ) . This increase in cell death was only seen in interneuromast cells and not in primary neuromasts , suggesting that interneuromast cells are particularly sensitive to the levels of Wnt/β-catenin signaling . Combined , our experiments demonstrate that Wnt/β-catenin signaling is sufficient and necessary for inducing interneuromast cell proliferation and is absolutely required for the extra neuromast formation that occurs in the absence of Schwann cells . 10 . 7554/eLife . 01832 . 022Figure 9 . Pharmacological activation of Wnt/β-catenin signaling is sufficient to induce interneuromast cell proliferation . To verify BIO induces Wnt/β-catenin signaling we treated 48 hpf zebrafish with DMSO or BIO for 6 hr and then fixed and stained for lef1 expression . Compared to DMSO treated fish , BIO induces expression of the Wnt/β-catenin target lef1 in neuromasts and interneuromast cells ( A–B ) . The large cluster of cells that are labeled in both ( A ) and ( B ) is primII ( arrowhead ) . To measure proliferation Tg ( SqET20:gfp ) fish were treated with BrdU plus DMSO or BIO at 48 hpf and fixed at 72 hpf . ( C–E ) DMSO treated interneuromast cells ( arrow ) show single chain morphology with rare BrdU incorporation . BrdU incorporation is strong in primary neuromasts . ( F–H ) BIO treated fish show increased interneuromast cells with BrdU incorporation ( arrows ) . Quantification of BrdU index ( I , Student’s t-test , p=0 . 0003 ) and Tg ( SqET20:gfp ) positive cell number ( J , Student’s t-test , p=2 . 75 E−5 ) shows a significant increase in both after BIO treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 02210 . 7554/eLife . 01832 . 023Figure 9—figure supplement 1 . Schwann cells are still present after BIO treatment . Tg ( foxd3:gfp ) larvae were treated with DMSO or BIO at 32 hpf then imaged at 48 hpf . ( A ) DMSO treated larvae show a line of Tg ( foxd3:gfp ) positive Schwann cells along the midline . ( B ) BIO treated larvae also show Schwann cells along the midline . Schwann cell migration is not complete due to stalling of the primordium . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 023 Fgf signaling is upregulated in interneuromast cells in nrg1-3z26 mutant larvae ( Figure 4 ) . To test if Fgf signaling is also required for intercalary neuromast formation we used both genetic and pharmacological methods to block Fgfr signaling . We added the Fgfr inhibitors SU5402 or PD173074 to 32 hpf Tg ( SqET20:gfp ) or Tg ( SqET20:gfp ) /Tg ( sox10:DNerbb4 ) larvae and observed neuromasts at 3 dpf . DMSO treated Tg ( sox10:DNerbb4 ) larvae show formation of intercalary neuromasts ( Figure 10B , arrowheads ) . In contrast , intercalary neuromast formation is blocked after Fgfr inhibition in Tg ( sox10:DNerbb4 ) ( Figure 10D , F ) . In Fgfr inhibitor treated Tg ( sox10:DNerbb4 ) larvae a thickening of the chain of interneuromast cells occurs that is not observed in Fgfr inhibited siblings ( Figure 10C–F , arrows ) . However , this increase in interneuromast cells does not lead to the formation of rosettes or differentiated neuromasts . Quantification of neuromast numbers revealed a significant reduction in Fgfr inhibitor treated Tg ( sox10:DNerbb4 ) larvae ( Figure 10G , red bars ) . Fgfr is required for maintenance of rosettes in the primordium ( Nechiporuk and Raible , 2008 ) . SU5402 and to a lesser extent PD173074 , also induced loss of the rosette shape of primary neuromasts , decreasing the number of neuromasts in treated control siblings ( Figure 10G , blue bars ) . As another means to inhibit Fgf signaling , we heat shock induced dominant negative Fgfr1 expression in 32 hpf Tg ( sox10:DNerbb4 ) and sibling larvae . Dominant negative Fgfr1 also inhibited extra neuromast formation in Tg ( sox10:DNerbb4 ) larvae by 3 dpf ( Figure 10H ) . The observation that Fgfr inhibited Tg ( sox10:DNerbb4 ) larvae still form large clusters of interneuromast cells that fail to differentiate into neuromasts , shows that Fgf signaling is crucial for interneuromast differentiation and rosette formation , but not proliferation . 10 . 7554/eLife . 01832 . 024Figure 10 . Fgf signaling is required for neuromast formation but not interneuromast proliferation in the absence of ErbB signaling . Control Tg ( SqET20:gfp ) or Tg ( sox10:DNerbb4 ) /Tg ( SqET20:gfp ) larvae were treated for 24 hr starting at 32 hpf with DMSO , SU5402 or PD173074 then allowed to develop until 3 dpf and imaged ( A–F ) . The first primary neuromast is to the left in all images . Control siblings ( A ) show no intercalary neuromasts , while Tg ( sox10:DNerbb4 ) have several ( B , arrowhead ) . Tg ( sox10:DNerbb4 ) treated with SU5402 ( D ) or PD173074 ( F ) show clumps of interneuromast cells ( arrows ) compared to sibling SU5402 ( C ) or PD173074 ( E ) treated but no intercalary neuromasts . ( G ) Quantification of neuromast number up to somite 14 shows a decrease when control siblings are treated with SU5402 or PD173074 ( blue bars , One-way ANOVA with Tukey pairwise comparison , p≤0 . 05 ) . ( G ) Extra neuromast formation is inhibited when Tg ( sox10:DNerbb4 ) fish are treated with SU5402 or PD173074 ( red bars , one-way ANOVA with Tukey pairwise comparison , p≤0 . 05 ) . ( H ) Tg ( sox10:DNerbb4 ) were crossed to Tg ( hsp70l:dnfgfr1-EGFP ) and larvae were heat shocked at 32 hpf then allowed to grow to 3 dpf . DNFgfr1 reduces neuromasts slightly in control siblings ( blue bars , Student’s t-test , p=7 . 1 E−9 ) and completely blocks extra neuromast formation in Tg ( sox10:DNerbb4 ) ( red bars , Student’s t-test , p=1 . 5 E−7 ) . To measure proliferation , control Tg ( SqET20:gfp ) or Tg ( sox10:DNerbb4 ) /Tg ( SqET20:gfp ) larvae were treated with BrdU plus DMSO , SU5402 or PD173074 at 32 hpf then fixed at 48 hpf . Fish were then processed for BrdU and GFP immunohistochemistry ( I–N ) . All Tg ( sox10:DNerbb4 ) /Tg ( SqET20:gfp ) treated fish ( J , L and N ) show higher BrdU incorporation when compared to control Tg ( SqET20:gfp ) siblings ( I , K and M ) . ( O ) Quantification of BrdU index shows no significant difference between DMSO , SU5402 or PD173074 treated Tg ( sox10:DNerbb4 ) larvae ( red bars , one-way ANOVA with Tukey pairwise comparison , p≥0 . 1 ) . ( P ) Quantification of Tg ( SqET20:gfp ) cell number shows no difference in Tg ( sox10:DNerbb4 ) larvae treated with DMSO , SU5402 or PD173074 ( red bars , one-way ANOVA with Tukey pairwise comparison , p≥0 . 9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 02410 . 7554/eLife . 01832 . 025Figure 10—figure supplement 1 . Posterior lateral line ganglion neurons are affected differently by SU5042 or PD173074 treatment . Tg ( HGN39D:gfp ) transgenics were treated with indicated drugs at 48 hpf , allowed to develop for 24 hr then imaged . DMSO ( A ) and AG1478 ( B ) have similar ganglion size . SU5402 reduces ganglion size ( C ) , while PD173074 increases ganglion size ( D ) . ( E ) Quantification of GFP positive cells per ganglion . There is no significant difference between DMSO and AG1478 ( one-way ANOVA with Tukey pairwise comparison , p=0 . 9 ) . SU5402 induces a reduction in cell number ( one-way ANOVA with Tukey pairwise comparison , p≤0 . 05 ) . PD173074 induces a mild increase in cell number ( one-way ANOVA with Tukey pairwise comparison , p≤0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 025 To test if Fgf signaling is required for proliferation we performed BrdU analyses of Fgfr inhibitor treated Tg ( SqET20:gfp ) and Tg ( sox10:DNerbb4 ) /Tg ( SqET20:gfp ) larvae . BrdU plus DMSO , SU5402 or PD173074 were added at 32 hpf , and the fish were fixed and stained at 48 hpf . Non-transgenic siblings treated with DMSO or Fgfr inhibitors display a single chain of interneuromast cells with rare BrdU positive cells ( Figure 10I , K , M ) . SU5402 treated wild type larvae show an increase in BrdU incorporation compared to DMSO treated larvae . The increase is significant by a Student’s t-test but not by an ANOVA analysis that compares all groups . PD173074 treated wild type fish show no significant change in BrdU incorporation compared to DMSO ( Figure 10O , blue bars ) . When characterizing SU5402 and PD173074 treated fish we noticed an effect on the posterior lateral ganglion ( pllg ) size . Quantification of the number of cells in the pllg showed a decrease induced by SU5402 but an increase induced by PD173074 ( Figure 10—figure supplement 1 ) . The increase in the posterior lateral line ganglion size after PD173074 treatment is similar to recent results that showed that Fgf inhibition resulted in a larger statoacoustic ganglion ( Vemaraju et al . , 2012 ) . Possibly , the two inhibitors affect a different set of Fgf receptors or pathways . Irrespective of their opposing effect on ganglia size , both SU5402 and PD173074 treated Tg ( sox10:DNerbb4 ) transgenic larvae show BrdU positive clusters of interneuromast cells ( Figure 10J , L , N ) . Likewise , quantification of BrdU indices and interneuromast cell number indicates that neither inhibitor significantly decreases the proportion of proliferating cells or cell number in Tg ( sox10:DNerbb4 ) larvae ( Figure 10O–P , red bars ) . These results confirm that Fgf signaling is not required for the increase in interneuromast cell proliferation in Schwann cell deficient zebrafish . In addition to rosette formation , Fgf signaling also regulates atoh1a , which is crucial for sensory hair cell differentiation in the mouse and zebrafish ear and lateral line ( Bermingham et al . , 1999; Sarrazin et al . , 2006; Millimaki et al . , 2007; Lecaudey et al . , 2008; Nechiporuk and Raible , 2008 ) . We tested if Fgf signaling plays similar roles in postembryonic and precocious intercalary neuromast formation in wild type and Schwann cell deficient larvae , respectively . We examined the expression of the Fgf targets pea3 and atoh1a after DMSO or PD173074 treatment from 32 to 48 hpf in wild type and Tg ( sox10:DNerbb4 ) larvae ( Figure 11 ) . In DMSO treated larvae pea3 and atoh1a are only expressed in primary neuromasts but not in interneuromast cells ( Figure 11A , E , inset ) . After Fgfr inhibition the expression of pea3 and atoh1a are downregulated in primary neuromasts ( Figure 11B , F , inset ) . pea3 and atoh1a are strongly upregulated in precociously differentiating interneuromast cells of DMSO treated Tg ( sox10:DNerbb4 ) larvae ( Figure 11C , G ) . However , this induction of pea3 and atoh1a in Tg ( sox10:DNerbb4 ) larvae is blocked by Fgfr inhibition ( Figure 11D , H ) . Therefore , Fgf signaling has the same function in rosette formation and hair cell differentiation in postembryonic intercalary neuromast formation , as in primary neuromast formation during primordium migration . 10 . 7554/eLife . 01832 . 026Figure 11 . Fgf signaling is required for neuromast differentiation . Control or Tg ( sox10:DNerbb4 ) siblings were treated with DMSO or PD173074 at 32 hpf then fixed at 48 hpf . To verify that Fgf signaling was blocked , we performed in situ hybridazation for the Fgf target pea3 ( A–D ) . ( A ) In controls treated with DMSO , pea3 is not expressed in interneuromast cells but is expressed in neuromasts ( inset ) . ( B ) Control siblings treated with PD173074 show downregulation of pea3 in neuromasts ( inset ) . ( C ) As shown for nrg1-3z26 , Tg ( sox10:DNerbb4 ) have an upregulation of pea3 within interneuromast cells . ( D ) This upregulation of pea3 is blocked by PD173074 , illustrating that Fgfr signaling is inhibited . ( E ) In controls , atoh1a is only expressed in neuromasts ( inset ) . ( F ) PD173074 decreases atoh1a in primary neuromasts ( inset ) . ( G ) atoh1a is upregulated in differentiating interneuromast cells in Tg ( sox10:DNerbb4 ) . ( H ) This upregulation of atoh1a in interneuromast cells is completely blocked by PD173074 , while some expression is still retained in primary neuromasts ( inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 026
Interneuromast cells are a latent lateral line progenitor cell population capable of giving rise to all cell types of the neuromast . We uncovered a novel mechanism whereby Schwann cells constitute part of an inhibitory niche , regulating interneuromast cell proliferation and differentiation through non-cell-autonomous regulation of Wnt/β-catenin and Fgf signaling downstream of ErbB signaling ( see model Figure 12 ) . 10 . 7554/eLife . 01832 . 027Figure 12 . Model of Schwann cell inhibition of intercalary neuromast formation . ( A ) Schwann cells ( red ) co-migrate with lateral line axons ( yellow ) and interneuromast cells ( green ) . Nrg1-3 present on the axon induces Schwann cell migration and proliferation through activation of ErbB2/3B . As interneuromast cells are deposited they remain in close proximity to Schwann cells . This interaction induces inhibition of Wnt/β-catenin signaling by an unknown mechanism . ( B ) As interneuromast cells migrate ventrally away from Schwann cells , this Wnt/β-catenin inhibition is released . ( C ) Release of inhibition leads to increased Wnt/β-catenin signaling and proliferation followed by Fgf signaling and differentiation . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 027 ErbB receptor expressing Schwann cells co-migrate with the primordium due to the presence of Nrg1-3 ligand in lateral line axons . As interneuromast cells are deposited by the primordium , they are in close contact with Schwann cells . Schwann cells keep interneuromast cells in a quiescent state via the action of an unknown inhibitor ( Figure 12A ) . After deposition , interneuromast cells migrate ventrally away from the midline , which coincides with the commencement of proliferation in interneuromast cells ( Figure 12B ) . Simultaneously , Schwann cells migrate medially , crossing the basement membrane ( Raphael et al . , 2010 ) . Both of these steps likely contribute to the release of the Schwann cell inhibitory signal during normal development . The release of this inhibitory signal causes the upregulation of Wnt/β-catenin signaling leading to interneuromast proliferation ( Figure 12C ) . Eventually , Notch and Fgf signaling is initiated inducing rosette formation and differentiation . In zebrafish , ErbB signaling serves several functions during the lifetime of a Schwann cell ( Lyons et al . , 2005; Raphael et al . , 2011 ) . Our ErbB inhibitor treatment , after Schwann cell migration is complete , suggests that ErbB signaling plays an additional role in Schwann cells by regulating an inhibitor of Wnt/β-catenin signaling in interneuromast cells . As pharmacological inhibition of ErbB signaling also causes a reduction in Schwann cell proliferation , it is possible that this secondarily affects interneuromast proliferation via reduction of the number of Schwann cells . However , pharmacological inhibition of ErbB signaling induces an increase in Wnt/β-catenin signaling before we observe a decrease in Schwann cell number ( Figure 6 and Figure 1—figure supplement 4 ) , suggesting that ErbB signaling directly regulates the expression or activity of the Wnt/β-catenin inhibitor . The inhibitor could either be expressed by Schwann or interneuromast cells . The list of potential candidates for the Wnt/β-catenin pathway inhibitor is growing ( Cruciat and Niehrs , 2013 ) . The Wnt/β-catenin inhibitor controlling interneuromast progenitor proliferation should be expressed in wild type Schwann cells until at least 48 hpf , when intercalary neuromasts begin to form . In embryos in which ErbB signaling is abrogated and interneuromast cells proliferate , the Wnt/β-catenin inhibitor should be downregulated . However , the known Wnt/β-catenin inhibitors sfrp1a , wif1 , dkk1b and dkk2 are upregulated in nrg1-3z26 mutants or Tg ( sox10:DNerbb4 ) larvae and wif1 , dkk1b and dkk2 are not expressed in 48 hpf wild type interneuromast or Schwann cells ( Figure 13 ) . Therefore , the expression patterns of these inhibitors do not correlate with a function in inhibiting Wnt/β-catenin signaling in 48 hpf wild type interneuromast cells or the release of inhibition in nrg1-3z26 mutant larvae . Our results are consistent with several recent reports that have shown the importance for Wnt/β-catenin signaling in regulating proliferation in mammalian inner ear support cells and zebrafish neuromasts ( Chai et al . , 2012; Jacques et al . , 2012; Jan et al . , 2013; Shi et al . , 2012; Head et al . , 2013; Wada et al . , 2013b ) . We identified wnt10a as the potential ligand required for intercalary neuromast formation , which should be either inhibited by Schwann cells or induced in their absence . wnt10a is expressed in interneuromast cells of Schwann cell deficient mutants ( Figure 4A–B , Figure 4—figure supplement 1 ) and is induced 6 hr post AG1478 treatment ( data not shown ) . However , even though lef1 and the Wnt/β-catenin reporter are expressed in wild type interneuromast cells at 48 hpf ( Figure 6 ) , we did not detect wnt10a expression , suggesting that additional Wnt ligands upregulate Wnt/β-catenin signaling at this time . 10 . 7554/eLife . 01832 . 028Figure 13 . Expression of several Wnt/β-catenin inhibitors are increased in nrg1-3z26 . Control sibling and nrg1-3z26 mutants were fixed at 48 hpf and processed for in situ hybridization . Control larvae show sfrp1a expression in interneuromast and mantle cells ( A ) . Expression of sfrp1a is increased in nrg1-3z26 ( B ) . Control larvae show no expression of wif1 ( C ) or dkk1b ( F ) in interneuromast or Schwann cells . Both wif1 ( D ) and dkk1b ( E ) are induced in interneuromast cells of nrg1-3z26 . ( G ) Control larvae show expression of dkk2 in primary neuromasts ( inset ) but not in interneuromast or Schwann cells . ( H ) In Tg ( sox10:DNerbb4 ) dkk2 is expressed in neuromasts ( inset ) , and is upregulated in interneuromast cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01832 . 028 Formation of the adult zebrafish pigment pattern depends on ErbB3b , in part non-cell-autonomously , by regulating the formation of dorsal root ganglion neurons , which act as a niche for adult melanophore precursors ( Budi et al . , 2011; Dooley et al . , 2013 ) . We found that nrg1-3z26 mutants mimic the erbb3b pigment pattern phenotype ( Figure 1—figure supplement 2 ) . Whether Nrg1-3 is required for melanophore progenitor niche formation or maintenance is currently unknown . The signaling pathways regulating the niche downstream of ErbB3b have not been identified , but it is interesting to speculate that Wnt/β-catenin signaling may also be involved . Another ErbB family member , ErbB1 , inhibits neural stem cell proliferation non-cell-autonomously through inhibition of Notch activity ( Aguirre et al . , 2010 ) . Notch signaling serves several functions in the development of the lateral line and therefore , is another likely candidate to regulate intercalary neuromast formation . Notch signaling is not active in wild type interneuromast cells but is activated in nrg1-3z26 mutants ( Figure 4 ) . However , the time course expression analysis after ErbB inhibition uncovered that Notch signaling is not upregulated until 24 hr post treatment , well after the initiation of interneuromast cell proliferation ( Figures 3 and 6 , Figure 4—figure supplement 1 ) . In addition , pharmacological inhibition of Notch activity , using the γ-secretase inhibitors DAPT or LY411575 , did not induce Wnt or Fgf ligand expression ( Figure 4—figure supplement 2 ) or intercalary neuromasts in wild type larvae ( data not shown ) . Therefore we conclude that , similar to Fgf , Notch signaling is only playing a later role in intercalary neuromast development , likely regulating the choice between support or hair cell specification as in primary neuromasts ( Itoh and Chitnis , 2001 ) . Recently , it was shown that non-myelinating Schwann cells constitute part of a niche , non-cell-autonomously controlling the quiescence of hematopoietic stem cells ( HSC ) ( Yamazaki et al . , 2011 ) . Removing Schwann cells in mice by axotomy lead to loss of active TGF-β and an increase in HSC proliferation ( Yamazaki et al . , 2011 ) . The TGF-β pathway has been implicated to interact with ErbB signaling ( Chow et al . , 2011 ) . tgfb1a is also robustly expressed in the migrating lateral line primordium and we therefore examined if this pathway acts as an intermediary between ErbB and Wnt/β-catenin signaling ( ZFIN , http://zfin . org ) . We treated wild type larvae with two pharmacological inhibitors of Tgf-β receptors , SB505124 or SB431542 ( Hagos and Dougan , 2007 ) . Neither drug induced an increase in interneuromast number or clusters of interneuromast cells , suggesting that Tgf-β signaling does not inhibit interneuromast cell proliferation and is not involved in their regulation ( data not shown ) . In Drosophila , glial cells control quiescence and proliferation of neural progenitors depending on the developmental stage ( Doe , 2008 ) . This non-cell-autonomous interaction is regulated by ErbB ( spitz/TGF-α ) , HSPGs ( trol/perlecan ) , and a secreted glycoprotein , anachronism ( Ebens et al . , 1993; Voigt et al . , 2002; Morante et al . , 2013 ) . Based on our recent results it would be interesting to examine if these molecules or pathways interact with the Wnt/β-catenin pathway . For example , Perlecan interacts with Wnt/β-catenin and Fgf signaling , however the other pathways have not been examined yet ( Park et al . , 2003; Kamimura et al . , 2013 ) . We find that a set of the same signaling pathways are involved in neuromast differentiation from interneuromast cells as in neuromast formation in the primordium . However , primary neuromasts form within the migrating primordium and are deposited , whereas interneuromast cells proliferate to form a cluster of progenitor cells that eventually differentiate ( Video 1 ) . Therefore some differences exist . Both Fgf and Wnt/β-catenin are required for proliferation in the primordium ( Aman et al . , 2011 ) , whereas interneuromast proliferation depends on Wnt/β-catenin signaling only . Nevertheless , both neuromast types require Fgf signaling for rosette formation and hair cell specification ( Figure 10A–F; 11E–H and Lecaudey et al . , 2008; Nechiporuk and Raible , 2008 ) . Fgf signaling is induced by Wnt/β-catenin signaling in the primordium , and likely also in interneuromast cells ( Figure 4—figure supplement 2 and 3 ) . However , it is not upregulated for many hours after the Wnt/β-catenin pathway is activated post ErbB inhibition ( Figures 5 and 6 ) or after BIO treatment ( Figure 4—figure supplement 3 ) . Also , activation of the Wnt/β-catenin pathway by GSK-3 inhibition does not induce pea3 in interneuromast cells during the first 24 hr post treatment ( data not shown ) . Therefore , in contrast to the primordium , Fgf may not be a direct target of Wnt/β-catenin signaling in interneuromast cells . Neuromast formation in the primordium occurs in an existing and migrating tissue whereas intercalary neuromasts form from a string of cells . Therefore , it is not surprising that some of the signaling interactions between these two modes of neuromast formations are different . A detailed analysis of ErbB receptor functions is important as upregulation of ErbB family receptor tyrosine kinases is characteristic of many human cancers , notably breast cancer , resulting in enhanced tumorigenesis . Consequently , ErbB receptors are important therapeutic targets ( Moasser , 2007 ) . Although cancers typically originate from unique cell types , tumors can be heterogeneous , containing multiple cell types . As the repertoire of pharmacological kinase inhibitors that are being developed to treat cancer increases , it becomes crucial to take into consideration that different ErbB receptor/ligand combinations can result in counterproductive cellular responses . Our results from the zebrafish lateral line stress the importance that ErbB signaling also has non-cell-autonomous , anti-proliferative functions , and that these anti-proliferative functions play a role during the regulation of neuronal progenitor regulation during development . In conclusion , we identified that ErbB signaling in Schwann cells non-cell-autonomously regulates a molecular signaling network consisting of Wnt/β-catenin , Fgf and Notch pathways within neural progenitors , regulating their quiescence and activation . Given how conserved signaling interactions are during development and across species , it is tempting to speculate that these four pathways also interact in other neural progenitor cell populations .
We used the following fish strains; rowgain/erbb2 and hypersensitive/erbb3b ( Grant et al . , 2005 ) , nrg1-3z26 ( Perlin et al . , 2011 ) , erbb2st61 ( Lyons et al . , 2005 ) , Tg ( foxd3:gfp ) zf15 ( Gilmour et al . , 2002 ) , Tg ( clndB:lyngfp ) zf106 ( Haas and Gilmour , 2006 ) , Et ( krt4:EGFP ) sqet20 ( Parinov et al . , 2004 ) , Tg ( hsp70l:dkk1b-GFP ) w32 ( Stoick-Cooper et al . , 2007 ) , Tg ( hsp70l:dnfgfr1-EGFP ) pd1 ( Lee et al . , 2005 ) , Tg ( OTM:d2EGFP ) kyu1 referred to as Tg ( Tcf/Lef-miniP:dGFP ) ( Shimizu et al . , 2012 ) , cntnap2ankhgn39dET referred to as HGN39D ( Nagayoshi et al . , 2008 ) and lef1zd11 ( Wang et al . , 2012 ) . To generate the Tg ( sox10:DNhsaerbb4-RFP ) , we used the zebrafish Tol2 kit ( Kwan et al . , 2007 ) . We cloned the zebrafish 7 . 2 kb sox10 promoter , obtained from Thomas Carney , into the 5′ entry vector . Human dominant-negative ErbB4 ( hsaDNerbb4 ) construct was a gift of Gabriel Corfas ( Rio et al . , 1997 ) . The flag tag on hsaDNerbb4 was replaced with mRFP and cloned into the middle entry vector . To generate the final vector the 5′ entry , middle entry and 3′ polyA entry vector were recombined with the destination vector containing the cmcl2:gfp expression cassette , in order to identify transgenics based on GFP expression in the heart ( Kwan et al . , 2007 ) . The recombined vector was then injected into one cell stage embryos from the Tubingen strain along with transposase mRNA . Founder GFP positive heart carriers were raised and identified by crossing to wild type fish . The positive carriers have been maintained over three generations . In situ hybridization was performed as previously described ( Kopinke et al . , 2006 ) . The following probes were used lef1 , fgf3 , fgf10 , fgfr1a , pea3 , atoh1a , dkk1b , klf4 ( Aman and Piotrowski , 2008 ) , sfrp1a ( Tendeng and Houart , 2006 ) , mbp ( Brosamle and Halpern , 2002 ) , dkk2 ( Wada et al . , 2013b ) . To clone additional probes the following primers were used , for myca; forward 5′-ggtcctggacactccaccta-3′ and reverse 5′-atgcactctgtcgccttctt-3′ , beta-catenin-2 ( ctnnb2 ) ; forward 5’-cgactctgctcatccaacaa-3’ and reverse 5′-aggatctgcaggcagtctgt-3′ , wnt10a; forward 5′-cttcagcaggggtttcagag-3′ and reverse 5′-tccctggctggtcttgttac-3′ , wif1; forward 5′-aaccaaaggatggxtttcagg-3′ and reverse 5′- aggtttaaaccacatagttggtttcag-3′ . For low magnification images , individual images were stitch together using ImageJ . For alkaline phosphatase staining larvae were fixed overnight at room temperature in 4% paraformaldehyde . Larvae were then washed three times 5 min each in PBS/0 . 3% Tween-20 followed by three 5-min washes in staining buffer ( 50 mM MgCl2 , 100 mM NaCl , 100 mM Tris pH 9 . 5 and 0 . 1% Tween-20 ) . Larvae were then placed in staining buffer plus NBT/BCIP ( Roche , USA ) and stained at room temperature in the dark . The staining reaction was stopped with 4% paraformaldehyde . To label hair cells embryos were placed in a 0 . 06 mg/ml solution of DASPEI ( 2- ( 4- ( dimethylamino ) styryl ) -N-ethylpyridinium iodide , [Invitrogen , USA] ) diluted in embryo media for 10 min . Embryos were then briefly washed and anesthetized with Tricane for imaging under a dissecting fluorescent microscope . For low magnification images , individual images were stitch together using ImageJ . Transplantation assays were performed as previously described ( Aman and Piotrowski , 2008 ) . Donor embryos were injected with 5% Alexa-568 and 3% lysine-fixable biotinylated-dextran ( Invitrogen , USA ) at the one cell stage . Tg ( foxd3:gfp ) or Tg ( clndB:lyngfp ) wild type donors were used to test rescue of erbb2 mutant hosts . Tg ( clndB:lyngfp ) donors were used to test rescue of nrg1-3z26 mutant hosts . Host embryos were screened for lateral line or Schwann cell clones at 24 hr post fertilization ( hpf ) . Embryos were imaged at 48 hpf to record the extent of the transplanted clone . To identify mutant hosts or donors , neuromasts were counted after DASPEI staining at 4 days post fertilization ( dpf ) . Both transplanted and untransplanted sides of the hosts were imaged with a Zeiss LSM 510 or 780 confocal microscope at 20X . For low magnification images , individual images were stitch together using ImageJ . All chemical inhibitors were added to embryo media with a final concentration of 1% DMSO . Negative controls consisted of 1% DMSO only . The ErbB inhibitor AG1478 was used at 3 μM . The Fgf receptor inhibitors PD173074 or SU5402 were added at 100 μM or 10 μM respectively . The Wnt/β-catenin inhibitors IWR-1 or XAV939 were used at 40 μM and 20 μM respectively . The GSK-3 inhibitor BIO was used at 2 μM . The γ-secretase inhibitor LY411575 was used at 100 μM . LY411575 was obtained from Santa Cruz ( USA ) and all other inhibitors were purchased from Tocris ( USA ) . BrdU incorporation was performed in Tg ( foxd3:gfp ) or Tg ( SqET20:gfp ) transgenics by addition of 10 mM BrdU ( Sigma , USA ) with 1% DMSO or chemical inhibitors for various lengths of times as indicated in the text . Embryos were fixed in 4% paraformaldehyde overnight . BrdU immunostaining was performed as described except embryos were treated for 15 min with proteinase K ( Aman et al . , 2011 ) . To visualize GFP , larvae were also immunostained with rabbit anti-GFP ( Invitrogen , USA ) at 1/400 dilution . All embryos were counterstained with DAPI ( Invitrogen , USA ) . To quantify BrdU index we counted BrdU and GFP double positive cells between the first and second deposited neuromasts . Immunostained embryos were imaged with a Zeiss LSM 510 or 780 confocal microscopes at 40X . Heat shock induction was done at various developmental ages as indicated in the text . Embryos were placed at 39°C for 20 min , room temperature for 20 min and then another 20 min at 39°C . Embryos were then allowed to develop at 28 . 5°C . Time-lapse imaging was performed similar as described ( Aman and Piotrowski , 2008 ) . Tg ( SqET20:gfp ) or Tg ( SqET20:gfp ) /Tg ( sox10:DNhsaerbb4-RFP ) were anesthetized with Tricaine and mounted in 0 . 8% low melting agarose on glass bottom dishes ( MatTek , USA ) . Embryos were imaged with a Zeiss 710 or 780 confocal microscopes using a 40X water objective in a climate-controlled chamber set to 28°C . | All the different types of cells that make up the body of an animal are descended from a single fertilized egg . As this egg develops into an embryo , the cells divide and specialize to become a specific type of cell , such as: a liver cell , a muscle cell or a nerve cell . The cells in the embryo that are destined to become specific cell types are called progenitor cells . However , these cells are also found within adult tissues , where they wait until they are needed to replace old or damaged cells . Zebrafish are commonly used in scientific research and , like other fish , they have a ‘lateral line’ that runs along both sides of the body and contains cells that detect movements in the surrounding water . During its development , the lateral line contains many progenitors that are primed to form more of these sense organs . The lateral line is also connected to nerve cells that relay information about water movements to the central nervous system , while other cells called Schwann cells support the nerve cells . The local environment or ‘niche’ created by the Schwann cells is known to prevent the progenitor cells within the lateral line from becoming their specific cell type too early . However , the molecules that cause progenitor cells to stop dividing , and later restart dividing and change in to their predestined cell type is not well understood . Now Lush and Piotrowski have discovered that signaling through a protein called ErbB causes the Schwann cells to multiply , but has the opposite effect on nearby progenitor cells in the lateral line . ErbB signaling in the Schwann cells inhibited various signaling pathways in the progenitor cells; and whilst some of these pathways normally encourage the progenitors to multiply , others cause them to change into their specific cell type . The findings of Lush and Piotrowski have important implications for understanding how the interactions between progenitor cells and the cells around them affect their development . These findings may be useful for understanding diseases caused when the control of cell multiplication or cell-type changes goes awry—such as developmental abnormalities or cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine"
] | 2014 | ErbB expressing Schwann cells control lateral line progenitor cells via non-cell-autonomous regulation of Wnt/β-catenin |
The eukaryotic 43S pre-initiation complex bearing tRNAiMet scans the mRNA leader for an AUG start codon in favorable context . Structural analyses revealed that the β-hairpin of 40S protein Rps5/uS7 protrudes into the 40S mRNA exit-channel , contacting the eIF2∙GTP∙Met-tRNAi ternary complex ( TC ) and mRNA context nucleotides; but its importance in AUG selection was unknown . We identified substitutions in β-strand-1 and C-terminal residues of yeast Rps5 that reduced bulk initiation , conferred ‘leaky-scanning’ of AUGs; and lowered initiation fidelity by exacerbating the effect of poor context of the eIF1 AUG codon to reduce eIF1 abundance . Consistently , the β-strand-1 substitution greatly destabilized the ‘PIN’ conformation of TC binding to reconstituted 43S·mRNA complexes in vitro . Other substitutions in β-hairpin loop residues increased initiation fidelity and destabilized PIN at UUG , but not AUG start codons . We conclude that the Rps5 β-hairpin is as crucial as soluble initiation factors for efficient and accurate start codon recognition .
Accurate identification of the translation initiation codon is critical to ensure synthesis of the correct cellular proteins . In eukaryotic cells this process generally occurs by a scanning mechanism , wherein the small ( 40S ) ribosomal subunit first recruits initiator tRNA ( Met-tRNAi ) in a ternary complex ( TC ) with eIF2-GTP in a reaction stimulated by eIFs 1 , 1A , and 3 . The resulting 43S pre-initiation complex ( PIC ) attaches to the mRNA 5′ end and scans the 5′ UTR for an AUG , using complementarity with the anticodon of Met-tRNAi to identify the start codon and assemble a 48S PIC . Nucleotides immediately surrounding the AUG , particularly the −3 and +4 positions ( referred to below as context nucleotides ) , also influence start codon selection . During scanning , the GTP bound to eIF2 in the TC is hydrolyzed in the 43S PIC in a manner dependent on the GTPase activating protein eIF5 , but Pi release is blocked by eIF1 , which also impedes stable binding of Met-tRNAi in the P site . Start codon recognition triggers dissociation of eIF1 from the 40S subunit , which allows interaction between eIF5 and the C-terminal tail ( CTT ) of eIF1A , Pi release from eIF2-GDP·Pi , and more stable TC binding in the P site ( Figure 1 ) . Subsequent dissociation of eIF2-GDP and other eIFs from the 48S PIC enables eIF5B-catalyzed subunit joining and formation of an 80S initiation complex with Met-tRNAi base-paired to AUG in the P site ( reviewed in Hinnebusch , 2014 ) . 10 . 7554/eLife . 07939 . 003Figure 1 . Model describing conformational rearrangements of the PIC during scanning and start codon recognition . Assembly of the PIC , scanning and start codon selection in WT cells . ( i ) eIF1 and the scanning enhancer ( SEs ) elements in the CTT of eIF1A stabilize an open conformation of the 40S subunit to which the TC loads rapidly . ( ii ) The 43S PIC in the open conformation scans the mRNA for the start codon with Met-tRNAi bound in the POUT state . The GAP domain in the N-terminal domain of eIF5 ( 5N ) stimulates GTP hydrolysis by the TC to produce GDP•Pi , but release of Pi is blocked . The unstructured NTT of eIF2β interacts with eIF1 to stabilize eIF1•40S association and the open conformation . ( iii ) On AUG recognition , Met-tRNAi moves from the POUT to PIN state , clashing with eIF1 and the CTT of eIF1A . Movement of eIF1 and the eIF1A CTT away from the P site disrupts eIF1's interaction with eIF2β-NTT , and the latter interacts with the eIF5-CTD . eIF1 dissociates from the 40S subunit , and the eIF5-NTD disengages from eIF2 and interacts with the eIF1A CTT instead , dependent on the SE elements , thereby facilitating Pi release from eIF2 . The eIF5-CTD moves into the position on the 40S subunit previously occupied by eIF1 and blocks reassociation of eIF1 . ( Below ) Arrows summarize that eIF1 and the eIF1A SE elements promote POUT and block transition to the PIN state , whereas the scanning inhibitor ( SI ) element in the NTT of eIF1A stabilizes the PIN state . ( Adapted from Hinnebusch and Lorsch , 2012; Nanda et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 003 eIF1 plays a dual role in the scanning mechanism . It promotes an open , scanning-conducive conformation of the PIC ( Pestova and Kolupaeva , 2002 ) to which TC rapidly loads , bound in a state capable of inspecting successive triplets entering the P site ( dubbed POUT ) ( Passmore et al . , 2007; Saini et al . , 2010 ) ; and it also blocks recognition of near-cognate start codons ( e . g . , UUG ) ( Yoon and Donahue , 1992 ) and AUG codons in poor sequence context ( Pestova and Kolupaeva , 2002 ) . Hence , eIF1 must dissociate from the 40S subunit ( Maag et al . , 2005; Cheung et al . , 2007 ) to allow Pi release ( Algire et al . , 2005 ) and rearrangement to a scanning-incompatible state with Met-tRNAi base paired with AUG and more tightly bound in the PIN conformation ( Passmore et al . , 2007; Saini et al . , 2010 ) . Consistent with this , structural analyses of different PICs reveal that eIF1 and eIF1A promote rotation of the 40S head relative to the body ( Lomakin and Steitz , 2013 ) ( Hussain et al . , 2014 ) , which is likely instrumental in TC binding in the POUT conformation , but that eIF1 physically obstructs Met-tRNAi binding in the PIN state ( Rabl et al . , 2011; Lomakin and Steitz , 2013 ) . Accordingly , eIF1 is deformed and displaced from its 40S location in the open complex during the POUT to PIN transition ( Hussain et al . , 2014 ) . Consequently , mutations that weaken eIF1 binding to the 40S subunit reduce the rate of TC loading , while elevating initiation at near-cognate codons or AUGs in poor context , by destabilizing the open/POUT conformation and favoring rearrangement to the closed/PIN state during scanning ( Martin-Marcos et al . , 2011 , 2013 ) . Moreover , decreasing wild-type ( WT ) eIF1 abundance reduces initiation accuracy , whereas overexpressing eIF1 suppresses initiation at near cognates or AUGs in poor context ( Valasek et al . , 2004; Alone et al . , 2008; Ivanov et al . , 2010; Saini et al . , 2010; Martin-Marcos et al . , 2011 ) . This tight link between eIF1 abundance and initiation accuracy is exploited to autoregulate eIF1 expression , as the AUG start codon of the eIF1 gene ( SUI1 in yeast ) occurs in poor context—a feature conserved throughout eukaryotic evolution—and the frequency of recognizing its own start codon is inversely related to eIF1 abundance ( Ivanov et al . , 2010; Martin-Marcos et al . , 2011 ) . The stability of the codon-anticodon duplex is an important determinant of initiation accuracy , as the rate of the POUT to PIN transition and stability of the PIN state are both favored by AUG vs non-AUG start codons ( Kolitz et al . , 2009 ) . It is possible that favorable context also contributes to the stability of PIN ( Pisarev et al . , 2006; Martin-Marcos et al . , 2011 ) , but the stimulatory effect of optimum context on initiation rate is not understood at the molecular level . There is evidence that the context nucleotides are recognized by the α-subunit of eIF2 , as replacement of heterotrimeric eIF2 with the eIF2βγ heterodimer reduced the efficiency of AUG recognition and diminished the stimulatory effect of optimum context on 48S PIC assembly in a reconstituted mammalian system ( Pisarev et al . , 2006 ) . Moreover , crosslinking experiments ( Pisarev et al . , 2006; Sharifulin et al . , 2013 ) and structural analyses of a mammalian 43S PIC ( Hashem et al . , 2013 ) and a partial yeast ( py48S ) PIC ( Hussain et al . , 2014 ) indicate that the N-terminal domain ( D1 ) of eIF2α is in proximity to the −3 nucleotide of the mRNA in the exit channel of the 40S subunit . These and other studies ( Lomakin and Steitz , 2013 ) revealed that the conserved β-hairpin of the 40S protein uS7 ( Rps5 in yeast ) lies in the vicinity of eIF2α-D1 and the −3 nucleotide of mRNA in reconstituted 43S/48S PICs ( Figure 2A , B ) ; however , functional evidence that eIF2α-D1 and the Rps5 β-hairpin have important roles in start codon recognition in vivo is lacking . 10 . 7554/eLife . 07939 . 004Figure 2 . Location in the py48S PIC , and sequence conservation , of the Rps5 β-hairpin loop . ( A , B ) Depiction of the partial yeast 48S PIC ( PDB 3J81 ) showing Rps5 ( gold ) , mRNA ( orange ) , Met-tRNAi ( green ) , eIF2α ( purple ) , eIF2γ ( yellow ) , eIF1 ( cyan ) and eIF1A ( blue ) . For clarity other ribosomal proteins , eIF2β and putative eIF5 densities are not shown . Residues implicated here in AUG recognition are shown with stick side-chains and highlighted ( as in panel C ) in blue or pink . ( C ) Alignment of Rps5 sequences from diverse eukaryotes using the Clustal Omega algorithm ( http://www . ebi . ac . uk/Tools/msa/clustalo/ ) . The β- hairpin loop is annotated below the alignment and residues implicated in this study in AUG recognition are highlighted in blue or pink and underlined . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 004 In this report , we establish that the β-hairpin of Rps5 is critically required for both efficient and accurate translation initiation in vivo . Substituting Glu-144 ( E144 ) in β-strand 1 of the hairpin , or the proximal C-terminal residue R225 ( Figure 2B ) , confers a marked reduction in the efficiency of AUG recognition . This defect exacerbates the effect of poor context at the SUI1 mRNA start codon to reduce eIF1 cellular abundance and thereby increase recognition of near-cognate UUG codons as a secondary effect . Analysis in the yeast reconstituted system reveals striking destabilization of the PIN state formed at AUG codons by E144R mutant 40S subunits . Substitutions in the loop portion of the Rps5 β-hairpin also destabilize the PIN state , but produce this effect selectively for near-cognate UUG start codons , and thereby dampen UUG initiation in vivo . These findings indicate that the Rps5 β-hairpin functions on a par with soluble initiation factors , such as eIF1 , eIF1A and eIF2 , to insure efficient and accurate start codon recognition in eukaryotic cells .
To examine the role of the Rps5 β-hairpin in start codon recognition , we introduced single substitutions into 3 residues of β-strand 1 ( 144EDT146 ) and 8 of the 10 residues in the hairpin loop ( 147TR148 and 151GGGARRQ158 ) ( Supplementary file 1 ) . Residues in the β-strands , and the loop residues proximal to the β-strands , are among the most highly conserved in evolution ( Figure 2C ) . We also substituted the last four residues of Rps5 ( 222KSNR225 ) in view of their strong conservation and proximity to the β-hairpin , and because invariant Glu144 in β-strand 1 ( E144 ) appears to form a salt-bridge with C-terminal residue R225 in the yeast 80S ribosome ( Ben-Shem et al . , 2011 ) ( Figure 2B ) . Residues were generally substituted with Ala to shorten the side-chain , or with basic or acidic residues to introduce or alter side-chain charge ( Supplementary file 1 ) . The mutations were generated in an RPS5 allele under its own promoter on a low-copy plasmid and examined in a yeast strain with WT chromosomal RPS5 under a galactose-inducible promoter ( PGAL1 ) . Mutant phenotypes were scored following a switch from galactose to glucose , where PGAL1-RPS5 expression is repressed . Despite strong sequence conservation of many β-hairpin residues ( Figure 2C ) , only the G151S substitution was lethal and prevented growth on glucose medium; however , several substitutions conferred a slow-growth ( Slg− ) phenotype , including E144R and R225K ( Figure 3A , glucose; Supplementary file 1 ) . 10 . 7554/eLife . 07939 . 005Figure 3 . RPS5 mutations E144R and R225K impair translation initiation and elevate UUG initiation without reducing 40S subunit abundance in vivo . ( A ) 10-fold serial dilutions of transformants of PGAL1-RPS5 his4-301 strain ( JVY07 ) with the indicated plasmid-borne RPS5 alleles were spotted on synthetic medium supplemented with histidine and containing galactose ( SGal + His + Ura + Trp ) or glucose ( SD + His + Ura + Trp ) as carbon source and incubated at 30°C for 3 days ( Glucose ) or 4 days ( Galactose ) . ( B ) Strains from ( A ) also harboring HIS4-lacZ reporters with AUG or UUG start codons ( plasmids p367 and p391 , respectively ) were cultured in SD + His + Trp at 30°C to A600 of ∼1 and β-galactosidase specific activities were measured in WCEs in units of nanomoles of o-nitrophenyl- β-D-galactopyranoside ( ONPG ) cleaved per min per mg of total protein . Ratios of mean expression of the UUG and AUG reporters calculated from four transformants are plotted with error bars ( indicating S . E . M . s ) . ( C ) Strains from ( A ) were cultured in SD + His + Ura + Trp at 30°C to A600 of ∼1 , and cycloheximide was added prior to harvesting . WCEs were separated by sucrose density gradient centrifugation and scanned at 254 nm to yield the tracings shown . Mean Polysome/Monosome and 40S/60S ratios ( and S . E . M . s ) from four replicates are indicated . Student's t-test indicates that the mean values for polysome/monosome ratio in the RPS5 mutants are reduced significantly from the WT ( p < 0 . 0005 ) . ( D ) Similar to ( C ) but the cultures were not treated with cycloheximide and lysed in buffers without MgCl2 to allow separation of the dissociated ribosomal subunits . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 005 To identify effects on fidelity of start codon selection , the mutant strains were assayed for expression of otherwise identical HIS4-lacZ reporters containing an AUG or UUG start codon . Substantial ( >twofold ) increases in expression of the UUG relative to AUG reporter ( UUG:AUG ratio ) were observed only for E144R and three different mutations substituting Arg225 . Mutations E144R and R225K elevated the UUG:AUG ratio by 5 . 9- and 3 . 6-fold , respectively ( Figure 3B and Supplementary file 1 ) . To test the importance of the Glu144/Arg225 salt-bridge , we constructed the E144R/R225E double mutant in which the salt-bridge should be reinstated . This strain displayed a Slg− phenotype and increased UUG:AUG ratio similar in magnitude to that of the R225K mutant but less severe than seen for the E144R strain ( Figure 3A , B ) . The fact that combining these mutations did not produce more severe phenotypes than those conferred by E144R alone is consistent with the possibility that reinstating the salt-bridge mitigates the effects of the E144R single mutation , with the stipulation that the WT identity of E144 or R225 is needed for robust Rps5 function . Thus , although the salt-bridge is reinstated , the substitution of one or both residues still impairs growth and initiation fidelity in the double mutant . Additional experiments are needed to establish the importance of the salt bridge for Rps5 function . Mutations in various initiation factors are known that elevate the UUG:AUG ratio and restore translation of mutant his4-301 mRNA , which lacks the AUG start codon , by enabling initiation at the third , UUG codon , thereby suppressing histidine auxotrophy and conferring a Sui− ( Suppressor of initiation codon mutation ) phenotype . However , neither E144R nor R225K suppress the His− phenotype of his4-301 to confer the Sui− phenotype . Based on previous observations of Sui− mutants , it is possible that the Rps5 substitutions do not elevate the UUG:AUG ratio sufficiently to produce enough his4-301 product for adequate histidine biosynthesis ( Dorris et al . , 1995; Martin-Marcos et al . , 2013 ) . For example , the eIF1 mutations sui1-K37A and sui1-R33A increase the UUG:AUG ratio by 4 . 8- and 7 . 7-fold , but only the latter suppresses the His− phenotype of his4-301 ( Martin-Marcos et al . , 2013 ) . Alternatively , the Rps5 mutations could interfere with an unknown aspect of histidine biosynthesis or utilization ( Nanda et al . , 2009 ) . Consistent with their Slg− phenotypes , E144R and R225K conferred significant reductions in the polysome:monosome ( P/M ) ratio ( p-value <0 . 0005 ) ( Figure 3C ) , indicating a reduced rate of bulk translation initiation relative to elongation , with the greater reduction conferred by the mutation ( E144R ) with the stronger Slg− phenotype ( Figure 3A ) . Neither mutant significantly perturbed the ratio of 40S to 60S subunits ( Figure 3C , D ) , suggesting that the initiation defects arise from altered 40S function rather than abnormalities in expression of Rps5 , 40S biogenesis , or stability of mature 40S subunits . Thus , it appears that E144R and R225K reduce the function of Rps5 in stimulating the rate of general translation initiation and promoting accurate start codon selection . In addition to increasing initiation from near-cognate UUG codons , certain Sui− mutations in eIF1 , eIF1A , and eIF2β are known to enhance initiation from AUG codons in poor context . As such , they suppress the effects of the suboptimal context of the AUG codon of SUI1 mRNA and increase expression of the encoded eIF1 protein ( Martin-Marcos et al . , 2011 ) . This phenotype is illustrated for the Sui− sui1-L96P mutation in Figure 4A ( lanes 7 , 8 vs 5 , 6 ) . However , unlike previously described mutations that enhance UUG recognition , rps5-E144R and -R225K paradoxically decrease eIF1 abundance ( Figure 4A , lanes 1–4 vs 5 , 6 ) to a degree that correlates with their elevated UUG:AUG HIS4-lacZ initiation ratios ( Figure 3B ) . Consistently , E144R and R225K reduce expression of a SUI1-lacZ reporter bearing the native , suboptimal context at the three nucleotides preceding the AUG codon ( −3CGU−1 ) , but not that of a modified SUI1opt-lacZ reporter with an optimized AUG context ( −3AAA−1 ) ( Figure 4B ) . Thus , the rps5 mutations exacerbate the effect of suboptimal context and decrease AUG recognition on WT SUI1 mRNA . The reduction in eIF1 abundance implies that the rps5 mutations overcome the autoregulation of eIF1 expression , wherein low eIF1 levels suppress the effect of poor context at the SUI1 AUG codon to boost eIF1 abundance ( Ivanov et al . , 2010; Martin-Marcos et al . , 2011 ) . Accordingly , it appears that the rps5 mutations evoke a pronounced defect in recognition of the native SUI1 AUG codon that prevails even at low cellular concentrations of eIF1 that would normally boost SUI1 translation . 10 . 7554/eLife . 07939 . 006Figure 4 . RPS5 mutations E144R and R225K exacerbate poor context at the native SUI1 AUG to reduce eIF1 expression and indirectly confer Sui- phenotypes , but evoke Ssu- phenotypes when eIF1 abundance is boosted . ( A ) WCEs of strains from Figure 3A , and from sui1-L96P strain H4564 , were subjected to Western analysis using antibodies against eIF1 or Gcd6 ( as loading control ) . Two amounts of each extract differing by a factor of two were loaded in successive lanes . ( B ) Strains from Figure 3A also harboring SUI1-lacZ ( pPMB24 ) or SUI1- opt-lacZ ( pPMB25 ) reporters were cultured and assayed for β-galactosidase activities as described in Figure 3B . Mean expression levels and S . E . M . s from four transformants are plotted , and relative ( Rel . ) mean expression levels normalized to that of the WT strain are listed below the histogram . Student's t-test indicates that the mean values for SUI1-lacZ expression in the RPS5 mutants are reduced significantly from the WT ( ***p < 0 . 0005 ) . ( C ) WCEs of strains from Figure 3A also harboring sc SUI1 plasmid pPMB21 or empty vector were subjected to Western analysis as in ( A ) . Signal intensities were quantified and mean eIF1/Gcd6 ratios are listed below the blot with S . E . Ms ( D ) HIS4-lacZ reporters with AUG or UUG start codons were assayed in strains from ( C ) as in Figure 3B . ( E ) his4-301 strains with the indicated WT or mutant RPS5 alleles ( from Figure 3A ) harboring sc SUI1 plasmid pPMB21 , SUI5 plasmid p4281 , or empty vectors were spotted on SD plates containing ( SD + His ) or lacking histidine ( SD-His ) and incubated for 3 days and 5 days , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 00610 . 7554/eLife . 07939 . 007Figure 4—figure supplement 1 . Increased SUI1 gene dosage partially rescues the Slg- phenotype of RPS5 mutations E144R and R225K . 10-fold serial dilutions of transformants of strains containing the indicated RPS5 alleles from Figure 3A harboring sc SUI1 plasmid pPMB21 or empty vector were spotted on SD + His + Ura and incubated at 30°C for 3 days . All strains were spotted in parallel on plates of identical medium . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 00710 . 7554/eLife . 07939 . 008Figure 4—figure supplement 2 . RPS5 mutation E144R confers a Gcd- phenotype , derepressing GCN4-lacZ expression with restored eIF1 expression . Strains of the indicated RPS5 genotype from Figure 3A harboring either sc SUI1 plasmid pPMB21 or empty vector , and the WT GCN4-lacZ reporter on plasmid p180 , were cultured in duplicate sets in SD + His at 30°C . One set ( unstarved ) was cultured to A600 of ∼1 and harvested . When the second set ( starved ) reached A600 of ∼0 . 5 , sulfometuron methyl ( SM ) was added to 0 . 5 μg/ml and grown for 6 hr and harvested . β-galactosidase specific activities were measured in WCEs as described in Figure 3B . Mean expression levels and S . E . M . s were calculated from four transformants . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 008 Interestingly , the discrimination against native poor context at SUI1 with attendant reduced eIF1 expression represents a hyperaccuracy phenotype displayed by known Ssu− ( Suppressor of Sui− ) mutations in eIF1 , eIF1A , and eIF2β , which additionally suppress the elevated UUG:AUG ratio conferred by various Sui− mutations ( Martin-Marcos et al . , 2011 ) . Hence , the fact that rps5-E144R and -R225K discriminate against poor context at SUI1 but elevate the UUG:AUG ratio seems paradoxical . However , their elevated UUG:AUG ratio ( hypoaccurate ) phenotype could be explained by the reduced levels of eIF1 present in these rps5 mutants ( Hinnebusch , 2011 ) . Indeed , we found that increasing the level of WT eIF1 by adding an extra plasmid-borne copy of WT SUI1 ( Figure 4C , scSUI1 vs Vector transformants ) mitigated the Sui− phenotypes of both rps5 mutants , reducing their UUG:AUG ratios to essentially WT levels ( Figure 4D ) . Introducing scSUI1 also mitigated their Slg− phenotypes , slightly for -R225K and substantially for -E144R ( Figure 4—figure supplement 1 ) . The resulting rps5/scSUI1 strains exhibit reduced eIF1 levels compared to WT cells containing an extra copy of SUI1 ( Figure 4C , lanes 3–4 and 7–8 vs 11–12 ) , indicating that the rps5 mutations still exacerbate the effect of poor AUG context at higher levels of eIF1 expression . We conclude that the increased recognition of UUG start codons conferred by the rps5 mutations is an indirect consequence of their reduced expression of eIF1 . As noted above , reducing eIF1 expression by discriminating against the poor context of the SUI1 AUG codon is a phenotype of known Ssu− ( hyperaccuracy ) mutants ( Martin-Marcos et al . , 2011 ) . Since rps5 mutations also reduced eIF1 expression in a context-dependent manner , we next examined whether they exhibit Ssu− phenotypes by testing them for the ability to suppress the dominant Sui− phenotype of the SUI5 variant of eIF5 ( eIF5-G31R ) ( Huang et al . , 1997 ) . This test was conducted using the RPS5 mutant strains harboring scSUI1 to compensate for the reduced eIF1 expression responsible for their Sui− phenotypes . Introducing SUI5 on a plasmid conferred the expected His+/Sui− phenotype in the his4-301 strain harboring WT RPS5 and native eIF1 levels ( Figure 4E , compare row 1 with 2 ) . Importantly , this His+ phenotype was eliminated in the corresponding rps5-E144R and -R225K mutants when eIF1 levels were boosted by introduction of scSUI1 , as the reduction in growth on −His medium was greater than that seen on +His medium in the rps5/SUI5 mutants vs the RPS5/SUI5 strain ( Figure 4E , rows 3–4 vs 1 ) . Suppression of the His+ phenotype of SUI5 could arise from defective induction of GCN4 mRNA with attendant impairment of HIS4 transcription ( Hinnebusch , 2005 ) ; however , at least rps5-E144R does not reduce the expression of a GCN4-lacZ reporter in amino acid starved cells containing an extra copy of SUI1 ( Figure 4—figure supplement 2 ) . Hence , similar to known Ssu− mutations in eIF1 or eIF1A ( Martin-Marcos et al . , 2011 ) , the rps5-E144R mutation appears to suppress recognition of the UUG start codon of his4-301 mRNA in addition to discriminating against poor context at the SUI1 AUG codon . As noted above , the rps5 mutations decrease recognition of the SUI1 start codon and suppress UUG initiation when eIF1 levels are restored . We next asked whether they also decrease recognition of an upstream AUG codon and allow leaky scanning to the downstream ORF . A GCN4-lacZ reporter was employed with a modified version of upstream ORF1 that is elongated to overlap the GCN4 ORF ( el . uORF1 ) . This construct is ideally suited for this query because virtually all scanning ribosomes normally recognize the uORF1 AUG ( uAUG-1 ) , and reinitiation at the downstream AUG of the GCN4 main ORF following termination of el . uORF1 translation is almost non-existent , so that translation of the main ORF is extremely low ( Grant et al . , 1994 ) . Remarkably , rps5-E144R confers a dramatic increase in leaky scanning through el . uORF1 , elevating GCN4-lacZ expression by 20-fold for the construct containing optimum context ( −3AAA−1 ) at uAUG-1 ( Figure 5 , row 1 , WT vs E144R ) . This effect is nearly comparable to the ∼40-fold increase in leaky scanning seen in WT cells for the extremely weak uAUG-1 context of −3UUU−1 ( Figure 5 , WT , row 5 vs row 1 ) . rps5-E144R also evokes a large ∼10-fold increase in leaky scanning for the el . uORF1 construct with the uAUG-1 context of intermediate strength ( −3UAA−1 ) , but only a ∼fourfold increase with the weakest context of −3UUU−1 ( Figure 5 , rows 3 and 5 , WT vs E144R ) . Expression of the construct lacking uAUG-1 is not significantly affected by rps5-E144R ( Figure 5 , row 7 , WT vs E144R ) , consistent with leaky scanning being the source of elevated GCN4-lacZ expression for the various el . uORF1 constructs . The R225K mutation also increases leaky scanning of uAUG-1 , but to a lesser degree: 3 . 5-fold for −3AAA−1 , 3 . 3-fold for −3UAA−1 , and 1 . 7-fold for −3UUU−1 ( Figure 5 , rows 1 , 3 , 5; WT vs R225K ) . The increases in leaky scanning conferred by the rps5 mutations were relatively unaffected by the restoration of WT eIF1 levels by introducing scSUI1 ( Figure 5 , rows 2 , 4 , 6 vs 1 , 3 , 5 , respectively ) . This was anticipated because the reduced levels of eIF1 in the rps5 mutants ( lacking scSUI1 ) would not be expected to reduce uAUG-1 recognition and confer leaky scanning , as decreased eIF1 abundance is associated with increased start codon recognition ( at least for near-cognate start codons or AUG codons in poor context ) ( Pestova and Kolupaeva , 2002; Ivanov et al . , 2010; Martin-Marcos et al . , 2011 ) . We conclude that the rps5 substitutions impair recognition of GCN4 uAUG-1 , whether located in perfect or poor surrounding sequence context , to allow increased translation of the downstream GCN4 coding sequences . 10 . 7554/eLife . 07939 . 009Figure 5 . RPS5 mutations E144R and R225K confer strong leaky scanning of GCN4 uAUG-1 in vivo . β-galactosidase activities were measured in WCEs of strains from Figure 4C harboring the scSUI1 plasmid ( as indicated ) and el . uORF1 GCN4-lacZ reporters pC3502 , pC4466 , or pC3503 containing , respectively , the depicted optimum , weak , or poor context of uAUG-1; or uORF-less GCN4-lacZ reporter pC3505 with a mutated uAUG-1 . Mean expression values with S . E . M . s were determined from four transformants as described in Figure 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 009 The multiple defects in start codon recognition conferred by rps5-E144R suggest that it destabilizes the PIN state of the 48S PIC . We tested this hypothesis by analyzing the effects of E144R on the equilibrium and rate constants governing TC binding to the 40S subunit in the yeast reconstituted translation system . To this end , we purified 40S subunits from rps5Δ::kanMX deletion strains harboring either plasmid-borne rps5-E144R or WT RPS5 as the only source of Rps5 . We began by measuring the affinity of WT TC , assembled with [35S]-Met-tRNAi , for mutant or WT 40S subunits in the presence of saturating eIF1 , eIF1A and a model mRNA containing an AUG start codon ( mRNA ( AUG ) ) , using native gel electrophoresis to separate 40S-bound and unbound fractions of TC . Reactions conducted with increasing concentrations of 40S subunits revealed that 43S∙mRNA ( AUG ) complexes assembled with either E144R or WT 40S subunits have relatively high affinities for TC ( Figure 6A ) , with Kd values of ≤1 nM ( Figure 6E ) . In the absence of mRNA , the affinities for TC are similar between 43S PICs assembled with mutant or WT 40S subunits ( Figure 6E ) ; however , the endpoint of the reaction is markedly reduced for the E144R complexes ( Figure 6B ) . It was previously proposed that the endpoints of TC binding reactions achieved at saturating 40S concentrations reflect the distribution of PICs between open and closed states . The open state was assumed to be unstable during electrophoresis , and thus could not be visualized , leading to endpoints of <1 ( measured as fractions of TC bound to 40S complexes ) for reactions using mRNA lacking an AUG codon ( Kapp et al . , 2006; Kolitz et al . , 2009 ) , or tRNAiMet mutants ( Dong et al . , 2014 ) , in which the open complex is favored over the closed state . Hence , the reduced endpoint seen in Figure 6B suggests that the closed state of 43S complexes formed with E144R mutant ribosomes is unstable and rearranges to the less stable , open conformation during electrophoresis . This interpretation supports the hypothesis that E144R destabilizes the closed state of the PIC . 10 . 7554/eLife . 07939 . 010Figure 6 . Rps5 Ssu- substitution E144R destabilizes the PIN state in vitro to a greater extent at UUG vs AUG start codons . ( A , B ) Determination of Kd values for TC with [35S]-Met-tRNAi binding to 40S·eIF1·eIF1A complexes assembled with WT or E144R mutant 40S subunits and either mRNA ( AUG ) ( A ) or without mRNA ( B ) . ( C ) Analysis of TC dissociation from 43S·mRNA complexes assembled with WT or E144R mutant 40S subunits and either mRNA ( AUG ) or mRNA ( UUG ) . Representative curves selected from at least three independent experiments are shown . ( D ) Determination of kon values for TC binding to 40S·eIF1·eIF1A complexes from plots of observed rate constants ( kobs ) vs 40S concentration for WT or E144R mutant 40S subunits and mRNA ( AUG ) . ( E , F ) Kd , koff and kon values with S . E . M . s determined in ( A–D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 01010 . 7554/eLife . 07939 . 011Figure 6—figure supplement 1 . RPS5 mutation E144R confers a Gcd- phenotype , derepressing GCN4-lacZ expression . Strains of the indicated RPS5 genotype from Figure 3A were transformed with p180 and analyzed for β-galactosidase expression as in Figure 3B . Mean expression levels and S . E . M . s were calculated from four transformants . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 011 Direct evidence for this last conclusion came from determining rate constants for TC association and dissociation for 43S complexes bound to mRNA . To measure the TC off-rate ( koff ) , 43S∙mRNA complexes were formed as above using TC assembled with [35S]-Met-tRNAi , and the amount of [35S]-Met-tRNAi remaining in the slowly-migrating PIC was measured at different times after adding a chase of excess unlabeled TC . In agreement with previous findings ( Kolitz et al . , 2009; Dong et al . , 2014; Martin-Marcos et al . , 2014 ) , TC dissociates very little from WT PICs formed with mRNA ( AUG ) over the time course of the experiment , yielding a rate constant of only 0 . 05 hr−1 ( Figure 6C; summarized in Figure 6F ) . By contrast , TC dissociation from WT PICs assembled on an otherwise identical mRNA containing a UUG start codon is ∼sevenfold faster ( koff = 0 . 34 hr−1 ) , reflecting the reduced stability of the PIN state at this near-cognate start codon ( Figure 6C , F ) ( Kolitz et al . , 2009 ) . Remarkably , for 43S∙mRNA ( AUG ) complexes assembled with E144R 40S subunits , the dissociation rate was increased ∼36-fold compared to that seen for the corresponding WT complexes ( from 0 . 05 hr−1 to 1 . 8 hr−1; Figure 6C , F ) . An even larger increase in koff of ∼130-fold was measured for mRNA ( UUG ) complexes assembled with E144R vs WT 40S subunits ( ≥46 hr−1 vs 0 . 34 hr−1; Figure 6C , F ) . These findings provide strong biochemical evidence that E144R destabilizes PIN at both AUG and UUG start codons with a relatively stronger effect on the near-cognate triplet , which coincides with the in vivo effects of E144R of reducing recognition of the SUI1 AUG and GCN4 uAUG-1 start codons , and of suppressing UUG initiation on his4-301 mRNA . The rates of TC association ( kon ) were measured by mixing labeled TC with different concentrations of WT or E144R 40S subunits and saturating eIF1 , eIF1A and mRNA ( AUG ) . Aliquots were removed at different time points , the reactions terminated with excess unlabeled TC , and the amount of labeled TC in PICs was measured by native gel electrophoresis . The slope of the plot of the pseudo-first-order rate constants ( kobs ) for PIC formation vs 40S concentration yields the second-order rate constant ( kon ) ( Kolitz et al . , 2009 ) . The kon values measured for WT and E144R 40S subunits were essentially identical ( Figure 6D , F ) , indicating that PICs formed with the mutant ribosomes assemble the POUT complex and rearrange to PIN at the same rates achieved with WT ribosomes , and that rps-E144R primarily reduces the stability of the PIN state . Calculation of Kd values using the measured rate constants koff and kon reveals that E144R decreases the affinity of TC for 43S∙mRNA ( AUG ) complexes by ∼40-fold ( Figure 6E , koff/kon ) . Together , the in vitro experiments demonstrate that E144R reduces the affinity of TC for 43S∙mRNA PICs by destabilizing the PIN state , with a relatively greater effect at UUG vs AUG start codons . Interestingly , we found that E144R confers the Gcd− phenotype , derepressing a GCN4-lacZ reporter by more than fourfold in non-starvation conditions ( Figure 6—figure supplement 1 ) , which indicates a defect in TC binding to 40S subunits in vivo . A decreased rate of TC binding derepresses GCN4-lacZ expression because scanning 40S subunits that have translated uORF1 and resumed scanning can bypass the start codons of the inhibitory uORFs 2–4 before rebinding TC , and then reinitiate further downstream at the GCN4 AUG codon ( Hinnebusch , 2005 ) . Derepression of GCN4-lacZ by E144R was evident even in the presence of scSUI1 ( Figure 4—figure supplement 2 , E144R/scSUI1 vs WT , unstarved ) , indicating that it does not result solely from the reduced eIF1 abundance in this mutant . Because E144R does not reduce the rate of TC binding to 43S·mRNA complexes ( ∼WT kon value , Figure 6F ) , but greatly increases its off-rate ( elevated koff , Figure 6F ) , we infer that the Gcd− phenotype of E144R arises instead from dissociation of TC from a fraction of re-scanning 40S subunits , enabling their bypass of uORFs 2–4 , followed by re-binding of TC in time to reinitiate at GCN4 . This mechanism was described previously for Gcd− substitutions of 18S rRNA residues in the P site of the 40S subunit that , like rps5-E144R , destabilize TC binding in vitro and confer leaky-scanning of an uAUG in vivo ( Dong et al . , 2008 ) . Having concluded that rps5-E144R can suppress UUG initiation once native eIF1 levels have been restored , we examined the remaining β-hairpin substitutions we constructed for this Ssu− phenotype . Remarkably , eight different mutations affecting various loop residues were found to suppress the His+ phenotypes conferred by dominant Sui− alleles SUI3-2 and SUI5 , including R148A/E , R156A/E , R157A/E , A154R and A155E ( Figure 7A , -His panel; cf . WT and rps5 strains harboring SUI3-2 or SUI5 ) . With the possible exception of R157E , they also suppressed the dominant Slg− phenotype of SUI5 ( Figure 7A , +His panel; cf . WT and rps5 strains harboring SUI5 ) —a hallmark of known Ssu− mutations in eIF1 ( Martin-Marcos et al . , 2014 ) . Furthermore , R148A/E , R156A/E and R157A suppressed the elevated UUG:AUG ratio of HIS4-lacZ expression conferred by SUI3-2 ( Figure 7B , C ) , demonstrating bona fide Ssu− phenotypes for these mutations . 10 . 7554/eLife . 07939 . 012Figure 7 . Substitutions in loop residues of the Rps5 β-hairpin confer Ssu- phenotypes . ( A ) 10-fold serial dilutions of PGAL1-RPS5 his4-301 strain ( JVY07 ) transformed with the indicated plasmid-borne RPS5 alleles and either SUI3-2 plasmid p4280 , SUI5 plasmid p4281 , or empty vector were spotted on SD + His + Ura ( +His ) or SD + Ura ( −His ) and incubated at 30°C for 3 days and 5 days , respectively . ( B , C ) Strains from ( A ) also harboring HIS4-lacZ reporters with AUG or UUG start codons ( plasmids p367 and p391 , respectively ) were analyzed as in Figure 3B . Ratios of mean expression of the UUG and AUG reporters calculated from four transformants are plotted with S . E . M . s . Student's t-test indicates that the mean UUG/AUG expression in the RPS5 mutants is significantly reduced when compared to WT ( *p < 0 . 05 , **p < 0 . 005 ) . ( D , E ) WCEs of his4-301 strains with the indicated RPS5 alleles were subjected to Western analysis as in Figure 4A . ( F ) WCEs of strains from ( D , E ) also harboring SUI1-lacZ ( pPMB24 ) or SUI1-opt-lacZ ( pPMB25 ) reporters were assayed for β-galactosidase activities as described in Figure 4B . Mean expression levels and S . E . M . s from four transformants are plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 01210 . 7554/eLife . 07939 . 013Figure 7—figure supplement 1 . Substitutions in the loop of the Rps5 β-hairpin do not increase leaky scanning of GCN4 uAUG-1 . Strains isogenic to those in Figure 3A but containing the indicated RPS5 alleles were transformed with the el . uORF1 GCN4-lacZ reporters containing optimum ( pC3502 ) or poor ( pC3503 ) contexts of uAUG-1 and analyzed for β-galactosidase activities as in Figure 3B . Mean expression values were determined from four independent transformants and normalized to the value obtained for the WT strain , and the normalized means and S . E . M . s are reported . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 013 In addition to discriminating against non-AUG codons , Ssu− mutations in eIF1 and eIF1A exacerbate the effect of suboptimal context of the SUI1 AUG start codon and reduce eIF1 expression . As such , they exacerbate the differential expression of SUI1-lacZ fusions containing native , suboptimal context vs optimized context by specifically reducing expression of the native-context reporter ( Martin-Marcos et al . , 2011 ) . However , none of the rps5 Ssu− mutants exhibit diminished eIF1 abundance ( Figure 7D , E ) , or selectively diminish expression of the SUI1-lacZ fusion with native context ( Figure 7F ) . Nor do they increase leaky scanning of uAUG-1 regardless of its context in the el . uORF1 reporters ( Figure 7—figure supplement 1 ) . Thus , unlike known Ssu− mutations affecting eIF1 and eIF1A , the rps5 Ssu− substitutions in the β-hairpin loop suppress recognition of UUG codons without affecting utilization of AUG codons in poor context . To reveal the molecular mechanism of the Ssu− substitutions in the β-hairpin loop , we analyzed mutant 40S subunits purified from rps5-R148E cells in the reconstituted yeast system . Measurements of TC binding to 43S·mRNA ( AUG ) complexes or 43S complexes without mRNA revealed reaction endpoints ( Figure 8A , B ) and Kd values ( <1 nM for 43S·mRNA ( AUG ) complexes , Figure 8E ) indistinguishable between WT and R148E mutant ribosomes , as were rates of TC dissociation ( koff ) from these complexes containing AUG start codons ( Figure 8C , AUG , WT vs R148E; Figure 8F , WT eIF2β , AUG values ) . However , R148E increased the koff for 43S·mRNA ( UUG ) complexes by ∼twofold , suggesting destabilization of PIN specifically at UUG codons ( Figure 8C , F , WT eIF2β , UUG ) . To support this conclusion , we repeated the koff measurements using eIF2 harboring the Sui− substitution in eIF2β encoded by SUI3-2 ( S264Y ) . Consistent with previous results ( Martin-Marcos et al . , 2014 ) , in reactions with WT 40S subunits , SUI3-2 eliminates detectable TC dissociation from AUG complexes and also delays TC dissociation from UUG complexes ( Figure 8D ) compared to that seen using WT eIF2 ( Figure 8C ) , thus decreasing the koff for UUG complexes by ∼threefold ( Figure 8F; WT eIF2β vs eIF2β-S264Y , WT RPS5 , UUG complexes ) . These results are consistent with the elevated UUG initiation conferred by SUI3-2 in vivo . Importantly , in assays with the SUI3-2 variant of eIF2 , rps5-R148E produced a marked , ∼fourfold increase in koff for the UUG complexes without affecting dissociation of the corresponding AUG complexes ( Figure 8D , F ) . Thus , rps5-R148E preferentially destabilizes the PIN conformation at UUG start codons , overriding the opposing effect of SUI3-2 of enhancing the stability of the UUG complex . These biochemical results are in accordance with our finding that rps5-R148E suppresses the elevated UUG:AUG initiation ratio conferred by SUI3-2 in vivo . 10 . 7554/eLife . 07939 . 014Figure 8 . Rps5 Ssu- substitution R148E destabilizes PIN in vitro selectively at UUG codons . ( A , B ) Determination of Kd values for TC with [35S]-Met-tRNAi binding to 40S·eIF1·eIF1A complexes assembled with WT or R148E mutant 40S subunits and either mRNA ( AUG ) ( A ) or without mRNA ( B ) . ( C , D ) Analysis of TC dissociation from 43S·mRNA complexes assembled with WT or R148E mutant 40S subunits and mRNA ( AUG ) or mRNA ( UUG ) , conducted using WT eIF2 ( C ) or eIF2β-S264Y mutant eIF2 ( D ) . Representative curves selected from at least three independent experiments are shown . ( E , F ) Kd , koff values with S . E . M . s determined in ( A–D ) . ND , no dissociation observed . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 014
In this study , we obtained genetic and biochemical evidence implicating the β-hairpin of Rps5 in achieving efficient and accurate start codon recognition in vivo . In the recent py48S cryo-EM structure ( Hussain et al . , 2014 ) , this domain projects into the mRNA exit channel of the 40S subunit and the hairpin loop approaches the key context nucleotide at the −3 position of mRNA . The β-hairpin also interacts with eIF2α-DI , which mimics an E-site tRNA and contacts the Met-tRNAi in the P site ( Figure 2A ) . Our genetic findings indicate that the E144R substitution in β-strand 1 of the hairpin reduces the rate of bulk translation initiation ( Figure 3C ) and dramatically impairs recognition of GCN4 uAUG-1 in optimal context by the scanning PIC , conferring a higher incidence of leaky scanning for the el . uORF1-GCN4-lacZ reporter than described thus far for any initiation factor mutation ( Elantak et al . , 2010 ) . The E144R mutation also impairs recognition of the SUI1 AUG codon in its native , suboptimal context , and suppresses utilization of the UUG start codon in his4-301 mRNA in different Sui− mutants to confer an Ssu− phenotype . Our biochemical analysis of E144R mutant 40S subunits revealed a drastic destabilization of the PIN state of reconstituted 48S PICs at AUG or UUG codons ( Figure 9 ) , with a stronger effect on the inherently less stable UUG complexes . These biochemical phenotypes can account for both the defects in AUG recognition and the reduction in UUG:AUG initiation ratio ( Ssu− phenotype ) conferred by rps5-E144R in vivo . 10 . 7554/eLife . 07939 . 015Figure 9 . β-hairpin of Rps5 has a critical role in start codon recognition during translation initiation by stabilizing initiator tRNA binding to the pre-initiation complex . Model summarizing the role of the conserved β-hairpin residues in Rps5 in start codon recognition . See Figure 1 for description of the open/POUT and closed/PIN states of the PIC and roles of eIF1 and the SE/SI elements of eIF1A in regulating conformational rearrangements and reactions accompanying AUG recognition . Results from this study indicate that Rps5 β-hairpin residues E144 and R148 function in stabilizing the PIN conformation of TC binding , with E144 having a stronger effect , as indicated by the thicker arrow . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 01510 . 7554/eLife . 07939 . 016Figure 9—figure supplement 1 . Rps5 β-hairpin loop is in proximity to rRNA helix 23 . Depiction of the py48S PIC ( PDB 3J81 ) showing Rps5 in gold , mRNA in orange , Met-tRNAi in green , eIF2α in purple and rRNA residues of helix 23 in grey . The backbone of helix 23 is ∼4 Å from Rps5 loop residue R156 ( indicated by dotted line ) , and rRNA residue G904 is within 3 . 6 Å of the −3 nucleotide in the mRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 07939 . 016 The E144R mutation conferred a greater increase in leaky scanning of GCN4 uAUG-1 in optimum context ( ∼20-fold ) vs uAUG-1 in poor context ( ∼fourfold ) . This difference could be interpreted to indicate that E144R disrupts recognition of optimum context nucleotides , so that its deleterious effect on AUG recognition is dampened when the optimum context nucleotides are absent . However , this interpretation would overlook the fact that it is impossible to detect an increase in leaky scanning of uAUG-1 in poor context by 20-fold , as only a ∼threefold reduction in its recognition yields GCN4-lacZ expression essentially indistinguishable from that seen in the absence of uAUG-1 itself ( Figure 5 , cf . constructs UUUAUG vs AAAAGC in WT cells ) . Thus , it seems plausible that E144R suppresses uAUG-1 recognition equally well for poor and optimal context . By contrast , E144R and R225K impair recognition of the SUI1 AUG only when it resides in poor context , suggesting that the effects of these Rps5 mutations are limited to the inherently weaker initiation site of native SUI1 mRNA . This conclusion would be consistent with the fact that E144R and R225K also suppress recognition of the UUG start codon of his4-301 mRNA , which resides in moderately strong context , as near-cognate codons are inherently weaker initiation sites even when present in optimum context . Extending this last interpretation of E144R and R225K—that they preferentially discriminate against poor initiation sites—to explain our findings on leaky scanning of GCN4 uAUG-1 would require a stipulation that uAUG-1 is a relatively weak initiation site even with the native optimum context present , which seems at odds with the fact that nearly all scanning ribosomes recognize native uAUG-1 in WT GCN4 mRNA ( Grant et al . , 1994 ) . However , one distinctive feature of WT uORF1 and the elongated el . uORF1 contained in our leaky scanning reporter is the absence of typical coding sequences 3′ of the uAUG-1 start codon . WT uORF1 is only 3 codons long , and much of el . uORF1 is derived from non-translated triplets that normally reside between uORF4 and the GCN4 coding sequence . This region has a relatively low propensity for secondary structure ( Kertesz et al . , 2010 ) , which might enable elongating ribosomes to clear the initiation region more rapidly than occurs with more typical coding sequences . In fact , the properties of the leaky scanning reporter insure that the increases in GCN4-lacZ expression observed in the Rps5 mutants reflect diminished recognition of uAUG-1 relative to the main ORF start codon by scanning PICs . Considering that the rps5 mutations appear to have no effect on recognition of the main ORF AUG in the GCN4-lacZ construct lacking el . uORF1 ( Figure 5 , row 7 ) , the presence of greater secondary structure in the main ORF could reduce the rate of scanning through the initiation region to promote start codon recognition in a way that would be lacking at the el . uORF1 start codon . If elongating ribosomes located in the initiation region can enhance initiation by this mechanism , this could compensate for the reduction in PIN stability conferred by E144R and R225K for the SUI1 and GCN4 initiation regions in a way that would not occur for el . uORF1 , making the latter sensitive to the destabilizing effects of these mutations even with optimum context present . Consistent with this scenario , it was shown recently that the presence of slowly-translated codons near the AUG codon can affect the initiation rate in yeast cells ( Chu et al . , 2014 ) . Thus , it seems plausible that E144R and R225K decrease the efficiency of start codon recognition only for weak initiation sites , including near-cognate UUG codons and sub-optimal AUG start sites , without reducing recognition of AUG codons in strong context that initiate structured coding sequences . The impaired recognition of the native SUI1 AUG codon and attendant reduced synthesis of eIF1 conferred by the rps5-E144R and -R225K mutations evokes increased recognition of near-cognate , UUG start codons . While this elevated UUG:AUG initiation ratio is the expected outcome of diminished eIF1 abundance ( Ivanov et al . , 2010; Martin-Marcos et al . , 2011 ) , previously described Ssu− substitutions in eIF1 and eIF1A reduce eIF1 levels by the same mechanism described here—discriminating against the weak context of the native SUI1 AUG—but they suppress , rather than elevate , the UUG:AUG initiation ratio despite reduced eIF1 levels ( Martin-Marcos et al . , 2011 ) . Thus , these eIF1 and eIF1A Ssu− substitutions appear to have a stronger effect than rps5-E144R and -R225K in blocking selection of UUG start codons . On the other hand , the eIF1A and eIF1 Ssu− substitutions confer much smaller increases in leaky scanning of GCN4 uAUG-1 ( Fekete et al . , 2007 ) ( Pilar Martin Marcos and AGH , unpublished observations ) compared to that seen here for rps5-E144R , thus indicating a relatively greater defect in AUG recognition for the Rps5 mutation . In addition to the mutations affecting the upper , structured portion of the β-hairpin loop ( E144R and R225K ) , we also identified Ssu− substitutions in the loop region that discriminate against UUG codons in the presence of Sui− substitutions in eIF5 or eIF2β . Consistent with this , an exemplar of such mutations , rps5-R148E , specifically destabilized the PIN state formed at UUG , but not AUG , start codons in reconstituted PICs in vitro . The Rps5 loop substitutions do not discriminate against the weak context of the SUI1 AUG codon , nor increase leaky scanning of el . uORF1 even when uAUG-1 resides in poor context , and thus exclusively destabilize PICs lacking a cognate ( AUG ) start codon . They differ from previously described Ssu− substitutions in eIF1 ( Martin-Marcos et al . , 2011 ) in that the Rps5 substitutions efficiently suppress UUG initiation but do not discriminate against the poor context of the native SUI1 AUG codon . This distinction might be explained by noting that eIF1 is the principal ‘gate-keeper’ that blocks utilization of weak initiation sites ( Hinnebusch , 2014 ) . As the eIF1 Ssu− substitutions delay eIF1 release from the 40S subunit on start codon recognition ( Martin-Marcos et al . , 2013 ) , they might discriminate more broadly against unstable PICs regardless of whether they lack strong context or a perfect codon:anticodon duplex in the P site . Our Rps5 loop substitutions , by contrast , appear to have a more nuanced effect that destabilizes PIN only when a mismatch occurs in the codon:anticodon duplex itself . As summarized in Figure 9 , our results indicate that both E144 and R148 promote start codon selection by stabilizing the PIN state , and the finding that E144R reduces initiation at both UUG and sub-optimal AUG codons , while R148E impairs only UUG recognition , can be explained as the result of a relatively stronger contribution of E144 vs R148 to the stability of the PIN state . There are several possibilities to explain how perturbing the Rps5 β-hairpin destabilizes the PIN state and reduces start codon recognition . The proximity of the hairpin loop to the E site ( Figure 2 ) suggests a disruption of Rps5 contacts with the context nucleotides in mRNA . Indeed , R156 in the loop interacts with the backbone of rRNA helix 23 , which in turn contacts the −3 context nucleotide ( Figure 9—figure supplement 1 ) ( Hussain et al . , 2014 ) . If this interaction promotes the PIN state , it would help explain why loop residue substitutions impair recognition of UUG start codons ( Ssu− phenotype ) . However , except for the lethal substitution G151S , all of the substitutions affecting loop residues we examined—G151 through Q158—have weaker phenotypes compared to the E144R substitution in β-strand 1 of the hairpin itself , distant from the context nucleotides . Thus , perhaps structural alteration of the β-hairpin by E144R indirectly perturbs the conformation of the N-terminal tail of yeast Rps5 , which promotes AUG recognition by its interaction with Rps16/uS9 ( Ghosh et al . , 2014 ) , whose C-terminal tail closely approaches the codon-anticodon duplex in the P site ( Hussain et al . , 2014 ) . Alternatively , E144R might affect the conformation or location of ribosomal proteins Rps28 and Rps14 , also located in the exit channel and in contact with the Rps5 β-hairpin ( Hussain et al . , 2014 ) , or of domain 1 of eIF2α , which interacts with other regions of Rps5 as well as Met-tRNAi in the PIN complex ( Figure 2 ) . In these latter scenarios , the inherent flexibility of the Rps5 hairpin loop could prevent loop substitutions from altering the orientation of the β-hairpin and attendant perturbations within the PIC compared to effects exerted by E144R or R225K on the structured portion of the hairpin . The β-hairpin of uS7 also protrudes into the mRNA exit channel of bacterial ribosomes in position to interact with mRNA residues just upstream from the P site codon ( Jenner et al . , 2007 ) . In bacterial elongation complexes , the hairpin is also in proximity to E-site tRNA , and truncation of the hairpin increases the frequency of frameshifting , most likely by allowing premature dissociation of the E-site tRNA ( Devaraj et al . , 2009 ) . Interestingly , in the yeast py48S PIC , eIF2α-D1 essentially occupies the position of E-site tRNA ( Hussain et al . , 2014 ) , in accordance with our suggestion that altering the β-hairpin of yeast uS7/Rps5 could impair start codon selection by perturbing the position or flexibility of eIF2α-D1 . Regardless of the exact mechanisms involved , the strong impairment of AUG recognition in vivo and marked destabilization of the PIN state in vitro conferred by E144R dramatically illustrates that a 40S ribosomal protein functions as an equal partner with soluble initiation factors in ensuring efficient and accurate start codon recognition .
Yeast strains and plasmids are listed in Supplementary files 2 , 3 , respectively . Yeast strains used in this study are listed in Supplementary file 2 . The PGAL1-RPS5 strain JVY07 was generated from HLV01a ( MATa ura3-52 trp1Δ-63 leu2-3112 his4-301 ( ACG ) ) by the one-step PCR strategy ( Longtine et al . , 1998 ) using the kanMX4 cassette and selecting for resistance to kanamycin on rich medium containing galactose as carbon source ( YPGal ) . Integration of the kanMX:PGAL1 promoter cassette at RPS5 was verified by PCR analysis of genomic DNA using the appropriate primers . JVY07 was shown to be inviable on glucose medium ( where the GAL1 promoter is repressed ) in a manner fully complemented by plasmid-borne RPS5 alleles on pJV01 and pJV09 . Derivatives of JVY07 harboring low copy LEU2 plasmids containing WT ( pJV09 ) or mutant RPS5 alleles ( pJV12-pJV53 ) , listed in Supplementary file 3 , were generated by transformation . To avoid possible contamination with WT 40S subunits ( from leaky expression of PGAL1-RPS5 on glucose medium ) transformants of JVY07 containing rps5 alleles were not used for purifying mutant 40S subunits , and haploid strains harboring the relevant rps5 alleles as the only source of Rps5 were generated for this purpose . Diploid strain F2009/YSC1021-672858 ( MATa/MATα ura3-∆0/ura3-∆0 leu2-∆0/leu2-∆0 his3∆-1/his3∆-1 lys2-∆0/LYS2 met15-∆0/MET15 rps5∆::kanMX/RPS5 ) was transformed with URA3 RPS5 plasmid pJV38 and sporulated . Tetrads were dissected and analyzed for resistance to G418 to identify rps5Δ::kanMX ascospores , which was verified by PCR analysis of genomic DNA with appropriate primers . One such strain was selected as JVY11 , and used as host to replace pJV38 with plasmids pJV09 , pJV13 and pJV39 , harboring RPS5 , rps5-E144R , rps5-R148E , respectively , by plasmid-shuffling on medium containing 5-FOA ( Boeke et al . , 1987 ) , resulting in strains JVY29 , JVY15 and JVY52 . Plasmids used in this study are listed in Supplementary file 3 . pJV01 was made by inserting into pRS315 a 1 . 6 kb BamHI restriction fragment containing RPS5 flanked by 640 bp upstream and 320 bp downstream of the coding sequences that was amplified from genomic DNA of strain HLV01a . A BglII restriction site was introduced into pJV01 120 bp upstream of the RPS5 ORF using the QuikChange site-directed mutagenesis system ( Agilent Technologies , Santa Clara , CA ) to create pJV09 , which was verified by DNA sequencing of the entire 1 . 6 kb insert . Introduction of the BglII site did not appreciably affect RPS5 expression , as pJV09 and pJV01 were indistinguishable for complementation of strain JVY07 for growth on glucose medium . The insert from pJV09 was sub-cloned into pRS316 to create pJV38 . RPS5 fragments were amplified by fusion PCR to introduce the desired site-directed mutations , using primers listed in Supplementary file 4 and pJV09 as template DNA . The mutagenized fragments were digested with BglII and NdeI and inserted between the same two restriction sites in pJV09 , to produce pJV12-pJV52 ( Supplementary file 3 ) . Plasmid pJV53 was constructed similarly by using primers R225K , R225K_r and pJV13 ( that was verified by sequencing ) as template DNA . All constructs were verified by DNA sequencing of 1 kb from the inserted BglII site beyond the NdeI restriction site , covering the entire RPS5 ORF . Assays of β-galactosidase activity in whole-cell extracts ( WCEs ) were performed as described previously ( Moehle and Hinnebusch , 1991 ) . The sequence context of the start codon for both AUG and UUG HIS4-lacZ reporters is: 5′-AUA ( AUG/UUG ) G-3′ . For Western analysis , WCEs were prepared by trichloroacetic acid extraction as described ( Reid and Schatz , 1982 ) , and immunoblot analysis was conducted as described previously ( Martin-Marcos et al . , 2011 ) with antibodies against eIF1 ( Valasek et al . , 2004 ) and Gcd6 ( Bushman et al . , 1993 ) . Enhanced chemiluminescence ( Amersham ) was used to visualize immune complexes , and signal intensities were quantified by densitometry using NIH ImageJ software . For polysome analysis , strains were grown in SD + His + Ura + Trp at 30°C to A600 , ∼1 . Cycloheximide was added ( 50 μg/ml ) 5 min prior to harvesting , and WCE was prepared in breaking buffer ( 20 mM Tris–HCl , pH 7 . 5 , 50 mM KCl , 10 mM MgCl2 , 1 mM dithiothreitol , 5 mM NaF , 1 mM phenylmethylsulfonyl fluoride , 1 Complete EDTA-free Protease Inhibitor Tablet ( Roche ) /50 ml buffer ) . 15 A260 units of WCE was separated by velocity sedimentation on a 4 . 5–45% sucrose gradient by centrifugation at 39 , 000 rpm for 3 hr in an SW41Ti rotor ( Beckman ) . Gradient fractions were scanned at 254 nm to visualize ribosomal species . Initiation factors eIF1A and eIF1 were expressed in Escherichia coli and purified using the IMPACT system ( NEB ) , and His6-tagged eIF2 was overexpressed in yeast and purified as described ( Acker et al . , 2007 ) . WT and mutant 40S subunits were purified from yeast as described previously ( Acker et al . , 2007 ) . Model mRNAs with the sequences 5′-GGAA[UC]7UAUG[CU]10C-3′ and 5′-GGAA[UC]7UUUG[CU]10C-3′ were purchased from Thermo Scientific . Yeast tRNAiMet was synthesized from a hammerhead fusion template using T7 RNA polymerase and charged with [35S]-methionine or unlabeled methionine as previously described ( Acker et al . , 2007 ) . Kd values of TC ( assembled with [35S]-Met-tRNAi ) and 40S∙eIF1∙eIF1A∙mRNA PICs , and rate constants of TC association/dissociation for the same PICs , were determined by gel shift assays as described previously ( Kolitz et al . , 2009 ) with the minor modifications described below . | To make a protein , the DNA sequence of a gene is first copied to make an mRNA molecule , which is then translated into a protein by a molecular machine called the ribosome . The first step of translation is known as initiation . Several proteins referred to as initiation factors can bind to the small subunit of the ribosome , which itself is composed of an RNA molecule and numerous proteins , and form a pre-initiation complex ( or PIC for short ) . This complex contains a molecule called initiator tRNA that is specialized for initiation . The PIC then attaches to the mRNA and starts scanning it , searching the sequence for an AUG triplet to serve as the start codon . The sequence immediately surrounding an AUG triplet , known as the context , influences the likelihood of its selection as the start site . When the start codon is recognized by the initiator tRNA in the PIC , the complete ribosome assembles and begins to build the protein . Choosing the correct start codon is crucial to ensure that the correct protein is made from every mRNA in the cell . The PIC can adopt an ‘open’ state , which makes it easier to scan the mRNA for the correct start codon and ignore triplets similar in sequence to AUG ( like UUG ) or AUG triplets in a poor context . Once the correct AUG start codon has been recognized , the PIC changes to a ‘closed’ state , ceases to scan , and assembles the complete ribosome . One of the proteins that make up the small ribosomal subunit ( called Rps5 in budding yeast ) is located near the channel where the mRNA molecule exits the PIC during scanning , and is also thought to be involved in translation initiation . However , the role of Rps5 in recognizing the start codon is poorly understood . Visweswaraiah et al . have now studied Rps5—in particular , a region of this protein that adopts a hairpin structure that dips into the exit channel—using genetic and biochemical methods . In mutant yeast cells in which the hairpin structure was mutated , translation initiation was diminished at suboptimal start codons—including a UUG start codon and an AUG codon in poor context—thus making translation initiation more accurate . Visweswaraiah et al . then performed experiments on PICs built from purified components to determine how the Rps5 mutations might affect the assembly and stability of the PIC . The results revealed that mutating the upper region of the Rps5 hairpin destabilized the closed state of the PIC when either an AUG or UUG start codon was present in the mRNA . However , other mutations of the hairpin structure destabilized the closed state of the PIC only at a UUG start codon . In both cases , the mutations made the PIC more likely to remain in the open conformation and continue scanning at incorrect or suboptimal start codons , making it more likely that translation begins at the correct AUG start codon . These results indicate that the Rps5 hairpin is crucial for both efficiently and accurately recognizing the start codon to begin translation . This suggests that ribosomal proteins not only contribute to ribosome structure but can actively participate with other initiation factors in choosing the correct start sites for protein synthesis on all mRNAs in the cell . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2015 | The β-hairpin of 40S exit channel protein Rps5/uS7 promotes efficient and accurate translation initiation in vivo |
The functional mechanisms of membrane proteins are extensively investigated with cysteine mutagenesis . To complement cysteine-based approaches , we engineered a membrane protein with thiol-independent crosslinkable groups using azidohomoalanine ( AHA ) , a non-canonical methionine analogue containing an azide group that can selectively react with cycloalkynes through a strain-promoted azide-alkyne cycloaddition ( SPAAC ) reaction . We demonstrate that AHA can be readily incorporated into the Shaker Kv channel in place of methionine residues and modified with azide-reactive alkyne probes in Xenopus oocytes . Using voltage-clamp fluorometry , we show that AHA incorporation permits site-specific fluorescent labeling to track voltage-dependent conformational changes similar to cysteine-based methods . By combining AHA incorporation and cysteine mutagenesis in an orthogonal manner , we were able to site-specifically label the Shaker Kv channel with two different fluorophores simultaneously . Our results identify a facile and straightforward approach for chemical modification of membrane proteins with bioorthogonal chemistry to explore their structure-function relationships in live cells .
Membrane proteins play fundamental roles in maintaining cellular homeostasis by transporting ions and organic molecules and triggering intracellular signaling pathways in response to external stimuli . One of the most commonly used methods to probe structural relationships and functional dynamics of membrane proteins is the chemical modification of cysteine residues , commonly known as the substituted cysteine accessibility method ( SCAM ) ( Akabas , 2015; Akabas et al . , 1992; Altenbach et al . , 1990; Falke and Koshland , 1987 ) . The distinct reactivity of the thiol group ( pKa 8 . 5 ) towards methanethiosulfonate ( MTS ) or maleimide conjugated reagents , low abundance in membrane proteins and the ease of introducing cysteine residues through site-directed mutagenesis , has facilitated the application of cysteine mutagenesis to study the conformation dynamics of a variety of membrane proteins including ion channels ( Akabas , 2015; Bezanilla , 2000; El Hiani and Linsdell , 2014; Forman and Miller , 2011; Horn , 2002; McNally and Prakriya , 2012; Nakajo and Kubo , 2015; Nys et al . , 2013 ) , transport proteins ( Javitch , 1998; Mulligan and Mindell , 2017; Rudnick , 2006; Schmidt-Rose and Jentsch , 1997; Takeuchi et al . , 2009; Zhu and Casey , 2007 ) and G-protein coupled receptors ( Liapakis et al . , 2001; Wess et al . , 2008 ) . Traditionally , cysteine mutagenesis experiments have been carried out by assessing either cysteine accessibility and/or disulfide and metal bridging in the presence or absence of a substrate or activating stimuli ( Linsdell , 2015 ) . Cysteine mutagenesis also provides a reactive chemical handle for site-specifically installing fluorophores or spin labels into membrane proteins to track their conformational transitions with voltage-clamp fluorometry ( Cha and Bezanilla , 1997; Chanda et al . , 2004; Gorraitz et al . , 2017; Hou et al . , 2017; Mannuzzu et al . , 1996; Priest and Bezanilla , 2015 ) , fluorescence resonance energy transfer ( FRET ) ( Cha et al . , 1999; Chanda et al . , 2005; Glauner et al . , 1999; Jarecki et al . , 2013; Ji et al . , 2016; Posson and Selvin , 2008 ) , electron paramagnetic resonance and double electron-electron resonance spectroscopy ( Altenbach et al . , 1990; Paz et al . , 2018; Pliotas , 2017 ) . With cysteine providing a single reactive group to investigate the conformational dynamics of membrane proteins , it becomes challenging to study multiple allosteric transitions underlying the function of the protein . Recently , a variety of non-canonical amino acids have been incorporated into membrane proteins to introduce reactive chemical groups , environmentally sensitive biophysical reporters and subtle chemical modifications ( Huber and Sakmar , 2014; Klippenstein et al . , 2018; Pless and Ahern , 2013; Rannversson et al . , 2016; Van Arnam and Dougherty , 2014; Young and Schultz , 2018 ) . Typically , non-canonical amino acids are incorporated through a nonsense suppression method using chemically charged tRNA ( Nowak et al . , 1995 ) or a specific pair of tRNA and amino acyl-tRNA synthetase ( Noren et al . , 1989 ) . Since non-canonical amino acid mutagenesis is compatible with cysteine mutagenesis , the two have been combined to expand the scope and precision of strategies for studying membrane proteins ( Dai et al . , 2019; Gordon et al . , 2018; Kalstrup and Blunck , 2013 ) . However , the success rate and efficiency of non-canonical amino acid mutagenesis varies considerably with the type of membrane protein being investigated , expression system , choice of the non-canonical amino acid and the target site in the protein ( Kalstrup and Blunck , 2015; Leisle et al . , 2015; Pless et al . , 2015 ) . The goal of the present study was to develop a facile and generalizable strategy to incorporate a bioorthogonal ( a chemical group which is absent in vivo ) crosslinkable amino acid into membrane proteins for installation of biophysical reporters that could then be combined with cysteine mutagenesis for the arsenal of approaches outlined above . We used an alternate to the nonsense suppression approach where an essential amino acid is simply replaced by supplying excess of an non-canonical analogue , resulting in the global replacement of the natural amino acid in the newly synthesized protein ( Budisa , 2004; Johnson et al . , 2010; Link and Tirrell , 2005 ) . One of the most successful examples of this strategy is the replacement of methionine residues by incorporation of selenomethionine into proteins for X-ray crystallography ( Cohen and Cowie , 1957; Saotome et al . , 2016; Yang et al . , 1990 ) . Incorporation of methionine analogues into proteins has been particularly successful due to the unique conformational flexibility in the amino acid binding site of methionyl tRNA synthetase ( Nadarajan et al . , 2013; van Hest and Tirrell , 1998 ) and indispensable dependence of most eukaryotic systems on external sources for this essential amino acid . To install a bioorthogonally crosslinkable amino acid , we chose azidohomoalanine ( AHA ) ( Figure 1A ) ( Kiick et al . , 2002 ) , a non-canonical amino acid that is nearly isosteric with methionine ( Figure 1A ) and contains an azide group capable of reacting selectively with strained alkynes ( Figure 1B ) . Indeed , AHA has previously been incorporated into proteins to identify and visualize newly synthesized proteins after attachment of biotin ( Dieterich et al . , 2006 ) or fluorescent probes ( Dieterich et al . , 2010 ) using bioorthogonal azide-alkyne cycloaddition reactions ( Agard et al . , 2004; Rostovtsev et al . , 2002; Tornøe et al . , 2002 ) , approaches that have been successful in both prokaryotic and eukaryotic systems ( Ma and Yates , 2018 ) . Using the Shaker voltage-activated potassium ( Kv ) channel as a test case , we find that AHA can be readily incorporated into the channel and enables site-specific installation of fluorophores in the protein via a catalyst-independent strain-promoted azide-alkyne cycloaddition ( SPAAC ) reaction in live Xenopus laevis oocytes ( Figure 1B ) ( Agard et al . , 2004 ) . We show that fluorophores attached to AHA can be used to track conformational changes of the voltage-sensor in a manner analogous to cysteine-based methods . Finally , we implement a straightforward strategy to carry out two-color labeling of membrane proteins in a site-specific manner using a combination of AHA incorporation and cysteine mutagenesis .
We began by testing whether AHA could be introduced into the Shaker Kv channel , an extensively studied ion channel protein that opens and closes in response to changes in membrane voltage ( Bezanilla , 2000; Horne and Fedida , 2009; Swartz , 2008 ) . Shaker is an oligomeric integral membrane protein containing four identical subunits , with each subunit containing six transmembrane ( TM ) segments . The S1-S4 segments from each subunit form peripheral voltage-sensing domains , while the S5-S6 segments from all four subunits constitute the central pore domain ( Long et al . , 2007 ) ( Figure 1C ) . Due to a close homology with other eukaryotic Kv channels and optimal expression in heterologous expression systems , Shaker has been extensively subjected to a variety of chemical modifications through cysteine mutagenesis ( Gandhi et al . , 2003; Gonzalez et al . , 2005; Gross and MacKinnon , 1996; Holmgren et al . , 1998; Horne and Fedida , 2009; Larsson et al . , 1996; Liu et al . , 1996 ) . In addition , many non-canonical amino acids have been successfully incorporated into the Shaker Kv channel ( Infield et al . , 2018; Kalstrup and Blunck , 2013; Pless et al . , 2015; Tao et al . , 2010 ) . We initially examined whether AHA could be incorporated in the Shaker Kv channel expressed in Xenopus oocytes using surface biotinylation to detect AHA incorporation into the channel ( Figure 1—figure supplement 1 ) . The Shaker Kv channel contains 12 methionine residues in each subunit , 5 of which are in TM regions of the channel ( Figure 1C , cyan ) ( Long et al . , 2007 ) . Cysteine accessibility experiments have shown that two out of five methionine residues in the TM regions of Shaker ( M356 and M448 ) face the extracellular side of the channel ( Figure 1C ) ( Larsson et al . , 1996; Liu et al . , 1996 ) , providing the means of detecting AHA incorporation into the Shaker Kv channel expressed on the surface of live Xenopus oocytes . In eukaryotic expression systems , AHA is typically incorporated into proteins by adding the non-canonical amino acid into methionine-free cell culture medium at concentrations ranging from 1 to 4 mM ( Dieterich et al . , 2006 ) , but has not been previously tested with Xenopus oocytes , a commonly used expression system for studying membrane proteins . To incorporate AHA into the Shaker Kv channel , we preincubated oocytes overnight with 4 mM AHA in the ND96 maintenance buffer to compete out the endogenous pool of free methionine . The next day , cRNA for a myc-tagged construct of the Shaker Kv channel was injected ( See Materials and methods ) , and oocytes were maintained in the presence of AHA . After 3–4 days at 17°C , excess AHA was removed by washing oocytes with ND96 and then incubated with membrane-impermeable biotinylation reagents , either azide-reactive dibenzocyclooctyne ( DBCO ) -sulfo-biotin ( Figure 1D , top ) to tag AHA-modified proteins or amine-reactive NHS-sulfo-biotin ( Figure 1D , bottom ) to tag all proteins expressed on the cell surface . After removing unreacted biotin probes by washing with ND96 , oocytes were lysed with a triton X-100 containing lysis buffer , followed by neutravidin agarose pull-down . Subsequently , we analyzed both the cell lysate ( total cell protein ) and the surface fraction with anti-myc western blotting . AHA-supplemented oocytes showed a single band for the mature Shaker Kv channel subunits in the surface fraction isolated with DBCO-sulfo-biotin ( Figure 1E , surface fraction , lane 2 ) , indicating successful incorporation of AHA into the Shaker Kv channel . No pull-down was observed with DBCO-sulfo-biotin in the absence of AHA ( Figure 1E , surface fraction , lane 1 ) , although the protein expression was similar in both the cases ( Figure 1E , total cell protein , lane 1 and 2 ) , demonstrating that the DBCO is chemically selective and does not react with other residues of the protein . In contrast , pull down with NHS-sulfo-biotin yielded similar amounts of protein both in the absence and presence of AHA ( Figure 1E , surface fraction and total cell protein , lane 3 and 4 ) , indicating that introduction of AHA into the Shaker Kv channel has no detectable effect on expression and trafficking of the channel in Xenopus oocytes . Moreover , AHA supplementation had no toxic effects on the survival of oocytes in ND96 at 17°C ( data not shown ) , consistent with previous reports using this non-canonical amino acid in other expression systems ( Dieterich et al . , 2006; Hinz et al . , 2013 ) . Next , we tested whether AHA replaces only methionine residues in the Shaker Kv channel when expressed in Xenopus oocytes by mutating the two methionine residues on the extracellular side of the Shaker Kv channel to Ala ( M356A ) or Leu ( M448L ) ( Figure 1C ) ( Larsson et al . , 1996; Liu et al . , 1996 ) . When oocytes expressing the M356A/M448L mutant channel were labeled with DBCO-sulfo-biotin , no detectable pull-down was observed in the absence or presence of AHA ( Figure 1F , surface fraction , lane 1 and 2 ) , although a comparable amount of protein expression was observed in the cell lysate ( Figure 1F , total cell protein , lane 1 and 2 ) , indicating that the expression of mutant Shaker Kv channel was not affected substantially . In contrast , NHS-sulfo-biotin showed a similar amount of pull down for the mutant protein in both cases ( Figure 1F , surface fraction and total cell protein , lane 3 and 4 ) , establishing that AHA specifically replaces methionine residues in membrane proteins expressed in Xenopus oocytes . These results unambiguously demonstrate that Xenopus oocytes can uptake AHA from the extracellular medium and their endogenous protein synthesis machinery can robustly incorporate this non-canonical amino acid into newly expressed heterologous proteins . To assess the efficiency of AHA incorporation into the Shaker Kv channel , we used densitometry to analyze the anti-myc western blots from wild-type protein ( Figure 1E ) . In general , AHA incorporation had no detectable effect on the total protein expression of the Shaker Kv channel ( Figure 1—figure supplement 2 ) . Moreover , AHA supplementation does not alter the surface expression of the Shaker Kv channel , as the amount of protein pulled down with NHS-sulfo-biotin , which provides an estimate for the total expression of Shaker on the surface of oocytes , was similar in the absence and presence of AHA ( Figure 1G ) . In contrast , DBCO-sulfo-biotin pulls down a comparable amount of protein in the presence of AHA and fails to pull down any protein in the absence of AHA in repeated trials ( Figure 1G ) . We did observe some variability in the pull down with DBCO-sulfo-biotin as compared to NHS-sulfo-biotin , suggesting that the extent of AHA incorporation can vary between different batches of oocytes . AHA-modification of the Shaker Kv channel will replace up to 12 methionine residues with the non-canonical amino acid in regions of the protein known to be critical for voltage-dependent gating ( Figure 1C ) ( Bezanilla , 2008; Swartz , 2008 ) . To explore whether AHA incorporation alters the gating behavior of the Shaker Kv channel , we initially expressed the channel in the absence and presence of AHA and used the two-electrode voltage clamp recording technique to obtain voltage-activation relationships from macroscopic ionic currents with K+ as the charge carrier ( See Materials and methods ) . When the membrane voltage was stepped between −100 mV and +50 mV , we observed robust voltage-activated K+ currents in the presence and absence of AHA , with voltage-activation relationships that were not discernably different ( Figure 2A , B ) , suggesting that AHA incorporation does not detectably alter the overall process of voltage-dependent gating in the Shaker Kv channel . Activation of the voltage-sensing domains in the Shaker Kv channel involves the movement of positively charged arginine residues , which can be directly measured as non-linear capacitive currents known as gating currents ( Bezanilla et al . , 1991 ) . Using the V478W non-conducting mutant of the Shaker Kv channel ( Hackos et al . , 2002; Kitaguchi et al . , 2004 ) , we also measured gating currents in the presence and absence of AHA and obtained gating charge ( Q ) -voltage ( V ) relationships that were similar ( Figure 2C , D ) , suggesting that voltage-sensor activation was also not detectably altered with AHA incorporation . Finally , we tested whether AHA incorporation alters the sensitivity of the Shaker Kv channel to a tarantula toxin that binds to the S1-S4 voltage-sensing domain to allosterically inhibit opening of the channel in response to membrane depolarization . Using a construct of the Shaker Kv channel that is sensitive to the tarantula toxin GxTx1E ( ShakerΔ5 ) ( Herrington et al . , 2006; Milescu et al . , 2009; Milescu et al . , 2013 ) , we observed that the toxin produced robust and indistinguishable shifts of voltage-activation relationships to more positive voltages , both in the absence and presence of AHA ( Figure 2—figure supplement 1 ) . Collectively , these results demonstrate that global replacement of methionine residues with AHA in the Shaker Kv channel does not result in detectable alterations in the gating properties of the channel . After successfully labeling the Shaker Kv channel with cyclooctyne-conjugated biotin probes through AHA , we explored the possibility of using the SPAAC reaction to install environmentally-sensitive fluorophores in a site-specific manner and monitor the conformational dynamics of the channel using voltage-clamp fluorometry ( VCF ) ( Cha and Bezanilla , 1997; Mannuzzu et al . , 1996 ) . This technique has been widely used to investigate a variety of membrane proteins , including the Shaker Kv channel ( Horne and Fedida , 2009 ) , after installation of different thiol-reactive fluorophores using cysteine mutagenesis ( Priest and Bezanilla , 2015 ) . For site-specific fluorescent labeling , we generated the Shaker-M356 construct ( M356/M448L ) , where the methionine residue in the pore domain ( M448 ) was mutated to leucine , leaving a single accessible methionine residue ( M356 ) on the extracellular side of the channel . M356 is located at the N-terminus of the S4 helix within the voltage-sensing domain ( Figure 1C ) , is accessible to extracellular thiol-reactive compounds when mutated to cysteine ( Cha and Bezanilla , 1997; Larsson et al . , 1996; Mannuzzu et al . , 1996 ) and fluorophores attached at this position exhibit changes in fluorescence as the protein undergoes conformational changes in response to changes in membrane voltage ( Cha and Bezanilla , 1997; Mannuzzu et al . , 1996 ) . To facilitate gating current measurements , all voltage clamp fluorometry experiments were carried out with the V478W non-conducting mutant . Since the advent of the SPAAC reaction , a variety of cyclooctynes with varying size , hydrophobicity and reactivity towards the azide group have been synthesized ( Dommerholt et al . , 2016; Sletten and Bertozzi , 2011 ) . In addition , many cyclooctyne conjugated fluorophores have been generated and are commercially available ( Supplementary files 1 and 2 ) , although in some instances their solubility is limited in aqueous solutions ( e . g . TAMRA-DBCO ) . In order to maximize the aqueous solubility of cyclooctyne-fluorophore conjugate , we chose to work with a relatively polar and charged fluorophore , Alexa 488 ( Hughes et al . , 2014 ) . We began by measuring baseline fluorescence signals at a holding voltage of −90 mV using a filter cube appropriate for Alexa 488 ( ex . 480/40 nm; em . 535/50 nm ) after labeling with a cyclooctyne-conjugated Alexa 488 fluorophore ( AF488-DBCO; Figure 3A ) and compared uninjected oocytes with those expressing Shaker-M356 in the absence and presence of AHA ( Figure 3B ) . In the absence of AHA , we observed a 2-fold increase in fluorescence intensity with both uninjected and Shaker-M356 expressing oocytes when compared to unlabeled oocytes ( Figure 3B , Uninjected and M356 ) , which may reflect non-specific interactions between the oocyte membrane and the hydrophobic cyclooctyne group in AF488-DBCO . In the presence of AHA , labeling of uninjected oocytes showed a 5-fold increase in the fluorescence intensity , suggesting incorporation of AHA into a fraction of endogenous oocyte proteins during basal protein turnover ( Figure 3B , Uninjected + AHA ) . In contrast , oocytes expressing AHA-modified Shaker-M356 exhibited ~15 fold higher baseline fluorescence intensity compared to unlabeled oocytes ( Figure 3B , Shaker-M356 + AHA ) , indicating that AHA modification enables the preferential labeling of newly synthesized Shaker Kv channels with cyclooctyne-conjugated fluorophores over other endogenous proteins in Xenopus oocytes . In addition , gating currents measured from oocytes with and without labeling with AF488-DBCO reveals that the Q-V relationship of labeled oocytes was detectably shifted towards more depolarized voltages ( Figure 3C ) , which is consistent with a bulky and charged fluorophore attaching to and modulating the movement of S4 helix ( Cha and Bezanilla , 1997 ) . Next , we investigated whether voltage-sensor activation would produce a change in the fluorescence of AF488-DBCO labeled oocytes expressing Shaker-M356 . As the oocyte membrane was depolarized to positive voltages from a holding voltage of −90 mV , we observed readily detectable increase in fluorescence signals that saturated at positive membrane voltages and returned to the baseline value on membrane repolarization ( Figure 3D ) . We observed no detectable change in fluorescence with AF488-DBCO labeled oocytes expressing the M356A mutant ( M356A/M448L ) in the presence of AHA ( Figure 3E ) , demonstrating that the voltage-dependent fluorescence response specifically originates from AF488-DBCO conjugated at the M356 position in the Shaker Kv channel . Similarly , oocytes expressing Shaker-M356 in the absence of AHA and labeled with AF488-DBCO did not produce any change in fluorescence on membrane depolarization ( Figure 3F ) . Taken together , these results demonstrate that AHA-modified Shaker Kv channels can be labeled with azide-reactive fluorophores in a site-specific and chemoselective manner to track the conformational changes of the channel in response to changes in membrane voltage . Having established AHA as a unique chemical handle for site-specific fluorescent labeling of the Shaker Kv channel , we wanted to compare the properties of AHA-mediated fluorescent labeling of the channel with the well-established cysteine-based method . To monitor the fluorescence response from a single methionine or cysteine residue placed at identical sites in the protein , we designed a construct , Shaker-M356* ( M356/M448L/C245V/C462A ) that has two endogenous cysteines ( C245V and C462A ) along with the methionine ( M448L ) residue in the pore domain mutated . For AHA-mediated labeling , we used the native methionine at M356 position and for the cysteine-mediated labeling , we mutated this methionine to a cysteine and generated the Shaker-M356C construct ( M356C/M448L/C245V/C462A ) . We chose AF488-C5-maleimide to carry out cysteine mediated labeling as the linker length in the two probes is similar ( Figure 4—figure supplement 1 ) . Although our comparison was done using the same fluorophore , labeling sites and background constructs , the residue at the labeling site and attachment chemistries are necessarily different , and therefore the results are not expected to be identical . In addition , the M356C mutant of the Shaker Kv channel has been shown to slow voltage sensor activation and shift the Q-V relationship to more positive voltages ( Cha and Bezanilla , 1997 ) , perturbations not observed with replacement of M356 with AHA ( Figure 2C , D ) . Oocytes injected with Shaker-M356* and Shaker-M356C were labeled with the complementary azide or thiol-reactive fluorophores using identical protocols ( See Materials and methods ) . For both AF488-DBCO and AF488-C5-maleimide labeled oocytes , we observed an increase in fluorescence intensity with membrane depolarization that saturated at positive membrane voltages ( Figure 4A , B ) , indicating that the labeling chemistry does not affect the qualitative behavior of the fluorophore in response to voltage-dependent conformational changes in the Shaker Kv channel . We characterized the behavior of the two fluorescent probes with respect to the gating behavior of the Shaker Kv channel and analyzed the relationship between gating charge movement and changes in fluorescence intensity in each case . The steady state F-V relationship obtained from the labeled Shaker-M356* construct exhibited a detectable shift towards depolarized voltages in comparison to the Q-V relationship ( Figure 4C ) , whereas a closer overlap was observed between the Q-V and F-V relationships obtained from the labeled Shaker-M356C construct ( Figure 4D ) . In addition , the onset of the fluorescence response from Shaker-M356* was discernibly slower than the displacement of gating charge upon depolarization ( Figure 4E ) but overlapped closely during repolarization ( Figure 4G ) . In the case of the Shaker-M356C channel , both the gating currents and fluorescence response showed multiple kinetic components during activation and deactivation of the channel ( Figure 4F , H ) . Our observations on the behavior of the AF488-C5-maleimide labeled Shaker-M356C channel are similar to those reported for the M356C mutant of Shaker after labeling with Oregon green maleimide , a thiol-reactive fluorophore with identical excitation and emission spectra to AF488 ( Cha and Bezanilla , 1997 ) . Taken together , this comparison shows that fluorescent labeling of AHA-modified Shaker Kv channels with cyclooctyne-conjugated fluorophores can be utilized to track the conformational rearrangements similar to cysteine-based methods . To compare the magnitude of fluorescence responses as a function of protein expression level for fluorophore installation using AHA- and cysteine-based approaches , we measured maximal fluorescence responses ( ΔF/F , % ) along with Qmax to estimate the total number of Shaker Kv channels expressed on the surface of oocytes ( Aggarwal and MacKinnon , 1996 ) . For both AF488-DBCO and AF488-C5-maleimide labeled oocytes ( Figure 5A , B ) , the magnitude of maximal fluorescence response increases along with channel expression on the surface of oocytes ( Figure 5C , D ) , although there is a considerable spread in both relationships . Variability in fluorescence responses is to be expected given the heterogeneity in the endogenous oocyte fluorescence around 480 nm excitation ( Lee and Bezanilla , 2019 ) . Nevertheless , comparison of the trends for AHA- and cysteine-mediated fluorescent labeling suggests that AHA-mediated labeling requires approximately two-fold higher protein expression when compared to cysteine-mediated labeling ( Figure 5C , D ) . This difference could arise from either incomplete incorporation of AHA and/or fluorophore labeling due to the slower rate of SPAAC reaction compared to the reaction between maleimide and cysteine ( Dommerholt et al . , 2016; Lang and Chin , 2014; Saito et al . , 2015 ) . Membrane proteins are also labeled simultaneously with two distinct biophysical reporters to investigate their conformational transitions through FRET ( Taraska , 2012; Taraska and Zagotta , 2010 ) or to independently track structural rearrangements in two different regions of the protein ( Kalstrup and Blunck , 2013; Kalstrup and Blunck , 2018 ) . Thus far , two-color labeling of membrane proteins has been achieved using pairs of cysteine residues ( Glauner et al . , 1999; Koch , 2005; Posson and Selvin , 2008; Wang et al . , 2018 ) , where it is difficult to monitor the site-specific attachment of fluorophores and often suffers from complexities arising from mixed populations of proteins containing one or both fluorophores . Cysteine mutagenesis has also been combined with fluorescently-labeled ligands ( Posson and Selvin , 2008 ) , transition metal binding sites ( Billesbølle et al . , 2016; Taraska et al . , 2009 ) , lanthanide metal binding peptide tags ( Vázquez-Ibar et al . , 2002 ) or fluorescent non-canonical amino acids ( Gordon et al . , 2018; Kalstrup and Blunck , 2013 ) to achieve site-specific labeling of membrane proteins with two different biophysical reporters . Interestingly , biorthogonal reactions including the SPAAC reaction and the copper mediated azide alkyne cycloaddition ( CuAAC ) reaction have also been combined with thiol-mediated reactions for two-color labeling , but exclusively with relatively small and soluble purified proteins containing azide- or alkyne-terminated amino acids introduced through the nonsense suppression method ( Sadoine et al . , 2017; Seo et al . , 2011; Tyagi and Lemke , 2013 ) . Given the straightforward nature of AHA incorporation , the biocompatible nature of the SPAAC reaction and the comparable fluorescence responses observed using AHA- and cysteine-based approaches ( Figures 4 and 5 ) , we explored the possibility of combining the two methods for two-color labeling of the Shaker Kv channel using azide and thiol-mediated chemical reactions in live cells . To install two different fluorophores simultaneously into the Shaker Kv channel , we added a cysteine mutation at position S424C in the outer mouth of pore domain of Shaker-M356* to generate Shaker-M356*-S424C ( M356/M448L/C245V/C462A/S424C ) . The S424C site is accessible to fluorescent labeling with thiol-reactive TAMRA-maleimide fluorophore ( Gandhi et al . , 2000; Loots and Isacoff , 1998; Loots and Isacoff , 2000 ) , the resulting voltage-dependent fluorescence responses are distinct from those measured when TAMRA fluorophores are attached to the external end of S4 ( Cha and Bezanilla , 1997 ) and have been proposed to report on conformational rearrangements associated with slow inactivation of the channel ( Claydon et al . , 2007; Loots and Isacoff , 1998 ) . Thus , fluorophores attached at M356 in the voltage-sensing domain and S424 in the pore domain of the Shaker-M356*-S424C construct should report on distinct conformational changes in these two regions of the protein . We first measured fluorescence responses of Shaker-M356*-S424C when labeled independently with AF488-DBCO or TAMRA-MTS . Oocytes injected with Shaker-M356*-S424C in the presence of AHA gave rise to functional channels after labeling with AF488-DBCO ( Figure 6A ) or TAMRA-MTS ( Figure 6B ) . AF488-DBCO labeled oocytes produced a similar fluorescence response as observed with Shaker-M356* through the 488 filter cube ( ex . 480/40 nm; em . 535/50 nm ) ( Figure 4A , C ) , indicating that the additional cysteine mutation in the pore did not substantially affect the fluorescence behavior of AF488 installed at M356 ( Figure 6C , G ) . In contrast , TAMRA-MTS labeled oocytes generated distinct fluorescence responses through the TAMRA filter cube ( ex . 535/50 nm; em . 610/75 nm ) when compared to the fluorophore on top of S4 and were consistent with responses reported when labeling with TAMRA-maleimide ( Figure 6F ) ( Claydon et al . , 2007 ) . The fluorescence-voltage relationships for AF488-DBCO on S4 and TAMRA-MTS within the pore domain were radically different from each other ( Figure 6G , H ) , and no fluorescence response was detected when AF488-DBCO labeled oocytes were subjected to TAMRA excitation/emission ( ex . 535/50 nm; em . 610/75 nm ) ( Figure 6E ) or when TAMRA-MTS labeled oocytes were subjected to AF488 excitation/emission ( ex . 480/40 nm; em . 535/50 nm ) ( Figure 6D ) . Thus , we could clearly distinguish between the fluorescence responses originating from AF488-DBCO or TAMRA-MTS labeled Shaker-M356*-S424C in the presence of AHA . For two-color labeling , oocytes expressing the Shaker Kv channel containing only the M356 or S424C sites were used to assess the degree of cross-reactivity between cyclooctyne and cysteine residues ( Beatty et al . , 2010; Conte et al . , 2011; van Geel et al . , 2012; Zhang et al . , 2018a ) . To minimize the cross-reactivity , oocytes were labeled sequentially with TAMRA-MTS and then AF488-DBCO ( See Materials and methods ) . As expected , two-color labeled Shaker-M356* produced voltage-dependent fluorescence changes through the AF488 filter cube ( Figure 7E , K ) which closely resembled Shaker-M356* labeled only with AF488-DBCO ( Figure 4A , C ) , and there was no measurable change in fluorescence intensity through the TAMRA filter cube ( Figure 7H ) . Similarly , two-color labeling of oocytes expressing Shaker-M356*-S424C in the absence of AHA showed no voltage-dependent fluorescence through the AF488 filter cube ( Figure 7F ) , while the fluorescence response through the TAMRA filter cube was consistent with previous reports for S424C labeled with TAMRA-maleimide ( Figure 7I , L ) ( Claydon et al . , 2007 ) . The lack of voltage-dependent fluorescence responses in the TAMRA channel ( ex . 535/50 nm; em . 610/75 nm ) after two-color labeling of M356* ( Figure 7H ) indicates that TAMRA-MTS does not cross-react with AHA at M356 , since TAMRA produces a robust response when attached at this position using cysteine chemistry ( Cha and Bezanilla , 1997; Mannuzzu et al . , 1996 ) . Similarly , the absence of voltage-dependent fluorescence responses in the AF488 channel for two-color labeling of M356*-S424C indicates that AF488-DBCO does not cross-react with S424C under these labeling conditions , since labeling S424C with AF488-C5-maleimide produces robust fluorescence responses ( Figure 7—figure supplement 1 ) . Finally , the two-color labeled Shaker-M356*-S424C showed distinct voltage-dependent fluorescence responses through both AF488 and TAMRA filter cubes ( Figure 7G , J , M ) , similar to the single color labeling ( Figure 6 ) . These results demonstrate that AHA- and cysteine-mediated fluorescent labeling approaches can be combined for chemically selective and site-specific installation of different fluorophores into the Shaker Kv channel . To determine whether individual Shaker Kv channels have been simultaneously labeled with both fluorophores , we looked for direct intramolecular energy transfer between AF488 and TAMRA . In the structure of the Kv1 . 2/2 . 1 paddle chimera ( Long et al . , 2007 ) , the Cα distances between the residues corresponding to M356 in S4 and S424 in the four subunits forming the pore domain are 23 . 7 Å , 38 . 5 Å , 45 . 8 Å and 55 Å ( Figure 8—figure supplement 1 ) , near enough to allow FRET from AF488 to TAMRA ( where R0 ~55 Å ) . To distinguish such intra-molecular FRET in our system , we must account for background sources of fluorescence ( e . g . fluorophores attached to other surface proteins , oocyte auto-fluorescence , etc . ) as well as ‘bleed through’ due to direct emission of the donor and direct excitation of the acceptor through the FRET filter cube . Conveniently , the background fluorescence is independent of voltage , so its contribution can be excluded by considering only the voltage-dependent fluorescence changes ( ∆F ) . Figure 8 shows the voltage-dependent fluorescent changes for constructs containing only the fluorescent donor site ( Shaker-M356* ) or the fluorescent acceptor site , ( Shaker M356A-S424C ) or both donor and acceptor sites ( Shaker M356*-S424C ) in the presence of AHA . Because the signal through the FRET cube ( Alexa 488 excitation: ex . 480/40 nm; TAMRA emission: em . 535/50 nm ) includes contributions from AF488 fluorophores emitting directly into the TAMRA channel and TAMRA fluorophores directly excited by the AF488 excitation , we used oocytes expressing the Shaker-M356*-S424C labeled only with AF488-DBCO or TAMRA-MTS to estimate and correct for this spectral bleed-through ( Figure 8—figure supplement 2 ) . Subtraction of AF488-DBCO direct emission into the TAMRA channel should be quite reliable as the emission in the AF488 channel and TAMRA channel have identical voltage dependent behavior with a mean bleed through ratio of 0 . 142 ± 0 . 004 at 50 mV ( Figure 8—figure supplements 2G and 3 ) . In contrast , the subtraction of the TAMRA direct excitation signal is more approximate because the voltage-dependence of TAMRA emission depends on the excitation wavelength and was not identical with AF488 and TAMRA excitation ( Figure 8—figure supplement 2H ) . The mean bleed through ratio for TAMRA was estimated to be 0 . 076 ± 0 . 004 at 50 mV . ( Figure 8—figure supplement 3 ) . Nevertheless , both the raw and corrected FRET signals ( Alexa 488 excitation; TAMRA emission ) are larger when both donor and acceptor are present ( Figure 8L , O ) compared to when only the donor ( Figure 8J ) or acceptor ( Figure 8K ) is present . Furthermore , the increase in the FRET signal upon depolarization correlates with the upward movement of the S4 helix ( and donor ) towards the acceptor with the corrected FRET F-V relationship closely following the Q-V relationship for the oocytes expressing Shaker-M356*-S424C in the presence of AHA ( Figure 8R ) . Collectively , these results establish that AHA incorporation and cysteine mutagenesis can be efficiently combined to carry out site-specific two-color labeling of individual Shaker Kv channels in live cells .
In the present study , we introduce a cysteine-independent method to engineer membrane proteins with crosslinkable chemical groups and subsequently modify them with spectroscopic probes using a bioorthogonal chemical reaction in live cells . We used the non-canonical amino acid , azidohomoalanine ( AHA ) ( Kiick et al . , 2002 ) , to introduce azide groups in place of methionine residues in the Shaker Kv channel . Our results establish that AHA can be readily incorporated into the Shaker Kv channel in an efficient and residue-specific manner . Using SPAAC chemistry with azide-reactive cyclooctyne conjugated reagents ( Agard et al . , 2004 ) , we demonstrate the utility of AHA incorporation for site-specific installation of fluorescent probes in the Shaker Kv channel to follow the conformational changes with voltage-clamp fluorometry in Xenopus oocytes ( Cha and Bezanilla , 1997; Mannuzzu et al . , 1996 ) . We were able to combine AHA and cysteine-mediated fluorescent labeling for simultaneous labeling with two different fluorophores at specific sites in the voltage-sensing and pore domains of the Shaker Kv channel . We also demonstrate that a voltage-dependent FRET response can be detected with the two-color labeled Shaker Kv channel exclusively when unique cysteine and AHA-substituted methionine residues are both present . Taken together , our results suggest that AHA incorporation and cysteine mutagenesis provide a straightforward and robust way of incorporating two distinct reactive groups into the Shaker Kv channel expressed in live cells . We believe this approach will work for other membrane proteins , but several important issues should be considered for each potential application . One of the most important considerations is the number and location of methionine residues in the protein of interest . Methionine residues have similar abundance as cysteine residues in membrane proteins and they tend to be located towards the center of the lipid bilayer ( Koehler Leman et al . , 2018 ) , suggesting that our AHA-based approach will have similar applications and limitations when compared to cysteine-based approaches . Although methionine residues are somewhat more abundant in other membrane proteins compared to the Shaker Kv channel , their prevalence is comparable to cysteine residues within regions potentially accessible to the extracellular solution ( Supplementary file 3 ) . Importantly , not all methionine residues in the extracellular half of the protein will be accessible to azide-reactive alkyne probes and would need to be removed . In the Shaker Kv channel , for example , out of the three methionine residues in the extracellular half of the protein , while M356 can be robustly labeled with DBCO-biotin in the presence of AHA , M448 exhibits barely detectable labeling ( data not shown ) and M440 is inaccessible , as seen by the absence of streptavidin pulldown for the M356A/M448L double mutant ( Figure 1F ) . Moreover , it has been previously demonstrated that site-specific fluorescence responses can be measured using cysteine-based approaches without removing the native cysteine residues ( Savalli et al . , 2006 ) . Thus , using AHA for fluorescent labeling on the extracellular side of membrane proteins should be generally applicable , even for some of the larger membrane proteins we analyzed ( Supplementary file 3 ) . The degree to which replacement of methionine with AHA perturbs the functional properties of a membrane protein is another important consideration in applying the approaches described here . In our experience with the Shaker Kv channel , AHA incorporation is well-tolerated , showing minimal perturbations in cellular expression level ( Figure 1 ) , voltage-dependent gating properties ( Figure 2 ) and response of the channel to a gating modifier toxin , GxTx1E ( Figure 2—figure supplement 1 ) . This can be attributed to the highly isosteric nature of methionine and AHA residues ( Kiick et al . , 2002 ) , making it a suitable substrate for endogenous protein translation machinery of Xenopus oocytes and precluding substantial changes to the allosteric transitions required for voltage-dependent activation and deactivation of the channel . In most circumstances , it is likely that replacement of methionine with AHA will be well tolerated , although the effect of substituting a new methionine residue will depend on the identity of that specific site . It is also important to consider that the relatively bulky DBCO group we used in this study and the product of SPAAC reaction with AHA may not be well-accommodated at all positions in a protein of interest . However , in our experience thus far , labeling the voltage-sensing domain of the Shaker Kv channel with DBCO-conjugated fluorophores does not dramatically alter voltage sensor activation ( Figures 3C , 4E and G ) . We also note that smaller azide-reactive cyclooctyne reagents like the bicyclononynes ( BCN ) group have also been conjugated to fluorophores and are commercially available for fluorescent labeling of AHA-modified membrane proteins ( Supplementary file 2 ) ( Dommerholt et al . , 2010; Dommerholt et al . , 2014; Leunissen et al . , 2014 ) . The extent of AHA incorporation will also be an important variable to consider in planning future applications of our methods and is clearly a limitation in comparison to cysteine mutagenesis . Although the pull-down observed with the azide-reactive DBCO-sulfo-biotin is comparable to that with the amine-reactive NHS-sulfo-biotin ( Figure 1E , lane 2 and 4 and Figure 1G ) , suggesting that the extent of AHA incorporation is robust , this assay is relatively qualitative . We are currently developing quantitative methods to measure the extent of AHA incorporation using mass spectrometry and/or azide-reactive cyclooctyne polyethylene glycol ( PEG ) polymers ( Darabedian et al . , 2018 ) . Background labeling of cysteine and methionine residues present within endogenously expressed proteins might also prevent detection of distinct fluorescent signals from proteins that do not express to high enough levels . However , it is important to appreciate that specifically detecting the stimulus-dependent changes in fluorescence may be sufficient to distinguish responses originating from the protein of interest . For instance , we did not observe any voltage-dependent changes in fluorescence with the Shaker-M356A construct in the presence of AHA ( Figure 3E ) , even though background labeling of other proteins was clearly detectable in the presence of AHA ( Figure 3B ) . Nevertheless , limitations imposed by partial AHA incorporation and incomplete SPAAC reactions combine to produce only 2-fold smaller fluorescence responses for our AHA-based labeling approach when compared to cysteine-based methods ( Figure 5 ) , providing a convenient benchmark to suggest that AHA-based methods should be viable for most proteins that have already been successfully studied using cysteine-based approaches ( Supplementary file 3 ) ( Cowgill and Chanda , 2019; Horne and Fedida , 2009; Priest and Bezanilla , 2015 ) . Finally , it is important to keep in mind that the array of DBCO- or BCN-conjugated fluorophores which are commercially available is currently more limited ( Supplementary files 1 and 2 ) compared to the thiol-reactive probes available for cysteine-based approaches . Nevertheless , there are certain DBCO-conjugated fluorophores ( e . g . AF555-DBCO and AF594-DBCO ) that may have better properties than AF488-DBCO because they absorb far from the UV region , and thus the fluorescence responses should have less contamination from endogenous oocyte fluorescence and improved signal/noise ratios . Our preliminary results with AF555-DBCO and AF594-DBCO show that these spectrally distinct fluorophores also exhibit voltage-dependent changes in fluorescence when installed at M356 using AHA ( data not shown ) . The ability to install different azide-reactive fluorescent reporters through AHA incorporation into a membrane protein will hopefully provide motivation to generate better reagents with optimal spectral properties and linkers between the reactive group and fluorophore . An important aspect of combining AHA incorporation and cysteine mutagenesis is the compatibility between azide and thiol-mediated chemical reactions for installing two different biophysical probes into a single protein . It is important to mention that some cyclooctyne groups can react with cysteine residues ( Beatty et al . , 2010; Conte et al . , 2011; van Geel et al . , 2012; Zhang et al . , 2018a ) , however , our approach of first labeling cysteine residues with thiol-reactive probes works efficiently for achieving site-specific fluorescence responses from azide and thiol-reactive fluorophores with negligible cross-reactivity between cyclooctyne and cysteine residues ( Figures 7 and 8 ) . Given that the reaction rates between cysteine and MTS/maleimide reagents are considerably faster ( ~103–104 M−1s−1 ) ( Saito et al . , 2015 ) than the reaction between AHA and cyclooctynes ( 0 . 1–1 M−1s−1 ) ( Dommerholt et al . , 2016; Lang and Chin , 2014 ) , it should be possible to achieve specific labeling when the two reactions are carried out at the same time . In addition , other azide-containing amino acids such as p-azidophenylalanine have been widely incorporated into membrane proteins through nonsense suppression methods ( Daggett and Sakmar , 2011; Rannversson et al . , 2016; Zhu et al . , 2014 ) , and thus can also be combined with cysteine mutagenesis similar to what we have shown here with AHA incorporation and cysteine mutagenesis ( Figure 7 ) . Our FRET measurements demonstrate that we can simultaneously install two different fluorophores site-specifically into the Shaker Kv channel using a combination of AHA incorporation and cysteine mutagenesis ( Figure 8 ) . Recently , the fluorescent non-canonical amino acid , Anap , has been incorporated and combined with cysteine mutagenesis to install two fluorescent reporters into the Shaker Kv channel and simultaneously monitor conformational changes on the intracellular and extracellular side of the protein ( Kalstrup and Blunck , 2013 ) . Anap incorporation has also been combined with TETAC , a cysteine-reactive transition metal binding cyclen , for measuring intramolecular distances using tmFRET ( Dai et al . , 2019 ) . Although Anap incorporation is highly site-specific and allows quantitative estimation of short distances ( 10–20 Å ) , its incorporation diminishes protein expression considerably ( Aman et al . , 2016; Shandell et al . , 2019; Zagotta et al . , 2016 ) . Moreover , Anap requires UV excitation , and thus suffers from contamination with cellular autofluorescence in some cell types ( Chatterjee et al . , 2013 ) . The combination of AHA incorporation and cysteine mutagenesis would provide flexibility to choose the donor and acceptor pairs for measuring a wide range of distances and it would be particularly exciting to carry out FRET measurements between fluorophores installed through AHA and TETAC installed with cysteine . In addition to two-color labeling , combining AHA incorporation and cysteine mutagenesis would be valuable for an array of other biophysical applications . Installation of fluorophores with AHA effectively frees up cysteine mutagenesis , providing an opportunity to spectroscopically monitor the impact of cysteine modification , or disulfide/metal bridge formation on the structural rearrangements of membrane proteins . It would also be exciting to combine AHA incorporation and cysteine mutagenesis for bioorthogonal installation of electron paramagnetic probes at two independent sites , greatly expanding the types of distance measurements that could be achieved . Finally , due to inherent orthogonality between AHA incorporation , cysteine mutagenesis and nonsense suppression methods , it is conceivable that three independent biophysical reporters can be installed within a protein to further constrain FRET based measurements and/or study multiple conformational changes simultaneously . Given that AHA incorporation has been widely used in a variety of different cell types ( Dieterich et al . , 2010; Dieterich et al . , 2006; Erdmann et al . , 2015; Glenn et al . , 2017; Hinz et al . , 2012; Link et al . , 2004; Ma and Yates , 2018 ) , the approach described here with Xenopus laevis oocytes should be readily applicable to studies in other cellular expression systems .
All the constructs were generated in the pGEMHE vector ( Liman et al . , 1992 ) with Shaker-IR ( ΔN , 6–46 ) ( Hoshi et al . , 1990 ) as the common background . The mutations for methionine and cysteine residues were carried out using the QuickChange Lightning site-directed mutagenesis kit as per manufacturer’s protocol ( Agilent Technologies ) . The DNA sequence of all constructs and mutants was confirmed by automated DNA sequencing and complementary RNA ( cRNA ) was synthesized using T7 polymerase after linearizing the DNA with NheI restriction enzyme . The RNA was purified using the RNAse easy kit ( Qiagen ) , eluted in RNAse-free water and stored at −80°C until use . All the chemicals were purchased from Sigma-Aldrich unless specified . Female Xenopus laevis animals were housed and surgery was performed according to the guidelines of the National Institutes of Health , Office of Animal Care and Use ( OACU ) ( Protocol Number 1253–18 ) . Oocytes were removed surgically and incubated with agitation for 1 hr in a solution containing ( in mM ) 82 . 5 NaCl , 2 . 5 KCl , 1 MgCl2 , 5 HEPES , pH 7 . 6 ( with NaOH ) , and collagenase ( 2 mg/ml; Worthington Biochemical , Lakewood , NJ ) . All surface biotinylation experiments were carried out with ShakerΔ5-V478W-myc ( Hackos et al . , 2002; Milescu et al . , 2013 ) containing a myc tag at the C-terminal . Defolliculated oocytes were injected with 50 nl of channel RNA ( ~500 ng/μl ) after preincubating them in the absence or presence of 4 mM AHA ( Bachem , 100 mM stock in ddH2O ) at 17°C overnight in an ND96 oocyte maintenance buffer , containing ( in mM ) : 96 NaCl , 2 KCl , 5 HEPES , 1 MgCl2 and 1 . 8 CaCl2 plus 50 mg/ml gentamycin , pH 7 . 6 with NaOH . After four days of maintaining the oocytes at 17°C , excess AHA was removed with 5–6 washes of ND96 and oocytes were labeled with amine or azide reactive biotin reagents , EZ-Link-sulfo-NHS-LC-biotin ( ThermoFisher ) or DBCO-sulfo-biotin ( Sigma ) , according to the previously published protocol with minor modifications ( Silberberg et al . , 2005; Zhang et al . , 2018b ) . Twenty healthy oocytes were incubated with 1 mM of each biotin probe ( 10X stock in ddH2O ) in separate wells of a 24-well plate in a final volume of 0 . 5 ml at room temperature . The reaction was terminated after 20 min for NHS-sulfo-LC-biotin and 60 min for DBCO-sulfo-biotin by transferring the oocytes to ND96 , followed by 6–8 washes to remove excess biotinylation reagent . Subsequently , oocytes were homogenized in 400 µl of lysis buffer containing ( in mM ) : 100 NaCl , 20 Tris⋅Cl , pH 7 . 4 , 1% Triton X-100 , 5 µl/ml protease inhibitor mixture ( Sigma ) . Homogenization and all subsequent steps were performed at 4°C . After centrifugation at 16 , 000 × g for 3 min , a 20 µl aliquot of the supernatant ( total cell protein ) was mixed with equal volume of 2 × NuPAGE LDS sample buffer ( ThermoFisher ) plus reducing agent: 50% 4 × LDS sample buffer ( Bio-Rad ) , 10% 2-mercaptoethanol and 40% 100 mM DTT . The remaining supernatant was diluted 1:1 with the lysis buffer and 60 µl of High Capacity NeutrAvidin agarose beads ( ThermoFisher ) were added followed by gentle tumbling overnight at 4°C . The NeutrAvidin agarose beads were washed six times with the lysis buffer with a 2 min centrifugation ( 16 , 000 × g ) step between each wash . At the end of the final wash , 40 µl of 2 × LDS sample buffer plus reducing agent was added to the beads and samples were heated at 70°C for 10 min . Following a 2 min centrifugation ( 16 , 000 × g ) , the supernatant ( surface protein ) and total cell protein ( collected earlier ) were separated in 10% Bis-Tris acrylamide gel ( ThermoFisher ) using a MOPS running buffer containing ( in mM ) : 20 Tris base , 20 MOPS , 1 . 25 EDTA , 0 . 1% SDS , pH 7 . 6 . Seeblue Plus2 prestained ladder ( ThermoFisher ) was used as the protein molecular weight marker . After SDS-PAGE , proteins in the gel were transferred to nitrocellulose membrane using the iBLOT semi-dry transfer apparatus ( ThermoFisher ) . The nitrocellulose membrane was probed with mouse anti-myc antibody ( ThermoFisher , Cat . No . 46–0603 ) diluted 1:1000 in TBS-T containing ( in mM ) : 25 Tris , 137 NaCl , 3 KCl , 0 . 05% Tween20 followed by HRP-conjugated anti-mouse secondary antibody ( 4 µl in 15 ml TBS-T ) . The blot was developed using Immobilon ECL Western detection reagents ( Millipore ) . Densitometry was performed with the Image lab software ( Bio Rad ) . All ionic currents were recorded using the Shaker-IR construct where residues 6–46 were deleted to remove N-type inactivation ( Hoshi et al . , 1990 ) . For experiments with the tarantula toxin GxTx1E , the toxin-sensitive ShakerΔ5 construct ( L327I , A328F , V330T , V331E and A332S ) was used ( Milescu et al . , 2013 ) . GxTx1E toxin was synthesized on an ABI peptide synthesizer using Fmoc chemistry , refolded in vitro and purified as previously described ( Gupta et al . , 2015 ) . Experiments with AHA-modified channel were performed after preincubating the oocytes in 4 mM AHA ( prepared in ND96 from a 100 mM stock in ddH2O ) overnight , followed by cRNA injection . Oocytes were injected with 50 nl of channel RNA ( 5–10 ng/μl ) in the absence or presence of AHA and studied after 1–4 days to allow for expression at 17°C in ND96 solution . All the recordings were performed using the two-electrode voltage-clamp recording technique ( OC-725C amplifier; Warner Instruments , Hamden , CT ) using a 150 μl recording chamber . Data were filtered at 1 kHz and digitized at 5–10 kHz using Digidata 1321A interface board and pCLAMP 10 software ( Molecular Devices , Sunnyvale , CA ) . Microelectrode resistances were 0 . 2–0 . 8 MΩ when filled with 3 M KCl . The external recording solution contained ( in mM ) : 50 KCl , 50 NaCl , 10 HEPES , pH 7 . 6 with NaOH at room temperature ( ~22°C ) . All voltage-clamp fluorometry experiments were performed using the non-conducting V478W mutant of the Shaker Kv channel ( Hackos et al . , 2002; Kitaguchi et al . , 2004 ) , with additional methionine or cysteine mutations as indicated in the text and figure legends . Oocytes were injected with 50 nl of channel RNA ( 100–500 ng/μl ) in the absence or presence of 4 mM AHA and maintained at 17°C for 1–5 days . Fluorescent labeling of oocytes was carried out by first removing excess AHA with 5–6 washes with ND96 , followed by incubation with 100 µM AF488-DBCO ( Alexa fluorophore 488-dibenzocyclooctyne , Click Chemistry Tools , 100X stock in DMSO ) or AF488-C5-maleimide ( Alexa fluorophore 488-C5-maleimide , Click Chemistry Tools , 100X stock in DMSO ) for 60 min at room temperature in 0 . 5 ml ND96 . Oocytes were transferred to fresh ND96 and washed five times ( 5 min each ) to remove the excess fluorophore and stored in the dark at 10–13°C prior to performing experiments . For TAMRA-MTS ( 2- ( ( 5 ( 6 ) -tetramethylrhodamine ) carboxylamino ) ethyl methanethiosulfonate; Toronto Research Chemicals ) , oocytes were incubated with 10 μM of the fluorophore ( 1000X stock in ddH2O ) in ND96 at 4°C for 60 min , followed by five washes . For the two-color labeling , oocytes were first labeled with TAMRA-MTS , followed by AF488-DBCO labeling as documented above . Two-electrode voltage clamp recordings were obtained using a Dagan CA-1B amplifier . Electrodes were filled with 3M KCl and had resistances between 0 . 2–0 . 8 MΩ . The external recording solution was ND96 without gentamycin , containing ( in mM ) : 96 NaCl , 2 KCl , 5 HEPES , 1 MgCl2 and 1 . 8 CaCl2 , pH 7 . 6 with NaOH . For all gating current measurements , Q was obtained by integrating the OFF gating current elicited by repolarization to the holding voltage . Fluorescence signals were acquired through a 40X , 0 . 8-NA objective ( Olympus LUMplanFLN ) on an Olympus BX51WI microscope and a photodiode . For Alexa 488 signals , excitation filter , emission filter and dichroic were ET480/40 , ET535/50 and T510nm , respectively ( Chroma Tech . ) . For TAMRA-MTS , excitation filter , emission filter and dichroic were HQ535/50 , HQ610/75 and T570pxrxt , respectively ( Chroma Tech . ) . For the FRET cube , the excitation filter , emission filter and dichroic were ET480/40 , HQ610/75 and T570pxrxt , respectively ( Chroma Tech . ) . The signal from the photodiode was low-pass filtered at 3 kHz and sampled at 20 kHz through a Digidata-1440A controlled by pClamp10 ( Molecular Devices ) . The light source used for the illumination was either a blue ( 470/24 nm ) or a green ( 550/15 nm ) LED ( Lumencor , Spectra X ) . The bleed through fluorescence ratio for AF488 and TAMRA were calculated by using single-color labeled oocytes expressing Shaker-M356*-S424C in the presence of AHA and normalizing the magnitude of the steady-state voltage-dependent change in fluorescence intensity ( ΔF , A . U . ) obtained through the FRET cube by the one obtained from 488 and TAMRA filter cubes , respectively . All the fluorescence traces represent single recordings without averaging . All data needed to evaluate the conclusions in this paper are available in the main text and supplementary materials . | Living cells can sense cues from their environment via molecules located at the interface between the inside and the outside of the cell . These molecules are mostly proteins and are made up of building blocks known as amino acids . To understand how these proteins work , fluorescent probes can be attached to amino acids within them – which can then tell when different parts of proteins move in response to a signal . Scientists often target fluorescent probes at the amino acid cysteine , because it has a chemically reactive side group and is rare enough so that unique positions can be labeled in the protein of interest . However , being able to target other amino acids would allow scientists to ask , and potentially solve , more precise questions about these proteins . Methionine is another amino acid that has a low abundance in most proteins . Previous research has shown that the cell’s normal protein-building machinery can incorporate synthetic versions of methionine into proteins . This suggested that the introduction of chemically reactive alternatives to methionine could offer a way to label membrane proteins with fluorescent probes and free up the cysteines to be targeted with other approaches . Gupta et al . set out to develop a straightforward method to achieve this and started with a well-studied membrane protein , called Shaker , and cells from female African clawed frogs , which are widely used to study membrane proteins . Gupta et al . found that the cells could readily take up a chemically reactive methionine alternative called azidohomoalanine ( AHA ) from their surrounding solution and incorporate it within the Shaker protein . The AHA took the place of the methionines that are normally found in Shaker , and just like in cysteine-based methods , fluorescent probes could be easily attached to the AHAs in this membrane protein . Shaker is a protein that allows potassium ions to flow across the cell membrane by changing shape in response to the membrane voltage . The fluorescence from those probes also changed with the membrane voltage in a way that was comparable to cysteine-mediated approaches . This indicated that the AHA modification could also be used to track structural changes in the Shaker protein . Finally , Gupta et al . showed that AHA- and cysteine-mediated labeling approaches could be combined to attach two different fluorescent probes onto the Shaker protein . This method will expand the toolbox for researchers studying the relationship between the structure and function of membrane proteins in live cells . In future , it could be applied more widely once the properties of the fluorescent probes for AHA-mediated labeling can be optimized . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics",
"tools",
"and",
"resources"
] | 2019 | Exploring structural dynamics of a membrane protein by combining bioorthogonal chemistry and cysteine mutagenesis |
Cytosine DNA methylation ( mC ) is a genome modification that can regulate the expression of coding and non-coding genetic elements . However , little is known about the involvement of mC in response to environmental cues . Using whole genome bisulfite sequencing to assess the spatio-temporal dynamics of mC in rice grown under phosphate starvation and recovery conditions , we identified widespread phosphate starvation-induced changes in mC , preferentially localized in transposable elements ( TEs ) close to highly induced genes . These changes in mC occurred after changes in nearby gene transcription , were mostly DCL3a-independent , and could partially be propagated through mitosis , however no evidence of meiotic transmission was observed . Similar analyses performed in Arabidopsis revealed a very limited effect of phosphate starvation on mC , suggesting a species-specific mechanism . Overall , this suggests that TEs in proximity to environmentally induced genes are silenced via hypermethylation , and establishes the temporal hierarchy of transcriptional and epigenomic changes in response to stress .
Phosphorus ( P ) is one of the most important macronutrients for all living organisms , being a key component of nucleic acids and membrane phospholipids , as well as being an essential element for energy-mediated metabolic processes . Plants preferentially absorb this nutrient as inorganic phosphate ( Pi ) , a form of P with low availability and mobility in the soil ( Poirier and Bucher , 2002 ) . As a consequence , Pi is one of the most limiting nutrients for plant growth and development in most agricultural soils . To overcome these issues , application of large quantities of Pi fertilizers has been the primary strategy to maintain crop yields . Yet , this approach is increasingly economically and environmentally unsustainable , with the reserves of Pi rocks greatly diminishing . It is therefore critical to better understand the molecular mechanisms involved in Pi homeostasis in order to generate plants with increased P acquisition and use efficiency , associated with sustained yields that will contribute to improve global food security . Plants have developed a wide set of sophisticated responses aimed at acquiring and utilizing Pi efficiently in order to maintain cellular Pi homeostasis even under Pi limiting conditions ( −Pi ) ( Rouached et al . , 2010; Chiou and Lin , 2011; Peret et al . , 2011 ) . In −Pi , the expression level of genes encoding high affinity Pi transporters ( PTs ) in the roots increases in order to increase Pi uptake , as well as inducing and secreting acid phosphatases and ribonucleases to mobilize organically bound P . In contrast , the levels of P-containing intermediates such as nucleotides , RNA and phospholipids are dramatically reduced under −Pi , with phospholipids being replaced with sulpho- and galactolipids . As a consequence , genes involved in sulpho- and galactolipids synthesis , such as sulpholipid synthases ( SQDs ) and monogalactosyl diacylglycerol synthases ( MGDs ) are highly up-regulated in −Pi conditions ( Misson et al . , 2005; Secco et al . , 2013a ) . Members of the SPX-domain containing protein family ( e . g . , SPX , PHO1 and NLA ) have also been shown to be key regulators of Pi homeostasis , being involved in Pi transport and signaling ( Wang et al . , 2009; Rouached et al . , 2010; Kant et al . , 2011; Secco et al . , 2012; Secco et al . , 2013a; Puga et al . , 2014; Wang et al . , 2014 ) . In addition to the complex regulation observed at the transcriptome level , studies have shown that Pi homeostasis is also regulated by several post-transcriptional mechanisms involving non-coding RNAs , such as miR827 , miR399 and IPS1 ( Franco-Zorrilla et al . , 2007; Chiou and Lin , 2011 ) as well as post-translational changes ( Bayle et al . , 2011; Lin et al . , 2013; Park et al . , 2014 ) . However , only a limited number of studies have assessed the potential involvement of altered DNA or histone modifications in response to Pi starvation , and stresses in general ( Sahu et al . , 2013 ) . Smith and colleagues previously reported that in Arabidopsis ( Arabidopsis thaliana ) the histone variant H2A . Z was deposited at a number of Pi starvation-induced ( PSI ) genes and that a loss H2A . Z resulted with their de-repression ( Smith et al . , 2010 ) . In addition , using low resolution , non-quantitative and locus-specific methods , several studies have shown the potential involvement of altered DNA methylation in response to stresses ( Labra et al . , 2002; Chinnusamy and Zhu , 2009; Wang et al . , 2011a , 2011b; Karan et al . , 2012; Chen and Zhou , 2013; Sahu et al . , 2013 ) . Deep sequencing technologies now enable whole genome single base resolution analysis of DNA methylation ( Cokus et al . , 2008; Lister et al . , 2008 ) , thus enabling global assessment of changes in DNA methylation in response to environmental and developmental cues . Indeed , Dowen et al . ( 2012 ) previously reported that biotic stress could induce dynamic changes in DNA methylation of repetitive sequences or transposons , often coupled to transcriptional changes of neighbouring genes ( Dowen et al . , 2012 ) . In addition , Zhong et al . ( 2013 ) recently reported that changes in DNA methylation patterns play a role in the process of tomato fruit ripening ( Zhong et al . , 2013 ) . Cytosine DNA methylation ( mC ) is involved in a range of important biological processes , including silencing of repetitive sequences and transposable elements ( TEs ) , genomic imprinting , and stable gene silencing . In plants , DNA methylation exists in all sequence contexts ( CG , CHG , CHH , where H = A , C or T ) through the activity of multiple genetically distinct pathways ( Law and Jacobsen , 2010; Matzke and Mosher , 2014; Mirouze and Vitte , 2014 ) . De novo DNA methylation is mediated by the RNA-directed DNA methylation ( RdDM ) pathways . In the canonical RdDM pathway , transcripts produced from the RNA polymerase IV ( Pol IV ) are then copied into dsRNAs by the RNA-dependent RNA polymerase ( RDR2 ) before being processed into 24-nucleotides ( nt ) small interfering RNAs ( siRNAs ) by DICER-LIKE 3 ( DCL3 ) . These newly generated siRNAs are then loaded onto ARGONAUTE 4 ( AGO4 ) before being guided towards the nascent scaffold of RNAs transcribed by the RNA polymerase V ( Pol V ) through sequence complementarity . Finally , this complex recruits the DNA methyltransferase DOMAINS REARRANGED METHYLTRANSFERASE 2 ( DRM2 ) to perform de novo methylation in all sequence contexts ( Matzke and Mosher , 2014 ) . Recently , another RdDM pathway , which is independent of Pol IV and DCL3 , has been identified and is referred to as RDR6-RdDM ( Nuthikattu et al . , 2013; Panda and Slotkin , 2013; Creasey et al . , 2014; Matzke and Mosher , 2014; Bond and Baulcombe , 2015 ) . In this pathway , Pol II derived transcripts are copied by RDR6 into dsRNAs before being processed into 21-22-nt siRNAs by DCL2 and DCL4 . These siRNAs can then either induce post-transcriptional gene silencing ( PTGS ) when loaded onto AGO1 or initiate de novo DNA methylation when associated with AGO2 , thus ultimately triggering the canonical RdDM pathway . Maintenance of DNA methylation through replication is mediated by METHYLTRANSFERASE 1 ( MET1 ) and CHROMOMETHYLASE 3 ( CMT3 ) methyltransferases in the CG and CHG contexts , respectively , and are thus referred to as symmetrical methylation , while methylation in the non-symmetrical CHH context has to be established de novo after DNA replication and involves the activities of the DOMAINS REARRANGED METHYLTRANSFERASE 1 and 2 ( DRM1/DRM2 ) and CMT2 methyltransferases ( Finnegan et al . , 1996; Du et al . , 2012; Zemach et al . , 2013 ) . Within the plant species studied to date , the general methylation state of particular genomic features are relatively conserved , with TEs often highly methylated in all contexts , and CG methylation commonly located in gene bodies ( Feng et al . , 2010; Zemach et al . , 2010; Mirouze and Vitte , 2014 ) . However , large differences in global DNA methylation levels can be observed amongst plant species potentially associated with different TE content in the various plant genomes . Indeed , the TE-rich genome ( ∼40% ) of rice has a much higher aggregate level of DNA methylation than the TE-poor ( ∼15% ) Arabidopsis genome ( Li et al . , 2012; Ragupathy et al . , 2013; Mirouze and Vitte , 2014 ) . Given the paucity of past studies assessing the impact of abiotic stresses upon the plant DNA methylome and the temporal relationship between DNA methylation and transcriptional changes , we performed a comprehensive spatio-temporal assessment of the impact of limiting a central plant macronutrient , Pi , upon DNA methylation patterns and transcription , in rice ( Oryza sativa ) and Arabidopsis . Using whole genome bisulfite sequencing , we identified species-specific , widespread and mitotically heritable changes in DNA methylation in response to Pi starvation that are particularly enriched at stress responsive genes . These changes in DNA methylation occur after changes in nearby gene expression , and are thus likely a consequence of induced transcription of nearby Pi responsive genes , as well as being largely independent of DCL3a . Altogether , we demonstrate a species-specific process in which Pi starvation causes highly localized changes in the genomic DNA methylation patterns in rice , which may act to repress the potentially deleterious activity of TEs located near genes that are highly induced upon stress .
A comprehensive time-course experiment of Pi-starved plants was undertaken , spanning medium ( 3 and 7 days ) , and long-term ( 21 days up to 52 days ) Pi deprivation ( −Pi ) , as well as both short term ( 1 and 3 days ) and long-term ( 31 days ) recovery ( Figure 1A ) . The 52 days time point consisting of 21 days starvation +31 days recovery enabled investigation of the effects of long term resupply on Pi starved plants , and coincided with the emergence of the first panicles and grains ( Figure 1A , B ) . Pre-germinated rice seedlings were grown for 14 days in Pi sufficient conditions ( 0 . 32 mM Pi ) before being transferred to either Pi sufficient ( 0 . 32 mM Pi ) or Pi deficient ( 0 mM Pi ) media for 21 days . After 21 days of Pi deficient treatment , half of the plants were either maintained under Pi deficient conditions or re-supplied with Pi ( 0 . 32 mM ) for 1 , 3 or 31 days . To confirm the effectiveness of the Pi starvation and resupply treatments , physiological and molecular analyses were performed . Growth analyses revealed that 21 days of Pi deprivation resulted in a 2 . 3 fold decrease in shoot biomass ( Figure 1—figure supplement 1A ) , however no significant differences could be observed for the root biomass after 21 days of treatment ( Figure 1—figure supplement 1B ) . Consequently , the root-to-shoot biomass ratio , a parameter often used to assess the efficiency of various nutrient stresses , was significantly altered ( t-test , p < 0 . 05 ) , with plants grown under Pi deficient conditions for 21 days showing a 2 . 2 fold increase compared to plants grown under Pi sufficient conditions ( Figure 1—figure supplement 1C ) , which is consistent with previous reports ( Reymond et al . , 2006; Jiang et al . , 2007; Secco et al . , 2013a ) . Pi concentration measurements revealed that 3 days of −Pi led to a threefold reduction in root Pi concentration , while 21 days −Pi resulted in a 6 fold decrease in root Pi concentration compared to plants continuously grown under Pi sufficient conditions ( Figure 1C ) . Resupplying Pi for 1 day was sufficient to increase the root Pi concentration >4 fold compared to plants grown under Pi deficient conditions , and within 3 days of Pi resupply the root Pi concentration was similar to that of plants continuously grown under Pi sufficient conditions . 10 . 7554/eLife . 09343 . 003Figure 1 . Effects of Pi starvation and resupply in rice . ( A ) Schematic representation of the experimental design . Seeds were germinated in water for 2 days and then transferred to a Pi-sufficient hydroponic ( 0 . 32 mM ) solution for 2 weeks before being either transferred to Pi-deficient media ( 0 mM ) or maintained in Pi-sufficient solution . After 21 days of treatment , half of the Pi-starved plants were resupplied with Pi-sufficient media for up to 31 additional days . Seeds from plants continually grown in + and −Pi were harvested and grown in Pi sufficient conditions for 10 days . Black arrows indicate the time of emergence of panicles . ( B ) Morphological appearance of rice seedlings after 52 days of treatment . White arrows indicate panicles . ( C ) Pi concentration in the roots . ( D ) Hierarchical clustering of significantly ( Cuffdiff , FDR < 0 . 05 ) differentially expressed genes ( DEGs ) in response to Pi starvation as determined by mRNA-seq . ( E ) Number of significantly differentially expressed genes in the roots for each time point . Source data for Figure 1 is available at Dryad ( Secco et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 00310 . 7554/eLife . 09343 . 005Figure 1—figure supplement 1 . Responses of rice to Pi starvation . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 005 To investigate the transcriptional responses to Pi starvation and resupply , RNA sequencing ( RNA-seq ) was performed on root samples from all time points , using three biological replicates per condition ( Figure 1—source data 1 , available at Dryad , Secco et al . , 2015 ) . Hierarchical clustering of the steady state transcript abundance of the 5570 genes identified as significantly differentially expressed ( Cuffdiff , FDR < 0 . 05 ) in at least one of the time points revealed a gradual increase in the number and fold change of differentially expressed genes upon −Pi , associated with the length of the Pi deprivation ( Figure 1D , E; Figure 1—source data 1 , available at Dryad , Secco et al . , 2015 ) . Of note , several phosphate starvation-induced ( PSI ) marker genes , including the SPX genes , MGD2 and PTs , were already induced and showed high steady state transcript abundance after only 3 days of Pi deprivation ( Figure 1—source data 1 , available at Dryad , Secco et al . , 2015 ) . Surprisingly , 52 days of Pi deprivation was associated with a decrease in the number and extent of significantly differentially abundant transcripts , including most of the PSI marker genes , potentially due to the concurrent occurrence of panicle development and grain filling . Indeed , a previous study aimed at profiling the shoots of rice grown in the field throughout their life cycle identified two major transcriptome changes , occurring just before panicle differentiation and straight after flowering ( Sato et al . , 2011 ) . In addition , the transcription of some of the PSI genes , including MGD2 and PHO2 , was reduced before the panicle differentiation , suggesting that the rice plants undergo a major change in Pi homeostasis at the vegetative-reproductive phase transition ( Sato et al . , 2011 ) . Resupplying Pi for 1 day after 21 days of starvation was sufficient for the transcript abundance of 40% of PSI differentially regulated genes to return to a level that was not significantly different from the +Pi condition at the matched time point ( Figure 1D , E ) . Within 3 days of Pi resupply , the internal root Pi content was similar to that of Pi sufficient plants and the transcript abundance of 80% of the 5570 PSI differentially expressed genes had already returned to levels equivalent to Pi sufficient conditions . After 31 days of Pi resupply , the transcript abundance of only 80 PSI genes remained significantly different compared to Pi sufficient conditions . Overall , the physiological and transcriptional changes associated with Pi starvation and resupply confirmed the effectiveness of the Pi treatments , as well as the capacity of the rice plants to rapidly sense and respond to these changing nutrient conditions . To test whether Pi starvation affects genomic DNA methylation in rice , whole genome base resolution profiling of DNA methylation by MethylC-seq was performed on rice roots in triplicate at each time point ( Figure 1A ) . Altogether , 45 root single-base resolution high coverage DNA methylomes were generated ( 77–87% of cytosines covered by at least one read , 82–88% of the genome ) ( Supplementary file 1 ) . In order to identify a set of conserved PSI differentially methylated regions ( DMRs ) in the roots , the 21 , 22 and 24 time points were utilized , resulting in the selection of 9 +Pi samples and 9 −Pi samples . The methylation levels in all sequence contexts ( CNN ) were then assessed for these samples , and only regions that showed significant differences in methylation levels ( FDR < 0 . 01 ) in at least 7 of the 9 samples in each of the conditions were considered for further analysis , resulting in the identification of 175 high confidence root PSI DMRs ( Figure 2—source data 1 , available at Dryad , Secco et al . , 2015 ) . Among these 175 PSI DMRs , 84% were hypermethylated in response to Pi starvation ( 147 hypermethylated , 28 hypomethylated regions , Figure 2A ) . As observed for transcript abundance , 3 and 7 days of Pi deprivation resulted in fewer changes in DNA methylation levels , which increased with the duration of the Pi starvation . Indeed , hierarchical clustering of the 45 root samples based on their CNN methylation levels in the 175 PSI DMRs revealed two main clusters , with cluster 1 containing all +Pi samples as well as the 3 and 7 days −Pi samples , and cluster 2 comprising all long-term −Pi time points ( ≥21 days ) in addition to the Pi re-supplied samples ( Figure 2—figure supplement 1 ) . Furthermore , unlike transcript abundance , where 3 days of Pi resupply was sufficient for the majority of PSI genes to return towards non-stressed level , DNA methylation levels were unaffected by 3 days of Pi recovery . While long-term ( 31 days ) Pi resupply resulted in the PSI DMR methylation level moving towards the Pi sufficient methylation levels ( Figure 2A ) , hierarchical clustering analysis revealed that the long-term resupply samples were still more closely related to the long-term Pi starvation samples rather than those of +Pi , as assessed by methylation level at the PSI DMRs in roots ( Figure 2—figure supplement 1 ) . Thus , nutrient stress-induced differential methylation states can persist for substantial periods of time following cessation of the stress conditions . 10 . 7554/eLife . 09343 . 006Figure 2 . Pi starvation triggers widespread changes in DNA methylation in rice roots . ( A ) Hierarchical clustering of the average difference in CNN methylation levels of the 175 PSI DMRs ( FDR < 0 . 01 ) identified upon long term Pi starvation ( 21 days , 22 days and 24 days ) in the roots in response to Pi starvation . ( B ) Normalised distribution of the distance of the DMRs to the nearest gene . The position of each DMR was calculated with respect to the nearest gene . DMRs were categorized in bins ( within gene body , 0–1 kb , 1–2 kb , 2–4 kb , 4–6 kb , and >6 kb from the TSS or TES ) , and the number of DMRs in each bin was normalised by the total number of regions present in that bin category in the genome . TSS , Transcription Start Site; TES , Transcription End Site . ( C ) Proportion of DMRs overlapping with transposable elements ( TEs ) , and their corresponding classes . ( D ) Distribution of non-redundant DMR-associated gene transcription levels identified by RNA-seq upon Pi deprivation . ( E ) Gene ontology enrichment analysis of non-redundant DMR-associated genes ( p-value < 0 . 05 ) . Source data for Figure 2 is available at Dryad ( Secco et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 00610 . 7554/eLife . 09343 . 010Figure 2—figure supplement 1 . Hierarchical clustering of the 45 root methylomes based on their CNN methylation levels in the 175 root PSI DMRs . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 01010 . 7554/eLife . 09343 . 011Figure 2—figure supplement 2 . Characteristics of the root PSI DMRs . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 01110 . 7554/eLife . 09343 . 012Figure 2—figure supplement 3 . DNA methylation in transposable elements . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 01210 . 7554/eLife . 09343 . 013Figure 2—figure supplement 4 . Examples of PSI DMRs . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 01310 . 7554/eLife . 09343 . 014Figure 2—figure supplement 5 . Hierarchical clustering of the 341 shoot PSI DMRs . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 014 The root PSI DMRs had an average size of 205 bp with a mean of 16 differentially methylated cytosines per DMR , and were preferentially localized within the first two kilobases upstream ( 40% ) and first kilobase downstream ( 15% ) of the nearby gene ( Figure 2B , Figure 2—figure supplement 2 ) . Given the role of DNA methylation in repressing TE activity , and the high frequency of TEs in the rice genome , accounting for 40% of the total genome length ( Ragupathy et al . , 2013 , Mirouze and Vitte , 2014 ) , the position of the DMRs with respect to annotated TEs was assessed . Among the 175 root PSI DMRs , 90% overlapped with TEs , with Miniature Inverted-repeat Transposable Elements ( MITEs ) , the most common class of TEs in the rice genome ( 69% of all TEs ) , accounting for 58% of the DMRs overlapping with TEs ( Figure 2C ) . To investigate the specificity of the changes in DNA methylation in TEs , DNA methylation levels were assessed in TEs overlapping PSI DMRs as well as in all annotated TEs after 21 days of Pi treatment . Altogether , while TEs overlapping PSI DMRs show significant changes in DNA methylation in response to Pi starvation , the majority of all annotated TEs are unaffected by Pi starvation , suggesting that Pi starvation triggers specific and localized changes in DNA methylation in a subset of all annotated TEs ( Figure 2—figure supplement 3 ) . In addition , RNA-seq analysis failed to reveal any significantly differentially expressed TEs ( Cuffdiff , FDR < 0 . 05 ) that overlapped with PSI DMRs , between Pi sufficient and deficient conditions ( 7 and 21 days ) , with more than 80% of them not being expressed ( Figure 2—source data 2 , available at Dryad , Secco et al . , 2015 ) . Next , each of the 175 PSI DMRs were assigned to the nearest gene , resulting in the association of 126 unique genes with PSI DMRs ( Figure 2—source data 1 , available at Dryad , Secco et al . , 2015 ) . Gene transcript abundance analyses revealed that 66 of these DMR-associated genes ( 52% ) were differentially expressed by >2 fold ( FDR < 0 . 05 ) in response to 21 days −Pi ( Figure 2D ) . The steady state transcript abundance of 35% of the 126 genes was >4 fold higher following 21 days Pi starvation . Furthermore , gene ontology analyses revealed a strong enrichment for genes encoding proteins involved in phosphate homeostasis , such as acid phosphatases , phosphate transporters , and factors involved in glycerophosphodiester phosphodiesterase activity ( Figure 2E , Figure 2—figure supplement 4 ) . Of note , while the majority of DMR-associated genes were associated with only one DMR , several genes were associated with multiple DMRs , including key PSI marker genes such as SPX1 , SPX2 , SPX5 , PT10 and MGD2 , reinforcing the strong association between differential DNA methylation and transcription in response to −Pi . To investigate whether Pi starvation could also induce differential methylation in tissues other than the root , which is the primary organ involved in Pi sensing and uptake , high coverage methylomes of rice shoots grown for 21 days under Pi sufficient or deficient conditions were also generated , performing 3 biological replicates per condition ( Supplementary file 1 ) . Firstly , changes in shoot methylation levels were assessed at the 175 PSI DMRs identified in the roots ( Figure 2A , Figure 2—source data 1 , available at Dryad , Secco et al . , 2015 ) , revealing similar changes in DNA methylation levels in response to Pi starvation in both roots and shoots , though to a lesser extent in the shoots . Secondly , shoot methylomes under Pi-sufficient and deficient conditions were analysed in order to identify shoot PSI DMRs . Due to the lower number of replicates used to identify shoot PSI DMRs ( n = 3 ) compared to the root PSI DMRs ( n = 9 ) , a less stringent FDR cut-off of < 0 . 05 was selected for further analysis , resulting in the identification of 341 shoot PSI DMRs ( FDR < 0 . 05 ) ( Figure 2—source data 3 , available at Dryad , Secco et al . , 2015 ) . Assigning the 341 shoot PSI DMRs to the nearest genes identified 43 distinct genes that were significantly differentially regulated by Pi starvation in the shoots ( Figure 2—figure supplement 5 ) . In addition , analysis of the methylation levels in all contexts ( CNN ) revealed that the 341 shoot PSI DMRs had similar patterns of methylation in both roots and shoots , suggesting that Pi starvation affects DNA methylation in roots and shoots in a similar manner . Altogether , more than 30 conserved regions showed significant changes in both roots and shoots , and could be associated to PSI marker genes such as SPX1 , SPX2 and MGD2 . To further examine the potential role of the PSI DMRs , only DMRs that were associated with significant changes in nearby gene transcript abundance following long-term Pi starvation were considered for downstream analysis . Among the 175 root PSI DMRs identified , 100 PSI DMRs were close to genes showing significant changes ( FDR < 0 . 05 ) in gene transcription upon long term Pi starvation , corresponding to 66 unique genes ( Figure 3A , Figure 2—source data 1 , available at Dryad , Secco et al . , 2015 ) . Notably , 63 ( 95% ) of these DMR-associated genes were induced by Pi starvation , and included key regulators of Pi homeostasis such as Pi transporters ( PT3 , PT4 , PT9 , PT10 ) , SPX genes ( SPX1 , SPX2 , SPX3 , SPX5 ) , MGD2 , IPS1 and pre-miR827 . Only three of the DMR-associated genes were down-regulated by long term Pi starvation . Hierarchical clustering of the differential methylation levels in all contexts for the root PSI DMRs in response to Pi deprivation revealed two distinct clusters , with DMRs in cluster 1 and 2 being hyper- and hypomethylated in response to Pi starvation , respectively ( Figure 3A , Figure 3—figure supplement 1 ) . The first group contained 81 PSI DMRs associated with 61 genes that were overwhelmingly hypermethylated in the CHH context , with a subset displaying CHG hypermethylation . Furthermore , these hypermethylated DMRs almost exclusively ( 80 of 81 ) overlapped with TEs ( Figure 3A ) . In contrast , the 19 hypomethylated PSI DMRs from Cluster 2 , associated with 13 unique genes , less frequently overlapped with TEs ( 42% overlap ) . Notably , most of the known key regulators of Pi homeostasis were present in both clusters , including the SPX genes , IPS1 , pre-miR827 and some PTs . Altogether , the majority of PSI DMRs are located in close proximity to genes , gain DNA methylation in response to Pi deprivation , preferentially in the CHH context , and almost exclusively overlap with TEs . 10 . 7554/eLife . 09343 . 015Figure 3 . Pi starvation-induced DMRs are enriched at key regulators of Pi homeostasis . ( A ) Hierarchical clustering of the differential methylation levels of the 100 root PSI DMRs associated with a significant change in nearby gene expression ( DEGs ) in response to Pi stress ( left panel ) . Middle panel represents the DMR-associated gene transcript abundance fold change ( log2 ) in response to Pi stresses , while right panel indicates DMRs that overlap with TEs ( coloured in black ) . ( B ) Scatter plots of the changes in DNA methylation ( compared to +Pi ) ( X ) against the changes in gene transcript abundance of the nearby associated gene ( FPKM , log2 ) ( Y ) for each of the 100 PSI DMRs , at various time points . Blue coloured plots represent Pi starvation [ ( −Pi ) − ( +Pi ) ] while red coloured plots represent Pi resupply [ ( Pi resupply ) − ( +Pi ) ] . Source data for Figure 3 is available at Dryad ( Secco et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 01510 . 7554/eLife . 09343 . 017Figure 3—figure supplement 1 . Changes in DNA methylation and gene transcription . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 017 The detailed time course analysed in this study provides the unique opportunity to decipher the temporal hierarchy between changes in transcription and changes in DNA methylation , which is critical to shed light on the potential causative relationships between the two , and the potential role of the PSI DMRs . The average changes in CNN methylation levels in the PSI DMRs in response to Pi stress were compared to the associated changes in nearby gene steady state transcript abundance ( Figure 3B , Figure 3—figure supplement 1 , Figure 2—source data 1 , available at Dryad , Secco et al . , 2015 ) . After 3 and 7 days of Pi starvation , 18% and 55% of the PSI DMR associated genes , respectively , showed significant changes in transcript abundance ( Cuffdiff , FDR < 0 . 05 ) , while only 5% and 9% of the 100 PSI DMRs showed significant changes in DNA methylation compared to +Pi samples ( t-test , FDR < 0 . 05 ) ( Figure 3B , Figure 3—figure supplement 1 , Figure 3—source data 1 , available at Dryad , Secco et al . , 2015 ) . Long term Pi deprivation ( ≥21 days −Pi ) was sufficient to induce significant changes in both gene transcript abundance and DNA methylation levels , with all PSI DMRs being significantly differentially methylated after 24 days of −Pi compared to +Pi ( t-test , FDR < 0 . 05 ) . Furthermore , while resupplying Pi starved plants with Pi for 3 days resulted in 87% of the DMR associated genes differentially expressed upon Pi starvation to return to Pi sufficient-like levels , 94% of PSI DMRs still showed significant differences in DNA methylation levels compared to +Pi conditions ( t-test , FDR < 0 . 05 ) . Finally , while 52 days of Pi starvation resulted in only 22% of PSI DMR associated genes showing significant changes in transcript abundance ( Cuffdiff , FDR < 0 . 05 ) compared to +Pi , likely a consequence of the floral transition and nutrient reallocation to the grains , the majority of PSI DMRs remained unaffected , with 83% of PSI DMRs still showing significant changes in DNA methylation after 52 days −Pi compared to 52 days +Pi ( t-test , FDR < 0 . 05 ) . Resupplying Pi for 31 days resulted in 11 of the 100 PSI DMRs showing significant persisting changes in DNA methylation ( t-test , FDR < 0 . 05 ) while no significant changes in gene transcription could be observed compared to +Pi . Taken together , Pi deprivation and Pi resupply appear to first rapidly modulate the transcript levels of genes induced by the stress before subsequently inducing changes in DNA methylation , indicating that the Pi starvation-induced changes in transcription precede , and are potentially causal , for the changes in DNA methylation . This induction of methylation may be involved in repressing the activity of specific TEs close to highly induced genes , via hypermethylation of the TEs in the CHH context . Due to the observation that differences in DNA methylation in some PSI DMRs may persist despite 31 days of Pi resupply , during which root fresh weight is increased by more than 2 . 2 fold ( Figure 1—figure supplement 1 ) , we wanted to assess the extent of potential transmission of changes in DNA methylation through mitosis . Hierarchical clustering of 45 root samples based on their methylation levels in all sequence contexts ( CNN ) in the 175 PSI DMRs revealed that the 31 days Pi recovered roots were more closely related to the −Pi samples than the +Pi samples , further indicating that some stress induced differences in DNA methylation can persist despite extended Pi recovery ( Figure 2—figure supplement 1 ) . However , resupplying Pi for 31 days resulted in 96 of the 175 PSI DMRs showing more than a 2 fold reduction in the differential DNA methylation levels induced by 52 days −Pi , when compared to +Pi , suggesting that the majority of PSI DMRs are returning towards Pi sufficient levels after 31 days of Pi resupply ( Figure 4—figure supplement 1 ) . Indeed , only 14 of the 175 PSI DMRs ( 11 hypermethylated , 3 hypomethylated ) showed significant differences in DNA methylation level ( CNN ) between 52 days +Pi compared to 52 days resupply ( t-test , FDR < 0 . 05 ) , while not presenting significant differences between 52 days −Pi vs 52 days resupply ( t-test , FDR > 0 . 05 ) , suggesting that changes in DNA methylation persist in only a limited subset of these PSI DMRs after 31 days of Pi resupply ( Figure 4A , Figure 4—source data 1 , available at Dryad , Secco et al . , 2015 ) . Of note , despite identifying significant persisting differences in DNA methylation , Pi resupplied samples often showed DNA methylation levels that were intermediate between +Pi and −Pi samples . These persistent differences in methylation levels preferentially occurred in the CHH context ( Figure 4A ) and were associated with 13 unique genes located nearby , including key regulators of Pi homeostasis such as SPX2 and MGD2 . Among these 14 persisting PSI DMRs , the single hypomethylated persisting DMR , namely the second DMR associated with SPX2 ( denoted SPX_DMR2 ) , showed the greatest change in DNA methylation level ( CNN ) in response to Pi stress , decreasing from 50% in +Pi to 1 . 3% in −Pi , as well as being maintained at a similar low level ( 1 . 5% ) despite 31 days of Pi resupply ( Figure 4A ) . Overall , it appears that only a small proportion of the Pi starvation induced changes in DNA methylation can be sustained despite 31 days of Pi resupply and active cell growth , while the majority of PSI DMRs have DNA methylation levels returning towards +Pi levels . 10 . 7554/eLife . 09343 . 018Figure 4 . Phosphate starvation induced DMRs are mainly transient . ( A ) Methylation levels ( CNN ) of the 14 regions showing persisting changes in DNA methylation despite 31 days of Pi resupply . Error bars indicates standard error . ( B ) Hierarchical clustering of the difference in methylation level of the 36 DMRs induced by 52 days of Pi deprivation in the panicles . Arrows indicate significant persisting changes in DNA methylation ( t-test , FDR < 0 . 1 ) . ( C ) Scatter plots representing the DNA methylation levels of the 175 PSI DMRs in the progeny of Pi deprived and non-stressed parents . Errors bars indicate standard error ( n = 5 ) . A linear trendline as well as its equation is shown . ( D ) Graphical representation of the changes in DNA methylation ( CNN ) of the 175 PSI DMRs between the first generation after 52 days of Pi treatment [ ( 52d −Pi ) − ( 52d +Pi ) ] and the offspring of stressed and non-stressed parents grown for 10 days in +Pi . PSI DMRs are sorted based on the change in DNA methylation levels at 52 days , in the first generation . Asterisks represent SPX_DMR2 , and error bars indicate standard error . ( E ) Genome browser screenshot of the SPX2 locus , showing the methylation levels in the first generation at 52 days , as well as in the second generation at 10 days . Source data for Figure 4 is available at Dryad ( Secco et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 01810 . 7554/eLife . 09343 . 022Figure 4—figure supplement 1 . Mitotic transmission of Pi starvation-induced changes in DNA methylation . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 02210 . 7554/eLife . 09343 . 023Figure 4—figure supplement 2 . Hierarchical clustering of the methylation levels ( CNN context ) of the 175 PSI DMRs in the progeny of stressed and non-stressed parents . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 023 In order to further investigate the observation that only a limited number of PSI DMRs can be transmitted through mitosis , we analysed a tissue that was only generated post-stress , the panicles from Pi recovered plants ( Figure 1A , B ) . Indeed , by the time the panicles appeared ( ∼40 days after initiation of treatment ) , the recovering plants had been re-supplied with Pi for ∼20 days , and should thus have a physiological status similar to that of Pi sufficient plants . After 52 days of Pi treatment , panicles were collected from plants continuously grown in +Pi or −Pi , as well as 21 days Pi starved plants resupplied for 31 days , using independent triplicates . Generation of high coverage methylomes enabled the identification of 36 PSI DMRs in the panicles of plants grown for 52 days in +Pi and −Pi ( FDR < 0 . 05 ) . However , no significant differences ( t-test , FDR < 0 . 05 ) could be observed in the DNA methylation levels of these 36 regions between panicles of plants grown for 52 days in −Pi compared to those resupplied with Pi for 31 days ( Figure 4B , Figure 4—source data 2 , available at Dryad , Secco et al . , 2015 ) . Yet , using a less stringent FDR cutoff of < 0 . 1 , two of the 36 PSI DMRs identified in the panicles , including SPX2_DMR2 , showed persisting differences in DNA methylation despite 31 days of Pi resupply , suggesting that a limited number of PSI DMRs could potentially be transmitted to newly generated tissue that has never experienced the stress . We next tested whether PSI DMRs could be transmitted from Pi stressed plants to their progeny , and thus potentially act as a transgenerational stress memory mechanism . To do so , individual seeds from five individual plants continuously grown under Pi sufficient or deficient conditions were harvested , and subsequently germinated and grown for 10 days under Pi sufficient conditions ( Figure 1A ) , before performing whole genome bisulfite sequencing on the corresponding root genomic DNA ( n = 5 for each condition , Supplementary file 1 ) . Analysis of the methylation levels in the progeny of stressed and non-stressed parents in the 175 root PSI DMRs that were identified in the first generation of plants failed to identify any significant differences in DNA methylation levels ( t-test , FDR < 0 . 05 ) ( Figure 4C , D , E , Figure 4—figure supplement 2 , Figure 4—source data 3 , available at Dryad , Secco et al . , 2015 ) . Indeed , independent of the treatment performed in the parental generation , DNA methylation levels were reset to a Pi-sufficient like level in the progeny ( Figure 4C , D , E , Figure 4—figure supplement 2 ) . While one DMR ( SPX2_DMR2 ) showed a difference in methylation level between the progeny of plants continuously grown under −P compared to the progeny of those continuously grown in +P , the difference was not statistically significant ( t-test , FDR < 0 . 05 ) . Therefore , it does not appear that these stress induced differential methylation states can be transmitted through meiosis . Thus , phosphate deprivation induces differential methylation in a variety of plant tissues in TEs that are close to PSI genes , and a limited subset of them , including DMRs associated with key regulators of Pi homeostasis such as SPX2 , can be mitotically transmitted to newly generated cells , despite an extended period of stress recovery . However , no evidence of transgenerational transmission of PSI changes in DNA methylation was observed . In order to shed light on the DNA methylation pathway mediating Pi starvation-induced changes in DNA methylation , the phosphate starvation experiment was repeated using an RNAi line that knocks down DCL3a , a key factor involved in the canonical RdDM pathway . In rice , there are two DCL3 genes , DCL3a and DCL3b , the latter also being known as DCL5 ( Fei et al . , 2013 ) . While DCL3b is mainly involved in the generation of stamen-specific 24-nt phased small RNAs , DCL3a is involved in producing 24-nt centromere-associated OsCentO siRNAs , MITE-derived siRNAs for abiotic stress responses , and non-canonical long miRNAs ( Wu et al . , 2010; Yan et al . , 2011; Song et al . , 2012; Wei et al . , 2014 ) . Furthermore , it has been shown that reducing the transcription of DCL3a via RNAi resulted in more than 80% of all 24-nt clusters being reduced by more than 3 fold , compared to WT ( Wei et al . , 2014 ) . In this study , WT and DCL3a RNAi plants were grown as previously described ( Figure 1A ) and subjected to 21 days of Pi sufficient or deficient conditions , before performing RNA-seq and whole genome bisulfite sequencing on the root tissues . RNA-seq analysis ( n = 3 ) confirmed that DCL3a was the most abundant DCL3 family member in the roots , with 11 times higher steady state transcript abundance than DCL3b under Pi sufficient conditions ( Figure 5—figure supplement 1A; Figure 5—source data1 , available at Dryad , Secco et al . , 2015 ) . In the DCL3a RNAi line , the transcript abundance of DCL3a decreased by 4 . 6 and 3 . 3 fold compared to WT in Pi sufficient or deficient conditions , respectively , while the abundance of DCL3b transcript was unaffected ( Figure 5—figure supplement 1A , B , Figure 5—source data1 , available at Dryad , Secco et al . , 2015 ) . In +Pi conditions , the reduced transcription of DCL3a was accompanied by widespread changes in DNA methylation , significantly altering DNA methylation levels in 9379 regions in the genome ( 6694 hypo- and 2685 hypermethylated regions in DCL3a RNAi line compared to WT; FDR < 0 . 05 ) , in addition to 3531 genes displaying a significant change in transcript abundance ( Cuffdiff , FDR < 0 . 05 ) compared to WT +Pi ( Figure 5A , Figure 5—figure supplement 1B ) . Similar results were also observed between WT and DCL3a RNAi plants under Pi deficient conditions , with 8164 DMRs and 1791 differentially expressed genes identified ( Figure 5A , Figure 5—figure supplement 1C , Figure 5—source data 1 , available at Dryad , Secco et al . , 2015 ) . Analysis of the DMRs induced in DCL3a RNAi plants revealed that the majority of changes in DNA methylation occurred in the CG and CHG contexts , under both Pi sufficient and deficient conditions ( Figure 5—figure supplement 1D , E , F , G ) . Among the 175 PSI DMRs identified previously ( Figure 2A ) , only 8 and 5 regions also showed significant differences in DNA methylation between WT and DCL3a plants under Pi sufficient and deficient conditions , respectively , and corresponded to 11 unique PSI DMRs ( 8 unique associated genes ) ( Figure 5A , B , C , D; Figure 5—source data 2 , available at Dryad , Secco et al . , 2015 ) . These 11 DMRs corresponded to 7 and 4 regions previously identified as hypo- and hypermethylated in response to −Pi in WT , respectively ( Figure 2A ) . Of the 4 hypermethylated PSI DMRs altered in the DCL3a RNAi plants , only one was associated with significant changes in nearby gene transcript abundance in response to −Pi in WT , while all the hypomethylated PSI DMR associated genes were significantly differentially regulated by −Pi in WT ( Figure 5—source data 2 , available at Dryad , Secco et al . , 2015 ) . Furthermore , while reduction in DCL3a transcript levels only moderately affected DNA methylation in 6 of the 11 PSI DMRs ( mC change < 15% ) , DNA methylation was almost completely abolished in all contexts in five of these hypomethylated PSI DMRs , which were associated with SPX1 , SPX2 and MGD2 ( Figure 5D , E , F ) . Among these , three DMRs were associated with SPX2 , and showed an average reduction in DNA methylation levels from 47% in WT +Pi to 1% in DCL3a RNAi plants +Pi , indicating a requirement for DCL3a in maintaining the methylation in these regions ( Figure 5E , F ) . Furthermore , among the 8 genes associated with DCL3a-dependant PSI DMRs , only SPX2 showed significant changes in transcript abundance between WT and DCL3a RNAi plants , decreasing by 69 fold under +Pi conditions ( 15 . 9 FPKM in WT +Pi and 0 . 2 FPKM in DCL3a RNAi plants +Pi ) ( Figure 5—figure supplement 1H ) . Despite a 2 fold difference in transcript abundance , no significant difference in SPX2 transcript abundance was observed between WT and DCL3a RNAi plants under −Pi conditions , suggesting that the Pi responsiveness of SPX2 is maintained in the DCL3a RNAi plants . 10 . 7554/eLife . 09343 . 024Figure 5 . DCL3a knockdown has limited effects on Pi starvation-induced DMRs . ( A ) Summary of the number of DMRs identified by comparing WT and DCL3a RNAi line ( referred to as dcl3a ) root methylomes ( 21 days ) under + and −Pi conditions . For each condition , the DCL3a RNAi line-induced DMRs were intersected with the 175 root PSI DMRs , revealing a total of 11 unique PSI DMRs that are significantly altered in DCL3a RNAi plants . ( B and C ) DNA methylation levels in the 175 PSI DMRs in the DCL3a RNAi line vs WT under −Pi and +Pi conditions , respectively . Asterisks represent the significant changes ( FDR < 0 . 05 ) in DNA methylation levels between WT and the DCL3a RNAi line . Black box indicates regions for which the DCL3a RNAi line almost completely abolishes DNA methylation , compared to WT . ( D ) Hierarchical clustering of the DNA methylation levels in the 175 PSI DMRs in the DCL3a RNAi line vs WT under −Pi ( blue ) and +Pi ( red ) conditions . ( E ) DNA methylation levels in 5 regions significantly affected in the DCL3a RNAi line . ( F ) Genome browser representation of the SPX2 locus , showing three DCL3a-dependant methylation regions , also responsive to Pi starvation , indicated by red boxes . Source data for Figure 5 is available at Dryad ( Secco et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 02410 . 7554/eLife . 09343 . 028Figure 5—figure supplement 1 . Down-regulation of DCL3a . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 02810 . 7554/eLife . 09343 . 029Figure 5—figure supplement 2 . Identification of PSI DMRs in the DCL3a RNAi line . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 029 In addition , in order to determine whether DCL3a is required for the stability of methylation at a distinct set of genomic loci upon Pi starvation , assessment of changes in DNA methylation between DCL3a RNAi plants grown in +Pi and −Pi conditions was conducted . Together , 409 PSI DMRs ( FDR < 0 . 05 ) were identified in the stressed DCL3a RNAi plants ( Figure 5—figure supplement 2 , Figure 5—source data 3 , available at Dryad , Secco et al . , 2015 ) . While the majority of changes in DNA methylation in these 409 DMRs were similar between WT and DCL3a RNAi plants ( Figure 5—figure supplement 2 ) , a subset of regions appeared to show changes in the DCL3a RNAi plants but not in WT , suggesting that DCL3a may mediate the stability of DNA methylation at some regions of the genome under these stress conditions . Associating each DMR to the nearest gene revealed 88 DMRs that were close to a gene that was significantly differentially expressed in WT upon Pi starvation ( Cuffdiff , FDR < 0 . 05 ) . However , these regions showed similar changes in DNA methylation upon Pi starvation in both WT plants and the DCL3a RNAi line . Thus , it appears that while DCL3a could potentially mediate the DNA methylation state of some genomic regions in response to Pi starvation , none of these were associated with changes in nearby gene expression . Overall , it appears that down regulation of DCL3a has a very limited effect on the methylation of hypermethylated PSI DMRs , while several regions , particularly those hypomethylated in response to −Pi in WT , fully require the presence of a functional DCL3a . Altogether , the majority of PSI DMRs appear to be independent of the canonical RdDM pathway and are thus likely regulated via a different pathway that does not require DCL3a . To determine whether the widespread PSI DMRs observed in rice could also be seen in other plant species , a similar approach was undertaken in Arabidopsis . Due to the shorter life cycle of Arabidopsis compared to rice , Arabidopsis seedlings were germinated and grown either under Pi sufficient ( 500 μM ) or deficient ( 13 μM ) conditions for 10 days , thus corresponding to a similar stress as the long-term Pi starvation performed in rice relative to the lifespan of the plant . In addition , 7 days after germination , half of the Pi starved plants were transferred to Pi sufficient conditions for 3 days to allow recovery . RNA-seq analysis was performed to assess the transcriptional response to the Pi treatments , identifying 4560 genes displaying differential transcript abundance in the roots after 10 days of Pi starvation ( Cuffdiff , FDR < 0 . 05 ) ( Figure 6A , Figure 6—source data 1 , available at Dryad , Secco et al . , 2015 ) . Known PSI marker genes such as SPX3 , IPS1 , miR399 and PHT1;9 showed >100 fold higher transcript abundance upon Pi starvation . In addition , resupplying Pi for 3 days was sufficient for the transcript abundance of the majority of the PSI genes to return to +Pi like levels . Indeed , of the 4560 PSI genes that displayed differential transcript abundance by 10 days of −Pi , only 10% showed significantly different transcript abundance after 3 days of resupply compared to +Pi . Together , this indicates that the Arabidopsis plants responded to both Pi deprivation and Pi resupply . 10 . 7554/eLife . 09343 . 030Figure 6 . Pi starvation in Arabidopsis results in a limited number of changes in DNA methylation . ( A ) Hierarchical clustering of significantly differentially regulated genes upon 10 days of Pi starvation and resupply . Arabidopsis seeds were germinated and grown in either Pi sufficient ( 500 µM Pi ) or deficient ( 13 µM Pi ) for 10 days . After 7 days of treatment , half of the Pi starved plants were re-supplied with 500 µM Pi for 3 days . ( B and C ) Genome browser representation of the two Arabidopsis PSI DMRs identified in this study ( FDR < 0 . 2 ) . AtCopeg1 ( Copia evolved gene 1 ) is a Copia like transposon previously shown to be induced by Pi starvation ( Duan et al . , 2008 ) , while MGD3 ( monogalactosyl diacylglycerol synthase 3 ) encodes the major enzyme for galactolipid metabolism during Pi starvation . ( C and E ) DNA methylation levels in the CNN , CG , CHG and CHH contexts in the three Pi treatments ( +Pi , −Pi and Pi resupply ) in the DMRs associated with Atcopeg1 and MGD3 , respectively . Asterisks represents significant differences in the methylation level ( t-test < 0 . 05 ) . Source data for Figure 6 is available at Dryad ( Secco et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 030 To examine the extent of DNA methylation changes in response to Pi starvation and recovery in Arabidopsis , high coverage whole genome bisulfite sequencing ( 91–96% cytosines covered , 92–96% of genome covered ) was performed on root samples pooled from 10 plants , in triplicate ( Supplementary file 1 ) . While no PSI DMRs could be detected using a FDR < 0 . 05 , two PSI DMRs were identified using lower stringency parameters ( FDR < 0 . 2 ) , requiring only two of the three replicates to show significant methylation differences for each condition . Assignment of these DMRs to the nearest gene revealed that they were associated with two known PSI genes , MGD3 , encoding for the major enzyme for galactolipid metabolism during phosphate starvation ( Kobayashi et al . , 2009 ) , and Atcopeg1 ( Copia evolved gene 1 ) , the only expressed gene derived from the AtCopia95 retrotransposon in the Arabidopsis genome , which has been shown to be involved in many developmental and adaptive processes , including Pi starvation ( Duan et al . , 2008 ) ( Figure 6B , D ) . Both DMRs overlapped with TEs and were hypermethylated in response to −Pi , preferentially in the CHH context ( Figure 6B , C , D , E ) . For both DMRs , resupplying Pi for 3 days resulted in a significant reduction in the CHG and CHH methylation levels compared to −Pi ( Figure 6C , E ) . For example , while under Pi sufficient conditions only 10% of the cytosines in the CHH context were methylated in the DMR associated with Atcopeg1 , Pi starvation increased the number of methylated cytosines in the CHH context to 40% , before returning to 24% upon 3 days of Pi resupply , suggesting that the changes in DNA methylation observed are transient . Overall , despite Pi starvation having a limited effect upon DNA methylation in Arabidopsis compared to rice , with only 2 PSI DMRs identified , the location of these PSI DMRs appears to be conserved , with hypermethylated regions overlapping with specific TEs that are close to highly induced PSI genes .
Plants have developed a wide array of mechanisms to tolerate changing environmental conditions . However , despite our understanding of Pi homeostasis greatly increasing during the past decade , very little is known about the potential role of changes in the epigenome in response to Pi starvation , and to stresses in general . Comprehensive methylome profiling coupled with an extensive time course experiment enabled investigation of the presence and potential role of dynamic epigenomic changes upon stress , revealing the temporal relationship between changes in DNA methylation and transcription , and the stability of altered DNA methylation states through mitosis and meiosis . Indeed , this study provides insights into the dynamic changes in DNA methylation in response to Pi starvation , with the identification of widespread changes in DNA methylation near PSI genes , in a species-dependent manner , likely correlated with the TE content of the genome . It is well known that plant genomes exhibit a very large diversity of size and repeat content as well as methylation levels , which were recently shown to be positively correlated , thus highlighting the central role of DNA methylation in repressing the potentially deleterious activity of TEs through hypermethylation ( Mirouze and Vitte , 2014 ) . However , the majority of DNA methylation analyses performed in plants to date have focused on Arabidopsis , despite being relatively depleted of TEs ( 15–20% of the genome ) and being poorly methylated compared to other plant genomes ( Mirouze and Vitte , 2014 ) . Potentially , the differences in DNA methylation changes observed in rice and Arabidopsis in response to Pi starvation could primarily be due to the differential frequency of TEs in the genome of each species . To date , only a limited number of studies have comprehensively investigated the involvement of DNA methylation in response to adverse environmental conditions ( Dowen et al . , 2012 , Liang et al . , 2014 ) . Several studies have reported that changes in the environment can affect the methylation status of some regions of the genome , using low resolution and non-quantitative techniques such as methylation-sensitive amplification polymorphism ( Choi and Sano , 2007; Hauser et al . , 2011; Wang et al . , 2011a , 2011b , Karan et al . , 2012; Shaik and Ramakrishna , 2012; Sahu et al . , 2013; Yu et al . , 2013 ) . However , these studies have lacked the resolution to provide the specific context and genomic location of the changes in DNA methylation , thus offering limited insights into the potential role of stress-induced changes in DNA methylation . While this study primarily focused in the main organ involved in nutrient sensing and uptake , namely the root , both roots and shoots appeared to exhibit similar changes in DNA methylation in response to Pi starvation . In addition , all but one of these DMRs overlapped with TEs . While , several studies have suggested that there is a negative relationship between methylation of TEs and nearby gene transcription ( Hollister and Gaut , 2009; Ahmed et al . , 2011; Eichten et al . , 2012 ) , we observed the opposite trend , with hypermethylated DMRs being located in close proximity to PSI genes . Consequently , we sought to shed light upon the temporal hierarchy , and thus potential causal relationships , between the observed environmentally-induced differential DNA methylation and the changes in gene transcript abundance . Using our time-course experiment , we demonstrated that upon Pi stress , changes in transcription occurred prior to changes in DNA methylation levels , thus suggesting that induction of gene transcription may be causal for the differential DNA methylation . In addition , since almost all the regions gaining DNA methylation in response to Pi deprivation overlap with TEs , such a process could potentially constitute a mechanism to repress the activity of specific TEs that are near highly induced PSI genes when plants are confronted with −Pi conditions . Indeed , vonHoldt and colleagues recently showed that the methylation pattern of TEs is a complex function of TE size , age and distance to a gene , with young TEs being preferentially poorly methylated ( vonHoldt et al . , 2012 ) . During Pi stress , PSI genes are highly expressed , presumably requiring increased chromatin accessibility to facilitate access of Pol II and other transcription machinery , which may facilitate the transcription of poorly methylated nearby TEs , with potentially deleterious effects upon the genome such as insertional inactivation of genes and ectopic recombination . This hypothesis is strengthened by the fact that the majority of the hypermethylated PSI DMRs are independent of DCL3a , a key component of the canonical RdDM pathway . Thus , we hypothesize that the RDR6-RdDM pathway , another entry point into RdDM that recognizes Pol II-derived TE transcripts , may act to methylate these TEs . Overall , the observed increase in DNA methylation in specific TEs located close to PSI genes provides insights into a cellular activity that may act to maintain localized suppression of transposable elements in genomic regions that must be transcriptionally activated to respond to environmental perturbation ( Figure 7 ) . 10 . 7554/eLife . 09343 . 032Figure 7 . Model of the role of DNA methylation in response to Pi starvation . Schematic of a PSI gene associated with two TEs . TE A represents TEs that are located distant from genes and are highly methylated , while TE B represents TEs that are close to genes and lowly methylated . Upon short term Pi deprivation , RNA polymerase II is recruited to Pi starvation-induced ( PSI ) genes , resulting in increased PSI gene expression . These high levels of RNA polymerase II could induce transcription of nearby and poorly methylated TEs , such as TE B , which could have deleterious effects upon the plant . As a consequence , upon prolonged Pi deprivation , TEs that are close to highly expressed stress-induced genes are hypermethylated , via a DCL3a-independent mechanism , thus preventing their transcription via RNA polymerase II . Black and white circles represent methylated and unmethylated cytosines , respectively . Thickness of the dashed lines represents the proportion of RNA polymerase II recruited to PSI genes or TEs . DOI: http://dx . doi . org/10 . 7554/eLife . 09343 . 032 In contrast , hypomethylated rice root PSI DMRs overlapped less frequently with TEs , and were often affected by knockdown of DCL3a , suggesting a different function of this subset of PSI DMRs , and likely involving the canonical RdDM pathway . Among the 11 DCL3a-dependent PSI DMRs , those associated with SPX1 and SPX2 were the most affected by reduced transcription of DCL3a . It has recently been shown that both SPX1 and SPX2 can inhibit phosphate starvation responses through Pi-dependant direct interaction with PHR2 , a key transcription factor controlling the majority of Pi-responsive genes . Under Pi sufficient conditions , SPX1 and SXP2 are tightly bound to PHR2 , preventing PHR2 from interacting with its cognate binding sites , and suppressing induction of Pi responsive genes ( Puga et al . , 2014 , Wang et al . , 2014 ) . Surprisingly , SPX2 transcript abundance was >65 fold lower in the DCL3a RNAi line compared to WT , potentially enabling PHR2 to interact with its target binding sites and induce transcription of PSI genes . Such a mechanism would thus explain the high number of differentially expressed genes observed between WT and the DCL3a RNAi line under +Pi conditions ( 3531 ) , compared to −Pi ( 1791 ) , where SPX2 is expressed at similar levels between WT and the DCL3a RNAi line . However , it remains unclear how the knockdown of DCL3a can repress SPX2 to such an extent in +Pi conditions . It is unlikely that the repression of SPX2 is a direct consequence of DCL3a-dependent changes in DNA methylation at this locus , since similar DNA methylation patterns are observed in DCL3a RNAi plants under +Pi and −Pi , while SPX2 is highly induced in −Pi and not expressed in +Pi ( Figure 5F ) . In addition , several studies have shown that changes in DNA methylation within or near DNA binding elements could interfere with the binding of cognate transcription factors ( Deng et al . , 2001; Bird , 2002; Zhong et al . , 2013 ) . Zhong and colleagues ( 2013 ) recently showed in tomato that binding sites for RIN ( Ripening Inhibitor ) , one the main transcription factors involved in fruit ripening , were frequently demethylated during ripening , thus enabling the induction of ripening genes ( Zhong et al . , 2013 ) . Such a mechanism could potentially occur in rice , fine-tuning the expression of specific genes involved in Pi homeostasis by regulating the binding capacity of specific transcription factors to PSI genes . However , the involvement of such a mechanism in the Pi starvation response remains to be determined . The Pi resupply experiment allowed us to determine whether PSI DMRs could be maintained despite an extensive period of Pi recovery and tissue growth . While the majority of PSI DMRs returned towards +Pi DNA methylation levels , 14 root PSI DMRs showed significant sustained changes in DNA methylation despite 31 days of Pi resupply , suggesting that these epialleles are likely to be transmitted to newly generated cells , and can thus be mitotically inherited . In addition , unlike animals where DNA methylation is reset in primordial germ cells and during embryogenesis , DNA methylation states in plants can be stably transmitted from parents to offspring ( Kinoshita et al . , 2007; Becker et al . , 2011; Schmitz et al . , 2011; Weigel and Colot , 2012 ) , and thus could potentially establish a transgenerational ‘memory’ of the stress . However , to date few reports have studied the potential heritability of stress-induced DNA methylation changes , and no evidence of transgenerational transmission of stress-induced differential DNA methylation exists ( Hauser et al . , 2011 , Pecinka and Scheid , 2012; Sahu et al . , 2013 ) . Boyko and colleagues reported that the progeny of stress-treated plants had increased homologous recombination frequency and global DNA methylation levels , as well as higher tolerance to stress ( Boyko et al . , 2010 ) . However , no evidence of transgenerational transmission of differences in DNA methylation was observed , and the increase in global genome methylation levels did not persist in successive generations during the absence of stress ( Boyko et al . , 2010 ) . Additionally , several recent studies suggest that specific mechanisms exist to prevent transgenerational inheritance of stress-induced epigenetic states ( Baubec et al . , 2014; Crevillen et al . , 2014; Iwasaki and Paszkowski , 2014 ) . Furthermore , Hagmann and colleagues , using a near-clonal North American A . thaliana population that has diverged under natural conditions for at least a century , recently showed that environment-induced changes are only minor contributors to durable genome-wide heritable epigenetic variation , with more than 97% of the total methylated genome space not being altered by the environment across dozens of generations ( Hagmann et al . , 2015 ) . In this study , we found no evidence of transgenerational inheritance of PSI DMRs , suggesting that PSI DMRs , and potentially most stress-induced differential methylation in general , are not transmitted to their progeny and are thus not likely to contribute to transgenerational stress memory . Altogether , this study reveals a species-specific process in which an abiotic stress induces dynamic and widespread changes in DNA methylation in rice . In addition , we establish the temporal relationship between differential transcription and DNA methylation , the limited stability of such induced DNA methylation events through mitosis , and the absence of their transmission through meiosis . These findings have important implications for interpreting the capacity of environmentally induced epialleles to influence genic transcription , and their stability through plant growth and reproduction .
Rice ( Oryza sativa L . cv . Nipponbare ) was used for all physiological experiments . Hydroponic experiments were performed under controlled conditions ( day/night temperature of 30/22°C and a 12 hr photoperiod , 200 µmol photons m−2 s−1 ) , allowing 0 . 5 l of hydroponic solution per plant . The hydroponic solution consisted of a modified solution as described in ( Secco et al . , 2013a ) , containing 1 . 425 mM NH4NO3 , 0 . 513 mM K2SO4 , 0 . 998 mM CaCl2 , 1 . 643 mM MgSO4 , 0 . 075 µM ( NH4 ) 6Mo7O24 , 0 . 25 mM NaSiO3 , 0 . 009 mM MnCl2 , 0 . 019 µM H3BO3 , 0 . 155 µM CuSO4 , 0 . 152 µM ZnSO4 and 0 . 125 mM EDTA-Fe , with or without 0 . 323 mM NaH2PO4 , resulting in the +Pi and −Pi conditions . The pH of the solution was adjusted to 5 . 5 and the solution was renewed every 3 day . Rice seeds were first pre-germinated in tap water for 2 days before being transferred into the hydroponic solution , containing 0 . 323 mM Pi ( +Pi ) for 2 weeks . Half of the seedlings were then transferred to a solution lacking Pi ( 0 mM Pi ) for 21 days , before being re-supplemented with 0 . 323 mM Pi for up to 31 days , while the other half of the seedlings continuously remained in +Pi conditions ( control ) . During the resupply experiment , half of the rice seedlings were left in Pi deficient media , to serve as control . After 24 days of Pi starvation , plants grown under Pi deficient conditions were supplemented with 0 . 03 mM Pi ( 1/10th of Pi sufficient Pi concentration ) until the end of the experiment to prevent them from dying . Roots and shoots were harvested separately at each time points . Furthermore , all ‘Materials and methods’ such as media replacement and sample collection were performed at similar time of the day ( 2 hr after light ) in order to minimize possible circadian effect . For Arabidopsis , seeds were germinated and grown vertically on Murashige and Skoog medium diluted 10-fold in Petri dishes supplemented with a Pi source of either 500 µM ( +P ) or 13 µM ( -P ) NaH2PO4 in a culture chamber under a 16-hr- light/8-hr-dark regime ( 25°C/22°C ) , as previously described ( Misson et al . , 2004 ) . Determination of Pi in tissues was measured by releasing the cellular content of cells into water by repeated freeze–thaw cycle , or by incubation for 1 hr at 85°C , and quantifying Pi by the molybdate assay according to the procedure of Ames ( Ames , 1966 ) . The total RNA from the roots and shoots tissues was extracted using TRIzol reagent ( Invitrogen , Carlsbad , CA ) , according to the manufacturer's instructions . For Arabidopsis , RNA was extracted using the Spectrum Plant Total RNA kit from Sigma ( St Louis , MO ) . The integrity and quality of the total RNA was determined using NanoDrop 1000 Spectrophotometer and formaldehyde-agarose gel electrophoresis . RNA was only used when the Abs260 nm/Abs280 nm ratio was >1 . 8 . For RNA-seq library synthesis ( 3 biological replicates per condition , 1 plant per replicate , except for Arabidopsis where only two replicates were used ) , total RNA was first depleted of rRNA using the Ribo-Zero rRNA removal kit ( Plant Leaf , and Plant Seed/Root kits , Epicentre , Madison , WI ) . To do so , 1 µg of total RNA from root samples was used as input for rRNA removal , while 2 µg of total RNA was used for shoot samples . Sequencing libraries were generated using the TruSeq RNA Sample Prep Kit ( Illumina , San Diego , CA ) . For the rice 24 day and 52 day time point samples and the Arabidopsis samples , the Illumina TruSeq Stranded Total RNA with Ribo-Zero Plant kit was used to generate the libraries ( Illumina ) . Genomic DNA ( gDNA ) was extracted plant tissues using the DNeasy Plant minikit ( Qiagen , the Netherlands ) . 600 ng of purified gDNA , spiked in with 0 . 5% ( wt/wt ) unmethylated Lambda DNA ( Promega , Madison , WI ) was used to prepare MethylC-seq libraries using three biological replicates per condition ( 1 plant per replicate for rice ) , as previously described ( Lister et al . , 2013 ) . In Arabidopsis , due to the reduced size of the roots and thus reduced amount of gDNA recovered , roots of 10 plants were pooled per replicate , and three independent replicates were used per time point . Briefly , gDNA was sonicated ( Covaris ) to a mean fragment size distribution of ∼200 bp . Sheared ends were repaired ( End-It DNA End Repair kit , Epicentre , Madison , WI ) , followed by 3′ A-tailing with 0 . 2 mM dATP and 15 U of Klenow fragment of DNA polymerase and ligation of single-read methylated adapters ( Illumina ) using 15 U of T4 DNA ligase ( New England Biolabs , Ipswich , MA ) . Sodium bisulfite conversion was then performed using the MethylCode Bisulfite conversion kit ( Life Technologies , Carlsbad , CA ) according to the manufacturer's instructions . Libraries were then amplified by PCR using 20 µl of bisulfite converted adapter-ligated DNA molecules were mixed with 25 µl of KAPA HiFi Hotstart Uracil+ ReadyMix ( Kapa Biosystems , Wilmington , MA ) and 5 µl of Truseq Primer Cocktail ( Illumina ) with cycle conditions of an initial denaturing at 95°C for 2 min , followed by 6 cycles of 98°C for 20 s , 60°C for 15 s , and 72°C for 1 min , followed by 10 min at 72°C . Libraries were then purified with AMPure XP beads ( Beckman Coulter , Brea , CA ) before being sequenced for 101 cycles using the Illumina HiSeq 1500 , as per manufacturer's instructions . All RNA-seq analyses were performed as previously reported ( Secco et al . , 2013a ) . Briefly TopHat2 and Cufflinks2 ( Trapnell et al . , 2012 , 2013 ) packages were used to map sequence reads to the rice IRGSP-1 . 0 and Arabidopsis TAIR10 reference genomes and quantitate differential gene transcription ( FDR < 0 . 05 ) . TEs were classified using the classification of Plant Repeat database ( http://plantrepeats . plantbiology . msu . edu/about . html#codes ) . Read mapping , processing , and analysis were performed as described previously ( Lister et al . , 2013 ) , aligning reads to the rice IRGSP-1 . 0 and Arabidopsis TAIR10 reference genomes . To estimate the bisulfite non-conversion frequency , the frequency of all cytosine basecalls at reference cytosine positions in the unmethylated control lambda genome was normalized by the total number of basecalls at cytosine reference positions in the lambda genome . DMRs were identified as previously described ( Lister et al . , 2013 ) . Briefly , for each cytosine in the CNN context , a root mean square test was performed as previously reported ( Perkins et al . , 2011 ) . To do so , a contingency table where rows indicated the position of each cytosine in the CNN context on the genome and the columns indicated the number of reads that supported a methylated cytosine or an unmethylated cytosine was generated . Next , p-values were simulated using 10 , 000 permutations and for each permutation , a new contingency table was generated by randomly assigning reads to cells with a probability equal to the product of the row marginal and column marginal divided by the total number of reads squared . To increase the efficiency of this process , the permutations were stopped when a p-value returned 100 permutations with a statistic greater than or equal to the original test statistic . The p-value cutoff that would control the FDR at our desired rate was determined as previously reported ( Bancroft et al . , 2013 ) . The largest p-value cutoff that still satisfied our FDR requirement was then chosen and significant sites showing changes in DNA methylation were combined into DMRs if they were within 200 bases of one another and had methylation changes in the same direction . Furthermore , blocks that contained fewer than 8 and 5 differentially methylated sites were discarded in the rice and Arabidopsis analyses , respectively . The sample comparison details for the identification of the DMR sets described in this study are described below . Rice root +Pi vs −Pi DMRs were identified ( FDR < 0 . 01 ) between ( 21 , 22 , 24 days ) +Pi and ( 21 , 22 , 24 days ) −Pi samples , requiring ≥7 of 9 root −Pi samples to be differentially methylated compared to root + Pi samples , as well as enabling 1 of 9 root −Pi samples to be similar to + Pi and vice versa . Rice shoot +Pi vs −Pi DMRs were identified ( FDR < 0 . 05 ) between 21 days +Pi and 21 days −Pi samples , requiring 3 of 3 −Pi shoot samples to be differentially methylated compared to +Pi shoot samples . Arabidopsis root +Pi vs −Pi DMRs were identified ( FDR < 0 . 2 ) between 10 days +Pi and 10 days −Pi samples , requiring 3 of 3 −Pi root samples to be differentially methylated compared to +Pi root samples . Persisting differences in DNA methylation levels were defined as being significantly different between 52 days + Pi and 52 days −Pi conditions ( t-test , FDR < 0 . 05 ) and significantly different between 52 days Pi resupply and 52 days +Pi ( t-test , FDR < 0 . 05 ) , but not significant different between 52 days −Pi and 52 days Pi resupply ( t-test , FDR >0 . 05 ) . Persisting DMRs were identified in CNN context . FDR was calculated using the Benjamini Hochberg method . Full browsing of the entire rice and Arabidopsis datasets can be found at http://www . plantenergy . uwa . edu . au/public/annoj/rice_Pi_starved_methylomes_Secco_et_al . htm and http://www . plantenergy . uwa . edu . au/public/annoj/Arabidopsis_Pi_starved_methylomes_Secco_et_al . htm , respectively . Transposable element track was obtained from Oryza Repeat Database ( http://rice . plantbiology . msu . edu/annotation_oryza . shtml ) . To calculate the methylation levels ( mCNN/CNN , mCG/CG , mCHG/CHG or mCHH/CHH ) , which is an estimate of the fraction of cytosines in the sequenced population which are methylated , we computed the fraction of all MethylC-Seq basecalls at cytosine reference positions that were cytosine ( protected from bisulfite conversion ) , and then subtracted these estimates for the failure of the chemical conversion of unmethylated cytosine ( non-conversion rate ) , based on the non-conversion rate to the unmethylated Lambda DNA spiked-in control . Illumina reads of all samples have been submitted to the Sequence Read Archive at the National Center for Biotechnology Information ( http://www . ncbi . nlm . nih . gov/sra ) under accession number SRP061678 ( Stranded RNA-seq for rice after 3 , 24 and 52 days of Pi stress ) , SRP032765 ( Rice methylomes under Pi stress ) , SRP061677 ( Stranded RNA-seq and methylomes for the rice DCL3a RNAi experiment ) , and SRP040029 ( Methylomes and stranded RNA-seq for Arabidopsis under Pi stress ) . Source data are deposited in the DRYAD repository: 10 . 5061/dryad . 40gd6 . ( Secco et al . , 2015 ) . | Phosphate is an important nutrient for all living organisms . This chemical group forms part of the backbone of DNA molecules , and has a crucial role in many chemical reactions that occur inside cells . Plants in particular need a source of phosphate to grow . This is why agricultural fertilizers are rich in phosphate , but unfortunately , the use of fertilizers is not sustainable . Many researchers are now looking for new ways to maintain high crop yields without chemical fertilizers , and understanding how crops are affected in times of shortage will be pivotal to achieving this goal . DNA contains coded information in the form of genes , which can either be switched on or off . Chemical marks added to the DNA can earmark genes for activation or inactivation , a bit like handwritten annotations in an instruction manual . One example is the addition of a chemical tag called a methyl group to one of the letters of the DNA code—so-called ‘cytosine methylation’ . However , little is known about how the pattern of these chemical marks on DNA changes in response to changes in the environment . Secco et al . investigated changes in cytosine methylation in both rice and Arabidopsis plants that had been deprived of phosphate . Arabidopsis , or thale cress , is a model plant that is often studied by plant biologists because it is small and grows quickly . The experiments showed that when rice plants were not given enough phosphate , the pattern of DNA methylation changed . This was particularly true around certain genes that help the plants survive in these difficult conditions . Notably , in the absence of phosphate , methylation also occurs more often in DNA sequences called transposable elements that sit close to these useful genes , and less often around other genes . Transposable elements , also known as ‘jumping genes’ can move within the genome and thus potentially have damaging effects through altering the DNA sequence . However , DNA methylation normally prevents this from happening . Therefore , the extra methylation observed by Secco et al . may be a cautionary measure to inactivate these transposable elements and limit their potential deleterious effects . Further experiments went on to show that these useful genes seem to be switched on before the DNA of these transposable elements is methylated , implying that the extra methylation observed in these transposable elements is a consequence of the activation of these nearby useful genes . By contrast , similar experiments performed in Arabidopsis reveal a very limited change in DNA methylation when the plants are grown under stressful conditions . This might be because Arabidopsis has considerably fewer transposable elements than rice . The next challenge will be to explore how significant the environmentally induced silencing of transposable elements is to the stress responses and genome integrity of crop plants . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"genetics",
"and",
"genomics"
] | 2015 | Stress induced gene expression drives transient DNA methylation changes at adjacent repetitive elements |
Single-cell RNA sequencing has spurred the development of computational methods that enable researchers to classify cell types , delineate developmental trajectories , and measure molecular responses to external perturbations . Many of these technologies rely on their ability to detect genes whose cell-to-cell variations arise from the biological processes of interest rather than transcriptional or technical noise . However , for datasets in which the biologically relevant differences between cells are subtle , identifying these genes is challenging . We present the self-assembling manifold ( SAM ) algorithm , an iterative soft feature selection strategy to quantify gene relevance and improve dimensionality reduction . We demonstrate its advantages over other state-of-the-art methods with experimental validation in identifying novel stem cell populations of Schistosoma mansoni , a prevalent parasite that infects hundreds of millions of people . Extending our analysis to a total of 56 datasets , we show that SAM is generalizable and consistently outperforms other methods in a variety of biological and quantitative benchmarks .
Single-cell RNA sequencing ( scRNAseq ) datasets typically contain tens of thousands of genes , although many of them may not be informative for differentiating between cell types or states . Feature selection is thus commonly used to select a subset of genes prior to downstream analyses , such as manifold reconstruction and cell clustering ( Crow et al . , 2018; Satija et al . , 2015; Vallejos et al . , 2015 ) . However , current approaches have two major limitations . First , feature selection methods filter genes based on arbitrarily or empirically chosen thresholds , small changes in which may result in different gene sets ( Vallejos et al . , 2017 ) . In addition , the selection of features typically operates under the assumption that genes with highly variable expression between individual cells capture biologically meaningful variation . Because single-cell transcriptomes are inevitably contaminated by a combination of random transcriptional and technical noise ( Grün et al . , 2014 ) , the variation in biologically relevant genes may be hard to distinguish from the background noise , especially when the differences between cell populations are subtle . Resolving these differences , or ‘signals’ , is essential to study a variety of biological problems , including identifying cell subtypes ( Olsson et al . , 2016; Treutlein et al . , 2014; Lönnberg et al . , 2017; Fincher et al . , 2018; Baron et al . , 2016; Schwalie et al . , 2018 ) and quantifying the effects of molecular perturbations to otherwise homogeneous populations of cells ( Lane et al . , 2017 ) . In such datasets , only a small fraction of the genes , and therefore only a small fraction of the total variation , may contain the signals relevant for distinguishing cell types or cell states . Choosing these features without a priori knowledge remains an unmet computational challenge . The second limitation is that existing methods have been almost exclusively benchmarked on well-annotated , gold standard datasets with clearly distinguishable cell types ( Wang et al . , 2017; Kiselev et al . , 2017; Duò et al . , 2019; Bahlo et al . , 2018 ) . These datasets are not informative for distinguishing the performance between methods , because the differences between cell types are relatively straightforward to detect . However , evaluating the performance of feature selection and/or dimensionality reduction methods on datasets with more subtle signals is difficult as their ground truth labels are typically ambiguous or nonexistent . To overcome the shortcomings of current feature selection approaches , here , we introduce the Self-Assembling Manifold ( SAM ) method , an unsupervised , ‘soft feature selection’ algorithm that iteratively rescales gene expressions to refine a nearest neighbor graph of cells until the graph converges to a stable solution . At each iteration , SAM assigns more weight to genes that are spatially variable across the constructed graph , and this weighted gene expression is then used to improve the next nearest neighbor assignment . SAM presents two advantages: it rescales all genes according to their weights , solving the problem of thresholding , and it prioritizes genes that are variable across the intrinsic manifold of the data rather than selecting genes that are variable across individual cells . In order to better distinguish the performance between methods , we define a network sensitivity measure to identify datasets with subtle signals . With limited annotations in most high-sensitivity datasets , we introduce unsupervised graph-based metrics to quantify the degree of structure within the reconstructed manifolds for comparison between methods . In addition , we perform benchmarking using known ground truth labels on simulated datasets spanning a wide range of sensitivities by introducing increasing levels of noise to well-annotated datasets . These analyses reveal that SAM consistently improves feature selection and cell clustering . To demonstrate the utility of SAM in practice , we provide an in-depth analysis of two datasets that are challenging to analyze using existing methods: stem cells in a human parasitic worm , Schistosoma , and activated macrophages ( Lane et al . , 2017 ) . We show that SAM can capture novel biology undetectable by other methods and validate these results with experimental evidence .
The SAM algorithm begins with a random k-nearest neighbor ( kNN ) graph and averages the expression of each cell with its k nearest neighbors: C=1kNE , where N is the directed adjacency matrix and E is the gene expression matrix ( Figure 1a ) . For each gene i , SAM computes a spatial dispersion factor of the averaged expressions Ci , which measures variation across neighborhoods of cells rather than individual cells ( Materials and methods ) . These dispersions are used to calculate the gene weights , which then rescale the expression matrix: E^=EWD , where WD is a diagonal matrix with gene weights along the diagonal . Using the rescaled expressions E^ , we compute a pairwise cell distance matrix and update the assignment of each cell’s k-nearest neighbors accordingly . This cycle continues until the gene weights converge . To demonstrate the implementation and utility of SAM , below we analyze a challenging dataset comprised of a few hundred relatively homogeneous stem cells isolated from Schistosoma mansoni ( Figure 1—figure supplement 1 ) , a widespread human pathogen ( Hoffmann et al . , 2014 ) . Using a protocol we have established previously ( Wang et al . , 2018 ) , these cells were collected by sorting dividing cells from juvenile parasites harvested from their mouse hosts at 2 . 5 weeks post infection . At this stage , the parasites use an abundant stem cell population ( ~15–20% of the total number of cells ) for rapid organogenesis and growth ( Wang et al . , 2013; Wang et al . , 2018 ) . Testing several existing methods ( Wang et al . , 2017; Kiselev et al . , 2017; Satija et al . , 2015 ) , we found that they were not able to identify distinct cell populations in this dataset . In contrast , SAM finds a stable solution independent of initial conditions ( Figure 1b ) . A graph structure with clearly separated cell populations self-assembles through the iterative process ( Figure 1c ) . In parallel , the gene weights converge onto the final weight vector . Eventually , only a small fraction of genes ( ~1% ) are strongly weighted and useful for separating cell clusters , reflecting the inherent difficulty of analyzing this dataset . Figure 1d shows that SAM iteratively improves a series of graph characteristics , including the network-average clustering coefficient ( NACC ) , modularity , and Euclidean norm of the spatial dispersions ( Materials and methods ) . The NACC and modularity quantify the degree of structure within the graphs – graphs with high NACC and modularity have regions of high density separated by regions of low density . The dispersion quantifies the spatial organization of gene expression – the higher the spatial dispersion the less uniformly distributed the gene expressions are along the graph . The final graph metrics are independent of initial conditions , which can start from a random graph or the output of an existing manifold reconstruction algorithm ( e . g . Seurat , Satija et al . , 2015 ) . Importantly , we verified that SAM does not artificially boost these metrics in data that lack inherent structure: when applying SAM to a randomly shuffled expression matrix , none of these metrics increased from the random initial conditions . Visualizing the converged graph in two dimensions using Uniform Manifold Approximation and Projection ( Becht et al . , 2019 ) , we find that cells can be separated into three main populations , with Louvain clustering ( Blondel et al . , 2008 ) further splitting one of these clusters into two subpopulations ( Figure 2a ) . In contrast , other commonly used dimensionality reduction methods , such as principal component analysis ( PCA ) , Seurat ( Satija et al . , 2015 ) , and SIMLR ( Wang et al . , 2017 ) , failed to distinguish these cell populations ( see Materials and methods for the selection of algorithms for comparison ) . Clustering the Seurat graph using Louvain clustering still results in a low-modularity partition and poor correspondence to the SAM cluster assignments . Supplementary file 1 lists genes with high SAM weights , which includes most markers previously implicated to be enriched in subsets of schistosome stem cells ( Wang et al . , 2013; Wang et al . , 2018 ) . Figure 2b shows that the three populations include previously characterized δ′-cells , which specifically express an RNA binding protein nanos-2 ( Smp_051920 ) , and ε-cells , which are marked by the expression of eledh ( eled , Smp_041540 ) ( Wang et al . , 2018 ) . More importantly , SAM reveals a novel stem cell population , μ , comprising ~30% of all sequenced cells ( μ denotes muscle progenitors as discussed below ) . While μ-cells express ubiquitous stem cells markers ( e . g . ago2-1 , Smp_179320; cyclin B , Smp_082490 ) and cell cycle regulators ( Figure 2—figure supplement 1a ) ( Collins et al . , 2013; Wang et al . , 2013; Wang et al . , 2018 ) , they are also strongly enriched for a large set of genes , with a calcium binding protein ( cabp , Smp_005350 ) , an actin protein ( Smp_161920 ) , an annexin homolog ( Smp_074140 ) , a helix-loop-helix transcription factor ( dhand , Smp_062490 ) , and a phosphatase ( dusp10 , Smp_034500 ) as the most specific markers of this population in comparison to other stem cells ( Figure 2—figure supplement 1b ) . Fluorescent in-situ hybridization ( FISH ) in conjunction with EdU labeling of dividing cells reveals that μ-cells ( cabp+EdU+ ) are distributed near the parasite surface right beneath a layer of post-mitotic differentiated cells that also express cabp ( Figure 2c ) . Close to the parasite surface , there are two major cell types intertwined in space: the skin-like tegumental cells and the body wall muscle cells . However , μ-cells express none of the recently identified markers in tegumental progenitors ( Wendt et al . , 2018 ) , suggesting that they may be associated with the muscle lineage . To test this idea , we performed double FISH experiments and observed in post-mitotic cabp+ cells the coexpression of a set of canonical muscle markers ( Witchley et al . , 2013 ) , including tropomyosin ( Smp_031770 ) , myosin ( Smp_045220 ) , troponin ( Smp_018250 ) , and collagen ( Smp_170340 ) ( Figure 2d ) . These results suggest that cabp may mark the parasite body wall muscles and μ-cells are likely muscle progenitors , although functional validation is required to support this observation . Why the juvenile parasites maintain such an active pool of muscle progenitors will be an important question for future studies . In addition , SAM identifies two subpopulations among ε-cells: εɑ-cells that are highly enriched for an aschaete-scute transcription factor ( astf , Smp_142120 ) , and εβ-cells that abundantly express another basic helix-loop-helix protein ( bhlh , Smp_087310 ) ( Figure 2b , right panels ) . FISH experiments confirm these cells to be in close spatial proximity but with no coexpression of astf and bhlh ( Figure 2e ) . Moreover , we observed with FISH that there are fewer astf+ cells in larger , more matured juveniles , suggesting εɑ-cells are a dynamic population during development . To verify this observation , we sequenced another ~370 stem cell from juveniles at a later developmental time point ( 3 . 5 weeks post infection ) . After correcting for batch effects in the combined 2 . 5- and 3 . 5- week datasets using the mutual nearest neighbors ( MNN ) algorithm ( Haghverdi et al . , 2018 ) , we find that δ′- , μ- , and εβ-cells remain relatively constant throughout both time points , whereas εɑ-cells comprise a significantly smaller fraction of the stem cells at 3 . 5 weeks ( 7% ) compared to 21% at 2 . 5 weeks ( Figure 2f ) . Taken together , these analyses demonstrate that SAM can identify experimentally validated stem cell populations that are previously too subtle to separate using other methods but are closely associated with the schistosome development . The critical difference between SAM and other methods lies in how they select genes for manifold reconstruction . SAM prioritizes genes with variable expressions across neighborhoods of cells rather than individual cells as in other methods ( e . g . Seurat ) . Figure 2g shows that genes with high standardized dispersion across individual cells often have low SAM weights , indicating that these highly variable genes ( HVGs ) are irrelevant to the topological relationships between cells . Other methods ( e . g . SC3 , Kiselev et al . , 2017 ) identify marker genes based on differential gene expression between cell clusters , but this approach suffers when cell cluster assignment is poor , especially when discrete cell groups are difficult to separate or absent . Indeed , SC3 failed in the default mode as it incorrectly predicted there to be only one cluster in the schistosome dataset . After we manually increased the number of clusters , SC3 could recover a few of the marker genes associated with only one ( μ-cells , blue symbols in Figure 2h ) of the populations detected by SAM . Furthermore , changing the number of clusters resulted in different solutions and large variability in SC3 scores for its top ranked genes . Below , we assess the general applicability of SAM by benchmarking its performance against state-of-the-art scRNAseq analysis methods on a large collection of datasets . We focus on three methods , that is , Seurat , SIMLR , and SC3 , as they are mostly unsupervised , have been broadly used , and were shown to outperform other methods through extensive benchmarking ( Kiselev et al . , 2017; Wang et al . , 2017; Duò et al . , 2019; Bahlo et al . , 2018; Tian et al . , 2019 ) . The criteria to select algorithms for comparison are explained in Materials and methods . We first benchmark against nine datasets ( Supplementary file 2 ) that have high-confidence annotations to evaluate the accuracy of SAM in assigning cell clusters . Seven of these datasets are of pancreatic islet cells , as their subpopulations have been extensively characterized with known marker genes ( Baron et al . , 2016 ) . For five out of the nine datasets , SAM has the highest Adjusted Rand Index ( ARI , a measure of clustering accuracy ) ( Hubert and Arabie , 1985 ) with respect to the provided annotations ( Figure 3a ) . On the remaining four Baron datasets , SAM and Seurat perform equally well with near perfect clustering accuracy , whereas SC3 and SIMLR tend to overestimate and underestimate the number of clusters , respectively . Supplementary file 3 lists the clustering scores for each method and for each annotated cell type in the benchmarking datasets ( Materials and methods ) . SC3 and SIMLR struggle to cleanly cluster cell types that constitute large fractions of the data , such as the alpha and beta cells in the pancreatic datasets . While Seurat performs well on the Baron datasets , it fails to identify alpha cells in the Wang and Muraro datasets when run with default parameters , although its performance is improved after optimizing parameters to maximize its clustering accuracy ( Materials and methods ) . We note that this parameter optimization is impossible to perform on an experimental dataset with no available ground truth labels . Nevertheless , even with optimal parameters , Seurat has accuracy lower than or equal to that of SAM on all datasets . SAM converges to the same set of gene weights for all datasets analyzed ( Figure 3b , Figure 3—figure supplement 1a ) and its performance is robust to the choice of parameters and random initial conditions ( Figure 3—figure supplement 1b–c ) . In contrast , applying SAM to randomly generated datasets ( Materials and methods ) , the resulting gene weights are highly dissimilar across random initial conditions ( Figure 3b ) , indicating that SAM does not converge to a stable solution on datasets with no intrinsic structure . Finally , the scalability of SAM is similar to that of Seurat , capable of analyzing hundreds of thousands of cells in minutes ( Figure 3c ) , whereas SIMLR and SC3 are orders of magnitudes slower and thus excluded from further benchmarking which requires the analysis of many more datasets . Because the nine benchmarking datasets are all comprised of clearly distinguishable cell types , they may not represent the performance of methods on other datasets that contain cell populations that are only subtly different . To identify such datasets , we introduce a network sensitivity metric that quantifies the changes in the cell-to-cell distances when randomly selecting a subset of features from the gene expression matrices ( Materials and methods ) . High network sensitivity indicates that changes to the selected features strongly alters the resulting topological network . Networks that are robust to the selected features correspond to datasets that have many redundant signals or genes corroborating the network structure . In the datasets we compiled ( Supplementary file 2 ) , all broadly used benchmarking datasets have lower sensitivities whereas the schistosome dataset , which we have shown to be challenging to analyze , has the highest sensitivity ( Figure 4a ) . The fraction of genes with large SAM weights ( >0 . 5 ) is negatively correlated with the network sensitivity , suggesting that the biologically relevant variation in datasets with high sensitivity is captured by relatively fewer genes ( Figure 4b ) . Analyzing all 56 datasets , we found that SAM improves the clustering , modularity , and spatial organization of gene expression across the graph in comparison to Seurat as the datasets become increasingly sensitive ( Figure 4c ) . Evaluating the clustering accuracy for the highly sensitive datasets , however , is challenging , because many of them have incomplete or nonexistent cell type annotations . Therefore , we use the nine well-annotated benchmarking datasets to simulate data across a wide spectrum of sensitivities . For this , we corrupt the data by randomly permuting gradually increasing fractions of the gene expressions . As illustrated by the Darmanis dataset ( Darmanis et al . , 2015 ) , Figure 5a shows that the sensitivity increases along with the corruption . Below ~50% corruption , SAM’s ARI scores only marginally decrease as the corruption ( and thereby sensitivity ) increases , whereas Seurat’s performance rapidly deteriorates , even when run with optimal parameters . A similar contrast was observed between SAM and Seurat with the NACC , modularity , and spatial dispersion . Importantly , passing the genes with high SAM weights into Seurat rescued its performance across all metrics , indicating that SAM is able to consistently capture the genes relevant to the underlying structure of the data even with increasing levels of noise and illustrating the robustness of its feature selection strategy compared to the HVG filtering approach used by Seurat . These observations generalize to all nine benchmarking datasets , quantified by the area under the curves ( AUC ) of the metrics with respect to corruption ( Figure 5b ) . We next highlight another dataset to show that SAM can recover biologically meaningful information that other methods fail to capture . We chose this example , which contains ~600 macrophages treated with lipopolysaccharide ( LPS ) when individually trapped in microfluidic channels ( Lane et al . , 2017 ) , because it has high network sensitivity ( Figure 4a ) and has accompanying single cell functional data of macrophage activation dynamics that may help to validate the results of our analysis . Applied to this dataset , SAM initially identifies two clusters ( Figure 6a , top ) . Performing gene set enrichment analysis ( GSEA , Subramanian et al . , 2005 ) , we find that genes with high SAM weights are dominated by cell cycle-related processes , with one of the clusters heavily enriched for cell cycle genes ( e . g . Top2a , Mki67 , Figure 6—figure supplement 1a ) . After removing the cell cycle effects ( Materials and methods ) , SAM identifies two different clusters in which cells are properly ordered by the time since LPS induction , with the highly weighted genes being primarily involved in immune signaling ( Figure 6a , bottom ) . These observations demonstrate that , in conjunction with GSEA , the quantitative gene weights output by SAM can be used to infer the biological pathways that drive the clustering of cells . One of the two clusters is enriched for TNFα expression ( Figure 6b ) . It is known that LPS activates two independent pathways , one through the innate immune signal transduction adaptor ( Myd88 ) and the other through the TIR-domain-containing adapter-inducing interferon-β ( TRIF ) ( Lee et al . , 2009 ) . While the Myd88 pathway directly activates NF-κB , the TRIF pathway first induces the secretion of TNFα , which subsequently binds to its receptor , TNFR , to prolong the activation of NF-κB ( Figure 6c ) . Figure 6d and Figure 6—figure supplement 1b show examples of genes that are highly enriched with TNFα , a number of which are inflammatory factors known to accumulate due to prolonged NF-κB activation ( Lane et al . , 2017 ) . These results suggest that SAM grouped the cells based on their activated signaling pathways: one cluster is activated through both Myd88 and TRIF pathways ( MT ) , while the other is only activated through Myd88 ( M ) . To further verify that the separation between the MT and M clusters truly reflects the dichotomy in cellular response to LPS induction , we noted that this dataset combines scRNAseq with live-cell imaging of NF-κB activity in single cells . This allows us to directly test if the MT and M clusters correspond to different signaling dynamics ( Materials and methods ) . We found that most of the cells with prolonged NF-κB response ( i . e . cells showing broad peaks of NF-κB activation in time ) are in fact in the MT cluster ( Figure 6e–f , and Figure 6—figure supplement 2a ) , consistent with the expectation that TNFα signaling prolongs NF-κB activation . Although our interpretation of the data matches that provided in the original study , we were able to analyze the dataset with almost no a priori knowledge . In contrast , the original study required extensive manual curation , analyzed only a subset of the dataset , and could not group cells by their NF-κB activation dynamics based on the gene expression data alone . Similarly , Seurat and SIMLR were unable to order the cells by the time since LPS induction or group cells based on their activation dynamics after removing the cell cycle effects ( Figure 6g , and Figure 6—figure supplement 2b–c ) .
Here , we introduced a scRNAseq analysis method , SAM , which uses an unsupervised , robust , and iterative strategy for feature selection and manifold reconstruction . As demonstrated by our analysis of the schistosome stem cells and activated macrophages , SAM can capture biology that is undetectable by other methods . While SAM has consistently higher clustering accuracy than other state-of-the-art methods on datasets containing clearly distinct cell types , its advantages are especially apparent on datasets in which cell states or types are only distinguishable through subtle differences in gene expression . The strength of SAM lies in the integration of three algorithmic components: spatial dispersion to measure feature relevance , soft feature selection , and the iterative scheme . By averaging the gene expression of a cell with that of its neighbors , the spatial dispersion quantifies the variation across neighborhoods of cells rather than individual cells . Genes with high spatial dispersion are more likely to be biologically relevant as they are capable of separating cells into distinct topological locations . Soft feature selection includes all genes and weights their contribution to the manifold reconstruction by their spatial dispersions . This mitigates the shortcoming of existing approaches in which the selection of features is a binary decision: genes are either included or not depending on arbitrarily chosen thresholds . The conceptual challenge here is that calculating the gene weights requires the manifold , but reconstructing the manifold requires the gene weights for feature selection . SAM thus uses an iterative strategy to converge onto both the gene weights and the corresponding graph topology from a random initial graph . Each successive iteration refines the gene weights and network structure until the algorithm converges . Empirically , for all datasets analyzed we have shown that SAM converges onto a stable solution and is robust to the random initial conditions . Practically , it is possible to initialize SAM using the graph output of another method such as Seurat ( Figure 1d ) , but using random initial conditions avoids potential biases in the analysis and enables the evaluation of the stability of SAM . To demonstrate the strengths of SAM in practice , we analyzed the schistosome stem cells and identified novel stem cell populations that were validated by FISH experiments ( Figure 2 ) . In the analysis of activated macrophages , we showed that SAM can simultaneously order cells by the time since LPS induction and group cells according to their respective activated signaling pathways . We have validated this result using the live-cell imaging data presented in the original study ( Figure 6 ) . We expect that the application of SAM is not limited to feature selection , cell clustering , and manifold reconstruction; it can be readily integrated with existing analytical pipelines as its gene weights and reconstructed manifolds can be used in downstream analyses . For example , we have shown how the genes ranked by their SAM weights can be used as input to GSEA to determine the biological processes enriched among the highly weighted genes ( Figure 6 ) , thus directly testing if the weights reflect the biological relevance of genes . Additionally , the manifold reconstructed by SAM can be used as input to pseudotemporal ordering algorithms ( Setty et al . , 2016; Trapnell et al . , 2014 ) . Beyond the two example case studies , we have rigorously evaluated SAM on a total of 56 datasets . While previous studies benchmarked on datasets with clearly defined cell populations , we defined a network sensitivity measure to rank the datasets based on the inherent difficulty of their analysis ( Figure 4 ) . Using these datasets , we showed that SAM consistently outperforms other methods in terms of both cell clustering accuracy measured by ground truth annotations , and manifold reconstruction measured by quantitative graph characteristics . These improvements can be attributed to the robust selection of features relevant for cell clustering and manifold reconstruction even in the presence of significant amounts of random noise , as shown in the corruption tests ( Figure 5 ) . Overall , the network sensitivity and quantitative benchmarking metrics should help in characterizing the performance of future scRNAseq analysis methods across a wider variety of datasets .
The SAM source code and tutorials can be found at https://github . com/atarashansky/self-assembling-manifold ( Tarashansky , 2019; copy archived at https://github . com/elifesciences-publications/self-assembling-manifold ) . We have included a number of tutorials describing in detail the various functions , parameters , attributes , and data structures of the SAM package , and provided the documentation ( docstrings ) for all functions available to users . In addition , we have developed an interactive user interface that facilitates the convenient exploration of single-cell data and SAM parameters ( Figure 1—figure supplement 2 ) . A Jupyter notebook tutorial explaining how to use the interface is provided as well . The schistosome stem cell scRNAseq data generated in this study were obtained in two sequencing batches and are available through the Gene Expression Omnibus ( GEO ) under accession number GSE116920 . Supplementary file 2 summarizes all datasets used in this study as well as the methods used to convert raw sequence read counts to gene expression , such as TPM ( transcripts per million ) , CPM ( counts per million ) , RPKM ( reads per kilobase per million ) , or FPKM ( fragments per kilobase per million ) . Datasets with asterisks next to their accession numbers are sourced from the conquer database ( Soneson and Robinson , 2018 ) . The nine benchmarking datasets used with high-confidence annotation labels are marked by crosses . Gene expression is measured in log space with a pseudocount of 1 ( e . g . log2 ( TPM+1 ) ) . Genes expressed ( log2 ( TPM+1 ) >1 ) in fewer than 1% or more than 99% of cells are excluded from downstream analysis as these genes lack statistical power . To reduce the influence of technical noise near the molecular detection limit , we set gene expression to zero when log2 ( TPM+1 ) <1 . We denote the resulting expression matrix as E . In the SAM algorithm ( see below ) , we either standardize the gene expression matrix E to have zero mean and unit variance per gene ( which corrects for differences in distributions between genes ) or normalize the expressions such that each cell has unit Euclidean ( L2 ) norm ( which prevents cells with large variances in gene expressions from dominating downstream analyses ) prior to dimensionality reduction . In the below section , we denote the standardized or normalized expression matrix as E¯ . Empirically , we have found that standardization performs well with large , sparse datasets that are expected to contain many subpopulations , whereas L2-normalization is more suitable for smaller datasets with fewer subpopulations . This is likely due to the fact that standardization amplifies the relative expression of genes specific to small populations in large datasets , thereby making them easier to identify . In contrast , standardization decreases the relative expression of genes specific to populations comprising larger fractions of the data , as is typically the case in smaller datasets , thereby making distinct populations more difficult to identify . Supplementary file 2 documents the preprocessing step used for each dataset . After first generating a random kNN adjacency matrix , the SAM algorithm goes through three steps that are repeated until convergence . To visualize the topological structure identified by SAM , we feed the final weighted PCA matrix , P^ , into UMAP ( Becht et al . , 2019 ) using Pearson correlation as the distance metric by default . To directly visualize the final kNN adjacency matrix ( Figure 1c ) , we used the Fruchterman-Reingold force-directed layout algorithm and drawing tools implemented in the Python package graph-tool ( Peixoto , 2017 ) . We used three main criteria for choosing the benchmarking scRNAseq analysis methods: they should be widely used , have done extensive benchmarking against other methods , and be mostly unsupervised . We found on Web of Science that among the highest cited scRNAseq analysis tools in 2017–2018 are Seurat , SC3 , SIMLR , Reference Component Analysis ( Li et al . , 2017 ) , Monocle ( Trapnell et al . , 2014; Qiu et al . , 2017 ) , zero-inflated factor analysis ( ZIFA , Pierson and Yau , 2015 ) , and Wishbone ( Setty et al . , 2016 ) , of which we chose Seurat , SC3 , and SIMLR . SC3 is a consensus clustering algorithm that has done rigorous benchmarking against other methods such as SINCERA ( Guo et al . , 2015 ) , SNN-Cliq ( Xu and Su , 2015 ) and pcaReduce ( Žurauskienė and Yau , 2016 ) on 12 datasets with ground truth annotation labels . SIMLR , a dimensionality reduction and clustering algorithm , evaluated its clustering performance on four annotated datasets against eight other dimensionality reduction methods , including PCA , Factor Analysis ( FA ) , t-SNE , multidimensional scaling ( MDS ) , and ZIFA . Both methods have demonstrated the highest clustering accuracy across most of the tested datasets . Additionally , as both methods have built-in functions to estimate the number of clusters present within the data , they are largely unsupervised . We also selected Seurat as one of the benchmarking methods , because it is arguably the most widely used tool for dimensionality reduction and clustering of scRNAseq data and has performed well in rigorous benchmarking studies against various methods including SC3 , SIMLR , RCA , and pcaReduce ( Duò et al . , 2019; Bahlo et al . , 2018 ) . We did not select Reference Component Analysis as it is primarily designed for cases in which an atlas of bulk , cell-type specific , reference transcriptomes is present . Additionally , we did not benchmark against Monocle and Wishbone , because they are pseudotime analysis methods and are meant for datasets with continuous branching processes such as cell differentiation . However , it is important to note that SAM can be used for dimensionality reduction upstream of pseudotime algorithms for such datasets . Finally , we did not benchmark against ZIFA as it has already been shown to have lower clustering accuracy than SIMLR . In addition to measuring clustering accuracy , we also introduce the NACC , modularity , and spatial dispersion metrics to quantify both the degree of structure and spatial organization of gene expression within a nearest-neighbor graph . These metrics can only be applied to dimensionality reduction methods that construct a graph representation of the dataset . Consequently , we cannot use these metrics to evaluate SC3 . Although it does technically produce a graph representation of the data , SIMLR should be considered as a hybrid between a clustering and dimensionality reduction method . Because its similarity graph is assumed to have a block structure where the number of blocks is equal to the prespecified number of clusters , the resulting nearest-neighbor graph will , by construction , tend to have a higher degree of structure and therefore artificially inflated NACC and modularity . Furthermore , the poor scalability of SC3 and SIMLR makes them difficult to run for many trials across a large number of datasets . Although SIMLR , in particular , does provide an alternative algorithm that can scale to run on much larger datasets , this alternative version has not been extensively benchmarked . Even so , despite the improved speed of this large-scale implementation , estimating the number of clusters using its built-in function remains a significant computational and memory bottleneck . For example , when applied to datasets with ~10 , 000 cells , neither implementations of SIMLR could estimate the number of clusters within 2 hr . As a result , we cannot run SIMLR in an unsupervised manner on datasets significantly larger than ~3000 cells . As there are few practical alternatives for manifold reconstruction that have been extensively benchmarked and widely used , we primarily compare SAM to Seurat in tests involving the unsupervised , graph-based metrics to highlight the key , advantageous characteristics of SAM as a manifold reconstruction and feature selection algorithm when applied to datasets with varying sensitivities ( Figure 4a–c ) . To generate the convergence curves in Figure 1b , we computed the root mean square error ( RMSE ) of the gene weights averaged across all pairwise comparisons of ten replicates starting from randomly generated initial graphs . In Figure 3b , we extend this analysis to all datasets analyzed and report the final error . We use randomly generated datasets of varying sizes ( ranging from 200 to 5000 cells ) as a negative control to show that SAM does not converge onto the same solution across initial conditions when the data has no intrinsic structure . These datasets were randomly generated by sampling gene expressions from a Poisson distribution with mean drawn from a gamma distribution . To generate the convergence curves in Figure 3—figure supplement 1a , we computed the RMSEs , which are ensemble-averaged across ten replicate runs , between the gene weights in adjacent iterations . We compute the adjacency error between kNN adjacency matrices Ni and Nj ( Figure 1b ) as ( 13 ) Ai , j= eT|Ni - Nj|e2eTNiewhere e is a column vector of ones . This simply measures the fraction of total edges that are different between the two graphs . To compute the standardized dispersion factors in Figure 2g , we used Seurat’s methodology implemented in Scanpy ( Wolf et al . , 2018 ) , which groups the genes into 20 bins based on their mean expression values and computes the z-score of each gene’s Fano factor with respect to the mean and standard deviation of all Fano factors in its corresponding bin . To generate the AUROC scores in Figure 2h , which quantify the likelihood of genes being cluster-specific markers , we ran SC3 on the schistosome data with the number of clusters ranging from 2 to 12 . We used the AUROC scores corresponding to four clusters for the points on the scatter plot and the standard deviations of the scores across all tested numbers of clusters for the error bars . We evaluated each analysis method on nine gold standard datasets ( Figure 3a ) using ARI , which measures the accuracy between two cluster assignments X and Y while accounting for randomness in the clustering: ( 14 ) ARI=∑nij2-∑ai2∑bj2/n212∑ai2+∑bj2-∑ai2∑bj2/n2where n is the number of cells , and nij , ai , and bj are elements from a contingency table that summarizes the overlap between the assignments X and Y ( Hubert and Arabie , 1985 ) . nij denotes the number of cells assigned to Xi that are also assigned to Yj , while ai and bj are the sums of the ith row jth column of the contingency table , respectively . To calculate the clustering accuracy for each ground truth annotation label in Supplementary file 3 , we decomposed the ARI into a vector of j elements if Y is the ground truth ( i otherwise ) by not summing up the j terms in the numerator , leaving it in vector form . Because the magnitudes of the cluster-specific scores depend on the number of cells in each cluster , a reference score was computed for each cluster using both X and Y as the true labels . Seurat was implemented using the Scanpy package in Python ( Wolf et al . , 2018 ) . For Seurat , we used both default and optimized parameters . In its default implementation , we selected the top 3000 variable genes according to their standardized dispersions and chose the number of PCs ( bounded between 6 and 50 ) which explain 30% of the variance for dimensionality reduction . From these PCs , we calculated a cell-cell correlation distance matrix . To keep the comparison between SAM and Seurat graphs consistent , this distance matrix was converted into a kNN adjacency matrix with the value of k used by SAM . We also ran a parameter sweep to optimize Seurat’s performance for each benchmarking dataset separately by changing the number of highly variable genes and principal components to maximize the clustering accuracy . To assign cluster labels for SAM and Seurat , we applied HDBSCAN ( McInnes et al . , 2017 ) , an unsupervised , density-based clustering algorithm to their respective PCA outputs . As HDBSCAN does not cluster any cell it deems an outlier , we assign the remaining outlier cells to clusters using kNN classification . For each outlier cell , we identify its 20 nearest neighbors among the clustered cells . Outliers are assigned to the same cluster as that of the majority of its neighbors . This minor extension to HDBSCAN is available as the built-in function hdbknn_clustering in SAM . SC3 was run using default parameters . The SIMLR package was implemented in R and run with the normalization parameter set to ‘True’ , which mean-centers gene expressions after normalizing them to be between 0 and 1 . Both SC3 and SIMLR provide their own functions to estimate the number of clusters and cluster assignments . To compare the quality of graphs generated by different methods , we use the NACC , modularity , and spatial dispersion . The NACC is the average of the local clustering coefficient for each node of a graph and quantifies the degree of structure in the graph ( Watts and Strogatz , 1998 ) . The local clustering coefficient is defined as ( 15 ) ai=Likiki-1where Li is the number of edges between the ki neighbors of node i and measures the degree of connectedness in a particular node’s local neighborhood . We calculate the NACC using the implementation in graph-tool ( Peixoto , 2017 ) . The modularity Q of a graph is defined as ( 16 ) Q=14m∑i , jcAij-kikj2mδijwhere Aij is one if there is an edge from cell i to cell j , ki is the degree of cell i , kj is the degree of cell j , m is the total number of edges , and δij is 1 if cells i and j are in the same cluster or 0 otherwise . High modularity indicates that clusters have on average more edges within clusters than between clusters . To find the optimal modularity for a particular graph , we used Louvain clustering , which searches for a partition with maximum modularity . To quantify the spatial organization of gene expression along the graph , we calculate the Euclidean norm of the largest 100 spatial dispersions . Spatial dispersion is defined as before in the SAM algorithm: Fi=σCi2μCi , where Fi is the Fano factor of the kNN-averaged expressions and Ci=1kNEi . N is the directed adjacency matrix output by SAM or Seurat and Ei is a column vector of expression values for gene i . To measure the inherent sensitivity of each dataset , we randomly perturbed the gene expression matrices of each dataset by randomly sampling 2000 genes and applied PCA to the subsampled data . A correlation distance matrix was calculated from the top 15 PCs and perturbations were repeated 20 times to generate distance matrix replicates . Sensitivity is then defined as the average error across all pairwise comparisons between replicates . The error between two distance matrices j and k , Sjk , is defined as the average correlation distance between corresponding pairs of rows in the distance matrices dj and dk: ( 17 ) Sjk=1n∑i=1nD{dj , i , dk , i}where D{dj , i , dk , i} is the Pearson correlation distance between the distances from cell i in distance matrices j and k . We simulated datasets with increasing sensitivity by introducing increasing degrees of corruption in each of the nine annotated datasets . To corrupt a dataset , we randomly permuted a fraction f of the elements in the expression matrix . The proportion of elements permuted corresponds to the degree of corruption , ranging from 0 to all elements . For each annotated dataset , we simulated 10 replicates per value of f . SAM and Seurat were run on each corrupted dataset , clustering was performed using the hdbknn_clustering function in SAM , and the ARI , NACC , modularity , and spatial dispersion metrics were recorded . The Area Under the Curve ( AUC ) was calculated for each metric with respect to f using the trapezoidal rule . Finally , to rescue the performance of Seurat , we used as input to Seurat the top 3000 genes with the highest SAM weights . GSEA ( Subramanian et al . , 2005 ) is typically run on a gene expression matrix with user-defined cluster assignments to quantify the differential expression for each gene . By default , differential expression is quantified using a signal-to-noise metric and the resulting scores are used to rank the genes in descending order . However , GSEA can also run in an alternative mode in which the user provides a predefined list of gene rankings . Therefore , we used the genes ranked by their SAM weights as input to GSEA to determine the biological processes enriched among the highly weighted genes . As shown in Figure 6a , we can directly test if SAM captures the relevant biological processes . GSEA provides a number of statistical measures to assess the significance of enriched gene sets , of which we use the False Discovery Rate ( FDR ) . The FDR quantifies the likelihood that a highly enriched gene set represents a false positive . The significance threshold typically used with FDR is 25% , which implies that the results are likely to be valid 75% of the time . To remove cell cycle effects from the macrophage dataset , we adopted a simpler version of the strategy used in ccRemover ( Barron and Li , 2016 ) , in which we subtract from the data PCs that are significantly associated with known cell cycle genes . Letting P represent the PCs and L be the gene loadings , we quantify the association between the set of cell cycle genes G and PC j as ( 18 ) Aj=1|G|∑i∈G|Lji| PC j is selected if its association Aj is at least two standard deviations above the mean of the associations for all PCs . In the particular case of the macrophage data , we identified the set of PCs S={P0 , P1 , P8} as being significantly associated with the cell cycle genes . We next reconstruct the data using these PCs , which thus captures the cell-cycle effects , and subtract the reconstructed data from the expression matrix E: ( 19 ) Eremoved=E−∑j∈SPjLjW When reconstructing the data , we scale the gene loadings by the SAM weights W so that only the highly weighted SAM genes ( which are initially dominated by cell cycle genes ) contribute to the cell cycle removal , as there may be other genes involved in other biological processes that could also be correlated with the PCs in S . To run SAM on the data with cell cycle effects removed , we use E as opposed to Eremoved for the calculation of spatial dispersions , because the latter may contain negative values , for which dispersion is ill-defined . This method is made available as a part of the SAM package in the functions calculate_regression_PCs and regress_genes . The original study combined imaging and transcriptomics to link NF-κB nuclear translocation dynamics to changes in gene expression within single cells . Macrophages stimulated with LPS were individually trapped in microfluidic chambers and imaged for various lengths of time ( 75–300 min ) prior to scRNAseq library preparation . NF-κB was tagged with a fluorescent protein , and its activation was measured as the nuclear-localized fluorescence intensity . Based on the imaging data , the authors identified three main classes of NF-κB dynamics , the first with a transient initial response , the second with a prolonged initial response , and the third with a recurrent response . Because the recurrent response is found only in the 300 min time point ( the latest time point in the study ) and comprises only ~8% of these cells , we primarily focused on clustering cells based on their initial dynamics . To do this , we used the tslearn ( Tavenard , 2017 ) python package to group cells based on their NF-κB activity time series . Because these time series are quite noisy , we were conservative in labeling cells as having a prolonged initial response in an effort to avoid false positives . As a result , these cells comprise only ~30% of the dataset . For the cells sampled at 75 and 150 min after LPS stimulation , we used the time series k-means algorithm with the softdtw distance metric to cluster them into three groups , which resulted in representative time series with transient , intermediate , and prolonged responses . Merging the cells with transient and intermediate responses into one cluster ( which we simply labeled as transient response ) , we obtained the 75 and 150 min representative time series shown in Figure 6e . Because the cells sampled at 300 min displayed much more variability in their NF-κB time series , we clustered them into six groups , labeling the cluster whose representative time series had the broadest initial peak as the prolonged response cluster ( blue in Figure 6e , right ) . The remaining groups were labeled as the transient response cluster ( blue in Figure 6e , left ) . We used the Mutual Nearest Neighbors algorithm ( Haghverdi et al . , 2018 ) with default values to generate an expression matrix Ecorrected in which the batch effects between the 2 . 5-week and 3 . 5-week datasets were corrected for . To run SAM on the batch-corrected data , we use E for the calculation of spatial dispersions as opposed to Ecorrected . Schistosome stem cells were isolated from juvenile parasites retrieved from infected mice at 2 . 5 and 3 . 5 weeks post infection . We followed the protocol as previously described ( Wang et al . , 2018 ) . Briefly , we retrieved juvenile parasites from schistosome-infected mice ( Swiss Webster NR-21963 ) by hepatic portal vein perfusion . Parasites were cultured at 37°C/5% CO2 in Basch Medium 169 supplemented with 1X Antibiotic-Antimycotic for 24–48 hr to allow complete digestions of host blood cell in parasite intestines . In adherence to the Animal Welfare Act and the Public Health Service Policy on Humane Care and Use of Laboratory Animals , all experiments with and care of mice were performed in accordance with protocols approved by the Institutional Animal Care and Use Committees ( IACUC ) of Stanford University ( protocol approval number 30366 ) . Before dissociation , parasites were permeabilized in PBS containing 0 . 1% Triton X-100% and 0 . 1% NP-40 for 30 s and washed thoroughly to remove the surfactants . The permeabilized parasites were dissociated in 0 . 25% trypsin for 20 min . Cell suspensions were passed through a 100 μm nylon mesh ( Falcon Cell Strainer ) and centrifuged at 150 g for 5 min . Cell pellets were gently resuspended , passed through a 30 μm nylon mesh , and stained with Vybrant DyeCycle Violet ( DCV; 5 µM , Invitrogen ) , and TOTO-3 ( 0 . 2 µM , Invitrogen ) for 30–45 min . As the stem cells comprise the only proliferative population in schistosomes , we flow-sorted cells at G2/M phase of the cell cycle on a SONY SH800 cell sorter . Dead cells were excluded based on TOTO-3 fluorescence . Single stem cells were gated using forward scattering ( FSC ) , side scattering ( SSC ) , and DCV to isolate cells with doubled DNA content compared to the rest of the population ( Wang et al . , 2018 ) . Cells that passed these gates were sorted into 384-well lysis plates containing Triton X-100 , ERCC standards , oligo-dT , dNTP , and RNase inhibitor . cDNA was reverse transcribed and amplified on 384-well plate following the Smart-Seq2 protocol ( Picelli et al . , 2013 ) . For quality control , we quantified the histone h2a ( Smp_086860 ) levels using qPCR ( the primers are listed in Supplementary file 4 ) , as h2a is a ubiquitously expressed in all schistosomes stem cell ( Collins et al . , 2013; Wang et al . , 2013; Wang et al . , 2018 ) . We picked wells that generated CT values within 2 . 5 CT around the most probable values ( ~45% of total wells , Figure 1—figure supplement 1 ) . cDNA was then diluted to 0 . 4 ng/µL for library preparation . Tagmentation and barcoding of wells were prepared using Nextera XT DNA library preparation kit . Library fragments concentration and purity were quantified by Agilent bioanalyzer and qPCR . Sequencing was performed on a NextSeq 500 using V2 150 cycles high-output kit at ~1 million reads depth per cell . Raw sequencing reads were demultiplexed and converted to fastq files using bcl2fastq . Paired-end reads were mapped to S . mansoni genome version WBPS9 ( WormBase Parasite ) using STAR . In 2 . 5 week dataset , 338 cells with more than 1700 transcripts expressed at >2 TPM were used for downstream analysis . In the 3 . 5 weeks dataset , 338 cells with more than 1350 transcripts expressed at >2 TPM were used for downstream analysis ( Figure 1—figure supplement 1 ) . RNA FISH experiments were performed as detailed in previous publications ( Collins et al . , 2013; Wang et al . , 2013; Wang et al . , 2018 ) . Clones used for riboprobe synthesis were generated as described previously , with oligonucleotide primers listed in Supplementary file 4 . Juvenile parasites were cultured with 10 µM EdU overnight , killed in 6 M MgCl2 for 30 s , and then fixed in 4% formaldehyde with 0 . 2% Triton X-100% and 1% NP-40 . Fixed parasites were sequentially dehydrated in methanol , treated in 3% H2O2 for 30 min , and rehydrated . Parasites were permeabilized by 10 μg/mL proteinase K for 15 min and post fixed with 4% formaldehyde . The hybridization was performed at 52°C with riboprobes labeled with either digoxigenin-12-UTP ( Roche ) or fluorescein-12-UTP ( Roche ) . For detection , samples were blocked with 5% horse serum and 0 . 5% of Roche Western Blocking Reagent , and then incubated with anti-digoxigenin-peroxidase ( 1:1000; Roche ) or anti-fluorescein peroxidase ( 1:1500; Roche ) overnight at 4°C for tyramide signal amplification ( TSA ) . For double FISH , the first peroxidase was quenched for 30 min in 0 . 1% sodium azide solution before the detection of the second gene . After FISH , EdU detection was performed by click reaction with 25 μM Cy5-azide conjugates ( Click Chemistry Tools ) . Samples were mounted in scale solution ( 30% glycerol , 0 . 1% Triton X-100 , 4 M urea in PBS supplemented with 2 mg/mL sodium ascorbate ) and imaged on a Zeiss LSM 800 confocal microscope . | New technologies have enabled scientists to closely examine the activity of individual cells . One increasingly popular technique to do this is called single-cell RNA-sequencing and it relies on the fact that although all cells in an organism carry the same DNA , different cell types use different genes . This technique is powerful but can struggle to identify meaningful distinctions between cell types , especially when the differences are subtle . In single-cell RNA-sequencing , the messenger RNA ( mRNA ) copied from each gene is collected and counted , and usually the more a gene is copied the more active it is . Differences in gene activity ( also called gene expression ) between two cells often imply that they are different types of cells . However , since only an infinitesimal amount of mRNAs can be collected from a single cell , the counting is often inaccurate . In addition , the transient changes in gene expression can make cells of the same type have different gene expressions . These factors make it challenging to determine what genes are informative for distinguishing between cell types . To address this problem , Tarashansky et al . have developed a computational approach called Self-Assembling Manifold ( or SAM for short ) to identify differences in gene expression that can lead to a better classification of cell types . First , SAM groups the cells randomly and looks for genes with different expression patterns between the groups . By looking at differences between groups instead of differences between individual cells , SAM is ‘averaging out’ individual differences within groups . SAM then uses the resulting information to re-classify the cells and start the process over again , taking the new groups and finding differences between them . SAM repeats these steps until the classification stops changing and becomes stable . SAM does not require any existing knowledge about cell types or gene expression , meaning it is unbiased and widely applicable . To test the usefulness of the algorithm , Tarashansky et al . used SAM to identify new cell types in the medically important parasitic worm Schistosoma mansoni , which infects hundreds of millions of people worldwide every year . SAM can tell cell types apart better than existing approaches , and it can find meaningful differences in systems with a lot of meaningless variability as demonstrated by evaluating SAM’s performance on 55 other datasets . The potential applications of this approach are many , including the creation of detailed cell atlases recording the different types of cells throughout entire organisms . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"computational",
"and",
"systems",
"biology",
"tools",
"and",
"resources"
] | 2019 | Self-assembling manifolds in single-cell RNA sequencing data |
We examined substrate-induced conformational changes in MjNhaP1 , an archaeal electroneutral Na+/H+-antiporter resembling the human antiporter NHE1 , by electron crystallography of 2D crystals in a range of physiological pH and Na+ conditions . In the absence of sodium , changes in pH had no major effect . By contrast , changes in Na+ concentration caused a marked conformational change that was largely pH-independent . Crystallographically determined , apparent dissociation constants indicated ∼10-fold stronger Na+ binding at pH 8 than at pH 4 , consistent with substrate competition for a common ion-binding site . Projection difference maps indicated helix movements by about 2 Å in the 6-helix bundle region of MjNhaP1 that is thought to contain the ion translocation site . We propose that these movements convert the antiporter from the proton-bound , outward-open state to the Na+-bound , inward-open state . Oscillation between the two states would result in rapid Na+/H+ antiport .
Na+/H+ antiporters are ubiquitous and essential secondary-active transporters found in the cell membranes of all organisms . They play crucial roles in the regulation of intracellular pH , sodium homeostasis , and cell volume . In humans , Na+/H+ antiporter dysfunction is associated with numerous serious or life-threatening diseases ( Donowitz et al . , 2013 ) , which makes them important drug targets ( Fliegel , 2009; Boedtkjer et al . , 2012; Loo et al . , 2012 ) . Na+/H+ antiporters belong to the superfamily of cation/proton antiporters ( CPA ) , which include the CPA1 and CPA2 subfamilies as main branches ( Brett et al . , 2005 ) . CPA1 transporters are electroneutral and most likely exchange protons and Na+ ions with a 1:1 stoichiometry , whereas CPA2 transporters , which mostly exchange two protons per Na+ , are electrogenic . The best-known member of the CPA2 subfamily is the Na+/H+ antiporter NhaA from E . coli . EcNhaA enables E . coli to survive at high salinity or alkaline pH ( Padan and Schuldiner , 1994 ) , making use of the proton-motive force to extrude sodium from the cell ( Figure 1 ) . Other members of the CPA2 subfamily include the plant CHX transporters and the mammalian NHA transporters . Well-known representatives of the CPA1 subfamily include the medically important mammalian NHE exchangers and the archaeal NhaP antiporters ( Brett et al . , 2005 ) . 10 . 7554/eLife . 01412 . 003Figure 1 . Physiological roles of EcNhaA and MjNhaP1 . ( A ) EcNhaA is a sodium pump driven by the proton-motive force , exchanging one sodium ion against two protons ( Taglicht et al . , 1991 ) , enabling E . coli to survive at high sodium and alkaline pH . ( B ) MjNhaP1 is thought to act mainly as a proton pump driven by the sodium gradient , exchanging one proton against one sodium ion . Like the homologous NHE1 in mammals ( Lee et al . , 2012 ) , MjNhaP1 plays a critical role in pH homeostasis and enables M . jannaschii to survive at low pH ( 4–6 ) and sodium concentrations up to 0 . 9 M ( Jones et al . , 1983; Hellmer et al . , 2002 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01412 . 003 The activity of Na+/H+ antiporters is highly dependent on the concentration of their substrate ions , H+ and Na+ . For EcNhaA , a maximum transport rate of 105 min−1 at pH 8 . 5 has been reported , which drops by two or three orders of magnitude at pH 6 . 5 ( Taglicht et al . , 1991; Rimon et al . , 1998; Padan , 2008 ) . This pH dependence has been attributed to a putative cytoplasmic pH sensor , which transfers the antiporter into an acidic-locked conformation at low pH ( Taglicht et al . , 1991; Padan , 2008 ) . A recent study of ΔpH and ΔNa+-driven transport indicated a symmetrical bell-shaped pH dependence of EcNhaA ( Mager et al . , 2011 ) . Transport activity was maximal at pH 8 . 5 , and more than doubled as the Na+ concentration increased from 10 mM to 100 mM . These observations are consistent with a simple kinetic model whereby H+ and Na+ compete for the same substrate binding site ( Mager et al . , 2011 ) , thought to include two conserved aspartates in TMH V of EcNhaA . According to this model , inactivation at low pH occurs because Na+ ions cannot compete effectively against protons for the binding site . Inactivation at high pH is simply due to the depletion of H+ substrate ions at the binding site , as the proton concentration becomes too low to drive transport . The archaeal CPA1 antiporters share significant sequence homology of functionally important regions with the mammalian NHE antiporters ( Goswami et al . , 2011 ) . A study of NHE1 in mammalian cells has shown a strong dependence of transport activity on both H+ and Na+ concentrations , with activity dropping to background level at pH 8 , and a strong dependence on extracellular Na+ ( Fuster et al . , 2008 ) . Like NHE1 , but unlike EcNhaA , the Na+/H+ antiporter NhaP1 from Methanocaldoccocus jannaschii ( MjNhaP1 ) is thought to use a sodium gradient to extrude protons from the cell ( Figure 1 ) ( Thauer et al . , 2008; Lee et al . , 2012 ) . Like NHE1 ( Fuster et al . , 2008 ) , but again unlike EcNhaA , MjNhaP1 is active at pH 6 , and down-regulated at pH 7 . 5 or above ( Hellmer et al . , 2002; Goswami et al . , 2011 ) . The first insight into the structure of a Na+/H+ came from the 3D map of the EcNhaA dimer in the membrane at 7 Å resolution , obtained by electron crystallography of 2D crystals ( Williams , 2000 ) . The map revealed 12 transmembrane α-helices ( TMHs ) per protomer , referred to as TMH I-XII . The TMHs of EcNhaA were arranged in two groups: a 6-helix bundle at the tip of each protomer , and a row of 6 helices at the dimer interface . The 3 . 45 Å X-ray structure of monomeric EcNhaA in an inward-open conformation provided further details , including a pair of discontinuous helices ( TMH IV and XI ) in the 6-helix bundle that were proposed to harbor the ion translocation site ( Hunte et al . , 2005 ) . By electron crystallography we obtained a projection structure ( Vinothkumar et al . , 2005 ) and , more recently , a 3D map of the MjNhaP1 dimer in the membrane at 7 Å resolution ( Goswami et al . , 2011 ) that indicated 13 TMHs , referred to as TMHs 1-13 . While the dimer interface looked very different to that of EcNhaA , the structure of the 6-helix bundle was similar , supporting the notion of related transport mechanisms . Two-dimensional ( 2D ) crystals are ideal for investigating conformational changes of membrane proteins in a native-like lipid environment ( Subramaniam et al . , 1993; Beroukhim and Unwin , 1995 ) . By this approach we discovered substrate-induced changes in MjNhaP1 ( Vinothkumar et al . , 2005 ) and in EcNhaA ( Appel et al . , 2009 ) . In both cases , the changes occurred in the 6-helix bundle . Here , we present a detailed study of in situ conformational changes in 2D crystals of MjNhaP1 in response to pH and Na+ . To separate the effects of H+ and Na+ , 2D crystals were grown without NaCl and examined by electron crystallography in a wide range of carefully controlled pH and ionic conditions . Our results provide new insights into the molecular mechanisms of activation and substrate binding in the CPA1 antiporters .
For electron crystallography , MjNhaP1 was purified and crystallized under Na+-free conditions . The 2D crystals looked similar to those obtained earlier ( Vinothkumar et al . , 2005; Goswami et al . , 2011 ) but were more highly ordered , diffracting up to 6 Å resolution . In total , 29 data sets of MjNhaP1 at different pH and salt concentrations were collected , processed and analyzed ( Supplementary file 1 ) . Peaks in the projection maps characteristic of membrane-spanning helices at 6 Å resolution were observed in the two protomers of the crystallographic MjNhaP1 dimer ( Figure 2 ) . The 7 Å 3D EM model of MjNhaP1 ( Goswami et al . , 2011 ) enabled us to assign peaks in the projection maps to the TMHs and to the cytoplasmic ( c ) or exoplasmic ( e ) halves of the discontinuous TMHs 5 ( 5c/5e ) and 12 ( 12c/12e ) in the protomer ( Figure 2 ) . TMH 1-3 and 7-10 form the dimer interface . Even though tilted or discontinuous TMHs overlap to some extent in projection , the assignment of helices 5e , 6 and 13 to well-defined peaks in the projection of the 6-helix bundle was unambiguous . Helix 5c overlaps with helix 12e in one prominent peak in the center of the MjNhaP1 protomer , whereas helix 12c coincides with the projected densities of TMH 4 and 11 ( Figure 2 ) . 10 . 7554/eLife . 01412 . 004Figure 2 . Helix assignment . 6 Å projection map of the MjNhaP1 dimer in two-dimensional membrane crystals at pH 4 with 500 mM NaCl . ( A ) Superposition of one protomer of the MjNhaP1 3D EM model ( Goswami et al . , 2011 ) . The model is seen from the extracellular space . The 6-helix bundle is colour-coded , while the 7 TMH at the dimer interface are neutral . ( B ) TMHs are shown as ovals and numbered as in the MjNhaP1 model . Identities of TMH 1 and 7 in the model were ambiguous . DOI: http://dx . doi . org/10 . 7554/eLife . 01412 . 004 Conformational changes in the MjNhaP1 dimer were visualized by projection difference maps , calculated from amplitudes and phases collected under defined pH and salt conditions . To observe pure pH-induced conformational changes , 2D crystals grown without Na+ were incubated on the EM grid with sodium-free buffers at pH 4 , 6 or 8 ( Figure 3; Supplementary file 1 ) . The resulting projection maps indicated only minor differences in the shape or position of density peaks . A small positive peak close to TMH 10 was observed upon a shift from pH 4 or pH 8 to pH 6 ( Figure 3A , B ) and a similarly-sized negative difference peak close to TMH 6 occurred in the transition from pH 8 to pH 6 . Differences between pH 4 and pH 8 without Na+ were at background level ( Figure 3C ) . These results indicate that a change in pH alone had no major effect on the conformation of MjNhaP1 . 10 . 7554/eLife . 01412 . 005Figure 3 . pH-induced conformational changes in absence of sodium . Difference maps at 6 Å resolution between projection maps of MjNhaP1 2D crystals grown without sodium at different pH: ( A ) pH 4 minus pH 6; ( B ) pH 8 minus pH 6; and ( C ) pH 4 minus pH 8 . ( D ) Helix assignment as in Figure 2B . Blue contours indicate positive differences , negative differences are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 01412 . 005 To examine pure Na+-induced conformational changes , 2D crystals of MjNhaP1 grown without Na+ were incubated with 20–500 mM NaCl at constant pH . Addition of Na+ at pH 8 caused evident changes in the 6-helix bundle ( Figure 4A ) , giving rise to major peaks in the difference maps ( Figures 4C and 5A ) . The unit cell changed gradually from 81 . 2 × 104 . 2 Å to 80 . 6 × 107 . 9 Å at 500 mM Na+ , whereas a pH shift in the absence of salt had no significant effect on the unit cell dimensions ( Supplementary file 1 ) . 10 . 7554/eLife . 01412 . 006Figure 4 . Sodium-induced conformational changes at pH 8 . ( A ) 6 Å projection maps of MjNhaP1 crystals at pH 8 without sodium ( left ) and at increasing sodium concentration . ( B ) Helix assignment as in Figure 2B . ( C ) Difference map at 6 Å resolution between projection maps of MjNhaP1 at pH 8 in absence of sodium and at 500 mM NaCl . Blue contours indicate positive differences , negative differences are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 01412 . 00610 . 7554/eLife . 01412 . 007Figure 5 . Sodium-induced conformational changes at different pH . Matrix of 6 Å difference maps between projections of MjNhaP1 2D crystals in absence of sodium and the sodium concentrations indicated at pH 8 ( A ) , pH 6 ( B ) and pH 4 ( C ) . Clusters of predominant positive/negative difference peaks are outlined by ovals , single positive difference peaks are circled . The arrow indicates a broad difference peak observed at pH 4 . Helices are assigned to regions of the projection maps at the end of each row . The first column shows control difference maps calculated from datasets split randomly into two halves to assess the level of background noise , which was estimated at around ±2 contour lines . Blue contours indicate positive differences , negative differences are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 01412 . 00710 . 7554/eLife . 01412 . 008Figure 5—figure supplement 1 . Substrate specificity . Difference maps of MjNhaP1 monomer at 6 Å resolution in response to ions other than Na+ . Respective pH and salt conditions are indicated . ( C ) Mg2+ ions have no major impact on the structure of MjNhaP1 . ( A and B ) K+ ions , which may bind ( Alhadeff et al . , 2011 ) but are not transported cause changes above background level which are different from those observed with the substrate ions Na+ ( Figure 5 ) and Li+ ( D ) . ( E ) Helix assignment as in Figure 2B . Blue contours indicate positive differences , negative differences are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 01412 . 00810 . 7554/eLife . 01412 . 009Figure 5—figure supplement 2 . Difference maps calculated for TtNapA . ( A ) The similarity of projection maps at 6 Å resolution indicates that TtNapA is a structural homologue of MjNhaP1 . On the left is the outward-open TtNapA X-ray structure ( PDB ID 4BWZ ) . The inward-open TtNapA model , which was built as described by Lee et al . 2013 , is shown on the right . ( B ) Difference map at 6 Å resolution between the two TtNapA projections in ( A ) . ( C ) Difference map at 6 Å resolution between projection maps of MjNhaP1 at pH 8 in absence of sodium ( outward-open ) and at 500 mM NaCl ( inward-open ) . Blue contours indicate positive differences , negative differences are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 01412 . 00910 . 7554/eLife . 01412 . 010Figure 5—figure supplement 3 . Comparison to the difference map obtained by Vinothkumar et al . ( 2005 ) . ( A ) Difference map of MjNhaP1 monomer at 6 Å resolution at pH4 + 150 mM NaCl minus projection map of crystals grown with 200 mM sodium and washed on the grid with pH 8 buffer without sodium , reproducing the experimental conditions in Vinothkumar et al . ( 2005 ) as closely as possible . ( B ) Difference map reproduced from Figure 7 in Vinothkumar et al . ( 2005 ) ( published in EMBO Journal , copyright original authors ) . ( C ) Helix assignment as in Figure 2B . Red contours indicate positive differences , while negative differences are shown in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 01412 . 010 All difference maps at pH 8 showed the same three sets of positive/negative difference peaks , outlined by ovals in Figure 5A . Difference peaks within each set became progressively stronger as the NaCl concentration increased from 20 mM to 500 mM . At 20 mM , there were 7–9 contour levels between the lowest negative and highest positive peaks in each set , rising to a peak-to-trough difference of 15 contour levels at 500 mM NaCl ( Figure 5A ) . A positive peak that was not matched by a corresponding negative peak appeared at 100 mM NaCl and above ( Figure 5A ) , indicating a gain of order in this protein region . Control experiments gave similar results for Na+ and Li+ , which are both substrate ions of MjNhaP1 , whereas Mg2+ , which is not transported , did not give rise to significant difference peaks under otherwise similar conditions ( Figure 5—figure supplement 1 ) . K+ , which is likewise not transported but has been reported to bind to EcNhaA ( Alhadeff et al . , 2011 ) , caused changes above background level that were however different from those observed with Na+ or Li+ ( Figure 5—figure supplement 1 ) . The strong changes observed with Na+ or Li+ are thus substrate-specific . At pH 4 , essentially the same pattern of difference peaks and changes in unit cell dimensions were found as at pH 8 ( Supplementary file 1 and Figure 5C ) . However , these differences occurred at NaCl concentrations that were roughly 10-fold higher than at pH 8 ( Figures 5 and 6 ) . Difference maps at 20 or 100 mM NaCl at pH 4 were largely featureless , except for one comparatively weak pair ( 6 or 7 contour levels peak-to-trough ) at the position of the strongest peaks at higher NaCl concentration ( Figure 5C ) . In addition , at NaCl concentrations of 150 mM or above , there was a broad difference peak to the left of the 6-helix bundle that was not observed at pH 8 ( arrow in Figure 5C ) . 10 . 7554/eLife . 01412 . 011Figure 6 . pH dependence of apparent dissociation constants for Na+ in MjNhaP1 . The number of contour lines ( peak-to-trough ) was counted for each set of positive/negative difference peaks in the difference maps shown in Figure 5 including maps generated for 50 mM NaCl at pH 8 and 150 mM NaCl at pH4 ( not shown , see Supplementary file 1 ) and plotted against the Na+ concentration at pH 8 ( A ) and pH 4 ( B ) . Apparent dissociation constants for Na+ were calculated ( C ) . Contour levels observed in the same regions of control difference maps in absence of sodium ( first column ) were set to zero and subtracted from all subsequent values . DOI: http://dx . doi . org/10 . 7554/eLife . 01412 . 011 At pH 6 , again the same sets of difference peaks were evident , but only at elevated NaCl concentrations of 250 mM or above ( Figure 5B ) . Differences were much weaker ( maximally 7–8 contour levels peak-to-trough ) and more diffuse than at pH 8 or at pH 4 ( Figure 5 ) , even up to 1 M NaCl ( not shown ) . Unit cell parameters at pH 6 did not change significantly ( Supplementary file 1 ) . To estimate the sodium affinity of MjNhaP1 under different pH regimes , peak heights for each of the three sets of positive/negative difference pairs shown in Figure 5 were plotted as a function of Na+ concentration ( Figure 6 ) . Analysis of these titration curves yielded apparent dissociation constants for Na+ ( KDNa+ , app ) ( Figure 6C ) . At pH 8 , where conformational changes were evident already at comparatively low NaCl concentrations , the three sets of difference pairs gave almost identical apparent KDNa+ values of around 30 mM . At pH 4 , the apparent KDNa+ of ∼280 mM indicated roughly 10-fold weaker binding than at pH 8 . Strong pairs of positive/negative peaks in projection difference maps at 6–8 Å resolution are indicative of helix movements , which can be either lateral displacement or helix tilts ( Subramaniam et al . , 1993; Appel et al . , 2009 ) . The three sets of strong difference peaks in Figure 5 all map to the 6-helix bundle of MjNhaP1 and indicate significant and specific rearrangements of TMHs in this part of the protein . The changes take place in the physiological pH and Na+ concentration range for M . jannaschii ( Jones et al . , 1983 ) and thus are likely to reflect the conformational changes in the transport cycle . Our model of MjNhaP1 ( Goswami et al . , 2011 ) superposed on the projection and difference maps enables us to assign these changes to individual TMHs , in favourable cases even to the side of the membrane on which they occur . Of the three sets of difference peaks in Figure 5 that are common to all three pH regimes , the strongest pair ( set 1 ) coincides with the projection of the cytoplasmic half of TMH 5 and the extracellular half of TMH 12 ( helix segments 5c/12e ) , indicating a joint movement of these two half helices . Inspection of the corresponding peaks in the projection map ( Figure 4 ) and the 3D model indicates a displacement , most likely a lateral movement , of the 5c/12e helix pair by ∼2 Å towards TMH 10 and 13 . The second-strongest pair of positive/negative difference peaks ( set 2 in Figure 5 ) maps to the projection of TMH 6 . In Figure 4 , the peak corresponding to TMH 6 is elongated in the absence of Na+ and becomes round at elevated NaCl concentration , indicating that this helix reorients in response to Na+ binding . In the 3D model the extracellular half of TMH 6 is tilted , so that the entire TMH 6 becomes more straight and perpendicular to the membrane in response to Na+ binding . The third set of difference peaks in Figure 5 ( set 3 ) coincides with the map region occupied by TMH 13 and helix 5e . This set consists of a central positive peak , between two smaller negative peaks . In principle , the positive peak could form a pair with either ( or both ) of the negative peaks , thus indicating a movement of the corresponding helices . Inspection of the projection map ( Figure 4 ) helps to resolve this ambiguity . The projection peak corresponding to TMH 13 is more elongated at 500 mM NaCl than in absence of sodium , indicating that this helix becomes more tilted with increasing Na+ concentration . A peak for helix 5e is not evident at low Na+ , but a clear peak is present at 100 mM NaCl and above in the projection maps . Therefore helix 5e is either highly tilted or disordered at low Na+ , which would both make it difficult to see in projection . Upon Na+ binding , it either reorients or becomes ordered and then runs more or less perpendicular to the membrane plane . The broad difference peak to the left of the 6-helix bundle in Figure 5C that appeared at pH 4 above 150 mM NaCl indicates a Na+-induced movement of the group of helices on the outside of the bundle , consisting of TMH 4 , 11 and helix 12c . Again , inspection of the corresponding projection maps indicates that this part of the bundle shifts by ∼2 Å towards the dimer interface . This movement is largely absent at pH 8 or 6 . The positive difference peak above 100 mM NaCl at pH 8 or 250 mM at pH 4 ( Figure 5A , C ) indicates a gain of order in the region between the 6-helix bundle and of TMH 10 , which is part of the dimer interface . Otherwise , no significant differences were recorded in the central part of the dimer . Figure 7 summarizes the observed helix movements in the context of the MjNhaP1 model ( Goswami et al . , 2011 ) . 10 . 7554/eLife . 01412 . 012Figure 7 . Na+ -induced conformational changes in MjNhaP1 . ( A ) Summary of observed helix movements in response to Na+ binding . Schematic helix positions in absence of NaCl ( left ) or at pH 4 and pH 8 at >250 mM NaCl ( right ) . Helix movements are indicated by arrows . Helix projections are shown as circles or ovals . The 6-helix bundle is color-coded as in Figure 2A . Helices at the dimer interface are grey . ( B ) Model drawing of changes in the positions of TMH 5 , 6 and 12 that respond most strongly to Na+ binding . Residue D161 in TMH 6 , thought to be directly involved in substrate binding , is shown in black . Na+ binding results in a transition from the apo or proton-bound state , where the putative ion-binding site is likely to be more accessible form the extracellular space ( left ) , to a Na+-bound state , which we propose to be open to the cytoplasm ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01412 . 012
The E . coli antiporter NhaA shares many features with MjNhaP1 . Both form dimers of similar shape and size in the membrane , and have the conserved 6-helix bundle thought to harbor the ion translocation site . In EcNhaA and other CPA2 antiporters , the ion-binding and transport site has been proposed to include two conserved aspartates ( Inoue et al . , 1995; Hunte et al . , 2005; Maes et al . , 2012 ) . In the CPA1 antiporters , one of the aspartates is replaced by an asparagine , which likewise is thought to participate in ion-binding and translocation ( Hellmer et al . , 2003; Goswami et al . , 2011 ) . In the EcNhaA X-ray structure ( Hunte et al . , 2005 ) the two conserved aspartates are located at the center of TMH V , which means that the corresponding residues in MjNhaP1 are in TMH 6 . Our difference maps show that this helix participates prominently in Na+-induced changes . We also found changes in helix 5e of MjNhaP1 , which corresponds to the periplasmic half of TMH IV of EcNhaA . Electron crystallography of 2D crystals , side-directed tryptophan fluorescence , ion accessibility studies and MD simulations have all found conformational changes or an intrinsically high degree of flexibility in this half helix ( Appel et al . , 2009; Kozachkov and Padan , 2011; Rimon et al . , 2012 ) , and we find that the corresponding region in MjNhaP1 also moves or rearranges upon Na+ binding . The events known to occur upon ion-binding and translocation in EcNhaA are thus consistent with the conformational changes we observe in MjNhaP1 . Recently , the 3 Å X-ray structure of the Na+/H+ antiporter NapA from T . thermophilus has been reported ( Lee et al . , 2013 ) . TtNapA is electrogenic , and thus functionally similar to EcNhaA . However , in terms of protein structure , it resembles MjNhaP1 much more closely , as judged from a 6 Å projection map we calculated from the TtNapA coordinates ( Figure 5—figure supplement 2A and Figure 2 ) . This refers in particular to the position of the 6-helix bundle relative to the dimer interface and the number of TMHs ( 13 in TtNapA and MjNhaP1 , only 12 in EcNhaA ) . Nevertheless , Lee et al . assume that the structures of TtNapA and EcNhaA show the outward-open and inward-open conformation of essentially the same antiporter . A model of TtNapA in the inward-open conformation based on the EcNhaA structure was generated , and compared it to the outward-open TtNapA structure . Not surprisingly , there were major differences between the inward-open model and the outward-open structure , in particular with respect to the position and orientation of the 6-helix bundle , which appeared to rotate by 20° within the membrane . On the basis of this comparison , they proposed a transport mechanism , which was claimed to apply to all Na+/H+ antiporters . However , such a massive domain movement is difficult to reconcile with the high turnover rates of EcNhaA and the comparatively subtle substrate-induced changes in and around the 6-helix bundle we observe experimentally in MjNhaP1 . We built an inward-open TtNapA model based on the EcNhaA structure , as described by Lee et al . ( 2013 ) , and calculated a projection difference map between it and the outward-open TtNapA structure ( Figure 5—figure supplement 2 ) . The very large and strong difference peaks in the region of the 6-helix bundle bear no resemblance to Figures 4 and 5 . We conclude that a 20º in-plane rotation of the 6-helix bundle does not occur in MjNhaP1 . It most likely also does not occur in EcNhaA , since Figure 5—figure supplement 2 also bears no resemblance to projection difference maps obtained under a range of physiological ion and pH conditions with this antiporter ( Appel et al . , 2009 ) . The projection and difference maps in Figures 3–5 indicate two different conformations of MjNhaP1: one in the absence of Na+ , which looks similar at any pH between 4 and 8 , and one in the presence of Na+ above 20 mM at pH 8 or 200 mM at pH 4 . The latter shows the Na+-bound state , whereas the former shows the apo or proton-bound state of the antiporter . At elevated Na+ concentrations , the pH 6 maps resemble those at pH 4 or 8 , but the difference peaks are less distinct and only roughly half as strong , as would be expected if the corresponding helices were disordered . We ascribe this apparent disorder to the fact that the antiporter is fully active at pH 6 , and the helices involved in ion translocation would oscillate between the sodium-free and sodium-bound state . Hence , when the 2D crystals are frozen in liquid nitrogen , the helices are trapped in a continuum of slightly different orientations . In a projection map , which takes the average of all molecules on the 2D crystal lattice , the helices would thus appear disordered , as observed . At pH 4 and 8 MjNhaP1 is down-regulated ( Vinothkumar et al . , 2005 ) and presumably more or less rigid . Under these conditions , the effect of Na+ on the structure is progressive , indicating that the maps are averages of an increasing number of molecules on the 2D crystal lattice in the Na+-bound state and a decreasing number in the apo state , as the Na+ concentration increases . The height of the difference peaks is then a direct measure of Na+ affinity , and enables us to determine apparent binding constants for Na+ in different pH regimes ( Figure 6 ) . We ( Goswami et al . , 2011 ) and others ( Hellmer et al . , 2002 ) have shown that MjNhaP1 is down-regulated at pH 8 . Down-regulation at acidic pH has not yet been shown experimentally ( largely because there is no suitable pH-sensitive fluorescent dye ) , but follows from physiological considerations , which indicate that the antiporter must be inactive at pH 4 to prevent uncontrolled influx of Na+ ions that would otherwise result from a large outward pH gradient ( Vinothkumar et al . , 2005 ) . A shutdown at acidic and basic conditions is a key feature of the simple kinetic model proposed by Mager et al . for EcNhaA . This model thus also appears to hold for MjNhaP1 . At pH 4 , this antiporter may be down-regulated due to competition of H+ and Na+ for a common binding site , and at pH 8 due to proton depletion . Together with the activity peak around pH 6 , these two effects would result in a bell-shaped pH dependence of transport activity . The observed 10-fold increase in apparent KDNa+ from pH 8 to pH 4 is consistent with a competition of Na+ and H+ for a common binding site , because more Na+ ions are needed to displace the protons as the H+ concentration rises . However , if Na+ can displace H+ and can arrest the antiporter in a Na+-bound state at pH 4 , the question arises why this does not also happen at pH 6 , where the antiporter is fully active . Apparently , in addition to substrate competition , the transport rate of MjNhaP1 is also modulated in some other way , for example by the protonation states of amino acid sidechains involved in ion-binding and translocation that render the antiporter more flexible at pH 6 than at pH 4 or pH 8 . In EcNhaA and NHE1 , a pH sensor is thought to transform the antiporters from an active into an acid-locked state at low pH ( Aronson et al . , 1982; Aronson , 1985; Taglicht et al . , 1991; Wakabayashi et al . , 2003; Padan et al . , 2009 ) . In the case of EcNhaA , this has been reported to involve a conformational switch at the level of secondary structure ( Herz et al . , 2010; Diab et al . , 2011; Schushan et al . , 2012 ) . In MjNhaP1 , we see no evidence of a pH-triggered conformational switch ( Figure 3 ) , so that down-regulation at low pH is not associated with a significant change in secondary structure . However , it may involve the re-orientation of amino acid sidechains , which would not be visible at 6 Å resolution . In an earlier study , Vinothkumar et al . investigated pH-induced changes in 2D crystals of MjNhaP1 ( Vinothkumar et al . , 2005 ) . The crystals had been grown in the presence of NaCl , and therefore incubating them in salt-free buffers changed both the pH and the Na+ concentration simultaneously . The effect observed by Vinothkumar et al . was thus due to both Na+ and pH , rather than to pH only . We were able to reproduce these differences with 2D crystals grown with or without NaCl , as indicated in Figure 5—figure supplement 3 .
By growing 2D crystals of the archaeal Na+/H+ antiporter MjNhaP1 in absence of sodium , we were able to separate effects of the two substrate ions , Na+ and H+ , on its conformation . Projection difference maps at 6 Å resolution show that pH in the absence of Na+ has no major effect on the structure of the antiporter . This contrasts with current models of Na+/H+ antiporter regulation , which postulate a significant pH-triggered conformational switch . If such a pH switch occurs in MjNhaP1 , it does not affect the secondary structure but may involve sidechain movements , which are not resolved by our method . On the other hand , Na+ ions cause a marked conformational change that is largely pH-independent . At pH 8 and 4 , where MjNhaP1 is down-regulated , the Na+-induced differences reflect a progressive change in the relative population of the two distinct conformational sodium-free and sodium-bound states . This enabled us to deduce apparent binding constants of MjNhaP1 for Na+ under acidic and basic conditions . At pH 6 , where this antiporter is fully active , the helices involved in ion translocation appear disordered , due to averaging over many slightly different conformations on the crystal lattice . The MjNhaP1 model ( Goswami et al . , 2011 ) allows us to conclude that the helix movements deduced from the projection difference maps in Figures 4 and 5 change the access to the proposed substrate binding site , as shown schematically in Figure 7 . In the apo or proton-bound state , the ion-binding site near the center of TMH 6 would be accessible from the extracellular side , making this the outward-open state . The helix movements brought about by Na+ ions then convert the antiporter into the Na+-bound , inward-open state . Oscillation between the two states would result in Na+/H+ antiport , as specified by the alternating access mechanism ( Jardetzky , 1966 ) . It is likely that the mammalian NHE antiporters , which share many similarities with the archaeal Na+/H+ exchangers , work in the same way .
MjNhaP1-His was expressed in the pET26b vector and purified by Ni-NTA affinity chromatography as described previously ( Goswami et al . , 2011 ) , with the following modifications: The Ni-NTA column was washed with 10 cv ( column volumes ) of buffer 1 ( 15 mM Tris/HCl pH 7 . 5 , 500 mM NaCl , 15 mM imidazole and 0 . 03% DDM ) , followed by 8 cv of sodium-free buffer 2 ( 15 mM Tris/HCl pH 7 . 5 , 200 mM KCl and 0 . 03% DDM ) . MjNhaP1 was eluted with 50 mM K+ acetate pH 4 , 100 mM KCl , 5 mM MgCl2 and 0 . 03% DDM , concentrated and stored at −80°C . Two-dimensional crystallization of MjNhaP1 with E . coli polar lipids ( Avanti Polar lipids ) was carried out at a final protein concentration of 1 mg/ml , and a final DM concentration of 0 . 15% . The mixture was incubated for 1 . 5 h at room temperature and transferred to a dialysis bag with a 14 kDa cutoff . Dialysis was performed at 37°C over 7–10 days , in 25 mM K+ acetate pH 4 , 200 mM KCl , 5% glycerol and 5% 2-4-methylpentanediol ( MPD ) . Crystals were obtained at a lipid-to-protein ( LPR ) range of 0 . 3–0 . 7 and were stable for several months . Samples were prepared by the back-injection method ( Wang and Kuhlbrandt , 1991 ) in 4% trehalose . The composition of the embedding medium used varied , accordingly . In total , a combination of three different buffers ( 25 mM K+ acetate pH 4 , 50 mM MES pH 6 and 50 mM Tris/HCl pH 8 ) and four different salts ( NaCl , LiCl , KCl , and MgCl2 ) at different salt concentrations were used ( Supplementary file 1 ) . 1 . 5 μl sample was applied to the carbon film , mixed thoroughly with excess of embedding buffer and incubated for 1 min . The grids were blotted and rapidly frozen in liquid nitrogen . Images were recorded on Kodak SO-163 film with a JEOL 3000 SFF electron microscope at a nominal temperature of 4 K , an acceleration voltage of 300 kV , a magnification of x 53 , 000 in spot scan mode at 0 . 2–1 μm defocus . Crystal quality was evaluated by optical diffraction . Well-ordered areas of 4000 × 4000 or 6000 × 6000 pixels were digitized at 7 μm step size on a Zeiss SCAI scanner . Images were processed using the 2dx software ( Gipson et al . , 2007 ) and data quality was improved by synthetic unbending ( Arheit et al . , 2012 ) . Projection maps were calculated from at least six lattices and were of similar quality to 6 Å resolution ( Figure 4 , Supplementary file 1 ) . For calculation of difference maps scripts from the CCP4 program suite package were manually modified and used at each step as indicated in brackets . Projection phases and amplitudes for each data set were calculated ( 2dx software ) , scaled ( sftools ) and subtracted ( overlapmap ) . Difference maps were plotted to reveal pH- or salt-induced conformational changes . To account for small differences in unit cell length , maps were excised ( fft , mapmask , maprot , and npo ) and placed into the same cell before subtraction . For an estimate of background noise level , image data collected under identical conditions were randomly divided into two sets ( Figure 5 first column ) , or image data from different conditions were mixed and randomly merged into two separate data sets . These control difference maps were plotted with a stepsize of 1 . 5 σ between contour levels , indicating a background level of ±2 contour levels , or ∼3 σ . The background level was set to be the same in all calculated difference maps by applying the same absolute plotting parameters as used for the control difference maps ( Figures 3–5 ) . | Antiporters are proteins that move molecules or charged particles , such as sodium ions and protons , in and out of cells . Antiporters therefore have an important role in controlling the conditions inside a cell , such as the pH ( which is a measure of acidity ) and sodium content ( which is a measure of saltiness ) . In human cells , defects in sodium/proton antiport result in heart or kidney failure and other serious diseases . Some aspects of sodium/proton antiporters are well understood , such as the levels of saltiness and acidity that trigger the flow of charged particles in and out of bacterial cells , but the details of how sodium ions or protons activate an antiporter are unknown . Now Paulino and Kühlbrandt have used a technique called electron crystallography to study how the structure of the sodium/proton antiporter changes as the acidity or salt conditions sensed by the antiporter vary . When the concentration of sodium ions was increased at acidic conditions ( low pH ) , the structure of the antiporter began to change so as to increase the ion flow . However , no such changes were observed when the concentration of sodium ions was held constant at a low level while the pH was increased . These findings suggest that , contrary to previous thinking , the operation of a sodium/proton antiporter is largely determined by the concentration of sodium ions . A better understanding of the operation of sodium/proton antiporters should help with the design new treatments for faulty antiporters . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Conclusions",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | pH- and sodium-induced changes in a sodium/proton antiporter |
Virus-host interactions drive a remarkable diversity of immune responses and countermeasures . We found that two RNA viruses with broad host ranges , vesicular stomatitis virus ( VSV ) and Sindbis virus ( SINV ) , are completely restricted in their replication after entry into Lepidopteran cells . This restriction is overcome when cells are co-infected with vaccinia virus ( VACV ) , a vertebrate DNA virus . Using RNAi screening , we show that Lepidopteran RNAi , Nuclear Factor-κB , and ubiquitin-proteasome pathways restrict RNA virus infection . Surprisingly , a highly conserved , uncharacterized VACV protein , A51R , can partially overcome this virus restriction . We show that A51R is also critical for VACV replication in vertebrate cells and for pathogenesis in mice . Interestingly , A51R colocalizes with , and stabilizes , host microtubules and also associates with ubiquitin . We show that A51R promotes viral protein stability , possibly by preventing ubiquitin-dependent targeting of viral proteins for destruction . Importantly , our studies reveal exciting new opportunities to study virus-host interactions in experimentally-tractable Lepidopteran systems .
Viruses represent a constantly evolving challenge to the fitness and survival of their cellular hosts . Thus , not surprisingly , investigations into virus-host interactions have produced important and fundamental new insights into both cellular and pathophysiology ( Panda and Cherry , 2012 ) . Invertebrate model organisms have proven useful in elucidating a wide range of host responses to infection and because many of these responses are well conserved , studies in model organisms are often directly relevant to human health ( Moser et al . , 2010; Panda and Cherry , 2012; Moy et al . , 2014 ) . Notably , studies of invertebrate antiviral RNA interference ( RNAi ) pathways ( Fire et al . , 1998; Zhou and Rana , 2013 ) have produced powerful tools for probing and manipulating gene function , with potential utility for direct therapeutic intervention ( Blake et al . , 2012 ) . Relatively small genomes , well-defined genetics , and efficient RNAi pathways make insect models attractive systems in which to study virus-host interplay ( Moser et al . , 2010; Cherry , 2011 ) . Dipteran organisms , such as Drosophila melanogaster , have been the primary focus of virus-host studies in invertebrates and Drosophila RNAi screens have greatly enhanced our understanding of how eukaryotic host factors can promote or inhibit virus replication ( Cherry , 2011 ) . These studies have almost exclusively focused on RNA viruses ( Xu and Cherry , 2014 ) . In contrast to Dipterans , most virus-host studies in the order Lepidoptera ( moths and butterflies ) have focused on DNA viruses , particularly baculoviruses ( Ikeda et al . , 2013 ) . These studies have provided key insights into highly conserved mechanisms by which Lepidopterans combat DNA virus infection ( Ikeda et al . , 2013 ) . Thus , while Lepidopterans provide a relevant model for studying DNA virus-host interaction they have not previously been used to probe restrictions to RNA virus replication . The gypsy moth ( Lymantria dispar ) has been one of the most prolific North American hardwood forest pests since its accidental release in the late 1800’s ( Sparks et al . , 2013 ) . Exploring the susceptibility and responses of L . dispar and other Lepidopterans to virus infection is of particular importance in designing new and effective virus-based biocontrol strategies to minimize the devastating economic impact these species continue to have on the forest industry ( Sparks et al . , 2013 ) . L . dispar-derived cell lines are susceptible to a wide variety of invertebrate DNA viruses , and as such , they are often used in virus-host studies ( Sparks and Gundersen-Rindal , 2011 ) . Interestingly , L . dispar-derived LD652 cells can also support a limited infection by vaccinia virus ( VACV ) , a vertebrate poxvirus encoding a large dsDNA genome ( Li et al . , 1998 ) . During infection of LD652 cells , VACV undergoes early gene expression , DNA replication and late gene expression , but the infection is abortive due to a defect in one or more steps of virion morphogenesis ( Li et al . , 1998 ) . VACV entry and early gene expression have also been documented in Drosophila cells , however viral DNA replication and subsequent late gene expression were not detected , indicating that VACV replication is blocked earlier in its life cycle in Drosophila cells than in LD652 cells ( Moser et al . , 2010 ) . Despite these limitations , RNAi screening of VACV-infected Drosophila cells has identified multiple host factors required for VACV entry in eukaryotic hosts ( Moser et al . , 2010 ) . Thus , the LD652 cell culture system provides a unique model in which to explore multiple aspects of vertebrate DNA virus biology , including basic replication strategies and suppression of host immune pathways by viral proteins . The extent of RNA virus studies in Lepidoptera is limited compared to DNA virus studies and largely restricted to non-enveloped dsRNA and ( + ) -sense ssRNA viruses such as cypoviruses ( Hill et al . , 1999 ) , iflaviruses ( van Oers , 2010 ) and tetraviruses ( Short and Dorrington , 2012 ) . These viruses only infect invertebrate hosts and several cannot productively replicate in cultured cells ( Short and Dorrington , 2012 ) . Furthermore , to our knowledge , ( − ) -sense ssRNA viruses have not been previously reported to productively infect Lepidopteran hosts . A new model system for studying RNA viruses in Lepidopteran hosts may be useful in the design of new biocontrol agents for pest species and improve our understanding of RNA virus-induced disease in vertebrates . Here we explore RNA virus-Lepidopteran host interactions by infecting LD652 cells with the ( − ) -sense ssRNA vesicular stomatitis virus ( VSV ) or the ( + ) -sense ssRNA Sindbis virus ( SINV ) , both of which replicate in a wide range of invertebrate and vertebrate hosts ( Letchworth et al . , 1999; Xiong et al . , 1989 ) . We unexpectedly found that LD652 cells restrict both VSV and SINV replication after virus entry . Using RNAi to knock down the expression of candidate L . dispar antiviral immunity factors , we show that specific RNAi and innate immune pathway components restrict RNA virus replication . We also uncover a role for the Lepidopteran ubiquitin-proteasome system ( UPS ) in restricting RNA virus replication . Surprisingly , co-infection with VACV strongly suppressed this restriction , suggesting that VACV encodes one or more factors that promote RNA virus replication . Using RNAi and genetic techniques , we found that the highly conserved , and previously uncharacterized , VACV A51R gene product is sufficient to alleviate the LD652 cell restriction to VSV and SINV replication . Interestingly , A51R formed aggregate- and filament-like structures that colocalize with microtubules ( MTs ) and protected MTs from depolymerization . Using alanine mutagenesis , we further show that an A51R point mutant with reduced RNA virus rescue ability still forms filamentous structures and stabilizes MTs , suggesting that A51R functions , in addition to MT stabilization , are required for disarming Lepidopteran antiviral immunity . Using mass spectrometry-based techniques , we found that A51R co-immunoprecipitates with several host proteins , including ubiquitin ( Ub ) . Using radiolabeling and immunoblotting , we show that A51R does not affect viral mRNA translation rates but does promote virus protein stability , possibly by inhibiting Ub-dependent host targeting of viral proteins for degradation . Importantly , we show that A51R is also required for efficient replication of VACV in vertebrate cells and for pathogenesis in mice , indicating that A51R is a VACV virulence factor . Collectively , our findings demonstrate the utility of Lepidopteran systems for the study of RNA- and DNA virus host interactions and shed light on how this economically-important order of insects restricts virus replication .
To determine the susceptibility of L . dispar cells to RNA virus infection , we challenged LD652 cells with recombinant strains of VSV and SINV that express either green fluorescent protein ( GFP ) or luciferase ( LUC ) from viral promoters . In single infections with VSV-GFP ( Kato et al . , 2005 ) or SINV-GFP ( Cristea et al . , 2006 ) , we found that , even at a high multiplicity of infection ( MOI ) of 10 , <4% of LD652 cells exhibited GFP fluorescence by 96 hr post-infection ( hpi ) ( Figure 1A , B ) . A previous report indicated that VACV enters LD652 cells and reaches the stage of late gene expression but ultimately fails to complete virion morphogenesis ( Li et al . , 1998 ) . VACV , like other poxviruses , encodes numerous immunomodulatory proteins that inhibit a wide variety of host antiviral pathways ( Smith et al . , 2013 ) . We therefore wondered if VACV co-infection ( although abortive ) might nevertheless overcome an immune response in LD652 cells that restricts RNA virus replication . Consistent with this idea , when LD652 cells were co-infected with VSV-GFP or SINV-GFP and the Western Reserve ( WR ) strain of VACV ( VACV-WR ) , the number of GFP-positive cells increased to ∼77% ( VSV-GFP ) and ∼45% ( SINV-GFP ) by 96 hpi ( Figure 1A , B ) . 10 . 7554/eLife . 02910 . 003Figure 1 . Restriction of RNA virus replication in LD652 cells is relieved by VACV co-infection . ( A ) GFP fluorescence ( top ) and phase contrast images ( PC , bottom ) of infected LD652 cells at 96 hpi . Images are shown at 20X magnification . ( B ) Percentage of GFP-positive LD652 cells from experiments in ( A ) . ( C and D ) LUC assay [arbitrary light units ( LU ) ] of lysates from mock-infected cells or cells infected with VSV-LUC ( C ) or SINV-LUC ( D ) , in the absence or presence of VACV-WR . Mock-infected data are identical in ( C ) and ( D ) . ( E and F ) Immunoblot of LUC , VSV M ( E ) or SINV E1 ( F ) , VACV F4L , and cellular actin proteins in lysates from ( C ) and ( D ) 72 hpi . ( G and H ) VSV-LUC ( G ) or SINV-LUC ( H ) titers , ( plaque-forming units ( PFU ) /ml ) in culture supernatants from ( C ) and ( D ) , respectively . Quantitative data represent means ( ±SEM ) . See also Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 00310 . 7554/eLife . 02910 . 004Figure 1—figure supplement 1 . Restriction of RNA virus replication in L . dispar cells occurs at a step post-entry . VSV N ( A ) and SINV E1 ( B ) virion protein staining inside LD652 cells by confocal microscopy 24 hpi . Scale bars represent 10 μm . ( C ) RT-PCR assay of VSV ( + ) -sense transcription and cellular actin mRNA levels in LD652 cells at the indicated times post-infection . ( D ) RT-PCR assay of SINV ( − ) -sense transcription and cellular actin mRNA levels in LD652 cells at the indicated time post-infection . ( E and F ) Effect of timing of VACV-WR co-infection on VSV-LUC ( E ) or SINV-LUC ( F ) gene expression . Cells were first infected with the indicated RNA virus and then infected with VACV immediately ( T = 0 ) or at the indicated hours ( T = 2–24 ) post-RNA virus infection . Lysates were collected for LUC assay ( shown as arbitrary light units [LU] ) 72 hr post RNA virus infection . Fold change in LU was calculated by dividing LU obtained from each co-infection condition by LU obtained from cells infected with VSV-LUC or SINV-LUC in the absence of VACV-WR . Data represent means ( +SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 004 We next employed VSV-LUC ( Cureton et al . , 2009 ) and SINV-LUC ( Cook and Griffin , 2003 ) recombinant strains along with sensitive chemiluminescence-based LUC assays to detect and measure virus gene expression . Using this strategy , we detected small , ∼3–5-fold increases in light units ( LU ) during VSV-LUC ( Figure 1C ) or SINV-LUC ( Figure 1D ) infection of LD652 cells between 8 and 24 hpi , with levels plateauing by 24 hpi . Notably , these single infections still produced LU levels ∼10-fold above mock-infected cells by 24 hpi . Co-infection of VSV-LUC or SINV-LUC with VACV-WR initially yielded LU levels similar to those seen in single infection by 8 hpi; however , LU readings increased logarithmically by 24–48 hpi ( Figure 1C , D ) . The trends observed in the LUC assays were further confirmed by immunoblotting for LUC and other viral proteins ( Figure 1E , F ) . Although we detected LUC enzymatic activity in VSV-LUC-infected cells in the absence of VACV-WR co-infection , we were typically unable to detect LUC protein on immunoblots under these conditions . We did , however detect a small amount of VSV Matrix ( M ) structural protein ( Figure 1E ) . Enhancement of LUC and SINV E1 capsid protein expression during VACV-WR co-infection with SINV-LUC was also confirmed by immunoblot ( Figure 1F ) . Importantly , measurement of VSV-LUC and SINV-LUC titers from LD652 cultures also reflected LUC activity , remaining unchanged over time in single infections but increasing during VACV-WR co-infection ( Figure 1G , H ) . Collectively , these data show that LUC activity mirrors RNA virus gene expression and virion production thus providing a sensitive measure of RNA virus replication in LD652 cells . The low level , but above background , LUC activity detected in VSV-LUC and SINV-LUC singly-infected LD652 cultures suggested that these viruses are blocked at one or more steps post-entry . Consistent with this , we detected VSV and SINV capsids inside singly-infected LD652 cells using confocal microscopy ( Figure 1—figure supplement 1A , B ) . Furthermore , at 8 hpi RT-PCR analysis of VSV- and SINV-infected cells detected a similar level of viral transcripts with or without VACV-WR co-infection ( Figure 1—figure supplement 1C , D ) . Remarkably , we could also rescue VSV and SINV gene expression ( albeit to reduced levels ) by the addition of VACV-WR as late as 24 hpi with RNA virus ( Figure 1—figure supplement 1E , F ) . Together these data indicate that RNA virus restriction occurs post-entry . To ask if host transcription was required to resist RNA virus infection , we treated VSV-LUC and SINV-LUC-infected LD652 cells with increasing doses of actinomycin D ( ActD ) ( Black and Brown , 1968 ) and then measured LUC activity 48 hpi . We chose an ActD dose range such that the highest dose ( 0 . 1 μg/ml ) reduced cell viability by ∼50% after 48 hr of treatment ( Figure 2A ) . We found that the higher ActD doses enhanced viral gene expression by as much as ∼100-fold compared to control treatments ( Figure 2B ) . These results suggest that host cell transcription is required for LD652 resistance to RNA virus infection and that LD652 cells express antiviral immunity factors . It should be noted that ActD treatment can induce apoptosis in invertebrate cells ( Wang et al . , 2008 ) and previous studies have found apoptosis induction to enhance SINV replication in mosquitoes ( Wang et al . , 2012 ) , thus it is possible that the enhanced viral replication observed in the presence of ActD may be in part due to apoptosis induction in LD652 cells . 10 . 7554/eLife . 02910 . 005Figure 2 . Host transcription and RNAi- , NF-κB- , and ubiquitin-related factors restrict RNA virus replication in LD652 cells . ( A ) Trypan blue exclusion assay to measure cell viability in the presence of ActD ( 48 hr ) . ( B ) Effect of ActD on virus LUC activity ( in LU ) 48 hpi . ( C and D ) LUC activity ( LU ) in lysates from cells 48 hpi with either VSV-LUC ( C ) or SINV-LUC ( D ) and after RNAi of L . dispar transcripts relative to LU generated in GFP ( control ) RNAi treatments . Data represent means ( +SEM ) . See also Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 00510 . 7554/eLife . 02910 . 006Figure 2—figure supplement 1 . RNA virus gene expression is enhanced upon inhibition of the ubiquitin-proteasome system in L . dispar cells . ( A ) RNAi knockdown of ubiquitin ( Ub ) transcripts increases viral gene expression . LD652 cells were transfected with the indicated dsRNAs for 24 hr and then infected with the indicated viruses for 48 hr . In parallel experiments , uninfected cells were transfected with a p166 vector encoding LUC ( p166-LUC ) along with each indicated dsRNA . Lysates were collected 72 hr post-transfection and analyzed for LUC activity ( in LU ) . Data are plotted as fold change when compared to LU generated in GFP dsRNA ( control ) treatments within each treatment group . ( B and C ) Effect of the E1 Ub-activating enzyme inhibitor PYR-41 ( B ) or the proteasome inhibitor MG132 ( C ) on virus gene expression . Cells were infected with the indicated viruses or transfected with p166-LUC . Media was replaced after 2 hr of infection ( VSV/SINV-LUC ) or 5 hr of transfection ( p166-LUC ) with either vehicle-containing medium ( control ) or medium containing 50 µM PYR-41 or 40 µM MG132 . Lysates were prepared 48 hr later and assessed for LUC activity . Data are plotted as fold change in LU when compared to LU generated in control medium treatments . ( D ) Effect of expression of coronavirus PLpro deubiquitinase on virus gene expression . LD652 cells were transfected with empty pIZ/His-V5 vector ( EV ) or vectors encoding PLpro or the catalytic mutant form PLproC112A for 24 hr followed by infection with the indicated viruses . In parallel cultures , uninfected cells were co-transfected with the indicated pIZ/His-V5 constructs along with p166-LUC . In each case , cells were collected 48 hpi/transfection and analyzed for LUC activity . Data are plotted as fold change in LU when compared to LU generated in empty vector transfection treatments . Data represent means ( +SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 006 To identify viral-resistance pathways in L . dispar cells , we used our published ( Sparks and Gundersen-Rindal , 2011 ) , as well as unpublished mRNA sequencing ( mRNA-seq ) data to identify candidate L . dispar homologs of eukaryotic innate immunity factors . We then prepared dsRNA to silence candidate immune-related factors by RNAi in LD652 cells . After RNAi knockdown , we challenged cells with VSV-LUC and SINV-LUC infection . As a positive control for knockdown , we used dsRNAs targeting luc sequences . Knockdowns of four transcripts , encoding homologs of the RNAi pathway proteins AGO2 and Dicer-2 ( Kingsolver et al . , 2013 ) , the Nuclear Factor kappa B ( NF-κB ) -homolog Relish ( Dushay et al . , 1996 ) , and the E2 ubiquitin ( Ub ) -conjugating enzyme Effete ( Treier et al . , 1992 ) , increased LUC expression ∼10-fold in both VSV-LUC- and SINV-LUC-infected cells . RNAi-targeting three Ub-encoding genes , including Polyubiquitin , Ub-RPL40 and Ub-RPS27A increased LUC expression nearly 1000-fold in VSV-LUC-infected cells ( Figure 2C ) and 10-fold in SINV-LUC-infected cells ( Figure 2D ) . Ub-RPL40 and Ub-RPS27A encode highly conserved fusion proteins that are cleaved by endogenous proteases into free Ub and either ribosomal protein L40 ( RPL40 ) or ribosomal protein S27A ( RPS27A ) , components of the 60S and 40S ribosomal subunits , respectively ( Finley et al . , 1989 ) . The dramatic effect of Ub RNAi on LUC expression in VSV-LUC-infected cells was not caused by increased LUC protein stability due to inhibition of the Ub-proteasome system ( UPS ) ( See Figure 2—figure supplement 1 ) . We found that viral-driven , but not plasmid-driven ( p166-LUC ) , LUC expression was enhanced in experiments aimed at inhibiting the UPS , including: ( i ) exposure to Ub RNAi ( Figure 2—figure supplement 1A ) , ( ii ) treatment with inhibitors of either E1 Ub-activating enzymes ( PYR-41; Figure 2—figure supplement 1B ) ( Yang et al . , 2007 ) , or the proteasome ( MG132; Figure 2—figure supplement 1C ) ( Lee and Goldberg , 1998 ) , and finally , by ( iii ) transient expression of a wild-type ( but not a catalytically inactive ) coronavirus-encoded deubiquitinase ( PLpro; Figure 2—figure supplement 1D ) ( Barretto et al . , 2005 ) . Taken together , these findings suggest that host restriction of RNA virus replication in LD652 cells is mediated by multiple host factors including those involved in RNAi- , NF-κB , and Ub-related pathways . We next sought to identify the VACV gene ( s ) responsible for relieving the LD652 restriction to RNA virus infection using several complementary lines of investigation . To test if VACV binding and entry into LD652 cells is sufficient to rescue RNA virus infection in the absence of VACV gene expression , we co-infected LD652 cells with VSV or SINV and a heat-inactivated strain of VACV expressing the late A5L core protein fused to GFP ( A5L-GFP ) ( Carter et al . , 2003 ) . Heat inactivated VACV , which can enter cells but cannot express its genes ( Dales and Kajioka , 1964 ) , failed to rescue VSV and SINV replication ( Figure 3A ) , indicating that VACV gene expression was required for rescue . Next we treated co-infected cells with the viral DNA polymerase inhibitor cytosine arabinoside ( AraC ) at a dose that blocks viral DNA replication and subsequent late poxvirus gene expression ( Li et al . , 1998 ) . We confirmed that AraC treatments blocked late VACV gene expression by immunoblotting for A5L-GFP in the absence or presence of AraC ( data not shown ) . We found that inhibiting post-replicative VACV gene expression did not prevent rescue of either VSV or SINV ( Figure 3B ) , narrowing the RNA virus rescue activity to one or more of the 118 early VACV genes ( Yang et al . , 2010 ) . 10 . 7554/eLife . 02910 . 007Figure 3 . Characterization of VACV-dependent RNA virus rescue in LD652 cells . ( A ) Effect of heat inactivation of VACV strain A5L-GFP on RNA virus rescue 48 hpi by LUC assay ( in LU ) . LU generated from co-infection lysates and are plotted as 'fold change' with respect to single infection conditions . ( B ) Effect of AraC ( 200 μg/ml ) treatment on A5L-GFP-mediated RNA virus rescue 48 hpi . Fold change in LU was calculated as in ( A ) . ( C and D ) Relative LUC activity in lysates from cells infected with VSV-LUC ( C ) or SINV-LUC ( D ) and co-infected with various VACV strains and in the presence of AraC for 72 hr . Fold change in LU was calculated as in ( A ) VACV strain Copenhagen ( VACV-COP ) ; modified VACV Ankara ( VACV-MVA ) . ( E ) Immunoblot of lysates from ( C ) for LUC , I3L , and actin . Data represent means ( +SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 007 We next screened a collection of VACV mutants for their ability to rescue RNA viruses in LD652 cells . We found that the attenuated modified VACV strain Ankara ( VACV-MVA ) ( Antoine et al . , 1998 ) failed to fully rescue either VSV or SINV as compared to either VACV-WR or Copenhagen strains ( Figure 3C , D ) , despite similar levels of viral gene expression by all three VACV strains ( Figure 3E ) . These results suggested that VACV-MVA is defective for one or more factors required for RNA virus rescue in LD652 cells . The above findings narrowed the number of candidates to ∼30 genes that are deleted or truncated in VACV-MVA relative to VACV-WR ( Meisinger-Henschel et al . , 2007 ) . We then used RNAi to knock down each of the candidate VACV genes during VSV and SINV co-infection with VACV-WR ( Figure 4A ) . RNAi of a single , uncharacterized VACV gene , A51R , caused RNA virus gene expression to drop below an arbitrary cut off of 50% of the control ( GFP ) RNAi treatment . The A51R knockdown reduced virus gene expression by ∼95% ( VSV ) and by ∼69% ( SINV ) ( Figure 4B , Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 02910 . 008Figure 4 . A51R is a major determinant of VACV-mediated rescue of RNA virus replication in LD652 cells . ( A ) Strategy to identify VACV 'rescue' factors using dsRNA-mediated RNAi in L . dispar cells . ( B ) Relative LUC activity 72 hpi in lysates from cells infected with VSV-LUC and VACV-WR after the indicated RNAi treatment . Fold change in LU was calculated as in Figure 3A . ( C and D ) Relative LUC activity in lysates from cells co-infected with recombinant VACV strains and VSV-LUC ( C ) or SINV-LUC ( D ) in the presence of AraC ( 200 μg/ml ) for 72 hr . ( E ) Relative LUC activity in lysates of cells transfected with empty p166 vector , Flag-GFP , or Flag-A51R for 24 hr and then infected with VSV-LUC or SINV-LUC for 48 hr . Fold change in LU are relative to LU obtained from empty vector treatment . ( F ) Immunoblot of Flag-GFP ( lower band ) and Flag-A51R ( upper band ) from lysates in ( E ) . Actin served as a loading control . ( G–H ) VSV-LUC ( G ) and SINV-LUC ( H ) titers from experiments in ( E ) . ( I ) Cell viability after VSV infection as measured by trypan blue exclusion assay . Cells were transfected for 24 hr with either Flag-GFP or Flag-A51R vectors and then mock-infected or infected with VSV-LUC . Data represent means ( +SEM ) . See also Figure 4—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 00810 . 7554/eLife . 02910 . 009Figure 4—figure supplement 1 . Identification of VACV A51R as a determinant of SINV rescue in L . dispar cells . RNA interference ( RNAi ) assay in LD652 cells to identify VACV-encoded determinants of SINV rescue . L . dispar cells were treated as described in the legend for Figure 4A , B and then analyzed by LUC assay . Data represent means ( +SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 00910 . 7554/eLife . 02910 . 010Figure 4—figure supplement 2 . Characterization of A51R-related recombinant VACV strains . ( A ) Agarose gel showing RT-PCR analysis of A51R , I3L , and actin mRNA expression in recombinant VACV strains 24 hpi in LD652 cells . ( B ) Immunoblot of cell lysates from BSC-40 cells infected for 24 hr with the indicated strains ( MOI = 1 ) for Flag-A51R , VACV I3L and actin . ( C ) Immunoblot of BSC-40 cell lysates infected with the indicated recombinant VACV strains ( MOI = 1 ) for 24 hr . Lysates were probed with antibodies for Flag ( to detect Flag-A51R/Flag-MVAA51R ) , I3L , and actin . ( D ) Immunoblot of Flag-A51R protein expression during ΔA51RFREV infection ( MOI = 1 ) of BSC-40 cells . Lysates were prepared at the indicated times post-infection and analyzed by immunoblot for: early VACV proteins ( F4L ) , late VACV proteins ( A27L ) , Flag ( Flag-A51R ) , and actin . Where indicated , 50 μg/ml AraC was present in the cell culture medium throughout the time course . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 010 To confirm the necessity of A51R for VACV rescue of RNA virus infection , we created a recombinant VACV-WR strain with a deletion of the A51R gene ( ΔA51R ) ( Figure 4—figure supplement 2A ) . Importantly , we found that co-infection with ΔA51R resulted in a 100-fold reduction in VSV gene expression relative to co-infection with VACV-WR ( Figure 4C ) . VSV was rescued by co-infecting with revertant VACV strains ( Figure 4C ) , expressing a Flag-tagged A51R gene ( Flag-A51R ) reintroduced into either the A51R locus ( ΔA51RFA51R ) or the VACV thymidine kinase ( J2R ) locus ( ΔJ2R/ΔA51RFA51R ) . In addition , we reverted the ΔA51R strain with a Flag-tagged form of the VACV-MVA A51R gene ( Flag-MVAA51R ) introduced into the J2R locus ( ΔJ2R/ΔA51RFMVAA51R ) . Flag-A51R and Flag-MVAA51R proteins migrated as single bands of ∼38 and 35 kDa , respectively , on immunoblots ( Figure 4—figure supplement 2B , C ) . The lower molecular weight of the VACV-MVA protein is likely due to a C-terminal truncation of ∼20 amino acids that results from a genomic deletion that occurred during VACV-MVA passage in culture ( Antoine et al . , 1998 ) . Furthermore , consistent with the idea that the VACV-MVA A51R protein is not functional , we found that the ΔJ2R/ΔA51RFMVAA51R strain failed to fully rescue VSV ( Figure 4C ) . Importantly , immunoblot of lysates from a time course of A51RFA51R infection confirmed that Flag-A51R is expressed early in VACV infection , as it was detected by 2 hpi in the absence or presence of AraC ( Figure 4—figure supplement 2D ) . Together these results indicate that A51R is an early VACV protein necessary to relieve the restriction of VSV in LD652 cells . A51R was also required for full rescue of SINV replication by VACV ( Figure 4D ) . However , we noticed that ΔA51R and ΔJ2R/ΔA51RFMVAA51R strains exhibited only a moderate fourfold reduction in their ability to promote SINV gene expression as compared to 100-fold reductions for VSV ( Figure 4C , D ) . These findings suggest that other VACV factors in addition to A51R promote SINV replication in LD652 cells . We next wished to determine if expression of A51R was sufficient to rescue VSV and SINV infection in LD652 cells in the absence of other VACV components . Therefore , we cloned Flag-A51R into a p166 expression vector and transiently transfected LD652 cells . Although only ∼40–60% of cells expressed detectable Flag-A51R protein ( data not shown ) , we observed an ∼60-fold increase in VSV and 12-fold increase in SINV gene expression relative to control transfections with empty p166 vector or vector encoding Flag-tagged GFP ( Flag-GFP ) ( Figure 4E ) . Immunoblots confirmed similar levels of Flag-GFP and Flag-A51R expression in these lysates ( Figure 4F ) . VSV and SINV titers were similarly enhanced by Flag-A51R transfection ( Figure 4G–H ) . These results indicate that Flag-A51R expression is sufficient to overcome the host restriction to RNA virus replication in the absence of other VACV proteins . The stronger A51R-dependent rescue of VSV compared to SINV led us to focus on VSV infection assays in our further studies of A51R-mediated rescue . We next tested if VSV replication affected LD652 cell viability . Transfection of Flag-GFP or Flag-A51R constructs into LD652 cells did not lead to notable changes in cell viability over a 7 day time course . However , when transfected cells were subsequently infected with VSV-LUC , cell viability began to drop by 96 hpi , ultimately reaching cell viabilities of ∼50% by day 7 of infection ( Figure 4I ) . This reduced cell viability was not observed when Flag-GFP-transfected cells were infected with VSV-LUC , suggesting that active viral replication was required to induce cell death ( Figure 4I ) . A51R genes are found in most vertebrate poxvirus genera ( Table 1 ) yet absent in the genomes of entomopoxviruses ( not shown ) . We generated constructs to express some of these A51R homologs , with amino acid identities ranging from 90% to 30% relative to VACV-WR A51R ( Table 1 ) , and asked whether the ability of A51R to rescue VSV replication is conserved between poxvirus genera . Strikingly , we found that A51R from each poxvirus enhanced VSV gene expression to a similar level ( Figure 5 ) . Thus , the ability of A51R to overcome the LD652 restriction to VSV infection is a conserved function of distantly-related A51R homologs . 10 . 7554/eLife . 02910 . 011Table 1 . Differential conservation of Chordopoxirinae A51R genesDOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 011GenusA51RExample species ( % amino acid identity to VACV-WR A51R ) Orthopoxvirus+VACV*HSPV† ( 96 ) CPXV ( 94 ) ECTV ( 93 ) VARV ( 92 ) Suipoxvirus+SPXV ( 33 ) Yatapoxvirus+TANV ( 35 ) YLDV ( 35 ) Leporipoxvirus+MYXV ( 35 ) SFV ( 35 ) Capripoxvirus+GTPV ( 32 ) SPPV ( 30 ) LSDV ( 29 ) Cervidpoxvirus+DPV ( 31 ) Parapoxvirus+ORFV ( 22 ) Molluscipoxvirus–MCVAvipoxvirus–FPVCNPVUnclassified–CRV*VACV strains MVA ( Antoine et al . , 1998 ) and Dryvax ( Qin et al . , 2011 ) contain a truncated and fragmented A51R gene , respectively . †HSPV contains a fragmented A51R gene ( Tulman et al . , 2006 ) . ‘+’ Indicates presence , and ‘−‘ indicates absence of A51R gene in viral genomes within each genus . Abbreviations: VACV , vaccinia virus; HSPV , horsepox virus; TATV , taterapox virus; VARV , variola virus; SPXV , swinepox virus; TANV , tanapox virus; yaba-like disease virus; MYXV , myxoma virus; SFV , Shope fibroma virus; GTPV , goatpox virus; SPPV , sheeppox virus; LSDV , lumpy skin disease virus; DPV , deerpox virus; FPV , fowlpox virus; CNPV , canarypox virus; MCV , molluscum contagiosum; ORFV , orf virus; CRV , crocodilepox virus . 10 . 7554/eLife . 02910 . 012Figure 5 . A51R proteins from disparate poxviruses rescue RNA virus replication in LD652 cells . Relative LUC activity in lysates of cells expressing the indicated A51R construct and infected with VSV-LUC for 48 hr . Fold change in LU are relative to LU obtained in Flag-GFP treatments . CPXV , cowpox virus; ECTV , ectromelia virus; DPXV , deerpox virus; YLDV , Yaba-like disease virus; and MYXV , myxoma virus . Data represent means ( +SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 012 The ability of a vertebrate poxvirus to rescue VSV replication in LD652 cells was so unexpected that we asked whether other cell lines from L . dispar or other Lepidopteran species similarly restrict VSV replication . Indeed , we found that A51R expression relieved restriction to VSV replication in each of the three additional Lepidopteran cell lines tested , including an L . dispar embryonic cell line ( LdEP ) ( Figure 6A ) , a Spodoptera frugiperda-derived cell line ( Sf9 ) ( Figure 6B , C ) , and a Manduca sexta-derived cell line ( GV-1 ) ( Figure 6D , E ) . 10 . 7554/eLife . 02910 . 013Figure 6 . A51R relieves RNA virus restriction in multiple Lepidopteran hosts . ( A ) Relative LUC activity in lysates of L . dispar-derived embryonic ( LdEP ) cells co-infected with VSV-LUC and VACV-WR or ΔA51R strains . Fold changes in LU are relative to LU obtained from VSV-LUC ( 8 hpi ) lysates . ( B ) Relative LUC activity in lysates of Spodoptera frugiperda-derived Sf9 cells expressing Flag-GFP or Flag-A51R and infected with VSV-LUC . Fold change in LU are relative to LU obtained from Flag-GFP ( 8 hpi ) lysates . ( C ) Virus titer of supernatants from ( B ) . ( D ) Relative LUC activity in lysates of Manduca sexta-derived GV-1 cells expressing Flag-GFP or Flag-A51R and infected with VSV-LUC . Fold change in LU were calculated as in ( B ) . ( E ) Virus titer of supernatants from ( D ) . Data represent means ( +SEM ) . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 01310 . 7554/eLife . 02910 . 014Figure 6—figure supplement 1 . Effect of Flag-A51R expression or VACV-WR co-infection on VSV-LUC gene expression in Drosophila cells . ( A ) Effect of Flag-A51R expression on VSV-LUC gene expression in DL1 cells . Cells were transfected with p166 vectors encoding Flag-A51R or Flag-GFP and infected with VSV-LUC ( MOI = 0 . 1 ) . 24 hpi , lysates were analyzed by LUC assay . Fold change in LU was calculated by dividing LU in Flag-A51R treatments with LU in Flag-GFP treatments . ( B ) Immunoblot of lysates from experiments in ( A ) for Flag-GFP ( lower band ) , Flag-A51R ( upper band ) , and actin . ( C ) Effect of VACV-WR co-infection on LUC activity . DL1 cells were infected with VSV-LUC ( at the indicated MOIs ) in the absence or presence of VACV-WR ( MOI = 10 ) . At the indicated hpi , lysates were analyzed by LUC assay . Data represent means ( ±SEM ) . ( D ) Immunoblot of lysates in ( C ) where cells were mock-infected or infected with VSV-LUC ( MOI = 0 . 01 ) in the absence or presence of VACV-WR ( MOI = 10 ) for 48 hr . Lysates were blotted for VSV M , VACV I3L , and actin proteins . Data represent means ( +SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 014 In contrast , A51R expression did not enhance VSV replication in D . melanogaster-derived DL1 cells ( Figure 6—figure supplement 1A ) , despite significant expression of Flag-A51R protein ( Figure 6—figure supplement 1B ) . VSV gene expression in DL1 cells was also not enhanced by co-infection with VACV-WR ( Figure 6—figure supplement 1C , D ) . Thus , if Drosophila cells do have restrictions targeted by A51R , these restrictions are not sufficient to measurably check VSV replication in these experiments . Bioinformatic analyses of A51R proteins with public protein databases failed to reveal any functional domains or motifs that might provide insight into A51R function . Therefore we used confocal microscopy to begin A51R characterization by examining A51R subcellular localization . We found that Flag-A51R , expressed under its natural promoter from the ΔA51RFA51R virus , formed both aggregate-like as well as filamentous structures in the cytoplasm of African green monkey-derived ( BSC-40 ) cells . This staining pattern significantly overlapped with that of cellular tubulin , suggesting that A51R associates with a subset of MTs ( Figure 7—figure supplement 1A ) . Indeed deconvolution of images from these experiments revealed Flag-A51R staining localized on intact MT tracks ( Figure 7—figure supplement 1B ) . Consistent with the idea that A51R not only co-localizes with MTs but also alters their properties , we observed enhanced co-localization on A51R-dependent , drug-resistant MT structures in infected cells treated with the MT-depolymerizing drug nocodazole . In the presence of nocodazole , MT structures were essentially absent in mock-infected and ΔA51R-infected cells . However filamentous , drug-resistant ‘MT pieces’ were abundant in ΔA51RFA51R-infected BSC-40 cells ( Figure 7—figure supplement 1C ) . Nocodazole-resistant MT pieces were previously observed in mammalian cells infected with VACV-WR but not with VACV-MVA ( Ploubidou et al . , 2000; Schepis et al . , 2006 ) . We therefore examined Flag-MVAA51R protein localization in nocodazole-treated cells and found that , although Flag-MVAA51R staining stained small punctate structures throughout the cytoplasm that overlapped with tubulin , large drug-resistant MT-like structures failed to form ( Figure 7—figure supplement 2A ) . These results suggest that Flag-MVAA51R protein is deficient in its ability to both form filament-like structures and to protect MTs from depolymerization . To ask if Flag-A51R localization to and stabilization of MTs requires other VACV proteins , we transfected BSC-40 cells with a plasmid expressing Flag-A51R . We found that Flag-A51R expression was sufficient for both localization to , and stabilization of , MTs ( Figure 7A ) . Moreover , using similar assays , we found that A51R homologs from other poxviruses also co-localize with and protect MTs from depolymerization ( Figure 7—figure supplement 2B ) . Finally , in LD652 cells , as in vertebrate cells , Flag-A51R formed aggregates and filamentous structures that co-localized with and protected MTs from depolymerization by the MT-depolymerizing agent vincristine ( Figure 7B ) . Taken together , our findings suggest that the ability of A51R to interact with the host MT cytoskeleton is a conserved property of A51R proteins . 10 . 7554/eLife . 02910 . 015Figure 7 . Poxvirus A51R proteins localize with and stabilize host microtubules . ( A ) Immunofluorescence of Flag-A51R ( red ) and tubulin ( green ) in BSC-40 cells in the presence or absence of nocodazole 24 hr post-transfection of Flag-A51R vector . ( B ) Immunofluorescence of Flag-A51R ( red ) and tubulin ( green ) in LD652 cells in the presence or absence of vincristine 24 hr post-transfection of Flag-A51R vector . In ( A ) and ( B ) arrows indicate A51R filaments and arrowheads indicate A51R aggregate structures that overlap with tubulin staining . Note the absence of these aggregates and filaments in cells that lack Flag-A51R staining . Scale bars represent 10 μm . See also Figure 7—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 01510 . 7554/eLife . 02910 . 016Figure 7—figure supplement 1 . A51R localizes with and stabilizes MTs . ( A ) Flag-A51R localization during VACV infection of BSC-40 cells . Cells were infected with the ΔA51RFREV strain ( MOI = 3 ) and methanol fixed 8 hpi for confocal microscopy analysis using antibodies against Flag ( red ) and tubulin ( green ) . DAPI staining ( blue ) was also used to mark nuclear and viral DNA in the merged image . ( B ) A single slice from the Z-stack shown in ( A ) illustrating Flag-A51R staining ( red ) colocalizing with tubulin staining ( green ) . The inset shows a magnified view of Flag-A51R staining overlapping with MT tracks . ( C ) Nocodazole-resistant , MT-like filamentous tubulin pieces are only observed in VACV-infected cells when Flag-A51R is also present . BSC-40 cells were either mock-infected or infected with the indicated VACV strains ( MOI = 3 ) for 8 hr . Coverslips were then methanol-fixed and processed for confocal microscopy using Flag ( red ) , VACV ( magenta ) , and tubulin ( green ) antibodies . DAPI staining ( blue ) was also used to mark DNA in merged images . Cell culture media contained 20 μM nocodazole from 2-8 hpi . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 01610 . 7554/eLife . 02910 . 017Figure 7—figure supplement 2 . Localization of Flag-A51R proteins encoded by VACV-MVA and disparate poxviruses in BSC-40 cells . ( A ) Localization of Flag-A51R and Flag-MVAA51R proteins in the presence of nocodazole ( 20 μM ) . Cells were infected with the indicated strains ( MOI = 3 ) and then paraformaldehyde-fixed 8 hpi . Cells were stained with antibodies against Flag ( red ) and tubulin ( green ) . DAPI staining ( blue ) was included in merge images . ( B ) Localization of Flag-tagged A51R proteins from the indicated poxviruses 24 hr post-transfection of BSC-40 cells with pCDNA3 expression plasmids . Where indicated , nocodazole ( 20 μM ) was present in culture media from 5-24 hr post-transfection . Cells were then methanol-fixed and stained with Flag ( red ) and tubulin ( green ) antibodies . Scale bars represent 10 μm in each figure . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 017 Based on the conservation of RNA virus rescue and MT stabilization functions of poxvirus A51R proteins , we wondered if these two functions could be separated . Therefore we performed systematic alanine mutagenesis of A51R residues that were directly conserved or conserved based on charge . After screening over 30 point mutations throughout the VACV A51R protein , we found that almost all of the mutants displayed significantly reduced protein stability , rendering virus rescue and MT-related functions difficult to assess ( data not shown ) . However , alanine substitution of residue R321 , which lies within the C-terminal region of A51R that is missing in VACV-MVA A51R ( Figure 8A ) , produced a suitable mutant for analysis . Flag-A51RR321A proteins displayed a reduced VSV rescue phenotype when compared to Flag-A51R proteins ( Figure 8B ) despite similar levels of expression ( Figure 8C ) . Importantly , when expressed in LD652 cells in the presence of vincristine , both Flag-A51R ( Figure 8D; Video 1 ) and Flag-A51RR321A ( Figure 8D; Video 2 ) formed filamentous structures that overlapped with tubulin staining ( Figure 8D ) . These results suggest that MT stabilization alone is insufficient to promote full RNA virus rescue . 10 . 7554/eLife . 02910 . 018Figure 8 . Microtubule stabilization is insufficient for full RNA virus rescue in LD652 cells . ( A ) C-terminal alignment of poxvirus A51R proteins with site of R321A substitution indicated by an asterisk . Multiple alignments were performed using eBioX Software ( v . 1 . 5 . 1 ) using T-COFFEE alignment parameters . CPXV , cowpox virus; ECTV , ectromelia virus; DPXV , deerpox virus; and YLDV , Yaba-like disease virus . ( B ) Relative LUC activity in lysates of cells transfected with Flag-GFP , Flag-A51R or Flag-A51RR321A p166 constructs for 24 hr and then infected with VSV-LUC for 48 hr . Fold change in LU are relative to LU obtained in Flag-GFP treatments . Data represent means ( +SEM ) . ( C ) Immunoblot of Flag-GFP ( lower band ) and Flag-A51R/A51RR321A ( upper bands ) from lysates in ( B ) . ( D ) Immunofluorescence of Flag-A51R/A51RR321A ( red ) and tubulin ( green ) proteins in LD652 cells in the presence of vincristine 48 hr post-transfection of Flag-A51R vectors . Arrows indicate A51R filaments that overlap with tubulin staining . Scale bars represent 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 01810 . 7554/eLife . 02910 . 019Video 1 . Three-dimensional rendering of Flag-A51R-transfected LD652 cell in the presence of vincristine 48 hr post-transfection . A merge between Flag ( red ) and tubulin ( green ) staining is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 01910 . 7554/eLife . 02910 . 020Video 2 . Three-dimensional rendering of Flag-A51RR321A-transfected LD652 cell in the presence of vincristine 48 hr post-transfection . A merge between Flag ( red ) and tubulin ( green ) staining is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 020 The previous results suggested that A51R may provide functions , beyond MT stabilization , that promote RNA virus replication . In order to explore A51R protein complexes we used highly-sensitive multidimensional protein identification technology ( MuDPIT ) ( Washburn et al . , 2001 ) to identify proteins associated with A51R immunoprecipitation ( IP ) complexes using mass spectrometry . We transfected LD652 cells with either empty vector or with Flag-A51R expression vector and then analyzed Flag antibody immunoprecipitates 48 hr post-transfection . We matched peptides identified in these LD652 cell immunoprecipitates to Flag-A51R sequence as well as to the available Bombyx mori proteome . Filtering differential normalized spectral abundance factor ( NSAF ) ( Zybailov et al . , 2006 ) values from both control and Flag-A51R immunoprecipitates identified the four most-enriched proteins in Flag-A51R treatments . These proteins are the target Flag-A51R , myosin II essential light chain ( a constituent of the actomyosin cytoskeleton [Clark et al . , 2007] ) , Ub , and ribosomal protein S15 ( a component of the 40S subunit of the ribosome that has also been shown to modulate the activity of specific E3 ubiquitin ligases [Daftuar et al . , 2013] ) ( Table 2 ) . Unfortunately we found that the majority of A51R associated with insoluble cell fractions , making direct immunoblot analysis of immunoprecipitated materials impossible . 10 . 7554/eLife . 02910 . 021Table 2 . Proteins most-enriched in Flag-A51R immunoprecipitates from LD652 cellsDOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 021ProteinNo . of spectral countsΔNSAF ( NSAFFlag-A51R-NSAFControl ) Myosin II essential light chain1100 . 0355Ub210 . 0113Flag-A51R330 . 0066Ribosomal protein S15180 . 0056 Given our knowledge that RNAi of Ub transcripts ( Figure 2C , D ) or inhibition of the UPS system ( Figure 2—figure supplement 1 ) enhances RNA virus gene expression in L . dispar cells and that A51R associates with Ub , we asked if A51R expression might affect viral protein stability . To first investigate if there were differences in viral protein translation rates in the presence of A51R , we used a two hour pulse of [35S]methionine to label newly-synthesized proteins in virus-infected LD652 cells . We then analyzed the abundance of newly synthesized , radiolabeled VSV N protein after immunoprecipitation from these lysates . We found that radiolabeled VSV N protein was undetectable in cells infected only with either VACV-WR ( control ) or VSV-LUC . In contrast , we found significant quantities of radiolabeled VSV N protein in VSV-LUC-infected cells that were also infected with either VACV-WR or ΔA51R strains ( Figure 9A , top panel ) . These results suggest that viral protein translation rates are too low to be detected by this method in VSV-LUC single infections but are detectable ( and surprisingly similar ) in co-infections with VACV-WR or ΔA51R strains . When whole cell extracts from these experiments were immunoblotted for total VSV N protein levels , again we found that VSV N was not detected in either VACV-WR or VSV-LUC single infection lysates . However , total VSV N protein levels were dramatically reduced in ΔA51R co-infections compared to co-infections with VACV-WR ( Figure 9A , middle panel ) . To determine if the reduced VSV N protein levels in ΔA51R co-infections might be due to altered VSV N protein stability , we calculated the half-life of newly-synthesized VSV N protein using pulse-chase analyses . Although newly synthesized VSV N protein could be detected during co-infection with either VACV-WR or ΔA51R strains ( Figure 9B ) , the levels of radiolabeled protein appeared to diminish more rapidly during ΔA51R co-infection ( Figure 9C ) . Regression analyses of pulse-chase experiments estimated the half-life of VSV N to be ∼5 . 3 hr during co-infection with VACV-WR while only ∼1 . 8 hr during co-infection with the ΔA51R strain . Collectively , these data suggest that A51R associates with Ub and promotes viral protein stability . 10 . 7554/eLife . 02910 . 022Figure 9 . A51R promotes RNA virus protein stability in LD652 cells . Cells were infected with the indicated strains for 72 hr and then pulsed for 2 hr with [35S]methionine . To visualize radiolabeled , nascent VSV N protein , cell lysates were subjected to immunoprecpitation with anti-VSV N protein antibodies and immunoprecipitated complexes were separated by SDS-PAGE and visualized by autoradiography ( top panel ) . Equal fractions of each total cellular lysate were also used in parallel immunoblot experiments to detect total VSV N ( middle panel ) or actin protein levels . ( B and C ) Cells were infected with VSV-LUC and the indicated VACV strains as in ( A ) , pulsed for 2 hr with [35S]methionine and then chased for the indicated times at which point cell lysates were collected and subjected to immunoprecipitation with anti-VSV N antibodies . Immunoprecipitated complexes were then separated by SDS-PAGE , visualized by autoradiography ( B ) and the percentage of radiolabeled VSV N protein remaining ( compared to T = 0 , representing the end point of the 2 hr pulse ) at each time point post-pulse was plotted ( C ) . Note that lysates containing radiolabeled VSV N protein from VACV-WR and ΔA51R co-infections time courses ( B ) were processed separately and thus band intensities between VACV infection treatments are not directly comparable . Data in ( C ) represent means ( +SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 022 Given that A51R is encoded by vertebrate poxviruses , we wished to determine if A51R contributes to VACV replication and pathogenesis in vertebrate hosts . To do this we first performed growth curve analyses after infection of BSC-40 cells . At a low MOI of 0 . 03 , the ΔA51R , VACV-WR , and ΔA51RFA51R strains replicated with similar kinetics until 24 hpi at which point ΔA51R titers began to plateau . In contrast , VACV-WR and ΔA51RFA51R titers continued to increase to ∼100-fold higher levels by 72 hpi ( Figure 10A ) . This plateau of ΔA51R titer was also observed at a higher MOI of 3 , but occurred earlier , at 12 hpi ( Figure 10B ) . These results indicate that A51R promotes VACV replication in vertebrate cell culture . 10 . 7554/eLife . 02910 . 023Figure 10 . A51R promotes VACV replication and pathogenesis in vertebrates . ( A and B ) VACV titers from BSC-40 cells infected with the indicated strains at low MOI ( 0 . 03 ) ( A ) or high MOI ( 3 ) ( B ) . Data represent means ( +/-SEM ) . ( C ) Body weight of NMRI mice infected with the indicated virus ( 10 , 000 PFU/animal ) . Data represent mean percent body weight change among surviving members in each group at the indicated day post-infection . ( D ) Percent survival of mice from ( C ) . p<0 . 05 indicates a statistically significant difference in survival between VACV-WR- and ΔA51R strain-infected animals . ( E and F ) Virus titer in lung ( F ) and ovary ( F ) tissue from mice infected with VACV-WR or ΔA51R . Each dot indicates the total virus titers from an individual mouse . Horizontal bars represent the mean of each group . Mice were infected as in ( C ) and euthanized at 4 or 6 days post-infection . p<0 . 05 indicates a statistically significant difference between infection group tissue titers . See also Figure 10—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 02310 . 7554/eLife . 02910 . 024Figure 10—figure supplement 1 . The ΔA51R strain displays attenuated pathogenesis in NMRI mice , releases increased levels of EEV , and does not display enhanced sensitivity to IFN treatment . Groups of 5 NMRI mice were intranasally-inoculated with the indicated doses of either VACV-WR ( A ) or ΔA51R ( B ) strains and percentage survival of these groups of mice was tracked over the indicated time course . ( C ) Effect of A51R deletion on extracellular enveloped virus ( EEV ) production . BSC-40 cells were infected with the indicated strains at a MOI of 3 for 24 hr at which point supernatants were collected by low speed centrifugation , neutralized for intracellular mature virus ( IMV ) infectivity by incubation with L1R antibody ( Backes et al . , 2012 ) and titered on BSC-40 cell monolayers to determine ‘EEV’ titers . In parallel cultures ‘total’ virus titer ( IMV+EEV ) was determined by collecting both supernatants and cells , followed by three rounds of freeze-thaw and titration on BSC-40 cell monolayers . Data represent means ( +SEM ) . ( D ) Effect of interferon ( IFN ) pretreatment on ΔA51R virus replication in BSC-40 cells . Cell culture medium containing 1000 U/ml recombinant IFN-α or normal growth medium ( control ) was used to replace growth medium of cells 16 hr prior to infection with the indicated viruses ( MOI = 0 . 03 ) . After 24 of infection , total virus titers were determined by plaque assay as in ( C ) . Data represent means ( +SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02910 . 024 To determine if A51R also promotes VACV pathogenesis , we intranasally-infected groups of NMRI mice with either VACV-WR or ΔA51R and then tracked body weight ( Figure 10C ) and survival ( Figure 10D ) of animals over time . Both VACV-WR and ΔA51R infections led to rapid weight loss of animals by days 4 and 5 , respectively . In animals infected with VACV-WR , weight loss continued on days 6–7 post-infection , with 5/5 animals ultimately succumbing to infection by day 7 . In contrast , only 1/5 animals infected with the ΔA51R strain died by day 7 , while the remaining animals recovered from infection , gained weight ( Figure 10C ) and survived until the end of the 28 day experiment ( Figure 10D ) . We further confirmed in separate experiments , and at multiple doses of virus , that the ΔA51R virus displays attenuated virulence compared to VACV-WR ( Figure 10—figure supplement 1A , B ) . To determine the relative amount of virus replication that occurred during these infections in vivo , we isolated lung and ovarian tissues on days 4 and 6 post-infection and measured virus titers by plaque assay . After four days of infection , both VACV-WR and ΔA51R viruses replicated to similar titers in lung tissue , the primary site of infection . By day 6 , however , a trend toward lower ΔA51R titers in the lungs was observed , although this difference was not statistically significant ( p=0 . 05; Figure 10E ) . Neither VACV-WR nor ΔA51R viruses were detected in isolated ovaries by day 4 post-infection , suggesting that these viruses had not yet spread to this distant tissue . By day 6 post-infection , however , we detected significantly higher virus titers ( p<0 . 05 ) in the ovaries of all VACV-WR-infected animals compared to ΔA51R-infected animals , in which virus was detectable in only 1/5 animals ( Figure 10F ) . These data suggest that A51R is not only important for VACV replication in cell culture but also for virulence in vivo . The reduced replication of the ΔA51R strain at later time points of infection during in vitro and in vivo infections was not caused by a reduction in the extracellular enveloped ( EEV ) form of VACV , which primarily mediates spread of the virus ( Smith et al . , 2002 ) . Our analysis revealed that , despite ∼10-fold lower titers of total ( intracellular and extracellular ) virus in ΔA51R cultures than in VACV-WR or ΔA51RFA51R cultures , ΔA51R cultures shed ∼twofold higher levels of EEV than VACV-WR or ΔA51RFA51R cultures ( Figure 10—figure supplement 1C ) . The early plateau of ΔA51R replication and reduced load of this mutant in ovarian tissues might be explained by an inability to overcome an immune response caused by initial infection . We therefore pre-treated BSC-40 cells with interferon ( IFN ) to induce an antiviral state in cells prior to VACV infection and asked whether A51R is required to overcome an IFN response . Comparing the titers of VACV-WR , ΔA51R , or ΔA51RFA51R cultures pre-treated with IFN to parallel cultures without IFN treatment , we found that IFN pre-treatment reduced VACV-WR and ΔA51RFA51R titers by ∼10-fold by 24 hpi relative to control treatment . ΔA51R titers , however , were similarly reduced at this time point , suggesting that loss of A51R does not make VACV more susceptible to the antiviral effects of IFN ( Figure 10—figure supplement 1D ) .
Insect model systems provide powerful tools for probing virus-host interactions but have primarily focused on Dipteran hosts ( Kingsolver et al . , 2013 ) . Development of new virus-host models in Lepidoptera is important for several reasons . First , these insects may encode antiviral mechanisms that are not found in Diptera . Second , our current understanding of Lepidopteran antiviral immunity is solely garnered from studies of invertebrate DNA viruses , leaving many questions regarding how these organisms restrict RNA virus infection . Third , understanding how Lepidopterans combat virus infection may lead to more rationale design of virus-based ‘bioinsecticides’ to control pest species . Here we have identified an unusual resistance of Lepidopteran cells to VSV and SINV infection , and in so doing , we have further developed LD652 cells as a useful tool for the study of virus-host interplay in an important pest species . Our mRNA-seq-based transcriptome data , combined with RNAi knockdowns studies , allowed us to identify host cell immunity factors that restrict RNA virus replication . We also used an RNAi screen to identify A51R as a poxvirus factor that overcomes Lepidopteran immunity to virus infection . A51R stabilizes a subset of host MTs and promotes RNA virus protein stability , which might be explained by its association with Ub . In addition we have shown that A51R is critical for VACV infection in vertebrate cells and is a virulence factor in mice . Our findings raise the intriguing possibility that A51R promotes both DNA and RNA virus infection by disarming host immunity mechanisms conserved in both vertebrates and invertebrates . Despite promiscuous host ranges in both vertebrates and invertebrates , VSV and SINV were completely restricted in LD652 cells . Restriction occurred post-entry , and was overcome either by VACV-co-infection or by blocking host cell transcription . Using RNAi we have identified several host-cell factors that restrict RNA viruses in Lepidopteran cells . These include; ( i ) homologs of the antiviral RNAi factors AGO2 and Dicer-2 ( Kingsolver et al . , 2013 ) , ( ii ) Relish , a homolog of vertebrate NF-κB family members of antimicrobial transcription factors ( Dushay et al . , 1996 ) , ( iii ) a homolog of ‘Effete’ an E2 Ub-conjugating enzyme ( Treier et al . , 1992 ) , and finally ( iv ) several genes encoding Ub . A role for AGO2 and Dicer-2 in Lepidopteran restriction to RNA virus replication is consistent with the deep conservation of RNAi-based antiviral immunity in vertebrates and invertebrates ( Kingsolver et al . , 2013; Li et al . , 2013; Maillard et al . , 2013 ) . Indeed , transient expression of the nodamura virus B2 protein , a well-characterized inhibitor of both invertebrate and vertebrate RNAi pathways ( Sullivan and Ganem , 2005 ) , also rescues VSV and SINV replication in multiple Lepidopteran cell lines ( DBG and CCM , unpublished data ) . Thus , Lepidopteran cells provide a new system to investigate the mechanism of antiviral RNAi . These studies will likely complement other RNA virus-invertebrate host models such as the Orsay virus-Caenorhabditis elegans system , in which RNAi has been shown to restrict RNA virus replication ( Felix et al . , 2011 ) . We are currently conducting deep-sequencing of small RNAs to further explore how Lepidopteran RNAi pathways respond to RNA virus infection . In Drosophila , the NF- κB homolog Relish is a component of the IMD pathway which is stimulated by receptor-dependent recognition of bacterial peptidoglycan ( Kleino and Silverman , 2014 ) . Pathway stimulation results in phosphorylation and cleavage of Relish , which then translocates to the nucleus to activate transcription of antimicrobial factors ( Silverman et al . , 2000 ) . Activation of Relish requires a series of ubiquitination reactions that are dependent on a complex of Ub-conjugating enzymes that includes Effete ( Paquette et al . , 2010 ) . The finding that homologs of Relish , Effete and Ub all function to restrict VSV and SINV replication in L . dispar cells suggests that an IMD-like pathway protects against virus infection in Lepidoptera . Our findings are consistent with recent reports that the IMD pathway plays an antiviral role in Diptera ( Avadhanula et al . , 2009; Costa et al . , 2009 ) . However , the mechanism by which viruses are recognized by this pathway is still unknown . While it is tempting to speculate that the Ub-encoding transcripts identified here promote viral immunity via Relish activation , Ub has a role in many cellular processes and therefore may be required for a variety of immune functions ( Oudshoorn et al . , 2012 ) . Interestingly , several natural viral pathogens of Lepidoptera encode Ub-like molecules , suggesting that Ub may be required for their life cycle ( Barry et al . , 2010 ) . On the other hand , viruses may also benefit from inhibiting or reversing ubiquitination . For example , baculovirus-encoded Ub homologs can inhibit Ub chain elongation ( Haas et al . , 1996 ) , and several viruses encode potent deubiquitinating enzymes ( Lindner , 2007 ) . Inhibiting or reversing Ub reactions might block the activation of the IMD pathway and/or may prevent the direct targeting of viral proteins for destruction by the host UPS . Future studies will be aimed at dissecting the roles of these L . dispar host factors in restricting virus replication . We were surprised by the rescue of VSV and SINV replication in LD652 cells by VACV , despite the fact that VACV replication is abortive in these cells ( Li et al . , 1998 ) . To our knowledge , this is the first example of heterologous virus rescue by a vertebrate virus in an invertebrate host . In mammalian cell culture , VACV also enhances VSV replication , but the VACV-encoded IFN response antagonists E3L and B18R are the major determinants of this rescue ( Le Boeuf et al . , 2010; Shors et al . , 1998 ) . Since the IFN response appears to be vertebrate-specific ( Mukherjee et al . , 2014 ) , it was perhaps not surprising that we failed to observe any effect of E3L or B18R deletion on VACV-dependent VSV rescue in L . dispar cells ( data not shown ) . Instead , we identified the previously uncharacterized VACV gene product A51R as a major determinant of VACV-mediated RNA virus rescue . Additional VACV proteins are likely to contribute to suppression of antiviral activity in L . dispar cells , since both VSV and SINV replication were partially rescued by co-infection with the ΔA51R strain . This might explain why newly synthesized , radiolableled VSV N proteins were undetectable in single infections yet at relatively abundant ( and similar ) levels during co-infection with either VACV-WR or ΔA51R strains , despite the lack of N protein accumulation in the latter co-infection . Thus , the highly restrictive nature of VSV and SINV infections in L . dispar cells , in combination with our sensitive LUC-based detection methods , may prove useful in identifying additional novel immunomodulators encoded by VACV and other viruses . Although our understanding of how A51R promotes RNA virus replication is incomplete there are several intriguing clues . First , our analysis indicates that A51R-mediated rescue occurs at a step after RNA virus entry . Second , the findings that A51R promotes RNA virus infection in multiple Lepidopteran ( but not Dipteran ) cells and promotes VACV infection in vertebrate hosts , suggests that A51R overcomes immunity mechanisms that are shared between Lepidopterans and mammals but absent or easily antagonized in Dipteran insects . Third , the suppression of host restriction to RNA virus replication and the association with MTs are both conserved features of A51R homologs from disparate vertebrate poxviruses , suggesting that both immunomodulatory and MT-stabilization functions may be important for poxvirus replication . This idea is supported by the reduced replication of the ΔA51R strain in vertebrate cell culture and in mice . Finally , the association of A51R with Ub and the A51R-dependent VSV protein stabilization we observed , along with the aforementioned roles of the UPS in restricting RNA virus replication , point to a role for A51R in usurping or inhibiting host Ub machinery . A51R proteins localize with a subset of host MTs in both invertebrate and vertebrate cells , often forming tubulin aggregates and filament-like structures that are resistant to MT depolymerizing agents . These findings suggest a close interaction of A51R with MT structures , and consistent with this , A51R co-sedimented with MTs in MT-pelleting assays ( DBG and CCM unpublished data ) which may explain our difficulty in recovering soluble A51R in immunoprecipitation studies . Importantly , these nocodazole-resistant MT structures are absent in ΔA51R infections and can be formed by expression of A51R in the absence of other poxvirus proteins . Furthermore , the truncated VACV-MVA A51R protein fails to form filamentous , nocodazole-resistant MTs . Thus , the presence or absence of functional A51R could explain why the nocodazole-resistant ‘MT pieces’ previously observed in VACV-infected cells were not observed in uninfected or VACV-MVA-infected cells ( Ploubidou et al . , 2000; Schepis et al . , 2006 ) . Our results suggest that A51R is the elusive MT-stabilizing factor encoded by VACV . When we initially observed an inability of VACV-MVA A51R to both rescue RNA virus replication and stabilize MTs , we hypothesized that stabilization of MTs may be the mechanism by which A51R promotes virus replication . Previous studies implicate MTs in promoting VSV and SINV gene expression ( Qiu et al . , 1998; Heinrich et al . , 2010 ) , and tubulin has been shown to promote VSV RNA polymerase activity in vitro ( Moyer et al . , 1986 ) . Furthermore , VSV transport to the cell surface is reported to be MT-dependent ( Das et al . , 2006 ) . However , a recent study found no effect of MT depolymerization on VSV and SINV replication ( Matthews et al . , 2013 ) . Thus the role of MTs in VSV and SINV replication remains unresolved . Considering the very early restriction of RNA virus gene expression in LD652 cells , it seems unlikely that A51R overcomes host restriction by promoting RNA virus transport on stabilized MTs . Furthermore , our studies with the A51RR321A mutant indicate that RNA virus rescue is not fully mediated by MT association or stabilization and therefore must require an additional function encoded by A51R . This finding however , does not rule out the possibility that MT-association is required for A51R immunosuppressive function or stability . It is possible that A51R subverts host antiviral responses partly by perturbing MT-dependent transport of host factors and/or organelles . Previously it has been shown that VACV-WR infection causes redistribution of endosomes from dispersed cytosolic locations to large perinuclear clusters , suggesting that outward , MT-dependent endosomal trafficking might be impaired during VACV infection ( Schepis et al . , 2006 ) . This endosome clustering phenomenon requires early gene expression and one or more activities lacking in VACV-MVA ( Schepis et al . , 2006 ) . A51R fits these criteria , and thus could be the VACV factor that induces endosome clustering , perhaps to overcome a host mechanism that could otherwise trap viruses in endosomes . It is interesting to note that even at relatively late times post-infection ( e . g . , 24 hr ) we detected punctate immunofluorescence staining patterns for VSV N and SINV E1 virion-associated proteins inside LD652 cells . It is possible that these punctate structures represent virions trapped inside endosomes as both VSV ( Mire et al . , 2010 ) and SINV ( DeTulleo and Kirchhausen , 1998 ) can utilize endocytic modes of entry . It will be interesting in the future to determine if the expression of wild-type or mutant forms of A51R alter endosomal distribution in L . dispar cells and whether endosomal redistribution correlates with virus replication . Unfortunately , we have not yet found suitable reagents for monitoring endosome distributions in L . dispar . Alteration of MT-dependent trafficking by A51R may help explain why both VACV-MVA ( Meiser et al . , 2003 ) and ΔA51R ( this study ) strains release 2-3-times more EEV particles than does VACV-WR . This phenomenon is particularly striking given the overall 10-fold lower total ( intracellular + extracellular ) virus titers of the ΔA51R strain . The release of EEV is MT-dependent ( Ploubidou et al . , 2000 ) , and EEV is known to elicit neutralizing antibodies ( Smith et al . , 2002 ) . Thus it is possible that A51R attenuates virion transport along MTs , and subsequent EEV release , in order to allow the virus to escape activation of host immune responses ( Schepis et al . , 2006 ) . Interestingly , the Dryvax strain of VACV , which is used for smallpox vaccination , encodes a fragmented A51R gene ( Qin et al . , 2011 ) . These observations , along with the attenuated virulence of ΔA51R strain in vivo , raise the intriguing possibility that loss of A51R function contributes to the induction of poxvirus immunity . It is also possible that MT structures might serve as a scaffold to allow A51R to interact with and sequester host factors , such as Ub . Although we do not yet know specifically how A51R promotes virus protein stability , it is possible that A51R-dependent sequestration of Ub ( or removal of Ub from viral proteins ) acts to overcome a Ub-dependent mechanism that either activates host antiviral protein effectors or directly targets nascent viral proteins for destruction . This latter hypothesis predicts that Lepidopterans cells might have a mechanism to target protein synthesis that is coupled to cytoplasmic transcription . Importantly , inhibition of the UPS in L . dispar cells only enhanced LUC activity when cells were infected with LUC-encoding viruses and not when cells were simply transfected with a LUC expression plasmid . Given that VACV , VSV and SINV are all cytoplasmically-replicating viruses and that plasmid-based transcription of luc mRNA would occur in the nucleus , this hypothesis is not unreasonable . Whether other potential A51R interactors identified here such as myosin II essential light chain or RPS15 play a role in A51R rescue function is currently unknown . Further studies of A51R function will not only help us to understand eukaryotic antiviral immunity but may also provide new strategies to overcome the antiviral responses of Lepidopteran pests , leading to more effective biocontrol agents . Indeed , the unexpected uncoupling of RNA virus restriction by A51R has provided an exciting new arena to explore virus-host interactions .
X-gal , nocodazole , vincristine , PYR-41 , and MG132 were obtained from Sigma-Aldrich ( St . Louis , Mo ) and dissolved in dimethylsulfoxide ( DMSO ) . AraC ( Sigma ) was dissolved in sterile water . Recombinant IFN-α solution was from Sigma . X-glu ( Clontech , Palo Alto , CA ) was dissolved in DMSO . Compounds were diluted to their final concentration in cell culture medium or in a 1:1 mixture of 2 × Dulbecco's minimal essential medium ( DMEM; Invitrogen , Carlsbad , CA ) and 1 . 7% agar ( X-gal; X-glu ) immediately prior to use . African Green Monkey kidney cells ( BSC-40 ) , baby hamster kidney ( BHK ) , and human embryonic lung ( HEL ) cells were obtained from American Type Culture Collection ( ATCC ) . BSC-40 and HEL cells were maintained in MEM containing 10% Fetal bovine serum ( FBS ) . BHK cells were maintained in DMEM containing 10% FBS . L . dispar-derived LD652 cells were obtained from Dr Basil Arif ( Natural Resources Canada , Canada ) and were maintained in a 1:1 mixture of Ex-Cell 420 ( Sigma ) and Graces insect medium ( Invitrogen ) that also contained 10% FBS . L . dispar-derived embryonic ( LdEP ) cells were obtained from the United States Department of Agriculture Research Service and were maintained in Ex-Cell 420 medium supplemented with 3% FBS . S . frugiperda-derived Sf9 cells were obtained from Invitrogen and were maintained in Sf-900 II medium ( Invitrogen ) supplemented with 10% FBS . M . sexta-derived GV-1 cells were obtained from Dr Que Lan ( University of Wisconsin ) and were maintained in Graces insect medium supplemented with 10% FBS . D . melanogaster-derived DL1 cells were obtained from Dr Sara Cherry ( University of Pennsylvania ) and were maintained in S2 medium ( Invitrogen ) containing 10% FBS . All medium also contained 1% antibiotic-antimycotic ( Invitrogen ) . Vertebrate cells were incubated at 37°C in a 5% CO2 atmosphere and invertebrate cells were incubated at 27°C in normal atmosphere . VACV-WR , VACV-COP , and VACV-MVA were obtained from Dr Anuja Mathew ( University of Massachusetts Medical School ) . The A5L-GFP VACV strain ( Carter et al . , 2003 ) was obtained from Dr Geoff Smith ( University of Cambridge , United Kingdom ) . VSV-GFP ( Kato et al . , 2005 ) was obtained from Dr Hiroki Kato ( Kyoto University , Japan ) . VSV-LUC ( rVSVluc; Cureton et al . , 2009 ) was obtained from Dr Sean Whelan ( Harvard Medical School ) . SINV-GFP ( TE/5’2J/GFP-3XFlag; Cristea et al . , 2006 ) was obtained from Dr Margaret MacDonald ( Rockefeller University ) and SINV-LUC [TRNSVluc; Cook and Griffin , 2003] was obtained from Dr Dianne Griffin ( Johns Hopkins University ) . VACV stocks were amplified in BSC-40 cells with the exception of VACV-MVA which was passaged in BHK cells . VACV strains were titrated on BSC-40 cells by plaque assay as previously described ( Gammon et al . , 2010 ) . VSV and SINV were amplified using low MOI infections in BHK cells . Viruses were collected from culture supernatants by ultracentrifugation ( 25 , 000 rpm , 2 hr , 4°C ) and titrated by plaque assay on BSC-40 cell monolayers overlayed with a 1 . 5% low-melting point agarose ( Invitrogen ) . All MOIs reported for vertebrate viruses ( VACV , VSV , SINV ) refer to those calculated from mammalian cell-based plaque assays . Virus infections were carried out for 2 hr in either serum-free DMEM at 37°C ( vertebrate cell infections ) or in Sf-900 II serum free media at 27°C ( invertebrate cell infections ) . After 2 hr , inocula were replaced with normal growth medium . When titering VSV/SINV from cultures co-infected with VACV , AraC ( 50 μg/ml ) was included in the agarose overlays . Where indicated , AraC was added to cultures at the indicated doses 2 hpi in the appropriate growth medium or VACV virions were heat-inactivated prior to infection by incubation at 55°C for 1 . 5 hr ( Dales and Kajioka , 1964 ) . Unless otherwise noted , invertebrate cells were infected at MOIs of 100 ( VACV ) or 10 ( VSV/SINV ) . At these MOIs ∼100% of cells are infected with VACV by 48 hpi ( Li et al . , 1998 ) . Total RNA was collected from cell cultures in TRIzol Reagent ( Life Technologies ) . The aqueous layer was isolated using phase lock columns ( 5 Prime Gaithersburg , MD ) and RNA was isopropanol-precipitated and resuspended in nuclease-free water . Genomic DNA was removed using a Turbo DNA-free kit ( Invitrogen ) . First-strand cDNA was synthesized using 0 . 5–1 μg of purified total RNA , gene-specific primers , and Superscript III Reverse Transcriptase ( Invitrogen ) , according to the manufacturer’s recommendations . PCR was performed with Takara Ex-Taq ( Clontech , Mountain View , CA ) . PCR cycling conditions were: 94°C for 3 min , followed by 94°C for 30 s , 50°C for 30 s , 72°C for 1 min for 36 cycles . PCR products were separated on 2% agarose gels containing ethidium bromide and images were captured using an ECII Darkroom System with Labworks acquisition software ( version 4 . 0 . 0 . 8; Bioimaging Systems , Upland , CA ) . VSV ( + ) -sense transcription was analyzed by RT-PCR using primers: 5′-ATGTCTACAGAAGATGTA-3′ & 5′-TAATATATAATAGGTGATCTGAGAATTATAGGGTC-3′ ( Wilkins et al . , 2005 ) . SINV ( − ) -sense transcript levels were analyzed by RT-PCR using primers: 5′-TAGACAGAACTGACGCGGACGT-3′ & 5′-TCCATACTAACTCATCGTCGATCTC-3′ ( Campbell et al . , 2008 ) . RT-PCR amplification of L . dispar actin transcripts was performed with primers: 5′-GGGACAGAAGGACTCGTACG-3′ & 5′-GCCTTAGGGTTGAGAGGAGC-3′ ( Chen et al . , 2003 ) . Total RNA from LD652 cells was submitted to the Georgia Genomics Facility ( Athens , GA ) for library preparation and SE100 sequencing . TruSeq RNA libraries were prepared from total RNAs and rRNA depletion was performed . SE100 reads were sequenced from resulting libraries using an Illumina HiSeq 1000 instrument . This sample yielded 238 , 502 , 918 raw reads , which were subjected to quality control procedures implemented using the FASTX-toolkit ( Provided by Dr G Hannon , Cold Spring Harbor Laboratories , NY ) . Specifically , artifact reads were eliminated , and end sequences whose Phred scores corresponded to an error rate exceeding 1% were clipped . Only end-trimmed reads of 36 bases or longer were retained , and these were also required to have not less than 90% of their bases possessing a Phred score of 21 or higher . Post cleaning , the sample had 122 , 295 , 683 sequences ( 12 , 148 , 224 , 605 bases ) . These were assembled into 115 , 739 putatively unique transcripts ( PUTs ) using the Trinity assembly program ( Grabherr et al . , 2011 ) and compared with the 4 September 2013 version of NCBI NR using Blastx ( Altschul et al . , 1997 ) . PUTs were partitioned into gold ( 7055 entries ) , silver ( 6353 ) and bronze ( 8251 ) tiers on the basis of alignment quality as described in Sparks et al . ( 2013 ) . The gold and silver tiers collectively defined a set of 13 , 408 unique PUTs mapping to 7593 unique NCBI NR records . These gold and silver tier PUTs were used to identify potential immunity-related L . dispar transcripts and for the design of dsRNAs for RNAi . The cDNA sequences of the L . dispar transcripts identified in Figure 2C , D are available in Supplementary file 1 . dsRNAs ( ∼400 bp in length ) were transcribed in vitro using the Megascript RNAi kit ( Life Technologies ) . Templates were generated by RT-PCR reactions using gene-specific primers tailed at the 5′ end with the T7 promoter sequence ( TAATACGACTCACTATAGGG ) . Sequences of primers used to make dsRNA targeting L . dispar mRNAs are listed , along with target sequences in Supplementary file 1 . Sequences of primers used to generate dsRNA against VACV targets are listed in Supplementary file 2 . dsRNA ( ∼1 μg ) was transfected into 105 LD652 cells using Cellfectin II ( Invitrogen ) in Sf-900 II media according to the manufacturer's guidelines . DNA transfections ( ∼1 μg ) were also performed using Cellfectin II in Sf-900 II media . For VACV RNAi , after 5 hr the transfection media was replaced with virus inocula for 2 hr and then with regular growth medium . For L . dispar RNAi , after 5 hr the transfection medium was replaced with regular growth medium . Transfected cells were then challenged with VSV or SINV 24 hr post-transfection . RNAi transfection conditions resulted in >80% knockdown of either VACV or host transcripts ( data not shown ) . At the indicated times post-infection/transfection , cells were briefly washed in phosphate buffered saline ( PBS ) collected by centrifugation ( 2000 rpm , 10 min , 4°C ) and lysed in reporter lysis buffer ( Promega , Madison , WI ) according to the manufacturer's guidelines . Lysates were spotted to 96-well dishes , mixed with Luciferase Assay Reagent ( Promega ) and arbitrary light units ( LU ) were measured using an Envision 2102 Multilabel Reader with Wallac EnVision Manager software ( version 1 . 12; PerkinElmer , Waltham , MA ) . All experiments were performed at least three times . For PCR amplifications for use in downstream cloning and gene expression studies , iproof DNA polymerase ( Bio-Rad , Hercules , CA ) was used . All final expression constructs were confirmed by DNA sequencing . Cellfectin II and Lipofectamine 2000 were used according to the manufacturer's guidelines for transfection of invertebrate and vertebrate cells , respectively . Invertebrate cell transfections were performed in SF-900 II medium and vertebrate cell transfections were performed in Opti-MEM for 5–6 hr after which the transfection medium was replaced with normal growth media . Where indicated , normal growth media containing nocodazole ( 20 µM ) or vincristine ( 4 μM ) was used to replace transfection media . To express genes in invertebrate cells , either a modified p166 vector ( Lin et al . , 2001 ) or the pIZ/V5-His vector ( Invitrogen ) was used for transient expression . To facilitate cloning , the p166 vector was modified by introduction of a new multiple cloning site ( MCS ) between the BamHI and XbaI sites using linker ligation and primers: 5′-GATCCCCCGGGACCGCGGCAAGTCGACCAATCGCGAAAGGAATTCAAGTTAATTAAT-3′ & 5′-CTAGATTAATTAACTTGAATTCCTTTCGCGATTGGTCGACTTGCCGCGGTCCCGGGG-3′ . A Flag-tagged gfp gene was amplified from pEGFP-C3 vector ( Clontech ) DNA using primers: 5′-CCGCGGATGGATTATAAGGATGATGATGATAAGATGGTG-3′ & 5′-TTAATTAATTACTTGTACAGCTCGTCCATGCC-3′ and cloned into the SacII/PacI sites of p166 . A plasmid encoding a codon-optimized , Flag-tagged VACV-WR A51R gene was synthesized by Invitrogen and used for cloning into the p166 vector using SacII/PacI sites . Plasmids encoding codon-optimized , Flag-tagged A51R genes from CPXV ( strain Brighton ) , ECTV ( strain Moscow ) , DPXV ( strain W-1170-84 ) , YLDV ( strain Davis ) , and MYXV ( strain Lausanne ) were also synthesized and these genes were digested from these initial vectors and ligated to SacII/PacI-cut p166 . To express Flag-A51R , and Flag-tagged forms of CPXV/ECTV/DPXV/YLDV/MYXV A51R proteins in vertebrate cells , each p166 vector was SacII/PacI digested and inserts were cloned into pCDNA3 ( Invitrogen ) that had been modified to encode the same MCS as the p166 vector using linker ligation and primers described above . A QuikChange II Site-Directed Mutagenesis Kit ( Agilent Technologies , Santa Clara , CA ) and primers: 5′-GTAACCGTGACTACATCCCCGGAGCTCGTGGTTACTCCTACTAC-3′ & 5′-GTAGTAGGAGTAACCACGAGCTCCGGGGATGTAGTCACGGTTAC-3′ were used to generate a VACV Flag-A51R p166 vector encoding the R321A amino acid substitution . To express the Severe Acute Respiratory Syndrome Coronavirus PLpro deubiquinase and its catalytically-inactive mutant form ( PLproC112A ) ( Barretto et al . , 2005 ) in invertebrate cells , pCDNA3 vectors encoding the wild-type and mutant forms ( kind gifts from Dr Susan Baker , Loyola University of Chicago , IL ) were digested with HindIII/EcoRI enzymes to isolate the genes which were then cloned into the corresponding sites in the pIZ/V5-His vector . To express LUC in invertebrate cells , the firefly luc gene from pGL3-Basic ( Promega ) was first amplified using primers: 5′-GGATCCATGGAAGACGCCAAAAACA-3′ & 5′-TCTAGAATTACACGGCGATCTTTCCG-3′ , Topo-cloned , and then digested out of the Topo vector with BamHI/XbaI and cloned into BamHI/XbaI-digested p166 , creating p166-LUC . Rabbit anti-Flag , mouse anti-Flag , mouse anti-tubulin , FITC-conjugated mouse anti-tubulin , and mouse anti-actin were from Sigma . Secondary antibodies used in IF microscopy were raised in donkey or goat and conjugated to Alexa 488 , 568 , or 647 were commercially obtained ( Invitrogen ) . In some cases , FITC-conjugated anti-tubulin antibodies were used after secondary antibody staining with Alexa 568 and 647 antibodies for 4-color imaging . Rabbit anti-tubulin antibody was from Cell Signaling Technology ( Danvers , MA ) . Mouse anti-LUC and rabbit anti-LUC antibodies were from Invitrogen and Abcam ( Cambridge , MA ) , respectively . Mouse anti-VACV I3L and F4L antibodies ( Gammon et al . , 2010 ) were from Dr David Evans ( University of Alberta , Canada ) . Mouse anti-VACV A27L and L1R antibodies were obtained through the NIH Biodefense and Emerging Infections Research Resources Repository ( NIAID , NIH ) . Rabbit anti-VACV antibody was from ViroStat ( Portland , ME ) . Mouse antibodies against VSV nucleocapsid and matrix proteins ( Lefrancois and Lyles , 1982 ) were from Dr Douglas Lyles ( Wake Forest School of Medicine , Winston–Salem , NC ) . Mouse anti-SINV E1 protein antibodies were from Dr Dianne Griffin ( Johns Hopkins Bloomberg School of Public Health , Baltimore , MD ) . Protein extracts for immunoblots were prepared from cell cultures by lysing cells in either reporter lysis buffer as described above or in NP-40 buffer containing 150 mM NaCl , 20 mM Tris ( pH 8 . 0 ) , 1 mM EDTA , and 0 . 5% NP-40 along with phenylmethylsulfonyl fluoride ( 100 μg/ml ) and protease inhibitor tablets ( Roche , Indianapolis , IN ) as described ( Gammon et al . , 2010 ) . Lysates were subjected to SDS-PAGE and subsequently blotted with appropriate primary antibodies after transfer to nitrocellulose membranes . Membranes were scanned using an Odyssey Infrared Imaging System ( Li-COR Biosciences , Lincoln , NE ) . For MuDPIT experiments , LD652 cell lysates were prepared in NP-40 buffer after 48 hr of transfection with either empty p166 vector or vector encoding Flag-A51R ( control ) . Equal quantities of each lysate were subjected to immunoprecipitation in NP-40 buffer containing protease-inhibitor cocktail ( Roche ) using anti-Flag antibodies and protein G Dynabeads ( Invitrogen ) . Immunoprecipitates were eluted from Dynabeads using glycine-HCl ( pH 2 ) treatment for 10 min and were subsequently neutralized with neutralization buffer ( 0 . 5 M Tris–HCl , 1 . 5M NaCl ) . MuDPIT analyses of these eluted fractions were performed using an Accela HPLC and a Thermo LTQ connected to a homemade electrospray stage . Protein identification was performed with Integrated Proteomics Pipeline–IP2 ( Integrated Proteomics Applications , Inc . , San Diego , CA . http://www . integratedproteomics . com/ ) . Tandem mass spectra were extracted from raw files using RawExtract 1 . 9 . 9 ( McDonald et al . , 2004 ) , searched against the Bombyx mori UniprotKB proteome ( as well as Flag-A51R sequence ) , reversed sequenced using ProLuCID , ( Peng et al . , 2003 ) , and peptide candidates were filtered using DTASelect with the parameters -p 1 -y 1 --trypstat --sfp 0 . 01 -in Tabb et al . ( 2002 ) and McDonald et al . ( 2004 ) . Radiolabeling of LD652 cell cultures ( ∼105 cells/well ) was performed 72 hr post-infection with the indicated strains by incubation of cultures with 100 μCi/well of [35S]methionine [in methionine-free Sf900-II medium ( Invitrogen ) ] for 2 hr . After the 2 hr pulse , [35S]methionine-containing medium was replaced with normal growth medium and lysates were then extracted either immediately ( T = 0 ) or at the indicated times post-pulse using radioimmunoprecipitation buffer ( 150 mM NaCl , 1 . 0% NP-40 , 0 . 5% sodium deoxycholate , 0 . 1% SDS , and 50 mM Tris , pH 8 . 0 ) . These whole cell extracts were then subjected to IP with anti-VSV N ( 10G4 ) antibodies or used directly in immunoblots to determine total VSV N protein levels . [35S]-labeled VSV N IP complexes were separated by SDS-PAGE and then either dried on Whatman chromatography paper or directly transferred to nitrocellulose membranes . Radiolabeling was detected using either a Bio-Rad Personal Molecular Imager System with Quantity One software or a FLA-5000 Imaging System ( Fujifilm Tokyo , Japan ) equipped with Multi Gauge v3 . 2 software . Half-life determinations were made from non-linear regression analyses using GraphPad Prism v6 software ( La Jolla , CA , USA ) . Live images were captured using a Nikon Eclipse TE200 inverted fluorescent microscope with Spot Software ( Diagnostics Instruments Inc . , Sterling Heights , MI , version 4 . 6 ) . ImageJ v1 . 04g software ( NIH , Bethesda , MD ) was used to set image thresholds and calculate the frequency of GFP-positive cells . BSC-40 cells were split into 24-well dishes containing 12 mm glass coverslips . LD652 cells were resuspended in the growth medium and plated on 12 mm glass coverslips treated with concanavalin A ( Sigma ) solution ( 0 . 5 mg/ml ) for 45 min prior to fixation . Cells were either fixed in 4% paraformaldehyde or methanol . Coverslips were incubated 1 hr in blocking buffer ( 0 . 5% triton X-100 , 1% BSA , in PBS ) prior to antibody staining . Primary and secondary antibodies were diluted in blocking buffer and incubated with rocking for 1 hr at room temperature . After antibody incubation , coverslips were washed 3–4 times with blocking buffer and mounted using ProLong Gold Antifade with DAPI ( Invitrogen ) . Confocal images were captured using a Nikon TE2000-E confocal microscope with MetaMorph v7 . 7 . 4 software ( Molecular Devices , Sunnyvale , CA ) . Point spread functions were calculated using the ImageJ plugin Diffractive PSF 3D plugin ( OptiNav Inc . , Bellevue , WA ) and deconvolved using the Iterative Deconvolve 3D plugin ( OptiNav Inc ) . Video files were generated from confocal microscopy Z-stacks using MetaMorph v7 . 7 . 4 software . Intranasal infections were performed under anesthesia using ketamine/xylazine in saline . 5 week-old female NMRI mice ( Laboratoire Elevage Janvier , Le Genest-ST-Isle , France ) were inoculated with 25 µl of PBS ( mock-infected group ) or with 25 µl of PBS containing virus inoculum at the indicated doses ( PFU ) . Body weight , morbidity and mortality were monitored for 20–28 days . When necessary , animals were euthanized by administering pentobarbital sodium . To determine the extent of viral replication , 5 mice per group were euthanized at day 4 and 6 post-infection , and tissues were collected and processed as previously described ( Duraffour et al . , 2013 ) . Viral loads in tissue homogenates were titrated on HEL cells . Survival curves between VACV-WR and ΔA51R virus-infected groups were compared using log-rank ( Mantel–Cox ) tests and tissue virus titers were compared using unpaired Mann Whitney tests . Statistical tests were performed using GraphPad Prism v6 software . | Viruses can infect species as diverse as bacteria , plants and animals , and once they have infected an organism they hijack its cells to rapidly replicate their own genetic material , which is made of DNA or RNA . Many animals , including insects , have been used as model organisms to investigate viral infections . These studies have , for example , provided insights into how viruses replicate and how they suppress their host's immune system . One insect species that has been used in many virus-host studies is the gypsy moth . This species of moth was accidently introduced into North America from Europe in the late 1800s , and its caterpillars have become a major pest because they destroy hardwood trees and forests . Gypsy moth outbreaks are still a serious problem , but their numbers can be kept in check by using biological control strategies , such as DNA viruses . However , the response of gypsy moths to infection by RNA viruses has not been studied extensively . Gammon et al . now show that , after being infected with one of two different RNA viruses , gypsy moth cells can slow down and eventually halt the replication of the RNA viruses . However , if the gypsy moth cells are also infected with a DNA virus , they lose their ability to restrict the replication of the RNA virus . Gammon et al . discovered that the moth’s immunity to RNA virus infection is disarmed by a protein called A51R from the DNA virus . This protein increases the stability of the proteins in the RNA virus , most likely by stopping the moth from breaking them down . The results of Gammon et al . suggest that it might be possible to use a combination of RNA viruses and the A51R protein to keep the number of gypsy moths in check . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2014 | A single vertebrate DNA virus protein disarms invertebrate immunity to RNA virus infection |
Duplicating chromosomes once each cell cycle produces sister chromatid pairs , which separate accurately at anaphase . In contrast , reduplicating chromosomes without separation frequently produces polytene chromosomes , a barrier to accurate mitosis . Chromosome reduplication occurs in many contexts , including: polytene tissue development , polytene tumors , and following treatment with mitosis-blocking chemotherapeutics . However , mechanisms responding to or resolving polyteny during mitosis are poorly understood . Here , using Drosophila , we uncover two distinct reduplicated chromosome responses . First , when reduplicated polytene chromosomes persist into metaphase , an anaphase delay prevents tissue malformation and apoptosis . Second , reduplicated polytene chromosomes can also separate prior to metaphase through a spindle-independent mechanism termed Separation-Into-Recent-Sisters ( SIRS ) . Both reduplication responses require the spindle assembly checkpoint protein Mad2 . While Mad2 delays anaphase separation of metaphase polytene chromosomes , Mad2’s control of overall mitotic timing ensures efficient SIRS . Our results pinpoint mechanisms enabling continued proliferation after genome reduplication , a finding with implications for cancer progression and prevention .
Regulating mitotic chromosome structure is critical to preventing genomic instability ( Gordon et al . , 2012; Pfau and Amon , 2012 ) . During mitosis , chromatids associate in sister pairs , which facilitates their bi-orientation and subsequent segregation to opposite spindle poles . A frequently occurring and long-recognized departure from this paired chromosome structure occurs when the genome reduplicates without chromatid separation ( hereafter: genome reduplication ) . Following a single extra S-phase , cells frequently form diplochromosomes: four sister chromatids conjoined at centromeres ( White , 1935 ) . A more general term for chromosomes formed by any degree of genome reduplication without chromatid separation is 'polytene' ( Painter , 1934; Zhimulev et al . , 2004 ) . While incompletely understood , it is appreciated that multiple layers of physical connections tightly intertwine the multiple sister chromatids of polytene chromosomes . These connections likely include cohesins ( Cunningham et al . , 2012; Pauli et al . , 2010 ) as well as topological entanglements that can be removed by Condensin II activity ( Bauer et al . , 2012; Smith et al . , 2013; Wallace et al . , 2015 ) . Additionally , recurring regions of DNA under-replication occur between chromatids in some polytene cells ( Beliaeva et al . , 1998; Gall et al . , 1971; Hannibal et al . , 2014; Nordman et al . , 2011; Yarosh and Spradling , 2014 ) whereas DNA replication is more complete in others ( Dej and Spradling , 1999; Fox et al . , 2010 ) . In addition to connections between sister chromatids , another layer of chromosome association - pairing between homologs - also occurs in some polytene cells . This pairing results in polyploid/polytene cells that exhibit only the haploid number of distinct chromosomes ( Metz , 1916; White , 1954 ) . Given these multiple physical connections between polytene chromatids , mitosis in polytene cells is considered 'ill-advised for mechanical reasons' ( Edgar and Orr-Weaver , 2001 ) . Indeed , separation of polytene diplochromosomes at anaphase causes chromosome mis-segregation ( Vidwans et al . , 2002 ) . Given the association of polytene chromosomes with mitotic errors , it is not surprising that these structures are often associated with aberrant development and disease . Polytene chromosomes have been observed in cells from spontaneous human abortions ( Therman et al . , 1978 ) , in muscular dystrophy patients ( Schmidt et al . , 2011 ) , in a variety of tumors ( Biesele and Poyner , 1943; Erenpreisa et al . , 2009; Therman et al . , 1983 ) and can also precede tumor formation in mice ( Davoli and de Lange , 2012 ) . Polytene chromosomes also occur after treatment with currently used anti-mitotic chemotherapeutics such as those that inhibit Topoisomerase II ( Cantero et al . , 2006; Sumner , 1998 ) . Disruption of numerous other processes crucial for mitosis , including spindle formation ( Goyanes and Schvartzman , 1981; Takanari et al . , 1985 ) sister chromatid cohesion ( Wirth et al . , 2006 ) or genome integrity control ( Davoli et al . , 2010 ) also cause genome reduplication and polyteny . Thus , polytene chromosomes , a source of mitotic instability , are a conserved and common outcome of ectopic genome reduplication . To understand how cells adapt the cell cycle machinery to the challenge of segregating the intertwined polytene chromatids found in genome-reduplicated cells , naturally occurring models of this problem can prove useful . Programmed genome reduplication cycles of successive S-phase without M-phase ( endocycles , Edgar et al . , 2014; Fox and Duronio , 2013 see nomenclature ) produce polytene chromosomes in many plant , insect , and mammalian species , including humans ( Zhimulev et al . , 2004; Zybina et al . , 1996 ) . However , many cells with programmed genome reduplication do not subsequently divide , preventing study of how nature has circumvented the issue of segregating polytene chromosomes . In contrast , we previously demonstrated that rectal papilla ( hereafter: papillar cells ) , ion-absorbing structures in the Drosophila hindgut , are built entirely by mitosis of endocycled cells ( Fox et al . , 2010; Schoenfelder et al . , 2014 ) . Surprisingly , we never observed polytene chromosomes in hundreds of papillar metaphases ( Fox et al . , 2010; Schoenfelder et al . , 2014 ) , suggesting papillar cells are programmed to either avoid or eliminate polyteny and its associated mitotic defects . Interestingly , previous studies suggest that polyteny can be at least partially undone without anaphase in both normal and tumorous tissue ( Dej and Spradling , 1999; Grell , 1946; Levan and Hauschka , 1953 ) . Thus , in some cases , polyteny may be actively regulated or eliminated . Taken together , the potential negative impact of genome reduplication on mitotic chromosome structure is clear . However , the responses that enable either developing or tumorous cells to continue dividing after reduplication , despite profound chromosome structure changes , remain unclear . Here , using Drosophila tissue models of both ectopic and naturally occurring genome reduplication , we uncover two distinct cellular responses to reduplicated chromosomes . Both reduplication responses require the conserved spindle assembly checkpoint ( SAC ) protein Mad2 , which inhibits the Anaphase Promoting Complex to both delay anaphase in response to unattached or tensionless kinetochores and to also regulate overall mitotic timing from nuclear envelope breakdown ( NEBD ) to anaphase onset ( London and Biggins , 2014; Musacchio , 2015 ) . In reduplicated cells that retain polytene chromosomes at metaphase , we show Mad2 is involved in a SAC wait-anaphase response . This anaphase delay does not fully prevent the mitotic errors and the resulting aneuploidy associated with mitosis of polytene chromosomes , but it substantially reduces apoptosis , tissue malformation , and organismal death . In contrast to this wait-anaphase response , we also define a second response in reduplicated cells that actively eliminates polyteny before anaphase . In this response , polytene chromosomes undergo a dynamic , spindle-independent process we term Separation Into Recent Sister chromatid pairs ( SIRS ) , which eliminates any trace of polyteny before anaphase . Unlike mitosis with metaphase polytene chromosomes , mitosis with SIRS does not trigger a Mad2-dependent anaphase delay . Yet , we find Mad2 promotes efficient SIRS by allowing sufficient time between nuclear envelope breakdown and anaphase , which allows polytene chromosomes to separate into conventional mitotic sister chromatid pairs . Our results therefore define two distinct responses to reduplicated chromosomes , each of which depends on a distinct Mad2 response .
We first ectopically induced genome reduplication in proliferating tissues of developing larvae by transiently re-programming mitotic cycles to endocycles . fizzy-related ( fzr , mammalian Cdh1 ) plays a conserved role in endocycles by targeting the anaphase promoting complex to destroy the mitotic Cyclins A , B , and B3 ( Larson-Rabin et al . , 2009; Sigrist and Lehner , 1997 ) . fzr overexpression was previously shown to transform mitotic cycles into endocycles ( Sigrist and Lehner , 1997 ) . To transiently induce endocycles , we used a brief heat shock ( HS ) pulse to express ectopic fzr ( HS>fzr , Figure 1A ) . Using the cell cycle marker system Fly-FUCCI ( Zielke et al . , 2014; Figure 1B ) we find that pulsed fzr overexpression temporarily eliminates expression of the S/G2/M mRFP-CyclinB reporter in wing imaginal disc cells ( Figure 1C vs . C’ , D ) . This same population of mRFP-CyclinB-negative cells continues to express the G2/M/G1 GFP-E2F1 reporter , but in greater proportion ( Figure 1C vs . C’ , D ) . Together , these data suggest HS>fzr promotes G1 accumulation ( 91% of cells compared to 36% in controls , Figure 1B–D ) . To test whether this G1 accumulation is due to direct conversion of G2 cells to G1 , as opposed to an acceleration of the cell cycle through G2/M , we stained for the mitotic marker Phospho-Histone H3 at several time points after pulsed fzr expression . For up to 7 hr after fzr overexpression , there is essentially no mitosis in the wing imaginal disc , whereas wing cells in heat shocked wild type flies continue to divide after heat shock ( Figure 1—figure supplement 1A , B ) . Based on previous studies of fzr function and our FUCCI and Phospho-Histone H3 data , we conclude that pulsed fzr expression converts G2 cells to a G1 state by eliminating mitotic cyclins ( Figure 1A ) . 10 . 7554/eLife . 15204 . 003Figure 1 . Induced genome reduplication in wing progenitors leads to polytene diplochromosomes and aneuploidy . ( A ) A model for the cell cycle progression following fizzy-related ( HS>fzr ) overexpression in a mitotically cycling tissue . Cells progress directly from G2 into G1 without an intervening mitosis , resulting in an additional S-phase . ( B ) A diagram depicting the Fly-FUCCI system in each stage of the cell cycle , and representative images of wing imaginal disc cells in each cell cycle state . GFP-E2F11-230 ( green ) is nuclear during G1 and G2 and fills the cell during mitosis . RFP-CycB1-266 ( magenta ) is cytoplasmic during S-phase and G2 and fills the cell during mitosis . ( C ) Representative micrographs of the wing imaginal disc pouch expressing UAS Fly-FUCCI under the control of engrailed-Gal4 in the absence of HS>fzr expression ( No HS , C ) as well as +2 hr ( C’ ) and +10 hr ( C’’ ) after a 60-min heat shock to induce HS>fzr expression . GFP-E2F11-230 is in green , RFP-CycB1-266 is in magenta . ( D ) The percentage of cells in G1 , S , G2 , and M based on Fly-FUCCI expression prior to HS>fzr expression ( No HS ) , +2 hr and +10 hr after a 60-min heat shock to induce fzr expression . Stacked bars represent mean plus standard error of the mean ( +S . E . M . ) , ***p<0 . 001 , NS = p>0 . 05 , t-test . Data are an average of three replicates with at least 5 animals per replicate and at least 50 cells counted per animal . ( E ) Representative karyotypes from a mitotic HS>fzr wing imaginal disc cells without heat shock . Chromosomes are pseudocolored according to each chromosome type and numbered . Prior to HS>fzr expression cells display a normal diploid karyotype . Tissue was incubated in colcemid for 30 min to enrich for mitotic cells . ( F ) Representative karyotypes from mitotic HS>fzr wing imaginal cells 10 hr after a 60-min heat shock . Chromosomes are pseudocolored according to the type as in Figure 1E . Transiently , closely aligned polytene chromosomes are seen when chromosomes first condense after genome reduplication ( F ) . Asterisks indicate the 2 groups of homologous centromeres of the X-chromosome . Diplochromosomes , characterized by the attachment of four centromeres of each sister chromatid ( F’ see inset ) , are seen at the first metaphase after genome reduplication . Tissue was incubated in colcemid for 30 min to enrich for mitotic cells . ( G ) Representative karyotype of a mitotic HS>fzr cell 24 hr after a 60 minheat shock , colored according to type as in Figure 1E . Aneuploid cells are observed at 24 hr after heat shock , during the second metaphase after genome reduplication , which follows the division of diplochromosomes . Tissue was incubated in colcemid for 30min to enrich for mitotic cells . ( H ) The percentage of wing imaginal disc karyotypes classified as euploid/diploid , euploid/tetraploid , euploid/diplo-tetraploid , or aneuploid/tetraploid prior to heat shock ( No HS ) , or +10 hr , +24 hr , or +120 hr after a 60-min heat shock . Stacked bars represent Mean ( +S . E . M . ) , ***=p<0 . 001 , NS = p>0 . 05 , t-test . Data are an average of 3 replicates with at least 50 karyotypes per replicate . ( I ) Representative time-lapse of a diploid wing imaginal disc cell dividing prior to HS>fzr expression ( No HS ) and a tetraploid cell with diplochromosomes dividing 10 hr after a 60-min heat shock to induce HS>fzr expression ( HS +10 hr ) . Yellow arrowhead shows a single lagging kinetochore . Red arrows highlight a single diplochromosome that segregates its chromatids in a 3:1 fashion . Cenp-C-Tomato showing kinetochores in cyan , His2av-GFP showing DNA in magenta . Time represents min from the last frame prior to anaphase . ( J ) The percentage of lagging chromosomes in diploid cells , in tetraploid cells with diplochromosomes ( 4N Diplo ) , and in tetraploid cells without diplochromosomes ( 4N ) after HS>fzr expression . Bars represent averages ( +S . E . M . ) between animals with at least five animals per condition . ***p<0 . 001 , NS = p>0 . 05 , t-test . ( K ) A model for a cell cycle that results in aneuploid daughter cells showing only the two homologs of a single chromosome for simplicity . The two homologs are shown in black and gray with a red centromere . Chromatids are replicated in S-phase and then reduplicated following a heat shocked induced endocycle . This results in polytene chromosomes . Diplochromosomes are seen as the genome-reduplicated cells progress into metaphase . At anaphase , diplochromosome segregation frequently produces lagging chromatids , which can segregate erroneously resulting in aneuploidy . Scale bars represent 5 μm , except in insets in F’ and G where it represents 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 00310 . 7554/eLife . 15204 . 004Figure 1—figure supplement 1 . Supporting data regarding the effect of HS>fzr on imaginal discs and brains . ( A ) Representative micrographs of 3rd instar wing imaginal discs from HS>fzr animals stained for Phospho-HistoneH3 ( PH3 , green ) and DAPI ( magenta ) prior to heat shock ( No HS , A ) as well as +2 hr ( A’ ) and +10 hr ( A’’ ) after a 60-min heat shock . ( B ) Bee-swarm plot depicting the number of Phospho-HistoneH3 ( PH3 ) positive cells per animal prior to heat shock ( No HS ) , immediately after heat shock ( 2–7 hr post ) or 10 hr after a 60-min heat shock for HS>fzr ( dark blue circles ) and wild type ( w1118 , light blue triangles ) . N ≥ 6 animals per time point . ( C ) The proportion of 3rd instar larval brain karyotypes prior to heat shock ( No HS ) , or +10 hr , +24 hr and +120 hr after a 60 min heat shock , classified as euploid/diploid , euploid/polyploid , euploid/diplo-polyploid , or aneuploid/polyploid . Stacked bars represent Mean + Standard Error of the Mean ( +S . E . M . ) , **p < 0 . 01 , ***p< 0 . 001 , t-test , data are an average of 3 replicates with at least 50 karyotypes per replicate . ( D ) The proportion of aneuploidies caused by 3:1 non-disjunction or 4:0 non-disjunction of wing imaginal disc diplochromosome divisions , inferred from subsequent divisions . Information is divided into gains and losses . N = 52 aneuploidies . ( E ) Representative time-lapse of a tetraploid cell without diplochromosomes dividing 24 hr after a 60-min heat shock to induce HS>fzr expression . Cenp-C-Tomato showing kinetochores in cyan , His2av-GFP showing DNA in magenta . Time represents min from the last frame prior to anaphase . ( F ) Representative micrographs depicting a cell within a wing-disc stained for Phospho-HistoneH3 ( PH3 , magenta ) and γ–tubulin ( green ) before ( no HS ) and after ( HS+10 hr ) a 60min heat shock . HS>fzr cells occasionally show evidence of centrosome amplification . ( F’ ) Quantification of the frequency of counting 2 centrosomes or greater than 2 centrosomes in diploid ( 2N ) and tetraploid ( 4N ) cells . Denominator represents the number of cells counted . ( G ) A time-lapse micrograph showing a tripolar anaphase in a HS>fzr cell 24 hr after a 60 min heat shock . Cenp-C-Tomato showing kinetochores is in cyan , His2av-GFP showing DNA is in magenta . Time represents min from the last frame prior to anaphase . ( G’ ) Quantification of the number of diploid ( 2N ) and tetraploid ( 4N ) cells that showed tripolar anaphases before heat shock , or 10 hr and 24 hr after heat shock . Only a single tripolar was observed in over 200 movies , which is shown in ( G ) . Scale bars represent 50 μm in A and 5 μm in E , F , and G . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 004 To determine if the G2 cells re-programmed to G1 proceed through a second genome duplication , we examined mitotic chromosome number when mitosis of HS>fzr tissue first resumes ( ten hours after heat shock ) . At this time-point , we observe frequent tetraploidy ( 41% of mitotic cells , equivalent to 93% of all G2 cells prior to heat shock , Figure 1E v . F , F’ , H ) . We obtained similar results when examining the results of fzr overexpression in diploid brain progenitors ( Figure 1—figure supplement 1C ) . These results confirm our ability to induce genome reduplication in normally diploid tissues . We also examined chromosome structure in our induced tetraploid cells . When HS>fzr-induced tetraploid DNA first condenses post-heat shock , all chromatids of each chromosome type are closely aligned in a polytene configuration , as evidenced by having the haploid number of distinguishable chromosomes ( four in females , Figure 1F ) . Frequently , we observe un-pairing of the homologous groups of centromeres within each polytene chromosome ( Figure 1F , asterisk , also see discussion ) . Later , at the first metaphase , homologous chromosomes of each polytene are now completely separated , but the four centromeres of each group of sister chromatids remain conjoined within diplochromosomes ( observed for 96% of tetraploid cells , Figure 1F’ inset , H ) . Thus , HS>fzr induces ectopic genome reduplication , resulting in tetraploid cells with metaphase polytene diplochromosomes . We next examined the mitotic fidelity of cells with diplochromosomes by two independent means: chromosome karyotype analysis and live imaging . By examining the metaphase chromosomes of the division immediately following diplochromosome division , we could detect whether aneuploidy results from diplochromosome segregation . Following the division of cells with diplochromosomes , we observe tetraploid-aneuploid cells with one or two extra or missing chromosomes ( 8 . 6% of mitotic cells ) . In these cells , diplochromosomes are no longer present and instead chromatids are found in distinct sister pairs ( Figure 1G ) . This suggests that during or after anaphase of the first post-reduplication division , diplochromosomes can separate into individual chromatids . Further , these diplochromosome divisions can produce aneuploid daughter cells , many of which continue to divide ( Figure 1G , H , K , Figure 1—figure supplement 1C ) . We also live imaged mitosis of wing imaginal disc and brain progenitor ( neuroblasts and ganglion mother ) cells , both with and without ectopic genome reduplication . In addition to using a histone marker to observe chromosomes , we used the Cenp-C-Tomato marker to observe kinetochores . Control diploid cells divide without errors ( Figure 1I No HS , J , Video 1 ) . In contrast , most ( 80% ) tetraploid divisions with diplochromosomes exhibit lagging chromosomes , DNA bridges , or both ( Figure 1I HS+10 hr , J , Video 2 ) . In our live imaging , diplochromosomes were identifiable as quartets of centromeres and their associated chromosome arms in very close proximity . In some of these divisions we clearly observe four chromatids of a diplochromosome quartet segregating 3:1 ( in agreement with prior work in embryos by Vidwans et al . , 2002 , suggesting incomplete or imprecise sister chromatid disjunction is the cause of chromosome gains and losses ( Figure 1I , Figure 1—figure supplement 1D , Video 2 ) . Mitotic errors in the first division of HS>fzr cells appear to result primarily from diplochromosomes and not tetraploidy itself , as tetraploid cells in the subsequent divisions ( which lack diplochromosomes ) do not exhibit obvious chromosome quartets and have a substantially reduced error rate ( Figure 1J 4N , Figure 1—figure supplement 1E , Video 3 ) . 10 . 7554/eLife . 15204 . 005Video 1 . This video accompanies Figure 1I . Live imaging of a diploid HS>fzr wing imaginal disc cell dividing prior to HS>fzr expression showing His2av-GFP in magenta to label DNA , and Cenp-C-Tomato in cyan to label kinetochores . No mitotic errors are detected . Time Indicates minutes to the last frame of metaphase , scale bars represent 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 00510 . 7554/eLife . 15204 . 006Video 2 . This video accompanies Figure 1I . Live imaging of a tetraploid HS>fzr wing imaginal disc cell with diplochromosomes 10 hr after a 60-min heat shock to induce HS>fzr expression with His2av-GFP in magenta to label DNA , and Cenp-C-Tomato in cyan to label kinetochores . Lagging chromosomes are evident . Time Indicates minutes to the last frame of metaphase , scale bars represent 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 00610 . 7554/eLife . 15204 . 007Video 3 . This video accompanies Figure 1—figure supplement 1E . Live imaging of a tetraploid HS>fzr wing imaginal disc cell without diplochromosomes 24 hr after a 60-min heat shock to induce fzr overexpression with His2av-GFP in magenta to label DNA , and Cenp-C-Tomato in cyan to label kinetochores . Time Indicates minutes to the last frame of metaphase , scale bars represent 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 007 We previously reported centrosome amplification to contribute to polyploid mitotic errors in Drosophila ( Schoenfelder et al . , 2014 ) . However , centrosome amplification does not appear to be a major contributor to mitotic errors in the first ( or subsequent ) polyploid division of HS>fzr animals , as few tetraploid cells amplify centrosomes , and multipolar division is very rare ( Figure 1—figure supplement 1F , G , Video 4 ) . In spite of the high initial error rate caused by separation of diplochromosomes , tetraploid-aneuploid cell divisions continue to occur for at least 5 days after genome reduplication , as determined by cytology ( Figure 1H HS+120 hr ) . We conclude that division of diplochromosomes in the mitotically expanding diploid progenitor tissues that we surveyed can lead to the generation of aneuploid cells , which can continue to divide ( Figure 1K ) . 10 . 7554/eLife . 15204 . 008Video 4 . This video accompanies Figure 1—figure supplement 1G . Live imaging from a tetraploid HS>fzr wing imaginal disc cell 24 hr after a 60-min heat shock , undergoing a tripolar anaphase with His2av-GFP in magenta to label DNA , and Cenp-C-Tomato in cyan to label kinetochores . Time Indicates minutes to the last frame of metaphase , scale bars represent 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 008 To determine the long-term effect of tetraploid-aneuploid divisions on tissue development , we took advantage of the fact that expression of HS>fzr occurs in adult progenitor tissues . We thus examined the survival of these animals to adulthood . Survival is only subtly affected in animals with mild ( 23 . 0% tetraploid [S . E . M . 4 . 9%] ) levels of induced error-prone tetraploid progenitor division ( Figure 2A ) , and resulting adult tissues appear normal ( Figure 2B ) . In contrast , when tetraploidy is further increased by increasing the duration of heat shock , organism survival decreases in a tetraploid-dependent fashion ( Figure 2A , Figure 2—figure supplement 1A ) . Together , these results show that ectopic genome reduplication in multiple progenitor tissues yields tetraploid cells with polytene metaphase diplochromosomes , which are aneuploid-prone ( Figure 1K ) . These conclusions are in agreement with a previous study in the terminal embryonic division of embryos ( Vidwans et al . , 2002 ) . We further show that such aneuploid-prone cells can continue to propagate , and that only at high frequencies are these error-prone tetraploid mitotic events lethal to the organism . 10 . 7554/eLife . 15204 . 009Figure 2 . The spindle assembly checkpoint wait-anaphase response is required after ectopic genome reduplication . ( A ) Quantitation of survival rates from third instar larvae to adulthood of the indicated genotypes without heat shock ( dark blue ) or following a 15-min heat shock ( light red ) ( which generates 23% tetraploid , see methods ) . Bars represent means + standard error of the mean ( S . E . M ) of at least 5 independent experiments , with 20 animals per experiment . *p<0 . 05 , ***p<0 . 001 , NS = p>0 . 05 , t-test . ( B ) Representative micrographs of eyes , wings , and abdomens from HS>fzr alone , mad2 alone , or HS>fzr , mad2 flies heat shocked for 15 min as third instar larvae and then allowed to develop to adults . Red arrow indicates an extra ectopic wing vein , and yellow arrow heads indicate melanotic masses both of which are found in in HS>fzr , mad2 adults following heat shock . ( C ) Representative time-lapse showing a HS>fzr wing imaginal disc 10 hr after a 60-min heat shock including a cell with diplochromosomes ( yellow dotted line ) and a diploid cell ( blue dashed line ) dividing within the same field ( one of the diploid daughters drifts vertically out of the frame ) . The cell with diplochromosomes takes more than four times as long to enter anaphase , and division is error prone . His2av-GFP showing DNA is in white . Time indicates minutes from the start of filming . ( D ) The length of metaphase without fzr overexpression ( No HS ) or +10 hr after a 60-min heat shock to induce overexpression from HS>fzr , and HS>fzr , mad2 larval wing imaginal disc cells . Points represent individual cell divisions , bars represent means , diploid cells are represented in dark blue , polyploid cells are represented in light red , HS>fzr is represented in circles , HS>fzr , mad2 is represented in triangles . N>17 cells per column , ***p<0 . 001 , Not Significant ( NS ) = p>0 . 05 , one-way ANOVA with correction for multiple hypothesis testing . ( E ) Third instar larval wing imaginal discs from HS>fzr or HS>fzr , mad2 stained for TUNEL in green and DAPI in magenta without heat shock ( No HS ) or +24 hr after a 15-min heat shock . ( F ) Quantification of the number of TUNEL positive foci per wing disc for HS>fzr and HS>fzr , mad2 without heat shock ( No HS , blue bars ) or 24 hr after a 15 min heat shock ( +HS , red bars ) . Points represent individual wing imaginal discs , bars represent mean , N ≥ 18 discs per condition . NS = p>0 . 05 , * = p<0 . 05 , *** = p<0 . 001 , ANOVA . Scale bars represent 500 μm in B , 5 μm in C , and 50 μm in E . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 00910 . 7554/eLife . 15204 . 010Figure 2—figure supplement 1 . Supporting data regarding Mad2’s role in response to diplochromosomes . ( A ) Graph of the survival rate from third instar larvae to adulthood following a 30 min heat shock from wild type , HS>fzr alone , mad2 alone , or HS>fzr , mad2 animals . Bars represent means + Standard Error of the Mean ( + SEM ) of at least 5 experiments with 20 animals per experiment . ( B ) Graph showing the number of Phospho-Histone H3 positive ( PH3+ ) cells per wing imaginal disc incubated in PBS for one hour with ( + ) or without ( - ) colcemid from wild type , or mad2 animals . ( C ) Representative time-lapse micrographs of wing disc cells expressing Cenp-C-Tomato ( magenta ) and BubR1-GFP ( cyan ) dividing 10 hr after a 60-min heat shock . In diploid cells ( top ) BubR1 is evenly distributed across the kinetochores . In cells with diplochromosomes ( bottom ) mitosis is longer and BubR1 is present for longer as well . BubR1 is not evenly distributed on kinetochores prior to anaphase but appears to concentrate on a subset of diplochromosomes kinetochores . Time indicates minutes to the last prior to anaphase . ( D ) Micrographs of antibody staining against Drosophila cleaved caspase 1 protein ( DCP1 ) in the wing disc pouch of HS>fzr or HS>fzr , mad2 animals without heat shock or 24 hr after a 15-min heat shock . DCP1 is in green and DAPI is in magenta . ( E ) Quantification of the relative amount of DCP1 staining per wing disc from HS>fzr and HS>fzr , mad2 without heat shock or 24 hr after a 15-min heat shock . Data was normalized so that the mean of HS>fzr without heat shock ( No HS ) is equal to 1 . Points represent individual wing discs , bars represent means without ( dark blue ) or with ( light red ) a 15 min heat shock . *p<0 . 05 , ***p< 0 . 001 , ANOVA , N > 12 animals per condition . ( F ) Representative live imaging of a HS>fzr , mad2 cell dividing with diplochromosomes 10 hr after a 60-min heat shock . Cenp-C-Tomato to label kinetochores is in cyan and His2av-GFP to label DNA is in magenta . Minutes indicate time to the last frame prior to anaphase . ( G ) Graph showing the frequency of errors in HS>fzr , mad2 diploid cells ( 0% ) , HS>fzr , mad2 tetraploid cells with diplochromosomes ( 100% ) , and HS>fzr cells with diplochromosomes ( 80% ) . Scale bar represents 5 μm in C and F and 50 μm in D . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 010 Our data ( Figure 1H , Figure 1—figure supplement 1C ) suggest that many polytene diplochromosome divisions in a variety of tissues do not lead to aneuploidy . Little is known about aneuploidy prevention mechanisms in cells with polytene chromosomes , despite the numerous mechanisms that can generate these aberrations . Through live imaging , we uncovered one such aneuploidy-prevention mechanism in cells with metaphase diplochromosomes . Because our heat shock protocol only affects cells in G2 , a single HS>fzr pulse creates a mixed population of unaltered diploid cells and diplochromosome-containing tetraploid cells , allowing us to simultaneously live image both cell types in the same tissue . Metaphase in cells with diplochromosomes ( Figure 2C , yellow dotted outline ) is significantly longer than in diploid cells ( Figure 2C blue dashed outline , Video 5 , Figure 2D ) , consistent with previous work on diplochromosomes formed in Drosophila Securin mutants ( Pandey et al . , 2005 ) . We thus hypothesized that diplochromosomes trigger the SAC , which activates a wait-anaphase signal until all kinetochores attach to microtubules and are under tension ( London and Biggins , 2014; Musacchio , 2015 ) . To test this model , we examined SAC-defective mad2 null animals ( Buffin et al . , 2007; Emre et al . , 2011 ) Figure 2—figure supplement 1B ) . Using live imaging of wing imaginal discs before and after heat shock , we find that loss of mad2 eliminates the lengthened period of metaphase caused by diplochromosomes ( Figure 2D ) . When the checkpoint is active unattached or misattached kinetochores generate a wait-anaphase signal by localizing SAC proteins such as BubR1 to those kinetochores ( Musacchio , 2015 ) . To confirm that diplochromosomes have localized SAC proteins we co-imaged kinetochores and BubR1 in wing disc cells after fzr expression ( Royou et al . , 2010 ) . We find that in diploid cells BubR1-GFP is clearly evident on kinetochores following nuclear envelope break down and remains there until anaphase . This signal is relatively evenly spread across all the kinetochores ( Figure 2—figure supplement 1C , diploid , Video 6 ) . In contrast BubR1-GFP remains localized for much longer in cells with diplochromosomes and is often localized strongly to a specific kinetochore group rather than evenly distributed , suggesting that a subset of diplochromosomes may have trouble forming attachments that satisfy the checkpoint ( Figure 2—figure supplement 1C , diplochromosomes , Video 7 ) From these data , we conclude that diplochromosomes trigger a SAC wait-anaphase response . 10 . 7554/eLife . 15204 . 011Video 5 . This video accompanies Figure 2C . Live imaging of a wing disc from a HS>fzr animal 10 hr after a 60-min heat shock showing mitosis by a polyploid diplochromosome-containing cell ( yellow dotted line ) and a diploid cell ( blue dotted and dashed line ) in the same field , His2av-GFP labelling DNA is shown . Time Indicates minutes from the start of filming , scale bars represent 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 01110 . 7554/eLife . 15204 . 012Video 6 . This video accompanies Figure 2—figure supplement 1C . Live imaging of Cenp-C-Tomato in magenta to label kinetochores and BubR1-GFP in cyan during the division of a diploid cell 10 hr after a 60-min heat shock . Time indicates minutes to the last frame of metaphase . Scale bar represents 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 01210 . 7554/eLife . 15204 . 013Video 7 . This video accompanies Figure 2—figure supplement 1C . Live imaging showing Cenp-C-Tomato to label kinetochores in magenta and BubR1-GFP in cyan during the division of a tetraploid cell with diplochromosomes 10 hr after a 60-min heat shock . Time indicates minutes to the last frame of metaphase . Scale bar represents 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 013 Although important for mitosis in cultured Drosophila S2 cells ( Orr et al . , 2007 ) , the Mad2-directed SAC is reported to be dispensable in Drosophila tissue mitosis ( Buffin et al . , 2007; Emre et al . , 2011 ) . mad2 null animals are viable with no obvious tissue defects ( Figure 2B; Buffin et al . , 2007 ) due in part to an apoptotic response ( Morais da Silva et al . , 2013 ) . In contrast , the Mad2-dependent wait-anaphase response is essential during development of HS>fzr animals . Even at low levels of tetraploidy , which affect the survival of HS>fzr animals only slightly , few HS>fzr , mad2 animals survive to adulthood ( 15 . 4% , Figure 2A , Figure 2—figure supplement 1A ) . To understand why HS>fzr , mad2 animals have survival defects , we analyzed the surviving animals . In these animals , we find a variety of developmental defects in normally diploid tissues , including smaller eyes , ectopic wing veins , and melanotic abdominal masses ( Figure 2B ) . Increased apoptosis is associated with these tissue malformation phenotypes , as progenitor tissue from HS>fzr , mad2 animals have much higher rates of apoptotic cell death as shown by both TUNEL labeling ( Figure 2E , F ) , and cleaved caspase staining ( Figure 2—figure supplement 1D , E ) . Further , 100% of mad2 diplochromosome divisions exhibit lagging chromosomes , or DNA bridges ( Figure 2—figure supplement 1F , Video 8 ) , compared with 80% of divisions in HS>fzr cells . These data may suggest that diplochromosomes are likely susceptible to at least two classes of mitotic errors , one that can be corrected by the SAC , and one that cannot . It is also likely that the mad2 diplochromosome divisions are qualitatively more erroneous than in WT , which may account for differences in survival and tissue phenotype between these genotypes . Taken together , our data identify an important role for the Mad2-dependent SAC in delaying anaphase in the presence of metaphase polytene diplochromosomes . 10 . 7554/eLife . 15204 . 014Video 8 . This video accompanies Figure 2—figure supplement 1F . Live imaging showing His2av-GFP in magenta and Cenp-C-Tomato in cyan during the division of a HS>fzr , mad2 tetraploid cell with diplochromosomes 10 hr after a 60-min heat shock . Time indicates minutes from the last frame of metaphase . Scale bar represents 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 014 Having defined a response to mitosis after ectopic genome reduplication , we next asked if this same mechanism operates in a tissue that we previously found to divide after programmed genome reduplication . In earlier work we found Drosophila rectal papillar cells ( hereafter: papillar cells ) naturally undergo two fzr-dependent endocycles during the 2nd larval instar to generate octoploid cells and then divide , on average , two times during pupal development . An intervening S-phase accompanies these polyploid divisions , and cells at the papillar base undergo one additional S-phase after the final polyploid mitosis ( Figure 3A; Fox et al . , 2010; Schoenfelder et al . , 2014 ) . Thus , as with HS>fzr induction in diploid tissues , papillar development naturally involves genome reduplication followed by mitosis . Our previous work established that papillar mitoses can be error prone , so the same problems with dissociating polytene chromosomes in HS>fzr tissues could be responsible for a portion of these errors during papillar divisions . However , we previously did not see , in hundreds of observed cells , any instances of metaphases with persistent polyteny in papillar cells , suggesting that papillar cells somehow avoid mitosis of polytene chromosomes . Through careful re-examination of the first octoploid metaphase , we confirmed that papillar chromosomes in these octoploid cells are arranged in individual sister chromatid pairs ( Figure 3B’’ inset , C ) . This suggested two possibilities: 1 ) papillar cells never form polytenes , or 2 ) papillar cells form polytenes , but somehow separate into recent sister pairs prior to the first metaphase . 10 . 7554/eLife . 15204 . 015Figure 3 . Programed genome reduplication in papillar cells is followed by Separation Into Recent Sisters ( SIRS ) , which individualizes polytene chromosomes into recent sister pairs . ( A ) A model of the cell cycles in Drosophila papillar cells . These cells undergo two rounds of the endocycle in the 2nd instar to reach 8N/16C , then enter a G2-like state , then undergo , on average , two cell divisions with intervening S-phases during pupation ( Fox et al . , 2010; Schoenfelder et al . , 2014 ) . ( B ) Karyotypes of papillar cells during the 1st polyploid division , ( B–B’’ ) . Chromosomes are pseudocolored according to type and labeled in panel B . Panel B inset shows the 4th chromosomes , which were out of frame . When chromosomes first condense following genome reduplication , they are in a polytene configuration ( asterisks indicate the 8 separated centromere pairs of an otherwise polytene X chromosome ) . This cell contains a heterozygous pericentric inversion on the third chromosome caused by the presence of a balancer chromsome . B’ Example of the clumped configuration in early mitosis of the first papillar division . B’’ Example of fully separated 1st division papillar chromosomes . No diplochromosomes are present ( compare X chromosome in inset to inset in Figure 1F’ ) . Note- one second chromosome contains a DNA break , which are common in wild type papillar cells ( Fox et al . , 2010; Bretscher and Fox , 2016 ) . ( C ) Karotype of papillar chromosomes during the 2nd polyploid division . Chromosomes are pseudocolored according to type as in Figure 3B . At the second division almost all cells show chromosomes fully separated into sister pairs . ( D ) Percentage of cells with polytene chromosomes , recent sisters clumped , or recent sisters clearly separated from four time points: prior to the first division ( following treatment with Calyculin A to visualize pre-mitotic chromosome structure- see Materials and methods ) , during the first division ( no drug treatment ) , during the first division ( following treatment for 30 min with colcemid to enrich for late metaphase ( 1st Div + colc ) ) , and during the second division ( no drug treatment ) . *p<0 . 05 compared to 1st Division , chi-squared test , N ≥ 26 karyotypes per treatment from at least 5 animals . ( E ) Quantification of the number of resolvable Cenp-C-Tomato foci in fixed papillar cells during the course of pupation ( expressed in hours post pupation ) . Before the first mitosis ( 18 hr ) each cell has an average of 4 . 1 kinetochore foci closely corresponding to the haploid chromosome number , following the first division ( 24 hr ) cells average 15 . 1 foci per cell . At 20 hr some cells have divided and others are yet to divide and the distribution is bimodal . Circles represent individual cells . Bars represent the mean of 3 animals per time point and 15 cells per animal . ***p<0 . 001 , **p<0 . 01 , by Kruskal-Wallis one-way ANOVA . ( F ) Live imaging of the 1st divisions from wild type papillar cells shows the SIRS process . Cenp-C-Tomato ( Cenp-C ) is in cyan , His2av-GFP ( His ) is in magenta . Time represents minutes to the last frame prior to anaphase . In the 1st division kinetochores from a group of homologs are tightly clustered prior to division . At -18:00 min . relative to anaphase , chromosome condensation has begun and polytene chromosomes are visible ( See His channel ) . Dispersal continues until individual pairs of sister kinetochores are evident at metaphase . The inset shows the Cenp-C-Tomato channel of a single kinetochore focus from time frames -36 min to -8 min . ( G ) Live imaging of the 2nd division from a wild type papillar cell . In contrast to the first division many discrete kinetochore foci are evident at time-points prior to the onset of mitosis , and polytene chromosomes are never evident . ( H ) A model for a pair of homologs undergoing 2 rounds of endo-S-phase to become a polytene 16C chromosome . The polytene chromosome then separates into pairs composed of only the most recent sister chromatids during mitosis , and each sister then segregates to opposite poles at anaphase . Scale bar represents 5 μm , except in insets in B’’ and F where it represents 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 01510 . 7554/eLife . 15204 . 016Figure 3—figure supplement 1 . Supporting data regarding SIRS . ( A ) Representative micrograph of the hindgut including the ileum ( anterior , left ) and rectum ( posterior , right ) stained for GFP driven by brachyenteron ( byn ) ( green ) , stg>LacZ4 . 9 ( magenta ) , and DAPI ( white ) at 16 hr post pupation , a time point shortly before the onset of the first papillar mitosis . The rectum but the not ileum stains strongly for expression of the G2/M regulator string . ( B ) Karyotypes of papillar cells from just prior to the first papillar division treated with 200 nM Calyculin A to induce Premature Chromosome Compaction ( PCC ) , showing polytene chromosome organization in the interphase prior to SIRS . Panels show genotypes without balancer chromosomes ( B ) , with one balancer chromosome ( B’ ) or with two balancer chromosomes ( B’’ ) . Balancer chromosomes are evident by the presence of pericentric inversions ( yellow arrowhead ) . Asterisks indicate where the 8 recent sister chromatid pairs of the acrocentric X-chromosome have already separated into eight pairs of recent sister centromeres . Chromosomes are false colored by homolog and labelled in ( B ) ( C ) Two representative micrographs taken from the same field showing Fluorescent in situ hybridization ( FISH ) to a single region on the left arm of chromosome 3 in green and DAPI in magenta during the 1st papillar division , at a time-point in which some cells have a pre-SIRS chromosome configuration ( C ) while others have undergone SIRS ( C’ ) . Yellow arrowheads indicate FISH foci . ( D ) Live imaging of a papillar cell 1st division from a male expressing MSL3-GFP ( magenta ) and Cenp-C-Tomato ( Cyan ) . MSL3-GFP is specific to the male X chromosome and only a single Cenp-C-Tomato focus is MSL3-GFP positive prior to SIRS ( yellow arrows ) indicating that each Cenp-C foci is composed of a single homolog . Time represents minutes prior to the final frame of metaphase . ( E ) Micrographs of polyploid pupal division mitotic chromosomes stained with Phospho-Histone H3 ( white ) from the ileum of Culex pipiens , showing cells in all stages of mitosis including an apparent pre-SIRS polytene phase ( E ) , as well as a post SIRS prophase ( E’ ) cell . Scale bars represent 5 μm except in A where it represents 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 016 To distinguish these two possibilities , we examined papillar karyotypes from the moment chromosome condensation could be detected . Papillar cells re-enter mitosis from a G2-like state , as evidenced by expression of the G2/M regulator Cdc25/string just before the onset of pupal cell cycles ( Fox et al . , 2010 ) ( Figure 3—figure supplement 1A ) . At time points early in the first mitosis , we indeed find that papillar chromosomes are polytene ( Figure 3B , Polytene ) . In these polytenes , we again see examples of cells where the centromeric regions are no longer tightly associated , as we did in our studies of HS>fzr-induced polyteny . However , unlike in cells with induced polyteny , the centromeres in papillar polytene cells are able to not only separate into groups of homologs , but to further separate into individual sister chromatid pairs ( asterisks in Figure 3B and Figure 3—figure supplement 1B vs . Figure 1F , also see discussion ) . In flies heterozygous for inversion-containing balancer chromosomes , papillar polytene structure is locally perturbed , likely due to the disruption of somatic homolog pairing ( Figure 3—figure supplement 1B ) . At this early mitotic time point , we also observe cells where polytenes are absent . Instead , in these cells , sister chromatid pairs and homologs of each chromosome type are separated but remain clumped closely together , as if the polytene chromosome recently separated into pairs containing only the most recent sister chromatids ( Figure 3B’ , D , Clumped ) . Neither the polytene nor the clumped configurations remain during the second division ( Figure 3C , D ) , suggesting a specific chromosome structure is present early in the first division of papillar cells . Similarly , by Fluorescent In Situ Hybridization ( FISH ) we find examples of both closely associated ( polytene ) and dispersed ( separated/non-polytene ) signals during the period of the first papillar mitosis ( Figure 3—figure supplement 1C , C’ ) . Thus , a key difference between the response to genome reduplication between papillar and HS>fzr cells is the elimination of polyteny before anaphase in papillar cells . To examine if a majority of ( if not all ) papillar cells transition from polytene to separate/non-polytene chromosomes during the first mitosis , we used drug treatment to isolate specific chromosome structures during the transition into the first papillar division . To enrich for early mitotic and pre-mitotic chromosomes , we induced Premature Chromosome Compaction ( PCC ) in papillar cells at a time point just prior to the first mitosis ( Methods ) . PCC causes interphase chromosomes to condense and makes it possible to visualize interphase chromosome structure by standard cytological methods ( Figure 3—figure supplement 1B ) . Using this technique , we find that in pre-mitotic papillar tissue , clear polytene chromosomes are present in nearly every cell ( Figure 3D Pre 1st Div , Figure 3—figure supplement 1B ) . If we instead enrich for cells in metaphase of the first mitosis by treating with the spindle poison colcemid , we find zero examples where chromosomes are still polytene . In these metaphase-enriched samples , all chromosomes are separated into recent-sister pairs , and even cells with clumped chromosomes are rare ( Figure 3D , 1st Div +colc ) . Thus , our pharmacological studies further suggest that essentially all genome-reduplicated papillar cells are programmed to completely eliminate polytene chromosomes as cells progress into the first metaphase . To observe the temporal dynamics of the pre-anaphase elimination of papillar polyteny , we used live imaging , using the same markers used to image diplochromosome division . Prior to the first papillar mitosis , the kinetochores from each homolog are closely associated into an average of 4 . 1 large foci , close to the haploid number of distinct chromosomes ( 4 for females and 5 for males due to X/Y un-pairing , Figure 3E ) . As time progresses in the first division , it is possible to watch these large kinetochore foci disperse into many smaller foci prior to metaphase ( Figure 3F , inset , Video 9 ) . In contrast , prior to the second division kinetochores are already separated into many more foci ( an average of 15 . 1 observably distinct foci per cell ) before entry into mitosis ( Figure 3E ) . During the second division , the number of resolvable foci remains essentially constant ( Figure 3G , Video 10 ) . Additionally , the histone marker reveals that polytene chromosomes are visible when chromosomes first condense and can then be seen to disperse during the first but not the second division ( Figure 3F v . 3G , -18:00 min ) . This result confirms the model that genome-reduplicated papillar cells eliminate polyteny during the first mitosis , then undergo an intervening S-phase before the next division ( Figure 3A ) . We also confirmed that each clump of 4 or 5 pre-first division centromeres only contains a single chromosome type . To do so , we took advantage of the fact that dosage compensation in flies relies on upregulation of transcription on the male X chromosome via the Dosage Compensation Complex ( Conrad and Akhtar , 2011 ) . By live imaging papillar cells expressing the DCC complex protein MSL3 tagged with GFP , which localizes only to the male X-chromosome ( Strukov et al . , 2011 ) , we find that indeed only a single Cenp-C-Tomato focus is MSL3-GFP positive prior to polytene dissociation ( Figure 3—figure supplement 1D , Video 11 ) . Taken together , we find papillar cells avoid mitosis of polytene chromosomes in part by undergoing a pre-anaphase chromosome separation process we term Separation Into Recent Sisters ( SIRS , Figure 3H ) . 10 . 7554/eLife . 15204 . 017Video 9 . This video accompanies Figure 3F . Live imaging of a papillar cell undergoing a first division , including the SIRS process showing His2av-GFP in magenta , and Cenp-C-Tomato in cyan . SIRS is most evident at the centromere which transitions from four tight foci prior to mitosis to many foci at anaphase . Polytene chromosomes are visible 18:00 min before anaphase . Time Indicates minutes to the last frame of metaphase , scale bars represent 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 01710 . 7554/eLife . 15204 . 018Video 10 . This video accompanies Figure 3G . Live imaging showing His2av-GFP in magenta , and Cenp-C-Tomato in cyan from a papillar cell undergoing a second division . Time Indicates minutes to the last frame of metaphase , scale bars represent 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 01810 . 7554/eLife . 15204 . 019Video 11 . This video accompanies Figure 3—figure supplement 1D . Live-imaging showing MSL3-GFP in magenta and Cenp-C-Tomato in cyan from a male papillar cell undergoing a first division . Only a single Cenp-C-Tomato foci is MSL-3-GFP positive . Time Indicates minutes to the last frame of metaphase , scale bars represent 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 019 Previously , we confirmed that a similar polyploid mitotic program occurs in the developing hindgut of the mosquito Culex pipiens ( Fox et al . , 2010 ) . Interestingly , classical descriptions of mitosis in this part of the Culex hindgut seem to suggest a polytene organization is present only early in the first polyploid mitosis ( Grell , 1946 ) . In agreement with this observations , we find Phospho-Histone H3 positive polytene chromosomes during the period of the first polyploid mitosis ( Figure 3—figure supplement 1E , Video 12 ) . From our Drosophila and Culex studies , we conclude that unlike cells that enter metaphase with polytene chromosomes , a separate mechanism , SIRS , can eliminate polyteny as cells enter metaphase . 10 . 7554/eLife . 15204 . 020Video 12 . This video accompanies Figure 3—figure supplement 1E . Video showing sequential z-planes from a fixed ileum of Culex pipiens with mitotic cells that are pre-SIRS ( left ) and post-SIRS ( right ) stained with Phospho-Histone H3 . scale bars represent 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 020 Our dual genome-reduplication systems identified two distinct cellular responses to polytene chromosomes: a Mad2-dependent response that delays anaphase when polytenes remain at metaphase , and a SIRS response that eliminates polyteny as cells enter metaphase . Despite the lack of metaphase polytene chromosomes in papillar cells , we also identified an important role for Mad2 during SIRS . Because mad2 loss has no reported mitotic defects in Drosophila animals ( Buffin et al . , 2007 ) , we were surprised to find that first division mad2 papillar cells exhibit a substantial increase in DNA bridges ( Figure 4A , B , Video 13 , Video 14 ) . We did not detect similar defects during papillar mitosis of animals null for mad1 , another SAC component ( Figure 4A , B , Video 15 ) . Thus , a Mad1 independent function of Mad2 is important in cells during SIRS . 10 . 7554/eLife . 15204 . 021Figure 4 . SIRS does not depend on the SAC wait-anaphase response or formation of a mitotic spindle . ( A ) Representative micrographs of wild type ( WT ) , mad2 , and mad1 cell 1st divisions beginning in the last frame of metaphase ( 0:00 ) and continuing through eight minutes of anaphase . Cenp-C-Tomato showing kinetochores in cyan , His2av-GFP showing DNA in magenta . Yellow arrowheads show kinetochores that are part of a bridge between the two poles in a mad2 cell . Time represents minutes from the last frame prior to anaphase . ( B ) Quantification of the frequency of persistent DNA bridging observed 4 min after the onset of the 1st division anaphase from papillar cells in wild type ( WT ) , mad2 , and mad1 animals . Bars represent the mean of all cell divisions , + Standard Error of the Mean ( S . E . M . ) ***p<0 . 001 , t-test . ( C ) Representative images of a single pupal rectums from wild type or mad2 animals treated with colcemid for 60 min prior to fixation and stained for Phospho-Histone H3 ( PH3 ) positive nuclei in magenta and expressing GFP under the control of brachyenteron ( byn , a hindgut marker ) in green . ( D ) The fold increase in the number of polyploid mitotic cells per hindgut from wild type and mad2 animals following treatment with colcemid compared to without colcemid . A value of 1 equals no difference . Bars represent mean fold change ( + SEM ) , and are labeled with the mean value . **p<0 . 01 , t-test , N ≥ 8 animals per condition . ( E ) The fold increase in metaphase length for HS>fzr polyploid cells with ( 1st ) and without polytene diplochromosomes ( 2nd ) compared to papillar cells with ( 1st division ) and without ( 2nd division ) polytene chromosomes . A value of one equals no difference between 1st and 2nd divisions . The increase in HS>fzr wing cells indicates that metaphase polytene diplochromosomes trigger the spindle assembly checkpoint , but papillar polytene chromosomes do not . Bars represent means ( +S . E . M . ) , and are labeled with the mean value . *p<0 . 05 , t-test , N ≥ 22 cells per condition from at least 5 animals . ( F ) Live imaging of a cell expressing Cenp-C-Tomato in cyan and Jupiter-GFP in magenta undergoing SIRS in the presence of a vehicle control . Time represent minutes before the onset of SIRS . Inset shows the dispersal of a single Cenp-C-Tomato kinetochore at all the time points between 0 min and 16 min . ( F’ ) shows the number of resolvable Cenp-C-Tomato foci from prior to SIRS ( 0 min ) and after SIRS ( 16 min ) , points represent individual cells with the two time points connected by a line . ***p<0 . 001 , t-test N = 12 divisions from 2 animals . ( G ) Live imaging of a cell expressing Cenp-C-Tomato in cyan and Jupiter-GFP in magenta undergoing SIRS in the presence of a colcemid . Time represent minutes from the onset of SIRS . Inset shows the dispersal of a single Cenp-C-Tomato kinetochore at all the time points between 0min and 16 min . ( G’ ) shows the number of resolvable Cenp-C-Tomato foci from prior to SIRS ( 0 min ) and after SIRS ( 16 min ) , points represent individual cells with the two time points connected by a line . ***p<0 . 001 , t-test , N = 15 cells from 5 animals . Scale bar represents 5 μm except in insets of F , and G where they represent 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 02110 . 7554/eLife . 15204 . 022Video 13 . This video accompanies Figure 4A . Live imaging showing His2av-GFP in magenta , and Cenp-C-Tomato in cyan from a wild type papillar cell during anaphase of the first mitosis . Minutes indicates time before the onset of anaphase . No DNA bridge is present . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 02210 . 7554/eLife . 15204 . 023Video 14 . This video accompanies Figure 4A . Live imaging showing His2av-GFP in magenta , and Cenp-C-Tomato in cyan from a mad2 papillar cell from DNA condensation through anaphase of the first mitosis , including formation of a DNA bridge . Minutes indicates the time to before the onset of anaphase . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 02310 . 7554/eLife . 15204 . 024Video 15 . This video accompanies Figure 4A . Live imaging showing His2av-GFP in magenta , and Cenp-C-Tomato in cyan from a mad1 papillar cell during from DNA condensation through anaphase of the first mitosis . No DNA bridging is present in mad1 papillar cells . Time indicates minutes from the onset of anaphase . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 024 In Drosophila , Mad2 plays a conserved , cell type-dependent role in regulating NEBD-to-anaphase onset timing ( Buffin et al . , 2007; Meraldi et al . , 2004; Rodriguez-Bravo et al . , 2014; Yuan and O'Farrell , 2015 ) . As for the wait-anaphase response , this mitotic timing role involves Mad2 inhibition of the Anaphase Promoting Complex . However , Mad2’s control of overall mitotic timing is irrespective of SAC kinetochore attachment surveillance ( Meraldi et al . , 2004; Rodriguez-Bravo et al . , 2014 ) . Interestingly , Drosophila Mad1 is reported to be dispensable for regulation of NEBD-to-anaphase timing ( Emre et al . , 2011 ) . Given the lack of mad1 phenotypes with respect to the first papillar division , we thus hypothesized that SIRS enables papillar cells to bypass the SAC-mediated anaphase delay , and that Mad2 control of overall mitotic timing is important during SIRS . If so , one would predict cells undergoing SIRS to not trigger an anaphase delay , but to still depend on mitotic timing . To first test if papillar cells employ the SAC wait-anaphase in response to polytene chromosomes , we treated animals with colcemid , a known SAC wait-anaphase trigger . This treatment increases the mitotic index of wild type papillar cells , whereas the mitotic index of mad2 null animals is unaffected ( Figure 4C , D ) Thus , spindle defects trigger the SAC wait-anaphase response in papillar cells . We next asked if the SAC wait-anaphase responds to polytene chromosomes during SIRS . If so , the first divisions ( polytenes present ) should have a longer metaphase than the second division ( polytenes absent ) . However we find that metaphase is not any longer in the first papillar division than in the second papillar division , while in contrast metaphase is almost twice as long in wing cells with diplochromosomes than in those that are polyploid but lack diplochromosomes ( Figure 4E ) . We then tested if triggering the SAC wait-anaphase response can prevent or delay SIRS completion . We find SIRS occurs on schedule even in the presence of colcemid concentrations that are sufficient to eliminate a detectable spindle and inhibit anaphase ( Figure 4F , G , Video 16 , Video 17 ) . We conclude that: a ) SIRS is not regulated by the SAC wait-anaphase response , and b ) chromosome separation during SIRS does not require a mitotic spindle . 10 . 7554/eLife . 15204 . 025Video 16 . This video accompanies Figure 4F . Live imaging showing Jupiter-GFP in magenta and Cenp-C-Tomato in cyan from a first division papillar cell undergoing SIRS in control imaging media . Time indicates minutes from the start of filming . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 02510 . 7554/eLife . 15204 . 026Video 17 . This video accompanies Figure 4G . Live imaging showing Jupiter-GFP in magenta and Cenp-C-Tomato in cyan from a first division papillar cell undergoing SIRS in the presence of a colcemid . Colcemid prevents spindle formation so the Jupiter remains diffuse . Time indicates minutes from the onset of filming . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 026 We next tested whether Mad2-dependent control of overall mitotic timing is crucial for efficient SIRS . Using an NEBD marker , we first confirmed that Mad2 regulates NEBD-to-anaphase timing and that mad2 cells spend significantly less time in mitosis than wild type cells ( Figure 5A , B , Video 18 , Video 19 ) . We also find that Nuclear Envelope Breakdown and the onset of SIRS are synchronous in wild type cells and that SIRS generally continues until up to the onset of anaphase ( Figure 5C ) . This suggests that the rapid mitosis in mad2 cells might lead to a failure of complete SIRS , which could cause the resulting DNA bridges . In our live imaging , we saw evidence that a pre-SIRS group of homologs would often fail to completely disperse prior to anaphase ( Figure 5A , mad2 , yellow arrowhead ) . To quantify this , we generated heat maps and line profiles of centromere signals at the metaphase plate . We performed this analysis just prior to the onset of mitosis in wild type cells , mad2 cells that did not generate bridges , and mad2 cells that did generate DNA bridges ( Figure 5D ) . From these measurements , we found that SIRS fails to complete before anaphase in mad2 animals , leading to a high variance of centromere intensity signal across the metaphase plate ( reflecting failure of centromere dissociation/SIRS completion ) . This high variance disrupts the bilateral symmetry of the metaphase plate in mad2 cells that form DNA bridges ( Figure 5D’ ) . Taken together , our data show mitotic fidelity after genome reduplication is improved by one of two Mad2-dependent functions: 1 ) in the presence of metaphase polytene chromosomes a Mad2-dependent wait-anaphase signal is generated , 2 ) the efficient elimination of polytenes as cells enter metaphase by SIRS requires a Mad2 ( but SAC wait-anaphase-independent ) NEBD-anaphase timer ( Figure 5E ) . 10 . 7554/eLife . 15204 . 027Figure 5 . SIRS is dependent on Mad2-dependent mitotic timing . ( A ) Live imaging of a representative wild type and mad2 cell expressing geminin-Azami ( magenta ) in and Cenp-C-Tomato ( cyan ) during the 1st papillar division . Just the Cenp-C-Tomato channel is also shown . Nuclear Envelope Breakdown ( NEBD ) can be seen when the geminin signal goes from nuclear to cytoplasmic . Time represents minutes from NEBD . The mad2 cell reaches anaphase more quickly after NEBD than the wildtype cell ( 14 min to 19 min ) . Yellow arrows indicate a Cenp-C foci in mad2 that appears to fail SIRS and is still partially clumped at anaphase . ( B ) The length of time from NEBD to anaphase in wild type and mad2 papillar cells . Points represent individual cell divisions , red bar represents mean ( 17 . 1 min for wild type , 14 . 0 min for mad2 ) . ***p<0 . 001 , t-test , N = 22 cell divisions for each condition from at least 5 animals . ( C ) Quantification of the intensity of geminin-Azami in magenta and a measure of kinetochore clusteredness in cyan over time from wild type cell . 0 min represents the onset of anaphase . Both measures decline synchronously at the onset of NEBD . Data represents the mean of 22 cells . ( D ) Representative images of Cenp-C-Tomato forming the metaphase plate of WT or mad2 cells immediately prior to the onset of anaphase with reds indicating more Cenp-C-Tomato signal and blue indicating less signal ( Top ) and line graphs measuring the total signal intensity from left to right ( Bottom ) , in call cases the eventual division is in the same left-right orientation . mad2 metaphases were split into those that did not generate a persistent DNA bridge at anaphase ( no bridge ) and those that did ( persistent bridge ) . ( D’ ) Aggregate plots of the line graph and the confidence interval for each category . mad2 cells that formed bridgs are significantly more variable than wild type or mad2 without bridging . N > 13 cells per category . ( E ) Model: A simplified model in which four sisters from a single round of genome reduplication are shown . In cells with SIRS ( down arrow ) polytenes fully split into individual sister pairs and with a functioning mitotic timer complete SIRS and divide evenly . However , in the absence of the timer anaphase is precocious and DNA bridges result from incompletely resolved polytene chromosomes . In cells without SIRS ( upper arrow ) , diplochromosomes result . The spindle assembly checkpoint ( SAC ) delays cells in metaphase and reduces but does not eliminate aneuploidy during the ensuing anaphase . In the absence of a checkpoint , cell death results from errant diplochromosome divisions . DOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 02710 . 7554/eLife . 15204 . 028Video 18 . This video accompanies Figure 5A . Live imaging showing Geminin-Azami in magenta , and Cenp-C-Tomato in cyan in a wild type papillar cell from prior to Nuclear Envelope Breakdown through anaphase . The geminin signal is nuclear before the onset of mitosis . Concurrent with NEBD is the onset of SIRS . Anaphase takes place 19 min after NEBD . Time indicates minutes from the first frame after NEBD . Scale bar represents 5 μmDOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 02810 . 7554/eLife . 15204 . 029Video 19 . This video accompanies Figure 5A . Live imaging showing Cenp-C-Tomato in cyan and Geminin-Azami in magenta in a mad2 papillar cell from prior to NEBD through anaphase . Anaphase is 14 min after NEBD and Cenp-C-tomato clumps are still evident . Time indicates minutes from the first frame after NEBD . Scale bar represents 5 μmDOI: http://dx . doi . org/10 . 7554/eLife . 15204 . 029
Despite a large body of literature describing reduplicated chromosomes in development and disease , the cellular and molecular responses enabling cells to progress through mitosis after genome reduplication have remained essentially unknown . Here , we define two such responses- one that prevents malformation of tissues with polytene chromosomes that persist until anaphase onset , and another ( SIRS ) that eliminates polyteny before anaphase onset . Both polyteny responses require the conserved mitotic fidelity regulator Mad2 , yet Mad2’s role in each response is distinct ( anaphase delay vs . control of overall mitotic timing ) . These findings identify new roles for Drosophila Mad2 , for which few roles have been identified . Further , our findings illuminate a likely recurring role for Mad2 in response to genome reduplication . In cells with metaphase polytenes ( e . g . diplochromosomes ) , our data suggest polytene chromosomes present a challenge for the mitotic spindle , leading to a prolonged period of unattached/tensionless kinetochores . What particular feature of diplochromosomes triggers the SAC wait-anaphase response is unclear . It seems likely that diplochromosome structure is at least partially incompatible with attachment to the spindle . For example , it may be that the outer kinetochores block spindle attachment to the inner kinetochores within a diplochromosome , or it could be that the spindle has trouble generating tension on four kinetochores simultaneously , both of which would trigger the SAC wait-anaphase response . Eventually , the spindle appears able to attach and bi-orient all kinetochores , but the resulting anaphase is frequently error prone- with lagging chromosomes ( Figure 1I ) . This result fits with the known ability of cells with erroneous merotelic kinetochore attachments to satisfy the SAC and proceed to anaphase with lagging chromosomes ( Gregan et al . , 2011 ) . Potentially , then , the AuroraB-mediated mechanism that can correct merotely is overwhelmed/inoperable in cells with reduplicated chromosomes ( Cimini et al . , 2006; Knowlton et al . , 2006 ) . Despite the inability of the SAC to prevent all instances of mitotic errors in cells with polyteny , our data suggest that development of normally diploid tissues with an operable SAC is not noticeably altered by up to 23% ± 4 . 9% of divisions being error prone tetraploid divisions . Given the conserved nature of SAC signaling , and the widespread occurrence of diplochromosomes in disease , it will be interesting to explore whether the SAC wait-anaphase response is a general mechanism used to enable the expansion of aneuploid cells formed by aberrant genome reduplication . In contrast to cells with polyteny at anaphase , in cells such as papillar cells , SIRS vastly improves mitotic fidelity . This process does not require the Mad2-dependent wait-anaphase response , but the efficient completion of SIRS before anaphase requires the Mad2-dependent mitotic timer . Little is known about distinct , checkpoint-independent regulation of the Mad2 timer . In the future , papillar cells may prove useful in further study of the timer , given the dependence of SIRS completion on this Mad2 function . We previously described the error-prone nature of papillar divisions , as well as the high tolerance of this tissue for chromosome mis-segregation . This raises the question of why papillar cells employ SIRS , if papillar aneuploidy is well tolerated ( Schoenfelder et al . , 2014 ) . Based on our study of HS>fzr cells , which lack SIRS , we propose that SIRS is required to prevent extreme polytene chromosome mis-segregation events during papillar development , which could result in inviable nullisomic cells . Additionally , we have recently found that papillar cells actively prevent accumulation of micronuclei resulting from broken DNA ( Bretscher and Fox , 2016 ) . Thus , while mitotic genome-reduplicated cells such as papillar cells do tolerate some degree of aneuploidy , processes such as micronucleus prevention and SIRS may act to ensure a viable degree of mitotic fidelity . Our results identify SIRS as a spindle-independent chromosome separation process that , remarkably , individualizes polytene chromosomes into recent sister pairs before anaphase . This process is distinct from another spindle-independent chromosome separation process known as C-mitosis , which involves complete sister chromatid separation before anaphase ( Levan , 1938; Östergren , 1944 ) . While future work will determine what differentiates cells capable of SIRS from cells with polytenes that persist until anaphase , our data thus far has examined three layers of polytene chromosome organization that either are or are not eliminated during papillar and HS>fzr mitosis , and has pinpointed one of these layers of polytene organization as distinct to cells undergoing SIRS . The first layer of polytene organization is homolog-homolog pairing . Given that we observe the haploid number of chromosomes after both papillar ( Figure 3B ) and HS>fzr ( Figure 1F ) endocycles , it is clear that homologous chromosomes associate within both types of polytene chromosomes by somatic homolog pairing ( Metz , 1916; Painter , 1934 ) . Both mitotic papillar and HS>fzr polytenes exhibit un-paired homologs before dividing ( Figure 3B’’ , Figure 1F’ ) , and this process appears to initiate at centromeres ( Figure 1F , Figure 3B , asterisks , most obvious for the acrocentric X chromosome ) . Thus , homolog-homolog dissociation is not unique to SIRS . The second layer of polytene organization is interaction between sister chromatid pairs . Importantly , the arrangement of chromatid pairs at metaphase differs between cells that do or do not undergo SIRS . In SIRS-capable ( e . g . papillar ) cells , only the product of the single most recent round of replication ( recent sisters , see nomenclature ) remain attached at metaphase whereas the products of previous rounds of replication are no longer attached . In contrast , in SIRS-incapable ( e . g . HS>fzr ) cells , all sister chromatids remain attached . Thus , the separation into chromatid pairs appears to be the critical function of SIRS . Future work can test if this separation requires the Condensin II complex activity during SIRS , which was shown previously to enable partial polytene chromosome dissociation ( Hartl et al . , 2008 ) . The third layer of organization within polytenes are contacts between recent sister chromatid arms . These are equally undone by metaphase in both papillar and HS>fzr cells ( Figure 3B’’ , Figure 1F’ ) so that , at metaphase , chromatids are only attached at the centromere . This process likely involves the prophase cohesin removal pathway ( Losada et al . , 2002; Sumara et al . , 2002 ) . Taken together , we conclude the key difference between SIRS-capable ( e . g . papillar ) and SIRS-incapable ( e . g , HS>fzr ) cells is the ability to separate into sister chromatid pairs before metaphase ( Figure 5E ) . We further hypothesize that a key prerequisite to SIRS is the careful regulation of chromosome structure during genome reduplication/endocycles . During the endocycle , papillar cells show no evidence of karyokinesis , which suggests these cells lack a mechanical method of separating chromosomes during the endocycle ( Fox et al . , 2010 ) . However , we speculate that during endocycles , periodic cohesin removal occurs at centromeres after each S-phase . Such cohesin removal would then allow each chromatid to both eliminate its cohesins between its sister from a previous S-phase and then establish cohesins with a new sister during the subsequent S-phase . While alterations in cohesins do not noticeably perturb interphase polytene structure ( Cunningham et al . , 2012; Pauli et al . , 2008 ) , future work can determine if such endocycle-mediated cohesin regulation confers cells with polytene chromosomes with the ability to undergo SIRS during a later mitosis . Future work can also determine if cohesin regulation differs during endocycles of cells that are destined to later divide . We previously defined features of a distinct pre-mitotic variant of the endocycle , which include centriole retention and the completeness of DNA replication ( Fox et al . , 2010; Schoenfelder et al . , 2014 ) . Here , we propose that cohesin regulation may be also be distinct during this endocycle variant , and is a key factor to promoting SIRS . An additional interesting layer of SIRS regulation to explore is how it is triggered , and whether the mitotic timer is an active or passive regulator . Our data suggest NEBD is coincident with SIRS onset , possibly by allowing chromosomes to access some cytoplasmic SIRS regulator , or to initiate SIRS by releasing chromosomes from the nuclear envelope . Regarding the role of the Mad2 timer , it will be interesting to ask if it somehow senses completedness of DNA replication , which may be a pre-requisite for SIRS initiation . SIRS is likely frequent and conserved . Inspired by classical reports ( Berger , 1938; Grell , 1946; Holt , 1917 ) , we found polytene chromosomes are present before polyploid mitosis in Culex , but are later apparent as individual chromosomes during mitosis ( Figure 3—figure supplement 1E ) . Based on our results , we also suggest that chromosome dispersal in polyploid Drosophila ovarian nurse cells represents an incomplete version of SIRS ( Dej and Spradling , 1999 ) , especially given that these chromosomes can further separate if mitotic cyclins are experimentally elevated ( Reed and Orr-Weaver , 1997 ) . In polyploid trophoblasts of some mammalian species , polytene chromosomes separate into numerous bundles of paired chromosomes at the polykaryocyte stage , and thus SIRS may also occur in mammals ( Zybina et al . , 1996; Zybina et al . , 2011 ) . Similarly , SIRS may eliminate polyteny in some polyploid tumors . One of the first descriptions of polyteny in tumors noted diplochromosomes 'fall apart' before anaphase ( Levan and Hauschka , 1953 ) . Whole genome duplication is common ( ~37% ) in human tumors ( Zack et al . , 2013 ) . Given the transient nature of polytene chromosomes in mitotic tissues demonstrated here , we suggest future studies of whole genome duplication in cancer models should closely examine the initial mitosis after multiple S-phases to identify potential polytene chromosome origins of tumor aneuploidy . Finally , while our studies agree with the notion that multiple S-phases and polyploidy precede aneuploidy ( Davoli and de Lange , 2012; Gordon et al . , 2012 ) , they also underscore the need for aneuploidy-prevention responses including SIRS and the SAC for continued propagation of viable polyploid/aneuploid cells . Future studies can reveal additional SIRS regulation , and other critical genome instability controls in normal or tumorous cells following genome reduplication .
All flies were raised at 25° on standard media ( Archon Scientific , Durham , NC ) . For experiments to measure the length of mitosis larvae or pupae were shifted to 29° for at least 18 hr before dissection . Heat shocks were performed on third instar larvae . Vials were heat shocked in a 37° water bath for 15 min , 30 min , or 60 min . Flybase ( flybase . org ) describes full genotypes for the following stocks used in this study: engrailed Gal4 ( Bloomington stock 1973 ) ; w1118 ( Bloomington stock 3605 ) ; His-2av-GFP ( Bloomington stock 24163 ) ; UAS>GFP . E2f1 . 1–230 , UAS>mRFP1 . CycB . 1–266 ( Bloomington stock 55117 ) . Kyoto DGRC ( kyotofly . kit . jp ) describes the genotype for the following stock: S/G2/M-Azami ( Kyoto stock 109678 ) . The other stocks were generous gifts: tomato-Cenp-C ( Althoff et al . , 2012 ) ; HS>fzr ( Sigrist and Lehner , 1997 ) ; byn>gal4 ( Singer et al . , 1996 ) ; mad11 , Df ( 2R ) W45-30n ( Emre et al . , 2011 ) ; mad2p ( Buffin et al . , 2007 ) , msl3-GFP ( Strukov et al . , 2011 ) , stg>LacZ4 . 9 ( Bruce Edgar ) , jupiter-GFP ( Karpova et al . , 2006 ) , and BubR1-GFP ( Royou et al . , 2010 ) . Culex pipiens larvae were obtained from Carolina Biological ( Burlington , NC ) . Culturing conditions were as in Fox et al . , 2010 . Larvae were monitored hourly for pupation , and the hindgut was dissected beginning 7 hr post-puparium formation . Antibody staining was as for Drosophila tissue . 20 wandering 3rd instar larvae per replicate of the indicated genotype were placed into a fresh vial with food and then heat shocked for 15 or 30 min . The number of adults that eclosed was counted . Chromosome preparations were based on previous protocols with modifications for the pupal hindgut ( Fox et al . , 2010; Gatti et al . , 1994 ) . For colcemid treatment , to enrich for metaphase cells , tissue was first incubated in colcemid ( Sigma , St . Louis , MO ) at 50 μg/ml for 30 min in PBS . For pre-mitotic chromosome spreads with Premature Chromosome Compaction , tissue was incubated in Calyculin A ( Cell Signaling Technology , Danvers , MA ) at 200 nM in PBS for 30 min , ( Gotoh et al . , 1995; Miura and Blakely , 2011 ) . FISH was performed as in Dej and Spradling , 1999 . BAC clone #BACN04H23 ( Chromosome 3L , region 69C3-C8 ) from the PacMan collection ( Venken et al . , 2006 ) was labeled using the BioNick labeling system ( Invitrogen , Carlsbad , CA ) . BAC probe signal was amplified through sequential labeling with Peroxidase-labeled Streptavidin followed by the TSA Peroxidase detection kit ( Perkin Elmer , Waltham , MA ) . Imaging was performed on a Zeiss Axio Imager 2 with a 63x oil immersion lens . Tissue was dissected and cultured based on previous protocols with modifications for the pupal hindgut ( Fox et al . , 2010; Prasad et al . , 2007 ) . For colcemid live imaging experiments , pupae were dissected and imaged in media containing 50 μg/ml of colcemid ( Sigma , St . Louis , MO ) from the initiation of dissection to the first frame was at least 15 min and up to 1 hr . Imaging was performed on a spinning disc confocal ( Yokogawa CSU10 scanhead ) on an Olympus IX-70 inverted microscope using a 60x/1 . 3 NA UPlanSApo Silicon oil , 100x/1 . 4 NA U PlanSApo oil , or a 40x/1 . 3 NA UPlanFl N Oil objective , a 488 nm and 568 nm Kr-Ar laser lines for excitation and an Andor Ixon3 897 512 EMCCD camera . The system was controlled by MetaMorph 7 . 7 . Tissue was dissected in PBS and immediately fixed in 3 . 7% formaldehyde + 0 . 3% Triton-X for 15 min . Immunostaining was performed in 0 . 3% Triton-X with 1% normal goat serum as in Fox et al . , 2010 . The Fluorescent Ubiquitination-based Cell Cycle Indicator ( FUCCI ) probes ( Zielke et al . , 2014 ) and mouse anti-Phospho-Histone H3 ( ser 10 ) ( Cell Signaling Technology , 1:1000 ) were used to determine cell cycle stages . Rabbit anti-RFP ( MBL , Woburn , MA , 1:500 ) was used to detect Cenp-C-Tomato Foci . Rabbit anti-DCP1 ( Cell Signaling Technology , 1:500 ) was used to measure cleaved caspases . Mouse anti-γ-tubulin ( Sigma , clone GTU-88 , 1:1000 ) was used to detect centrosomes in mitotic cells . TUNEL staining was performed with the in situ cell death detection kit ( Roche , Basel , Switzerland ) according to the protocol in Schoenfelder et al . , 2014 . Tissue was stained with DAPI at 5 μg/ml . Images were obtained using a Leica SP5 inverted confocal with a 40x or 100x oil objective . Emission was done using a 405 nm diode laser , an argon laser tuned to 488 nm emission , a 561 nm Diode laser , and a 633 HeNe laser . All image analysis was performed using ImageJ ( Schneider et al . , 2012 ) . nuclear envelope brightness was calculated by measuring Geminin-Azami intensity or a single cell . The brightness for each cell was normalized from 1 to 0 , with 1 being the highest pixel intensity value and 0 being the dimmest value for a single cell across all time-points . Anaphase was determined as the first frame with poleward movement of the kinetochores as evident by Cenp-C-Tomato . Time from NEBD to anaphase was determined as the point from half-maximal Azami signal to anaphase . To calculate kinetochore clustering we closely cropped around the cell of interest for all frames . We then used a thresholding approach to outline each centromere or group of centromeres and stored those as ROIs . We then used those ROIs to measure the average intensity of each centromere pixel and the total area of all the pixels . We reasoned that as centromeres disperse the total area they cover increases and there is a corresponding decrease in fluorescence intensity from each individual point , therefore we divided the average pixel intensity by the area and normalized that on a scale from 1 to 0 . To measure symmetry of metaphase centromere alignment ( Figure 5D , D’ ) , we generated a line plot of each metaphase plate at the frame immediately prior to metaphase . We then generated aggregate plots of each genotype . 'N' refers to the haploid number of chromosome sets , while C refers to the haploid DNA content ( a diploid cell in G2 is 2N but 4C , a tetraploid cell in G1 is 4N and 4C ) . For chromosomes we use 'homolog' to distinguish maternally and paternally contributed chromosomes of the same chromosome type . All chromatids of the same homolog are considered 'sisters . ' We use 'recent sister' to refer to two chromatids that are the product of the most recent S-phase . 'Polytene' refers to the chromosome state of any cell formed by genome reduplication that has not fully separated its chromosomes into recent sisters . In a polyploid/polytene cell with somatic homolog pairing , polytene cells exhibit the haploid number of distinguishable chromosomes , whereas in a cell without homolog pairing , this number doubles . 4 chromatids conjoined at centromeres are 'diplochromosomes' . Please note that , using metaphase spreads , the presence/absence of diplochromosomes as well as the number of individual chromatids can only be scored when chromosomes are significantly condensed as in mitosis or in the presence of Calyculin A . We use the term “endocycle” to refer to any cell cycle involving successive genome reduplication without any sign of M-phase . We note the use in the literature of terms such as 'endoreduplication' , 're-replication' , and 'endoreplication' to often refer to the same phenomenon . | Before a cell divides , it duplicates all its genetic information , which is stored on chromosomes . Then , each chromosome evenly divides into two new cells so that each cell ends up with identical copies of the genetic information . Because the cellular machinery that evenly divides chromosomes is built to recognize chromosomes that were duplicated exactly once , it is important to maintain this pattern of alternating one round of duplication with one round of division . Cells that instead duplicate their chromosomes more than once can make mistakes during division that are associated with diseases such as cancer . Chromosomes with extra duplications are present in normal tissues such as the placenta of mammals . They can also occur in human diseases and may even result from chemotherapy treatment . However , we know almost nothing about how cells respond to these problematic chromosomes when dividing . By studying cells from the Drosophila melanogaster species of fruit fly , Stormo and Fox discovered two distinct ways in which cells respond to extra chromosome duplications . One response occurs in cells that were experimentally engineered to undergo an extra chromosome duplication . These cells delay division so that the chromosome separation machinery can somehow adapt to the challenge of separating more than two chromosome copies at once . The second response occurs in cells that naturally undergo extra chromosome duplications before division . In these cells , Stormo and Fox discovered a new type of chromosome separation , whereby the extra chromosome copies move apart from each other before cell division . In doing so the chromosomes can better interact with the chromosome separation machinery during division . Stormo and Fox also found that a protein named Mad2 is important in both responses , and gives the cell enough time to respond to extra chromosome copies . Without Mad2 , the separation of chromosomes with extra duplications is too hasty , and can lead to severe cell division errors and cause organs to form incorrectly . Having uncovered two new responses that cells use to adapt to extra chromosomes , it will now be important to find other proteins like Mad2 that are important in these events . Understanding these processes and the proteins involved in more detail could help to prevent diseases that are associated with extra chromosomes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"developmental",
"biology"
] | 2016 | Distinct responses to reduplicated chromosomes require distinct Mad2 responses |
Nanos RNA-binding proteins are required for germline development in metazoans , but the underlying mechanisms remain poorly understood . We have profiled the transcriptome of primordial germ cells ( PGCs ) lacking the nanos homologs nos-1 and nos-2 in C . elegans . nos-1nos-2 PGCs fail to silence hundreds of transcripts normally expressed in oocytes . We find that this misregulation is due to both delayed turnover of maternal transcripts and inappropriate transcriptional activation . The latter appears to be an indirect consequence of delayed turnover of the maternally-inherited transcription factor LIN-15B , a synMuvB class transcription factor known to antagonize PRC2 activity . PRC2 is required for chromatin reprogramming in the germline , and the transcriptome of PGCs lacking PRC2 resembles that of nos-1nos-2 PGCs . Loss of maternal LIN-15B restores fertility to nos-1nos-2 mutants . These findings suggest that Nanos promotes germ cell fate by downregulating maternal RNAs and proteins that would otherwise interfere with PRC2-dependent reprogramming of PGC chromatin .
In animals , formation of the germline begins during embryogenesis when a few cells ( ~30 in mice , two in C . elegans ) become fated as primordial germ cells ( PGCs ) – the founder cells of the germline . PGC specification requires the activity of chromatin regulators that induce genome-wide changes in gene expression . For example , in mice , the transcriptional repressor BLIMP1 initiates PGC specification by blocking the expression of a mesodermal program active in neighboring somatic cells ( Ohinata et al . , 2005; Saitou et al . , 2005 ) . In C . elegans , the NSD methyltransferase MES-4 and the Polycomb Repressive Complex ( PRC2 , including MES-2 , 3 and 6 ) cooperate to place active and repressive histone marks on germline and somatic genes , respectively ( Gaydos et al . , 2012 ) . Despite their critical roles during germ cell development , the BLIMP1 and MES/PRC2 chromatin regulators are not germline-specific factors and also function during the differentiation of somatic lineages ( Cui et al . , 2006; Gaydos et al . , 2012; Seydoux and Braun , 2006 ) . How the activities of these global regulators are modulated in germ cells to promote a germline-specific program is not well understood . In C . elegans , genetic analyses have shown that MES-dependent activation of germline genes is antagonized in somatic lineages by a group of transcriptional regulators first identified by their effects on vulval development ( Curran et al . , 2009; Petrella et al . , 2011; Unhavaithaya et al . , 2002 ) . Among these , components of the DRM ( named for its Dp/E2F , pRB , and MuvB subunits ) class of transcriptional regulators and LIN-15B , a THAP domain DNA binding protein , have been implicated in the silencing of germline genes in somatic cells ( Petrella et al . , 2011; Wu et al . , 2012 ) . Loss of DRM factors or LIN-15B causes ectopic activation of germline genes in somatic cells leading to growth arrest at elevated temperatures ( 26°C ) . Inactivation of mes-2 , mes-3 , mes-4 and mes-6 suppresses the ectopic germline gene expression and restores viability to lin-15B mutants at 26°C ( Petrella et al . , 2011 ) . These observations have suggested that DRM factors and LIN-15B antagonizes MES activity in somatic lineages to keep germline genes off ( Petrella et al . , 2011 ) . A similar antagonism , but in reverse , has been uncovered in the adult germline between the NSD methyltransferase MES-4 and the DRM transcription factor LIN-54 ( Tabuchi et al . , 2014 ) . The X chromosome is a major focus of MES repression in C . elegans germline . The X chromosome is silenced throughout germ cell development except in oocytes , which activate the transcription of many X-linked genes ( Kelly et al . , 2002 ) . mes mutants prematurely activate the transcription of X-linked genes in pregametic germ cells leading to germ cell death ( Bender et al . , 2006; Gaydos et al . , 2012; Seelk et al . , 2016 ) . Reducing the function of the synMuvB transcription factor lin-54 in mes-4 mutants restores the expression of X-linked genes closer to wild-type levels ( Tabuchi et al . , 2014 ) . Therefore in the germline , MES activity antagonizes DRM activity to keep the X chromosome silent . Together , these genetic studies suggest that competition between the MES chromatin modifiers and the DRM/LIN-15B transcription factors balance the transcription of somatic and germline genes in somatic and germline tissues . How this competition is biased during development to ensure appropriate gene expression in each tissue is not known . In this study , we have discovered a link between Nanos and LIN-15B that provides an explanation for how MES activity becomes dominant in the nascent germline ( Figure 1A ) . The C . elegans PGCs arise early in embryogenesis from pluripotent progenitors ( P blastomeres ) that also generate somatic lineages ( Figure 1B ) . RNA polymerase II activity is repressed in the P lineage until the 100 cell stage when the last P blastomere P4 divides to generate Z2 and Z3 , the two PGCs ( Seydoux et al . , 1996 ) . RNA polymerase II becomes active in PGCs , but these cells remain relatively transcriptionally quiescent , and exhibit reduced levels of active chromatin marks compared to somatic cells throughout the remainder of embryogenesis ( Kelly , 2014 ) . Active marks and robust transcription return after hatching when the L1 larva begins to feed and the PGCs resume proliferation in the somatic gonad ( Fukuyama et al . , 2006; Kelly , 2014 ) . The mechanisms that maintain PGC chromatin in a silenced state during embryogenesis are not known , but embryos lacking the nanos homologs nos-1 and nos-2 have been reported to display abnormally high levels of the active mark H3meK4 mark in PGCs ( Schaner et al . , 2003 ) . nos-1nos-2 PGCs initiate proliferation prematurely during embryogenesis and die during the second larval stage ( Subramaniam and Seydoux , 1999 ) . Nanos proteins are broadly conserved across metazoans and have been shown to be required for PGC survival in several phyla , from insects to mammals ( Asaoka-Taguchi et al . , 1999; Beer and Draper , 2013; Deshpande et al . , 1999a; Lai et al . , 2012a; Tsuda et al . , 2003 ) . Nanos proteins are cytoplasmic RNA-binding proteins that regulate gene expression post-transcriptionally by recruiting effector complexes that silence and degrade mRNAs in the cytoplasm . Six direct Nanos mRNA targets have been identified to date [Drosophila hunchback , cyclin B and hid ( Asaoka-Taguchi et al . , 1999; Dalby and Glover , 1993; Kadyrova et al . , 2007; Murata and Wharton , 1995; Sato et al . , 2007; Wreden et al . , 1997 ) , Xenopus VegT ( Lai et al . , 2012a ) , and sea urchin CNOT6 and eEF1A ( Oulhen et al . , 2017; Swartz et al . , 2014 ) ] ) , but none of these targets are sufficient to explain how Nanos activity might affect PGC chromatin . In this study , we characterize the gene expression defects of PGCs lacking nanos activity in C . elegans . Our findings indicate that nanos activity is required in PGCs to silence maternal transcripts inherited from the oocyte . We identify maternal lin-15B as a critical target of Nanos regulation that must be turned-over to establish MES dominance in PGCs .
nos-2 is provided maternally and functions redundantly with zygotically-expressed nos-1 ( Subramaniam and Seydoux , 1999 ) . To generate large numbers of larvae lacking both nos-1 and nos-2 activities , we fed hermaphrodites homozygous for a deletion in nos-1 [nos-1 ( gv5 ) ] bacteria expressing nos-2 dsRNA and collected their progenies at the L1 stage [hereafter designated nos-1 ( gv5 ) nos-2 ( RNAi ) L1 larvae] . We used fluorescence-activated cell sorting ( FACS ) to isolate PGCs based on expression of the germ cell marker PGL-1::GFP and processed the sorted cells for RNA-seq ( L1 PGCs ) . Two independent RNA-seq libraries ( biological replicates ) were analyzed for each genotype ( wild-type and nos-1 ( gv5 ) nos-2 ( RNAi ) ) using Tophat 2 . 0 . 8 and Cufflinks 2 . 0 . 2 software ( Trapnell et al . , 2012 ) . These analyses identified 461 underexpressed transcripts and 871 overexpressed transcripts in nos-1 ( gv5 ) nos-2 ( RNAi ) L1 PGCs compared to wild-type ( q < 0 . 05 , Figure 1C and Supplementary file 5 for list of misregulated genes ) . qRT-PCR of 11 genes confirmed the result of the RNA-seq analysis ( Figure 1—figure supplement 1A ) . To determine the types of genes affected , we used published gene expression data ( Gaydos et al . , 2012; Meissner et al . , 2009; Ortiz et al . , 2014; Reinke et al . , 2004; Wang et al . , 2009 ) to generate non-overlapping gene lists with preferential expression in pregametic germ cells ( germline stem cells and early meiotic cells ) , oocytes , sperms , or somatic cells ( Materials and methods and Supplementary file 1 ) . The oocyte list includes both genes required for oogenesis and maternal genes required for embryonic development . We found that 31% ( 144/461 ) of underexpressed transcripts in nos-1 ( gv5 ) nos-2 ( RNAi ) L1 PGCs correspond to genes expressed preferentially in pregametic germ cells ( Figure 1D ) . These include sygl-1 , a gene transcribed in germline stem cells in response to Notch signaling from the somatic gonad ( Kershner et al . , 2014 ) . The sygl-1 transcript was decreased by 4 . 7-fold in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs ( Supplementary file 4 ) . In contrast , overexpressed transcripts in nos-1 ( gv5 ) nos-2 ( RNAi ) L1 PGCs correspond primarily to genes expressed in oocytes ( 380/871 ) ( Figure 1D ) . These include lin-41 , a master regulator of oocyte fate ( Spike et al . , 2014a; 2014b ) and tbx-2 and hnd-1 , transcription factors that promote muscle development during embryogenesis ( Fukushige et al . , 2006; Smith and Mango , 2007 ) . The lin-41 , tbx-2 , and hnd-1 transcripts were upregulated 5 . 1-fold , 11-fold and 91-fold , respectively , in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs ( Supplementary file 4 ) . We conclude that nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs overexpress oogenic and maternal genes normally expressed in oocytes , and fail to activate pregametic genes normally expressed in PGCs . The overexpressed oocyte-class transcripts in nos-1 ( gv5 ) nos-2 ( RNAi ) L1 PGCs could correspond to maternal transcripts that failed to turnover during embryogenesis or to zygotic transcripts synthesized de novo in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs . To distinguish between these possibilities , we first examined the fate of maternal RNAs in PGCs . We isolated PGCs from embryos with fewer than 200 cells , at a time when PGCs are still mostly transcriptionally silent ( EMB PGCs ) ( Schaner et al . , 2003; Seydoux and Dunn , 1997 ) . By comparing the EMB PGC transcriptome to a published oocyte transcriptome ( Stoeckius et al . , 2014 ) , we observed an excellent correlation in relative transcript abundance between oocytes and EMB PGCs ( Figure 2—figure supplement 1A ) , suggesting that the transcriptome of EMB PGCs consists primarily of maternal mRNAs inherited from the oocyte . This finding is consistent with in situ hybridization experiments that showed that many maternal RNAs persist in the embryonic germ lineage at least to the P4 germline founder cell ( Seydoux and Fire , 1994 ) . Next , we compared the transcriptome of EMB PGCs to that of L1 PGCs to identify PGC transcripts whose abundance decline during embryogenesis . We identified 411 down-regulated transcripts , including 197 oocyte transcripts ( Figure 2A and B and Supplementary file 5 ) , consistent with turnover of many maternal mRNAs in PGCs after the 200 cell stage . Strikingly , the amplitude of this turnover was diminished in nos-1 ( gv5 ) nos-2 ( RNAi ) mutants: the abundance of the 411 transcripts remained high overall during the transition from EMB PGCs to L1 PGCs in nos-2 ( RNAi ) nos-1 ( gv5 ) embryos ( Figure 2C ) . Furthermore , when comparing wild type and nos-1 ( gv5 ) nos-2 ( RNAi ) EMB PGCs , we identified 182 differentially expressed transcripts ( 11 down- and 171 upregulated ) , including 71 of oocyte transcripts that were more abundant in nos-1 ( gv5 ) nos-2 ( RNAi ) EMB PGCs ( Figure 2—figure supplement 1B ) . Together these findings suggest a defect in maternal mRNA turnover in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs that is already detectable at the 200 cell stage and persists through embryogenesis . To test this hypothesis directly , we performed in situ hybridization experiments against three maternal mRNAs . In wild-type embryos , mex-5 , C01G8 . 1 and Y51F10 . 2 are turned over rapidly in somatic lineages ( before the 28 cell stage ) and more slowly in the germ lineage ( 200–300 cell stage for mex-5 and C01G8 . 1; bean-stage for Y51F10 . 2 ) . We found that , in nos-1 ( gv5 ) nos-2 ( ax3103 ) embryos , turnover was not affected in somatic lineages , but was delayed in PGCs , with C01G8 . 1 and mex-5 persisting to the ~500 cell stage and Y51F10 . 2 persisting to 1 . 5-fold stage ( Figure 2D ) . We conclude that turnover of maternal mRNAs is compromised in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs . By the L1 stage , PGCs are transcriptionally active . To explore whether inappropriate transcription also occurs in nos-1nos-2 PGCs by the L1 stage , we examined transcripts that increase in abundance during the transition from EMB and L1 PGCs . We identified 130 such transcripts in wild-type PGCs , including 30% in the pregamete category ( 39/130 , Figure 2B ) , consistent with PGCs transitioning to a pregamete fate by the L1 stage . In contrast , in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs , many more ( 510 ) transcripts increased in abundance during embryogenesis , and these were distributed among all genes categories , including oocyte genes ( 16% , 84/510 Figure 2E and F and Supplementary file 5 ) . This finding suggests that , unlike wild-type PGCs , nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs fail to transition to a pregamete program and instead adopt a hybrid transcriptional profile that includes activation of oocyte genes . To explore this possibility further , we used ATAC-seq to identify regions of ‘open’ chromatin that differ between wild-type and nos-1 ( gv5 ) nos-2 ( RNAi ) L1 PGCs ( Buenrostro et al . , 2015 ) . We identified 221 genes that showed increased chromatin accessibility at their promoter region in nos-1 ( gv5 ) nos-2 ( RNAi ) L1 PGCs compared to wild-type ( ‘ATAC-seq+’ genes; Figure 3 , Figure 3—figure supplement 1 , and Supplementary file 2 ) . Consistent with transcriptional activation , most of the ATAC-seq+ genes were overexpressed in nos-1 ( gv5 ) nos-2 ( RNAi ) L1 PGCs compared to wild-type ( Figure 3A ) . Furthermore , 108/221 of the ATAC-seq +genes were oocyte genes ( Figure 3B ) . In contrast , we identified 29 genes with decreased chromatin accessibility in nos-1 ( gv5 ) nos-2 ( RNAi ) compared to wild-type ( Figure 3—figure supplement 1 and Supplementary file 2 ) , most of which ( 13/29 ) were in the pregametic category ( Figure 3—figure supplement 1 ) . These observations confirm that nos-1nos-2 PGCs fail to fully activate the transcription of pregamete genes and instead activate many oocyte genes . Transcription of the X chromosome is silenced in all germ cells except in oocytes ( Kelly et al . , 2002 ) . If nos-1nos-2 PGCs are adopting an oocyte-like transcriptional program , we would expect X-linked genes to be active . Strikingly , 63% ( 139/221 ) of the ATAC-seq+ genes were X-linked ( Figure 3C ) . Furthermore , we found that , while transcripts from X-linked genes are rare in wild-type L1 PGCs ( average 4 . 7 FPKM per X-linked genes compared to 50 . 9 for autosomal genes ) , they are more abundant in nos-1 ( gv5 ) nos-2 ( RNAi ) L1 PGCs ( 9 . 6 FPKM for X-linked genes compared to 43 . 8 for autosomal genes ) ( Supplementary file 3 ) . We conclude that silencing of the X chromosome is defective in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs , consistent with these cells adopting an oocyte-like transcriptional profile . Failure to silence X-linked genes has been reported for germ cells lacking the chromatin regulators mes-2 and mes-4 ( Bender et al . , 2006; Gaydos et al . , 2012 ) . To directly compare the effect of loss of nos versus mes function in PGCs , we purified PGCs from L1 larvae derived from hermaphrodites where mes-2 or mes-4 was inactivated by RNAi ( Materials and methods ) . As expected , loss of mes-2 and mes-4 led to a significant upregulation of X-linked genes in L1 PGCs ( Figure 4A–B , Figure 4—figure supplement 2 , and Supplementary file 5 for lists of misregulated genes ) . To directly compare these changes to those observed in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs , we compared , for each genotype , the log2 fold change over wild-type for X-linked genes and for autosomal oocyte genes . As expected , we observed a strong positive correlation between mes-2 and mes-4 in both gene categories ( R = 0 . 91 and R = 0 . 76 , X-linked and autosomal oocyte genes , respectively ) ( Figure 4C and D ) . We also observed a strong correlation between mes-4 ( RNAi ) and nos-1 ( gv5 ) nos-2 ( RNAi ) ( R = 0 . 75 , Figure 4E , Figure 4—figure supplement 1 ) and mes-2 ( RNAi ) and nos-1 ( gv5 ) nos-2 ( RNAi ) ( R = 0 . 73 , not shown ) for X-linked genes . Interestingly , the correlations were weaker for autosomal oocyte genes ( R = 0 . 35 , Figure 4F ) , which tended to be more overexpressed in nos-1 ( gv5 ) nos-2 ( RNAi ) L1 PGCs . This finding is consistent with the notion that , while nos-1nos-2 and mes PGCs share a defect in the silencing of X-linked loci , nos-1nos-2 PGCs also have an additional defect in maternal mRNA turn over . MES-2 , 3 , 4 and 6 proteins are maternally-inherited and are maintained in PGCs throughout embryogenesis ( Fong et al . , 2002; Holdeman et al . , 1998; Korf et al . , 1998; Strome , 2005 ) . We observed no significant changes in mes transcripts in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs compared to wild-type ( Supplementary file 4 ) . Direct examination of MES-2 , MES-3 and MES-4 proteins confirmed that their expression patterns were unchanged in nos-1 ( gv5 ) nos-2 ( ax3103 ) or nos-1 ( gv5 ) nos-2 ( RNAi ) embryos ( Figure 4—figure supplement 2 ) . Together , these results suggest that nos-1 and nos-2 do not affect MES expression despite being required for MES-dependent silencing . MES-dependent silencing in somatic cells and adult germlines is antagonized by members of the synMuvB class of transcriptional regulators ( Petrella et al . , 2011; Tabuchi et al . , 2014 ) . To test whether synMuvB activity contributes to the nos-1nos-2 PGC phenotype , we tested whether inactivation of synMuvB genes could reduce the sterility of nos-1nos-2 animals using combinations of RNAi and mutants ( Figure 5—figure supplement 1 ) and verified positives by analyzing the sterility of triple mutant combinations ( Figure 5 ) . We found that loss-of-function mutations in lin-15B , lin-35 and dpl-1 reduced the sterility of nos-1 ( gv5 ) nos-2 ( ax3103 ) from >70% to<30% . ( Figure 5A ) . The most dramatic reduction was seen with lin-15B ( n744 ) , which reduced nos-1 ( gv5 ) nos-2 ( ax3103 ) sterility to 3 . 4% ( Figure 5 ) . lin-15B is a THAP domain DNA binding protein that has been implicated with the DRM class of transcriptional regulators , including lin-35 and dpl-1 , in the silencing of germline genes in somatic cells ( Araya et al . , 2014; Petrella et al . , 2011; Wu et al . , 2012 ) . Other DRM components ( efl-1 , lin-37 , lin-9 , lin-52 , lin-54 ) , however , did not suppress nos-1 ( gv5 ) nos-2 ( ax3103 ) sterility ( Figure 5—figure supplement 1 ) . Since PGCs lacking mes and nos-1nos-2 shared the same defect in X chromosome silencing ( Figure 4E , Figure 4—figure supplements 1 and 2 ) , we tested whether loss of lin-15B could also suppress mes-2 and mes-4 maternal-effect sterility . Hermaphrodites derived from mes-2 ( ok2480 ) and mes-4 ( ok2326 ) mothers are 100% sterile ( Figure 5A and Figure 5—figure supplement 2 ) . We found that lin-15B ( n744 ) suppressed mes-2 ( ok2480 ) and mes-4 ( ok2326 ) sterility weakly and only for one generation . Animals derived from mes-2 ( ok2480 ) ;lin-15B ( n744 ) mothers were 83% sterile in the first generation and 98% sterile in the second generation and could not be maintained as a selfing population ( Figure 5A and Figure 5—figure supplement 2 ) . In contrast , nos-1 ( gv5 ) nos-2 ( ax3103 ) ; lin-15B ( n744 ) triple mutant animals were almost fully fertile ( 96 . 6% fertile , Figure 5A ) and could be maintained as a selfing population for >10 generations . The fertility of nos-1 ( gv5 ) nos-2 ( ax3103 ) ;lin-15B ( n744 ) hermaphrodites was dependent on mes activity: inactivation by RNAi of mes-2 or mes-4 in nos-1 ( gv5 ) nos-2 ( ax3103 ) ;lin-15B resulted in 100% sterility ( Figure 5—figure supplement 3 ) . These genetic observations suggest that the sterility of nos-1nos-2 mutants is due , at least in part to , inappropriate inhibition of MES function in PGCs by LIN-15B . LIN-15B has been reported to be broadly expressed in somatic cells ( Sarov et al . , 2012 ) , but its expression pattern during germline development was not known . We used a polyclonal antibody generated against LIN-15B protein ( modencode project , personal communication with Dr . Susan Strome ) to examine LIN-15B expression in the adult germline and in embryos . We confirmed the specificity of this antibody by staining lin-15B ( n744 ) mutant , which showed no nuclear staining ( Figure 6—figure supplement 1 ) . We first detected LIN-15B expression in the germline in the L4 stage in nuclei near the end of the pachytene region where germ cells initiate oogenesis ( Figure 6A ) . Nuclear LIN-15B was present in all oocytes and inherited by all embryonic blastomeres , including the germline P blastomeres ( Figure 6B and Figure 6—figure supplement 1 ) . LIN-15B remained present at high levels in all somatic nuclei throughout embryogenesis . In contrast , in the germ lineage , LIN-15B levels decreased sharply during the division of the germline founder cell P4 that generates the two PGCs ( Figure 6B Left panels ) . LIN-15B expression remained at background levels in PGCs throughout embryogenesis . lin-15B transcripts were modestly elevated in nos-1 ( gv5 ) nos-2 ( RNAi ) EMB PGCs compared to wild-type EMB PGCs , suggesting that lin-15B may be one of the maternal RNAs that requires Nanos activity for rapid turnover in PGCs ( Supplementary file 4 ) . During the transition from EMB to L1 PGCs , lin-15B transcript levels rose by ~2 fold in nos-1 ( gv5 ) nos-2 ( RNAi ) embryos but not in wild-type embryos , suggesting that the lin-15B locus is also inappropriately transcribed in nos-1 ( gv5 ) nos-2 ( RNAi ) L1 PGCs . Unfortunately , we were not able to confirm these RNA-seq observations by in situ hybridization due to the low abundance of lin-15B RNA and its presence in all somatic cells . To determine whether LIN-15B protein expression is under the control of Nanos activity , we stained embryos derived from nos-1 ( gv5 ) nos-2 ( ax3103 ) hermaphrodites with the anti-LIN-15B antibody . We found that , in contrast to wild-type , nos-1 ( gv5 ) nos-2 ( ax3103 ) embryos maintained high LIN-15B levels in embryonic PGCs ( Figure 6B Right panels ) . nos-1 ( gv5 ) nos-2 ( ax3103 ) embryos could misregulate LIN-15B by delaying the turnover of maternal LIN-15B or by activating premature zygotic transcription of the lin-15B locus . To distinguish between these possibilities , we created a lin-15B transcriptional reporter by inserting a GFP::H2B fusion at the 5’ end of lin-15B locus in an operon configuration to preserve endogenous lin-15B expression ( Figure 6—figure supplement 1 and Supplementary file 6 ) . We crossed nos-1 ( gv5 ) males carrying the lin-15B transcriptional reporter to wild-type or nos-1 ( gv5 ) nos-2 ( ax3103 ) hermaphrodites and examined crossed progenies for GFP expression . In both cases , we observed strong GFP expression in somatic cells , but no expression in PGCs during embryogenesis ( data not shown ) , indicating that zygotic expression of LIN-15B in PGCs is not activated in embryogenesis in either wild-type or nos-1 ( gv5 ) nos-2 ( ax3103 ) embryos . In wild-type , we first observed zygotic expression of the lin-15B transcriptional reporter in the germline of L4 stage animals ( Figure 6—figure supplement 1 ) , in germ cells that have initiated oogenesis . In contrast , in animals derived from nos-1 ( gv5 ) nos-2 ( ax3103 ) mothers , zygotic expression of the lin-15B transcriptional reporter could be detected as early as the L1 stage in PGCs and their descendants ( Figure 6C ) . This expression was maintained until the L2 stage when nos-1nos-2 PGC descendants undergo cell death . We conclude that nos-1nos-2 activity is required both to promote the turnover of maternal LIN-15B in PGCs during embryonic development and to prevent premature zygotic transcription of the lin-15B locus in PGCs in the L1 stage . To determine whether misregulation of maternal or zygotic lin-15B is responsible for nos-1nos-2 sterility , we compared the sterility of nos-1 ( gv5 ) nos-2 ( ax3103 ) animals that lack either maternal or zygotic lin-15B ( Figure 6D and Figure 6—figure supplement 2 ) . We found that loss of maternal lin-15B was sufficient to fully suppress nos-1 ( gv5 ) nos-2 ( ax3103 ) sterility , even in the presence of one zygotic copy of lin-15B ( Figure 6D ) . The penetrance of the suppression was dependent on the dosage of maternal lin-15B . nos-1 ( gv5 ) nos-2 ( ax3103 ) animals with only one copy of maternal lin-15B were only 32% sterile compared to 70% sterility for animals with two copies of maternal lin-15B and 0% with animals with zero copies of maternal lin-15B ( Figure 6D ) . Interestingly , animals with only one copy of maternal LIN-15B appeared sensitive to the zygotic dosage of lin-15B ( Figure 6D , compare the sterility M1Z2 , M1Z1 and M1Z0 ) . We conclude that maternal lin-15B is primarily responsible for the sterility of nos-1nos-2 animals , although zygotic Lin-15B activity may also contribute . LIN-15B is a transcription factor with many targets in somatic cells but no known function in the germline ( Niu et al . , 2011 ) . To determine the effect of ectopic LIN-15B on the transcriptome of nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs , we profiled nos-1 ( gv5 ) nos-2 ( RNAi ) ;lin-15B ( RNAi ) PGCs and compared the log2 fold change of transcripts in nos-1 ( gv5 ) nos-2 ( RNAi ) ;lin-15B ( RNAi ) PGCs and nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs to wild-type . We found that loss of lin-15B reduced gene misexpression in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs ( Figure 6E ) . Of the 1430 upregulated genes in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs , 31% ( 452 ) had significantly lower expression levels in nos-1 ( gv5 ) nos-2 ( RNAi ) ;lin-15B ( RNAi ) PGCs ( Figure 6F ) . Both upregulated and down-regulated gene categories were at least partially rescued , as well as X-linked and oocyte genes ( Figure 6—figure supplement 3 ) . These data indicate that ectopic lin-15B activity is responsible for a significant number of misexpressed genes in nos-1 ( gv5 ) nos-2 ( RNAi ) PGCs . To determine whether loss of lin-15B also rescues the defect in maternal RNA turnover in nos-1nos-2 PGCs , we performed in situ hybridization on nos-1 ( gv5 ) nos-2 ( ax3103 ) ;lin-15B ( n744 ) embryos . We found that the turnover of mex-5 and C01G8 . 1 mRNAs was still delayed in these embryos , as is observed in nos-1 ( gv5 ) nos-2 ( ax3103 ) ( Figure 2—figure supplement 2 ) . We conclude that loss of lin-15B does not rescue the delay in maternal mRNA turnover observed in nos-1 ( gv5 ) nos-2 ( ax3103 ) PGCs . By comparing the lists of upregulated genes in nos-1 ( gv5 ) nos-2 ( RNAi ) and mes-4 ( RNAi ) PGCs and of down-regulated genes in nos-1 ( gv5 ) nos-2 ( RNAi ) ;lin-15B ( RNAi ) PGCs ( Supplementary file 5 ) , we identified 88 shared genes , including 70 X-linked genes . Among these is utx-1 , an histone demethylase specific for the H3K27me3 mark generated by mes-2 ( Agger et al . , 2007; Seelk et al . , 2016 ) . Like other X-linked genes , utx-1 transcripts are rare in wild-type PGCs ( FPKM <0 . 2 ) ( Supplementary file 4 ) and are overexpressed 9 . 1-fold in mes-2 ( RNAi ) PGCs . In nos-1 ( gv5 ) nos-2 ( RNAi ) L1 PGCs , the utx-1 locus acquires a new ATAC-seq peak ( Figure 3—figure supplement 1 ) and utx-1 transcripts are overexpressed 160-fold . This overexpression was reduced significantly by 2 . 4-fold in nos-1 ( gv5 ) nos-2 ( RNAi ) ;lin-15B ( RNAi ) PGCs . These observations suggest that utx-1 may function downstream or in parallel to lin-15B to further antagonizes MES activity as X-linked genes become desilenced . If so , loss of utx-1 should alleviate nos-1nos-2 sterility . Consistent with this prediction , we found that reduction of utx-1 activity by RNAi partially suppressed nos-1 ( gv5 ) nos-2 ( ax3103 ) sterility ( Figure 5—figure supplement 1 ) . Suppression by utx-1 was not as extensive as that observed with lin-15B , suggesting that utx-1 is not the only gene activated in nos-1 ( gv5 ) nos-2 ( ax3103 ) PGCs that leads to sterility . These results suggest that activation of utx-1 may participate in a regulatory loop downstream of maternal LIN-15B that further weakens mes activity in nos-1nos-2 animals .
During oogenesis , oocytes stockpile mRNAs and proteins in preparation for embryogenesis . These include mRNAs and proteins with housekeeping functions as well as factors required to specify embryonic cell fates ( somatic and germline ) . During embryogenesis , maternal products are eventually turned over to make way for zygotic factors ( maternal-to-zygotic transition ) . Our findings suggest that Nanos facilitates this transition in PGCs by accelerating the turnover of maternal mRNAs . Nanos family members are thought to silence mRNAs by interacting with the sequence-specific RNA-binding protein Pumilio and with the CCR4-NOT deadenylase complex , which interferes with translation and can also destabilize RNAs . ( Lai et al . , 2012a; Suzuki et al . , 2012; Swartz et al . , 2014; Wharton et al . , 1998 ) . In the C . elegans genome , there are eight genes related to Drosophila pumilio . Depletion of five of these ( fbf-1 , fbf-2 , puf-6 , puf-7 and puf-8 ) phenocopies the nos-1nos-2 PGC phenotypes , including failure to incorporate in the somatic gonad , premature proliferation , and eventually cell death ( Subramaniam and Seydoux , 1999 ) . These observations suggest that NOS-1 and NOS-2 function with Pumilio-like proteins to repress the translation of certain maternal RNAs . Paradoxically , in sea urchins , Nanos silences the mRNA coding for the CNOT6 deadenylase , which indirectly stabilizes other maternal mRNAs ( Swartz et al . , 2014 ) . In that system , Nanos was also found to silence eEF1A expression , leading to a transient period of translational quiescence in PGCs ( Oulhen et al . , 2017 ) . In combination , these effects could promote the turnover of maternal mRNAs and proteins that promote somatic development ( e . g . LIN-15B ) while preserving germline mRNAs ( e . g . mes ) whose translation could be reactivated at a later time . In C . elegans , the redundant nanos homologs nos-1 and nos-2 are expressed sequentially in PGCs during the maternal-to-zygotic transition and may have distinct targets . Genetic analyses already have suggested that nos-1 and nos-2 have both shared and unique functions ( Kapelle and Reinke , 2011; Mainpal et al . , 2015 ) . It will be important to determine whether nos-1 and nos-2 both target lin-15B , and whether they do so directly , by silencing lin-15B mRNA translation , or indirectly , by silencing other factors required for LIN-15B protein translation and/or stability . Two lines of evidence indicate that LIN-15B is responsible for much of the abnormal gene expression observed in nos-1nos-2 PGCs by the first larval stage . First , elimination of maternal LIN-15B restores fertility to nos-1nos-2 mutants and lessens the upregulation of many misregulated genes ( Figure 6 ) . Second , LIN-15B is a known genetic antagonist of MES function in somatic cells ( Petrella et al . , 2011; Wang et al . , 2005 ) , and PGCs lacking mes activity upregulate many of the same genes upregulated in nos-1nos-2 PGCs . The strongest correlation is seen for genes on the X chromosome ( Figure 4E , Figure 4—figure supplement 1 ) , a well-documented focus of MES-dependent silencing ( Bender et al . , 2006; Garvin et al . , 1998; Gaydos et al . , 2012 ) . Together these findings indicate that failure to downregulate maternal LIN-15B interferes with MES-dependent reprogramming of PGC chromatin and is the primary cause of PGC death in Nanos mutants . The lin-15B locus is on the X chromosome and is ectopically transcribed in nos-1nos-2 PGCs at hatching . These observations raise the possibility that maternal LIN-15B potentiates zygotic lin-15B expression as MES-dependent silencing of the X-chromosome becomes compromised . How LIN-15B opposes MES activity is not known , but another X-linked gene , utx-1 , may oppose MES activity directly . utx-1 encodes a de-methylase that removes the silencing mark deposited by the PRC2 complex . Upregulation of utx-1 was shown recently to promote reprogramming of adult germline stem cells into neurons ( Seelk et al . , 2016 ) . In nos-1nos-2 PGCs , utx-1 is upregulated in a lin-15B-dependent manner , and RNAi of utx-1 partially suppresses nos-1nos-2 sterility ( Figure 5—figure supplement 1 ) . Suppression by loss of utx-1 is weaker than that observed when inactivating lin-15B , suggesting that utx-1 is not the only lin-15B target that opposes PRC2 . Loss of two other synMuvB genes , lin-35/Rb and dpl-1 , also suppresses nos-1nos-2 sterility ( Figure 5A ) , albeit again less stringently than loss of lin-15B . It will be interesting to determine whether these genes function with , or in parallel to , LIN-15B to oppose PRC2 activity in PGCs . Recently , nos-2 was shown to function redundantly with xnd-1 to repress histone active marks in PGCs ( Mainpal et al . , 2015 ) . XND-1 is a chromatin-associated protein expressed in PGCs throughout embryogenesis . An exciting possibility is that XND-1 directly activates MES-dependent remodeling in PGCs . In that context , Nanos could promote germ cell fate simply by eliminating any maternal factors that would interfere with that remodeling . Because PGCs derive from embryonic blastomeres that also give rise to somatic lineages , they inherit many transcripts with somatic functions . In addition to LIN-15B , we have found that Nanos accelerates the turn-over of several maternal mRNAs coding for transcription factors that function in somatic embryonic lineages , including pha-4 , hlh-1 and tbx-2 ( Supplementary file 4 ) . We speculate that perdurance of these somatic transcription factors contributes to the complex transcriptional profile of nos-1nos-2 PGCs . The primary function of Nanos may be , therefore , to clear the PGCs of any mRNAs that promote somatic development . This interpretation is consistent with previous studies in Drosophila and Xenopus that reported the expression of somatic transcripts in PGCs lacking Nanos ( Deshpande et al . , 1999b; Hayashi et al . , 2004; Kadyrova et al . , 2007; Lai et al . , 2012b; Oulhen et al . , 2017; Swartz et al . , 2014 ) . Our genetic findings indicate that , in PGCs , Nanos opposes LIN-15B and DRM transcription factors . Studies in Drosophila and mammals have reported that , in somatic cells , the reverse is true: DRM transcription factors silence Nanos . Loss of the DRM subunit lethal ( 3 ) malignant brain tumor [l ( 3 ) mbt] leads to tumorous growth in Drosophila imaginal disks and ectopic expression of germline genes , including nanos ( Janic et al . , 2010 ) . Similarly , loss of the retinoblastoma transcription factor ( Rb ) leads to activation of nanos transcription in mammalian tissue culture cells and in Drosophila wings , which in turn is thought to repress the translation of Rb targets ( Miles and Dyson , 2014; Miles et al . , 2014 ) . A complex regulatory feedback loop has also been reported between the LSD1 demethylase and the Nanos partner Pumilio in Drosophila and human bladder carcinoma cells ( Miles et al . , 2015 ) . Taken together , these observations suggest that Nanos functions in an ancient transcriptional/post-transcriptional regulatory switch that controls gene expression during development . Key questions for the future will be to understand how the switch is activated in the embryonic germline to favor germ cell development ( what turns on Nanos expression in PGCs ? ) , how the switch is flipped back during oogenesis to favor somatic development ( what activates LIN-15B expression in oocytes and embryos ? ) , and how the switch becomes deregulated in malignancies .
C . elegans was cultured according to standard methods ( Brenner , 1974 ) . RNAi knockdown experiments were performed by feeding on HT115 bacteria ( Timmons and Fire , 1998 ) . Feeding constructs were obtained from Ahringer or OpenBiosystem libraries or PCR fragments cloned into pL4440 . The empty pL4440 vector was used as negative control . Bacteria were grown at 37°C in LB +ampicillin ( 100 µg/mL ) media for 5–6 hr , induced with 5 mM IPTG for 30 min , plated on NNGM ( nematode nutritional growth media ) +ampicillin ( 100 µg/mL ) +IPTG ( 1 mM ) plates , and grown overnight at room temperature . Embryos isolated by bleaching gravid hermaphrodites , or synchronized L1s hatched in M9 , were put onto RNAi plates . For sterility counts , the progeny of at least six gravid adult hermaphrodites were tested . Adult progenies were scored for empty uteri ( ‘white sterile’ phenotype ) on a dissecting microscope . For all immunostaining and smFISH experiments shown in Figures 2D , 6A and B , Figure 2—figure supplement 2 , Figure 4—figure supplement 1 and Figure 6—figure supplement 1 , worms were grown at 25°C . For live embryo imaging and synMuvB related experiments shown in Figure 5 , Figure 5-figure supplement 1 , Figure 6C and D and Figure 4—figure supplement 1C , worms were grown at 20°C . To verify the efficiency of RNAi treatments used to create sequencing libraries , we scored animals exposed to the same RNAi feeding conditions for maternal-effect sterility . For nos-1 ( gv5 ) strain on nos-2 RNAi , sterility was 81 ± 10% at 20°C and 86 ± 6% at 25°C; mes-2 ( RNAi ) maternal effect sterility was 51 ± 1 . 4% and mes-4 ( RNAi ) maternal effect sterility was 95 . 5 ± 3 . 5% . To test the efficiency of the double RNAi treatment for nos-1 ( gv5 ) nos-2 ( RNAi ) ;lin15B ( RNAi ) RNA-seq libraries , we performed two additional controls . First we exposed a nos-2::FLAG strain ( Paix et al . , 2014 ) to the same RNAi feeding conditions and stained the embryos with α-FLAG antibody to confirm knock down of nos-2 ( 4/15 embryos showed weak staining , compared to 15/15 embryos with strong staining in the untreated controls ) . Second , we exposed a lin-15B::GFP strain ( Paix et al . , 2014 ) to the same double RNAi feeding conditions and observed no GFP expression in embryos . nos-1 ( gv5 ) nos-2 ( RNAi ) ;lin15B ( RNAi ) animals gave 34 ± 19% sterile progenies . See Supplementary files 6 ( CRISPR/Cas9 strain table ) and key resources table for lists of strains and CRISPR reagents . The nos-2 ( ok230 ) allele removes the nos-2 coding region and a flanking exon in the essential gene him-14 , resulting in embryonic lethality . To create a nos-2 null allele that does not affect him-14 function , we deleted the nos-2 open reading frame using CRISPR/Cas9-mediated genome editing ( Paix et al . , 2015 ) . Consistent with previous reports ( Mainpal et al . , 2015; Subramaniam and Seydoux , 1999 ) , nos-2 ( ax3103 ) animals are viable and fertile and nos-1 ( gv5 ) nos-2 ( ax3103 ) double mutants are maternal effect sterile ( Figure 5A ) . Adult worms were placed on 3-wells painted slides in M9 solution ( Erie Scientific co . ) and squashed under a coverslip to extrude embryos . Slides were frozen by laying on pre-chilled aluminum blocks for >10 min . Embryos were permeabilized by freeze-cracking ( removal of coverslips from slides ) followed by incubation in methanol at −20°C for 15 min , and then in pre-chilled acetone at −20°C for 10 min . Slides were blocked twice in PBS-Tween ( 0 . 1% ) -BSA ( 0 . 1% ) for 15 min at room temperature , and incubated with 75 μl primary antibody overnight at 4°C in a humidity chamber . Antibody dilutions ( in PBST/BSA ) : Rabbit α-LIN-15B 1:20 , 000 ( SDQ3183 , gift from Dr . Susan Strome ) , Rabbit α-MES-4 1:400 ( Gift from Dr . Susan Strome ) , mouse K76 1:10 ( DSHB ) , Rat α-OLLAS-L2 1:200 ( Novus Biological Littleton , CO ) , Rat α-OLLAS 1:80 ( Gift from Dr . Jeremy Nathans ) , mouse α-FLAG M2 1:500 ( Sigma F3165 ) . Secondary antibodies ( Molecular Probes/Thermo Fisher Sci . ) were applied for 1 ~ 2 hr at room temperature . MES-3 was tagged with the OLLAS epitope at the C-terminus using CRISPR genome editing ( Paix et al . , 2015 ) . Fluorescence microscopy was performed using a Zeiss Axio Imager with a Yokogawa spinning-disc confocal scanner . Images were taken and stored using Slidebook v6 . 0 software ( Intelligent Imaging Innovations ) using a 40x or 63x objective . Embryos were staged by DAPI-stained nuclei in optical Z-sections and multiple Z-sections were taken to include germ cells marked by α-PGL-1 ( K76 ) staining . For images of embryonic PGCs , a single Z-section was extracted at a plane with the widest area of DAPI staining for nuclear signal of LIN-15B , MES-3 , and MES-4 . For MES-2-GFP , the Z-section was determined based on widest area of GFP signal . Equally normalized images were first taken by Slidebook v6 . 0 , and contrasts of images were equally adjusted between control and experimental sets using Image J . RNAi treatments for sorting experiments were done by seeding synchronized L1 ( hatched from embryos incubated in M9 overnight ) onto RNAi plates and growing them to gravid adults . Additional RNAi or control bacteria were added once to ensure enough food to support development . Early embryos were harvested from gravid adults . These embryos were either used directly to isolate embryonic PGCs or incubated for 12 ~ 16 hr in M9 solution until reaching the L1 stage for PGCs isolation . To isolate L1 PGCs from fed animals , the L1s were plated onto RNAi plates for additional 5 hr before processing for PGC isolation . For RNA-seq experiments described in Figure 1 and Figure 2 , RNAi treatments were done at 25°C . For the rest of RNA-seq experiments , RNAi treatments were done at 20°C . See Supplementary file 7 for sequencing library information . To isolate PGCs from embryos , cell dissociation was performed as described in Strange et al . ( 2007 ) ( Strange et al . , 2007 ) with the following modifications: 106 embryos were treated in 500 µl chitinase solution ( 4 . 2 unit of chitinase ( Sigma # C6137 ) in 1 ml of conditioned egg buffer ) . After chitinase treatment , embryos were collected by centrifugation at ~900 xg for 4 mins at 4°C and resuspended in 500 μl accumix-egg buffer solution for dissociation ( Innovative Cell Techologies , AM105 , 1:3 dilution ratio in egg buffer ) . In the final step , cells were resuspended in chilled egg buffer before sorting using BD FACSAriaII . 65 , 000 ~ 120 , 000 PGL-1::GFP PGCs were used for RNA isolation . To isolate PGCs from L1 larvae , total of >5 million L1 divided into ~500 , 000 L1 per reaction were used for cell dissociation as described in Zhang and Kuhn ( Zhang and Kuhn , 2013 ) ( www . wormbook . org/chapters/www_cellculture/cellculture . html#sec6-2 ) with the following modifications: starved and fed ( for 5 hr ) L1 were incubated with freshly thawed SDS-DTT solution for 2 min and 3 min , respectively , with gentle agitation using a 1000 μl pipette tip . Pronase treatment was performed using 150 μl of 15 mg/ml pronase ( Sigma P6911 ) . Pronase treatment was stopped by adding 1000 μl conditioned L-15 medium and spin at 1600xg for 6 min . Cells were resuspended in chilled egg buffer and washed three times to remove debris before sorting using BD FACSAriaII or Beckman Coulter MoFlo sorter . ~75 , 000 sorted cells were pelleted at 1600xg for 5 min , snap frozen and saved in −80°C for later RNAseq analysis . To assay the purity of isolated PGCs , aliquot of sorted PGCs were either passed through FACS sorter again to re-analyzed their GFP expression or subjected to GFP positive cell counting under microscope . PGC purity is defined by the percentage of GFP positive and propidium iodide negative in the sorted population . The purity of sorted embryonic PGCs is 95 . 7 ± 3 . 8% ( N = 3 ) ; The purity of sorted L1 PGCs is 94 . 7 ± 4 . 7% ( N = 10 ) . From purified embryonic cells , we identified 1347 PGC enriched genes ( enrichment over somatic blastomeres ) . We cross-reference our embryonic PGC enriched gene set with other published PGC or germline enriched gene sets . 392/1347 embryonic PGC enriched genes were identified as PGC enriched genes in Spencer et al . , 2011 ( in which 979 genes with enriched expression in Z2/Z3 ) ; 700/1347 were characterized as either germline specific or germline enriched genes in Gaydos et al . , 2012 . The result is summarized in Supplementary file 8 . The reproducibility of sorting/RNAseq procedure is demonstrated by PCA analysis as described in the section of RNAseq library preparation and analysis . RNA was extracted from sorted cells using TRIZOL . The aqueous phase was transferred to Zymo-SpinTM IC Column ( Zymo research R1013 ) for concentration and DNase I treatment as described in manual . RNA quality was assayed by Agilent Bioanalyzer using Agilent RNA 6000 Pico Chip . All RNAs used for library preparation had RIN ( RNA integrity number ) >8 . Three different RNA-seq library preparation methods were used for this study: SMART-seq , which uses poly-A selection ( Figures 1 and 2 ) , NuGEN Ovation , which uses random priming ( Figure 2—figure supplement 1 , top ) , and Truseq combined with Ribozero to remove ribosomal RNAs ( all other figures ) . The first two methods allow library construction from <10 ng of total RNA , whereas the later method requires >50 ng total RNA . We compared SMART-seq and Truseq-Ribozero performance on L1 PGCs isolated from wild-type and nos-1 ( gv5 ) nos-2 ( RNAi ) and observed identical trends , with an overall higher number of misregulated genes identified with Truseq-Ribozero ( Compare Figure 1 ( SMART-seq ) and Figure 1—figure supplement 1B , C and D ( Truseq/Ribozero ) . For the experiment shown in Figure 2—figure supplement 1 ( top panels ) where we compared RNA levels between embryonic PGCs and an oocyte library reference , we used Nugen Ovation libraries which avoids any bias due to poly-A selection while allowing library construction from <3 ng of RNA . For all experiments , control and experimental libraries were made using the same method . Supplementary file 5 contains lists of misregulated genes from analyses . Supplementary file 7 lists all the RNA-seq libraries used in this study and the corresponding figures . SMART-seq libraries: libraries were made from 2 ng of total RNA isolated from sorted PGCs from worms grown at 25°C . Libraries were constructed using SMART-seq v4 Ultra Low input RNA kit ( Clontech , Cat . No . 634888 ) followed by Low Input Library Prep Kit ( Clontech , Cat . No . 634947 ) . The cDNAs were then fragmented using Covaris AFA system at the Johns Hopkins University microarray core and cloned using the Low Input library prep Kit . NuGEN Ovation libraries: libraries were made from 3 ng of total RNA isolated from sorted cells from worms grown at 25°C . Libraries were constructed using Nugen Ovation system V2 ( #7102–08 ) followed by Nugen Ultralow library system . TruSeq libraries: 50 ng of total RNA isolated from sorted PGCs from L1 worms grown at 20°C were subjected to Ribozero kit ( illumina , MRZE706 ) to remove rRNA . Libraries were constructed using Truseq Library Prep Kit V2 . All cDNA libraries were sequenced using the Illumina Hiseq2000/2500 platform . Differential expression analysis was done using Tophat ( V . 2 . 0 . 8 ) and Cufflink ( V . 2 . 0 . 2 ) . Cuffdiff accepts multiple biological replicates and uses Benjamini–Hochberg multiple hypothesis to compute false discovery rate ( FDR ) . The cutoff of FDR ( q value ) =0 . 05 was used as a significance cutoff for all the analyses in this study . The command lines for Tuxedo suit are listed as below: For each biological sample , sequencing reads were first mapped to ce10 reference genome using tophat2: $ tophat2 -p 12 g 1 --output-dir<output > segment-length 20 --min-intron-length 10 --max-intron-length 25000 G < gene . gtf> --transcriptome-index<Name . fastq> For differential gene expression analysis , sets of independent mutant and control mapped reads ( e . g biological replicates ) were used in cuffdiff analysis: $ cuffdiff -p 12 -o < output > compatible-hits-norm --upper-quartile-norm -b < genome . fa> <genes . gtf> <tophat output_sample 1 , tophat output_sample 2 , tophat output_sample 3 , . . > <tophat output_control1 , tophat output_control2 , tophat output_control3 , . . > Gene set enrichment analysis for four different categories and correlation of gene expression were done using R functions . R function intersect ( ) was used to extract overlapping lists . Plots were drawn using R package and Prism software . For correlation plots of gene expression shown in Figure 4C–F , information from different pairs of cuffdiff analyses ( WT vs mes-2 , WT vs mes-4 and WT vs nos-1/2 ) was used . Genes with sufficient aligned reads to pass statistical test ( OK status in test status from cuffdiff output ) were kept , and those without enough alignments ( NOTEST , LOWDATA in test status ) , or other conditions prevent statistical testing were excluded . Values of Log2 fold change were extracted from each cuffdiff output file and list of genes were further consolidated to generated correlation plots . The data process results in different number of genes in selected categories ( 1173 vs 1250 in X-linked genes , and 1063 vs 1092 in autosomal oocyte genes ) . However , majority of genes were overlapped between comparisons ( 1117 for X-linked genes and 1062 for autosomal oocyte genes ) In Figure 6F , the area-proportional Venn diagram was created using the VennDiagram R package . For comparisons shown in Figure 2—figure supplement 1A , oocyte transcriptome data was extracted from Stoeckius et al . ( 2014 ) , and embryonic soma and germ cells expression profiles were from this study ( Supplementary file 7 ) . Expression of each gene was log10 transformed , ranked and ordered . Correlations were plotted using custom R codes and can be found in Figure 2—figure supplement 1A source code . Principal component Analysis ( PCA ) was used to evaluate reproducibility of RNA-seq experiments . PCA revealed clustering of biological replicates with the same library preparation procedure as shown in Figure 2—figure supplement 3 . In Figure 2—figure supplement 3A , two different sets of libraries ( one set was made with NuGEN protocol and the other was made with SMART-seq protocol ) were generated using the same RNA and clustered differently , suggesting different library making procedures could introduce biases . Sequence reads were mapped to transcriptome version ce10 using Hisat2 . HTseq-count was used to generate raw counts for each gene . The command lines are listed as below . $hisat2 -x < hisat2-index> -S < output file> -q < iinput file> --known-splicesite-infile<elegans_splicesites . txt> --no-softclip $htseq-count -s no <genes . gtf> > outputfile . genecount The gene count information from HTseq-count ( Supplementary file 9 ) was subject to regularized log transformation ( rlog ) and plotPCA in DEseq2 package . We defined four gene categories based on expression characteristics reported in published microarray , serial analysis of gene expression ( SAGE ) , and RNAseq data sets that profiled specific tissues or whole worms with or without a germline ( Gaydos et al . , 2012; Meissner et al . , 2009; Ortiz et al . , 2014; Reinke et al . , 2004; Wang et al . , 2009 ) . The oocyte category ( 1594 genes ) and sperm category ( 2042 genes ) are genes with differential enrichment in dissected female gonads from adult fog2 ( q71 ) animal compared to dissected male gonads from adult fem-3 ( q96 ) animals ( Ortiz et al . , 2014 ) . The soma category ( 2684 genes ) was obtained by taking genes with SAGE tags in at least one somatic tissue ( intestine , muscle , or nerve ) as described in Gaydos et al . , 2012 , and substracting from that list all genes in the oocyte and sperm categories described above . The pregamete category ( 1694 genes ) was constructed by adding the germline-enriched and the germline-specific gene sets from Gaydos et al . , 2012 and substracting from that list all genes in the oocyte and sperm categories described above . Germline-enriched genes include genes whose expression is significantly higher in germline based on comparison of adults with and without a germline ( Reinke et al . , 2004 ) . Germline-specific genes are those with SAGE tags in dissected germlines and not in somatic tissues ( intestine , muscle and nerve cells ) . For gene sets enrichment test , we used total number of 15851 expressed genes with RPKM >0 . 1 as the cutoff from our PGC RNA-seq experiments to calculate ‘expected’ values for each category . Expected value = ( No . of significantly changed genes ) x ( No . of genes in category/15851 ) . Hypergeomatric test was performed to derived p-values ( hypergeometric probability ) , and listed in figure legends . ATAC-seq was performed as described in Buenrostro et al . , 2015 . Experimental pipeline was described as follows: 30 , 000 sorted L1 PGCs were washed with 60 μl cold cell culture grade PBS once and spun at 2000xg for 10 min . Cell nuclei were isolated by resuspending cell pellets in cold lysis buffer ( 10 mM Tris-Cl pH7 . 4 , 10 mM NaCl , 3 mM MgCl2 , 0 . 1% Igepal CA-630 ) followed by centrifugation at 3500xg for 10 min at 4°C . The transposition reaction was performed with a 50 μl reaction mixture ( 25 μl TD , 2 . 5 μl TDE , 22 . 5 μl nuclease-free H2O . Illumina , Nextera DNA library preparation Kit FC-121–1030 ) at 37°C for 30 min . Transposed DNA was purified using Qiagen MinElute kit and saved in −20°C . qPCR was used to determine appropriate PCR cycle number for PCR amplification as detailed in Buenrostro et al . 6–7 cycles of PCR amplification were used . Final cDNA libraries ( 150 bp to 700 bp ) were selected using Agencourt AMPure beads ( Beckman-Coulter A63880 ) . Two biological samples for wild type and nos-1 ( gv5 ) nos-2 ( RNAi ) were sequenced with Hiseq2500 platform . Two biological replicates for control and nos-1 ( gv5 ) nos-2 ( RNAi ) samples were independently mapped to C . elegans ce10 reference genome using bowtie2 ( v2 . 1 . 0 ) . Peaks from individual ATAC-seq sample were called using MACS2 packing with options -p 1e-3 --nomodel --shift −100 --extsize 200 . To evaluate the correlation between two biological replicates , Diffbind package was then used to perform PCA analysis and RPKM for peaks were extracted from matadata using function dba . peakset ( DBA object , bRetrieve = T , DataType = DBA_DATA_FRAME ) . For correlation plots , peaks with RPKM >1 were kept and subjected to log2 transformation and correlations for replicates were calculated using Pearson correlation ( Figure 3—figure supplement 2 ) . To identify locus with nos-1/2-dependent features ( peaks ) , mapped reads from wild-type were used as reference sample and the callpeak function in MACS2 package was used as described below: $Macs2 callpeak -t [nos_rep1 . sam] -c [con_rep1 . sam] --outdir –f SAM –g ce –n exp1_vs_reference1 –p 0 . 01 --to-large $Macs2 callpeak -t [nos_rep2 . sam] -c [con_rep1 . sam] --outdir –f SAM –g ce –n exp2_vs_reference1 –p 0 . 01 --to-large $Macs2 callpeak -t [nos_rep1 . sam] -c [con_rep2 . sam] --outdir –f SAM –g ce –n exp1_vs_reference2 –p 0 . 01 --to-large $Macs2 callpeak -t [nos_rep2 . sam] -c [con_rep2 . sam] --outdir –f SAM –g ce –n exp2_vs_reference2 –p 0 . 01 --to-large To identify nos-1/2-dependent feature ( peaks ) with confidence , we followed the principle of ENCODE Irreproducibility Discovery Rate ( IDR ) framework as described in https://sites . google . com/site/anshulkundaje/projects/idr#TOC-FLAGGING-REPLICATES-FOR-LOW-CONSISTENCY . For IDR analysis , pairwise consistency analysis was done on replicate peak files as described below: $Rscript batch-consistency-analysis . r [exp1_vs_refernece1_ peakfile] [exp2_vs_reference1_ peakfile] −1 [output_set1 . perfix] 0 F p . value $Rscript batch-consistency-analysis . r [exp1 vs reference2_ peakfile] [exp2_vs_reference2_ peakfile] −1 [output_set2 . perfix] 0 F p . value To obtain a list of overlapped peaks between replicates , IDR cutoff was set to 0 . 1 . 1414 peaks were selected based on IDR cutoff and peaks were annotated using PAVIS ( https://manticore . niehs . nih . gov/pavis2/ ) . At the end , 221 peaks with location at upstream region of genes were extracted and gene IDs were cross-referenced with RNA-seq analysis for downstream analysis . To plot heatmap for ATAC-seq analysis , bamCompare and computeMatrix in deepTools package ( http://deeptools . readthedocs . io/en/latest/ ) were used to visualize merged ATAC-seq profile of nos-1/2-dependent genes as shown in Figure 3—figure supplement 1 . ATAC-seq reads from replicates were merged and mapped to C . elegans ce10 reference genome using bowtie2 ( v2 . 1 . 0 ) . Command lines were listed as below: $bamCompare -b1 < nos-1/2 . bam> -b2 < wild type . bam> -o < Name1 . bw> --ratio ratio --normalizeUsingRPKM -ignore chrM -bs 10 p max/2 $computeMatrix reference-point --referencePoint TSS -b 2000 -a 2000 R < nos-1/2- dependent_gene . bed> -S < Name1 . bw> -o < Name2 . gz> --sortUsing max -- skipZeros -bs 10 p 2 $plotHeatmap -m < Name2 . gz> --zMin 0 --colorList --heatmapHeight 20 -- heatmapWidth 5 -out < heatmap . png> To verify our analysis pipeline for RNAseq data , quantitative RT-PCR ( qRT-PCR ) reactions using sequencing libraries as templates were performed . The cDNA libraries were diluted to 1 nM before performing qRT-PCR . Primers for qRT-PCR were listed in key resources table . Enrichment of target mRNAs between wild-type and nos-1/2 was calculated using ΔΔCt with tbb-2 expression then normalized to wild-type control . Fold changes were plotted and significance was calculated by paired t-test . Biological replicates refer to experiments performed on independently treated hermaphrodites ( in the case of RNA-seq libraries , this refers to worms exposed to independent RNAi treatments followed by cell sorting and RNA extraction ) . All in vivo technical replicates refer to observations in the same strain from separate zygotes . Datasets generated in this paper are available at GEO accession GSE100651 for ATAC-seq and GSE100652 for RNA-seq . | Every new embryo inherits a set of starting instructions from its mother . These instructions are called a ‘maternal dowry’ and help a fertilized egg through the first few stages of development . Later , the maternal dowry is removed so that the embryo’s genetic instructions can take over . In animals , some of the cells in this early embryo become specialized to produce eggs ( technically called oocytes ) or sperm . These cells are called germ cells , and they are needed for reproduction . A protein called Nanos helps germ cells become different to other cells , but it is not clear how Nanos has this effect . Lee et al . studied Nanos in the embryos of the worm Caenorhabditis elegans . In many ways , early development is the same in the worm as in many other animals . By examining worms that did not have Nanos , Lee et al . showed that germ cells without Nanos do not lose their maternal dowry . As a result , the cells still contain a molecule called LIN-15B , which makes other types of cells in the worm . Ultimately , without Nanos , the germ cells do not develop and die leaving the worm sterile . Germ cells are essential for living things to reproduce and have children . Understanding how they are created can teach scientists a lot about how embryos develop before birth . This could eventually help to boost fertility in endangered species or to treat human sterility . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2017 | Nanos promotes epigenetic reprograming of the germline by down-regulation of the THAP transcription factor LIN-15B |
Building a genotype-phenotype-fitness map of adaptation is a central goal in evolutionary biology . It is difficult even when adaptive mutations are known because it is hard to enumerate which phenotypes make these mutations adaptive . We address this problem by first quantifying how the fitness of hundreds of adaptive yeast mutants responds to subtle environmental shifts . We then model the number of phenotypes these mutations collectively influence by decomposing these patterns of fitness variation . We find that a small number of inferred phenotypes can predict fitness of the adaptive mutations near their original glucose-limited evolution condition . Importantly , inferred phenotypes that matter little to fitness at or near the evolution condition can matter strongly in distant environments . This suggests that adaptive mutations are locally modular — affecting a small number of phenotypes that matter to fitness in the environment where they evolved — yet globally pleiotropic — affecting additional phenotypes that may reduce or improve fitness in new environments .
Laboratory evolution experiments are opening an unprecedented window into the dynamics and genetic basis of adaptive change by de novo mutation ( Crozat et al . , 2010; Good et al . , 2017; Huang et al . , 2018; Lang et al . , 2013; Levy et al . , 2015; Tenaillon et al . , 2012; Venkataram et al . , 2016a ) . One of the key insights revealed by these studies is that in many systems , evolution can initially proceed rapidly via many large-effect single mutations . While the identities of these adaptive mutations are often unique to a specific replicate of the evolutionary experiment , across many replicates they tend to occur in similar functional units ( e . g . genes and pathways ) ( Crozat et al . , 2010; Fumasoni and Murray , 2020; Good et al . , 2017; Huang et al . , 2018; Lang et al . , 2013; Levy et al . , 2015; Tenaillon et al . , 2012; Venkataram et al . , 2019 , Venkataram et al . , 2016a ) . Thus , although the diversity of mutations suggests that there might be many ways to adapt , the much smaller number of apparent functional units implies , in contrast , that most adaptive mutations affect a small set of key phenotypes ( Figure 1A ) . Consider the seminal study by Tenaillon et al . , 2012 in which 115 populations were evolved at high temperature for ~2000 generations . While the authors identified over a thousand mutations that were largely unique to each population , the number of affected genes was much smaller with 12 genes being hit over 25 times each . Even greater convergence was seen at higher levels of organization such as operons . Similarly , Venkataram et al . , 2016a found that , of the hundreds of unique genetic mutations that occur during adaptation to glucose-limitation , the vast majority fall into a relatively small number of genes ( mostly IRA1 , IRA2 , GPB2 , PDE2 ) and primarily two pathways — Ras/PKA and TOR/Sch9 . Thus , despite the diversity of mutations , it is possible that all their effects can be mapped in one or few dimensions required to describe their effects on the Ras/PKA or TOR/Sch9 pathways . These are just two examples , but the pattern has been seen repeatedly ( Barghi et al . , 2019; Crozat et al . , 2010; Good et al . , 2017; Lang et al . , 2013; Lind et al . , 2015 ) . Note that this pattern is seen not only in experimental evolution but also in cancer evolution . Individual tumors are largely unique in terms of specific mutations , but these mutations affect a much smaller set of driver genes and an even smaller number of higher functional units such as signaling pathways ( Bailey et al . , 2018; Hanahan and Weinberg , 2011; Hanahan and Weinberg , 2000; Sanchez-Vega et al . , 2018; Sondka et al . , 2018 ) . The mapping of adaptive mutations to a smaller number of functional units and thus a low-dimensional space representing the small number of phenotypes that they collectively affect ( Figure 1A ) is consistent with theoretical models of adaptation . These theoretical models argue that adaptive mutations , especially those of substantial fitness benefit , cannot affect too many phenotypes at once as most such effects should be deleterious and thus inconsistent with the overall positive effect on fitness ( Fisher , 1930; Orr , 2000 ) . More recent studies likewise suggest that selection against mutations with high pleiotropy , that is mutations that affect many phenotypes , has resulted in a modular architecture of the genotype-phenotype map , in which genetic changes can influence some phenotypes without disturbing others ( Altenberg , 2005; Collet et al . , 2018; Hartwell et al . , 1999; Melo et al . , 2016; Wagner et al . , 2007; Wagner and Altenberg , 1996; Wagner and Zhang , 2011; Welch and Waxman , 2003 ) . This architecture would allow single mutations to have a large effect on a small number of important phenotypes . It would also explain the observation that even very large collections of mutations that provide a fitness benefit in a particular condition are not diverse in terms of affected genes , pathways , and phenotypes . The reason for this is that only mutations that affect the genes , pathways , and phenotypes corresponding to the module most relevant to adaptation in that condition will be observed . We term this model in which mutations only affect a small number of phenotypes ‘strict modularity’ . While theoretically appealing , the possibility that observed adaptive mutations indeed affect only a very small number of phenotypes is difficult to reconcile with the notion that organisms are tightly integrated ( Kacser and Burns , 1981; Paaby and Rockman , 2013; Rockman , 2012 ) . Further , there is experimental evidence of widespread pleiotropy , for example , from genome-wide association studies that suggest that every gene can influence every trait , at least to some extent ( Boyle et al . , 2017; Chesmore et al . , 2018; Sella and Barton , 2019; Sivakumaran et al . , 2011; Visscher and Yang , 2016 ) . It is possible that pleiotropy is common , but strongly adaptive mutations observed in experimental evolution are unusual in that they have few phenotypic effects . Another possibility is that these mutations do have pleiotropic side effects , but these matter little to fitness in the condition where these mutants evolved ( Figure 1B , left side ) . We term this model ‘fitness-relevant modularity’ because these mutations are not strictly modular with respect to all the phenotypes they affect , but they are effectively modular because only a subset of these phenotypes are relevant to fitness in the evolution condition . Here , we do not need to claim that these phenotypic effects never matter to fitness but rather that they do not matter substantially to fitness in the condition where they evolved . In fact , the key prediction of this model is that one should be able to detect latent pleiotropy and reveal the additional phenotypic effects of these mutants by demonstrating their varied fitness consequences in other conditions or environments ( Figure 1B , right side ) . Note that we cannot test this prediction by demonstrating antagonistic pleiotropy , that is that mutations that are adaptive in one environment have fitness tradeoffs in other environments ( Dillon et al . , 2016; Jerison et al . , 2020 ) . Antagonistic pleiotropy could indeed indicate that the mutations affect many phenotypes , some of which only hinder fitness in certain environments . But it could also indicate that the adaptive mutations all change the same phenotype in a way that improves fitness in some environments and hinders fitness in others . If the ‘fitness-relevant modularity’ model depicted in Figure 1B is true then it is possible that adaptive mutations are locally modular — that they affect very few phenotypes that matter to fitness in the evolution condition — and globally pleiotropic . Under this model , the large number of distinct mutations available to adaptation becomes important . Indeed while these mutations tend to influence similar genes and pathways , their phenotypic effects do not simply collapse to a low-dimensional space . Instead , this genetic diversity becomes a source of consequential phenotypic diversity , but only once these genetic variants leave the local environment in which they originated . In order to test this model and better understand the genotype-phenotype-fitness map , we face the difficult task of identifying which phenotypes are affected by the adaptive mutations and then determining how these phenotypes contribute to fitness . This is a challenging problem as the possible number of phenotypes one can measure is effectively infinite , for example the expression level of every gene or the quantity of every metabolite ( Coombes et al . , 2019; Mehlhoff et al . , 2020 ) . Further , many measurable phenotypes are related in complex ways ( Geiler-Samerotte et al . , 2020 ) . Mapping their contribution to fitness requires a complete understanding of how genetic changes lead to molecular changes and how these percolate to higher functional levels and ultimately influence fitness ( Kemble et al . , 2020 ) . This might be possible to do in some cases where the phenotype to fitness mapping is simple ( e . g . antibiotic resistance driven by a specific enzyme or tRNA or protein folding mediating specific RNA or protein function; Baeza-Centurion et al . , 2019; Cowperthwaite et al . , 2005; Diss and Lehner , 2018; Domingo et al . , 2019; Harmand et al . , 2017; Karageorgi et al . , 2019; Li and Zhang , 2018; Otwinowski et al . , 2018; Pressman et al . , 2019; Sarkisyan et al . , 2016; Starr et al . , 2018; Weinreich , 2006 ) but is exceptionally difficult for complex phenotypes . In the case of the adaptive mutations from Venkataram et al . , 2016a mentioned above , we might be able to use our knowledge of the Ras/PKA pathway to make a guess about what phenotypes they affect . We know that many of these mutations result in the loss of negative regulators of the Ras/PKA pathway ( IRA1 , IRA2 , GPB2 , PDE2 ) . Thus , we might guess that these adaptive mutations all lead to an increase in the amount of active PKA . Then we could use more traditional approaches to confirm this hypothesis , for example , by measuring the levels of PKA through functional assays . However , even if these mutations do increase PKA activity , it is not clear how this effect percolates through the system , or what other phenotypic effects we might miss by using such a directed approach to investigate the genotype-phenotype-fitness map . Moreover , to distinguish between the model in which mutations affect a small number of phenotypes ( ‘strict modularity’ as shown in Figure 1A ) and the model in which mutations affect many phenotypes , albeit with few contributing substantially to fitness in the evolution condition ( ‘fitness-relevant modularity’ as shown in Figure 1B ) , we need to understand these genotype-phenotype-fitness maps not only in the environment in which adaptive mutants evolved but also in other environments . And we need to do this for many adaptive mutants so that we can assess the extent to which different mutants affect different phenotypes . Considering the scope of this challenge , it is not surprising that despite much theoretical discussion of modularity and pleiotropy as it relates to adaptation , experimental approaches to address these questions have lagged behind . Here , we suggest a way to model the genotype-phenotype-fitness relationship that avoids the problem of measuring each phenotype and its effect on fitness explicitly . We argue that it is possible to investigate the genotype-phenotype-fitness map by comparing how the fitness effects of many mutations change across a large number of environments . The way each mutant’s fitness varies across environments must be related to its phenotype , and thus the way mutants co-vary in fitness across environments tells us whether they affect similar fitness-relevant phenotypes . We can use these profiles of fitness across a set of environments to identify the total number of fitness-relevant phenotypes that must be affected across a collection of adaptive mutants , the extent to which different mutants affect different phenotypes , and whether the contribution of each phenotype to fitness changes across environments . Importantly , the phenotypes we identify with this approach are abstract entities rather than measured cell properties . Nevertheless , these abstract phenotypes reflect the causal effects of adaptive mutations on fitness . Here , we build a genotype- ( abstract ) phenotype-fitness model for hundreds of adaptive yeast mutants that originally evolved in a glucose-limited environment . We use this model to accurately predict the fitness of these mutants across a set of 45 environments that vary in their similarity to the evolution condition . We find that the fitness behavior of adaptive mutations near the evolution condition can be described by a low-dimensional phenotypic model . In other words , these mutants affect a small number of phenotypes that matter to fitness in the glucose-limited condition in which they evolved . We find that this low-dimensional phenotypic model makes accurate predictions of mutant fitness in novel environments even when they are dissimilar to the evolution condition . Moreover , we find that some phenotypes that contribute very little to fitness in the evolution condition become surprisingly important in some novel environments . This suggests that adaptive mutations are globally pleiotropic in that they affect many phenotypes overall , but that they are locally modular in that only a small number of these phenotypes have substantial effects on fitness in the environment they evolved in . Overall , we suggest that this set of adaptive mutations contains substantial and consequential latent phenotypic diversity , meaning that despite targeting similar genes and pathways , different adaptive mutants may respond differently to future evolutionary challenges . This finding has important consequences for understanding how directional selection can generate consequential phenotypic heterogeneity both in natural populations and also in the context of diseases , such as cancer and viral or bacterial infections . In addition , our results show that our abstract , top-down approach is a promising route of analysis for investigating the phenotypic and fitness consequences of mutation .
A previous evolution experiment generated a collection of hundreds of adaptive yeast mutants , each of which typically harbors a single independent mutation that provides a benefit to growth in a glucose-limited environment ( Levy et al . , 2015 ) . Many of these mutants , which began the evolution experiment as haploids , underwent whole-genome duplication to become diploid , which improved their relative fitness ( Venkataram et al . , 2016a ) . Some of these diploids acquired additional mutations , including increased copy number of either chromosome 11 or 12 as well as point mutations , which generated additional fitness benefits . The adaptive mutants that remained haploid acquired both gain- and loss-of-function mutations in nutrient-response pathways ( Ras/PKA and TOR/Sch9 ) . Some other mutations were also observed , including a mutation in the HOG pathway gene SSK2 ( Venkataram et al . , 2016a ) . Although these mutants have been well-characterized at the level of genotype and fitness , it is unclear what phenotypes they affect . The first question we address is whether these diverse mutations collectively affect a large number of phenotypes that matter to fitness , or whether these mutants are functionally similar in that they collectively alter a small set of fitness-relevant phenotypes . Understanding the map from genotype to phenotype to fitness is extremely challenging because each genetic change can influence multiple traits , not all of which are independent or contribute to fitness in a meaningful way . We contend with this challenge by measuring how the relative fitness of each adaptive mutant changes across a large collection of similar and dissimilar environments , which we term the ‘fitness profile’ . When a group of mutants demonstrate similar responses to environmental change , we conclude that these mutants affect similar phenotypes . By clustering mutants with similar fitness profiles across a collection of environments , we can learn about which mutants influence similar phenotypes , as well as estimate the total number of fitness-relevant phenotypes represented across all mutants in all investigated environments . Because our mutant strains are barcoded , we can use previously established methods to measure their relative fitness in bulk and with high precision ( Venkataram et al . , 2016a ) . Specifically , we compete a pool of the barcoded mutants against an ancestral reference strain over the course of several serial dilution cycles . During each 48 hr cycle , the yeast are given fresh glucose-limited media which supports eight generations of exponential growth after which glucose is depleted and cells transition to non-fermentable carbon sources . After every 48 hr cycle , we transfer ~5×107 cells to fresh media to continue the growth competition . We also extract DNA from the remaining cells to PCR amplify and sequence their barcodes . We repeat this process four times , giving us an estimate of the frequency of each barcode at five time-points . By quantifying the log-linear changes in each barcode’s frequency over time and correcting for the mean-fitness change of population , we can calculate the fitness of each barcoded mutant relative to the reference strain ( Figure 2A; Materials and methods ) . Using this method , we quantify the fitness of a large number of adaptive mutants in 45 environments . We focus on a set of 292 adaptive mutants that have been sequenced , show clear adaptive effects in the glucose-limited condition in which these mutants evolved ( hereafter ‘evolution condition’; EC ) ( Figure 2B; Supplementary file 1 ) , and for which we obtained high-precision fitness measurements in all 45 environments . These environments include some experiments from previously published work ( Li et al . , 2018; Venkataram et al . , 2016a ) , as well as 32 new environments including replicates of the evolution condition , subtle shifts to the amount of glucose , changes to the shape of the culturing flask , changes to the carbon source , and addition of stressors such as drugs or high salt ( Supplementary file 2 ) . In order to determine the total number of phenotypes that are relevant to fitness in the EC , we focus on environments that are very similar to the EC but still induce small yet detectable perturbations in fitness . We do so because the phenotypes that are the most relevant to fitness may change with the environment ( Figure 1B ) . Thus , we partition the 45 environments into a set of ‘subtle’ perturbations , from which we will detect the phenotypes relevant to fitness near the EC , and ‘strong’ perturbations which we will use to study whether these mutants influence additional phenotypes that matter in other environments ( Figure 1B ) . To partition environments into subtle and strong perturbations of the EC , we rely on the nested structure of replicate experiments performed in the EC . We assayed fitness in the EC on nine different occasions which we term ‘batches’ . Each batch contained multiple replicates . We observe much less variation across replicates than across batches ( p<1e-5 from permutation test ) . Variation across batches likely reflects environmental variability that we were unable to control ( e . g . slight fluctuations in incubation temperature due to limits on the precision of the instrument , slight differences in the media reflective of the limits on the precision of our scale ) . These differences between batches are as subtle as possible in our experimental setup , as they represent the limit of our ability to minimize environmental variation . Thus , variation in fitness across the EC batches serves as a natural benchmark for the strength of other environmental perturbations . If the deviations in fitness caused by an environmental perturbation are substantially stronger than those observed across the EC batches , we call that perturbation ‘strong’ . More explicitly , to determine whether a given environmental perturbation is subtle or strong , we subtract the fitness of adaptive mutants in this environment from their average across the EC batches . We then compare this difference to the variation in fitness observed across the EC batches . Sixteen environmental perturbations provoked fitness differences that were similar to those observed across EC batches ( Z-score <2 ) . These environments , together with the nine EC batches , make up a set of subtle environmental perturbations . The remaining 20 environments , where the average deviation in fitness is substantially larger than that observed across batches ( Z-score >2 ) , were classified as strong environmental perturbations ( Figure 2C , top; Materials and methods ) . Note that when we use different subsets of the subtle environmental perturbations , our qualitative conclusions hold , indicating they are not sensitive to our particular choice of which environments to classify as subtle or strong ( Figure 4—figure supplement 1 ) . The rank order of the fitnesses of many mutations is largely preserved across the 25 environments that represent subtle perturbations ( Figure 2C , bottom ) . For example , IRA1 nonsense mutants , which are the most adaptive in the EC , generally remain the most adaptive across the subtle perturbations . Additionally , the GPB2 and PDE2 mutants have similar fitness effects across EC batches and only occasionally switch order across the subtle environmental perturbations . In contrast , the 20 environments that represent strong perturbations reveal clear genotype-by-environment interactions ( Figure 2C , bottom ) . For example , altering the transfer time from 48 to 24 hr ( the ‘1 Day’ environment in Figure 2C ) affects GPB2 mutants more strongly compared to the other mutants in the Ras/PKA pathway , including IRA1 and PDE2 . The strongest environmental perturbations reveal clear tradeoffs for some of these adaptive mutants . For example , PDE2 and IRA1 nonsense but not GPB2 mutants are particularly sensitive to osmotic stress as indicated by the NaCl and KCl environments . Additionally , IRA1 nonsense mutants become strongly deleterious in the long transfer conditions that experience stationary phase ( 5- , 6- , 7-Day environments ) ( Li et al . , 2018 ) . In contrast to complex behavior exhibited by the adaptive haploids , the diploids appear to be relatively robust to strong tradeoffs , appearing similarly adaptive across all perturbations , subtle and strong . The observation that different mutants have different and fairly complex fitness profiles suggests that they have different phenotypic effects . Even PDE2 and GPB2 , which have similar fitnesses in the EC and are negative regulators of the same signalling pathway , have different fitness profiles . Do these diverse phenotypic effects contribute to fitness in the EC ? To examine how many phenotypes matter to fitness in the EC , we test whether it is possible to create low-dimensional models that capture the complexity of the fitness profiles of all adaptive mutants across all subtle perturbations . We utilize these complex fitness profiles to estimate the number of phenotypes that contribute to fitness in the EC . Given that many of these mutants affect genes in the same nutrient response pathway , the number of unique phenotypes they affect may be small . Alternatively , given the observation that these mutants have different interactions with environments that represent strong perturbations ( Figure 2C ) , this number may be large . We use singular value decomposition ( SVD ) to ask how much of the complexity in these fitness profiles can be captured by a low-dimensional phenotypic model ( Figure 3A ) . SVD is a dimensionality reduction approach which here decomposes fitness profiles into two abstract multi-dimensional spaces described below . The first space , P , represents the phenotypic effects of mutants , where each phenotype is represented as a dimension ( there are k phenotypic dimensions depicted in Figure 3A ) . Each mutant is represented by coordinates specifying a location in the phenotype space P ( e . g . mutant one having coordinates ( p11 , p12 , p13 , . . . , p1k ) ) . The ancestral reference lineage , which , by definition , has relative fitness zero in every environment , is placed at the origin ( e . g . ( 0 , 0 , 0 , … 0 ) ) in this phenotypic space . In this sense , we can think of a mutation's effect on any phenotype as a measure of the distance from the location of the mutant in that phenotypic dimension to the origin . The second space , E , represents the contribution of each of the phenotypes in P to fitness , and thus has the same number of dimensions as P . If a phenotype does not contribute substantially to fitness in any environment , it is not represented as a dimension in either space . Therefore , our model captures only fitness-relevant phenotypes . In space E , each environment is represented by coordinates specifying a location ( e . g . environment one having coordinates ( e11 , e21 , e31 , … , ek1 ) ) . These coordinates in E reflect the contribution ( weight ) of each of the k phenotypic dimensions on fitness in that environment . For example , an environment where only a single phenotype matters to fitness would be placed at the origin for all the axes , except for the axis corresponding to the single phenotypic dimension that matters . Environments for which the same phenotypes contribute to fitness will be placed closer together in the space E . In this model , each phenotype contributes to fitness independently , by definition , such that the fitness of mutant i in environment j is determined by each phenotypic effect of mutant i , scaled by the contribution of that phenotype to fitness in environment j . A linear combination of these weighted phenotypic effects determines the fitness of mutant i in environment j:fij=pi1e1j+pi2e2j+pi3e3j+ . . . +pikekj In this model , mutants with similar fitness profiles , for example mutants 1 and 2 in Figure 3A , will be inferred as having similar phenotypic effects , and thus be located near each other in the phenotypic space P . Mutants with dissimilar fitness profiles , for example mutants 3 and 4 in Figure 3A , can be inferred to have at least some differing phenotypic effects , which might be mediated by a different effect on a single phenotypic component or different effects on many . Mutants with dissimilar fitness profiles are informative about the number of dimensions needed in this abstract model of phenotypic space . This genotype-phenotype-fitness model that we generate using SVD harkens to Fisher’s geometric model ( FGM ) , which defines an abstract space of orthogonal phenotypes relevant to fitness ( Fisher , 1930 ) . Others have utilized FGM to answer questions about the number of phenotypes affected by mutations , although most previous work focuses on deleterious mutations and how their impacts vary across genetic backgrounds rather than environments ( Blanquart et al . , 2014; Blanquart and Bataillon , 2016; Lourenço et al . , 2011; Martin and Lenormand , 2006; Poon and Otto , 2000; Tenaillon et al . , 2007; Weinreich and Knies , 2013 ) . A key difference between FGM and our model is that our model does not make assumptions about the distribution of phenotypic effects or whether the relationship between mutations in phenotype space is additive . Here , we utilize SVD to count the number of phenotypes that contribute to fitness in the original glucose-limited environment in which these adaptive mutants evolved . We used SVD to build an abstract model that captures fitness profiles of all 292 adaptive mutants across the 25 subtle perturbations . This model suggests that the majority of the variation in fitness for the 292 adaptive mutants across the 25 subtle perturbations can be explained by eight phenotypic dimensions . The first phenotypic component is very large and explains 95% of variation in fitness across all mutants and all subtle perturbations ( Figure 3B ) . This component captures the variation in fitness explainable in the absence of genotype-by-environment interactions , where each mutation has a single effect that is scaled by the environment . As such , this first component effectively represents each mutant’s average fitness in the EC ( Figure 3—figure supplement 2A ) and the average impact of each subtle perturbation on mutant fitness ( Figure 3—figure supplement 2B ) . It is not surprising that this component explains much of this variation , as the fitness of mutants in the EC should be predictive of fitness in similar environments . The next seven components capture additional variation not detectable from the simple one-component model and thus represent genotype-by-environment interactions . Of these , the first four capture 87% of the variation not captured by component one ( 67 . 8% , 8 . 3% , 5 . 6% , and 5 . 3% , respectively ) . The remaining three interaction components each capture less than 2% of the variation not captured by component one ( Figure 3B ) . We cannot distinguish any additional components , beyond these eight , from noise . This is because we see components that explain a similar amount of variation when we apply SVD to datasets composed exclusively of values generated by our noise model ( Figure 3B; see Materials and methods and Figure 3—figure supplement 1 for additional details ) . We confirm that these eight phenotypic components capture meaningful biological variation in fitness by using bi-cross-validation . Specifically , we designate a balanced set of 60 of the 292 mutants as a training set , chosen such that the recurrent mutation types — diploids , high-fitness diploids , Ras/PKA mutants — are roughly equally represented ( see Materials and methods ) . The remaining 232 mutants comprise the test set . This set contains all mutation types represented by only a single mutant , including all TOR/Sch9 ( TOR1 , SCH9 , KOG1 ) and HOG ( SSK2 ) pathway representatives , as well as the rest of the recurrent mutants that were not picked for the training set . We include these diverse mutants in the test set so that we can measure the ability of our genotype-phenotype-fitness model to predict the fitness of mutants in genes and pathways that are absent from the training set . We iteratively construct phenotype spaces using the 60 training mutants while holding out one subtle perturbation at a time and creating the space with the data from the remaining 24 subtle perturbations . We then predict the fitness of the 232 held-out testing mutants in the held-out condition . We do so using all eight components , and again with only 7 , 6 , and so on . Then , we ask whether the eight component model does a better job at predicting mutant fitness than the other , lower dimensional models . If a component reflects measurement noise rather than biological signal , then the inclusion of this component would lead to overfitting and should harm the model’s ability to predict fitness in the held-out data . Instead we find that , on average across the 25 iterations , prediction power improves from the inclusion of each of the eight components . This confirms that even the smallest of these components captures biologically meaningful variation in fitness across the 25 subtle perturbations of the EC . However , the gain in predictive power decreases for each component . The model with only the first component explains on average 85% of weighted variance for the test mutants in the left-out conditions . A model with only the top five components explains 95 . 1% , and all eight components explain 96 . 2% of variation . This suggests that the last few components have very small contributions to fitness in the environments near the EC . We next ask whether the eight-dimensional phenotypic model clusters adaptive mutants found in similar genes or pathways ( e . g . Ras/PKA or TOR/Sch9 ) , or that represent similar mutation types ( haploid v . diploid ) . Alternatively , our model may classify mutations into functional units ( i . e . mutations that have similar phenotypic effects ) in a way that does not conform to gene or pathway identity . We use Uniform Manifold Approximation and Projection ( UMAP ) to visualize the distance between all the mutants in this phenotypic space . As the first phenotypic dimension captures the average fitness of each mutant in the EC , and since we already know that mutations to the same gene have similar fitness in the EC ( Figure 2B ) , we exclude the first phenotypic dimension from this analysis , although the inclusion of the first component does not change the identity of the clusters ( Figure 3—figure supplement 4A ) . By focusing on the other seven components , we are asking whether genotype-by-environment interactions also cluster the mutants by gene , mutation type , and pathway . These seven genotype-by-environment interactions indeed tend to cluster the adaptive mutants by type and by gene ( Figure 3C ) . Specifically , the diploids , IRA1 nonsense , GPB2 , and PDE2 mutants each form distinct clusters ( p=0 . 0001 , p=0 . 006 , p=0 . 0001 , and p=0 . 0001 , respectively ) . To generate p-values , we calculated the median pairwise distance , finding that multiple mutations in the same cluster are indeed more closely clustered than randomly chosen groups of mutants . Interestingly , the three smallest components , which capture very little variation in fitness across the environments that reflect subtle perturbations of the EC , also cluster some mutants by gene ( Figure 3—figure supplement 4B ) . Specifically , PDE2 , GPB2 , and IRA1 nonsense mutants are each closer to mutants of their own type than to other adaptive haploids ( p=0 . 0001 , p=0 . 0001 , and p=0 . 03 , respectively ) . Note that the space defined by the three smallest components does not cluster IRA1 nonsense mutants away from diploids ( p=0 . 718 ) . This suggests that some mutants , for example IRA1 nonsense and diploids , have smaller effects on these three phenotypic components . Overall , our abstract phenotypic model , which reflects the way that each mutant’s fitness changes across environments , reveals that mutations to the same gene tend to interact similarly with the environment . Our approach also detects cases where mutations to the same gene or pathway do not cluster together . This suggests that our model captures phenotypic effects that would be obscured by assuming mutations to the same gene affect the same traits . For example , genotype-by-environment interactions do not cluster IRA1 missense mutations ( p=0 . 317 ) ( Figure 3C; light blue points ) , despite clustering the IRA1 nonsense mutations . Perhaps , IRA1 missense mutations have more diverse impacts on phenotype than do IRA1 nonsense mutations because the latter all likely result in a loss of the IRA1 protein , albeit not necessarily to the same extent . Our model also does not cluster the eight mutations in IRA2 ( p=0 . 086 ) ( Figure 3C; dark gray points ) . At the pathway level , our model does not cluster the three mutations to the TOR/Sch9 pathway away from the rest of the mutants , which are mainly in the Ras/PKA pathway ( p=0 . 155 ) ( Figure 3C; purple points ) . Our model also does not cluster all diploids that possess additional mutations , including those with increased copy number of chromosome 11 or chromosome 12 and those with mutations in IRA1 or IRA2 ( p=0 . 863 ) ( Figure 3C; dark red points ) . Interestingly , our model does find a distinct cluster of six diploids that have higher than average diploid fitness in the EC ( p=0 . 0001 ) despite whole genome sequencing having revealed no mutations in their coding sequences ( Figure 3C ) . This likely indicates that these diploids harbor difficult-to-sequence additional adaptive mutations that all have similar phenotypic consequences . In sum , these observations suggest that our genotype-phenotype-fitness model reveals new insights about which mutations affect the same functional units , specifically that these units do not always correspond to genes and pathways . Overall , these results suggests that our approach , like others that compare genotype-by-environment interactions ( Li et al . , 2018 ) , is a useful and unbiased way to identify mutations that share functional effects . Now that we have identified the phenotypic components that contribute to fitness in environments that represent subtle perturbations of the EC , we can test the ability of these phenotypic components to predict fitness in more distant environments . Specifically , we can measure how the contribution of each of these components to fitness changes in new environments . We can also determine whether the phenotypic components that contribute very little to explaining fitness variation near the EC might at times have large explanatory power in distant environments ( as depicted in the ‘fitness-relevant modularity’ model shown in Figure 1B ) . To test this we performed bi-cross-validation , using the eight component model constructed from fitness variation of 60 training mutants across 25 subtly different environments to predict the fitness of 232 test mutants in the environments that represent strong perturbations of the EC . To evaluate the predictive power of the model , we compare our model’s fitness predictions in each environment to predictions made using the average fitness in that environment . Thus , negative prediction power indicates cases where the model predicts fitness worse than predictions using this average ( Figure 4A ) . The eight-dimensional phenotypic model , which was generated exclusively with the data from subtle environmental perturbations , has substantial predictive power in distant environments ( Figure 4 ) . Predictions explain 29–95% of the variation in fitness of the 232 test mutants across strong environmental perturbations . For instance , in an environment where glucose concentration was increased from 1 . 5% to 1 . 8% and the flask was changed to one that increases the oxygenation of the media ( the ‘Baffle , 1 . 8% Glucose’ environment ) , we predict 95% of weighted variance with the full eight-component phenotypic model , in contrast to 51% with a one-component model ( Figure 4B ) . This ability to predict fitness is retained even when the first component ( effectively the fitness in EC ) is a poor predictor of mutant fitness . For example , in the environment where salt ( 0 . 5 M NaCl ) was added to the media , the one-component model predicts fitness worse than predictions based on the average fitness for this environment , resulting in negative variance explained ( Figure 4A and B ) . This is due to the fact that mutant fitness in this environment reflects extensive genotype-by-environment interactions , such that the fitness of mutants in this environment is uncorrelated with EC fitness . However , our predictions of mutant fitness in the 0 . 5 M NaCl environment improve when made using the eight-component phenotypic model , which predicts 72% of weighted variance . Astoundingly , the eight-component model captures strong tradeoffs between mutants with high fitness in the EC and very low fitness in this high-salt environment , specifically for IRA1 nonsense and , to a lesser extent , PDE2 mutants ( Figure 4B ) . This was surprising because there appears to be very little variation in fitness of these mutants across the subtle compared to the strong perturbations ( Figure 2C ) . This ability to predict fitness is also observed for mutations in genes and pathways that are not represented in the 60 that comprise the training set ( e . g . those with mutations in TOR/Sch9 and HOG pathway genes ) . For example , the eight-component model explains 93% of variation in the ‘Baffle , 1 . 8% Glucose’ environment and 71% of variation in the 0 . 5M NaCl environment for these mutations , compared to 76% and 31% variance explained for the one-component model , respectively . This indicates that our model is able to capture shared phenotypic effects that extend beyond gene identity . Altogether , our ability to accurately predict the fitness of new mutants in new environments suggests that the phenotypes our model identifies reflect causal effects on fitness . Most strikingly , phenotypic models that include the three smallest phenotypic components , which together contribute only 1 . 1% to variance explained across the subtle environmental perturbations ( Figure 4A ) , often explain a substantial amount of variance in the distant environments ( Figure 4A; lower panel ) . For example , the three minor components contribute 17% of the overall weighted variance explained in the 1 Day condition ( R~2 = 0 . 6–5-component model , R~2 = 0 . 73–8-component model; ( 0 . 73–0 . 6 ) /0 . 73 = 0 . 17 ) and 45% in the 6-Day environment , ( R~2 = 0 . 25–5-component model , R~2 = 0 . 46–8-component model ) ( Figure 4A and B ) . In contrast , for other strong environments ( e . g . Baffle — 1 . 8% Glucose , 8 . 5 µM GdA ( B9 ) and Baffle — 2 . 5% Glucose ) , the three smallest components do not add much explanatory power ( Figure 4A ) . These observations demonstrate that phenotypic components that make very small contributions to fitness in the EC can contribute substantially to fitness in other environments . Overall , these observations suggest an answer to questions about how adaptation is possible when mutations have collateral effects on multiple phenotypes: not all of those phenotypes contribute substantially to fitness in the EC ( Figure 1B ) . The strength of our predictions depends on how many subtle environments we used to generate our phenotype model . When we use too few , we robustly detect the largest phenotypic components , but lose power to detect minor components , which can lead to less accurate predictions of fitness in strong environmental perturbations . We show this by randomly subsampling our 25 subtle environments and repeating all of our downstream analyses ( Figure 4—figure supplement 1 ) . We see a similar pattern when we reduce the number of mutation types used in the training set . Randomly excluding many mutation types from the training set decreases our ability to predict fitness , though the exclusion of any one mutation type from the training set has limited impact on our overall predictive accuracy ( Figure 4—figure supplement 2 ) . Next , we explore the extent to which the contribution of a phenotypic component to fitness is isolated to a specific environment and/or a specific type of mutation ( Figure 5 ) . We find that many phenotypic components matter more to fitness in some environments than others . For instance , component two adds on average 36% of the weighted variance in fitness across strong perturbations , despite adding only 7% on average across the subtle environmental perturbations . This contribution is , however , variable , with the second component adding over 90% of variance explained for the two environments with Benomyl and Baffled flasks ( the ‘Baffle , 0 . 4 μg/mL Benomyl’ and ‘Baffle , 2 μg/mL Benomyl’ environments ) and only 0 . 3% for the environment in which the transfer time was lengthened from 2 to 3 days ( Figure 5A ) . This environment-dependence is also true for the smallest two components . Specifically , predictions of mutant fitness in the 0 . 5 M NaCl environment are improved from the inclusion of component 7 , adding 7 . 5% to weighted variance explained ( Figure 5A ) . Predictions of mutant fitness in the 6-Day transfer environment show improvement from the inclusion of the 8th component , which adds over 15% to weighted variance explained ( Figure 5A ) . However , the predictions of fitness in the 6 Day environment are not improved from the inclusion of the 7th component and the predictions in 0 . 5 M NaCl are not improved markedly by the inclusion of the 8th component ( Figure 5A ) . This suggests that the phenotypic effects represented by these small components contribute substantially in some environments and not others . We further asked whether these effects are not only environment-specific but also mutant-specific . To do so , we focused on environments for which the two smallest components contribute substantially to fitness ( e . g . 0 . 5 M NaCl ) . We looked at the extent to which each of these components improves power to predict the fitness of each of the 232 held-out mutants . We found these components improve the fitness predictions for some classes of mutants far more than for others . For example , fitness predictions for mutations in GPB2 , diploids with chromosome 11 amplifications , and high-fitness diploids with no known mutations each improved by over four standard deviations of measurement error in the 0 . 5 M NaCl environment due to the inclusion of the 7th component ( Figure 5B ) . This phenotypic component also has importance in the 1 Day transfer environment , albeit to a lesser degree , resulting in improvements of roughly one standard deviation for each of these mutation types . This suggests that these mutants have some phenotypic effect that contributes only slightly to fitness in many environments , including those that represent subtle perturbations of the EC , but that are particularly important in the 0 . 5 M NaCl and 1-Day transfer environments . Similarly , we find that the 8th component also improves predictive power for specific types of mutants in specific environments . In this case , diploids with chromosome 11 amplifications and PDE2 mutants have particularly strong improvements in the 6-Day transfer environment ( 11 and 5 standard deviations , respectively ) and thus likely have a shared phenotypic effect that is captured by component 8 ( Figure 5B ) . In sum , not all mutations affect all eight phenotypic components to the same degree and not all phenotypic components contribute substantially to fitness in all environments . This idiosyncrasy suggests that directional selection has the potential to generate rather than reduce phenotypic diversity in cases where multiple adaptive mutants persist within a population or across populations . Although directional selection ‘chooses’ mutations that affect a small number of similar phenotypes relevant to fitness in the EC , these mutations may have latent effects on a larger number of diverse phenotypes . When the environment changes , these latent phenotypic effects are revealed , exposing the phenotypic diversity generated by the adaptive process .
Here , we succeeded in building a low-dimensional statistical model that captures the relationship from genotype to phenotype to fitness for hundreds of adaptive mutants . Mapping the complete phenotypic and fitness impacts of genetic change is a key goal of biology . Such a map is important in order to make meaningful predictions from genetic data ( e . g . personalized medicine ) and to investigate the structure of biological systems ( e . g . their degree of modularity and pleiotropy ) ( Collet et al . , 2018; Eguchi et al . , 2019; Exposito-Alonso et al . , 2019; Zan and Carlborg , 2020 ) . Our model allows us to do both of these things . We made accurate predictions about the fitness of unstudied mutants across multiple environments , and we gained novel insights about the degree to which adaptive mutations are modular versus pleiotropic . Specifically , we learned that adaptation is modular in the sense that hundreds of diverse adaptive mutants collectively influence a small number of phenotypes that matter to fitness in the evolution condition . We also learned that different mutants have distinct pleiotropic side effects that matter to fitness in other conditions . Building genotype-phenotype-fitness maps of adaptation has long been an elusive goal due to both conceptual and technical difficulties . Indeed , the very first part of this task , namely the identification of causal adaptive mutations , presents a substantial technical challenge ( Barrett et al . , 2019; Barrett et al . , 2008; Exposito-Alonso et al . , 2019 ) . Fortunately , in some systems , such as in microbial experimental evolution and studies of cancer and resistance in microbes and viruses , genomic methodologies combined with availability of repeated evolutionary trials allow us to detect specific genetic changes responsible for adaptation . In the context of microbial evolution experiments , lineage tracing and genomics have opened up the possibility of not only detecting hundreds of specific adaptive events but also measuring their fitness precisely and in bulk ( Good et al . , 2017; Levy et al . , 2015; Li et al . , 2019; Li et al . , 2018; Nguyen Ba et al . , 2019; Venkataram et al . , 2016a ) . Thus , in these cases , we are coming close to solving the technical challenge of building the genotype to fitness map of adaptation . However , adding phenotype into this map remains a huge challenge even despite substantial progress in mapping genotype to phenotype ( Burga et al . , 2019; Camp et al . , 2019; Exposito-Alonso et al . , 2018; Geiler-Samerotte et al . , 2016; Jakobson and Jarosz , 2019; Lee et al . , 2019; Paaby et al . , 2015; Yengo et al . , 2018; Ziv et al . , 2017 ) . In principle , we now have advanced tools to measure a large number of phenotypic impacts of a genetic change , for instance through high-throughput microscopy , proteomics , or RNAseq ( Manzoni et al . , 2018; Ritchie et al . , 2015; Zhang and Kuster , 2019 ) . The conceptual problem is how to define phenotypes given the interconnectedness of biological systems ( Geiler-Samerotte et al . , 2020; Paaby and Rockman , 2013 ) . If a mutation leads to complex changes in cell size and shape , should each change be considered a distinct phenotype ? Or if a single mutation changes the expression of hundreds or thousands of genes , should we consider each change as a separate phenotype ? Intuitively , it seems that we should seek higher order , more meaningful descriptions . For example , perhaps these expression changes are coordinated and reflect the upregulation of a stress-response pathway . Unfortunately , defining the functional units in which a gene product participates remains difficult , especially because these units re-wire across genetic backgrounds , environments , and species ( Geiler-Samerotte et al . , 2020; Pavličev et al . , 2017; Sun et al . , 2020; Zan and Carlborg , 2020 ) . If mutations influence more than one phenotype , then the mapping from phenotype-to-fitness also becomes challenging . To investigate this map , we would need to find an artificial way to perturb one phenotype without perturbing others such that we could isolate and measure effects on fitness . Mapping phenotype to fitness is further complicated by the environmental dependence of these relationships ( Fragata et al . , 2019; Price et al . , 2018 ) . For example , a mutation that affects a cell’s ability to store carbohydrates for future use might matter far more in an environment where glucose is re-supplied every 6 days instead of every 48 hr . In our study , we turned the challenge of environment-dependence into the solution to the seemingly intractable problem of interrogating the phenotype layer of the genotype-phenotype-fitness map . We rely on the observation that the relative fitness of different mutations changes across environments . We assume that differences in how mutant fitness varies across environments must stem from differences in the phenotypes each mutation affects . Rather than a priori defining the phenotypes that we think may matter , we use the similarities and dissimilarities in the way fitness of multiple mutants vary across environments to define phenotypes abstractly via their causal effects on fitness . This allows us to dispense with measuring the phenotypes themselves and instead focus on measuring fitness with high precision and throughput , since tools for doing so already exist ( Venkataram et al . , 2016a ) . This approach has the disadvantage of not identifying phenotypes in a traditional , more transparent way . Still , it represents a major step forward in building genotype-phenotype-fitness maps because it makes accurate predictions and provides novel insights about the phenotypic structure of the adaptive response . We successfully implemented this approach using a large collection of adaptive mutants evolved in a glucose-limited condition . The first key result is that the map from adaptive mutant to phenotype to fitness is modular , such that it is possible to create a genotype to phenotype to fitness model that is low dimensional . Indeed , our model detects a small number ( 8 ) of fitness-relevant phenotypes , the first two of which explain almost all of the variation in fitness ( 98 . 3% ) across 60 adaptive mutants in 25 environments representing subtle perturbations of the glucose-limited evolution condition . This suggests that the hundreds of adaptive mutations we study — including mutations in multiple genes in the Ras/PKA and TOR/Sch9 pathways , genome duplication ( diploidy ) , and various structural mutations — influence a small number of phenotypes that matter to fitness in the evolution condition . This observation is consistent with theoretical considerations suggesting that mutations that affect a large number of fitness-relevant phenotypes are not likely to be adaptive ( Orr , 2000; Wagner and Altenberg , 1996 ) . It also explains findings from other high-replicate laboratory evolution experiments and studies of cancer that show hundreds of unique adaptive mutations tend to hit the same genes and pathways repeatedly ( Hanahan and Weinberg , 2011; Hanahan and Weinberg , 2000; Sanchez-Vega et al . , 2018; Tenaillon et al . , 2012; Venkataram et al . , 2016a ) . Our work confirms the intuition that these mutations all affect similar higher-order phenotypes ( e . g . the level of activity of a signalling pathway ) . This suggests that , despite the genetic diversity among adaptive mutants , adaptation may be predictable and repeatable at the phenotypic level . Note that although we detect only eight fitness-relevant phenotypes , we expect the true number to be much larger as the detectable number is limited by the precision of measurement ( see Materials and methods and Figure 2—figure supplement 1 ) and the number of environments used to construct the phenotypic model ( Figure 4—figure supplement 1 ) . We expect this partly because we know that if we had worse precision in this experiment we would have detected fewer than eight phenotypic components ( Figure 3 ) . Still , these additional undetected components cannot be very consequential in terms of their contribution to fitness in the evolution condition , given how well the first eight components capture variation in environments that are similar to the evolution condition . Surprisingly , the model built only using subtle environmental perturbations was also predictive of fitness in environments that perturbed fitness strongly . In some of these environments , such as the environment where 0 . 5 M NaCl was added to the media or the time of transfer was extended from 2 to 6 days , many of the mutants are no longer adaptive and some of them become strongly deleterious . Here , the fitness of the mutants in the evolution condition is a very poor predictor of fitness . Despite this , the eight-dimensional phenotypic model built using subtle perturbations of the evolution condition explains from 29% to 95% of the variance in environments that represent strong perturbations . What was particularly interesting is that the explanatory power of different phenotypic components was very different for the strong compared to subtle perturbations . For instance , the second component , which explained 7% of weighted variation on average in the subtle perturbations , explained 36% on average in the environments that represent strong perturbations . The pattern was particularly striking for the smallest three components which at times explained 15% in the strong environmental perturbations while again explaining at most 1% in the subtle environmental perturbations . This discovery emphasizes that , although the smaller phenotypic components contribute very little to fitness in the evolution condition , they can at times have a much larger contribution in other environments , as predicted by the fitness-relevant modularity model ( Figure 1B ) . This makes intuitive sense . For instance , we know that some of the strongest adaptive mutations in our experiment , the nonsense mutations in IRA1 , appear to stop cells from shifting their metabolism toward carbohydrate storage when glucose levels become low ( Li et al . , 2018 ) . This gives these cells a head start once glucose again becomes abundant and does not appear to come at a substantial cost , at least not until these cells are exposed to stressful environments ( e . g . high salt or long stationary phase ) ( Li et al . , 2018 ) . This example , and more generally the observation that phenotypic effects that are unimportant in the evolution condition can become more important in other environments , supports the idea that adaptation can happen through large effect mutations because many of the pleiotropic effects will be inconsequential in the local environment ( Figure 1B ) . We can thus argue that our low-dimensional model representing the genotype-phenotype-fitness map near the evolution condition hides consequential phenotypic complexity across the collection of adaptive mutants . This complexity is hidden from natural selection in the evolution condition but becomes important once the mutants leave the local environment and are assessed globally for fitness effects . Thus , with respect to their effects on fitness-relevant phenotypes , adaptive mutants may be locally modular , but globally pleiotropic . The notion of latent phenotypic complexity is exciting as it generates a mechanism by which directional selection generates rather than removes phenotypic diversity . Although directional selection may promote multiple mutants that affect similar fitness-relevant phenotypes in the evolution condition , each mutant could have disparate latent phenotypic effects that do not contribute immediately to fitness . When the environment changes , these disparate phenotypic effects may be revealed , imposing fitness costs of different magnitudes or allowing for diverse solutions to a variety of possible new environments ( Bono et al . , 2017; Chavhan et al . , 2020; Jerison et al . , 2020; Li et al . , 2019 ) . This latent phenotypic complexity also has the potential to alter the future adaptive paths that a population takes even in a constant environment . Indeed , these phenotypically diverse mutants are likely to affect the subsequent direction of adaptation given that subsequent mutations can shift the context in which phenotypes are important in the same way as do environmental perturbations ( Blount et al . , 2018; Blount et al . , 2008; Dillon et al . , 2016 ) . Latent phenotypic complexity among adaptive mutations is thus similar to cryptic genetic variation in that it can influence a population’s ability to adapt to new conditions ( Paaby and Rockman , 2013 ) , but dissimilar in that it evolves under directional rather than stabilizing selection . The end result is that directional selection can generate diversity both within a population in which multiple adaptive mutants are segregating and across populations that are adapting to the same stressors . The phenomenon of latent phenotypic complexity being driven by adaptation is dependent on there being multiple mutational solutions to an environmental challenge , such that different adaptive mutations might have different latent phenotypic effects . Latent phenotypic diversity might be less apparent in cases where adaptation proceeds through mutations in a single gene and certainly would not exist if adaptation relies on one unique mutation . Thus , in some ways , latent phenotypic diversity reflects redundancies in the mechanisms that allow cells to adapt to a challenge . One such putative redundancy in the case investigated in this paper is that the Ras/PKA pathway can be constitutively activated by loss-of-function mutations to a number of negative regulators including IRA1 , PDE2 , and GPB2 . Mutations in these genes might be redundant in the sense that they influence the same fitness-relevant phenotype in the evolution condition , which in this case is likely flux through the Ras/PKA pathway . This type of redundancy is commonly observed in laboratory evolutions ( Barghi et al . , 2020 ) and is particularly apparent in studies that analyze individuals with several adaptive mutations . Such studies find that multiple mutations in the same functional unit occur less than expected by chance presumably because those mutations would have redundant effects on fitness ( Tenaillon et al . , 2012 ) . Similarly , studies also find that second-step adaptive mutations tend to be in different pathways or functional modules than the first adaptive step ( Aggeli et al . , 2020; Fumasoni and Murray , 2020 ) . The novel observation from our paper is that mutations with redundant effects on fitness in the evolution condition are not necessarily identical because they may influence different latent phenotypes . This observation adds to a long list of examples demonstrating that redundancies , such as gene duplications and dominance , allow evolution the flexibility to generate diversity . One disadvantage of our approach is that the phenotypic components that we infer from our fitness measurements are abstract . They represent causal effects on fitness , rather than measurable features of cells . For this reason , perhaps we should not refer to them as phenotypes but rather ‘fitnotypes’ ( a mash of the terms ‘fitness’ and ‘phenotype’ ) that act much like the causal traits in Fisher’s geometric model ( Blanquart et al . , 2014; Blanquart and Bataillon , 2016; Fisher , 1930; Harmand et al . , 2017; Lourenço et al . , 2011; Martin and Lenormand , 2006; Poon and Otto , 2000; Tenaillon , 2014; Tenaillon et al . , 2007; Weinreich and Knies , 2013 ) or a selectional pleiotropy model ( Paaby and Rockman , 2013 ) . Despite this limitation , these fitnotypes have proven useful in allowing us to understand the consequences of adaptive mutation . In addition to insights discussed above , we also learned that adaptive mutants in the same gene do not always affect the same fitnotypes . For example , we found that IRA1 missense mutations have varied and distinct effects from IRA1 nonsense mutations . Another way that identifying fitnotypes may ultimately prove useful is in identifying the phenotypic effects of mutation . The fitnotypes can serve as a scaffold onto which a large number of phenotypic measurements can be mapped . Even though fitnotypes are independent with respect to their contribution of fitness and contribute to fitness linearly , the mapping of commonly measured features of cells ( e . g . growth rate , the expression levels of growth supporting proteins like ribosomes ) onto fitnotypes may not be entirely straightforward . Nonetheless , methods such as Sparse Canonical Correlation Analysis ( Suo et al . , 2017 ) hold promise in such a mapping and might help us relate traditional phenotypes to fitnotypes . An important question for future research is whether our observation of local modularity and global pleiotropy are also apparent in other cases of adaptation . The method we described is generic and can be applied to any system as long as the fitness of a substantial set of mutants can be profiled across multiple environments or genetic backgrounds . This is becoming possible to do in many systems ( Flynn et al . , 2020; Jerison et al . , 2020; Li et al . , 2019; Martin et al . , 2015; Pan et al . , 2018; Rogers et al . , 2018 ) and presents an opportunity to understand how the number of fitness-relevant phenotypes that a collection of mutations affects depends on the environment in which those mutations evolved and the environment in which their fitness effects are assessed . The notion that diverse genetic changes can have redundant effects in one environment but distinct and consequential effects in other environments is important to our understanding of adaptation in other settings , including in the context of antibiotic resistance and cancer . For example , tumors representing the same type of cancer ( e . g . lung adenocarcinoma ) tend to be genetically diverse even if considering only driver mutations ( Cancer Genome Atlas Research Network , 2014 ) . However , the driver mutations often fall into a smaller number of key driver genes and even fewer pathways ( Bailey et al . , 2018; Hanahan and Weinberg , 2011; Hanahan and Weinberg , 2000; Sanchez-Vega et al . , 2018; Sondka et al . , 2018 ) . While this apparent redundancy might suggest that the tumors are functionally similar , the notion of latent diversity we propose here suggests that the specific mutational paths taken by different tumors might matter once the environment changes , for example when the tumors are treated by a cancer therapy . Substantial heterogeneity of tumor response to therapy is consistent with this notion ( Li et al . , 2020 ) . Despite the accumulation of large amounts of genomic and phenomic data , integrating this information to identify the phenotypic consequences of mutation that are ultimately responsible for fitness remains incredibly challenging . Our approach allows us to create an abstract representation of the causal effects of genetic mutation and their changing contribution to fitness across environments . This top-down view of the genotype-phenotype-fitness map simplifies the complex and multifaceted phenotypic consequences of mutation by focusing on those that contribute to fitness . Integrating this new perspective with the influx of precise and high-throughput data might allow us to answer age-old questions about the structure of biological systems and adaptation .
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact , Dmitri Petrov ( dpetrov@stanford . edu ) . The yeast strains used in this study can be grown and maintained using standard methods ( e . g . YPD media in test tubes , glycerol stocks for long term storage at −80°C ) , but should be propagated in the appropriate selection environment ( a glucose-limited minimal media - M3 medium for the evolution condition ) for comparable fitness and phenotypic measurements . All the strains we study are of genetic background MATɑ , ura3Δ0 , ybr209w::Gal-Cre-KanMX-1/2URA3-loxP-Barcode-1/2URA3-HygMX-lox66/71 . Experiments were performed with barcoded mutants isolated from a previous evolution experiment ( Levy et al . , 2015 ) . To measure their fitness , these mutants were competed against a constructed reference strain with a restriction site in the barcode region ( Venkataram et al . , 2016a ) . The majority of the fitness measurement experiments were conducted with a collection of 500 adaptive barcoded mutants where each strain starts at equal frequency ( Li et al . , 2018; Venkataram et al . , 2016a ) . We focus on a subset of 292 strains for which we obtained fitness measurements in all 45 environments and for which mutations conferring fitness advantages have been previously identified , either by whole genome sequencing or using a drug to test ploidy ( Li et al . , 2018; Venkataram et al . , 2016a; Supplementary file 1 ) . Note that because we utilize some data from previous experiments ( Li et al . , 2018; Venkataram et al . , 2016a ) , some of the experiments contained additional barcoded mutants not analyzed here , namely a pool consisting of a total of 4800 strains , including the 292 focused on in this study . These differences in the number of strains included in the experiment are partially accounted for in our inference of mean fitness , and any remaining effects can be thought of as another parameter that varies across the environments ( e . g . in addition to glucose or salt concentration ) . In a few experiments , we spiked in re-barcoded mutants and additional neutral lineages as internal controls . Since re-barcoded mutants are identical , except for the barcode , these teach us about the precision with which we can measure a mutant’s fitness . Specifically , we spiked in 10 re-barcoded IRA1 nonsense mutants ( each with a frameshift insertion AT to ATT mutation at bp 4090 ) and 10 IRA1 missense mutants ( each with a G to T mutation at bp 3776 ) . Neutral lineages teach us about the behavior of the unmutated reference strain , which we must infer because its barcode is eliminated from the experiment before sequencing . The spiked in neutrals include ten barcoded lineages from the original evolution experiment ( Levy et al . , 2015 ) for which whole genome sequencing did not reveal any mutations ( Venkataram et al . , 2016a ) and previous fitness measurements did not reveal any deviation from the reference ( Li et al . , 2018; Venkataram et al . , 2016a ) . After a growth competition is complete , we extracted DNA from frozen samples following either a protocol described previously ( for batches 1–6 and 10 ) ( Venkataram et al . , 2016a ) or a modified protocol that improves the ease and yield of extraction . Our modified protocol is as follows . For each sample , a single tube of the three that were frozen for each sample ( see Conducting the barcoded fitness measurements ) was removed from the freezer and thawed at room temperature . We extracted DNA from that sample using the following modification of the Lucigen MasterPure yeast DNA purification kit ( #MPY80200 ) . We transferred the thawed cells into a 15 mL conical and centrifuge for 3 min at 4000 RPM . After discarding the supernatant , the pellet was then resuspended with 1 . 8 mL of the MasterPure lysis buffer , and 0 . 5 mm glass beads were added to help with disruption of the yeast cell wall . The mix of pellet , lysis buffer , and beads was then vortexed for 10 s and incubated for 45 min at 65°C , with periodic vortexing . The solution was then put on ice for 5 min and then 900 μL of MPC Protein Reagent was mixed with the solution . We then separated protein and cell debris by centrifugation at 4000 RPM , transferring 1900 μL of supernatant to a 2 mL centrifuge tube . We further separated remaining protein and cell debris by centrifuging at 13 , 200 RPM for 5 min . The supernatant was then divided into two 2 mL centrifuge tubes , with 925 μL of the supernatant into each . Next , we added 1000 μL of isopropanol to each tube , mixed by inversion , centrifuged at 13 , 200 RPM for 5 min , and discarded the supernatant . The pellet , containing the DNA was then resuspended in 250 μL of Elution Buffer and 10 μL of 5 ng/μL RNAase A was added . This was either left at room temperature overnight or incubated at 60°C for 15 min . Next the two tubes per sample were combined into a single tube and 1500 μL of ethanol was added . This was then mixed by inversion , and strands of precipitating DNA appeared . This was centrifuged at 13200 RPM for 2 min , and the supernatant was discarded . We again precipitated the DNA by resuspending with 750 μL of ethanol , and collected the DNA by centrifuging 13200 RPM for 2 min . The supernatant was discarded , and the tubes were left to air dry . Finally , we resuspended the pellet in Elution Buffer to a final concentration of 50 ng/μL for later use in PCR reactions ( approximately 3600 ng of DNA were used for the PCR reactions ) . To avoid the vast majority of our sequencing reads mapping only to the reference strain ( and thus not being informative to relative fitness of the mutants ) , we use restriction digest to cut the ApaLI restriction site in the middle of the reference strain’s barcode region . We mixed 43 μL of the second step PCR product with 2 μL of ApaLI ( NEB #R0507L ) and 5 μL of 10X Cutsmart and incubated at 37°C for at least 2 hr ( up to overnight ) . After digestion , we conducted size selection by running the digested sample on a gel , removing all product less than 300 bp , and isolating the DNA using a standard ThermoScientific GeneJET Gel Extraction protocol . Our expected product is 350 bp . We did not remove longer sequences via gel extraction because of the possibility that some barcode sequences may selectively form complexes with themselves or other barcodes . Note that for some samples , we also digested the reference strain before PCR , in addition to after PCR , to decrease the amount of reference strain barcode . For these samples , we mixed 80 μL of genomic DNA ( at concentration 50 ng/μL ) with 10 μL of 10X Cutsmart and 2 μL of ApaLI and incubated 37°C for at least 2 hr ( up to overnight ) . This product was then used as the template for PCR step 1 ( with appropriate water volume adjustments to ensure 50 μL reactions ) . We used the Qubit High Sensitivity ( ThermoFisher #Q32854 ) method to quantify the concentration of the final product for each sample , then pooled samples with different dual indices in equal frequency for sequencing . Our samples were then sent to either Novogene ( https://en . novogene . com/ ) or Admera Health ( https://www . admerahealth . com/ ) for quality control ( qPCR and either Bioanalyzer or TapeStation ) and sequencing . We used 2 × 150 paired-end sequencing along with index sequencing reads on Illumina HiSeq machines using patterned flow cells ( either HiSeq 4000 or HiSeq X ) . We also used Illumina Nextseq machines with unpatterned flow cells . We found that the former was more subject to index hopping errors , please see Mitigating the effects of index hopping for a discussion of how our dual indexing reduces effects of index hopping . All amplicon samples were sequenced with at least 20% genomic DNA spiked in ( either whole genomes from an unrelated project or phi-X ) to ensure adequate diversity on the flow cell . To reduce the effects of index hopping observed on Illumina patterned flow cell technology ( including HiSeq 4000 , HiSeq X , and Novaseq machines ) ( Illumina , 2017; Sinha et al . , 2017 ) , we devise a nested unique-dual-indexing approach . This approach uses a combination of inline indices attached during the first step of PCR , as well as Nextera indices attached during the second step of PCR . The latter indices are not part of the sequencing read ( they are read in a separate Index Read ) . This process uniquely labels both ends of all DNA strands such that DNA strands from multiple samples can be multiplexed on the same flow cell . Had we only labeled one end of each DNA strand , index hopping could have caused us to incorrectly identify some reads as coming from the wrong sample . One approach to label samples with unique-dual-indices is to use 96 forward primers , each of which is paired to one of 96 reverse primers , instead our nested approach allows us to uniquely dual-index samples with only 40 total primers ( 12 forward inline , eight reverse inline , 12 Nextera i7 , 8 Nextera i5 ) . Specifically , we can use combinations of the Nextera and inline primers . One way to think of this is that there are 96 possible ways to combine the forward inline and Nextera i5 primers that are on the same side of the read , effectively creating 96 unique labels for that end of the read . To reduce the effect of index hopping contamination on our results , we included only samples that were sequenced on non-patterned flow cell technology ( HiSeq 2000 and 2500 for samples in batches 1–6 , 10 , NextSeq for samples in batch 9 ) or were sequenced on patterned flow cell technology ( patterned flow cell HiSeq ) with nested unique-dual indexing . We processed the amplicon sequencing data by first using the index tags to de-multiplex reads representing different conditions and timepoints . Then , using Bowtie2 ( Langmead and Salzberg , 2012 ) , we mapped reads to a known list of barcodes generated by Venkataram et al . , 2016a , removed PCR duplicates using the UMIs from the first-step primers , and counted the number of reads for each barcode in each sample . The source code for this step can be found at Venkataram , 2020 . We processed all raw data for this study using this pipeline , including re-processing the raw sequencing files for data from previous studies ( Li et al . , 2018; Venkataram et al . , 2016a ) so that all data was processed together using the most recent version of the code . Several samples included technical replicates where the sample was split at various times in the process , including before DNA extraction , before PCR , and prior to sequencing . Read counts across these technical replicates were merged in order to calculate the best estimate of barcode frequencies . Counts were merged after appropriately accounting for PCR duplicates as identified from Unique Molecular Identifiers . | One of the goals of evolutionary biology is to understand the relationship between genotype , phenotype , and fitness . An organism's genes – its genotype – determine its physical and behavioral traits – its phenotype . Phenotypes , in turn , affect the organisms’ chances of survival and reproduction – its fitness . However , mapping the relationships among these three variables is far from easy . Recently researchers have become able to identify many genetic mutations that increase an organism's fitness , but it is more difficult to work out how these mutations affect an organism’s phenotype , and why they are beneficial . The mutations that help organisms thrive in a particular environment are often limited to a handful of genes that affect similar biological processes . For example , microbes that grow in environments with limited sugar tend to accumulate mutations in genes involved in systems that determine whether to grow fast and carelessly or to be careful in case the sugar is never replenished . It is possible that these mutations all affect the same one or two phenotypes , such as the decision to grow or to hunker down . If this were the case , researchers should be able to easily predict how well these organisms adapt to new environments . However , it is possible that specific mutations affect several phenotypes , but these extra effects remain invisible until the environment changes and these phenotypes are revealed . To explore this possibility , Kinsler , Geiler-Samerotte , and Petrov obtained hundreds of individual yeast strains that each contained a different mutation that improved the yeast's fitness in a low sugar environment . They placed these strains into similar environments and measured their fitness . The patterns observed were used to build several models that predicted how many phenotypes each mutation must affect to explain the changes in fitness . Kinsler , Geiler-Samerotte and Petrov found that the model in which only five phenotypes were affected by the mutations was able to predict the fitness of the yeast in low-sugar environments . However , to predict the fitness of the same mutations in environments that were very different , the model had to include eight phenotypes . This suggests that although the mutations that helped yeast do well in the low sugar environment were similar in their benefits in this environment , they were not truly all the same . In fact , some mutations were quite different from the others in terms of their hidden phenotypic effects . The hidden effects of mutations can be positive or negative . One mutation might cause an organism to die in a new environment , whereas another might allow it to thrive . Understanding how this works has implications not only for evolutionary biology , but also for medical research . Pathogens that cause infection , and cells that cause cancer , often accumulate mutations in small numbers of crucial genes . Understanding how these mutations affect phenotypes that become important as the environment changes – for instance as the cells encounter new challenges as a tumor grows – and whether different mutations have different hidden effects , could improve treatments in the future . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2020 | Fitness variation across subtle environmental perturbations reveals local modularity and global pleiotropy of adaptation |
Fructophily is a rare trait that consists of the preference for fructose over other carbon sources . Here , we show that in a yeast lineage ( the Wickerhamiella/Starmerella , W/S clade ) comprised of fructophilic species thriving in the high-sugar floral niche , the acquisition of fructophily is concurrent with a wider remodeling of central carbon metabolism . Coupling comparative genomics with biochemical and genetic approaches , we gathered ample evidence for the loss of alcoholic fermentation in an ancestor of the W/S clade and subsequent reinstatement through either horizontal acquisition of homologous bacterial genes or modification of a pre-existing yeast gene . An enzyme required for sucrose assimilation was also acquired from bacteria , suggesting that the genetic novelties identified in the W/S clade may be related to adaptation to the high-sugar environment . This work shows how even central carbon metabolism can be remodeled by a surge of HGT events .
Comparative genomics is a powerful tool for discovering links between phenotypes and genotypes within an evolutionary framework . While extraordinary progress in this respect has been observed in all domains of life , analyses of the rapidly increasing number of fungal genomes available has been particularly useful to highlight important aspects of eukaryotic genomes , including a broader scope of evolutionary mechanisms than was thus far deemed likely . For example , horizontal gene transfers ( HGT ) are thought to have played a very important role in domestication ( Gibbons et al . , 2012; Marsit et al . , 2015; Ropars et al . , 2015 ) and in the evolution of metabolism in fungi ( Alexander et al . , 2016; Wisecaver and Rokas , 2015 ) . Instances of the latter are best showcased by the high frequency of HGT events involving gene clusters related to fungal primary and secondary metabolism ( Campbell et al . , 2012; Khaldi and Wolfe , 2011; Slot and Rokas , 2010; 2011; Wisecaver and Rokas , 2015 ) . When considering the horizontal transfer of single genes , those encoding nutrient transporters seem to be among the most frequently transferred ( Coelho et al . , 2013; Gonçalves et al . , 2016; Richards , 2011 ) . While the identification of HGT events can be straightforward given sufficient sampling of the lineages under study , inferences concerning the evolutionary driving forces behind HGT are often difficult and uncertain , because most HGT events identified are ancient . However , available evidence suggests that HGTs are often associated with rapid adaptation to new environments ( Cheeseman et al . , 2014; Gojković et al . , 2004; Qiu et al . , 2013; Richards et al . , 2011; Richards and Talbot , 2013 ) . In line with these findings , we recently reported on the evolutionary history of a unique , high-capacity , specific fructose transporter , Ffz1 , which is intimately associated with fructophilic metabolism in ascomycetous budding yeasts ( subphylum Saccharomycotina ) ( Gonçalves et al . , 2016 ) . Fructophily is a relatively rare trait that consists in the preference for fructose over other carbon sources , including glucose ( Cabral et al . , 2015; Gonçalves et al . , 2016; Sousa-Dias et al . , 1996 ) . The evolution of FFZ1 involved the likely horizontal acquisition of the gene from filamentous fungi ( subphylum Pezizomycotina ) by the most recent common ancestor ( MRCA ) of a lineage in the Saccharomycotina , composed so far entirely of fructophilic yeasts ( Gonçalves et al . , 2016 ) . Most of the approximately one hundred species forming this clade ( Wickerhamiella and Starmerella genera , as well as closely related Candida species ) , are associated with the floral niche and are often isolated from fructose-rich nectar ( Canto et al . , 2017; de Vega et al . , 2017; Lachance et al . , 2001 ) . Interestingly , fructophilic lactic acid bacteria , whose metabolism has been dissected in detail , also populate the floral niche ( Endo et al . , 2009; Endo and Salminen , 2013 ) . These bacteria have been shown to grow poorly on glucose , which can be at least partly explained by their lack of respiratory chain enzymes and alcohol dehydrogenase activity , deficiencies that hinder NAD+ regeneration during growth on this sugar , as shown for Lactobacillus kunkei ( Maeno et al . , 2016 ) . In contrast to glucose , fructose can be used both as a carbon source and as an electron acceptor for the re-oxidation of NAD ( P ) H ( Zaunmüller et al . , 2006 ) , providing an explanation of why it is favored over glucose . Hence , fructophily in lactic acid bacteria seems to be linked to redox homeostasis ( Endo et al . , 2014 ) . In yeasts , it is still unclear how preferential consumption of fructose may be beneficial , partly because unlike fructophilic bacteria , fructophilic yeasts grow vigorously on glucose when it is the only carbon and energy source available ( Sousa-Dias et al . , 1996; Tofalo et al . , 2012 ) . Our previous work suggested that , although a strict correlation was found so far between the presence of Ffz1 and fructophily in all species investigated ( Cabral et al . , 2015; Gonçalves et al . , 2016; Leandro et al . , 2014 ) and the requirement for FFZ1 was genetically confirmed in the fructophilic species Zygosaccharomyces rouxii ( Leandro et al . , 2014 ) , it is very likely that there are additional requirements for fructophily . Thus , the FFZ1 gene does not seem to be sufficient to impart a fructophilic character to a previously glucophilic species . To gain insight into the genetic underpinnings of fructophily in budding yeasts and how it may have become evolutionarily advantageous , here we used comparative genomics to identify traits , in addition to the presence of the FFZ1 gene , that might differentiate yeasts in the fructophilic Wickerhamiella/Starmerella ( W/S ) clade , focusing on central carbon metabolism . Our results suggest that the evolution of fructophily may have been part of a process of adaptation to sugar-rich environments , which included a profound remodeling of alcoholic fermentation involving the acquisition of bacterial alcohol dehydrogenases , which turned out to be particularly important for glucose metabolism , and an invertase , which is essential for sucrose assimilation . In general , we found a surge of bacterial-derived HGT events in the W/S clade when compared with other lineages in the Saccharomycotina ( Marcet-Houben and Gabaldón , 2010 ) , many of which seem to impact redox homeostasis .
We previously reported the acquisition of a high-capacity fructose transporter ( Ffz1 ) through HGT by the MRCA of W/S-clade species . This transporter was lost in the MRCA of the Saccharomycotina and was later transferred from a Pezizomycotina-related species to the MRCA of the W/S clade , and then from the W/S clade to the MRCA of the Zygosaccharomyces genus ( Gonçalves et al . , 2016 ) . A putative role for Ffz1 in fructophily in the W/S clade was hypothesized based on its kinetic properties ( Pina et al . , 2004 ) and the evidence that it is indispensable for fructophily in the phylogenetically distant species Z . rouxii ( Leandro et al . , 2014 ) . To test this hypothesis , a FFZ1 deletion mutant was constructed in the genetically tractable W/S-clade species Starmerella bombicola . The sugar-consumption profile in YP medium supplemented with 10% ( w/v ) fructose and 10% ( w/v ) glucose ( conditions where fructophily is apparent , hereafter referred to as 20FG medium ) , showed that fructophilic behavior was completely abolished in the ffz1∆ mutant ( Figure 1 ) , similarly to what was found in Z . rouxii ( Leandro et al . , 2014 ) . A slight increase in the glucose consumption rate was also observed for the ffz1∆ mutant compared to the wild type when cultures were grown in 20FG medium ( Figure 1 ) . One of the most distinctive metabolic characteristics found in fructophilic bacteria that distinguished them from closely related non-fructophilic species was the lack of the enzymatic activity required for ethanol production and concomitant NAD ( P ) + regeneration . In one well-studied species , the gene encoding the bifunctional alcohol dehydrogenase ( ADH ) /aldehyde dehydrogenase ( ALDH ) normally responsible for ethanol production was absent ( Endo et al . , 2014 ) , whereas in another species it was present but the encoded protein lacked the domain responsible for ethanol production ( ADH ) , while maintaining the domain that conducts the preceding reaction ( Maeno et al . , 2016 ) . On the other hand , some W/S-clade yeasts were previously known to be efficient producers of sugar alcohols or lipids to the detriment of ethanol ( Magyar and Tóth , 2011;Lee et al . , 2003a; Kurtzman et al . , 2010 ) . These observations prompted us to investigate whether alcohol dehydrogenase genes in fructophilic yeasts might also provide clues pertaining to a relation between fructophily and cofactor recycling in yeasts , as a first step toward unraveling other metabolic determinants of fructophily . In S . cerevisiae , the ADH1 gene encodes the enzyme mainly responsible for the conversion of acetaldehyde into ethanol ( de Smidt et al . , 2008 ) . Hence , we started by retrieving homologs of S . cerevisiae ADH1 from the genomes of six W/S-clade species as well as from four of their closest relatives ( Figure 2A ) using tBLASTx . Among the non-W/S-clade species considered , Candida infanticola ( Kurtzman , 2007 ) occupies a particularly informative position , since it was phylogenetically placed as an outgroup of the W/S clade in our species phylogeny ( Figure 2A ) , being its closest relative among the species included in this analysis . It has presently not been considered part of the W/S clade because it lacks the Ffz1 transporter ( Figure 2A ) and has not been isolated so far from sugar-rich habitats ( Kurtzman , 2007 ) . Notably , while the phylogenetic distance between all the species surveyed and S . cerevisiae was similar , protein sequence identity , E-value , and bitscore values denoted that predicted Adh1 proteins retrieved from W/S-clade species and C . infanticola as top hits of the tBLASTx search were much less similar to the S . cerevisiae ADH1 query than the genes recovered from their non-fructophilic counterparts Sugiyamaella lignohabitans , Blastobotrys adeninivorans , and Yarrowia lipolytica ( Figure 2B ) . Moreover , when Adh1 protein sequences from W/S-clade species and C . infanticola were used as queries in BLASTp searches in the NCBI non-redundant ( nr ) database , the top 1000 hits consisted entirely of bacterial proteins , while when Adh1 sequences of Su . lignohabitans , B . adeninivorans , and Y . lipolytica yeasts were similarly employed as queries , the top 1000 hits recovered were fungal proteins . Taken together , these results suggest that W/S-clade species and C . infanticola have Adh1 homologs of bacterial origin , in contrast to the remaining three species . The first step in the alcoholic fermentation pathway consists in the conversion of pyruvate in acetaldehyde , catalyzed by pyruvate decarboxylase ( Pdc ) ( Hohmann and Cederberg , 1990 ) . Since the gene encoding the ‘native’ enzyme catalyzing the second step , ADH1 , is missing from W/S-clade genomes and seems to have been ‘replaced’ by a bacterial version , we next examined whether PDC genes mirrored somehow the peculiarities in the evolution of ADH1 observed in the W/S clade . To this end , the sequence of the gene encoding the enzyme mainly responsible for conversion of pyruvate to acetaldehyde in S . cerevisiae , Pdc1 , was used to retrieve its homologs in the set of species identified in Figure 2A . Remarkably , the Pdc sequences retrieved in this manner from the genomes of W/S-clade species and C . infanticola were also found to be more dissimilar to S . cerevisiae Pdc1 than those recovered from the three remaining non-fructophilic species , based on sequence identity , E-values , and bitscores ( Figure 2B ) . In line with the observations for Adh1 sequences , two out of the three Pdc sequences identified in C . versatilis , were more closely related to bacterial Pdc proteins than to fungal Pdc enzymes ( Figure 2B ) . However , a BLASTp search using the third Pdc sequence from C . versatilis and the remaining Pdc sequences from W/S-clade species as queries , showed that their closest relatives were fungal proteins . In this case , the lower E-value appears to reflect the fact that the Pdc orthologs found in W/S-clade species and C . infanticola seem to belong to a decarboxylase family that is phylogenetically related to S . cerevisiae Aro10 . In S . cerevisiae , Aro10 acts preferentially on substrates other than pyruvate and is not involved in alcoholic fermentation ( Kneen et al . , 2011; Romagnoli et al . , 2012; Vuralhan et al . , 2005 ) . To better assess the phylogenetic relation between Aro10 and Pdc1-related sequences , and determine the evolutionary origin of the sequences identified in the W/S clade , a Maximum Likelihood ( ML ) phylogeny was reconstructed using the top 500 NCBI BLASTp hits using S . cerevisiae Pdc1 ( CAA97573 . 1 ) , St . bombicola putative Pdc ortholog , and C . versatilis Pdc sequences from apparent bacterial origin as queries . Putative Pdc sequences from the other W/S-clade species not available at the NCBI database were also included . This phylogeny ( Figure 3A and B ) confirmed the clustering of the W/S-clade sequences with Aro10 proteins from fungi , which indicates that PDC1 was lost in the W/S clade . Additionally , as suggested by the BLASTp results , the two Pdc1-like proteins from C . versatilis were clustered with bacterial pyruvate decarboxylases ( Figure 3C ) . All glycolytic genes were examined in the same set of species and were all found to be present and to exhibit the expected level of similarity to S . cerevisiae query proteins ( Figure 2—source data 3 ) . This , together with the fact that other inspected publicly available genome assemblies of W/S-clade species are of very high quality ( e . g . Wickerhamiella domercqiae JCM 9478 , PRJDB3620 or St . bombicola JCM 9596 from RIKEN Center ) , makes it very unlikely that alcoholic fermentation genes were missed in W/S-clade species because of insufficient coverage or quality of the genome assemblies used . The tBLASTx analyses revealed that extant ADH1 genes in W/S-clade species and also in C . infanticola were likely horizontally transferred from bacteria , and the same was observed for PDC1 in C . versatilis . To identify other genes of bacterial origin in the W/S clade that might be related to central carbon metabolism and therefore to fructophily , a systematic high-throughput analytical pipeline based on the Alien Index ( AI ) score ( Alexander et al . , 2016; Gladyshev et al . , 2008 ) was employed for HGT-detection in the previously defined group of species comprising W/S-clade representatives and close relatives outside the clade . We found a considerably larger number of genes of bacterial origin in W/S-clade species and also in C . infanticola when compared with the two species more distantly related to the W/S clade ( Su . lignohabitans and B . adeninivorans , Figure 2A , Figure 2—source data 2 ) . Notably , C . versatilis displayed the highest number of putative HGT-derived genes from bacteria ( 211 ) , followed by W . domercqiae with 80 genes for which phylogenetic clustering with bacteria could be confirmed ( Figure 2A ) . Given that all W/S-clade species possess a higher number of HGT-derived genes when compared to their closest relatives , it is possible that a surge of HGT events occurred in the MRCA of the W/S clade . In line with this hypothesis , we predicted that it would be possible to identify a meaningful number of genes that were retained in more than one W/S-clade species . Indeed , after implementation of alignment thresholds ( protein sequences > 150 amino acids ) and collapsing the replicate phylogenies and lineage-specific gene duplications , it was possible to define 52 ortholog groups with representatives in two or more W/S-clade genomes . Not excluding the possibility that independent events also occurred in the various species , this suggests that a surge of HGT events took place in the MRCA of the W/S clade and that different species subsequently retained different sets of genes , C . versatilis being the species that retained the most genes of bacterial origin ( Figure 2A ) . As expected , ADH1 was present in this group ( Figure 2—source data 2 ) . PDC1 was absent because only ortholog groups of bacterial origin detected in at least two W/S-clade species were selected , and bacterial Pdc1 proteins were only identified in C . versatilis . Another ortholog group relevant to fructose metabolism found among the 52 defined in this manner was SUC2 , which encodes an invertase in S . cerevisiae that extracellularly hydrolyzes sucrose into fructose and glucose ( Carlson et al . , 1981 ) . In S . cerevisiae , the MAL and the IMA genes have also been shown to play a role in sucrose hydrolysis ( Deng et al . , 2014; Voordeckers et al . , 2012 ) , but these genes are absent in the genomes of W/S-clade species , meaning that the horizontally transferred invertase appears to be the only enzyme with sucrose-hydrolyzing capacity encoded in the W/S-clade genomes investigated . In a phylogenetic tree reconstructed using the 200 top phmmer hits to W/S-clade species Candida magnoliae ( protein ID 2301 , see Figure 2—source data 2 ) , strong support for clustering of W/S-clade sequences to Acetobacteraceae SacC sequences was observed ( Figure 4A and Figure 4B ) . A topology comparison test ( Approximately Unbiased , AU ) also strongly supported the HGT hypothesis ( p-value=4e−07 ) . To check for putative enrichment in other protein functions among the remainder of the transferred genes , the 52 ortholog groups were subsequently cross-referenced with the Kyoto Encyclopedia of Genes and Genomes ( KEGG ) , Gene Ontology ( GO ) , and InterPro annotations . Notably , out of 43 proteins to which a GO molecular function was assigned , 16 impacted redox homeostasis and were associated with oxidoreductase activity ( GO:0016491 and GO:0016616 , 14 genes ) and peroxidase activity ( GO:004601 , two genes ) , while a BLAST KOALA annotation ( Kanehisa et al . , 2016 ) indicated that the biological processes most frequently involved were amino acid metabolism , carbohydrate metabolism , and metabolism of cofactors and vitamins ( Figure 2—source data 4 ) . Notably , some of these genes appeared to have undergone several intraspecific duplications , in particular those encoding oxidorreductases participating in various metabolic pathways . We noted that , in addition to ADH1 , other alcohol dehydrogenase genes seem to have been horizontally acquired by W/S-clade species ( Figure 2—source data 4 ) , including putative ADH6 and SFA1 orthologs , which can also participate in alcoholic fermentation in S . cerevisiae ( Drewke et al . , 1990; Ida et al . , 2012 ) . In all cases , except for SFA1 , the ‘native’ yeast orthologs appear to have been lost in W/S-clade genomes . This could imply that ethanol production is conducted by alcohol dehydrogenases of bacterial origin in W/S-clade species . To learn more about the evolutionary history of these genes , detailed phylogenetic analyses were conducted for Adh1 and Adh6 , for which maximum likelihood phylogenies were reconstructed using the top phmmer hits obtained using St . bombicola Adh1 and Adh6 proteins as queries . The resulting Adh1 tree ( Figure 5A ) included protein sequences from both bacteria and fungi . All W/S-clade species clustered with strong support with the Acetobacteraceae ( Proteobacteria ) Adh1 proteins ( Figure 5A and Figure 5B ) . Within the Adh1 W/S-clade cluster , the overall phylogenetic relationships were in line with the expected relationships between the species ( Figure 2A ) , suggesting that a single HGT event occurred in the MRCA of this clade . Topology comparison tests ( AU ) strongly supported the Adh1 HGT event to the W/S clade ( p-value=8e−03 ) , adding to the strong evidence of HGT provided by the robustly supported branch that clusters the W/S-clade xenologs with bacteria and the AI results . Adh1 sequences from C . infanticola also clustered with those of proteins of bacterial origin . However , the two Adh1 sequences from C . infanticola are not similar to those of the W/S clade ( Figure 5C ) , as might be expected if a single HGT event were responsible for the acquisition of the ADH1 gene in both lineages . These sequences instead grouped , albeit with weak support , with Adh1 sequences from the distantly related Lactobacillales and Enterobacteriales ( Figure 5C ) , implying that an independent HGT event may have occurred in the C . infanticola lineage . Nonetheless , topology tests did not strongly support an independent origin for W/S and C . infanticola Adh1 proteins ( p-value=0 . 073 ) . An extended Adh6 ML phylogeny ( Figure 6A ) was reconstructed with the top 10 , 000 phmmer hits to show that the W/S-clade sequences are indeed Adh6 orthologs . In this phylogeny , W/S-clade sequences clustered with strong support ( >95% ) with Proteobacteria Adh6 sequences , within a large cluster that also encompasses known fungal Adh6 proteins . Remarkably , while Adh1 W/S-clade sequences grouped with those of the Acetobacteraceae ( Figure 5A and Figure 5B ) , the Adh6 sequences are more closely related to those of other bacterial families , as highlighted in Figure 6B . The phmmer search failed to uncover Adh6 sequences in the C . infanticola proteome . The absence of an ADH6 ortholog was further confirmed by a tBLASTx search against the C . infanticola genome using Adh6 sequences from both S . cerevisiae ( KZV09178 . 1 ) and St . bombicola as queries . This result suggests that both ADH1 and ADH6 were lost in an ancestor of C . infanticola and the W/S clade . Interestingly , the ADH6 xenologs were apparently duplicated several times within each W/S-clade species ( Figure 6A , Figure 6B and Figure 2—source data 4 ) , and Starmerella bacillaris and C . magnoliae harbor the most paralogs ( four in total ) . Acquisition of bacterial alcohol dehydrogenases by W/S-clade yeasts could have been driven by putative benefits afforded by an enzyme with kinetic characteristics that provide some advantage that the ‘native’ enzyme presumably lacked or , alternatively , by the need to restore alcoholic fermentation after an ancestral loss event , possibly in connection to adaptation to a new environment . To help elucidate this and also taking into account the link found between Adh activity and fructophily in bacteria , we set out to expound putative functional differences between Adh xenologs found in the W/S clade and ‘native’ Adh enzymes . Specifically , we compared alcohol dehydrogenase ( Adh ) activity in three W/S-clade species , in the closely related yeast B . adeninivorans , and in the model species S . cerevisiae , as well as in a distantly related fructophilic yeast species Zygosaccharomyces kombuchaensis . All W/S-clade species tested were capable of using ethanol as carbon and energy source and have a Crabtree-negative behavior when growing on sugars , meaning that ethanol production in aerated batch cultures starts only when cell densities are very high , limiting oxygen availability ( typically OD640 nm 30 , ~5–10 g/L ethanol in St . bombicola ) . In all non-W/S-clade species tested , the characteristic NADH-dependent Adh activity was readily observed , but no NADPH-dependent Adh activity was detected ( Figure 5—figure supplement 2A ) , in line with available information concerning yeast enzymes ( Cho and Jeffries , 1998; Dashko et al . , 2014; Ganzhorn et al . , 1987; Leskovac et al . , 2002 ) . Conversely , all W/S-clade species tested ( St . bombicola , C . magnoliae , and St . bacillaris ) exhibited Adh activity when either NADH or NADPH was added to the reaction mixture ( Figure 5—figure supplement 2B ) . In fact , although both cofactors could be used for conversion of acetaldehyde into ethanol , there was a lower affinity for the substrate ( higher Km ) for NADPH-dependent activity in St . bombicola ( Figure 5—figure supplement 2C ) . Interestingly , in Acetobacter pasteurianus a bacterial species in the Acetobacteraceae , the same family as the likely donor of W/S-clade Adh1 , alcohol dehydrogenase activity was found to be NADH-dependent , although it is unclear whether NADPH was also tested as a cofactor ( Masud et al . , 2011 ) . In S . cerevisiae , the paralogous enzymes Adh1 , Adh2 , Adh3 , and Adh5 were all shown to contribute to different degrees to the inter-conversion of ethanol and acetaldehyde in a NADH-dependent manner . Although their participation in alcoholic fermentation is not substantial , as may be inferred from the lack of detectable ( NADP+ dependent ) activity in S . cerevisiae crude cell extracts ( Figure 5—figure supplement 2A ) , Adh6 and Adh7 can , in principle , also catalyze this type of reaction , using NADPH instead of NADH ( de Smidt et al . , 2008 ) . Since both ADH1- and ADH6-like genes are present in the genomes of all W/S-clade species studied , it was not clear which enzyme ( Adh1-type or Adh6-type ) was responsible for the NADPH-dependent inter-conversion of ethanol and acetaldehyde observed in W/S-clade species . To elucidate this , and to evaluate the impact of alcohol dehydrogenases on metabolism , three deletion mutants were constructed in St . bombicola ( adh1∆ , adh6a∆ , and adh6b∆ ) . During aerated growth , deletion of ADH1 ( adh1∆ ) did not seem to significantly affect specific growth rates in glucose or fructose when compared to the wild type ( Figure 5—figure supplement 3A ) . However , we noted a five-fold decrease in ethanol production ( Figure 7A ) and the absence of growth on ethanol as sole carbon and energy source in the adh1∆ mutant ( Figure 7—figure supplement 1 ) . Although some ethanol was produced , no Adh activity was detected in cell-free extracts of the adh1∆ mutant when either NADH or NADPH was used ( Figure 5—figure supplement 2E ) . These results suggest that Adh1 is the main enzyme used in alcoholic fermentation in St . bombicola and that it therefore likely accepts both NADH and NADPH as cofactors . We predicted that , if W/S-clade Adh1 enzymes were mainly used in the recycling of NADPH , which does not normally occur in yeasts , in its absence , compensation would be expected to occur in other NADP+ regenerating reactions . If on the contrary , W/S-clade Adh1 enzymes were mainly used in the recycling of NADH , the compensatory increase of a NAD+ producing reaction would be observed . In line with the cofactor preference measured in cell extracts , the significant decrease in ethanol yield seems to be counterbalanced by a concomitant increase in glycerol production ( Figure 7A ) , similarly to what has been observed in the S . cerevisiae adh1∆ mutant ( de Smidt et al . , 2012 ) . Moreover , growth of the adh1∆ mutant cultivated on identical growth medium but under limited aeration was severely affected ( Figure 5—figure supplement 3B ) . Taken together , these observations strongly suggest that Adh1 plays an important role in redox homeostasis , namely in NAD+ regeneration in the absence of oxygen because , similarly to S . cerevisiae , glycerol formation in St . bombicola is probably a NAD+ regenerating reaction . Consistent with this hypothesis , we did not detect NADP+ dependent glycerol dehydrogenase activity in cell-free extracts ( Figure 5—figure supplement 2D ) , and in at least one W/S-clade species , this reaction was shown to be NADH-dependent ( Van Bogaert et al . , 2008 ) . In contrast , mannitol production , which is a NADP+ regenerating reaction ( Lee et al . , 2003b ) , was significantly decreased in the adh1∆ mutant ( Figure 7A ) . The deletion of each of the two ADH6 paralogous genes ( adh6a∆ and adh6b∆ mutants ) did not significantly affect ethanol production ( Figure 7A ) . This means that , although some of the enzymes encoded by these genes might be involved in the production of ethanol in the absence of Adh1 , when Adh1 is functional , they are not essential for ethanol metabolism and are probably mainly involved in other metabolic reactions . Finally , to ascertain how perturbations in alcoholic fermentation might affect the relative preference of St . bombicola for fructose over glucose , we monitored the consumption of both sugars in aerated cultures over time in the adh1∆ , adh6a∆ , and adh6b∆ mutants . There was a significant decrease in sugar consumption rates in the adh1∆ mutant ( Figure 7B ) , but fructophily was still observed in all mutants and became even more pronounced as the lack of Adh1 affected glucose consumption more than it did fructose consumption . Orthologs of S . cerevisiae Pdc1 , which catalyzes the first step in the fermentative pathway , appeared to be absent in W/S-clade genomes . Since all W/S-clade species investigated produce ethanol , which requires prior decarboxylation of pyruvate to acetaldehyde , it follows that the role normally fulfilled by Pdc1 in S . cerevisiae must have been taken over by a different enzyme in W/S-clade yeasts . According to our phylogenetic analysis , the only candidate likely to assume this function would be the product of the ARO10 gene ( Figure 3A ) . However , and although it displays some amino acid sequence similarity with Pdc1 , S . cerevisiae Aro10 displayed extremely low affinity for pyruvate as a substrate ( Kneen et al . , 2011 ) . Therefore , to ascertain whether Aro10 is fulfilling a role in alcoholic fermentation in W/S-clade yeasts , an ARO10 deletion mutant ( aro10Δ ) was constructed in St . bombicola . This mutant failed to produce ethanol , as might be expected in the complete absence of pyruvate decarboxylase activity ( Figure 7A ) . Similarly to the adh1Δ mutant , it exhibited increased glycerol production , but unlike the former , it grew considerably slower than the wild type strain in aerated conditions ( Figure 7C ) . This decrease was probably mainly due to the fact that the absence of pyruvate decarboxylase activity affects other metabolic routes in addition to alcoholic fermentation , such as the production of acetate and acetoin ( Flikweert et al . , 1996 ) . The phenotype observed in this mutant confirmed that Aro10 is the only decarboxylase involved in alcoholic fermentation in St . bombicola and that , therefore , the modification of enzymatic specificities was also involved in remodeling alcoholic fermentation in W/S-clade yeasts . Importantly , in the aro10∆ mutant , glucose consumption seems to be even more seriously affected than in the adh1Δ mutant ( Figure 7B ) and , in fact , glucose was left all but untouched even after 120 hr of growth ( Figure 7C ) , while fructose was almost totally consumed . The present study describes a series of evolutionary events affecting the genes involved in alcoholic fermentation in all the species studied so far in the W/S clade , a lineage of more than 100 species that is very distantly related to S . cerevisiae . Two alternative hypotheses can be put forward concerning the order of the events underlying the observed remodeling of the fermentative pathway . Both our comparative genomics data and our experimental results suggest that loss of ‘native’ ADH1 and PDC1 orthologs preceded acquisition of bacterial counterparts , whose extant functions seem to be similar to the roles normally fulfilled by alcoholic fermentation enzymes in yeasts . However , while we found one species currently placed outside the W/S clade ( C . infanticola ) lacking PDC1 , which is consistent with loss of this gene having preceded acquisition of the bacterial versions of this gene by C . versatilis , all species in our analysis possessed either a ‘native’ or the bacterial version of the ADH1 gene . In fact , all publicly available genomes across the entire subphylum Saccharomycotina possess at least one ADH1 ortholog . Remarkably , sequencing of the genome of Candida galacta ( Figure 8 ) in the context of a distinct project ( the Y1000+ Project sequencing the genomes of all known species of Saccharomycotina; http://y1000plus . org ) ( Hittinger et al . , 2015 ) showed that it lacks both PDC1 and ADH1 ( Figure 8 , Figure 2—source data 3 ) . The phylogenetic position of this species strongly supports the hypothesis that loss of native ADH1 preceded the ( likely independent ) acquisition of the bacterial versions of the gene by the C . infanticola lineage and by the MRCA of the W/S clade . Our results so far are consistent with the hypothesis that the surge in HGT observed in the MRCA of the W/S clade is related to its adaptation to the high-sugar environment in the floral niche , as exemplified by the acquisition of the Ffz1 fructose transporter from filamentous fungi and by the likely reacquisition of alcoholic fermentation involving HGT for alcohol dehydrogenase and modification of enzyme specificity ( of Aro10 ) . Acquisition of a bacterial invertase ( sacC ) ( Martin et al . , 1987 ) by a lineage lacking this pivotal enzyme for sucrose metabolism is also in line with this hypothesis , since most floral nectars are very rich is sucrose ( Mittelbach et al . , 2015;Canto et al . , 2017 ) . To assess whether the horizontally acquired invertase gene is indeed responsible for sucrose assimilation in the W/S clade , a sacC deletion mutant ( referred henceforth to as suc2∆ to emphasize the functional relation to the well-known S . cerevisiae SUC2 gene ) was constructed in St . bombicola . Growth assays in medium supplemented with sucrose as sole carbon and energy source , showed that suc2∆ mutants were unable to grow ( Figure 4—figure supplement 1A ) , while the wild-type strain attained high-cell densities . Furthermore , the suc2∆ mutant failed to consume measurable amounts of sucrose , even after 72 hr ( Figure 4—figure supplement 1A ) . During growth of the wild-type strain on sucrose , the decrease in sucrose concentrations was accompanied by the appearance of fructose and glucose in the growth medium , strongly suggesting that the horizontally transferred SUC2 gene encodes an extracellular invertase ( Figure 4—figure supplement 1B ) . This conclusion is consistent with the apparent absence of genes encoding sucrose transporters in the W/S-clade genomes analyzed , which indicates that sucrose must be first hydrolyzed outside the cell to be used as a carbon and energy source .
The yeast lineage here named the Wickerhamiella/Starmerella ( W/S ) clade comprises several species that have previously attracted attention due to their unusual metabolic features . The most prominent example is St . bombicola , a species used for the production of sophorolipids , which are amphipathic molecules that are employed as biosurfactants ( Samad et al . , 2015; Takahashi et al . , 2011 ) . Starmerella bacillaris is often found in wine fermentations and is known for diverting an important fraction of its carbon flux towards the production of glycerol instead of ethanol ( Englezos et al . , 2015 ) . Candida magnoliae has been reported to be capable of producing large amounts of erythritol ( Ryu et al . , 2000 ) . More recently , we reported that fructophily was an important common trait that unified these species and others belonging to the W/S clade , and we also consubstantiated a strict correlation between the presence of the transporter Ffz1 and fructophily . Here , we show that presence of the Ffz1 transporter is a pre-requisite for fructophily in St . bombicola , as previously observed in Z . rouxii ( Leandro et al . , 2014 ) . The stronger emphasis on the production of sugar alcohols as byproducts of metabolism at the expense of ethanol seemed also to be a common trait between the species examined ( Lee et al . , 2003b; Ryu et al . , 2000 ) , which led us to hypothesize that the preference of these yeasts for fructose was likely to be part of a broader remodeling of metabolism connected to the adaptation to the high-sugar environments where these yeasts thrive . The present work reflects our effort to uncover other aspects of this adaptation using comparative genomics as a starting point . Our examination of genes acquired from bacteria showed that the number of HGT events from bacteria into the W/S clade and its neighbor lineage ( represented by the species C . infanticola ) , far exceeded the number of events reported for other Saccharomycotina lineages ( Marcet-Houben and Gabaldón , 2010 ) and was also considerably higher than those we identified in two species closely related to , but phylogenetically clearly outside the W/S clade ( Su . lignohabitans and B . adeninivorans ) . The largest number of HGT events was detected in the earliest-diverging species in the W/S clade , C . versatilis , which together with the phylogenetic signal in the genes that were acquired through HGT , suggests that a large surge of acquisitions probably occurred in the MRCA of the clade ( Figure 2A ) . Under this model , most extant lineages subsequently discarded a large portion of the xenologs originally present in the common ancestor . In line with this hypothesis , it was possible to identify 52 xenolog ortholog groups that were present in at least two W/S-clade species . It seems likely that , in addition to HGT events common only to W/S-clade species , additional HGT events took place in the MRCA of C . infanticola and the W/S clade because , from the inspection of the phylogenies constructed for C . infanticola , at least six genes of apparent bacterial origin in this species also have bacterial origin in the W/S clade . While the bacterial donor lineage seems to be quite different between C . infanticola and the W/S clade for one gene ( ADH1 ) , strongly suggesting that they were acquired in separate events , the other five genes share a common ancestor , possibly pointing to a single event . As far as can be presently assessed from the output of our AI pipeline , the remaining HGT-derived genes seem to be specific to C . infanticola . In addition , we noted that the set of 52 xenologs present in at least two extant W/S-clade species is enriched for genes encoding proteins that affect redox balance in the cell . In fact , changes in fluxes through main metabolic pathways were previously shown to impact redox balance and oxidative stress in yeasts ( González-Siso et al . , 2009 ) , which is consistent with our hypothesis that associates the acquisition of bacterial genes with adaptive changes in metabolism . The most striking finding concerning the function of the transferred genes is the profound remodeling of the ubiquitous alcoholic fermentation pathway used by yeasts to convert pyruvate into ethanol with concomitant regeneration of NAD+ . The first step of the pathway , consisting of the conversion of pyruvate to acetaldehyde , is normally catalyzed by the enzyme Pdc1 . However , in most W/S-clade yeasts , Pdc1 is absent . Here , we provided genetic evidence for one W/S-clade species , St . bombicola , that the role of Pdc1 was taken over by a related decarboxylase encoded in S . cerevisiae by the ARO10 gene . In S . cerevisiae , phenylpyruvate is the primary substrate of Aro10 , which links this enzyme to amino acid catabolism , rather than alcoholic fermentation ( Romagnoli et al . , 2012; Vuralhan et al . , 2005 ) . However , it has been shown that site-directed mutagenesis of a few selected sites was capable of considerably increasing the affinity of S . cerevisiae Aro10 for pyruvate ( Kneen et al . , 2011 ) , which supports the notion that this shift in substrate specificity may have occurred naturally in the course of evolution of W/S-clade yeasts . Intriguingly , C . versatilis , also lacks a native PDC1 gene but possesses two genes of bacterial origin encoding PDC1 orthologs , which coexist with ARO10 . In this species , it is possible that the xenologs , and not ARO10 , carry out the conversion of pyruvate into acetaldehyde . The second step in alcoholic fermentation is the conversion of acetaldehyde into ethanol , which is conducted in S . cerevisiae mainly by Adh1 . Again , ‘native’ ADH1 genes were absent from all W/S-clade genomes examined , and it seems that the MRCA of the W/S clade acquired a bacterial ADH1 gene , from which extant ADH1 xenologs found in all extant W/S-clade species examined were derived . A similar occurrence was detected involving the loss of the ‘native’ ADH6 ortholog , encoding a NADPH-dependent branched chain alcohol dehydrogenase and the acquisition of bacterial orthologs , although the bacterial donor lineages of the ADH1 and ADH6 xenologs seem to be distinct . We showed that , in St . bombicola , the ADH1 xenolog is absolutely required for growth on ethanol and is also mainly responsible for alcoholic fermentation , thereby performing the functions fulfilled by two different enzymes in S . cerevisiae ( where Adh2 catalyzes ethanol assimilation ) . The two ADH6 xenologs played a minor role , if any , in alcoholic fermentation , similarly to what happens in S . cerevisiae . All our results are consistent with the hypothesis that alcoholic fermentation was first lost in an ancestral lineage and was subsequently reacquired by W/S-clade yeasts through horizontal acquisition of genes . In one instance , a pre-existing yeast gene ( ARO10 ) also modified its enzymatic specificities to become involved in alcoholic fermentation . On the other hand , no evidence was found for the alternative hypothesis stating that acquisition preceded loss , such as the co-occurrence in the same genome of ‘native’ and bacterial Adh1 or a distinctive role for the bacterial enzyme in yeast metabolism , which might have driven the fixation of the bacterial version of Adh1 . Loss and subsequent reacquisition of a metabolic pathway through multiple HGT events was reported previously for the unicellular red algae Galdieria phlegrea where massive gene loss occurred concomitantly with adaptation to a specialized niche in the common ancestor of Cyanidiophytina red algae ( Qiu et al . , 2013 ) . The lines of evidence supporting a similar event for alcoholic fermentation in the W/S clade are threefold , as follows . First , one key aspect backing the ‘loss followed by reacquisition’ hypothesis is the identification of a yeast lineage , here represented by C . galacta , which lacks an ADH1 ortholog , either ‘native’ or bacterial . The species phylogeny presented in Figure 8 places this species close to C . infanticola , in a position consistent with our prediction for an extant representative of an Adh1- lineage pre-dating the acquisition of bacterial ADH1 genes . In addition , the two genes forming the alcoholic fermentation pathway seem to have been lost in quick succession because no genome was found that possessed only ‘native ‘ADH1 or only ‘native’ PDC1 genes . This also strongly suggests that loss of the entire pathway pre-dated acquisition of new genes or functions . While ADH1 and ADH6 xenologs persist in all extant W/S-clade species examined , bacterial PDC1 xenologs , if they were indeed also acquired by the MRCA of the clade , were subsequently lost in most species . These hypothesized losses may have occurred as Aro10 evolved to fulfill the function of Pdc1 . C . versatilis , which is the earliest-diverging species in the W/S clade , is a notable exception that possesses two PDC1 xenologs in addition to ARO10 . Hence , C . versatilis on the one hand , and the remaining W/S clade species on the other hand , seem to represent two distinct solutions restoring pyruvate decarboxylase activity , which is a second observation in line with gene loss being the ancestral event . Our assessment of the performance of the Adh1 enzymes in W/S-clade species constitutes a third argument in favor of the ‘loss followed by reacquisition’ hypothesis . We showed that W/S-clade enzymes of bacterial origin possess a potentially advantageous characteristic when compared to their yeast counterparts; they are capable of regenerating both NAD+ and NADP+ , while the yeast enzymes accept only NADH as a cofactor . However , at least in St . bombicola , elimination of Adh1 was compensated by an increase in glycerol formation , which is very likely a NAD+ regenerating reaction , and not of mannitol formation , which regenerates NADP+ . It is therefore reasonable to infer that the Adh xenologs are playing a similar role to that normally fulfilled by the native Adh enzymes . Morever , complete elimination of alcoholic fermentation , as observed in the aro10Δ mutant , resulted in an even more pronounced compensation at the level of glycerol production . In conclusion , given the evidence presented here , the presence of xenologs involved in alcoholic fermentation in extant W/S-clade yeasts can be best explained by the need to restore alcoholic fermentation in a lineage that had previously lost it . In this case , the involvement of bacterial genes seems to have been circumstantial and possibly a consequence of the availability of the donor in the same environment . The ecological opportunity for the horizontal exchange of genetic material underlying the acquisition of bacterial ADH1 seems to have existed for a long time , since W/S-clade Adh1 proteins are most closely related to those of a bacterial lineage that is also frequently associated with the floral niche ( Acetobacteraceae ) ( Iino et al . , 2012; Suzuki et al . , 2010; Tucker and Fukami , 2014 ) . SUC2 is also among the genes acquired by the MRCA of the W/S clade . Unlike the ‘native’ version of ADH1 , which is ubiquitous in the Saccharomycotina except in the W/S clade , evolution of ‘native’ SUC2 seems to be punctuated by multiple loss events resulting in a patchy extant distribution ( Carlson et al . , 1985 ) , which is probably linked to its role as a ‘social’ gene ( Sanchez and Gore , 2013 ) . Taken together , our results suggest the enticing hypothesis that the basis for fructophily arose in yeasts in a lineage devoid of alcoholic fermentation , but probably well equipped to use fructose as electron acceptor with concomitant production of mannitol ( Baek et al . , 2010; Lee et al . , 2003a ) , through the horizontal acquisition of the Ffz1 transporter that enabled the efficient utilization of fructose as carbon and energy source . Such a lineage would resemble the aro10∆ mutant in that it would be unable to use glucose in high-sugar conditions . This putative ancestor might subsequently have acquired fermentative capacity to enable efficient glucose consumption and SUC2 to permit the utilization of sucrose , thereby completing the set of tools required to use the sugars abundant in the floral niche . In conclusion , our previous ( Gonçalves et al . , 2016 ) and present results uncovered , to our knowledge for the first time , an instance of major remodeling of central carbon metabolism in fungi involving the horizontal acquisition of a multitude of genes mostly from bacteria . Thereby , fructophilic yeasts seem to have been able to overcome the inability to use glucose efficiently , as observed in extant fructophilic bacteria . As progress in the Y1000+ Project provides additional genomic information concerning species within and in the vicinity of the W/S clade , the improved phylogenetic resolution will hopefully help further extricate the complex pattern of acquisitions of genes .
Yeast strains were obtained from PYCC ( Portuguese Yeast Culture Collection , Caparica , Portugal ) or NRRL ( USDA ARS Culture Collection ) . All strains were maintained in YPD medium . A custom query consisting of ADH1 , ADH6 , PDC1 , and SUC2 genes from S . cerevisiae was used in a tBLASTx search against W/S-clade genomes ( C . versatilis JCM 5958 , St . bombicola PYCC 5882 , St . bacillaris 3044 , C . magnoliae PYCC 2903 , C . apicola NRRL Y-50540 and W . domercqiae PYCC 3067 ) . The genomes of closely related species Candida galacta NRRL Y-17645 ( Y1000+ Project ) , Sugiyamaella lignohabitans CBS 10342 , Blastobotrys adeninivorans LS3 , Trichomonascus petasosporus NRRL YB-2093 , and Saprochaete clavata CNRMA 12 . 647 were also added to the database ( Figure 8 , see Figure 2—source data 3 ) . The best tBLASTx hit sequences ( E-value <e−10 ) retrieved from the analysis were subsequently blasted against the NCBI non-redundant protein database , and orthology was assumed whenever the best hit in S . cerevisiae was the corresponding protein in the query . Genes related with glycolysis were also inspected ( Figure 2—source data 3 ) . For this analysis , distantly related fructophilic species Zygosaccharomyces kombuchaensis ( strain CBS 8849 ) was also added to the database . For the genes of bacterial origin in the W/S clade , genome location was assessed for each gene in order to discard possible assembly contaminations ( as scaffolds containing only bacterial genes ) . Although assembly artifacts cannot be excluded , all the key genes are located in considerably long scaffolds and flanked by other genes of yeast origin . Genomes of C . apicola ( Vega-Alvarado et al . , 2015 ) and C . versatilis ( PRJDB3712 ) are publicly available , while the draft genomes of the remaining W/S-clade species examined and of Z . kombuchaensis were generated in the course of a previous study ( Gonçalves et al . , 2016 ) and are also publicly available ( PRJNA416493 and PRJNA416500 ) . The genome assembly of Candida galacta NRRL Y-17645 was obtained in the context of the Y1000+ Project ( Hittinger et al . , 2015 ) using standardized sequencing ( Hittinger et al . , 2010 ) and assembly ( Zhou et al . , 2016 ) protocols and is published here for the first time . This Whole Genome Shotgun project has been deposited at DDBJ/ENA/GenBank under the accession PPSZ00000000 . The version described in this paper is version PPSZ01000000 . For St . bombicola PYCC 5882 , St . bacillaris PYCC 3044 , C . magnoliae PYCC 2903 , W . domercqiae PYCC 3067 , and C . infanticola DS-02 ( PRJNA318722 ) the complete proteome was predicted with AUGUSTUS ( Stanke et al . , 2008 ) using the complete model and S . cerevisiae , Scheffersomyces stipitis , and Y . lipolytica as references . To assess the completeness of the predicted proteomes in each case , the number of predicted proteins was compared to the number of proteins reported for the annotated proteomes of W . domercqiae JCM 9478 ( PRJDB3620 ) and S . bombicola JCM 9596 ( PRJDB3622 ) . Predictions that used S . cerevisiae as a reference turned out to be the most complete ( ~4 . 000 predicted proteins ) and were therefore used for all downstream analyses . For C . apicola NRLL Y-50540 , C . versatilis JCM 5958 , Su . lignohabitans ( NCBI ) and B . adeninivorans ( JGI ) , publicly available proteomes were used . Query protein sequences were searched against a local copy of the NCBI refseq protein database ( downloaded May 5 , 2017 ) using phmmer , a member of the HMMER3 software suite ( Eddy , 2009 ) using acceleration parameters --F1 1e-5 --F2 1e-7 --F3 1e-10 . A custom perl script sorted the phmmer results based on the normalized bitscore ( nbs ) , where nbs was calculated as the bitscore of the single best-scoring domain in the hit sequence divided by the best bitscore possible for the query sequence ( i . e . the bitscore of the query aligned to itself ) . The top ≤10 , 000 hits were retained for further analysis , saving no more than five sequences per unique NCBI Taxonomy ID . The alien index score ( AI ) was calculated for each query protein ( modified from Gladyshev et al . , 2008 ) . Two taxonomic lineages were first specified: the RECIPIENT into which possible HGT events may have occurred ( Saccharomycetales , NCBI Taxonomy ID 4892 ) , and a larger ancestral GROUP of related taxa ( Fungi , NCBI Taxonomy ID 4751 ) . The AI is given by the formula: AI=nbsO-nbsG , where nbsO is the normalized bitscore of the best hit to a species outside of the GROUP lineage , nbsG is the normalized bitscore of the best hit to a species within the GROUP lineage ( skipping all hits to the RECIPIENT lineage ) . AI is greater than zero if the gene has a better hit to a species outside of the group lineage and can be suggestive of either HGT or contamination . Note that the original Gladyshev et al . ( 2008 ) AI calculation was based on relative E-values and ranged from −460 to +460 , with AI >45 considered a strong HGT candidate . By converting the AI to a bitscore-based metric , the results are not impacted by BLAST version , database size , or computer hardware . The bitscore-based AI score ranges from −1 to +1 , with AI >0 . 1 considered strong HGT candidates . Full-length proteins corresponding to the top 200 hits ( E-value <1 × 10−10 ) to each query sequence were extracted from the local database using esl-sfetch ( Eddy , 2009 ) . Sequences were aligned with MAFFT v7 . 310 using the E-INS-i strategy and the BLOSUM30 amino acid scoring matrix ( Katoh and Standley , 2013 ) and trimmed with trimAL v1 . 4 . rev15 using its gappyout strategy ( Capella-Gutiérrez et al . , 2009 ) . Proteins with trimmed alignments < 150 amino acids in length were excluded . The topologies of the remaining proteins were inferred using maximum likelihood as implemented in IQ-TREE v1 . 5 . 4 ( Nguyen et al . , 2015 ) using an empirically determined substitution model and rapid bootstrapping ( 1000 replications ) . The phylogenies were midpoint rooted and branches with local support less than 95 were collapsed using the ape and phangorn R packages ( Paradis et al . , 2004; Schliep , 2011 ) . Phylogenies were visualized using ITOL version 3 . 0 ( Letunic and Bork , 2016 ) . For the data included in Figure 2A , the number of genes for which a bacterial origin was confirmed after inspection of the correspondent phylogenetic tree are shown . AI results per species are shown as in Figure 2—source data 2 . We subsequently selected HGT candidates in the W/S clade ( Figure 2—source data 4 ) according to two criteria: the putative bacterial ortholog should be present in at least two W/S-clade species , and it should cluster with a bacterial lineage with strong bootstrap support ( >90% ) . Putative HGT-derived genes shared with non-W/S-clade species were excluded . Application of these criteria yielded 200 strong candidate trees , many of which referred to the same ortholog ( genes shared by several W/S species and/or paralogs , as the case of ADH1 which corresponds to 10 trees ) . A final number of 52 different ortholog groups was established after collapsing the replicate phylogenies and lineage-specific gene duplications . The resulting set of 52 orthologs was subsequently cross-referenced with GO and InterPro annotations provided by InterproScan and the Joint Genome Institute MycoCosm Portal ( Grigoriev et al . , 2014 ) . KEGG annotation was performed using the KAAS database ( Moriya et al . , 2007 ) . A BLAST KOALA annotation ( Kanehisa et al . , 2016 ) was also conducted in the final dataset ( 52 proteins ) . Species phylogenies were constructed according to Gonçalves et al . ( 2016 ) . The same dataset was used with the addition of W/S-clade species C . apicola NRRL Y-50540; C . versatilis JCM 9598; and close relatives C . infanticola DS-02 , Candida galacta NRRL Y-17645 ( Y1000+ Project ) , Sugiyamaella lignohabitans CBS 10342 ( Su . lignohabitans ) , Trichomonascus petasosporus NRRL YB-2093 ( T . petasosporus ) , and Saprochaete clavata CNRMA 12 . 647 ( Sa . clavata ) . Rpa1 , Rpa2 , Rpb1 , Rpb2 , Rpc1 and Rpc2 protein sequences for each species were used to construct the ML tree with RAxML ( Stamatakis , 2006 ) v7 . 2 . 8 using the PROTGAMMAILG model of amino acid substitution and 1000 rapid bootstraps . Species names abbreviations and accession numbers of the proteins used to construct the phylogeny are indicated in Figure 2—source data 1 . Phylogenetic relationships are in agreement with the recently published phylogeny of 86 yeast species based on genome sequences ( Shen et al . , 2016 ) . For the construction of the Pdc1 phylogeny , the top 500 NCBI BLASTp hits from searches against the non-redundant database using Pdc1 from S . cerevisiae ( CAA97573 . 1 ) , St . bombicola Pdc1-like , and C . versatilis Pdc1-like from apparent bacterial origin as queries , were selected . Sequences with more than 90% similarity were removed using CD-HIT v 4 . 6 . 7 ( Li and Godzik , 2006 ) . Pdc1 sequences from the closest relative species T . petasosporus , B . adeninivorans , and Sa . clavata of the W/S clade were also added to the alignment . A total of 479 proteins were aligned using MAFFT v 7 . 2 . 15 , ( Katoh and Standley , 2014 ) using the fast but progressive method ( FFT-NS-2 ) . Poorly aligned regions were removed with trimAl ( Capella-Gutiérrez et al . , 2009 ) using the ‘gappyout’ option . The ML phylogeny in Figure 3 was constructed with IQ-TREE v 1 . 4 . 3 ( Nguyen et al . , 2015 ) using the LG + I + G substitution model . For the Suc2 phylogeny ( Figure 4 ) , the top 200 hits from the phmmer search against the local database were selected and the ML phylogeny was constructed as above . Given the absence of other Saccharomycotina sequences in the top 200 hits for Adh1 , the top 4000 hit sequences were selected instead . For Adh6 , no Saccharomycotina sequences were found even in the top 4000 phmmer hits , so the top 10 , 000 top hit sequences were used in this case . For both phylogenies , CD-HIT ( Li and Godzik , 2006 ) was used to remove sequences with more than 85% ( Adh1 ) and 80% ( Adh6 ) similarity . For Adh1 , a preliminary ML tree was constructed to eliminate sequences outside the Adh1 family . A final set of 976 Adh1 sequences were subsequently aligned and trimmed as aforementioned and used to construct the final Adh1 phylogeny . ML phylogenies were constructed as previously described . Original raw phylogeny files can be accessed using the following links: https://figshare . com/s/c59f135885f31565a864 . The likelihood of HGT for ADH1 and SUC2 was investigated assuming monophyly of Saccharomycotina as the constrained topology ( Figure 5—figure supplement 1A and Figure 5—figure supplement 1C ) . The best tree was inferred in RAxML , and ML values for constrained and unconstrained trees were also calculated . The AU test ( Shimodaira , 2002 ) implemented in CONSEL ( Shimodaira and Hasegawa , 2001 ) was used to compare the unconstrained best tree and the best tree given a constrained topology . To test the independence of Adh1 acquisition , monophily of W/S clade and C . infanticola was assumed ( Figure 5—figure supplement 1B ) . Phylogenies were visualized using iTOL version 3 . 0 ( Letunic and Bork , 2016 ) . Cultures were grown overnight in YPD medium until late exponential phase ( OD640nm ~15–25 ) . Cells were then collected by centrifugation ( 3000 x g for 5 min ) , washed twice with cold Tris-HCL buffer ( pH = 7 . 6 ) , and disrupted with glass beads in 500 µL of Lysis Buffer ( 0 . 1 M triethanolamine hydrochloride , 2 mM MgCl2 , 1 mM DTT and 1 µM PMSF ) with six cycles of 60 s vortex-ice . Cell debris were removed by centrifugation at 4°C and 16 , 000 x g for 20 min and the extracts were stored at −20°C . Alcohol dehydrogenase activity ( Adh ) assays were performed at 25°C in 500 µL reaction mixtures containing 50 mM Potassium Phosphate buffer ( pH = 7 . 5 ) , 1 mM of NADH or NADPH , and 25 µL of cell-free extract . The reaction was started by adding acetaldehyde to a final concentration of 100 mM , and reduction of NADH and NADPH was monitored spectrophotometrically by the decrease in absorbance at 340 nm for two minutes . For St . bombicola PYCC 5882 , 5 mM , 12 . 5 mM , 50 mM , and 100 mM as final concentrations of acetaldehyde were also used . The absence of Adh activity in the adh1∆ mutant was also confirmed with protein extract of adh1∆ cells grown in 20FG medium ( conditions where ethanol was detected by HPLC in the mutant , Figure 7A ) , using up to three times more protein extract . For the detection of NADP+-dependent glycerol dehydrogenase activity ( Klein et al . , 2017 ) in St . bombicola , cultures and cell-free extracts were obtained as above . Glycerol dehydrogenase activity was measured in a reaction mixture containing 50 mM Tris-HCl ( pH = 8 . 5 ) buffer , 1 mm NADP+ and 25 µL of cell-free extract . The reaction was started by adding glycerol to a final concentration of 100 mM , and NADPH formation was monitored spectrophotometrically for two minutes . Standard molecular biology techniques were performed essentially as described in Sambrook and Russell ( 2001 ) using E . coli DΗ5α as host . St . bombicola PYCC 5882 was used in all procedures involving this species . Disruption constructs were designed essentially as outlined by Van Bogaert et al . ( 2008 ) ( Van Bogaert et al . , 2008 ) . The St . bombicola GPD promoter was first amplified by PCR and fused to the hygromycin B phosphotransferase gene ( hygB ) from E . coli and the CYC1 terminator from S . cerevisiae . Phusion High Fidelity ( Thermo Fisher Scientific , Waltham , MA ) was used for hygB and GPD promoter amplifications . A 491 bp fragment of the GPD promoter ( Van Bogaert et al . , 2008 ) was amplified using the primer pair GPD_SacI_Fw/GPD_Hind III_Rv ( Figure 7—source data 2 ) . The TEF1 promoter from p414TEF-CYC ( Mumberg et al . , 1995 ) vector was replaced by the GPD promoter using Sac I and Hind III . The primer pair Hyg_Hind III_Fw/Hyg_Xho I_Rv was used to amplify the hygB gene from a commercial plasmid ( pBlueScript-hyg , [Niklitschek et al . , 2008] ) . The amplicon was then cloned into the previously obtained p414GPD-CYC plasmid . The resulting plasmid harbors the hygromycin resistance gene controlled by the St . bombicola GPD promoter and followed by the CYC1 terminator of S . cerevisiae . For disruption of the ADH1 , SUC2 , and ARO10 genes the coding sequences ( CDS ) with 1 kb upstream and downstream were amplified from genomic DNA using the primer pairs listed in Figure 7—source data 2 . The two fragments , corresponding to each of the genes , were separately cloned into the PJET1 . 2 plasmid . The GPD-HYG-CYC cassette was then cloned into each of the resulting PJET1 . 2 plasmids using the restriction enzymes listed in Figure 7—source data 2 . For FFZ1 , ADH6a , and ADH6b , two sets of primers were used to amplify 1 kb upstream and 1 kb downstream of the CDS . The upstream and downstream fragments of each of the genes were subsequently cloned into the p416GPD-HYG-CYC plasmid using suitable enzymes ( Figure 7—source data 2 ) , yielding three plasmids each containing a distinct disruption cassette . St . bombicola was transformed by electroporation with each gene disruption construct in turn , amplified by PCR from the respective plasmid template using Phusion High Fidelity DNA polymerase and the primers listed in ( Figure 7—source data 2 ) , using the protocol described by Saerens et al . ( 2011 ) ( Saerens et al . , 2011 ) . Two different transformants from each gene disruption transformation were subsequently used for all phenotypic assays . For metabolite and sugar consumption profiling ( Figure 7A and B ) , 10 mL cultures of St . bombicola ( wild type and mutants ) were grown overnight at 30°C with orbital shacking ( 180 rpm ) in YP medium supplemented with 10% ( w/v ) of fructose and 10% ( w/v ) of glucose . The overnight culture was used to inoculate a fresh culture in 30 mL of the same medium to an OD640nm of 0 . 2 , which was incubated under the same conditions . Growth was monitored until late stationary phase was reached ( typically after 150 hr ) , and 2 mL samples were taken at several time points , centrifuged at 12 , 000 x g for 5 min , and analyzed by HPLC , as previously described ( Gonçalves et al . , 2016 ) . Statistical significance was tested using a one way ANOVA using the Bonferroni’s correction for multiple testing , implemented in GraphPad Prism v5 . Growth on sucrose and on ethanol was assessed in wild type and mutants ( suc2∆ , adh1∆ and aro10∆ ) cultivated overnight in YP medium supplemented with 2% ( w/v ) sucrose ( suc2∆ ) or 2% ( v/v ) ethanol ( adh1∆ and aro10∆ ) with orbital shacking ( 180 rpm ) at 30°C . Cultures were transferred into the same medium ( OD640nm = 0 . 2 ) , and growth was monitored over time . Consumption of sucrose at different time points was monitored by HPLC . St . bombicola wild type and adh1∆ mutants were tested for growth aerobically ( favoring respiration ) and in microaerophily ( favoring fermentation ) conditions . For the mutants , two biological replicates from independent transformations were used . For microaerophily experiments , a 24 hr pre-culture in SC medium supplemented with 0 . 2% glucose was performed . These cultures were used to inoculate 200 µL of SC medium supplemented with the desired carbon source ( 2% ( w/v ) glucose or 2% ( w/v ) fructose ) , at a 1:40 ratio in a 96 well plate . The absorbance of each well was read by an unshaken BMG FLUOstar Omega plate reader ( Kuang et al . , 2016 ) every 120 min at 600 nm for five days . For aerobic growth , a 10 mL pre-culture in SC medium supplemented with 0 . 2% glucose was performed overnight with shaking ( 200 rpm ) . Cells were transferred to 30 mL ( in a 250 mL flask ) of SC medium supplemented with 2% ( w/v ) glucose or 2% ( w/v ) fructose until a final OD640nm=0 . 2 . Cultures were grown for 5 days , with shaking . | Cells build their components , such as the molecular machinery that helps them obtain energy from their environment , by following the instructions contained in genes . This genetic information is usually transferred from parents to offspring . Over the course of several generations , genes can accumulate small changes and the molecules they code for can acquire new roles: yet , this process is normally slow . However , certain organisms can also obtain completely new genes by ‘stealing’ them from other species . For example , yeasts , such as the ones used to make bread and beer , can take genes from nearby bacteria . This ‘horizontal gene transfer’ helps organisms to rapidly gain new characteristics , which is particularly useful if the environment changes quickly . One way that yeasts get the energy they need is by breaking down sugars through a process called alcoholic fermentation . To do this , most yeast species prefer to use a sugar called glucose , but a small group of ‘fructophilic’ species instead favors a type of sugar known as fructose . Scientists do not know exactly how fructophilic yeasts came to be , but there is some evidence horizontal gene transfers may have been involved in the process . Now , Gonçalves et al . have compared the genetic material of fructophilic yeasts with that of other groups of yeasts . Comparing genetic material helps scientists identify similarities and differences between species , and gives clues about why specific genetic features first evolved . The experiments show that , early in their history , fructophilic yeasts lost the genes that allowed them to do alcoholic fermentation , probably since they could obtain energy in a different way . However , at a later point in time , these yeasts had to adapt to survive in flower nectar , an environment rich in sugar . They then favored fructose as their source of energy , possibly because this sugar can compensate more effectively for the absence of alcoholic fermentation . Later , the yeasts acquired a gene from nearby bacteria , which allowed them to do alcoholic fermentation again: this improved their ability to use the other sugars present in flower nectars . When obtaining energy , yeasts and other organisms produce substances that are relevant to industry . Studying natural processes of evolution can help scientists understand how organisms can change the way they get their energy and adapt to new challenges . In turn , this helps to engineer yeasts into ‘cell factories’ that produce valuable chemicals in environmentally friendly and cost-effective ways . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2018 | Evidence for loss and reacquisition of alcoholic fermentation in a fructophilic yeast lineage |
The association between pregnancy and altered cutaneous pigmentation has been documented for over two millennia , suggesting that sex hormones play a role in regulating epidermal melanocyte ( MC ) homeostasis . Here we show that physiologic estrogen ( 17β-estradiol ) and progesterone reciprocally regulate melanin synthesis . This is intriguing given that we also show that normal primary human MCs lack classical estrogen or progesterone receptors ( ER or PR ) . Utilizing both genetic and pharmacologic approaches , we establish that sex steroid effects on human pigment synthesis are mediated by the membrane-bound , steroid hormone receptors G protein-coupled estrogen receptor ( GPER ) , and progestin and adipoQ receptor 7 ( PAQR7 ) . Activity of these receptors was activated or inhibited by synthetic estrogen or progesterone analogs that do not bind to ER or PR . As safe and effective treatment options for skin pigmentation disorders are limited , these specific GPER and PAQR7 ligands may represent a novel class of therapeutics .
Cutaneous pigmentary changes have been long recognized as common side effects of pregnancy . The British physician Daniel Turner , in his 1714 De Morbis Cutaneis ( Turner , 1726 ) , references Hippocrates ( 460–370 B . C . E . ) , “There is a spot on the face…more peculiar , according to our great master Hippoc . , to Big Belly’d women , and recon’d as one of the Signs of Conception . ” Modern physicians recognize this common pregnancy-associated hyperpigmentation as melasma ( Sheth and Pandya , 2011; Nicolaidou and Katsambas , 2014 ) . Hippocrates also thought that the pigment was predictive of the sex of the fetus: Quae utero gerentes , maculum in facie veluti ex solis adustione habent , eae faemellas plerumque gestant . Translated to English: pregnant women who have a mark on the face as though stained by the sun , quite often give birth to girls . While Turner noted this association with fetal sex to be 'fallible' , Hippocrates was remarkably astute in linking the pigment increases to the tanning response to DNA-damaging solar ultraviolet ( UV ) radiation . While early physicians attributed the pigment changes to “Retention of the menstrual Flux” ( Turner and Cutaneis , 1726 ) , the molecular mechanisms through which pregnancy-associated hormonal changes modulate skin color have remained elusive for over 2 , 000 years . Melanocytes in the basal epidermis control skin pigmentation through synthesis of melanin , a complex process thought to be primarily regulated by alpha-melanocyte stimulating hormone ( αMSH ) ( Figure 1—figure supplement 1A and B ) . The αMSH peptide is secreted centrally by the anterior pituitary gland , and locally by keratinocytes in response to UV damage ( Cui et al . , 2007 ) . αMSH binding to the melanocortin receptor 1 ( MC1R ) , a G protein-coupled receptor ( GPCR ) , activates adenylate cyclase , and increases cAMP . This secondary messenger activates a cascade of downstream transcriptional events leading to expression of genes required for melanin synthesis ( Rodríguez and Setaluri , 2014 ) . Exogenous broadly-acting adenylate cyclase activators such as plant-derived forskolin , also stimulate melanin production ( D'Orazio et al . , 2006 ) , but the degree to which other endogenous molecules , other than αMSH regulate melanin synthesis in tissue is unclear . However , the observation that melasma frequently occurs in non-pregnant women using oral contraceptive pills , which contain only steroid hormone analogs ( Sheth and Pandya , 2011; Resnik , 1967a ) , suggests that humans may maintain αMSH-independent pigment control mechanisms . Identifying these pathways , and strategies to specifically access them pharmacologically to modulate skin pigmentation , may have substantial therapeutic utility .
To examine whether sex steroids influence melanin synthesis , we treated primary human melanocytes with estrogen ( 17β-estradiol ) . This resulted in a dose-dependent melanin increase ( Figure 1—figure supplement 1C ) . After 4 days of exposure to 25 nM estrogen , a medically-relevant concentration observed during pregnancy ( Abbassi-Ghanavati et al . , 2009 ) , melanin was markedly increased ( 208% +/- 27% ) in three individual isolates of primary human melanocytes ( Figure 1A ) . The magnitude of this change was similar to that observed with αMSH ( Figure 1—figure supplement 1D ) , and is consistent with prior in vitro studies implicating estrogen in melanin synthesis ( McLeod et al . , 1994; Ranson et al . , 1988 ) . Hormonal oral contraceptives , most of which incorporate ethinyl estradiol , are associated with melasma ( Resnik , 1967b ) . Ethinyl estradiol also increased melanin to levels similar to those observed with native estrogen . To examine the effects of estrogen on melanocyte homeostasis in the context of intact human epidermis , architecturally-faithful three-dimensional organotypic skin was established utilizing normal primary epidermal keratinocytes and melanocytes in native human stroma ( Ridky et al . , 2010; Monteleon et al . , 2015; Duperret et al . , 2014; McNeal et al . , 2015 ) . After one week , estrogen-treated skin displayed a threefold increase in melanin content ( Figure 1B ) , without changes in melanocyte number or density ( Figure 1C ) . 10 . 7554/eLife . 15104 . 003Figure 1 . Estrogen and progesterone reciprocally regulate melanin synthesis . ( A ) Melanin content of primary human melanocytes treated with estrogen ( E2 ) , compared to vehicle-treated controls . ( B ) Fontana-Masson ( melanin ) staining of organotypic skin treated with vehicle or estrogen . Relative melanin content is quantified below . ( C ) MITF immunohistochemistry of organotypic skin treated with vehicle or estrogen . Melanocyte population density is quantified below . ( D ) Melanin content of primary human melanocytes treated with progesterone ( P4 ) , compared to vehicle . ( E ) , Fontana-Masson ( melanin ) staining of organotypic skin tissues treated with progesterone or vehicle . Relative melanin content is quantified below . ( F ) MITF immunohistochemistry of organotypic skin tissues treated with vehicle or progesterone . Melanocyte population density is quantified below . n=3 biologic replicates for each experiment . Error bars denote +/- s . d . , *p<0 . 05 , scale bar = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 00310 . 7554/eLife . 15104 . 004Figure 1—figure supplement 1 . Melanin production in melanocytes . ( A ) Schematic representation of the classical melanin production pathway . ( B ) Melanin production in response to αMSH . ( C ) Melanin production in response to estrogen ( E2 ) . ( D ) Melanin production by melanocytes treated with vehicle , αMSH , or estrogen . ( E ) Melanin production in response to ethinyl estradiol ( EE2 ) . ( F ) Melanin production in response to progesterone ( P4 ) . ( G ) Melanin production in response to estrogen and progesterone using iPS-derived female melanocytes . ( H ) Melanin production in response to estrogen and progesterone using facial , aged-adult melanocytes . n=3 biologic replicates for each experiment . Error bars denote +/- s . d . , *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 00410 . 7554/eLife . 15104 . 005Figure 1—figure supplement 2 . Relative proliferative response to estrogen and progesterone treatment . ( A ) Identical numbers of melanocytes from 3 donors were seeded in parallel and treated with vehicle or estrogen; relative cell number after 5 days was determined . ( B ) Identical numbers of melanocytes from 3 donors were seeded in parallel and treated with vehicle or progesterone; relative cell number after 5 days was determined . n=3 biologic replicates for each experiment . Error bars denote +/- s . d . , *p<0 . 05 . 1DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 005 Estrogen effects in other tissue types are often counter-balanced by progesterone ( Ismail et al . , 2015 ) , which also increases during pregnancy . To determine whether this reciprocal relationship is active in melanocytes , we treated cells with physiologic levels of progesterone , which resulted in a dose-dependent decrease in melanin production ( Figure 1—figure supplement 1F ) . At 500 nM , a concentration observed in third trimester pregnancy ( Abbassi-Ghanavati et al . , 2009 ) , progesterone decreased melanin production by half ( 58% +/- 11 . 4% ) , both in culture ( Figure 1D ) and in skin tissue ( Figure 1E ) , without altering melanocyte cell number ( Figure 1F ) . Most of our primary melanocytes were derived from newborn male foreskin . To determine whether female cells also responded similarly , we treated female iPS-derived melanocytes with estrogen and progesterone and noted responses similar to those observed with the male cells ( Figure 1—figure supplement 1G ) . To determine whether melanocytes isolated from body sites other than foreskin also responded similarly to sex hormones , we treated melanocytes from adult facial skin with estrogen and progesterone and again observed responses that were similar to those observed with the foreskin melanocytes ( Figure 1—figure supplement 1H ) . Consistent with other groups who have noted that steroid hormones have variable effects on melanocyte proliferation in culture ( Im et al . , 2002 ) , we observed modest changes in proliferation when isolated primary cells were treated with estrogen or progesterone in vitro . Estrogen treated melanocytes tended to proliferate slightly slower , while progesterone treated cells tended to proliferate slightly faster ( Figure 1—figure supplement 2A–B ) . The effects varied with the basal level of melanin production . Melanocytes from dark skin were more sensitive to progesterone than estrogen , while melanocytes from light skin were more sensitive to estrogen . These proliferation changes are likely an in vitro artifact , as adult interfollicular epidermal melanocytes are relatively nonproliferative in vivo , and we did not note any changes in melanocyte numbers in sex steroid-treated 3-D organotypic tissues . Consistent with this lack of melanocyte proliferation in interfollicular epidermis , another group thoroughly examined 280 tissue sections from normal human skin from 18 donors , and identified only 2 proliferative interfollicular melanocytes ( Jimbow et al . , 1975 ) . To determine the mechanisms mediating estrogen and progesterone pigment effects , we examined components of the canonical pigment production pathway , and observed a cAMP increase upon estrogen treatment ( Figure 2A ) , suggesting that estrogen accesses the canonical pigment production pathway downstream of MC1R . Consistent with this , pCREB and MITF proteins were similarly induced ( Figure 2B ) . In contrast , progesterone reciprocally decreased melanin , cAMP , pCREB and MITF ( Figures 2C–D ) . 10 . 7554/eLife . 15104 . 006Figure 2 . Estrogen and progesterone access the classical melanin production pathway through nonclassical receptors . ( A ) cAMP ELISA from estrogen-treated melanocytes ( B ) Western blot demonstrating changes in classical melanin pathway regulators after a 16 hr estrogen treatment . ( C ) cAMP ELISA from progesterone-treated melanocytes . ( D ) Western blot demonstrating changes in classical melanin pathway regulators after a 16 hr progesterone treatment . ( E ) Melanin assay from melanocytes treated with estrogen and progesterone simultaneously . ( F ) Western blot for estrogen and progesterone receptors in MCF7 cells and melanocytes . ( G ) Melanin content of melanocytes transduced with control shRNA or shRNA targeting GPER . Cells were treated with either vehicle or estrogen . ( H ) Melanin assay performed on melanocytes transduced with control shRNA or shRNA targeting PAQR7 . Cells were treated with either vehicle or progesterone . n=3 biologic replicates for each experiment . Error bars denote +/- s . d . , *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 00610 . 7554/eLife . 15104 . 007Figure 2—source data 1 . List of GPCR transcripts expressed in primary human melanocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 00710 . 7554/eLife . 15104 . 008Figure 2—figure supplement 1 . Hormone receptors in melanocytes . ( A ) Relative gene expression of classical hormone receptors in MCF7 cells and melanocytes , as determined by qRT-PCR . Ct values were normalized to actin , and set relative to the expression of androgen receptor ( AR ) in MCF7 cells . ( B ) Average RPKM values for classical and nonclassical estrogen and progesterone receptor transcripts in human melanocytes , by convention , RPKM values >1 indicate the gene is expressed . ( C ) Expression of GPER and PAQR7 displayed as 1/Ct value . ( D ) Relative expression of GPER and PAQR7 transcripts in melanocytes , fibroblasts , and keratinocytes , as determined by qRT-PCR , displayed relative to the expression level in melanocytes . ( E ) qRT-PCR showing mRNA knockdown efficiency of the two hairpins targeting GPER . ( F ) qRT-PCR showing mRNA knockdown efficiency of the two hairpins targeting PAQR7 . ( G ) Melanin content of melanocytes transduced with LentiCRISPRV2 with guide RNA targeting GFP or GPER . Cells were treated with either vehicle or estrogen . ( H ) Melanin content of melanocytes transduced with LentiCRISPRV2 with guide RNA targeting GFP or PAQR7 . Cells were treated with either vehicle or progesterone . Error bars denote +/- s . d . , *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 00810 . 7554/eLife . 15104 . 009Figure 2—figure supplement 2 . Progesterone signals through Gi in melanocytes . ( A ) Melanin synthesis in response to Pertussis Toxin ( PTX ) , progesterone ( P4 ) , or both . n=3 biologic replicates for each experiment . Error bars denote +/- s . d . , *p<0 . 05DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 009 These data indicate that estrogen , progesterone , and αMSH converge on adenylate cyclase to reciprocally modulate melanin synthesis , and suggest that individual steroid effects may counter-balance each other . Consistent with this , the estrogen effects were significantly attenuated in the presence of progesterone ( Figure 2E ) . This likely helps explain why pregnancy-associated hyperpigmentation is characteristically limited to specific areas where melanocyte or UV radiation exposure is highest including the face , genital , and areolar regions ( Szabo , 1954; Staricco and Pinkus , 1957 ) . It is also possible that in the complex hormonal milieu of pregnancy , additional factors beyond the sex steroid activated pathways described here also contribute to skin color modulation . As steroid hormones are not predicted to signal through MC1R , whose natural ligand is the peptide αMSH , we sought to identify the specific receptors mediating the estrogen and progesterone pigment effects . We did not detect classical estrogen ( ER ) or progesterone ( PR ) receptors in melanocytes using qRT-PCR , although transcripts were observed in breast cells ( Figure 2—figure supplement 1A ) . Previous RNAseq studies in human melanocytes , conducted for unrelated experimental questions ( Flockhart et al . , 2012 ) , also failed to detect ER or PR transcripts ( Figure 2—figure supplement 1B ) . Consistent with this , ER or PR protein was not observed in MC via western blotting , although both receptors were readily apparent in breast cells ( Figure 2F ) . Considering MC1R is a G protein-coupled receptor ( GPCR ) , we hypothesized that alternative GPCRs mediate sex steroid pigment effects . To identify possible candidates , we analyzed whole transcriptome melanocyte RNAseq data . Of 412 known or predicted 7-pass human GPCRs ( Alexander et al . , 2013 ) , 61 distinct GPCRs were expressed in MCs , including the membrane-bound , G protein-coupled estrogen receptor ( Filardo et al . , 2002 ) ( GPER ) ( Figure 2—source data 1 and Figure 2—figure supplement 1B ) . Given that prior work in breast cancer cell lines and fish oocytes determined that estrogen binding to GPER modulates cAMP ( Filardo et al . , 2002; Thomas et al . , 2005; Cabas et al . , 2013; Majumder et al . , 2015; Pang and Thomas , 2010 ) , and that cAMP signaling stimulates melanin synthesis , we thought it possible that GPER may be the physiologically relevant human melanocyte estrogen receptor . The melanocyte RNAseq studies also demonstrated that an analogous , noncanonical G protein-coupled progesterone receptor , progestin and adipoQ receptor 7 ( PAQR7 ) ( Zhu et al . , 2003; Tang et al . , 2005 ) , was also expressed ( Figure 2—figure supplement 1B ) . We next used qRT-PCR to verify that GPER and PAQR7 are both expressed in primary human MC ( Figure 2—figure supplement 1C ) . Notably , GPER and PAQR7 expression was markedly lower in other skin cells including keratinocytes and fibroblasts ( Figure 2—figure supplement 1D ) . To establish the necessity of GPER and PAQR7 in mediating sex hormone effects in MCs , we first depleted GPER using either of two independent shRNA hairpins , which completely eliminated the melanocyte pigmentation response to estrogen ( Figure 2G and Figure 2—figure supplement 1E–F ) . Analogous shRNA-mediated PAQR7 depletion ablated the pigmentary response to progesterone ( Figure 2H and Figure 2—figure supplement 1F ) . To verify these results , we next used a complementary genetic approach based on CRISPR-Cas9 mediated gene disruption of GPER or PAQR7 , which also completely blocked the pigmentary response to estrogen and progesterone , respectively ( Figure 2—figure supplement 1G–H ) . Consistent with our model in which these receptors access melanin synthesis at the level of adenylate cyclase , PAQR7 was found to bind progesterone and regulate the final stages of sea trout oocyte meiosis through cAMP reduction ( Zhu et al . , 2003; Tokumoto et al . , 2012 ) . In that fish system , PAQR7 signals through G protein complexes containing the inhibitory G subunit ( Gi ) , which represses adenylate cyclase . To examine whether this signaling mechanism is functional in melanocytes , we treated melanocytes with progesterone in the presence of pertussis toxin ( PTX ) , an exotoxin that specifically inactivates Gi subunits via ADP-ribosylation . With PTX treatment alone , we observed a small increase in melanin production . This likely reflects inhibition of Gi released from basal PAQR7 activity , as well Gi subunits from other GPCRs that collectively contribute to the basal level of cAMP signaling observed in culture . Importantly , PTX blocked progesterone effects , establishing that progesterone signals through Gi subunits ( Figure 2—figure supplement 2A ) . To complement these genetic studies establishing that GPER and PAQR7 are the melanocyte sex steroid receptors , we utilized a pharmacologic approach employing synthetic steroid analogs with specific agonist or antagonist activity on ER , PR , GPER , or PAQR7 . Tamoxifen , an ER antagonist , is associated with melasma in breast cancer patients ( Kim and Yoon , 2009 ) . The mechanistic basis for the pigment change was previously unknown . However , tamoxifen is a GPER agonist ( Thomas et al . , 2005; Li et al . , 2010 ) , and increased melanin to levels comparable to those observed with estrogen ( Figure 3—figure supplement 1A ) . To determine whether GPER signaling was sufficient to increase melanin , we utilized the specific GPER agonist G-1 ( Bologa et al . , 2006 ) , an estrogen analog developed for mechanistic studies in other systems that does not bind ER . G-1 drove a dose-dependent increase in melanin production through pCREB and MITF that was GPER dependent ( Figure 3A–C and Figure 3—figure supplement 2A–D ) . Further establishing that GPER is the melanocyte estrogen receptor , G-1 and estrogen effects were blocked by either of two specific GPER antagonists , G-15 and G-36 ( Figure 3—figure supplement 2E ) , which do not have inhibitory activity against ER ( Dennis et al . , 2009; 2011 ) . To establish that PAQR7 signaling is sufficient to decrease melanin production , we used a specific PAQR7 agonist Org OD-02 ( CH2P4 ) , which does not bind PR ( Kelder et al . , 2010 ) . CH2P4 caused a dose-dependent decrease in melanin production through pCREB and MITF that was PAQR7 dependent ( Figure 3D–F and Figure 3—figure supplement 3A–D ) . 10 . 7554/eLife . 15104 . 010Figure 3 . GPER and PAQR7 signaling is sufficient to alter melanin production in organotypic human tissue . ( A ) Organotypic skin treated with vehicle ( left ) or G-1 ( right ) . ( B ) Fontana-Masson ( melanin ) staining of organotypic skin treated with vehicle or G-1 . Quantification of melanin content is shown on the right . ( C ) MITF immunohistochemistry of organotypic skin treated with vehicle or G-1 . Quantification of melanocyte population density is shown on the right . ( D ) Organotypic skin treated with vehicle ( left ) or CH2P4 ( right ) . ( E ) Fontana-Masson ( melanin ) staining of organotypic skin treated with vehicle or CH2P4 . Quantification of melanin content is shown on the right . ( F ) MITF immunohistochemistry of organotypic skin treated with vehicle or CH2P4 . Quantification of melanocyte population density is shown on the right . n=3 biologic replicates for each experiment . Error bars denote +/- s . d . , *p<0 . 05 , scale bar = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 01010 . 7554/eLife . 15104 . 011Figure 3—figure supplement 1 . Melanin production is altered by sex steroid analogs currently in clinical use . The effect of clinically relevant GPER agonists . ( A ) Melanin production in response to tamoxifen ( TMX ) . n=3 biologic replicates for each experiment . Error bars denote +/- s . d . , *p<0 . 05DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 01110 . 7554/eLife . 15104 . 012Figure 3—figure supplement 2 . Targeting GPER with specific agonists and antagonists . ( A ) Melanin production in response to G-1 , a specific GPER agonist . ( B ) Western blot demonstrating an increase in pCREB after a 30 min G-1 treatment . ( C ) Western blot demonstrating an increase in MITF after a 16 hr G-1 treatment . ( D ) Melanin assay performed on melanocytes lentivirally transduced with control shRNA or shRNA targeting GPER . These cells were treated with either vehicle or G-1 . ( E ) Melanin production by melanocytes treated with vehicle , G-1 , or estrogen , in the presence of selective GPER antagonists G-15 or G-36 . n=3 biologic replicates for each experiment . Error bars denote +/- s . d . , *p<0 . 05DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 01210 . 7554/eLife . 15104 . 013Figure 3—figure supplement 3 . Targeting PAQR7 with specific agonists . ( A ) Melanin production in response to CH2P4 , a specific PAQR7 agonist . ( B ) Western blot demonstrating a decrease in pCREB after a 30 min CH2P4 treatment . ( C ) Western blot demonstrating a decrease in MITF after a 16 hr CH2P4 treatment . ( D ) Melanin assay performed on melanocytes lentivirally transduced with control shRNA or shRNA targeting PAQR7 , cells were treated with either vehicle or CH2P4 . n=3 biologic replicates for each experiment . Error bars denote +/- s . d . , *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 013 To demonstrate that GPER was sufficient to promote melanin production in vivo , we synthesized G-1 to 95% purity ( Figure 4—figure supplement 1A–B ) , and formulated G-1 for topical application . We treated the right ears of mice daily for 3 weeks with vehicle or 2% ( w/v ) G-1 in DMSO , and observed a gradual increase pigmentation compared to vehicle-treated controls over 3 weeks ( Figure 4A ) . Melanin content was increased 1 . 6-fold , a cosmetically significant change , and was consistent with the magnitude of change seen in vitro ( Figure 4B–C ) . Clinically apparent skin darkening on mice increased over 2–3 weeks while pigment changes in culture were more rapid . It is likely that the synthetic GPCR ligands are metabolized and/or distributed differently in vivo then in in vitro culture , such that the effective local concentration of steroid in the in vivo setting is relatively transient . Optimization of an ideal topical formulation and dosing schedule will require additional study in the context of a human clinical trial . 10 . 7554/eLife . 15104 . 014Figure 4 . Topical GPER agonists increase pigmentation in vivo . ( A ) Mouse ear skin treated for 3 weeks with vehicle only on the left ear , and 2% ( w/v ) G-1 on the right ear . ( B ) Melanin assay on whole ear tissue that was treated with either vehicle or 2% G-1 for 3 weeks . ( C ) Fontana-Masson ( melanin ) staining of tissue sections from ears treated with either vehicle or 2% G-1 , quantification of staining on right . ( D ) Schematic model of estrogen and progesterone signaling in melanocytes . n=3 biologic replicates for each experiment . Error bars denote +/- s . d . , *p<0 . 05 , scale bar = 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 01410 . 7554/eLife . 15104 . 015Figure 4—figure supplement 1 . NMR spectrometry of synthesized G-1 . ( A ) H-NMR spectrometry of synthesized G-1 . 1H-NMR ( 500 MHz , CDCl3 ) δ 7 . 70 ( s , 1H ) , 7 . 61 ( dd , J = 5 . 0 , 2 . 5 Hz , 1H ) , 7 . 09 ( s , 1H ) , 7 . 01 ( s , 1H ) , 6 . 61 ( d , J = 8 . 5 Hz , 1H ) , 5 . 99 ( d , J = 6 . 0 Hz , 2H ) , 5 . 93 ( m , 1H ) , 5 . 67 ( d , J = 4 . 5 Hz , 1H ) , 4 . 97 ( d , J = 3 . 0 Hz , 1H ) , 4 . 12 ( d , J = 9 . 0 Hz , 1H ) , 3 . 19 ( q , J = 8 . 5 Hz , 1H ) , 2 . 54–2 . 48 ( m , 4H ) , 1 . 84–1 . 79 ( m , 1H ) . ( B ) 13C-NMR spectrometry of G-1 . 13C-NMR ( 125 MHz , CDCl3 ) δ 196 . 5 , 150 . 0 , 147 . 5 , 147 . 4 , 133 . 8 , 133 . 5 , 130 . 4 , 130 . 0 , 128 . 5 , 127 . 6 , 125 . 0 , 115 . 1 , 112 . 9 , 112 . 8 , 107 . 6 , 101 . 8 , 56 . 1 , 45 . 3 , 42 . 0 , 31 . 3 , 26 . 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 15104 . 015
Safe and effective approaches for modulating skin melanocyte function for therapeutic benefit are lacking , largely because the factors normally regulating melanocyte homeostasis are complex and incompletely deciphered . Defining these mechanisms is important however , as myriad genetic and acquired conditions including common afflictions such as acne , eczema , vitiligo , ultraviolet ( UV ) radiation exposure , traumatic injury , and pregnancy are associated with alterations in skin pigmentation that can be extensive and long-lasting ( James et al . , 2011 ) . Another population that could potentially benefit from modulating skin pigment are people with naturally light skin , especially those with red hair , who have a markedly decreased ability to synthesize UV-protective brown eumelanin as a result of inactivating mutations in MC1R ( Valverde et al . , 1995 ) . This large population is especially susceptible to photodamage , sunburns , and has an increased lifetime risk of keratinocyte and melanocyte-derived skin cancers ( Han et al . , 2006 ) . There is currently no available therapeutic that promotes protective eumelanin pigment production . However , the specific activation of GPER alternatively activates cAMP signaling , bypassing MC1R , to stimulate melanin synthesis , and could therefore be especially useful in this sun-vulnerable population . Selective GPER activation in skin could potentially be a safe alternative to intentional UV radiation exposure ( via natural sunlight or tanning beds ) for individuals seeking what they perceive as an aesthetically desirable tan . The only method currently available to increase skin melanin is UV exposure . While effective at darkening skin , the requisite DNA damage promotes premature aging , wrinkles , and skin cancer . Commonly utilized approaches for decreasing skin melanin are also often unsafe , and involve application of toxic mercury or arsenic compounds , especially common in India , China , Japan , and Korea , but also encountered in the U . S . , and recently highlighted in a report from the California Department of Public Health ( Report #14–046 , 2014 ) , or hydroquinone , a tyrosinase inhibitor , which has been banned in Europe because of concerns regarding its possible association with cancer ( McGregor , 2007 ) . Our findings describe small molecule sex steroid analogs , without these toxicities , that modulate pigment production ( Figure 4D ) . Therapeutic use of GPER or PAQR7 agonists/antagonists could potentially have effects on cells other than epidermal melanocytes . While topical delivery of such agents would likely avoid off target effects in distant tissues , there exits the theoretical possibility of off-target effects within the skin . However , we did not note any significant abnormalities in the epidermis from our in vitro or in vivo skin tissues treated with the sex steroids . GPER and PAQR7 have been identified only relatively recently , but are expressed in several tissues , and may mediate at least some of the estrogen and progesterone effects that cannot be attributed to the classical nuclear hormone receptors . GPER has been identified in the reproductive , nervous , cardiovascular , renal , pancreas , and immune systems ( Prossnitz and Barton , 2011 ) . In immune cells , GPER expression on T cells has been shown to play a role in 17β-estradiol-induced thymic atrophy and autoimmune encephalomyelitis . PAQR7 is expressed in the reproductive and nervous systems ( Tokumoto et al . , 2016 ) , and in murine macrophages ( Lu et al . , 2015 ) and in human T cells ( Dosiou et al . , 2008 ) , although the functional role of PAQR7 in those tissues remains relatively unclear . Given that the increased systemic estrogen and progesterone associated with pregnancy does not typically result in skin cancer or significant pathology in other tissues , we think it likely that the specific GPER and PAQR7 agonists will be well-tolerated . Nonetheless , formal toxicity studies and careful evaluation of human skin treated in clinical trials will be important . The finding that PAQR7 works through inhibitory Gi subunits is especially interesting , as it is the first example of a melanocyte cellular signaling cascade that actively represses melanin synthesis at the level of G-protein signaling , as opposed to classically defined pigment control mechanisms that modulate the strength of the stimulatory MC1R signal . In many animal systems , the Agouti protein decreases pigment production via physically binding to MC1R and inhibiting αMSH stimulation ( Ollmann et al . , 1998 ) , rather than through an actively suppressive mechanism . Our finding that normal primary melanocytes lack nuclear ER or PR contradicts a prior report ( Im et al . , 2002 ) . This group performed immunohistochemistry and RT-PCR to support the claim that nuclear estrogen receptors are expressed in melanocytes , but in our view , the data presented in that work is not especially convincing , and there is no evidence in that work that nuclear hormone signaling drives changes in melanin synthesis . Another group demonstrated that melanocyte protein extracts have the ability to bind radioactive estrogen , but that work did not identify the specific protein ( s ) responsible for the binding activity ( Jee et al . , 1994 ) . We do not exclude the possibility that in some tissue settings , including neoplastic lesions and possibly hair follicles , melanocytes express nuclear ER/PR . Still , as there is no known direct signaling pathway linking nuclear sex hormone receptors to the melanin synthesis machinery , it is most likely that the major effects of estrogen and progesterone on pigment production are mediated through the Gs and Gi coupled GPCRs identified in our current work . We have shown that signaling through GPCRs other than MC1R directly affects melanin production . While surveying all the 7-pass transmembrane proteins expressed in melanocytes , we noted expression of several additional receptors that may also influence melanin production . These include histamine ( Yoshida et al . , 2000; Lassalle et al . , 2003 ) and leukotriene receptors , which in other contexts are known to signal through Gs and Gi subunits ( Mondillo et al . , 2005; Arcemisbéhère et al . , 2010 ) . Future functional analysis of these and other GPCRs in melanocytes may elucidate the mechanisms responsible for the pigmentation changes that frequently accompany many skin diseases associated with inflammation . These studies may identify additional 'drugable' and therapeutically useful receptors in melanocytes , and will help advance an understanding of how cumulative GPCR signaling is integrated to regulate melanin production in human skin .
Primary melanocytes were extracted from fresh discarded human foreskin and surgical specimens as described previously described with some modifications detailed as follows . After overnight incubation in Dispase , the epidermis was separated from the dermis and treated with trypsin for 10 min . Cells were pelleted and plated on selective MC Medium 254 ( Invitrogen , Carlsbad , CA ) with Human Melanocyte Growth Supplement , and 1% penicillin and streptomycin . Lightly pigmented primary melanocytes were utilized for experiments assaying estrogen and GPER agonist effects , and heavily pigmented primary melanocytes were utilized for experiments assaying progesterone and PAQR7 agonist effects in melanin production . Female iPS-derived human melanocytes were a gift from Meenhard Herlyn ( Wistar Institute , Philadelphia , PA , USA ) . Progesterone ( P8783 ) , 17β-Estradiol ( E8875 ) , and αMSH ( M4135 ) were purchased from Sigma-Aldrich ( St . Louis , MO ) . G-1 ( 10008933 ) , G-15 ( 14673 ) and G-36 ( 14397 ) were purchased from Cayman Chemical ( Ann Arbor , MI ) . CH2P4 ( 2085 ) was purchased from Axon Medchem ( Groningen , Netherlands ) . Pertussis toxin was purchased from R&D systems ( Minneapolis , MN ) . These compounds were diluted to working stock solutions in Medium 254 . 2 x 105 melanocytes were seeded uniformly on 6-well tissue culture plates . Cells were treated with vehicle controls , sex steroids , hormone derivatives , or pertussis toxin for 4 days . Cells were then trypsinized , counted , and spun at 300 g for 5 min . The resulting cell pellet was solubilized in 120 μL of 1M NaOH , and boiled for 5 min . The optical density of the resulting solution was read at 450 nm using an EMax microplate reader ( Molecular Devices , Sunnyvale , CA ) . The absorbance was normalized to the number of cells in each sample , and relative amounts of melanin were set based on vehicle treated controls . For tissue melanin assays , tissue was weighed prior to boiling in 1M NaOH for 20 min . Samples were spun down to eliminate insoluble materials , and then the optical density of the sample was measured as previously described and normalized to the weight of tissue . Organotypic skin grafts containing MCs were established using modifications to previously detailed methods ( Ridky et al . , 2010; Chudnovsky et al . , 2005 ) . The Keratinocyte Growth Media ( KGM ) used for keratinocyte-only skin grafts was replaced with modified Melanocyte Xenograft Seeding Media ( MXSM ) . MXSM is a 1:1 mixture of KGM , lacking cholera toxin , and Keratinocyte Media 50/50 ( Gibco ) containing 2% FBS , 1 . 2 mM calcium chloride , 100 nM Et-3 ( endothelin 3 ) , 10 ng/mL rhSCF ( recombinant human stem cell factor ) , and 4 . 5 ng/mL r-basic FGF ( recombinant basic fibroblast growth factor ) . 1 . 5 x 105 melanocytes and 5 . 0 x 105 keratinocytes were suspended in 80 μL MXSM , seeded onto the dermis , and incubated at 37˚C for 8 days at the air-liquid interface . 2% ( w/v ) G-1 was prepared in DMSO , 20 μL of this solution was applied daily to the right ear , with vehicle only applied to the left ear , of 4-week-old C57BL/6 mice . These studies were preformed without inclusion/exclusion criteria , randomization , or blinding . Based on a twofold anticipated effect , we preformed this experiment with 3 biological replicates . All procedures were performed in accordance with IACUC-approved protocols at the University of Pennsylvania . Adherent cells were treated with 1 µM doses of E2 and P4 overnight , washed once with DPBS , and lysed with 1% NP-40 buffer ( 150 mM NaCl , 50 mM Tris , pH 7 . 5 , 1 mM EDTA , and 1% NP-40 ) containing 1X protease inhibitors ( Roche , Basel , Switzerland ) ) and 1X phosphatase inhibitors ( Roche ) . Lysates were quantified ( Bradford assay ) , normalized , reduced , denatured ( 95°C ) and resolved by SDS gel electrophoresis on 4–15% Tris/Glycine gels ( Bio-Rad , Hercules , CA ) . Resolved protein was transferred to PVDF membranes ( Millipore , Billerica , MA ) using a Semi-Dry Transfer Cell ( Bio-Rad ) , blocked in 5% BSA in TBS-T and probed with primary antibodies recognizing MITF ( Cell Signaling Technology , #12590 , 1:1000 , Danvers , MA ) , p-CREB ( Cell Signaling Technology , #9198 , 1:1000 ) , CREB ( Cell Signaling Technology , #9104 , 1:1000 ) , and β-Actin ( Cell Signaling Technology , #3700 , 1:4000 ) . After incubation with the appropriate secondary antibody , proteins were detected using either Luminata Crescendo Western HRP Substrate ( Millipore ) or ECL Western Blotting Analysis System ( GE Healthcare , Bensalem , PA ) . cAMP ELISA was performed on primary human melanocytes using the Cyclic AMP XP Assay Kit ( Cell Signaling Technology , #4339 ) following manufacturer instructions . Formalin-fixed paraffin embedded tissue was sectioned at 5 µM and collected on superfrost plus slides ( Fisher , Pittsburgh , PA ) , and subjected to Fontana-Masson stain for melanin ( Masson , 1928 ) . Briefly , sections were deparaffinized , rehydrated , and incubated in the following solutions: 2 . 5% aqueous silver nitrate for 10 min , 0 . 1% aqueous gold chloride for 15 min , and 5% aqueous sodium thiosulfate for 5 min . Distilled deionized water was used for rinsing and incubations were done at room temperature except for silver nitrate at 60°C . Slides were counterstained with 0 . 1% nuclear fast red Kernechtrot for 5 min , dehydrated , cleared , and coverslipped using MM24 mounting media ( Leica , Wetzlar , Germany ) . All staining reagents were from Polyscientific R and D Corporation ( Bay Shore , NY ) . Formalin fixed paraffin embedded ( FFPE ) human skin tissue sections from organotypic tissue was stained for MITF protein expression using a primary antibody to MITF ( Leica Biosystems , NCL-L-MITF , 1:15 ) . Staining was performed following the manufacturer protocol for high temperature antigen unmasking technique for immunohistochemical demonstration on paraffin sections . Tissue sections from organotypic culture were stained using methods described above . Quantification was performed according to Billings et al . ( 2015 ) . Briefly , 20X photomicrograph images of representative tissue sections were taken using the Zeiss Axiophot microscope . Tiff files of the images were saved and transferred to Adobe Photoshop where pixels corresponding to Fontana-Masson staining and epidermal counter stain were selected using the color selection tool . Images corresponding to the single specific color were then analyzed using FIJI ( Image J ) to determine the number of pixels in each sample . The numbers of pixels representing Fontana-Masson staining were normalized to the total amount epidermal counter staining . Final ratios Fontana-Masson staining in the epidermis were set relative to amount of staining in vehicle treated controls . mRNA was extracted from melanocytes according to the RNeasy Mini Kit protocol ( Qiagen , Venlo , Netherlands ) , and reverse transcribed to cDNA using the High Capacity RNA-to cDNA kit ( Applied Biosystems , Foster City , CA ) . Quantitative PCR of the resulting cDNA was carried out using Power SYBR Green Master Mix ( Applied Biosystems ) and gene-specific primers , in triplicate , on a ViiA 7 Real-Time PCR System ( Life Technologies ) . The following primers were used for detection; B-Actin forward: 5’-CAT GTA CGT TGC TAT CCA GGC-3’; B-Actin reverse: 5’-CTCCTTAATGTCACGCACGAT -3’; ER-A forward: 5’- AAA GGT GGG ATA CGA AAA GAC C -3’; ER-A reverse: 5’-AGC ATC CAA CAA GGC ACT GA-3’; ER-B forward: 5’ – GGC TGC GAG AAA TAA CTG CC -3’; ER-B reverse: 5’-AAT GCG GAC ACG TGC TTT TC-3’; PGR forward: 5’- AGG TCT ACC CGC CCT ATC TC -3’; PGR reverse: 5’-AGT AGT TGT GCT GCC CTT CC -3’; AR forward: 5’- GTG CTG TAC AGG AGC CGA AG -3’; AR reverse: 5’- GTC AGT CCT ACC AGG CAC TT -3’; GPER forward: 5’-ACA GAG GGA AAA CGA CAC CT -3’; GPER reverse: 5’- AAT TTT CAC TCG CCG CTT CG -3’; PAQR7 forward: 5’- GTG CAC TTT TAT ACC GTC TGC TT -3’; PAQR7 reverse: 5’- CCT GGG CAG GGA GCT AAG AT -3’ . Relative expression was determined using the 2-[delta][delta] Ct method followed by normalization to the AR receptor transcript levels in MCF7 cells . The following shRNAs were expressed from the GIPZ vector and are available through GE Dharmacon ( Lafayette , CO ) . shPAR7 . 3 ( V3LHS_364596 , TGTGGTAGAGAAGAGCTGG ) , shPAQR7 . 4 ( V3LHS_364598 , AGAAGTGTGCCAAGGCACT ) , shGPER . 1 ( V2LHS_132008 , TCCTTCTCCTCTTTAACTC ) , shGPER . 3 ( V3LHS_390319 , TGATGAAGTACAGGTCGGG ) . Guide RNAs were designed using software tools developed by the Zhang Lab and provided on the website http://www . genome-engineering . org/ ( Hsu et al . , 2013 ) . Guide RNAs were subsequently cloned into lentiCRISPRv2 ( Addgene # 52961 ) according to the accompanying protocol ( Sanjana et al . , 2014 ) . Guide RNA sequences are as follows: lentiCRISPR GFP 5’ GAA GTT CGA GGG CGA CAC CC 3’; lentiCRISPR GPER . 1 5’ ACAGGCCGATCACGTACTGC 3’; lentiCRISPR GPER . 2 5’ GAGCACCAGCAGTACGTGAT 3’; lentiCRISPR PAQR7 . 1 5’ CGTACATCTATGCGGGCTAC 3’; lentiCRISPR PAQR7 . 5 5’ CGTGCGGAAATAGAAGCGCC 3’ G-1 ( ± ) 1- ( 4- ( 6-bromobenzo[d][1 , 3]dioxol-5-yl ) -3a , 4 , 5 , 9b-tetrahydro-3H-cyclopenta[c]quinolin-8-yl ) ethan- 1-one was prepared by the method of Baudelle et al . ( 1998 ) 6-bromopiperonal ( 1 . 110 g , 4 . 85 mmol ) and 4-aminoacetophenone ( 656 mg , 4 . 85 mmol ) were dissolved in anhydrous acetonitrile ( 16 . 2 mL , 0 . 3M ) and allowed to stir at 25°C under argon . After approximately 1 . 5 hr , trifluoroacetic acid ( 350 μL , 4 . 61 mmol ) was added and the reaction was allowed to stir at 25°C for 45 min . Freshly prepared cyclopentadiene ( 1 . 63 mL , 19 . 4 mmol ) was added dropwise to the reaction mixture . After 2 hr at 25°C the reaction mixture was concentrated in vacuo . The crude product was purified by silica gel chromatography using 30% EtOAc in hexanes as eluent to provide racemic 1- ( 4- ( 6-bromobenzo[d][1 , 3]dioxol-5-yl ) -3a , 4 , 5 , 9b-tetrahydro-3H-cyclopenta[c]quinolin-8-yl ) ethan-1-one ( 1 . 05 g , 53% ) . The G-1 was >95% pure as determined by high pressure liquid chromatography analysis . 1H and 13C NMR were identical to the data reported by Burai et al . ( 2010 ) . *denotes a P-value of less than 0 . 05 in an unpaired , two-tailed Students T-Test , assuming a normal distribution and equal variance . Due to the anticipated effect size of a twofold change , experiments were performed with 3 biological replicates . | Factors controlling pigment production in skin are complex and poorly understood . Cells called melanocytes produce a pigment called melanin , which makes the skin darker . It has been known for a long time that skin color often changes during pregnancy , which suggests that sex hormones may be involved . However , the specific hormones and signaling mechanisms responsible for the changes have remained largely undefined . Estrogen and progesterone are two of the main female sex hormones . Natale et al . now show that estrogen increases pigment production in human melanocytes , and progesterone decreases it . For hormones to signal to cells , they must bind to and activate particular receptor proteins . Further investigation by Natale et al . revealed that estrogen and progesterone regulate pigment production by binding to receptors that belong to a family called G protein-coupled receptors . These receptors can signal rapidly once activated by sex hormones , and may serve as therapeutic targets for treating pigmentation disorders . Skin diseases that cause inflammation often also cause changes in skin color . Natale et al . noticed several other G protein-coupled receptors that are likely to control pigmentation through similar mechanisms . Future analyses of the roles that these other receptors perform in melanocytes may therefore reveal how inflammation-based pigmentation changes occur . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2016 | Sex steroids regulate skin pigmentation through nonclassical membrane-bound receptors |
The internal brain dynamics that link sensation and action are arguably better studied during natural animal behaviors . Here , we report on a novel volume imaging and 3D tracking technique that monitors whole brain neural activity in freely swimming larval zebrafish ( Danio rerio ) . We demonstrated the capability of our system through functional imaging of neural activity during visually evoked and prey capture behaviors in larval zebrafish .
A central goal in systems neuroscience is to understand how distributed neural circuitry dynamics drive animal behaviors . The emerging field of optical neurophysiology allows monitoring ( Kerr and Denk , 2008; Dombeck et al . , 2007 ) and manipulating ( Wyart et al . , 2009; Boyden et al . , 2005; Zhang et al . , 2007 ) the activities of defined populations of neurons that express genetically encoded activity indicators ( Chen et al . , 2013; Tian et al . , 2009 ) and light-activated proteins ( Kerr and Denk , 2008; Boyden et al . , 2005; Zhang et al . , 2007; Luo et al . , 2008 ) . Larval zebrafish ( Danio rerio ) are an attractive model system to investigate the neural correlates of behaviors owing to their small brain size , optical transparency , and rich behavioral repertoire ( Friedrich et al . , 2010; Ahrens and Engert , 2015 ) . Whole brain imaging of larval zebrafish using light sheet/two-photon microscopy holds considerable potential in creating a comprehensive functional map that links neuronal activities and behaviors ( Ahrens et al . , 2012; Ahrens et al . , 2013; Engert , 2014 ) . Recording neural activity maps in larval zebrafish has been successfully integrated with the virtual reality paradigm: closed-loop fictive behaviors in immobilized fish can be monitored and controlled via visual feedback that varies according to the electrical output patterns of motor neurons ( Ahrens et al . , 2012; Engert , 2012 ) . The behavioral repertoire , however , may be further expanded in freely swimming zebrafish whose behavioral states can be directly inferred and when sensory feedback loops are mostly intact and active . For example , it is likely that vestibular as well as proprioceptive feedbacks are perturbed in immobilized zebrafish ( Engert , 2012; Bianco et al . , 2012 ) . The crowning moment during hunting behavior ( Bianco et al . , 2011; Patterson et al . , 2013; Trivedi and Bollmann , 2013 ) — when a fish succeeds in catching a paramecium — cannot be easily replicated in a virtual reality setting . Therefore , whole brain imaging in a freely swimming zebrafish may allow optical interrogation of brain circuits underlying a range of less explored behaviors . Although whole brain functional imaging methods are available for head-fixed larval zebrafish , imaging a speeding brain imposes many technical challenges . Current studies on freely swimming zebrafish are either limited to non-imaging optical systems ( Naumann et al . , 2010 ) or wide field imaging at low resolution ( Muto et al . , 2013 ) . While light sheet microscopy ( LSM ) has demonstrated entire brain coverage and single neuron resolution in restrained zebrafish ( Ahrens et al . , 2013 ) , it lacks the speed to follow rapid fish movement . Moreover , in LSM , the sample is illuminated from its side , a configuration that is difficult to be integrated with a tracking system . Conventional light field microscopy ( LFM ) ( Broxton et al . , 2013; Prevedel et al . , 2014 ) is a promising alternative due to its higher imaging speed; however , its spatial resolution is relatively low . Specialized LFMs for monitoring neural activity utilizing temporal information were also developed recently ( Pégard et al . , 2016; Nöbauer et al . , 2017 ) , which rely on spatiotemporal sparsity of fluorescent signals and cannot be applied to moving animals . Here , we describe a fast 3D tracking technique and a novel volume imaging method that allows whole brain calcium imaging with high spatial and temporal resolution in freely behaving larval zebrafish . Zebrafish larvae possess extraordinary mobility . They can move at an instantaneous velocity up to 50 mm/s ( Severi et al . , 2014 ) and acceleration of 1 g ( 9 . 83 m/s2 ) . To continuously track fish motion , we developed a high-speed closed-loop system in which ( 1 ) customized machine vision software allowed rapid estimation of fish movement in both the x-y and z directions; and , ( 2 ) feedback control signals drove a high-speed motorized x-y stage ( at 300 Hz ) and a piezo z stage ( at 100 Hz ) to retain the entire fish head within the field of view of a high numerical aperture ( 25× , NA = 1 . 05 ) objective . Larval zebrafish can make sudden and swift movements that easily cause motion blur and severely degrade imaging quality . To overcome this obstacle , we developed a new eXtended field of view LFM ( XLFM ) . The XLFM can image sparse neural activity over the larval zebrafish brain at near single cell resolution and at a volume rate of 77 Hz , with the aid of genetically encoded calcium indicator GCamp6f . Furthermore , the implementation of flashed fluorescence excitation ( 200 μs in duration ) allowed blur-free fluorescent images to be captured when a zebrafish moved at a speed up to 10 mm/s . The seamless integration of the tracking and imaging system made it possible to reveal rich whole brain neural dynamics during natural behavior with unprecedented resolution . We demonstrated the ability of our system during visually evoked and prey capture behaviors in larval zebrafish .
The newly developed XLFM is based on the general principle of light field ( Adelson and Wang , 1992 ) and can acquire 3D information from a single camera frame . XLFM greatly relaxed the constraint imposed by the tradeoff between spatial resolution and imaging volume coverage in conventional LFM . This achievement relies on optics and in computational reconstruction techniques . First , a customized lenslet array ( Figure 1a , Figure 1—figure supplement 1 ) was placed at the rear pupil plane of the imaging objective , instead of at the imaging plane as in LFM . Therefore , in ideal conditions , a 2D spatially invariant point spread function ( PSF ) could be defined and measured; in practice , the PSF was approximately spatially invariant ( see Materials and methods ) . Second , the aperture size of each micro-lens was decoupled from their interspacing and spatial arrangement , so that both the imaging volume and the resolution could be optimized simultaneously given the limited imaging sensor size . Third , multifocal imaging ( Abrahamsson et al . , 2013; Perwass and Wietzke , 2012 ) was introduced to further increase the depth of view by dividing the micro-lenses array into two groups whose focal planes were at different axial positions ( Figure 1b and c , Figure 1—figure supplements 3 and 4 ) . Fourth , a new computational algorithm based on optical wave theory was developed to reconstruct the entire 3D volume from one image ( Figure 1—figure supplement 5 ) captured by a fast camera ( see Materials and methods ) . We first characterized the XLFM by imaging 0 . 5 μm diameter fluorescent beads . In our design , the system had ~ Ø800 μm in plane coverage ( Ø is the diameter of the lateral field of view ) and more than 400 μm depth of view , within which an optimal resolution of 3 . 4 μm × 3 . 4 μm × 5 μm could be achieved over a depth of 200 μm ( Figure 1—figure supplements 6 and 7 , Materials and methods ) . In the current implementation , however , the imaging performance suffered from the variation in the focal length of the micro-lenses ( Figure 1—figure supplement 8 ) , which led to spatial variance of the PSF . As a result , the reconstruction performance and the achievable optimal resolution were shown to degrade beyond the volume of Ø500 μm × 100 μm ( Figure 1—figure supplements 9 and 10 ) . To minimize the reconstruction time while assuring whole brain coverage ( ~250 μm thick ) , all imaging reconstructions were carried out over a volume of Ø800 μm × 400 μm . We next characterized the imaging performance by considering more fluorescent light sources distributed within the imaging volume . The achievable optimal resolution depends on the sparseness of the sample , because the information captured by the image sensor was insufficient to assign independent values for all voxels in the entire reconstructed imaging volume . Given the total number of neurons ( ~80 , 000 [Hill et al . , 2003] ) in a larval zebrafish brain , we next introduced a sparseness index ρ , defined as the fraction of neurons in the brain active at a given instant , and used numerical simulation and our reconstruction algorithm to characterize the dependence of achievable resolution on ρ . We identified a critical ρc ≈ 0 . 11 , below which active neurons could be resolved at the optimal resolution ( Figure 1—figure supplement 11b ) . As ρ increased , closely clustered neurons could no longer be well resolved ( Figure 1—figure supplement 11c–d ) . Therefore , sparse neural activity is a prerequisite in XLFM for resolving individual neurons at the optimal resolution . Moreover , the above characterization assumed an aberration and scattering free environment; complex optical properties of biological tissue could also degrade the resolution ( Ji , 2017 ) . We demonstrated the capabilities of XLFM by imaging the whole brain neuronal activities of a larval zebrafish ( 5 d post-fertilization [dpf] ) at a speed of 77 volumes/s and relatively low excitation laser exposure of 2 . 5 mW/mm2 ( Figure 1d , Video 1 ) . The fluorescent intensity loss due to photobleaching reached ~50% when the zebrafish , which expressed pan-neuronal nucleus-labelled GCamp6f ( huc:h2b-gcamp6f ) , was imaged continuously for ~100 min and over more than 300 , 000 volumes ( Figure 1—figure supplement 12 , Video 2 and 3 ) . To test whether XLFM could monitor fast changes in neuronal dynamics across the whole brain at high resolution ( close to single neuron level ) , we first presented the larval zebrafish , restrained in low melting point agarose , with visual stimulation ( ~2 . 6 s duration ) . We found that different groups of neurons in the forebrain , midbrain , and hindbrain were activated at different times ( Figure 1e–f , Video 1 and 4 ) , suggesting rapid sensorimotor transformation across different brain regions . To track freely swimming larval zebrafish , we transferred fish into a water-filled chamber with a glass ceiling and floor . The 20 mm × 20 mm × 0 . 8 mm-sized chamber was coupled with a piezo actuator and mounted on a high-speed 2D motorized stage ( Figure 2 ) . A tracking camera monitored the lateral movement of the fish , and an autofocus camera , which captured light field images , monitored the axial movement of the fish head ( Figure 2 , Figure 2—figure supplement 1 ) . Real-time machine vision algorithms allowed quick estimate of lateral ( within 1 ms ) and axial ( ~5 ms ) head positions ( see Materials and methods ) . The error signals in three dimensions , defined as the difference between the head position and set point , were calculated ( Figure 3a ) and converted to analog voltage signals through proportional-integral-derivative ( PID ) control to drive the motorized stage and z-piezo scanner . Tracking and autofocusing allowed for rapid compensation of 3D fish movement ( 300 Hz in x and y , 100 Hz in z , Figure 3a ) and retainment of the fish head within the field of view of the imaging objective . Our tracking system permitted high-speed and high-resolution recording of larval zebrafish behaviors . With two cameras acquiring head and whole body videos simultaneously ( Figure 2 , Figure 3b ) , we recorded and analyzed in real time ( see Materials and methods ) the kinematics of key features during larval zebrafish prey capture ( Figure 3b and c , Video 5 and 6 ) . Consistent with several earlier findings ( Bianco et al . , 2011; Patterson et al . , 2013; Trivedi and Bollmann , 2013 ) , eyes converged rapidly when the fish entered the prey capture state ( Figure 3c ) . Other features that characterized tail and fin movement were also analyzed at high temporal resolution ( Figure 3c ) . The integration of the XLFM and 3D tracking system allowed us to perform whole brain functional imaging of a freely behaving larval zebrafish ( Figure 2 ) . We first replicated the light-evoked experiment ( similar to Figure 1 ) , albeit in a freely behaving zebrafish with pan-neuronal cytoplasm-labeled GCaMP6s ( huc:gcamp6s ) , which exhibited faster and more prominent calcium response ( Video 7 ) . Strong activities were observed in the neuropil of the optical tectum and the midbrain after stimulus onset . The fish tried to avoid strong light exposure and made quick tail movement at ~60 Hz . Whole brain neural activity was monitored continuously during the light-evoked behavior , except for occasional blurred frames due to the limited speed and acceleration of the tracking stage . Next , we captured whole brain neural activity during the entire prey capture process in freely swimming larval zebrafish ( huc:gcamp6s , Video 8 ) . When a paramecium moved into the visual field of the fish , groups of neurons , indicated as group one in Figure 4b , near the contralateral optical tectum of the fish were first activated ( t1 ) . The fish then converged its eyes onto the paramecium and changed its heading direction in approach ( t2 ) . Starting from t2 , several groups of neurons in the hypothalamus , midbrain , and hindbrain , highlighted as groups two , three , and four in Figure 4b , were activated . It took the fish three attempts ( Figure 4c ) to catch and eat the paramecium . After the last try ( t4 ) , neuron activity in group one decreased gradually , whereas activities in the other groups of neurons continued to rise and persisted for ~1 s before the calcium signals decreased . The earliest tectal activity ( group 1 ) responsible for prey detection found here is consistent with previous studies ( Semmelhack et al . , 2014; Bianco and Engert , 2015 ) . Moreover , our data revealed interesting neural dynamics arising from other brain regions during and after successful prey capture . We also monitored similar behavior in a zebrafish expressing nucleus-localized GCamp6f ( huc:h2b-gcamp6f ) with better resolution but less prominent calcium response ( Video 9 ) .
Whole brain imaging in freely behaving animals has been previously reported in Caenorhabditis elegans , by integrating spinning-disk confocal microscopy with a 2D tracking system ( Venkatachalam et al . , 2016; Nguyen et al . , 2016 ) . In the more remote past , Howard Berg pioneered the use of 3D tracking microscopy to study bacteria chemotaxis ( Berg , 1971 ) . However , the significant increase of animal size imposes challenges both in tracking and imaging technologies . The XLFM , derived from the general concept of light field imaging ( Broxton et al . , 2013; Adelson and Wang , 1992; Ng et al . , 2005; Levoy et al . , 2006 ) , overcomes several critical limitations of conventional LFM and allows optimization of imaging volume , resolution , and speed simultaneously . Furthermore , it can be perfectly combined with flashed fluorescence excitation to capture blur-free images at high resolution during rapid fish movement . Taken together , we have developed a volume imaging and tracking microscopy system suitable for observing and capturing freely behaving larval zebrafish , which have ~80 , 000 neurons and can move two orders of magnitude faster than C . elegans . Tracking and whole brain imaging of naturally behaving zebrafish provide an additional way to study sensorimotor transformation across the brain circuit . A large body of research suggests that sensory information processing depends strongly on the locomotor state of an animal ( Niell and Stryker , 2010; Maimon et al . , 2010; Chiappe et al . , 2010 ) . The ability to sense self-motion , such as proprioceptive feedback ( Pearson , 1995 ) and efferent copy ( Bell , 1981 ) , can also profoundly shape the dynamics of the neural circuit and perception . To explore brain activity in swimming zebrafish , several studies have utilized an elegant tail-free embedding preparation ( Severi et al . , 2014; Portugues and Engert , 2011; Portugues et al . , 2014 ) , in which only the head of the fish is restrained in agarose for functional imaging . Nevertheless , it would be ideal to have physiological access to all neurons in defined behavioral states , where all sensory feedback loops remain intact and functional . Our XLFM-3D tracking system is one step towards this goal , and could be better exploited to explore the neural basis of more sophisticated natural behaviors , such as prey capture and social interaction , where the integration of multiple sensory feedbacks becomes critical . In the XLFM , the camera sensor size limited the number of voxels and hence the number of neurons that could be reliably reconstructed . Our simulation suggested that the sparseness of neuronal activities is critical for optimal imaging volume reconstruction . A growing body of experimental data indeed suggests that population neuronal activities are sparse ( Hromádka et al . , 2008; Buzsáki and Mizuseki , 2014 ) and sparse representation is useful for efficient neural computation ( Olshausen and Field , 1996; Olshausen and Field , 2004 ) . Given the total number of neurons in the larval zebrafish brain , we found that when the fraction of active neurons in a given imaging frame was less than ρc ≈ 0 . 11 , individual neurons could be resolved at optimal resolution . When population neural activity was dense ( e . g . , neurons have high firing rate and firing patterns have large spatiotemporal correlation ) , we obtained a coarse-grained neural activity map with reduced resolution . To retain the fish head within the field of view of the imaging objective , our tracking system compensated for fish movement by continuously adjusting the lateral positions of the motorized stage . As a result , self-motion perceived by the fish was not exactly the same as that during natural behaviors . The linear acceleration of the swimming fish , encoded by vestibular feedback , was significantly underestimated . The perception of angular acceleration during head orientation remained largely intact . The relative flow velocity along the fish body , which was invariant upon stage translation , can still be detected by specific hair cells in the lateral line system ( Coombs , 2014; Liao , 2010 ) . Together , the interpretation of brain activity associated with self-motion must consider motion compensation driven by the tracking system . Both tracking and imaging techniques can be improved in the future . For example , the current axial displacement employed by the piezo scanner had a limited travelling range ( 400 µm ) , and our swimming chamber essentially restrained the movement of the zebrafish in two dimensions . This limitation could be relaxed by employing axial translation with larger travelling range and faster dynamics . Furthermore , to avoid any potential disturbance of animal behaviors , it would be ideal if the imaging system moved , instead of the swimming chamber . In XLFM , the performance degradation caused by focal length variation of the micro-lenses could be resolved by higher precision machining . In addition , the capability of XLFM could be further improved with the aid of technology development in other areas . With more pixels on the imaging sensor , we could resolve more densely labelled samples , and achieve higher spatial resolution without sacrificing imaging volume coverage by introducing more than two different focal planes formed by more groups of micro-lenses . With better imaging objectives that could provide higher numerical aperture and larger field of view at the same time , we could potentially image the entire nervous system of the larval zebrafish with single neuron resolution in all three dimensions . Additionally , the fast imaging speed of XLFM holds the potential for recording electrical activity when high signal-to-noise ratio ( SNR ) fluorescent voltage sensors become available ( St-Pierre et al . , 2014 ) . Finally , the illumination-independent characteristic of XLFM is perfectly suitable for recording brain activities from bioluminescent calcium/voltage indicators in a truly natural environment , where light interference arising from fluorescence excitation can be eliminated ( Naumann et al . , 2010 ) .
The imaging system ( Figure 1 ) was a customized upright microscope . Along the fluorescence excitation light path , a blue laser ( Coherent , OBIS 488 nm , 100 mW , USA ) was expanded and collimated into a beam with a diameter of ~25 mm . It was then focused by an achromatic lens ( focal length: 125 mm ) and reflected by a dichroic mirror ( Semrock , Di02-R488−25×36 , USA ) into the back pupil of the imaging objective ( Olympus , XLPLN25XWMP2 , 25X , NA 1 . 05 , WD 2 mm , Japan ) to result in an illumination area of ~1 . 44 mm in diameter near the objective’s focal plane . In the fluorescence imaging light path , excited fluorescence was collected by the imaging objective and transmitted through the dichroic mirror . A pair of achromatic lenses ( focal lengths: F1 = 180 mm and F2 = 160 mm ) , arranged in 2F1 +2F2 , were placed after the objective and dichroic mirror to conjugate the objective’s back pupil onto a customized lenslet array ( Figure 1—figure supplement 1 ) . The customized lenslet array was an aluminum plate with 27 holes ( 1 . 3 mm diameter aperture on one side and 1 mm diameter aperture on the other side , Source code file 1 ) housing 27 customized micro-lenses ( 1 . 3 mm diameter , focal length: 26 mm ) . The 27 micro-lenses were divided into two groups ( Figure 1—figure supplement 1 ) and an axial displacement of 2 . 5 mm was introduced between them . Apertures of 1 mm diameter on the aluminum plate were placed right at the objective’s pupil plane so that all micro-lenses samples light at pupil plane even though they were displaced axially after apertures . Due to the blockage of light by the aluminum micro-lenses housing , 16% of the light after a 1 . 05 NA imaging objective was effectively collected by the camera . This efficiency is equivalent to using a 0 . 4 NA imaging objective . Finally , the imaging sensor of a sCMOS camera ( Hamamatsu , Orca-Flash 4 . 0 v2 , Japan ) was placed at the middle plane between two focal planes formed by two different groups of micro-lenses . The total magnification of the imaging system was ~4 , so one camera pixel ( 6 . 5 µm ) corresponded to ~1 . 6 µm on the sample . We developed a computational algorithm for 3D volume reconstruction , which required an accurately measured PSF ( Figure 1—figure supplement 2 ) . The PSF was measured by recording images of a 500 nm diameter fluorescent bead sitting on a motorized stage under the objective . A stack of 200 images was recorded when the bead was scanned with a step size of 2 µm in the axial direction from 200 µm below the objective’s focal plane to 200 µm above . Since the images formed by two different groups of micro-lenses were from different axial locations and had different magnifications , the measured raw PSF data were reorganized into two complementary parts: PSF_A and PSF_B ( Figure 1—figure supplements 3 and 4 ) , according to the spatial arrangement of the micro-lenses . We took PSF_A stack , PSF_B stack , and a single frame of a raw image ( 2048 × 2048 pixels ) as inputs , and applied a newly developed algorithm to reconstruct the 3D volume . The reconstruction algorithm was derived from the Richardson-Lucy deconvolution . The goal was to reconstruct a 3D fluorescent object from a 2D image:Obj ( x , y , z ) The algorithm assumes that the real 3D object can be approximated by a discrete number of x-y planes at different z positions:Objx , y , z~Objx , y , zk The numbers and positions of these planes can be arbitrary , yet the Nyquist sampling rate should be chosen to optimize the speed and accuracy of the reconstruction . As the imaging system consisted of two different groups of micro-lenses ( Figure 1—figure supplement 1 ) , their PSFs ( Figure 1—figure supplements 3 and 4 ) each consisted of a stack of planes that were measured at the same chosen axial positions zk:PSFA ( x , y , zk ) Although the PSF was measured in imaging space , here we denote x and y as coordinates in object space to follow conventions in optical microscopy . Here and below , the combination of PSFA and PSFB is the total PSF . Additionally , the images formed by two different groups of micro-lenses had different magnifications , which could be determined experimentally . The ratio between two different magnifications can be defined as:γ=Magnification of group A microlensesMagnification of group B microlenses Then , the captured image on the camera can be estimated as:ImgEst ( x , y ) =∑k=1n{ObjA ( x , y , zk ) ⨂PSFA ( x , y , zk ) +ObjB ( x , y , zk ) ⨂PSFB ( x , y , zk ) } , where ObjA ( x , y , zk ) =ObjB ( γx , γy , zk ) The operator ⨂ represents 2D convolution . Here , x and y on the left hand side of the equation also represent coordinates in object space so that 2D convolution was carried out in the same coordinates . The goal of the algorithm is to estimate the Obj ( x , y , zk ) from the measured camera frame:ImgMeasx , y According to the Richardson-Lucy deconvolution algorithm , the iterative reconstruction can be expressed as:ImgEsti ( x , y ) =∑k=1n{ObjAi−1 ( x , y , zk ) ⨂PSFA ( x , y , zk ) +ObjBi−1 ( x , y , zk ) ⨂PSFB ( x , y , zk ) }ObjAtmp ( x , y , zk ) =ObjAi−1 ( x , y , zk ) {ImgMeas ( x , y ) ImgEsti ( x , y ) ⨂PSFA ( −x , −y , zk ) }ObjBtmp ( x , y , zk ) =ObjBi−1 ( x , y , zk ) {ImgMeas ( x , y ) ImgEsti ( x , y ) ⨂PSFB ( −x , −y , zk ) }ObjAi ( x , y , zk ) =w ( zk ) ObjAtmp ( x , y , zk ) + ( 1−w ( zk ) ) ObjBtmp ( γx , γy , zk ) ObjBi ( x , y , zk ) =w ( zk ) ObjAtmp ( xγ , yγ , zk ) + ( 1−w ( zk ) ) ObjBtmp ( x , y , zk ) Here , 0≤w ( zk ) ≤1 is the weighting factor at different axial positions . The choice of w ( zk ) can be arbitrary . Because the resolutions achieved by different groups of micro-lenses at different z positions were not the same , the weighting factor can take this effect into consideration by weighing higher quality information more than lower quality information . One simple choice is wzk=0 . 5 , that is , to weigh information from two groups of micro-lenses equally . The starting estimate of the object can be any non-zero value . Near the end of the iterations , ObjAix , y , zk and ObjBix , y , zk are interchangeable , except with different magnifications . Either can be used as the resulting estimate of the 3D object . In XLFM , together with its reconstruction algorithm , the diffraction of the 3D light field is properly considered by experimentally measured PSF . The raw imaging data can be fed into the algorithm directly without any preprocessing . Given that the PSF is spatially invariant , which is satisfied apart from small aberrations , the algorithm can handle overlapping fish images ( Figure 1—figure supplement 5 ) . As a result , the field of view can be increased significantly . The reconstruction algorithm was typically terminated after 30 iterations when modifications in the estimated object became very small . The computation can speed up significantly via GPU . It took about 4 min to reconstruct one 3D volume using a desktop computer with a GPU ( Nvidia Titan X ) . In comparison , the reconstruction ran ~20 × slower using a CPU ( Intel E5-2630v2 ) on a Dell desktop . The source code written in MATLAB can be found in the Source code file 2 . The 3D deconvolution method has been developed for conventional LFM ( Broxton et al . , 2013 ) . Our method differs from Broxton et al . ( 2013 ) in several ways . ( 1 ) The optical imaging systems are different . ( 2 ) The definitions of PSFs are different . Ours defines a spatially invariant PSF ( see below for detailed characterization ) , whereas Broxton et al . ( 2013 ) defined a spatially variant PSF , leading to increased computational complexity in the deconvolution algorithm . ( 3 ) The PSF in Broxton et al . ( 2013 ) was simulated based on a model derived from an ideal imaging system , whereas ours was measured experimentally . Furthermore , our system took practical conditions , such as a non-ideal imaging objective , actual positions of microlenses , the spectrum of received fluorescence signal et al . , into consideration . The definition of a 2D spatially invariant PSF fundamentally means that in an ideal optical microscopy system , the resulting image can be described as a 2D convolution between object and PSF . As discussed in the previous section , this operation forms the basis of our reconstruction algorithm . One of the fundamental differences between XLFM and conventional LFM is the location of the microlens array . In XLFM , the microlens array is placed at the pupil plane and the image sensor is at imaging plane , whereas in conventional LFM , the microlens array is placed at the image plane and the image sensor is| at pupil plane . It is possible to define a spatially invariant PSF in XLFM because: By definition , the imaging formation in an ideal optical imaging system is linear and spatially invariant , so spatially invariant PSFs for sub-imaging systems consisting of micro-lens A1 and A2 can be defined as:ImageA1=Object⨂PSFA1ImageA2=Object⨂PSFA2 where ImageA1/2 are sub-images behind individual micro-lens . If we perform the convolution in the imaging space , the coordinates of Object ( x , y ) should be scaled by the magnification factors of their sub-image systems , respectively . Now if the magnifications of different sub-image systems are the same , the summation of all PSFs formed by individual micro-lenses can be defined as a single PSF . In other words , ImageA=ImageA1+ImageA2=Object⨂ ( PSFA1+PSFA2 ) =Object⨂PSFAwhere PSFA=PSFA1+PSFA2 Experimentally , the small variation of individual micro-lenses’ focal length ( Figure 1—figure supplement 8 ) resulted in spatial variance of PSFA or PSFB , but it does not affect the imaging formation theory of XLFM . The spatial variance led to degraded reconstruction performance , as shown in Figure 1—figure supplement 9 . This degradation was negligible near the center of the field of view , but became more evident near the edge of the field of view . This is because the PSF was measured near the center of the field of view . The reconstruction algorithm produces 27 estimates of the same object based on 27 sub-images . In the meanwhile , it tries to combine and align these estimates all together in the same coordinates . The position where the PSF is measured determines the origin of this coordinates . If the magnifications of different micro-lenses are different , the reconstruction will yield an image that is clear near the origin of the coordinates but blurred at the edge , as shown in in Figure 1—figure supplement 9 . Unlike conventional microscopy , where the performance of the imaging system is fully characterized by the PSF at the focal plane , the capability of XLFM is better characterized as a function of positions throughout the imaging volume . We first characterized the spatial resolution in the x-y plane by analyzing the spatial frequency support of the experimentally measured PSF from individual micro-lenses using a 0 . 5 µm diameter fluorescent bead . The optical transfer function ( OTF ) , which is the Fourier transform of the PSF in the x-y plane , was extended to a spatial frequency of ~1/3 . 4 µm−1 ( Figure 1—figure supplement 6 ) , a result that agreed well with the designed resolution at 3 . 4 μm , given that the equivalent NA of individual micro-lenses was 0 . 075 . The lateral resolution , measured from the raw PSF behind individual micro-lenses , was preserved across the designed cylindrical imaging volume of Ø800 μm × 200 μm ( Figure 1—figure supplement 6 ) . However , the reconstruction results ( Figure 1—figure supplement 9 ) , which used total PSF ( Figure 1—figure supplement 2 ) , exhibited resolution degradation when the fluorescent bead was placed more than 250 μm away from the center ( Figure 1—figure supplement 9 ) . This discrepancy resulted from the variation in focal length of the micro-lenses ( Figure 1—figure supplement 8 ) , which , in turn , led to spatial variance of the defined PSFA and PSFB . In principle , the designed lateral resolution of 3 . 4 µm could be preserved over a volume of Ø800 μm × 200 μm by reducing focal length variation to below 0 . 3% We next characterized the axial resolution of the XLFM . The XLFM gained axial resolution by viewing the object from large projection angles achieved by micro-lenses sitting near the edge of the objective’s back pupil plane . For example , if two points of light source were located at the same position in the x-y plane , but were separated by ∆z in the axial direction , then one micro-lens in the XLFM could capture an image of these two points with a shift between them . The shift can be determined as:d=∆z*tanθ where θ is the inclination angle inferred from the measured PSF ( Figure 1—figure supplement 2 ) . If the two points in the image can be resolved , the two points separated by ∆z can be resolved by the imaging system . Since a micro-lens sitting in the outer layer of the array offered the largest inclination angle of 40 degree in our system , the axial resolution dz can be directly calculated as:dz=dxytanθmax=3 . 4 μmtan ( 40° ) =4 μm The best way to confirm the theoretical estimate is to image two fluorescent beads with precisely controlled axial separations . However , this is technically very challenging . Instead , we pursued an alternative method that is equivalent to imaging two beads simultaneously: The above method allowed us to experimentally characterize the axial resolution afforded by individual micro-lenses focusing at different z positions . We used a single fluorescent bead ( 0 . 5 μm in diameter ) with a high SNR ( Figure 1—figure supplement 7a ) . We imaged at different axial positions: z = −100 μm , z = 0 μm , and z = 100 μm ( Figure 1—figure supplement 7b ) . The third column is the combined images in column 1 and 2 . The capability of resolving the two beads in the third column can be demonstrated by spatial frequency analysis ( fourth column in Figure 1—figure supplement 7b ) . The two line dips , indicating the existence of two beads instead of one rod in the fourth column , were confirmations of the resolving capability . This becomes more evident after deconvolution of the raw images ( fifth column in Figure 1—figure supplement 7b ) . Micro-lenses 1 and 2 could resolve two beads , separated by 5 μm , within the range of -100 μm≤z ≤ 0 and 0≤z≤100 μm , respectively . In other words , the complementary information provided by the two micro-lenses allowed the system to maintain a high axial resolution at 5 μm across a 200 μm depth . Next , we imaged densely packed fluorescent beads ( 0 . 5 μm in diameter ) with a low SNR ( Figure 1—figure supplement 10a ) , and used our reconstruction algorithm to determine the minimum axial separation between beads that could be resolved ( Figure 1—figure supplement 10b–c ) . In this case , 5 μm axial resolution could be preserved across a depth of 100 μm . The resolution decayed gradually to ~10 μm at the edge of an imaging volume with a 400 μm axial coverage ( Figure 1—figure supplement 10b ) . We believe that the optimal axial resolution at 5 µm could be achieved over an axial coverage of 200 μm by minimizing micro-lens focal length variation ( Figure 1—figure supplement 8 ) . Finally , we characterized how the imaging performance depended upon the sparseness of the sample . Given the total number of neurons ( ~80 , 000 ) in a larval zebrafish brain , we introduced a sparseness index ρ , defined as the fraction of neurons in the brain active at an imaging frame , and used numerical simulation to characterize the dependence of achievable resolution on ρ . To this end , we simulated a zebrafish larva with uniformly distributed firing neurons ( red dots in Figure 1—figure supplement 11a ) . By convolving the simulated zebrafish with the experimentally measured PSFs ( Figure 1—figure supplements 3 and 4 ) , we generated an image that mimicked the raw data captured by the camera . We then reconstructed the simulated neurons from this image , represented by green dots . When ρ was equal to or less than 0 . 11 , which corresponded to ~9000 neurons activated at a given instant , all active neurons , including those closely clustered , could be reconstructed with optimal resolution ( Figure 1—figure supplement 11b inset ) . As the sparseness index ρ increased , the resolution degraded: nearby neurons merged laterally and elongated axially ( Figure 1—figure supplement 11c–d ) . In all calculations , the Poisson noise was properly considered by assuming that each active neuron emitted 20 , 000 photons , 2 . 2% of which were collected by our imaging system . In vivo resolution characterization is challenging due to a lack of bright and spot-like features in living animals . Additionally , achievable resolution depends on the optical properties of biological tissues , which can be highly heterogeneous and difficult to infer . The light scattering and aberration induced by biological tissue usually leads to degraded imaging performance ( Ji , 2017; Ji et al . , 2010; Wang et al . , 2014; Wang et al . , 2015 ) . To compensate for lateral fish movement and retain the entire fish head within the field of view of a high NA objective ( 25× , NA = 1 . 05 ) , a high-speed camera was used to capture fish motion ( 2 ms exposure time , 300 fps or higher , Basler aca2000-340kmNIR , Germany ) . We developed an FPGA-based RT system in LabVIEW that could rapidly identify the head position by processing the pixel stream data within the Cameralink card before the whole image was transferred to RAM . The error signal between the actual head position and the set point was then fed into the PID to generate output signals and control the movement of a high-speed motorized stage ( PI M687 ultrasonic linear motor stage , Germany ) . In the case of large background noise , we alternatively performed conventional imaging processing in C/C++ ( within 1 ms delay ) . The rate-limiting factor of our lateral tracking system was the response time of the stage ( ~300 Hz ) . We applied the principle of LFM to determine the axial movement of larval zebrafish . The autofocus camera ( 100 fps or higher , Basler aca2000-340kmNIR , Germany ) behind a one-dimensional micro-lens array captured triplet images of the fish from different perspectives ( Figure 2—figure supplement 1a ) . Z motion caused an extension or contraction between the centroids of the fish head in the left and right sub-images , an inter-fish distance ( Figure 2—figure supplement 1b ) that can be accurately computed from image autocorrelation . The inter-fish distance , multiplied by a pre-factor , can be used to estimate the z position of the fish , as it varies linearly with axial movement ( Figure 2—figure supplement 1c ) . The error signal between the actual axial position of the fish head and the set point was then fed into the PID to generate an output signal to drive a piezo-coupled fish container . The feedback control system was written in LabVIEW . The code was further accelerated by parallel processing and the closed loop delay was ~5 ms . The rate-limiting factor of the autofocus system was the settling time of the piezo scanner ( PI P725KHDS , Germany , 400 μm travelling distance ) , which was about 10 ms . Two high-speed cameras acquired dark-field images at high and low magnification , respectively , and customized machine vision software written in C/C ++ with the aid of OpenCV library was used to perform real-time behavioral analysis of freely swimming larval zebrafish . At high magnification , eye positions , their orientation , and convergence angle were computed; at low magnification , the contour of the whole fish , centerline , body curvature , and bending angle of the tail were computed . The high mag RT analysis was run at ~120 fps and the low mag RT analysis was run at ~180 fps . The source code can be found in the Source code file 3 . All animal handling and care were conducted in strict accordance with the guidelines and regulations set forth by the Institute of Neuroscience , Chinese Academy of Sciences , University of Science and Technology of China ( USTC ) Animal Resources Center , and University Animal Care and Use Committee . The protocol was approved by the Committee on the Ethics of Animal Experiments of the USTC ( permit number: USTCACUC1103013 ) . All larval zebrafish ( huc:h2b-gcamp6f and huc:gcamp6s ) were raised in embryo medium under 28 . 5°C and a 14/10 hr light/dark cycle . Zebrafish were fed with paramecium from 4 dpf . For restrained experiments , 4–6 dpf zebrafish were embedded in 1% low melting point agarose . For freely moving experiments , 7–11 dpf zebrafish with 10% Hank’s solution were transferred to a customized chamber ( 20 mm in diameter , 0 . 8 mm in depth ) , and 10–20 paramecia were added before the chamber was covered by a coverslip . To extract neural activity induced by visual stimuli ( Figure 1e and f ) , time series 3D volume stacks were first converted to a single 3D volume stack , in which each voxel represented variance of voxel values over time . Candidate neurons were next extracted by identifying local maxima in the converted 3D volume stack . The region-of-interest ( ROI ) was set according to the empirical size of a neuron . The voxels around the local maxima were selected to represent neurons . The fluorescence intensity over each neuron’s ROI was integrated and extracted as neural activity . Relative fluorescent changes ∆F/F0 were normalized to their maximum calcium response ∆Fmax/F0 over time , and sorted according to their onset time when ∆F first reached 20% of its ∆Fmax ( Figure 1e and f ) after the visual stimulus was presented . A short wavelength LED was optically filtered ( short-pass optical filter with cut-off wavelength at 450 nm , Edmund #84–704 ) to avoid light interference with fluorescence . It was then focused by a lens into a spot 2 ~ 3 mm in diameter . The zebrafish was illuminated from its side . The total power of the beam was roughly 3 mW . Each experiment was repeated at least three times with similar experimental conditions . Imaging and video data acquired from behaviorally active larval zebrafish with normal huc:h2b-gcamp6f or huc:gcamp6s expression were used in the main figures and videos . | How do neurons in the brain process information from the senses and drive complex behaviors ? This question has fascinated neuroscientists for many years . It is currently not possible to record the electrical activities of all of the 100 billion neurons in a human brain . Yet , in the last decade , it has become possible to genetically engineer some neurons in animals to produce fluorescence reporters that change their brightness in response to brain activity and then monitor them under a microscope . In small animals such as zebrafish larvae , this method makes it possible to monitor the activities of all the neurons in the brain if the animal’s head is held still . However , many behaviors – for example , catching prey – require movement , and no existing technique could image brain activity in enough detail if the animal’s head was moving . Cong , Wang , Chai , Hang et al . have now made progress towards this goal by developing a new technique to image neural activity across the whole brain of a zebrafish larva as it swims freely in a small water-filled chamber . The technique uses high-speed cameras and computer software to track the movements of the fish in three dimensions , and then automatically moves the chamber under the microscope such that the animal’s brain is constantly kept in focus . The newly developed microscope can capture changes in neural activity across a large volume all at the same time . It is then further adapted to overcome problems caused by sudden or swift movements , which would normally result in motion blur . With this microscope set up , Cong et al . were able to capture , for the first time , activity from all the neurons in a zebrafish larva’s brain as it pursued and caught its prey . This technique provides a new window into how brain activity changes when animals are behaving naturally . In the future , this technique could help link the activities of neurons to different behaviors in several popular model organisms including fish , worms and fruit flies . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"tools",
"and",
"resources",
"neuroscience"
] | 2017 | Rapid whole brain imaging of neural activity in freely behaving larval zebrafish (Danio rerio) |
Tumor initiation is often linked to a loss of cellular identity . Transcriptional programs determining cellular identity are preserved by epigenetically-acting chromatin factors . Although such regulators are among the most frequently mutated genes in cancer , it is not well understood how an abnormal epigenetic condition contributes to tumor onset . In this work , we investigated the gene signature of tumors caused by disruption of the Drosophila epigenetic regulator , polyhomeotic ( ph ) . In larval tissue ph mutant cells show a shift towards an embryonic-like signature . Using loss- and gain-of-function experiments we uncovered the embryonic transcription factor knirps ( kni ) as a new oncogene . The oncogenic potential of kni lies in its ability to activate JAK/STAT signaling and block differentiation . Conversely , tumor growth in ph mutant cells can be substantially reduced by overexpressing a differentiation factor . This demonstrates that epigenetically derailed tumor conditions can be reversed when targeting key players in the transcriptional network .
During development , epigenetic regulators are responsible for controlling and restraining cellular plasticity . This tight regulation allows cells to differentiate faithfully and heritably towards a specific fate ( Roy and Hebrok , 2015; Wainwright and Scaffidi , 2017 ) . An appropriate balance between proliferation and differentiation is fundamental and multiple regulatory layers of transcription factors and epigenetic regulators are employed to accomplish the underlying transcriptional control ( Gonda and Ramsay , 2015; Piunti and Shilatifard , 2016 ) . Many epigenetic regulators are evolutionary conserved and among the most commonly mutated genes in human cancer ( Piunti and Shilatifard , 2016 ) . Disruption of epigenetic constraints leads to global reorganization of the epigenome and changes in transcriptional profiles , which might provide a cellular state permissive for tumorigenesis ( Wainwright and Scaffidi , 2017 ) . Indeed , disturbed transcriptional profiles and putative oncogenic transcriptional regulators have recently gained significance as better alternatives for therapeutic targets in comparison to signaling pathways . Transcription factors ( TFs ) are less prone to be bypassed by alternative mutational events , and their perturbation can affect several cancer hallmarks . In addition , due to the complexity of transcriptional networks , it is unlikely that one TF functioning as an oncogenic driver can be entirely replaced by another ( Gonda and Ramsay , 2015 ) . For these reasons it is fundamental to identify the core transcriptional networks defining cancer cell types , and target those regulators crucial for survival ( Bonifer and Cockerill , 2015 ) . Epigenetic regulators involved in preserving cellular identity are composed of two classes of chromatin proteins , the Polycomb ( PcG ) and the Trithorax group ( TrxG ) , whose complementary functions maintain the repressed and active gene expression state , respectively ( Geisler and Paro , 2015 ) . PcG proteins are organized into two basic complexes , Polycomb repressive complex 1 and 2 ( PRC1 and PRC2 ) ( Piunti and Shilatifard , 2016 ) . One example of classical PcG targets are homeotic genes encoding Homeobox TFs , first identified in Drosophila and responsible for correct spatial body development in flies ( Shah and Sukumar , 2010; Abate-Shen , 2002 ) . The altered expression of Hox genes in human tumors suggests important roles for both oncogenesis and tumor suppression ( Shah and Sukumar , 2010 ) , which further hints towards an role for PcG proteins in oncogenesis . The tumor suppressive role of PcG proteins , in particular PRC1 members in Drosophila imaginal discs , has been extensively investigated ( Classen et al . , 2009; Martinez et al . , 2009 ) . However , the effects on the transcriptional landscape after PRC1 deregulation in tumorigenesis has only recently started to be assessed ( Bunker et al . , 2015; Loubière et al . , 2016 ) . Here , we show that loss of Polyhomeotic ( ph ) , a member of the PRC1 complex , in eye-antennal imaginal discs of larvae leads to a reprogramming of cellular identity towards an embryonic state and a concomitant loss of differentiation markers . Among the reactivated genes is knirps , an orphan nuclear hormone receptor . Depletion of knirps revealed its vital role in tumor maintenance , while misexpression showed its capacity to drive tumorigenesis in otherwise wild-type tissues . Tumors initiated by ph disruption or misexpression of knirps share features such as ectopic activation of JAK/STAT signaling and a differentiation block . We conclude that the embryonic TF knirps is an oncogene in eye-antennal imaginal discs and is crucial for the tumorigenic capacity of the epigenetic tumor under study . Additionally , we found that overexpressing a pro-neural TF leads to reduction of proliferation and suppression of the tumor phenotype .
To better understand the molecular consequences of polyhomeotic loss-of-function in vivo , we analyzed the transcriptome of ph505-tumor cells by RNA sequencing ( RNA-seq ) . We used fluorescence activated cell sorting ( FACS ) of dissociated eye-antennal imaginal discs to separate GFP-labeled tumor cells from surrounding unlabeled and non-mutant cells ( Martinez et al . , 2009; Dutta et al . , 2013; Harzer et al . , 2013 ) . This proved to be an essential step for an accurate diversification of the tumor transcriptome . Transcriptome analysis showed substantial deregulation of gene expression in mutant cells compared to neighboring non-mutant cells . We identified 1337 differentially expressed genes ( Benjamini adjusted p-value , padj . <0 . 01 ) , with 275 genes being upregulated in ph505-tumor cells ( Figure 1A , Figure 1—figure supplement 2A , Figure 1—source data 1 ) . Furthermore , gene set enrichment analysis revealed that neurogenesis-related genes were mainly downregulated in ph505-tumor cells , while genes regulating transcription were upregulated ( Figure 1B ) . Since PcG target genes encode crucial developmental regulators , such as TFs ( Simon and Kingston , 2009 ) , our expression data corroborates its impaired function . Moreover , we observed deregulation of genes involved in tissue development ( e . g . , GO-terms for genital disc development , imaginal disc development ) and enrichment for TFs among the upregulated genes ( GO-term sequence-specific DNA binding transcription factor activity ) ( Figure 1B , Figure 1—source data 2 ) . The observed global modulation of transcription output is in agreement with PcG proteins constituting a global regulatory system ( Simon and Kingston , 2009 ) . TFs are essential to define cell types and are among the main targets of PcG silencing ( Simon and Kingston , 2009 ) . As such , we decided to focus on the fraction of differentially expressed TF-encoding genes , due also to the enrichment of genes in this category in our RNA-seq dataset . We evaluated which upregulated TFs , and thus the primary response to the ph knock-out , could be contributing to the overall deregulation of gene expression observed . Employing the iRegulon tool ( Janky et al . , 2014 ) , predictions based on motif enrichment revealed caudal , grain and knirps ( direct Ph targets in eye discs [Loubière et al . , 2016] ) as TFs putatively responsible for all differentially expressed genes in our RNA-seq dataset . To further reveal the transcriptional identity of ph505-tumor cells , we integrated available datasets from the Gene Expression Omnibus ( Edgar et al . , 2002 ) and compared the gene expression signature of ph505-tumors with other tissues and/or developmental stages of Drosophila . In total , 83 samples including different tissues and cell types ( namely ovaries , larval brain , adult head , wing disc , eye-antennal disc , larval neurons and larval neuroblasts ) and developmental stages ( embryo , larva and pupa ) were considered for the analysis ( Figure 1—source data 3 ) . Specifically , we compared 124 differentially expressed genes involved in transcriptional regulation ( GO0006355 ) ( Figure 1—source data 4 ) . Strikingly , hierarchical clustering of 83 transcriptome samples showed that ph505-tumor cells clustered close to samples from early embryonic developmental stages ( Figure 1C and Figure 1—figure supplement 2B ) . As expected , our RNA-seq control samples ( neighboring unlabeled cells ) clustered with wild-type eye-antennal imaginal disc transcriptomes . This result might reflect the re-establishment of an earlier developmental program in ph505-tumors as a consequence of a reprogrammed epigenomic state . Additionally , this is not a general feature shared by all tumors , as depicted by other tumor types ( i . e . brat [Jüschke et al . , 2013] and RasV12/scrib- tumors [Atkins et al . , 2016] ) not clustering with embryos . We hypothesize that the clustering of ph505-tumor cells with early embryos was not only the result of the ectopic expression of embryonic TFs in the ph505-tumor cells , but also due to reduced expression of the TFs , which characterize differentiated tissues . Downregulation of neurogenesis-related genes suggests that these tumor cells may be unable to differentiate , losing cell fate markers and their normally established identity . We confirmed the downregulation of neurogenesis-related markers at the protein level for ELAV ( Embryonic Lethal Abnormal Vision ) , which is normally expressed in the differentiated neuronal cells that make up the eye imaginal disc ( Figure 1—figure supplement 3A–B ) and for Eya ( Eyes absent , Figure 1—figure supplement 3C–D ) . This supports previous findings that suggested that neoplastic Drosophila epithelial cells reverse their developmental commitments and switch to primitive cell states ( Khan et al . , 2013 ) . In this particular report , the switch in the eye primordium was shown to be Homothorax ( Hth ) -dependent ( Khan et al . , 2013 ) . Conversely , in our RNA-seq dataset hth was downregulated and at the protein level we confirmed that Hth is not ectopically expressed in the ph505 clones ( Figure 1—figure supplement 2E–F ) . Thus , our study reveals that ph505-tumors do not depend upon ectopic expression of Hth to keep cells in a non-differentiated state and support neoplastic growth . The similarity of the ph505-tumor TF signature with Drosophila early embryos was reinforced by confirmation of the presence of embryonic-TF misexpression across tumor-tissue samples . We performed immunostaining for additional embryonic TFs , namely Even-skipped , Abdominal-B and Caudal , and observed ectopic expression of these proteins specifically in mutant clones ( Figure 1D–E and Figure 1—figure supplement 3A–D ) . Overall , these results suggest that ph505-tumor cells previously committed to a neurogenesis-related path switch their cell fate as they fail to differentiate during the process of tumorigenesis due to the modulation of the transcriptional regulatory program of the cells . In order to pinpoint key regulators of tumorigenesis in ph505-tumors , we performed an in vivo screen for a subset of selected candidates . Among all the TFs upregulated in ph505-tumor cells , we chose , based on literature search , 24 to assess their importance in promoting tumorigenic potential in these cells . Our approach to test the ability of candidate genes playing a key role in ph505-tumorigenesis was to combine generation of ph505-tumor clones with RNAi-mediated knock-down ( KD ) of a target of interest within these clones . We compared effects of RNAi to the baseline ph505 neoplastic phenotype and observed that some RNAi lines targeting TFs ( in ph505 clones ) resulted in a strong increase in viability ( close to 90–100% , for example cad , drm , kni , bgcn ) , while others did not change or only slightly changed pupal viability ( Figure 2A ) . We further characterized 6 RNAi lines: crocodile ( croc ) , lateral muscles scarcer ( lms ) , caudal ( cad ) , drumstick ( drm ) , knirps ( kni ) and benign gonial cell neoplasm ( bgcn ) ( Figure 2B–H ) , which showed significant differences in eclosion rate in comparison to flies carrying ph505 clones ( Figure 2B and Figure 2—figure supplement 1A ) . The eclosion rate for three of these perturbations ( drm- , kni- and bgcn-KD ) reached similar levels as control flies ( carrying FRT19A neutral clones ) and rescue experiment ( ph505 , UAS-ph ) . By quantifying tumor volumes relative to tissue size of the six above-mentioned perturbations , we observed that only cad- , drm- , kni- and bgcn-KD showed a significant difference compared to the baseline of 46% tumor volume in the ph505 condition ( 14 , 13 , 5 and 14% tumor volume , respectively ) . In addition , the two perturbations ( croc- and lms-KD ) that did not show a significant effect on tumor volume were also those with less remarkable differences in eclosion rate ( Figure 2B and Figure 2—figure supplement 1B–D ) . These results suggest that higher eclosion rate is a good approximation for decreased tumor volume . Furthermore , the tissue volume of the eye-antennal imaginal disc in the drm- , kni- and bgcn-KD conditions was closer to the control tissue volume than the ph505 condition ( Figure 2—figure supplement 1C ) . From all the RNAi conditions tested , kni-KD in ph505 mutant clones showed the most striking decrease in tumor volume ( 9 . 2 fold decrease ) , similar to the rescue experiment ( ph505 , UAS-ph ) ( Figure 2B ) . Additionally , the phenotype of adult eyes of this genotype suggests a recovery of the differentiation program ( Figure 2—figure supplement 2A–D ) . This is supported by immunostaining against ELAV showing that it is no longer disrupted when expression of kni is blocked in ph505 ( Figure 2—figure supplement 2E ) . Altogether , this shows that the differentiation block observed in ph505-tumors is prevented upon reducing the level of knirps expression by RNAi KD . We confirmed that a second , independent RNAi line against kni also led to a significant decrease of tumor volume ( 14% of tumor volume vs . 46% in ph505 ) and an increase in eclosion rate ( 85% ) ( Figure 2—figure supplement 2F–J ) . We can thus minimize the chance that the effects observed using either kni-RNAi were due to off-target effects . We characterized the tumorigenic potential of ph505-tumors and ph505 , kni-KD by conducting transplantation assays ( Rossi and Gonzalez , 2015 ) of these tissues into the abdomen of adult host flies ( Figure 3A–D ) . In the case of the ph505-tumors the percentage of tumor-bearing hosts increased from 40% to 60% , from the first week to subsequent weeks after transplantation indicating hyperproliferation of the transplanted tissues ( Figure 3A ) . The tumorigenic potential of ph505-transplanted tissue was already detected on day seven after transplantation ( Figure 3B ) . By contrast , when transplanting ph505 , kni-KD clones we did not observe any tumors in the host flies within the first three weeks . Even after up to 5 weeks post-transplantation , we could only find a single fly with GFP+ tissue overgrowth ( Figure 3C–D ) . Our data demonstrate that the TF Knirps plays a crucial role in tumorigenesis of ph505-tumors given that kni-KD in these tissues not only led to a reduction of tumor volume but also the remaining clones were not able to proliferate in the host fly abdomen . Evasion of apoptosis is one of the hallmarks of cancer ( Hanahan and Weinberg , 2000 ) . As we observed a significant reduction of tumor volume upon depleting kni in ph505 mutant cells , we hypothesized that kni-KD could trigger cell death of tumor cells . We blocked apoptosis within mutant clones ( via expression of anti-apoptotic protein p35 [Hay et al . , 1994] ) . Levels of apoptosis as assessed by immunostaining against Death caspase-1 ( Dcp-1 ) confirmed a decrease in apoptosis in tissues where p35 was expressed in ph505 , kni-KD clones ( Figure 3E–F and Figure 3—figure supplement 1A–B ) . However , we observed that blocking apoptosis in ph505 , kni-KD clones was not sufficient to revert the anti-oncogenic effects of kni-KD ( Figure 3G–H and Figure 3—figure supplement 1C–F ) . Furthermore , the tumor volume of ph505 , kni-KD , UAS-p35 was similar to ph505 , kni-KD ( Figure 3G and Figure 3—figure supplement 1D–F ) . We also tested the effect of this particular RNAi line in the context of neutral clones generated with the same driver as for ph505-tumors . These FRT19A , kni-KD flies showed neither difference in eclosion rate nor in the adult eye phenotype , compared to control flies ( Figure 3E–F and Figure 3—figure supplement 1C ) . This suggests that the RNAi line targeting kni does not per se affect eye-antennal imaginal disc development . Ectopic expression of cell fate-specifying TFs was recently shown to lead to the formation of epithelial cysts ( Bielmeier et al . , 2016 ) . Cyst formation in wing and eye imaginal discs represents a response to cell fate mis-specification , compromising tissue integrity and potentially promoting precancerous lesions . We thus assessed the effect of ectopic kni expression in eye-antennal imaginal discs by generating mitotic clones using again the eyFlp system . We observed that FRT19A clones expressing kni ( Figure 4A ) displayed a more pronounced round shape in comparison to the notchy-shape of FRT19A neutral clones ( Figure 1—figure supplement 1A ) . Additionally , ectopic kni expression compromised the viability of the flies , evidenced by an eclosion rate of only 35% ( Figure 4B ) , and the defective development of the adult eye structures ( Figure 4C ) . Furthermore , we confirmed that ectopic expression of kni leads to the formation of cysts ( Figure 4D ) and thus interferes with epithelial polarity ( Figure 4D and Figure 5—figure supplement 1A ) . To test if ectopic expression of kni alone is sufficient to drive tumorigenesis we conducted transplantations of eye-antennal imaginal disc tissues ectopically expressing kni . We observed that knirps is sufficient to generate tumors in the host flies , visible 3 weeks after transplantation ( ranging from 15–50% , from week 3 to 5 after transplantation respectively ) ( Figure 4E–F ) . Our data suggest that ectopic expression of knirps interferes with the normal course of development and that knirps is a new oncogene , possibly acting in a context/tissue-dependent manner . Since knirps-KD alone was sufficient to reduce the tumorigenic potential of ph505-tumors , we hypothesized that some features of clones ectopically expressing knirps in a wild-type context could resemble ph505-tumor clones . We therefore evaluated the activation of signaling pathways in this context . We observed that of the JNK , JAK/STAT and Notch signaling pathways ( all are activated in ph505-tumors [Classen et al . , 2009; Martinez et al . , 2009; Beira et al . , 2018] ) , only JAK/STAT was ectopically activated , particularly in knirps cyst-like clones ( Figure 5A–C ) . This observation suggests that ectopic expression of knirps alone is sufficient to activate the JAK/STAT pathway in mitotic clones . Hence , ectopic activation of the other signaling pathways in ph505-tumors is likely attributable to other factors regulated by Ph and independent of knirps . In light of the compromised eye development seen in kni-ectopic flies , and suggestions that the JAK/STAT pathway needs to be switched off to allow differentiation ( Amoyel and Bach , 2012 ) , we investigated the expression of a number of neurogenesis-related markers in kni-ectopic eye-antennal imaginal discs . Similarly to what we observed with ph505 clones , ELAV expression was disrupted in kni-expressing cyst-like structures , as shown in Figure 5D , as well as Eya ( Figure 5E ) , without ectopic activation of Hth ( Figure 5F ) . These observations are thus in agreement with the hypothesis that knirps alone is sufficient to initiate tumorigenesis . Our data argue in favor of a role for JAK/STAT in contributing to the differentiation block in ph505 and kni-ectopic tumors . We decided to block this pathway in ph505-tumors ( ph505 , dome∆CYT ) and examine cellular differentiation in these eye-antennal imaginal discs . Upon blocking JAK/STAT in ph505-tumors , we observed that ELAV expression is re-established almost to a normal situation , even in the presence of clones ( Figure 5—figure supplement 1B in comparison to Figure 1—figure supplement 3B ) . Moreover , the viability of these flies is increased , close to normal levels ( Figure 5—figure supplement 1C , eclosion rate 85% ) and some adult flies presented eye structures similar to wt individuals Figure 5—figure supplement 1D ) . Blockage of normal differentiation appears to be a common feature between ph505 and kni-ectopic tumors in eye tissues , suggesting that kni expression in the ph505-tumors contributes to the differentiation defects observed ( Figure 1—figure supplement 3 and Figure 2—figure supplement 2C ) . Hence , we expected that apart from the knock-down of an embryonic TF with tumorigenic capacity , forcing differentiation of tumor cells could restrain the tumorigenic phenotype . Atonal ( ato ) , encoding a pro-neural TF , was previously shown to have an anti-oncogenic role in the fly retina , where it instructs tissue differentiation ( Bossuyt et al . , 2009 ) . Notably , ato is also among our downregulated set of genes ( padj . <0 . 01 ) . We ectopically expressed ato in ph505 clones , which led to the rescue of the phenotype by a reduction of the tumor volume from 46% baseline to 3% and an increase in the eclosion rate from 12% to 84% ( Figure 6A–C and Figure 6—figure supplement 1A–E ) . Hence , expression of ato in ph505 clones was sufficient to restore the normal pattern of differentiation of this tissue , as confirmed by the expression of ELAV ( Figure 6D and Figure 6—figure supplement 1G ) . Indeed , also the eye phenotype of the hatched flies resembled the phenotype of wild-type flies ( Figure 6—figure supplement 1F ) . We then asked whether these effects can be attributed to the capacity of atonal in preventing proliferation , as previously shown in a different tumor model ( Bossuyt et al . , 2009 ) . To test this hypothesis , we assessed levels of phospho-histone H3 ( pH3 ) as a measure of proliferation ( Figure 6E–G and Figure 6—figure supplement 2 ) . Quantitative analysis showed an overall increase in proliferation levels in ph505-tumor tissues in comparison to control tissues . This was largely due to an increase of proliferative cells outside of ph505 clones ( Figure 6—figure supplement 2C ) . The analysis also showed a decrease in pH3+ cells inside clones co-expressing ph505 and atonal ( in comparison to ph505 clones ) ( Figure 6G ) . Thus , atonal antagonizes ph tumor growth by counterbalancing proliferation , ultimately leading to a reduction of tumor burden and to a normal eye differentiation pattern .
With the analysis of a ph mutant transcriptome we highlight the complexity of disrupting global gene expression programs and , with that , newly established transcriptional dependencies . Previous approaches of generating PcG-negative transcriptomes investigated gene expression of mutant cells that were deprived from contact with non-mutant cells ( using the cell lethal system ) , which was then compared with wild-type discs composed of neutral clones ( Loubière et al . , 2016; Bunker et al . , 2015 ) . In contrast , we set out to compare ph505 mutant cells with their surrounding wild-type cells to gain potential additional information by taking into account non-autonomous growth effects previously reported ( Feng et al . , 2011 ) . The RNA-seq dataset presented here reveals enrichment for TFs in the upregulated gene set . It also indicates that tumor cells fail to differentiate , supported by the downregulation of neural-cell fate markers characteristic of this tissue and by the upregulation of embryonic TFs . This is also highlighted by the clustering of the TF-signature of ph505-tumors with embryonic stages of Drosophila development . Moreover , we also found several Hox genes in our set of upregulated genes ( e . g . , Antp , Ubx , Abd-A , Abd-B ) , which are classical embryonic PcG-targets shown to be important in oncogenesis . Although not regarded as a traditional hallmark of cancer ( Hanahan and Weinberg , 2000 ) , a key event in tumorigenesis is the perturbation of normal cell fate ( Gonda and Ramsay , 2015 ) . Re-expression of particular embryonic genes in an aberrant spatial-temporal pattern could contribute to oncogenesis by maintenance of a more embryonic state through the activation of anti-apoptotic pathways or suppression of differentiation ( Shah and Sukumar , 2010 ) . For example , re-establishment of an earlier developmental program has been proposed in human pediatric gliomas that frequently have mutations in histone H3 lysine 27 ( H3K27M ) and compromised PRC2 function ( Funato et al . , 2014; Wainwright and Scaffidi , 2017 ) . Since several classic TFs with important functions during embryogenesis are among the upregulated genes in the ph505-tumor transcriptome , we subsequently blocked their expression and showed for some TFs their potential to rescue the ph knock-out phenotype and reduce tumor growth . Quantitative measurements of tumor volume in various conditions ensured the reproducibility of the data , excluding an observer bias . The observed effects of TF-KD on eclosion rate and tumor volume did not necessarily correlate with the genes that are direct targets of Ph silencing in eye discs . This is illustrated for example by the strong effects of bgcn-KD that has not been identified as a direct Ph target ( Loubière et al . , 2016 ) . These observations on TFs are particularly important since transcription has a direct influence on the balance between proliferation and differentiation . Furthermore , when transcriptional regulators ( TFs , co-regulators or epigenetic modifiers ) are misregulated , differentiation is blocked and pre-cancer cells can proliferate ( Gonda and Ramsay , 2015 ) . kni is a gap gene involved in the subdivision of the embryo anterior-posterior axis that can function as an activator ( Langeland et al . , 1994 ) or a repressor ( Pankratz et al . , 1990 ) . Besides its classic function in embryonic development , kni is subsequently also required for vein formation in wing imaginal discs ( Lunde et al . , 2003 ) . We show that KD of knirps in ph505-tumors is sufficient to reduce tumor volume by 90% . It also reduces the tumorigenic capacity of ph505-tumors , as assessed by a transplantation assay . Misexpression of TFs in imaginal discs and formation of cysts has been suggested to be an indicator of precancerous lesions ( Bielmeier et al . , 2016 ) . Here we show that ectopic expression of knirps in eye-antennal imaginal discs leads to the formation of cysts and is sufficient to recapitulate the phenotypic tumor appearance . Moreover , we believe that this TF , with its dual regulatory role , could activate or repress other genes and thus form a regulatory circuit that is beneficial for tumor initiation and progression . Inducing a cell fate switch can be achieved by forcing expression of a TF that can activate the transcriptional network of the resulting cell type ( Yamada et al . , 2014 ) . We show that impairment of a global silencing regulator leads to reversion of neurogenesis-lineage committed cells to a less differentiated cell state , but also that this can be achieved by single ectopic expression of kni . This raises the possibility that embryonic TFs such as kni drive the establishment of a regulatory circuit that blocks differentiation . Although the involved factors of such mechanisms remain to be identified , we consider the identification of kni as a strong oncogene a valuable starting point for future studies . The identification of a tumorigenic role of the embryonic TF Kni in Drosophila , is in line with the identification of other embryonic TFs playing a role in several different tumor models . For instance , aberrant expression of the embryonic TF Oct-4 blocks progenitor-cell differentiation and causes dysplasia in mouse adult epithelial tissues ( Kumar et al . , 2012; Hochedlinger et al . , 2005 ) . In humans , activation of the TF TAL1 , normally expressed early in the erythroid lineage , has been shown to alter a core transcriptional regulatory circuit that in turn leads to tumor onset ( T cell leukemia ) ( Bradner et al . , 2017 ) . Additionally , other relevant embryonic TFs , such as FOXF1 , normally expressed in mesenchyme-derived cells , activate MAPK signaling when expressed in prostate epithelial cells and contribute to tumorigenesis ( Fulford et al . , 2016 ) . Taken together , our data show that knirps can drive tumor onset and is a strong oncogene in ph505-tumors . Moreover , our work is consistent with a growing understanding between the connections of developmental gene expression and cancer . We hope that in the long-term these findings can contribute to the development of new therapies for cancers driven by misexpression of TFs . Loss of differentiation capabilities , as well as the emergence of a progenitor-like state that promotes cellular transformation and tumor initiation are common processes observed in cancer ( Roy and Hebrok , 2015; Bossuyt et al . , 2009 ) . The concept of dedifferentiation preceding tumorigenesis has been shown in Drosophila neurons , where neurons lacking the TF Lola dedifferentiate , turning on neural stem cell genes , begin to divide , and form tumors ( Southall et al . , 2014 ) . We provide evidence that kni-ectopic tumors , very similar to ph505-tumors , also fail to undergo differentiation . Besides commonalities such as loss of polarity and loss of cell identity , these two tumor models also share the ectopic activation of the JAK/STAT signaling pathway . Developmental studies suggest the cooperation between JAK/STAT and gap genes ( e . g . knirps ) in regulating expression of pair-rule genes for segmentation during embryogenesis ( Hou et al . , 2002 ) . Hyperactivation of the JAK/STAT pathway has been observed in different human cancers , where it activates survival and proliferation genes ( Buchert et al . , 2016 ) . Also , cells can be maintained in a less differentiated and more proliferative state by JAK/STAT pathway activation , as highlighted by its activation in stem cell niches in Drosophila ( Hou et al . , 2002; Amoyel and Bach , 2012; Christofi and Apidianakis , 2013 ) and in mouse embryonic stem cells ( Hao et al . , 2006 ) . Furthermore , there is evidence suggesting that this pathway must be switched off to allow differentiation of hematopoietic progenitors in flies ( Amoyel and Bach , 2012 ) . In agreement with this , blocking JAK/STAT activity suppresses the PRC1 mutant tumor phenotype ( Classen et al . , 2009 ) and in our hands induces the re-establishment of the differentiation program characteristic of eye-antennal imaginal discs . However , using the blockage of signaling pathways as a therapeutic target has been shown to be difficult due to the redundancy of the signaling networks and thus acquired drug resistance is common in cancer cells ( Gonda and Ramsay , 2015; Buchert et al . , 2016 ) . Alternatively , forced differentiation by means of TF activation might solve this issue . We used atonal , a pro-neural TF in eye discs ( Bossuyt et al . , 2009 ) and downregulated in ph505-tumors , to ultimately restore differentiation in eye-antennal tissues . This approach proved to be sufficient to prevent tumor cells from proliferating , reduce tumor burden and recover the normal pattern of differentiation . The significance of these observations , referred to as ‘differentiation therapy’ , is supported by work done in acute myeloid leukemia where therapies to overcome the cellular differentiation arrest have led to favorable outcomes ( Gocek and Marcinkowska , 2011 ) . Moreover this strategy has also been suggested to restrict the cellular plasticity of cancer stem cells ( Wainwright and Scaffidi , 2017 ) . Our findings highlight the importance of embryonic transcription factors in oncogenesis and favor the potential of re-establishing differentiation as an attractive alternative in future considerations for cancer therapy .
Further information and requests for resources and reagents should be directed to and will be fulfilled by the corresponding authors . Flies were maintained on standard food at 25°C and 60% relative humidity , under a 12 hr light: 12 hr dark cycle . All fly stocks used are listed in the Key Resources Table . Mitotic recombination was induced by the expression of FLP recombinase under the control of eyeless promoter ( eyFlp ) . Additionally , using the mosaic analysis with a repressible cell marker ( MARCM ) system ( Wu and Luo , 2006 ) , clones were fluorescently labeled with GFP . For our mutant experiments , we used ph505 allele to knock-out both genes in the ph locus ( ph-p and ph-d ) . For control experiments , MARCM clones were generated with a FRT19A blank stock line . Specifically , ‘19A tester’ stock line was crossed either with FRT19A , ph505/FM7 act-GFP or with FRT19A in order to generate mutant or control clones in eye-antennal imaginal discs , respectively . Larvae were examined at the late third instar stage . RNAi strains were initially balanced ( #2 , Cyo or #3 TM6b ) and subsequently crossed with the strain carrying the mutant allele and maintained as a stock . For generation of clones and simultaneous expression of RNAi-target , the stock mentioned above was crossed with ‘19A tester’ strain . For all final crosses 25 female virgins were crossed with eight males , in order to insure that number of larvae per fly food vial would be similar and not overcrowded . Two independent crosses for each RNAi were performed . Up to three replicates were collected from each RNAi cross . Confirmation of the results obtained by RNAi KD with a knirpsmut allele could not be realized . We did not succeed in generating a recombinant mutant allele ( Kni[FC13] ) with a FRT element , probably caused by the expected low frequency of recombination between the two elements . For determination of eclosion rates , larvae were selected accordingly to GFP expression in eye discs , counted and transferred to a new food vial . After eclosion the number of adults was counted . Eclosion rate was measured as the ratio of number larvae over the number of adults that hatched . Images of adult eyes were acquired with Nikon SMZ1270 . Third instar larvae were dissected in PBS 1x and fixed in 4% paraformaldehyde ( SIGMA , #P6148 ) in PBS 1x for 20 min at room temperature ( RT ) and washed with PBS with 0 . 1% TritonX-100 ( SIGMA , #T9284 ) ( 0 . 1% PBS-T ) for 30 min ( 3 × 10 min ) and blocked ( 0 . 1% Bovine Serum Albumin ( Serva , #11930 . 04 ) in 0 . 1% PBS-T ) for 1 hr at RT . Larvae were then incubated with primary antibodies in blocking solution overnight at 4°C , washed with 0 . 1% PBS-T ( 3 × 15 min ) and incubated with secondary antibodies in blocking solution for 2 hr at RT . After washing with 0 . 1% PBS-T for 15 min , DAPI ( Invitrogen #62248 , 1:500 ) was added and incubated for 15 min at RT . Imaginal discs were then dissected in PBS 1x and mounted in a slide with Vectashield mounting medium ( Vector Laboratories ) . The primary antibodies used in this study were: rabbit anti-Ph ( Paro lab; 1:100 ) , mouse anti-Arm ( DSHB N27A1; 1:5 ) , mouse anti-MMP1 ( DSHB 5H7B11; 1:300 ) , rat anti-ELAV ( DSHB 7E8A10; 1:30 ) , mouse anti-eve ( DSHB Eve3C10; 1:100 ) , rabbit anti-cad ( Macdonald lab; 1:500 ) , mouse anti-Eya ( DSHB eya10H6; 1:500 ) , goat anti-Hth ( H . Sun; 1:100 ) , mouse anti-Abd-B ( DSHB; 1:10 ) , rabbit anti-Dcp-1 ( Cell Signaling 9578S; 1:200 ) , rabbit anti-pH3 ( Millipore 06–570; 1:200 ) . Appropriate combinations of Alexa-coupled secondary antibodies were subsequently applied . Phalloidin-633 ( Life Technologies A22284 , 1:100 ) was used for actin staining . The secondary antibodies used were: goat anti-Rabbit Alexa 568 ( Life Technologies . , Bleiswijk , Netherlands , A-11036 ) , goat anti-mouse Alexa 568 ( Life Technologies , A-11031 ) , goat anti-rat Alexa 568 ( Life Technologies , A-11077 ) , donkey anti-goat Alexa 594 ( Life Technologies , A-11058 ) . All secondary antibodies were used at 1:500 dilutions . Samples were analyzed with a Leica SP5 or SP8 confocal microscope . Images were processed using ImageJ and were assembled with Adobe Photoshop . Transplantation assays were performed according to previous reports ( Rossi and Gonzalez , 2015 ) . Briefly , eye-antennal discs of genotypes of interest ( either ph505; ph505 , kni-KD; or FRT19A , UAS-Kni ) were cut into small pieces and transplanted into the abdomen of female hosts ( w[1118] or wild-type Oregon R ) . Transplanted hosts were kept at 25°C and monitored for GFP+ overgrowth mass . Number of tumor-bearing hosts was assessed every week upon transplantation . Transplanted hosts with ph505 tissues were used as control to account for pathogen contaminations , temperature changes or other issues that could affect the survival of the flies . Adult hosts were analyzed and images were acquired with Nikon SMZ1270 . Protocol for sample preparation for RNA-sequencing was adapted from published work ( Harzer et al . , 2013; Martinez et al . , 2009; Dutta et al . , 2013 ) . Each biological replicate for FACS was composed of a total of 200–250 eye-antennal imaginal discs of third instar larvae dissected in PBS 1x . After spinning down and removing PBS 1x , imaginal discs were placed in low-binding 1 . 5 mL tube with 200 uL of saline solution containing collagenase ( 25 discs/tube ) ( collagenase SIGMA , C1639 - 1 . 5 mg/mL diluted in Rinaldini’s saline solution ) and incubated at RT for 45 min , 300rpms . Tubes were agitated every 15 min and mechanical digestion was performed twice during collagenase incubation ( pipetting up-and-down with 27G syringe ) . After digestion , tubes were pooled in a total of 2 1 . 5 mL tubes and centrifuged for 25 min , 300 g , 4°C . Supernatant was removed and pellet was resuspended in PBS 1x . Solution was filtered and shortly kept on ice before proceeding for FACS . Several rounds of FACS-sorting were performed from pools of ph505 eye-antennal discs , using a BD FACS Aria cell sorter ( BD Biosciences ) of the FMI FACS facility ( FMI , Basel ) and data was collected on the basis of FSC/SSC parameters . Sorting time was kept below 45 min to insure the maximum viability of the cells . Two populations of cells were collected separately , GFP+- ( mutant cells ) and GFP--sorted cells ( control ) , directly into extraction buffer ( 200 µL , PicoPure RNA isolation kit , Thermo Fisher , KIT0204 ) . RNA extraction was performed accordingly to manufacturers instructions , including a step of DNase treatment ( Qiagen , catalog #79254 ) . Samples were eluted in the final volume of 11 µL and kept at −80°C . RNA concentration ( RiboGreen , ThermoFisher , #R11490 ) and integrity ( Fragment analyzer , AATI ) of sorted samples was assessed by the Genomics Facility Basel ( D-BSSE , Basel ) . From the several rounds of samples’ preparation , we choose 4 pairs of samples ( tumor and matched-control ) and three extra tumor samples from batches where control cells did not have the desired quality , to prepare libraries for sequencing . Due to the low amount of RNA in these samples , libraries were prepared by the Genomics Facility Basel using a method conceived for single cell RNA-seq ( Smart-seq2 ) ( Picelli et al . , 2014 ) . The following steps were performed on 22 libraries . There were two technical replicates per sample corresponding to a total of 11 samples ( seven tumor , four control ) . The libraries were sequenced in paired-end mode ( 2 × 150 bp ) in a NextSeq500 ( Illumina ) , and insert sizes around 300 bp ( ungapped forward and reverse tags ) . Adaptor clipping and quality trimming was performed with Trimmomatic ( Bolger et al . , 2014 ) ( v0 . 30 ) , after initial quality checks with FastQC ( v0 . 11 . 2 , www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) . Reads were aligned using the splice aware aligner STAR ( Dobin et al . , 2013 ) ( v2 . 3 . 0e ) and subsequently filtered to remove potential PCR-duplicates with Picard Tools ( v1 . 121 , broadinstitute . github . io/picard/ ) . Transcript counts were produced with HTSeq ( Anders et al . , 2015 ) ( v0 . 6 . 1 ) using the Ensembl 78 annotation ( Aken et al . , 2016 ) . The subsequent differential expression analysis was performed in R ( v3 . 1 . 0 , www . r-project . org ) using the DESeq2 package ( Love et al . , 2014 ) ( v1 . 6 . 1 ) , neglecting one library ( technical replicate ) , which did not meet quality standards . All the differentially expressed genes were submitted to the”WEB-based GEne SeT AnaLysis Toolkit’ ( WebGestalt ( Wang et al . , 2013 ) , www . webgestalt . org ) , submitting either all differentially expressed genes at the same time , or splitting them into up- and down-regulated genes . For the in vivo screen , we decided not to exclude candidates based on their log2 fold change , as is commonly done , but rather selected candidates based on a stringent adjusted p value ( padj . <0 . 01 ) and a-priori knowledge . RNA-seq profiles of our tumor and control samples were compared with available D . melanogaster datasets ( Figure 1—source data 3 ) , specifically comparing 124 differentially expressed TF-encoding genes ( Figure 1—source data 4 ) . All additional samples ( fastq-files ) were obtained from the Gene Expression Omnibus ( GEO , www . ncbi . nlm . nih . gov/geo/ ) and processed in a similar fashion as the original 11 samples . For single-end-libraries , the removal of duplicates was not performed . Settings in Trimmomatic were adjusted for each sample , taking into account the sequencer type and read lengths . All samples were aligned with STAR and counting was performed with HTSeq . Hierarchical clustering was performed after normalizing gene-expression values with DESeq2 . The expression values after variance stabilizing transformation were then mean-centered for each gene . Hierarchical clustering was performed between samples using 1-Pearson correlation as distance measure , while genes were clustered using Euclidean distance . The datasets used for comparison were retrieved from the following references: ( Graveley et al . , 2011; Gan et al . , 2010; Jüschke et al . , 2013; Potier et al . , 2014; Berger et al . , 2012; Naval-Sánchez et al . , 2013; Czech et al . , 2013; Atkins et al . , 2016 ) . Images of eye-antennal imaginal tissues were acquired using 20x or 40x objectives on the Leica SP5/SP8 confocal microscopes and processed using ImageJ or Imaris . Images of adult eyes or transplantation hosts were acquired with Nikon SMZ1270 . As a measure of tumor volume , we quantified the space taken up by the tumor in these tissues employing a quantification pipeline developed in our lab ( Beira et al . , 2018 ) . To automate image segmentation and identification of clones across imaginal discs , we used Ilastik ( Interactive Learning and Segmentation Toolkit , [Sommer , 2011] ) to build an unbiased supervised learning classification of clone regions and surrounding tissue ( with 5 ph505-tumor eye-antennal imaginal discs ) . Confocal images of tissues of interest were acquired with a 0 . 8–1 . 1 μm z-stacks . The classification method was then used for the test set of ph505-tumor tissues ( N = 50 ) , as well as upon perturbation ( either TF-KD or overexpression of ph , p35 and ato ) . After unbiased classification of clones , a Matlab script ( kindly developed by Aaron Ponti , SCF , D-BSSE ) was used to enable us to use Imaris ( Bitplane ) in order to obtain volume data for each spatially defined clone , total clone number per tissue , and tissue volume ( DAPI ) . Tumor volume ( % ) was then calculated as the ratio of tumor volume ( sum up volumes of all GFP-clones in a tissue ) over the size of the respective tissue ( volume , DAPI ) . In order to measure proliferation levels , we quantified the number of phospho-histone H3 ( pH3 ) positive cells within eye-antennal imaginal discs in the four conditions of interest ( FRT19A; FRT19A , UAS-ato; ph505; ph505 , UAS-ato ) . We used Imaris ( Bitplane ) for semi-automated image segmentation of total tissue volume ( DAPI ) , total volume of clones ( GFP+ cells ) and number of pH3+ cells . In addition , we used the segmented GFP signal to mask voxels of the pH3 +channel inside and outside of GFP positive cells to zero . In this way we were able to measure pH3+ cells inside and outside the clones . To account for differences in the size of tissues and clones , we normalized the data accordingly . For the ‘whole tissue’ condition , total numbers of pH3+ cells were normalized to total tissue volume ( per tissue ) ; for ‘inside clones’ , numbers of pH3+ cells within clones were normalized to volume of GFP+ cells per tissue; for ‘outside clones’ , numbers of pH3+ cells outside of GFP+ clones were normalized to volume of GFP- cells per tissue . Values of pH3+ cells are represented per mm3 . GraphPad Prism 7 . 0 was used for statistical analysis and generation of the graphical output . No statistical analysis was used to predetermine sample size . Sample sizes ( N ) and p-values are indicated in the figures and/or figure legends . Statistical tests used: Kruskal-Wallis with Dunn's multiple comparisons test for eclosion rate , tumor volume ( % ) , number of clones and number of pH3+ cells; one-way ANOVA with Dunnett's multiple comparisons test for tissue size and average tumor volume . ****p<0 . 0001; ***p<0 . 001; **p<0 . 01; *p<0 . 05 . All data points represented by dots in the plots for tumor volume , average tumor volume , tissue volume , number of clones and number of pH3+ cells per tissue are randomly distributed along x-axis . The accession number for the sequencing data reported in this paper is GEO: GSE101463 . | When an animal is developing as an embryo , different cells start to specialize into the specific cell types needed to form the tissues and organs of the body . How an individual cell commits to become a certain type of cell is mostly determined by which of the genes in its DNA are active . In animal cells , DNA is wrapped around proteins called histones , and one way that cells can maintain their distinct pattern of gene activity is via chemical tags on the histones . These tags can switch nearby genes on or off , and are added or removed by other proteins called epigenetic regulators . The epigenetic tags are also stably inherited when the cell divides , meaning that a cell’s identity can be maintained over many cell generations . If epigenetic regulators fail to work properly or get disrupted , the pattern of gene activity in a cell becomes altered . As a consequence , that cell can lose its identity and will often turn into a cancer cell . In fact , mutations in epigenetic regulators are found in several human cancers . It is not yet understood how these changes in gene expression lead cells to become cancerous . Torres et al . have now analyzed an epigenetic regulator called Polyhomeotic in developing larvae of the fruit fly , Drosophila melanogaster . The results show that when Polyhomeotic is not produced the fly larvae develop tumors . Moreover , the mutant cells without Polyhomeotic had different gene expression profiles compared to normal cells . This in turn caused the mutant cells , which had previously committed to a certain fate , to become more like the unspecialized cells found in early embryos . Torres et al . next showed that , among the genes that were incorrectly regulated when Polyhomeotic’s activity was compromised , one gene called knirps was switched on by mistake , which led the mutant cells to become tumor cells . When the activity of knirps was reduced instead , almost no tumors grew . Additionally , Torres et al . found that the protein encoded by knirps activates a signaling pathway that keeps tumor cells unspecialized by blocking their normal progress to a more mature and specialized state – a process known as differentiation . Experimentally raising the levels of a different molecule that ultimately promotes differentiation caused the tumor cells to grow less . These findings suggest that tumors caused when epigenetic regulation goes awry may be reversed by targeting key genes such as knirps . Further work is now needed to test whether these findings will also extend to humans . Forcing cancer cells from a highly dividing , non-specialized state into a dead-end , mature state may lead to new ways to treat cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2018 | A switch in transcription and cell fate governs the onset of an epigenetically-deregulated tumor in Drosophila |
Metzincin metalloproteases have major roles in intercellular communication by modulating the function of membrane proteins . One of the proteases is the a-disintegrin-and-metalloprotease 10 ( ADAM10 ) which acts as alpha-secretase of the Alzheimer's disease amyloid precursor protein . ADAM10 is also required for neuronal network functions in murine brain , but neuronal ADAM10 substrates are only partly known . With a proteomic analysis of Adam10-deficient neurons we identified 91 , mostly novel ADAM10 substrate candidates , making ADAM10 a major protease for membrane proteins in the nervous system . Several novel substrates , including the neuronal cell adhesion protein NrCAM , are involved in brain development . Indeed , we detected mistargeted axons in the olfactory bulb of conditional ADAM10-/- mice , which correlate with reduced cleavage of NrCAM , NCAM and other ADAM10 substrates . In summary , the novel ADAM10 substrates provide a molecular basis for neuronal network dysfunctions in conditional ADAM10-/- mice and demonstrate a fundamental function of ADAM10 in the brain .
Proteolysis of cell surface membrane proteins is a basic cellular mechanism that controls intercellular communication and the interaction of cells with their extracellular environment . Proteolysis typically occurs close to the luminal or extracellular side of the membrane and results in release of the membrane protein’s ectodomain . This process , referred to as ectodomain shedding , is a mechanism that shapes the cell surface by controlling the ectodomain length of cell surface membrane proteins thus modulating their function ( Weber and Saftig , 2012 ) . Membrane-bound metalloproteases of the metzincin family have a key role in catalyzing ectodomain shedding of membrane proteins , in particular members of the ‘a disintegrin and metalloprotease’ ( ADAM ) subfamily ( Weber and Saftig , 2012 ) . One of them is ADAM10 , which is a ubiquitously expressed type I membrane protein whose active site is located within its ectodomain , well positioned to shed the ectodomains of its substrates ( Weber and Saftig , 2012 ) . A key substrate for ADAM10 is the Notch receptor , which requires ADAM10-mediated shedding for its signaling during differentiation and development ( Hartmann et al . , 2002; Qi et al . , 1999 ) . Consequently , constitutive ADAM10-deficient mice die at embryonic day 9 . 5 most likely due to a loss of Notch signaling ( Hartmann et al . , 2002 ) . Another major substrate of ADAM10 is the amyloid precursor protein ( APP ) for which ADAM10 acts as the constitutive alpha-secretase ( Postina et al . , 2004; Lammich et al . , 1999; Kuhn et al . , 2010; Jorissen et al . , 2010 ) and thus possesses the ability to prevent the generation of the pathogenic Aβ peptide in Alzheimer’s disease ( AD ) ( Lammich et al . , 1999; Kuhn et al . , 2010 ) . This makes ADAM10 a major drug target for AD ( Postina et al . , 2004 ) , and an activator of ADAM10 expression has been tested in a clinical trial for AD ( Endres et al . , 2014 ) . Whether such a therapeutic approach is safe , remains to be seen , in particular because relatively little is known about ADAM10 substrates in brain . Besides Notch and APP , additional ADAM10 substrates , such as E-cadherin and CX3CL1 , have been identified in different organs and tissues ( Hundhausen et al . , 2003 ) . Given the embryonically lethal phenotype of constitutive ADAM10-deficient mice , little is known about ADAM10 substrates in the central nervous system ( Hartmann et al . , 2002 ) . A few ADAM10 substrates have been identified in the brain . Some of them have neuronal and synaptic functions , such as APP ( Ring et al . , 2007; Weyer et al . , 2011; 2014 ) , Neuroligin-1 ( Blundell et al . , 2010; Kim et al . , 2008 ) or N-Cadherin , which is in line with the phenotypes reported for conditional ADAM10-/- mice lacking ADAM10 expression in most neurons ( Jorissen et al . , 2010; Gibb et al . , 2010; Prox et al . , 2013 ) . While these mice circumvent ADAM10 dependent embryonic lethality , they show epileptic seizures , learning deficits and an altered morphology of postsynaptic structures in the brain and die postnatally . This demonstrates that ADAM10 is essential for synaptic and neuronal network functions in the mouse brain ( Jorissen et al . , 2010; Prox et al . , 2013 ) . Yet , the known ADAM10 substrates only partly explain these phenotypes and more ADAM10 substrates are expected to exist in brain . Their identification will allow a better mechanistic understanding of ADAM10 function in brain . Moreover , new ADAM10 substrates may be useful as biomarkers to evaluate how a patient responds to an ADAM10-modulating drug , for example in clinical trials for AD . To systematically identify neuronal ADAM10 substrates , we used the quantitative proteomic ‘secretome protein identification with click sugars’ ( SPECS ) method , which has already been successfully applied to the identification of substrates for the membrane proteases BACE1 and SPPL3 ( Kuhn et al . , 2012; 2015 ) . SPECS allows specific enrichment of cell-derived glycoproteins which otherwise would escape detection by mass spectrometry due to their low abundance . Using high-resolution mass spectrometry and label-free quantification we systematically analyzed both the levels of membrane proteins in the neuronal membrane as well as their shed ectodomains in the secretome of primary , murine neurons , either expressing or lacking ADAM10 . We identified 91 , mostly novel ADAM10 substrate candidates . Selected substrates , including NRCAM , LDLR , MT4MMP and CDH6 were validated by quantitative immunoblots in neurons and in mouse brain and point to a central function of ADAM10 in synapse formation and axon targeting . In line with this we detected an axon targeting defect in the olfactory bulb of conditional ADAM10-/- mice , similar to what has been observed in mice deficient in the novel ADAM10 substrate NRCAM .
To identify novel ADAM10 substrates we quantitatively compared the secretome of neurons with and without ADAM10 activity following the rationale that a lack in ADAM10 activity would reduce or almost abolish ectodomain shedding of a defined fraction of single span membrane proteins with other membrane proteins and soluble proteins being unaffected . To this aim , we isolated and cultured primary neurons from E15/E16 brains of a conditional Adam10 knockout mouse model ( Ad10 fl/fl ) allowing an Adam10 specific knockout upon iCre expression ( Gibb et al . , 2010 ) . After two days in vitro ( DIV ) neurons were either transduced with an empty control lentivirus or a lentivirus that coded for iCre ( Figure 1A ) . Two days later , neurons were incubated for another two days with the chemically modified sugar ManNAz , which is metabolized and incorporated as sialic acid into the glycan moieties of newly synthesized , cellular glycoproteins . Staining for beta III tubulin suggests that knockout of ADAM10 did not alter neuronal differentiation of primary neurons in vitro ( Figure 1A ) . Media of both experimental conditions ( conditioned for 48 hr ) were processed with the SPECS method to selectively enrich the endogenous , cellular glycoproteins in the medium . Considering at least two unique peptides per protein and detection of the protein ( protein group ) in at least 4 out of 5 biological replicates the mass spectrometric analysis identified 313 proteins annotated as glycoproteins . For every glycoprotein we mapped all identified peptides to its extracellular , transmembrane and cytoplasmic domains , using the QARIP webserver ( Ivankov et al . , 2013 ) . The majority of peptides matched to the extracellular domain , but not to the transmembrane or cytoplasmic domains ( data not shown ) , which demonstrates that we had detected the proteolytically released ectodomain and not the full-length membrane proteins in the neuronal secretome . 10 . 7554/eLife . 12748 . 003Figure 1 . Identification of candidate ADAM10 substrates in conditional ADAM10-/- neurons . ( A ) Workflow for the identification of ADAM10 substrates in the secretome of primary cortical neurons comprising plating of neurons , lentiviral transduction , metabolic glycan labeling , purification and protein identification and quantification via mass spectrometry . Staining of primary cortical neurons for the neuron specific marker beta III tubulin and the nuclear stain DAPI followed by quantification of beta III tubulin staining supports that neuronal differentiation was unaffected by deletion of ADAM10 . ( B ) Topology of all glycoproteins in the secretome . ( C ) Volcano plot of the quantitative comparison between the secretomes of wild type ( wt ) and Adam10 knockout neurons of all in at least 4 out of 5 experiments identified glycoproteins in the secretome of Adam10-/- and wt neurons . The p-value is depicted as negative decadic logarithm while the fold-change ( Adam10-/-/wt ) is depicted as log2 value . Significantly changed protein: p≤0 , 05 ( -log10≥1 , 3 ) . Substrates that are significantly reduced are marked red . Proteins which were validated with immunoblot have a yellow ring . Proteins which have been formerly described in literature as ADAM10 substrates are marked with a green cross . Hits whose p-values survived correction for multiple hypothesis testing yielded a p-value of at least 0 . 00135 as new significance niveau . ( D ) Topology of all significantly changed proteins in the secretome between Adam10-/- and wt neurons . ( E ) Percentage of all membrane proteins with type-I topology in the total secretome and among the significantly changed proteins upon Adam10 knockout . ( F ) Remaining shedding after Adam10 deletion in percent for members of selected protein families like the Neurexin ( NRXN ) , Neuroligin ( NLGN ) , receptor tyrosine phosphatase ( PTPR ) , Slit and receptor tyrosine kinase domain ( SLITRK ) and the L1 family . DOI: http://dx . doi . org/10 . 7554/eLife . 12748 . 003 We classified these glycoproteins according to their topology ( Figure 1B ) . These comprised 113 secreted proteins and 200 membrane proteins of which 108 had type I orientation ( Figure 1B ) . Proteins with other membrane orientations or with a GPI-anchor were also detected ( Figure 1B ) which is similar to the data in our seminal SPECS study that identified BACE1 substrates in the neuronal secretome ( Kuhn et al . , 2012 ) . Label-free-quantification of our mass spectrometric analysis revealed that the levels of 91 glycoproteins were significantly reduced in the secretome upon Adam10 deletion considering a p-value cut off of 0 . 05 based on 5 biological replicates ( Figure 1C , Supplementary file 1 – A10 knockout quantified secretome data set ) . When applying false-discovery rate ( FDR ) -based multiple hypothesis testing according the method of Benjamini and Hochberg considering an FDR = 0 . 1 and all identified glycoproteins as hypotheses , 46 membrane proteins remain significantly reduced ( Figure 1C and Table 1 ) . 10 . 7554/eLife . 12748 . 004Table 1 . Proteins that are significantly reduced in the secretome upon Cre recombinase induced deletion of ADAM10 . The table contains the 42 most reduced proteins upon ADAM10 deletion . Indicated are the names of the proteins , the gene name , number of unique peptides , topology , the mean of the ratio between neurons devoid of ADAM10 and neurons expressing endogenous levels of ADAM10 of 5 biological replicates and the p-value calculated with a two-sided , heteroscedastic t-test based on the intensity ratios for the control and the ADAM10 deletion condition . Protein names of previously published ADAM10 substrates are highlighted in italic bold . ( MEAN ) Mean value of all 5 biological replicates , ( SEM ) Standard error of the mean . ( Peptides ) Number of peptides identified for every protein . Gene symbols of proteins validated with immunoblot are marked bold . DOI: http://dx . doi . org/10 . 7554/eLife . 12748 . 004Protein NameGene SymbolPeptidesTopologyMEAN ( A10KO/CON ) TTESTMatrix metalloproteinase-17Mmp173GPI0 , 051 , 77E-03Podocalyxin-like protein 2Podxl23Type-I0 , 053 , 13E-03Cadherin-10Cdh104Type-I0 , 051 , 15E-03Protocadherin-8Pcdh816Type-I0 , 061 , 06E-03SLIT and NTRK-like protein 1Slitrk14Type-I0 , 088 , 52E-03Sulfhydryl oxidase 2Qsox23Type-I0 , 087 , 47E-03FractalkineCx3cl12Type-I0 , 096 , 08E-03Cell adhesion molecule 4Cadm45Type-I0 , 091 , 11E-03Poliovirus receptor-related protein 1Pvrl13Type-I0 , 101 , 05E-02Leucine-rich repeat and immunoglobulin-like domain-containing nogoLingo22Type-I0 , 118 , 21E-04Semaphorin-7ASema7a8GPI0 , 121 , 24E-02Receptor-type tyrosine-protein phosphatase UPtpru6Type-I0 , 121 , 35E-02Latrophilin-1Lphn17Type-I0 , 132 , 81E-02Neural cell adhesion molecule 1Ncam116Type-I/GPI0 , 141 , 24E-03Leucine-rich repeat-containing protein 4BLrrc4b16Type-I0 , 167 , 03E-04Cell adhesion molecule 3Cadm35Type-I0 , 162 , 30E-02Low-density lipoprotein receptorLdlr13Type-I0 , 161 , 04E-02Neuroligin-1Nlgn15Type-I0 , 178 , 12E-04Receptor-type tyrosine-protein phosphatase gammaPtprg4Type-I0 , 174 , 03E-03H ( + ) /Cl ( - ) exchange transporter 3Clcn32Polytopic0 , 197 , 17E-03Cadherin-6Cdh65Type-I0 , 197 , 46E-03VPS10 domain-containing receptor SorCS1Sorcs13Type-I0 , 232 , 96E-03Neuronal growth regulator 1Negr13GPI0 , 232 , 37E-02Vesicular integral-membrane protein VIP36Lman23Type-I0 , 252 , 86E-02Receptor-type tyrosine-protein phosphatase kappaPtprk12Type-I0 , 261 , 12E-03SLIT and NTRK-like protein 3Slitrk33Type-I0 , 261 , 84E-02Cell adhesion molecule 1Cadm16Type-I0 , 263 , 85E-02Transmembrane protein 132ATmem132a2Type-I0 , 261 , 52E-03Immunoglobulin superfamily member 3Igsf325Type-I0 , 278 , 04E-03Protein tweety homolog 1Ttyh15Type-I0 , 281 , 79E-02Torsin-1A-interacting protein 1Tor1aip15Polytopic0 , 292 , 62E-02Protocadherin 9Pcdh92Type-II0 , 293 , 15E-02Protocadherin-20Pcdh202Type-I0 , 303 , 28E-02Plexin-B2Plxnb24Type-I0 , 303 , 64E-02Semaphorin-4BSema4b4Type-I0 , 323 , 98E-02Neuronal cell adhesion moleculeNrcam4Type-I0 , 334 , 91E-03RGM domain family member BRgmb3Type-I0 , 351 , 06E-02Chondroitin sulfate synthase 2Chpf2Type-I0 , 357 , 63E-03NeogeninNeo127Type-I0 , 353 , 27E-03Collagen alpha-1 ( XI ) chainCol11a14GPI0 , 364 , 53E-02Beta-1 , 3-galactosyltransferase 6B3galt62Type-II0 , 384 , 85E-02Neuroligin-3Nlgn328Type-I0 , 382 , 97E-04 Surprisingly , not a single glycoprotein was significantly increased in the secretome upon Adam10 deletion . Topological classification of the 91 glycoproteins ( p<0 . 05 ) with reduced levels in the Adam10-/- neuronal secretome revealed a strong enrichment of type-I membrane proteins ( Figure 1B ) . Among the 91 significantly changed glycoproteins we identified 78 membrane proteins comprising 57 type-I , 6 GPI-anchored , 1 type-I/GPI-anchored , 5 polytopic and 9 type-II membrane proteins which we considered as potential ADAM10 substrate candidates due to their membrane localization ( Supplementary file 1 – A10 knockout quantified secretome dataset ) . Among the significantly reduced proteins upon Adam10 deletion , type-I membrane proteins were enriched to 70% compared to the total secretome that contains only 39% type-I membrane protein ( Figure 1E ) . This enrichment of membrane proteins with a type-I orientation was even more pronounced when we assigned the topology to the 42 most significantly reduced glycoproteins upon Adam10 deletion in the secretome of which 90% possess a type-I orientation ( Table 1 ) , which is in line with the observation that most previously identified ADAM10 substrates also have a type I orientation ( Weber and Saftig , 2012 ) . Among all identified ADAM10 substrate candidates we found previously described ADAM10 substrates like Neuroligin-1 ( NLGN1 ) , fractalkine ( CX3CL1 ) or the amyloid precursor protein ( APP ) ( Figure 1C , yellow ring with green cross ) which validated our experimental approach ( Lammich et al . , 1999; Kuhn et al . , 2010; Hundhausen et al . , 2003; Suzuki et al . , 2012; Reiss et al . , 2005; Colombo et al . , 2013 ) . The formerly known ADAM10 substrate N-Cadherin ( CDH2 ) was borderline significant due to one outlier . Our SPECS-based substrate identification was characterized by high sensitivity as even small reductions in total shedding were detected in ADAM10-/- cells . For example , in neurons APP is mostly shed by the aspartyl protease BACE1 and only to a low extent by ADAM10 ( Colombo et al . , 2013 ) . Yet a 20% reduction in APP shedding upon Adam10 deletion was clearly detected in our SPECS analysis . Interestingly , some substrates , such as Neuroligin-1 are ‘exclusive’ ADAM10 substrates , as their shedding was nearly completely abolished upon ADAM10 deletion . However , for other substrates including APP and CHL1 , shedding was only partly reduced in ADAM10-/- neurons , indicating that they are not only substrates for ADAM10 , but also for other proteases . The list of identified ADAM10 substrate candidates comprises entire receptor families involved in synapse function and formation and axon targeting ( Figure 1F ) . This became apparent for example for the receptor tyrosine phosphatase ( Ptpr ) , the Neuroligin ( NLGN ) , the Neurexin ( NRXN ) and the SLIT and NTRK ( Slitrk ) families , which are involved in synapse function , and the L1 adhesion molecule family , which is involved in axon targeting . In case of the receptor tyrosine phosphatase family shedding of PTPRU , PTPRG , PTPRK , PTPRS , PTPRT and PTPRD was reduced down to 12% , 17% , 26% , 40% and 43% and 42% respectively while shedding of NLGN1 , NLGN3 and NLGN4 was reduced down to 17% , 38% and 56% in the Neuroligin family . Interestingly , NLGN2 shedding was not significantly affected upon ADAM10 knockout . Presynaptic Neurexins 2 and 3 , binding partners of Neuroligins , showed only a mild reduction in shedding upon ADAM10 deletion ( Figure 1F ) suggesting that ADAM10 plays only a minor role in their proteolytic processing . Besides these families ectodomain cleavage of member 1 and 3 of the SLITRK family was strongly reduced ( Figure 1F ) . Ectodomain cleavage of Protocadherin 8 and 9 , which have been proposed to play a role in synaptogenesis , was also reduced ( Yasuda et al . , 2007 ) . Axon targeting is another important physiological process , which contributes to proper function of the brain . The L1 family , which consists of NgCAM-related cell adhesion molecule ( NRCAM ) , L1 cell adhesion molecule ( L1 ) , Close homologue to L1 ( CHL1 ) and Neurofascin has been proposed to play a role in axon targeting and function . We observed that shedding of NRCAM was strongly reduced down to 33% . Shedding of CHL1 was mildly reduced down to 60% . L1 shedding was reduced down to 64% ( Figure 1F ) , which is in agreement with their additional cleavage by the protease BACE1 ( Kuhn et al . , 2012 ) . However , the reduction of L1 shedding did just not reach statistical significance . Another axon targeting molecule whose ectodomain shedding was reduced upon Adam10 deletion was Neogenin ( NEO1 ) , which belongs to the DCC family . Adam10 deletion also resulted in reduced ectodomain cleavage of CADM1 , 3 and 4 which are members of the Cellular adhesion molecule family that has been proposed to play a role in axon myelination ( Maurel et al . , 2007; Park et al . , 2008 ) . Interestingly , we identified the Low density lipoprotein receptor ( LDLR ) as a substrate of ADAM10 whose ectodomain shedding was almost completely abolished . Finally , our data revealed that ectodomain shedding of the GPI-anchored membrane-tethered matrix metalloprotease 4 ( MT4MMP/MMP17 ) depends to a great extent on ADAM10 activity . We validated selected ADAM10 substrate candidates with immunoblots . We focused on ADAM10 substrate candidates ( marked with yellow rings in Figure 1C ) , where antibodies were available to detect the reduction of the shed ectodomain in the conditioned medium of ADAM10-/- neurons as this allowed a quantitative comparison to the proteomic SPECS analysis . Furthermore , we analyzed full length protein levels to exclude that reduced ectodomain release simply resulted from a reduced full length protein expression in Adam10 knockout neurons . First , we verified successful Adam10 deletion upon iCre expression by immunoblot ( Figure 2A , ADAM10 ) . Cre infection strongly reduced levels of immature and mature ADAM10 levels in the neuronal cell lysate . Significantly reduced or almost abolished ectodomain levels confirmed ADAM10 cleavage of the previously described ADAM10 substrates Neuroligin-1 ( NLGN1 ) and N-Cadherin ( CDH2 ) and matched the results in the SPECS experiment ( Figure 2B ) , while their full length levels were unchanged or slightly increased ( Suzuki et al . , 2012; Reiss et al . , 2005 ) . The following novel ADAM10 substrate candidates were validated by immunoblots ( Figure 2A ) : LDLR , MT4MMP , LRRC4B , NRCAM , NEO1 and CNTN2 . For all proteins ectodomain levels were reduced , while full-length protein levels in the lysate were either unchanged or increased . Importantly , the quantitative reductions of the ectodomain levels measured by immunoblots corresponded very well to the reductions measured by SPECS ( Figure 2B ) , demonstrating the quantitative accuracy of SPECS . 10 . 7554/eLife . 12748 . 005Figure 2 . Validation of ADAM10 substrates by immunoblot or ELISA . ( A ) Western blots of supernatants ( sup , conditioned for 48 hrs ) and cell lysates ( lys ) of neurons expressing endogenous levels of ADAM10 ( Con ) or no active ADAM10 upon Cre-induced ( Cre ) ADAM10 knockout . ADAM10 blot shows absent expression of ADAM10 in Cre-transduced neurons . Immunoblots are representative examples from n = 6 experiments . ( B ) Quantification of all biological replicates of representative Western blots in A and comparison to the quantified relative intensity values of remaining shedding in the SPECS experiment: Depicted is the mean of substrate ectodomain levels detected by immunblots of the supernatant under control ( CON ) and Adam10 knockout condition by Cre infection ( Cre ) and the corresponding standard error of the mean ( SEM ) of 6 biological replicates and p-value calculated with a two-tailed , unpaired t-test . Additionally , we depicted the quantified reduction of ectodomain shedding including SEM and p-value of the SPECS experiment . For CX3CL1 a sandwich ELISA was used to quantify the shedding products in the supernatant . ( C ) Quantification of the mean change in the cell lysate upon ADAM10 deletion of 6 biological replicates including the standard error of the mean and comparison to the SPECS quantified changes in the cellular glycoproteome of 5 biological replicates . For proteins that were also detected in the proteomic analysis of the cellular membrane proteins ( Figure 3 ) , the proteomic ( SPECS ) data are also indicated in the graph . DOI: http://dx . doi . org/10 . 7554/eLife . 12748 . 005 Another protein , which was validated is L1 ( Figure 2B ) . In ADAM10-/- neurons L1 shedding was reduced by about 30% , both by SPECS and immunoblot ( Figure 2B ) , demonstrating that in neurons L1 is predominantly cleaved by a protease different from ADAM10 . This is in line with previous reports demonstrating that L1 is mostly cleaved by BACE1 in neurons ( Kuhn et al . , 2012; Zhou et al . , 2012 ) , whereas it is mostly cleaved by ADAM10 in non-neuronal tissues and cell lines ( Riedle et al . , 2009; Gavert et al . , 2007; Maretzky et al . , 2005 ) . For Cadherin-6 ( CDH6 ) no antibody was available against the ectodomain , but an antibody against the cytoplasmic C-terminus . Given that ADAM10 cleavage of a membrane protein does not only lead to the secreted ectodomain , but also to the generation of a C-terminal , membrane bound fragment ( CTF ) , we detected the CDH6 CTF as a read-out of CDH6 cleavage by ADAM10 . Hence , we used an antibody directed against the CDH6 cytoplasmic domain and thus detected CDH6 full-length protein and CDH6 C-terminal fragment ( CTF ) resulting from proteolytic cleavage in the cell lysate . Adam10 deletion strongly reduced CDH6 CTF formation with CDH6 full-length protein levels being unaltered ( Figure 2A , CDH6 ) , indicating that ADAM10 is required for CDH6 proteolysis . In case of the Notch ligand Delta-Notch EGF receptor ( DNER ) we observed increased full-length protein levels in the cell lysate , which indicates that DNER is cleaved by ADAM10 . However , the DNER ectodomain was not detectable in directly loaded conditioned media of neurons . In case of fractalkine ( CX3CL1 ) we confirmed its known cleavage by ADAM10 ( Hundhausen et al . , 2003 ) using a sandwich ELISA to detect soluble fractalkine in conditioned media of neurons . Upon Adam10 deletion , we observed a strong reduction of fractalkine ectodomain levels , which corresponded to our prior quantified levels with the SPECS method ( Figure 2B , CX3CL1 ) . Taken together , all membrane proteins analyzed by immunoblot could be validated as ADAM10 substrates . When we compared full length protein levels of all investigated proteins to the full length protein levels quantified by mass spectrometry after their SPECS mediated enrichment described below ( Figure 3 ) , we were able to detect for selected substrates the same increase in the mass spectrometry read out as observed in Western blot ( Figure 2C ) . 10 . 7554/eLife . 12748 . 006Figure 3 . Quantitative analysis of glycoproteins in the neuronal cell lysate of conditional Adam10-/- neurons . ( A ) Workflow for the identification of alterations in the cellular glycoproteome upon Adam10 deletion in primary cortical neurons comprising plating , lentiviral infection , metabolic glycan labeling , biotinylation of glycoproteins on intact neurons and affinity purification protocol and finally measurement via mass spectrometry . ( B ) Volcano plot of the quantitative comparison between the cellular glycoproteomes of wild type ( wt ) and Adam10 knockout neurons of all glycoproteins identified in 4 out of 4 experiments in the cellular glycoproteome of Adam10-/- and wt neurons . The p-value is depicted as negative decadic logarithm while the fold-change ( Adam10-/-/wt ) is depicted as log2 value . Significantly changed protein: p≤0 . 05 ( -log10≥1 . 3 ) . Additionally , we corrected for multiple hypothesis testing with Benjamini Hochberg correction which gave a significance cut-off of p≤0 . 02 . Substrates that are significantly reduced are marked red . Proteins which were validated with immunoblot have a yellow ring . ( C ) Topology of all in four experiments identified glycoproteins in the cellular glycoproteome of Adam10-/- and wildtype neurons . ( D ) Topology of identified glycoproteins with significantly increased protein levels in Adam10-/- compared to wt neurons . ( E ) Overlap between proteins significantly reduced in the secretome and proteins significantly increased in the cellular glycoproteome . DOI: http://dx . doi . org/10 . 7554/eLife . 12748 . 006 For the ADAM10 substrates investigated above , reduced shedding in ADAM10-/- neurons was accompanied by unchanged or slightly increased full-length protein levels . For other substrate candidates identified by SPECS no antibodies were available for validation by immunoblot . Thus , to be independent of antibodies we complemented our SPECS analysis of the secretome by a similar analysis of the levels of membrane proteins in the neuronal membrane , which may provide additional validation of more ADAM10 substrate candidates . An added value of this analysis is the opportunity to identify membrane proteins that are not direct ADAM10 substrates , but whose protein levels are instead indirectly changed . For example , it is conceivable that an ADAM10 substrate is part of a protein complex . Hence , its increased full-length protein levels upon ADAM10 deficiency might indirectly stabilize the other complex partner , such as ion channels or neurotransmitter receptors . Identical to the SPECS method intact neurons were labeled for 48 hr with ManNAz . Instead of collecting the supernatant , labeled cells were then reacted with the biotinylated alkyne in order to label the cellular glycoproteins carrying the modified sugar ( Figure 3A ) . Because sialic acid – to which ManNAz is converted - occurs as a terminal glycan modification in the Golgi , our approach is not expected to label all cellular glycoproteins , but only mature glycoproteins within the Golgi and beyond , including the plasma membrane . As the biotin-reagent is membrane-permeable , SPECS is expected to label glycoproteins not only at the plasma membrane , but also in membranes of the secretory and endocytic pathway . Using SPECS for labeling primary murine neurons , we identified 432 glycoproteins of which 37 were significantly changed upon Adam10 deletion considering a p-value of less than 0 . 05 based on 4 biological replicates ( Figure 3B , C , Table 2 , Supplementary file 2 – A10 knockout surfaceome dataset ) . Applying multiple hypothesis testing none of the hits would remain significantly changed . However , we could confirm increased levels of APP , CNTN2 and L1CAM by Western blot ( see below , Figure 4 ) indicating that correction for multiple hypothesis testing was too strict . Of the 37 candidate proteins , 5 showed reduced levels , while 32 proteins had increased levels ( Table 2 ) . 62% of the significantly changed proteins possessed a type-I topology ( Figure 3E ) while 22% were polytopic proteins ( Figure 3D ) . Among the significantly changed proteins we found previously known substrates like APP , L1CAM or CX3CL1 ( Figure 3E ) ( Lammich et al . , 1999; Kuhn et al . , 2010; Hundhausen et al . , 2003; Colombo et al . , 2013; Maretzky et al . , 2005 ) . Furthermore , we found substrate candidates that we had already identified in the secretome of ADAM10 knockout neurons like ISLR2 , CNTN2 or Neurexin 3 ( Figure 2B , C ) . NRCAM which showed increased levels in neurons after ADAM10 knockout in immunoblots ( Figure 2A ) was borderline significant in the surface analysis . The small overlap of only 13 proteins between significantly changed glycoproteins in the secretome and significantly changed glycoproteins in neurons upon ADAM10 knockout showed that ADAM10 cleavage regulated full-length protein levels only in a fraction of cases . Potentially , ADAM10 regulates substrate function by cleavage mainly at the cell surface without affecting total full-length substrate levels , which instead may be subject to lysosomal degradation . 10 . 7554/eLife . 12748 . 007Table 2 . Proteins that are significantly changed in the cellular glycoproteome upon Cre recombinase induced deletion of ADAM10 . The table contains all significantly changed proteins upon ADAM10 deletion . Indicated are the names of the proteins , the gene name , number of unique peptides , the mean of the ratio between neurons devoid of ADAM10 and neurons expressing endogenous levels of ADAM10 of 4 biological replicates and the p-value calculated with a two-sided , heteroscedastic t-test based on the relative label-free quantification ratios ( LFQ ) for the control and the ADAM10 deletion condition . Gene Symbols of previously published ADAM10 substrates are highlighted in bold . ( MEAN ) Average change between A10 deleted ( A10KO ) and Control ( CON ) cells of 4 biological replicates , ( SEM ) Standard error of the mean . ( Peptides ) Number of Peptides identified for every protein . DOI: http://dx . doi . org/10 . 7554/eLife . 12748 . 007Protein NameGene SymbolPeptidesTopologyMEAN ( A10KO/CON ) TTESTIsoform APP695 of Amyloid beta A4 proteinApp15Type I3 , 711 , 30E-02FractalkineCx3cl16Type I2 , 913 , 00E-03Protein Ptprz1Ptprz125Type I2 , 411 , 66E-03Receptor-type tyrosine-protein phosphatase UPtpru7Type I2 , 107 , 82E-03Immunoglobulin superfamily containing leucine-rich repeat protein 2Islr216Type I1 , 896 , 18E-03Semaphorin-4BSema4b4Type I1 , 812 , 36E-02Contactin-associated protein-like 4Cntnap417Type I1 , 653 , 55E-02Fibronectin leucine rich transmembrane protein 3Flrt316Type I1 , 645 , 05E-02Netrin receptor UNC5DUnc5d5Type I1 , 614 , 66E-02Neural cell adhesion molecule L1L1cam33Type I1 , 611 , 12E-02Voltage-dependent L-type calcium channel subunit beta-1Cacnb19Type I1 , 602 , 78E-02Phospholipase D3Pld33Type II1 , 604 , 49E-02Melanoma inhibitory activity protein 3Mia323Type I1 , 552 , 04E-02Receptor-type tyrosine-protein phosphatase kappaPtprk12Type I1 , 532 , 41E-02Tetraspanin-7Tspan75Polytopic1 , 491 , 63E-02Protein Pcdhgb6Pcdhgb69Type I1 , 499 , 66E-03Serine/threonine-protein kinase LMTK3Lmtk319Type I1 , 472 , 53E-02Protocadherin-8Pcdh832Type I1 , 462 , 73E-03Protein Pcdha9Pcdha912Type I1 , 433 , 90E-02Glutamate receptor 1Gria121Type I1 , 421 , 00E-02Tenascin-RTnr17Secreted1 , 404 , 69E-03Neurocan core proteinNcan6Secreted1 , 357 , 81E-03Protein disulfide-isomerase A6Pdia67Secreted1 , 345 , 00E-02Contactin-2Cntn218GPI1 , 301 , 87E-02Receptor-type tyrosine-protein phosphatase F ( Fragment ) Ptprf19Type I1 , 304 , 74E-02Reticulon-4Rtn439Polytopic1 , 282 , 94E-04Isoform 2 of Seizure protein 6Sez66Type I1 , 273 , 76E-02Large neutral amino acids transporter small subunit 1Slc7a59Polytopic1 , 251 , 50E-03Cell adhesion molecule 4Cadm410Type I1 , 233 , 02E-02Transferrin receptor protein 1Tfrc22Type II1 , 183 , 52E-02Cadherin EGF LAG seven-pass G-type receptor 3Celsr335Type I1 , 161 , 37E-02Protocadherin Fat 4Fat448Type I1 , 123 , 05E-02Sodium/potassium-transporting ATPase subunit alpha-3Atp1a362Polytopic0 , 954 , 09E-02Sarcoplasmic/endoplasmic reticulum calcium ATPase 2Atp2a252Polytopic0 , 862 , 14E-02Slc8a1 proteinSlc8a124Polytopic0 , 853 , 18E-02Synaptic vesicle glycoprotein 2ASv2a15Polytopic0 , 856 , 99E-03Solute carrier organic anion transporter family member 3A1Slco3a14Polytopic0 , 758 , 95E-0310 . 7554/eLife . 12748 . 008Figure 4 . Analysis of proteolytic cross talk of ADAM10 and BACE1 for selected substrates . ( A ) Representative Western blots of 48 hr conditioned supernatants and cell lysates of ADAM10 fl/fl neurons treated with or without C3 to inhibit BACE1 and infected with a control virus ( Con ) or Cre recombinase virus ( Cre ) to delete Adam10 . A representative ADAM10 blot shows abolished expression of immature and mature ADAM10 . ( B ) Quantification of the mean reduction in ectodomain shedding in the conditioned medium ( sup: supernatant ) , the respective standard error of the mean ( SEM ) and the significance of six biological replicates calculated with a two-tailed heteroscedastic t-test . A10: ADAM10 . ( C ) Quantification of the mean increase of selected substrates in the cellular glycoproteome ( lys: lysate ) , the respective standard error of the mean ( SEM ) and the significance of 6 biological replicates calculated with a two-tailed heteroscedastic t-test . ( D ) Brain membrane immunoblots from Adam10fl/fl and CamkII-Cre Adam10fl/fl mice which have lost ADAM10 expression in excitatory neurons , but not in other neurons and non-neuronal cells . DOI: http://dx . doi . org/10 . 7554/eLife . 12748 . 008 Thus , for the majority of candidate ADAM10 substrates – identified in the secretome analysis – the full-length cellular levels were neither significantly increased nor reduced upon ADAM10 deletion which was also reported previously for the ADAM10 substrates Neuroglin-1 and N-cadherin ( Prox et al . , 2013; Suzuki et al . , 2012 ) . We observed a mild but statistically significant accumulation in the full-length protein levels of the presynaptically localized Neurexin family member NRXN3 which matches the observed concomitant mild modulation of shedding in the secretome . On the contrary , full-length protein levels of postsynaptic Neurexin binding partners Neuroligin 1 , 3 and 4 ( NLGN-1/-3/-4 ) were not increased upon Adam10 deletion . Besides accumulation of full length protein levels of ADAM10 substrates , we additionally detected an accumulation of polytopic membrane proteins like the ionotropic glutamate receptor GLUR1 that plays a role in excitatory synaptic transmission and Reticulon-4 ( RTN4 ) which has been implicated to play a role in axon outgrowth ( Schmandke et al . , 2014 ) . Furthermore , we detected membrane proteins , which accumulate in the cellular glycoproteome but were not detected in the secretome like LRRTM1 that has been implicated to play a role in modulating the synapse architecture ( Soler-Llavina et al . , 2013 ) . Hence , changes in shedding of a given protein do not necessarily result in accumulation of its precursor in the membrane and , conversely , accumulation of a given protein can also be caused indirectly . Some of the significantly changed proteins in the secretome as well as in the cellular glycoproteome analysis were previously also identified as BACE1 substrates , such as L1 , CHL1 and contactin-2 . This suggests a potential cross-talk between both proteases and raises the possibility of redundancy between ADAM10 and BACE1 for the cleavage of some of its substrates . To test this possibility , we used immunoblots and analyzed cellular full-length protein levels as well as ectodomain levels in response to BACE1 inhibition , Adam10 deletion or a combination of both ( Figure 4A ) . We investigated ectodomain cleavage and cellular levels of 11 substrates ( CHL1 , L1 , CNTN2 , NLGN1 , NRCAM , LDLR , CDH2 , MT4MMP , SEZ6 , APP , NEO1 ) , which we selected according to their quantitative reduction in ectodomain shedding in the current ADAM10 and the previous BACE1 SPECS study ( Kuhn et al . , 2012 ) . We selected substrates that were predominantly ( NLGN1 , NRCAM , MT4-MMP ) or partially ( CDH2 , NEO1 ) cleaved by ADAM10 or that are predominantly cleaved ( SEZ6 ) and partially cleaved by BACE1 ( APP , CHL1 , L1 , CNTN2 ) . Blotting for ADAM10 confirmed deletion of Adam10 upon lentivirus-mediated expression of Cre recombinase ( Figure 4A ) . This was further corroborated by abolishment of sAPPα in ADAM10 deleted neurons . BACE1 inhibition was monitored with the abolishment of sAPPβ ( Figure 4A , B ) and led to a compensatory increase in sAPPα which previously has been described ( Colombo et al . , 2013 ) . SEZ6 ectodomain cleavage was almost abolished and its cellular levels were increased exclusively upon BACE1 inhibition with the validated inhibitor C3 , which shows that SEZ6 ectodomain cleavage is predominantly cleaved by BACE1 , in agreement with a previous study ( Kuhn et al . , 2012 ) . However , CDH2 , NRCAM , NLGN1 , MT4MMP , NEO1 and LDLR ectodomain levels were reduced and their full length protein levels in some cases increased exclusively upon Adam10 deletion while simultaneous BACE1 inhibition had no additional effect . In contrast to the L1 family member NRCAM , Adam10 deletion and BACE1 inhibition had an additive effect in the reduction of ectodomain cleavage and accumulation of full length protein levels for the members CHL1 and L1CAM in a similar fashion as for APP ( Figure 4A , B ) . This also held true for CNTN2 . However , in case of APP the lack of BACE1 cleavage seems to be compensated by ADAM10 , which does not appear to be the case for CHL1 , L1CAM or CNTN2 . In summary , the 11 investigated substrates can be subdivided into three classes , namely predominant ADAM10 substrates , predominant BACE1 substrates and substrates which are cleaved by ADAM10 and BACE1 , demonstrating that there is a cross-talk between ADAM10 and BACE1 for some but not all substrates . Finally , we tested whether the substrates identified in neurons would also be cleaved by ADAM10 in vivo . Therefore , we analyzed the brain membrane fraction of conditional Adam10 knockout mice that had been crossed with a postnatal neuron-specific CamKII-Cre driver line at P20 ( Prox et al . , 2013 ) . Due to technical limitations we were only able to analyze the membrane fraction of ADAM10 knockout brains . We observed a 50% reduction of total ADAM10 in the brain membrane fraction , which is in line with the excitatory neuron-specific CamkII-Cre driver line sparing GABAergic neurons and the fact that ADAM10 besides neurons is also expressed in glial cells . Similar to our in vitroexperiments with primary cortical neurons full-length levels of NLGN1 were not changed in brain membrane extracts . In contrast to NLGN1 , we observed a clear increase of full-length protein levels for NEO1 , NRCAM , CHL1 and MT4-MMP ( Figure 4D ) similar to what we had observed previously in neuronal lysates upon Adam10 deletion ( Figures 2A , 4A ) . As expected , the predominant BACE1 substrate SEZ6 showed no change upon Adam10 deletion ( Figure 4D ) . Several of the newly identified ADAM10 substrates , including the newly identified substrates NRCAM and CHL1 , have functions in axon targeting ( Demyanenko et al . , 2011; Heyden et al . , 2008; Montag-Sallaz et al . , 2002 ) . Thus , it appears possible that Adam10-/- mice present defects in brain connectivity . To test this we analyzed P20 conditional Adam10fl/fl knockout mice that had been crossed with a postnatal neuron-specific CamKII-Cre driver line , resulting in forebrain specific Adam10 deletion in excitatory neurons , while the cerebellum is not affected ( Prox et al . , 2013; Casanova et al . , 2001 ) . Brain areas , where connectivity changes can be well studied , are the olfactory bulb and the hippocampus . Staining of olfactory glomeruli in the olfactory bulb with the plant lectin DBA revealed 40% of diffuse olfactory glomeruli in Adam10-/- mice compared to 10% diffuse olfactory glomeruli in control Adam10fl/fl mice . Additionally , individual axons ( white arrows ) that seemed to project to two glomeruli , were detected in the ADAM10-/- , but not in control mouse olfactory bulbs ( Figure 5A ) . Such mistargeted projections have previously been observed in mice lacking our newly identified ADAM10 substrate NRCAM or lacking the BACE1 substrate CHL1 and may result from mistargeted axons that have not been correctly eliminated during development ( Heyden et al . , 2008 ) . 10 . 7554/eLife . 12748 . 009Figure 5 . Axon targeting and synaptic alterations in ADAM10 deficient mice ( A ) Olfactory nerve axons in adult ADAM10-deficient mice . Confocal microscopy picture of olfactory bulb frontal sections stained for the detection of the olfactory nerve axons labeled with the plant lectin DBA conjugated to biotin ( in magenta ) . Cell nuclei labeled with DAPI appear in green . In wild-type mice ( upper panel ) , olfactory nerve terminals generally project to only one olfactory glomerulus ( AD10-fl/fl ) . In contrast , in ADAM10-deficient mice ( AD10-fl/fl+CamkII-Cre ) ( lower panel ) , some olfactory axons terminate in two glomeruli or pass through the glomerular layer and terminate in the external plexiform layer . In addition , the arborizations of some olfactory axons extend outside a particular olfactory glomerulus . ( B ) Quantification of olfactory glomeruli morphology in adult ADAM10-deficient mice . In wild-type mice the majority ( ~90% ) of glomeruli stained with the lectin DBA display a compact defined morphology . In ADAM10-deficient mice ( AD10-fl/fl+CamkII-Cre ) , significantly less glomeruli are compact and many glomeruli appear more diffuse . ( C ) Mossy fiber organization in adult ADAM10-deficient mice . Sagittal sections stained for the presence of synaptophysin ( red ) . Confocal microscopy showing the distribution of mossy fiber terminals in the CA3 subfield of the hippocampus . In wild-type mice ( AD10-fl/fl ) ( upper panel ) the mossy fibers are organized in the infra-inter ( IIP-MF ) and supra ( SP-MF ) pyramidal bundles both terminating in large synaptic boutons on pyramidal cell dendrites in the stratum lucidum ( sl ) of the CA3 . CA3 can be clearly distinguished from the stratum pyramidale ( sp ) containing the pyramidal cell bodies and small synatophysin positive inhibitory synapses . In ADAM10-deficient mice ( AD10-fl/fl+CamkII-Cre ) ( lower panel ) , mossy fiber terminals are also detected throughout the sp surrounding the pyramidal cell soma . ( D ) Quantitative comparison of mossy fiber terminals in the stratum lucidum with respect to number and size between adult wild type and ADAM10-deficient mice . Large synaptophysin-labeled puncta indicative for mossy fiber synaptic boutons are significantly more frequent in the stratum lucidum of ADAM10-deficient mice ( AD10-fl/fl+CamkII-Cre ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12748 . 009 Staining in the hippocampus for the synaptic marker protein synatophysin revealed additional changes in ADAM10-/- mice . Synaptophysin stains synaptic terminals of mossy fibers , which project from the dentate gyrus to the dendritic arbor of hippocampal pyramidal cells in CA3 ( Figure 5C ) . In control mice the pyramidal somata were not stained and clearly separated from synaptophysin staining in the stratum lucidum . This separation was lost in Adam10-/- mice , where a significantly increased number and area of synaptophysin positive mossy fiber terminals in the pyramidal layer revealed aberrant mossy fiber projections on hippocampal pyramidal somata in CA3 ( Figure 5D ) . Interestingly , this phenotype is also observed in mice lacking the ADAM10 substrate NCAM ( Cremer et al . , 1998 ) and in mice lacking CHL1 ( Montag-Sallaz et al . , 2002 ) , which is a substrate for both ADAM10 and BACE1 ( Heyden et al . , 2008 ) . Taken together , these results reveal mistargeted axons in the hippocampus and olfactory bulb of Adam10-/- mice which may result from impaired processing of NRCAM , CHL1 and NCAM . Other ADAM10 substrates , including APP and NLGN1 , have physiological roles in brain development , for example in synapse formation/plasticity ( Ring et al . , 2007; Weyer et al . , 2014; Suzuki et al . , 2012; Heyden et al . , 2008; Montag-Sallaz et al . , 2002; Brennaman et al . , 2013; Hick et al . , 2015 ) . In fact , one of the previously reported phenotypes in ADAM10-/- brains are synaptic alterations . In the stratum radiatum of hippocampal region CA1 in ADAM10-deficient but not in wild-type mice , enlarged stubby dendritic spines filled with mitochondria were observed ( Prox et al . , 2013 ) . Interestingly , APP also has a role in synapse formation and maintenance , suggesting that loss of the ADAM10 cleavage product sAPPα in ADAM10-/- mice may be responsible for their synaptic alterations . To this end we crossed ADAM10-/- mice with sAPPα knockin ( sAPPαki ) mice that do no longer express APP-FL but solely the secreted APPsα ectodomain due to a stop codon insertion into the mouse APP locus behind the α-secretase cleavage site ( Ring et al . , 2007 ) . However , knock-in of sAPPα in ADAM10 -/- mice , confirmed by Western blot analysis ( Figure 6A ) , as well as ultrastructural analysis ( Figure 6B ) failed to rescue this synaptic phenotype , demonstrating that dysregulation of another or even several different ADAM10 substrates cause the deficits in synaptic morphology and function . Additionally , increased astrocytic activation demonstrated by enhanced GFAP expression could not be rescued by sAPPα expression ( Figure 6A ) . 10 . 7554/eLife . 12748 . 010Figure 6 . Synaptic alterations in ADAM10 deficient mice are not rescued by APPsα overexpression ( A ) Mice deficient for ADAM10 in adult brain ( ADAM10-fl/fl+CamkII-Cre ) and transgenic for a sAPPα knockin into the APP locus ( ADAM10ko sAPPki ) thus expressing only sAPPα , were generated by crossing sAPPα knockin mice ( sAPPki ) and ADAM10-fl/fl+CamkII-Cre knockout mice ( ADAM10ko ) for two generations . Immunoblot analysis of extracts from cortex , cerebellum and hippocampus from 21 days old mice revealed that ADAM10-expression was clearly reduced in cortex and hippocampus of conditional ADAM10 knockout mice ( A10fl/fl , CamkII-Cre pos . ) . No reduction of ADAM10 was observed in cerebellum since this tissue was not targeted by the CamkII-Cre driver . The proform ( pADAM10: 100 kDa ) and the mature form ( mADAM10: 70kDa ) of ADAM10 are depicted . Expression of sAPPα ( 105 kDa ) was detectable in sAPPα transgenic mice ( sAPPki + ) but absent in wild type mice . Full length APP ( APPfl ) ( 105 kDa ) expression was lost in sAPPα knockin mice but is present in wild type mice . As reported ( Prox et al . , 2013 ) GFAP ( 45 kDa ) expression in cortex of 21 days old brains was increased in the ADAM10 knockout mice ( ADAM10 ko ) . This pathology was not reversed upon sAPPα expression ( sAPPki + ) , hinting to an ongoing astrogliosis . ( B ) High resolution electron micrographs of spines in the hippocampal CA1 stratum radiatum from 21 days old mice . Conditional ADAM10 deficiency ( ADAM10ko ) leads to enlarged and stubby spines ( light red shaded ) . The presence of sAPPα ( sAPPki ) in ADAM10 knockout mice is not sufficient to rescue the alterations in spine morphology . In wildtype hippocampus spines are characterized as tiny spine heads with no organelles . Scale bars: 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12748 . 010
Ectodomain shedding of membrane proteins is a fundamental mechanism to control intercellular communication , the interaction between cells and their environment . Our study demonstrates that ADAM10 is a major shedding enzyme in the nervous system and that a loss of ADAM10 leads to defects in neuronal connectivity . The metalloprotease ADAM10 plays manifold roles in development and disease . Studying ADAM10 substrates in primary cells or tissue has not been possible until the advent of conditional Adam10 knockout mice as constitutive Adam10 knockout mice die latest at embryonic day 9 . 5 ( E9 . 5 ) . These mice succumb to malformations like a retarded heart development , extracorporeal formation of blood vessels and improper formation of somites , which are considered to result from a lack in proteolytic processing of Notch by ADAM10 ( Hartmann et al . , 2002; Jorissen et al . , 2010; Krebs et al . , 2000; Kageyama and Ohtsuka , 1999 ) . However , postnatal neuron-specific disruption of Adam10 in the brain leads to impaired learning , defects in long term potentiation and seizures which are considered to result from defects in synapse function and architecture and microglial and astrocytic activation ( Prox et al . , 2013 ) . These phenotypes point to an important postnatal function of ADAM10 via cleavage of additional substrates beyond Notch . So far , Notch and other ADAM10 substrates like N-Cadherin , fractalkine , Neuroligin-1 or E-Cadherin have already been described ( Hundhausen et al . , 2003; Suzuki et al . , 2012; Reiss et al . , 2005; Maretzky et al . , 2005 ) . However , only few substrates like N-Cadherin , Neuroligin-1 , Notch-1 and Notch-2 have been shown to be cleaved in primary cells like neurons due to the aforementioned embryonic lethality of constitutive Adam10 knockout mice and the lack of conditional Adam10 knockout models until recently ( Prox et al . , 2013; Suzuki et al . , 2012; Zheng et al . , 2012 ) . Here , we provide the first systematic analysis of ADAM10 substrates in primary cortical neurons of a conditional Adam10 knockout mouse . Applying SPECS to enrich newly synthesized glycoproteins from conditioned media and the cell surface of primary neurons we identified more than 300 glycoproteins in the secretome , which is slightly more than in our previous study identifying BACE1 substrates in neurons ( Kuhn et al . , 2012 ) . Additionally , we identified and quantified more than 400 membrane glycoproteins on or close to the cell surface of neurons which is 10 times more membrane proteins than in a previous study where cell surface biotinylation was used , but without the SPECS labeling ( Sanz et al . , 2015 ) . This demonstrates the power of the metabolic SPECS labeling to an in-depth quantification of the neuronal membrane proteome . Our analysis revealed that almost 50% of all membrane proteins released into the secretome are reduced upon Adam10 deletion of which 70% possess a type-I topology . Furthermore , we found that 13 secreted proteins like clusterin were reduced . Because ADAM10 cleaves membrane-bound substrates , the reduced clusterin levels are likely to be the consequence of secondary effects such as an increased internalization due to stabilization of its surface receptor in ADAM10-deficient neurons . The large number of substrates makes ADAM10 a major sheddase in the nervous system and an important modifier of the neuronal secretome . Among the candidate ADAM10 substrates in the secretome we could identify whole protein families having roles in synapse function and architecture . Their reduced cleavage upon Adam10 deletion fits to the phenotypes described in neuron-specific conditional Adam10 knockout mice , including defects in long term potentiation and seizures ( Prox et al . , 2013 ) . For example , Neurexins and Neuroligins are known to form trans-synaptic complexes which are especially necessary for synapse function ( Figure 7 ) ( Bang and Owczarek , 2013; Südhof , 2008 ) . Adam10 ablation caused a strong reduction in ectodomain shedding of postsynaptic NLGN1 , 3 and slightly less pronounced for NLGN4 . Hence , shedding of Neuroligins might negatively regulate their synaptic function ( Suzuki et al . , 2012 ) . Surprisingly , NLGN2 shedding was not affected by Adam10 deletion , which might be explained by the distinct roles played by these proteins . In fact , NLGN1 and 3 contribute to the function of excitatory , glutamatergic synapses while NLGN2 contributes to the function of inhibitory , GABAergic synapses ( Kang et al . , 2014 ) suggesting that ADAM10 might preferably be involved in the modulation of glutamatergic , excitatory but not GABAergic inhibitory synapse functions . This is additionally supported by the fact that the ionotropic glutamate receptor GLUR1 ( Gria1 ) , previously described to be modulated by synaptic NLGN1 levels ( Wittenmayer et al . , 2009 ) , was increased in the cellular glycoproteome upon Adam10 deletion while GABA receptors which are known to interact with NLGN2 were not increased ( Kang et al . , 2014 ) . Besides the reduction of NLGN1 , 3 and 4 shedding , Adam10 deletion resulted in a concomitant shedding reduction of the NLGN receptors NRXN 2 and 3 while the full length NRXN precursors mildly accumulated in the cellular glycoproteome . We additionally observed reduced ectodomain shedding for almost all members of the receptor tyrosine phosphatase family ( PTPR ) and the Slit and receptor tyrosine kinase domain ( SLITRK ) family which both play roles in synapse function and formation ( Figure 7 ) ( Um et al . , 2014; Yim et al . , 2013 ) . Specific members of both families like for example PTPRD and SLITRK3 have been shown to interact with each other transsynaptically ( Um et al . , 2014; Takahashi et al . , 2012 ) . While SLITRK1 , 2 , 4 and 5 and PTPRS have been implicated in the formation of excitatory synapses , SLITRK3 and PTPRD seem to be involved in the formation of inhibitory synapses ( Yim et al . , 2013 ) . 10 . 7554/eLife . 12748 . 011Figure 7 . Overview of subcellular localization and interaction partners of ADAM10 substrates . Overview of the subcellular localization and interaction partners of selected substrates of ADAM10 and their respective behavior in terms of ectodomain shedding and cellular amount upon Adam10 deletion . Upper symbol refers to ectodomain shedding while lower symbol indicates the cellular levels of the protein . DOI: http://dx . doi . org/10 . 7554/eLife . 12748 . 011 Impaired cognitive functions can also result from impaired axon targeting and function leading to impaired signal transmission and wrong connection of cortical regions in the brain . We observed mistargeted axons of the olfactory epithelium projecting instead of usually to one olfactory glomerulus to two olfactory glomeruli in the olfactory bulb of ADAM10-deficient mice . Similar axon targeting phenotypes have also been reported upon knock-out of L1 family members Nrcam and Chl1 while L1 knockout mice succumb to the much more severe CRASH syndrome with corpus callosum hypoplasia and mental retardation ( Heyden et al . , 2008; Montag-Sallaz et al . , 2002; Demyanenko et al . , 2011; Kolata et al . , 2008 ) . Indeed , we observed reduced ectodomain shedding of all L1 family members ( L1 , CHL1 , NRCAM and Neurofascin ( NFASC ) ) with NRCAM shedding being reduced the strongest . In contrast , shedding of L1 and CHL1 , previously identified as major substrates of BACE1 in neurons ( Kuhn et al . , 2012; Zhou et al . , 2012 ) were only mildly modulated upon ADAM10 knockout in line with previous data that demonstrate cleavage of CHL1 and L1 by metalloproteases in non-neuronal cells ( Maretzky et al . , 2005; Naus et al . , 2004 ) . Simultaneous BACE1 inhibition and Adam10 knockout resulted in a further reduction of ectodomain shedding of CHL1 and L1 in a similar fashion to APP ( Colombo et al . , 2013; Zhou et al . , 2012 ) . While for L1 differential functional outcomes of ADAM10 and BACE1 cleavage have not been investigated so far , it recently has been shown that CHL1 cleavage by BACE1 is indispensible for Semaphorin 3a induced axon repulsion while inhibition of metalloprotease mediated CHL1 cleavage had no impact which suggests that modulation of axon growth in the axonal or presynaptic compartment via CHL1 is only modulated by BACE1 ( Barão et al . , 2015 ) . Interestingly , as shown here for conditional ADAM10 knockout CamKII-Cre mice , misplaced hippocampal mossy fiber terminals were also observed in CHL1- or NCAM-deficient ( Colombo et al . , 2013; Yasuda et al . , 2007 ) but also in BACE1-deficient mice ( Hitt et al . , 2012 ) . Likewise , CHL1- , NrCAM- , BACE1- , or ADAM10-deficiency all result in mistargeted axons in olfactory glomeruli ( Montag-Sallaz et al . , 2002 ) , which may indicate functions for the refinement of connectivity ( Montag-Sallaz et al . , 2002 ) . Nevertheless , other identified substrates , such as neogenin ( NEO1 ) and contactin-2 ( CNTN2 ) may also contribute ( Denaxa et al . , 2005; Braisted et al . , 2000 ) , which makes a substrate phenotype correlation difficult . Conditional Adam10 knockout mice also show activated microglia upon postnatal , neuronal deletion of Adam10 ( Prox et al . , 2013 ) . One important chemokine expressed by neurons is fractalkine ( CX3CL1 ) which has previously been proposed to be cleaved by ADAM10 and ADAM17 ( Hundhausen et al . , 2003; Garton et al . , 2001 ) . Abolished constitutive shedding of CX3CL1 and increased CX3CL1 full length protein levels in neurons upon knockout of ADAM10 demonstrate that ADAM10 regulates CX3CL1 levels in neurons ( Figure 3B ) . The biological role of ADAM10 cleaved soluble CX3CL1 might be a reduction of microglial activation leading to neuroprotection . This is supported by the finding that microglial activation and neuronal cell death in the substantia nigra upon MPTP treatment are increased in a CX3CL1 knockout mouse model and can be rescued by soluble but not membrane bound CX3CL1 suggesting a neuroprotective role of soluble CX3CL1 ( Ueno et al . , 2013; Nash et al . , 2013; Morganti et al . , 2012 ) . We also identified low density lipoprotein receptor ( LDLR ) as a novel substrate of ADAM10 . LDLR has been described to undergo induced release upon treatment with the phorbolester PMA and its c-terminal fragment is subject to intramembrane proteolysis by γ-secretase ( Tveten et al . , 2013 ) . Here , we show that constitutive shedding of LDLR is catalyzed by ADAM10 . Soluble LDLR has been shown to be able to bind to LDL . However , functional consequences of ADAM10 catalyzed cleavage of LDLR like an altered LDL uptake have not been investigated so far . Another important finding was that in contrast to other proteases like ADAM17 or ADAM9 ( Kuhn et al . , 2010 ) , cellular levels of the GPI-anchored , neuronally expressed metalloprotease MT4MMP increased while MT4MMP ectodomain shedding was abolished upon Adam10 deletion ( Rikimaru et al . , 2007 ) . MT4MMP is assumed to mostly cleave soluble proteins in the extracellular matrix ( Itoh , 2015 ) . However , since little is known about the physiological substrate spectrum of MT4MMP in neurons , we cannot exclude that some of our observed changes in the ADAM10-/- neurons and their secretome may be an indirect consequence of the reduced MT4MMP shedding . Our data revealed that some proteins are exclusive substrates of ADAM10 ( NRCAM , CX3CL1 ) while others are exclusively cleaved by BACE1 ( SEZ6 , APLP1 ) . Finally , there is a group of proteins that can be cleaved by both proteases ( APP , CHL1 , L1 ) . This specificity may result from differential sorting of the substrates towards the pre- and postsynaptic compartment , or from primary sequence constraints in the juxtamembrane region . However , this classification will be extended in the future as substrates of other in neuron expressed proteases like MT4MMP and MT5MMP have not been determined in the brain yet ( Rikimaru et al . , 2007; Jaworski , 2000 ) . Taken together , the identification of novel ADAM10 substrates demonstrates that ADAM10 has a large spectrum of neuronal substrates , but the ADAM10 substrate repertoire is presumably even larger . For example Notch1 , an established ADAM10 substrate , was not detected in our analysis , potentially because its expression level was below the detection limit or because it was not expressed in the neurons at the time-point of analysis or because its ADAM10-mediated shedding is ligand-dependent . Additionally , while we identified ADAM10 substrates in neurons , ADAM10 is also expressed in astrocytes and oligodendrocytes and certain proteins are expressed at later developmental stages than what we have investigated ( Sakry et al . , 2014; Ludwig et al . , 2005 ) . Furthermore , it is possible that certain biological effects of Adam10 deletion cannot be investigated in dissociated in vitro cultures of neurons but require the context of brain tissue to provide the correct extracellular matrix and tissue architecture , which is provided by slice cultures or even better in vivo . In summary , our study demonstrates that ADAM10 is a major sheddase in the brain and is required for correct axon targeting in the olfactory bulb and the hippocampus of conditional ADAM10-/- mice . The newly identified substrates demonstrate that ADAM10 has a fundamental role in the brain and provide a molecular explanation for cognitive deficits , seizures and axon mistargeting in neuron-specific ADAM10 knockout mice . Finally , these findings will also aid the prediction of potential side effects of ADAM10-activating drugs as they are considered for the treatment of AD and prion disease .
Antibodies against murine CHL1 ( clone 7B2 , IgG2A ) and murine Neogenin ( NEO1 ) ( clone 21A8 IgGG1 ) were produced by fusing murine CHL1 ectodomain ( aa26-1006 ) and murine Neogenin ectodomain ( aa45-1101 ) to an N-terminal CD5 signal peptide and a C-terminal 2XStrepII tag which subsequently were expressed in HEK293T cells . Recombinant ectodomain was purified from 300 ml conditioned media with 300 µl Streptactin sepharose ( IBA GmbH , Göttingen , Germany ) according to the instructions of the manufacturer . 50 µg of each purified fusion protein ( mCHL1-2XStrepII , mNEO1-2XStrepII ) were injected intraperitoneally ( i . p . ) and subcutaneously ( s . c . ) into LOU/C rats using incomplete Freund's adjuvant supplemented with 5 nmol CpG 2006 ( TIB MOLBIOL , Berlin , Germany ) . After a six week interval a final boost with 50 µg hAPP and CpG 2006 was given i . p . and s . c . three days before fusion . Fusion of the myeloma cell line P3X63-Ag8 . 653 with the rat immune spleen cells was performed according to standard procedures . Hybridoma supernatants were tested in a solid-phase immunoassay with mCHL1-2XStrepII , mNEO1-2XStrepII or an irrelevant StrepII fusion protein coated to ELISA plates . Antibodies from tissue culture supernatant bound to mCHL1-2XStrepII or mNEO1-2XStrepII were detected with HRP conjugated mAbs against the rat IgG isotypes ( TIB173 IgG2a , TIB174 IgG2b , TIB170 IgG1 all from ATCC , R-2c IgG2c homemade ) , thus avoiding mAbs of IgM class . HRP was visualized with ready to use TMB ( 1-StepTM Ultra TMB-ELISA , Thermo , Braunschweig , Germany ) . MAbs that reacted specifically with mCHL1-2XStrepII or mNEO1-2XStrepII were further analyzed in non-reducing Western blot and immunocytochemistry . Murine DNER ectodomain ( aa26-620 ) was fused to an N-terminal CD5 signal peptide and a C-terminal biotin accepting peptide and 6XHIS tag . This construct was expressed in HEK293T cells with biotin ligase ( BirA ) fused to an N-terminal CD5 signal peptide and a C-terminal KDEL endoplasmic reticulum ( ER ) retention motif to target BirA to the ER . Recombinant ectodomain was purified from serum free media with a 1 ml Nickel NTA column ( GE Health Care , Solingen , Germany ) according to the instructions of the manufacturer . Briefly , the column was equilibrated with 5 column volumes ( CV ) 5 mM Imidazol in phosphate buffered saline ( PBS ) , 300 ml serum free media containing the recombinant protein and supplemented with 5 mM Imidazol were loaded on the column at a flow rate of 2 ml/min . Afterwards the column was washed with 5 CV 5 mM Imidazol in PBS . Bound DNER-BAP/HIS was eluted with 5 CV 200 mM Imidazol in PBS . LSL chicken hatching eggs were obtained from LSL Rhein-Main ( Gut Heinrichsruh , Berglern ) . Birds were hatched and raised at the Institute for Animal Physiology , Department of Animal Science , LMU Munich . The birds were fed ad libitum with a standard chicken diet . At an age of eight months hens were immunized i . m . with 300 µg purified DNER protein mixed with Freund’s adjuvant complete ( Sigma-Aldrich , Munich , Germany ) . Birds were boosted i . m . with 300 µg protein plus Freund’s incomplete adjuvant ( Sigma-Aldrich ) three weeks after the initial immunization . Eggs were collected two weeks after the boost for a period of two weeks . Animal experiments were authorized by the Regierung von Oberbayern ( 55 . 2-1-54-2532 . 6-12-09 ) . DNER chicken antibodies were purified by antigen affinity purification . 300 µg DNER-BAP-HIS was bound to 300 µl streptavidin bead slurry ( 50% w/v ) . The purified immunoglobulin fraction of chicken egg yolk was loaded on the DNER-BAP-HIS Streptavidin column to isolate DNER specific antibodies . Bound DNER antibodies were eluted with acidic elution at pH 2 . 5 . The following antibodies were used in this study: ADAM10 ( Abcam/Epitomics , Cambridge , UK , EPR5622 ) , Actin ( Sigma-Aldrich , A2066 ) , sAPPα 5G11 ( Colombo et al . , 2013 ) , sAPPβ 8C10 ( Kuhn et al . , 2010 ) , APP 22C11 ( Millipore , Billerica , MA ) , MAB348sAPPα ( Covance , Princeton , NJ , Sig-39151 ) , APP ( Sigma-Aldrich , A8717 ) , Cadherin 6 ( Abcam/Epitomics , EPR217 ) , Calnexin ( Enzo , Stressgen , Farmingdale , NY , USA , ADI-SPA-860 ) , Calbindin , ( rabbit polyclonal anti-calbindin D-28k antibodies Swant , Bellinzona , Switzerland ) , mContactin ( R&D Systems , Wiesbaden , Germany , AF4439 ) , DNER ( R&D Systems , AF2254 ) , GFAP ( Sigma-Aldrich , G3893 ) , L1CAM Clone 555 ( kindly provided by Peter Altevogt ) , LDL receptor ( R&D systems , AF2255 ) , Lrrc4b ( R&D systems , AF4995 ) , MT4-MMP ( MMP17 ) , ( Abcam/Epitomics , EP1270Y ) , N-Cadherin ( Abcam/Epitomics , EPR1791-4 ) , NrCAM ( Abcam , ab24344 ) , Neuroligin-1 ( Synaptic Systems , Göttingen , Germany 4C12 ) , Synaptophysin ( mouse monoclonal Sigma-Aldrich ) . Biotin-SP-conjugated goat anti-mouse secondary antibodies ( Jackson Immunoresearch Laboratories , West Grove , Pa . ) , and Cy3-conjugated streptavidin ( Dianova , Hamburg , Germany ) , Alexa 488 goat anti-rabbit secondary antibodies ( Molecular Probes , Leiden , The Netherlands ) . Levels of murine fractalkine ( CX3CL1 ) were measured using an Elisa Kit ( R&D systems , detection limit 98 pg/ml ) . Samples were stored at -80°C and analyzed undiluted after thawing . The generation and genotyping of knockout lines was formerly described: ADAM10 conditional KO embryos ( Gibb et al . , 2010 ) , ADAM10 conditional KO mouse crossed with CamkII-Cre ( Prox et al . , 2013 ) , APPsα-KI ( Ring et al . , 2007 ) . Homozygous Adam10 deficient sAPPα knockin ( sAPPαki ) mice were generated from Adam10F/F CaMKIIα-Cretg/+ mice ( Prox et al . , 2013 ) crossed with APPsα-KI mice ( Ring et al . , 2007 ) . Genotypes were confirmed by specific PCR on tail biopsies . PCR on CaMKIIα-Cre and floxed Adam10 was performed as previously described ( Prox et al . , 2013 ) . PCR on sAPPα was performed with primers 5’GGCTGACAAACATCAAGACGGAAGAG3’ , 5’CACACCTCCCCCTGAACCTGAAAC3’ and 5’CTGCGAGAGAGCATCCCTACAACC3’ . Embryonic primary cortical neurons were prepared as described ( Kuhn et al . , 2010; Colombo et al . , 2013 ) . Briefly , cortex samples from ADAM10 conditional KO embryos ( Gibb et al . , 2010 ) were collected and dissociated in DMEM plus 200 U of papain ( Sigma Aldrich ) . Neurons were plated in 6 well plates precoated with poly-D-lysine ( 1 . 5×106 cells/well ) . Plating medium was 10% ( v/v ) fetal calf serum FCS/DMEM which was changed after 4 hr to B27/Neurobasal ( Thermo , Braunschweig , Germany ) supplemented with 0 . 5 mM glutamine and 1% P/S . All experimental procedures on animals were performed in accordance with the European Communities Council Directive ( 86/609/EEC ) . Codon-improved Cre recombinase ( iCre ) ( Shimshek et al . , 2002 ) expressing lentiviral particles were generated as previously described ( Kuhn et al . , 2010; Colombo et al . , 2013 ) . Briefly , lentiviruses were generated by transient cotransfection of HEK293T cells ( DSMZ , Braunschweig , Germany ) with the plasmids psPAX2 , pCDNA3 . 1 ( − ) -VSV-G and as transfer vector F2UΔZeo-iCre using Lipofectamine 2000 ( Thermo ) . Lentiviral particles for infection of murine primary cortical neurons were concentrated and purified by ultracentrifugation . Lentiviral stocks were stored at −80°C until use . Secretome enriched and subsequent proteomic analyses were essentially performed as described earlier ( Kuhn et al . , 2012 ) . In brief , 40 million neurons were plated . After two div , neurons were infected with concentrated lentiviruses coding for iCre or Control . After five div neurobasal medium was exchanged for neurobasal medium supplemented with 200 nM tetraacetyl-N-azidoacetyl mannosamine ( ManNAZ ) . Two days conditioned media were subsequently collected and filtered through 0 . 45 µm PVDF Millex filter ( Millipore , Darmstadt , Germany ) into a VivaSpin 20 centrifugal concentrator ( 30 kDa ) at 4°C . VivaSpin 20 columns were centrifuged at 4600 rpm at 4°C to remove non-metabolized ManNAZ . The retentate was filled with 20 ml H2O . This procedure was repeated three times . In the last step , the ddH2O refill step was omitted . Instead , 100 nM of DBCO-PEG12-biotin ( Click Chemistry Tools , Scottsdale , AZ ) diluted in 1 ml ddH2O was added to biotinylate metabolically azide-labelled glycoproteins . Centrifugal concentrators were incubated overnight at 4°C . For removal of non-reacted DBCO-PEG12-Biotin , VivaSpin20 centrifugal concentrators were subjected to three times of centrifugation with subsequent ddH2O refill . After the last centrifugation step , the retentate was diluted in 5 ml PBS with 2% SDS ( v/v ) and 2 mM Tris 2-carboxyethyl-phosphine ( TCEP ) . For purification of biotinylated proteins , the sample was loaded on a 10-ml polyprep column with a streptavidin bead bed formed of 300 µl streptavidin slurry . After binding of proteins , streptavidin beads were washed three times with 10 ml PBS supplemented with 1% SDS . To elute the biotinylated and azide-labelled glycoproteins , streptavidin beads were boiled with urea sample buffer containing 3 mM biotin to compete for the binding of biotinylated proteins . Neurons that were labeled with 200 nM ManNAz for 48 hr were washed twice with 5 ml cold PBS . Afterwards 100 nM DBCO-PEG12-biotin ( Click Chemistry Tools ) diluted in 2 ml PBS were evenly distributed on the neurons and incubated at 4°C for 2 hr . After the incubation period , neurons were washed twice with 5 ml PBS to be lysed in 5 ml STE buffer ( NaCl , 150 mM , Tris 50 mM , 2 mM EDTA ) with 1% ( v/v ) NP40 per flask . Lysates were subjected to a clarifying spin at 4000 g . The clarified lysate was subsequently loaded on a polyprep column with a streptavidin bead bed formed of 300 µl streptavidin slurry . After binding of proteins , streptavidin beads were washed three times with 10 ml PBS supplemented with 1% SDS . To elute the biotinylated and azide-labelled glycoproteins , streptavidin beads were boiled with urea sample buffer containing 3 mM biotin to compete for the binding of biotinylated proteins . Brain homogenates were prepared as previously described ( Kuhn et al . , 2012 ) from P20 brains of conditional ADAM10 ( AD10 fl/f ) CamkII-Cre mice ( Prox et al . , 2013 ) . Mice were sacrificed at postnatal day P20 and brain regions were prepared ( cortex , hippocampus and cerebellum ) . Samples were homogenized and lysed in cell lysis buffer ( 5 mM Tris base pH 7 . 4 , 1 mM EGTA , 250 mM sucrose , 1% Triton X-100 supplied with protease inhibitors ) by passing 15 times through a 23 G syringe and incubated for 1 hr at 4°C . Cell debris was pelleted and protein concentration in the supernatant was determined by BCA assay ( Pierce , Thermo Fisher Scientific , Waltham , Ma ) . To analyze APP processing , samples were homogenized and lysed in RIPA buffer ( 50 mM HEPES pH 7 . 4 , 150 mM sodium chloride , 1 mM EDTA , 0 . 1% SDS , 1% NP-40 , 0 . 5 mM sodium deoxycholate supplied with protease inhibitors ) by passing 15 times through a 23 G syringe . Cell debris was pelleted and supernatant was centrifuged ( 1 hr , 120000 x g , 4°C ) . Protein concentration in the supernatant was determined by BCA assay ( Pierce ) . Samples were separated in either a 10% SDS-PAGE or a 4–12% gradient BIS/Tris NuPAGE Novex gels ( Thermo Fisher Scientific ) and transferred to nitrocellulose membranes ( Roth , Karlsruhe , Germany ) to perform Western blot analysis . Ultrastructural sample preparation and analysis was performed as previously described ( Prox et al . , 2013 ) . Proteins were separated on a 10% Tris/glycine SDS gel . Afterwards , qualitatively equal gel slices were cut out from the gel with the exception of the remaining albumin band at around 60 kDa . Proteins in the gel slices were subject to trypsinization according to standard protocols ( Kuhn et al . , 2015 ) . For Western blot analysis of conditioned media and cell lysates of ADAM10 knockout neurons 20 µl of 48 hr conditioned supernatant ( 1 . 5 ml ) of a six well plate with 1 . 5x106 neurons and 20 µl of 250 µl cell lysate of a six well plate with 1 . 5x106 neurons were loaded onto Tris/glycine SDS gels respectively . Proteins were transferred to 0 , 45 µm nitrocellulose membranes and blocked with skim milk prior to decoration with primary and secondary antibody . Six biological replicates were analyzed for each substrate . Mass spectrometry experiments were performed on an Easy NLC 1000 nanoflow HPLC system II ( Proxeon , Odense , Denmark ) connected to an LTQ-Velos Orbitrap Pro ( Thermo Fisher Scientific , Braunschweig , Germany ) . Peptides were separated by reverse phase chromatography using in-house made 30 cm columns ( New Objective , FS360-75-8-N-S-C30 , Woburn , MA ) packed with C18-AQ 2 , 4 µm resin ( Dr Maisch GmbH , Ammerbuch-Entringen , Germany , Part No . r124 . aq ) . A 90-min gradient ( 5–40% ) at a flow rate of 200 nl/min was used . The measurement method consisted of an initial FTMS scan recorded in profile mode with 30000 m/z resolution , a mass range from 300 to 2000 m/z and a target value of 1000000 . Subsequently , collision-induced dissociation ( CID ) fragmentation was performed for the 15 most intense ions with an isolation width of 2 Da in the ion trap . A target value of 10000 , enabled charge state screening , a monoisotopic precursor selection , 35% normalized collision energy , an activation time of 10 ms , wide band activation and a dynamic exclusion list with 30 s exclusion time were applied . Considering that every biological replicate had been analyzed twice with mass spectrometry ( technical replicate ) after sample processing , five biological replicates with 2 technical replicates each of the neuronal secretome upon ADAM10 knockout and four biological replicates with two technical replicates each of the neuronal ADAM10 knockout surface proteome were analyzed with the freely available MaxQuant suite ( version 1 . 4 . 1 . 2 ) ( Suzuki et al . , 2012 ) . Protein identification was performed using the integrated Andromeda search algorithm ( Reiss et al . , 2005 ) . First search , mass recalibration and main search of tryptic peptides were performed using a murine Uniprot database downloaded on the 08/21/2014 ( 86749 entries ) for the ADAM10 secretome dataset and a murine Uniprot database downloaded on the 05/16/2014 ( 86749 entries ) for the ADAM10 surfaceome dataset . Two missed cleavages were allowed . Peptide as well as protein false discovery rate was set to 1% . Mass accuracy was set to 20 ppm for the first search and 5 ppm for the main search . Quantification was performed between the respective control and ADAM10 knockout conditions on the basis of unique and razor peptides . Missing values were imputed in Perseus 1 . 5 . 16 following a standard distribution . p-values were calculated from log2 transformed relative intensity ratios of 5 biological replicates for the neuronal secretome and 4 biological replicates for the neuronal surface proteome with a heteroscedastic , two-sided student’s t-test . Proteins with a p-value of p≤0 . 05 were considered as hits . To correct for multiple hypothesis testing the Benjamini-Hochberg post hoc was applied with an adjusted false-discovery rate of 0 . 10 . MaxQuant output files ( protein groups and peptides ) are attached as Supplementary file 3 for the neuronal secretome and the neuronal surface proteome . Calculated values on which the volcanos are based are attached as Supplementary file 1 for the neuronal secretome and Supplementary file 2 for the neuronal surface proteome . The mass spectrometry secretome raw data including MaxQuant output files have been deposited at the ProteomeXchange Consortium ( http://proteomecentral . proteomexchange . org ) via the PRIDE partner repository with the data set identifier PXD003426 . Animals were deeply anesthetized with chloral hydrate ( 7% in saline , intraperitoneally ) , and perfused intracardially with PBS followed by 4% paraformaldehyde in PBS . The brains were post-fixed overnight at 4°C in the same fixative . Sections 60 µm thick were cut using a vibratome and collected in PBS . The free floating sections were incubated 15 min in methanol/PBS 1:1 containing 1% H2O2 and blocked 45 min in a 5% BSA solution in PBS . Sections were then incubated with primary antibodies in PBS containing 0 . 25% BSA , 0 . 1% Triton X-100 and 0 . 05% NaN3 for two nights at 4°C with gentle shaking . For the detection of synaptophysin , sections were incubated 90 min with biotinylated secondary antibody diluted in PBS , washed three times , and incubated 45 min with CyTM3-conjugated streptavidin . Then , sections were mounted on glass slides in PBS-glycerol 1:1 containing 0 . 1% Dapco ( 1 , 4-diazabicyclo [2 . 2 . 2 . ] octane , Sigma-Aldrich ) , and analyzed using an oil-immersion ( HCX APO 63X/1 . 40 NA ) objective coupled to a TCS SP5 confocal microscope ( Leica , Bensheim , Germany ) . An experimenter not aware of the genotype of the mice was able to identify all mutants according to the synaptophysin immunostaining detected in the CA3 subfield of the hippocampus ( Montag-Sallaz et al . , 2002 ) . For quantification of synaptophysin positve puncta , pictures of 164 x 164 µm ( 512x512 px , 320nm pixel width and height ) from the CA3 region were taken . The stratum pyramidale was delineated with straight lines and analyzed with the counting particles option of ImageJ . The average area of small synaptophysin reactive puncta ( 0 . 64817 µm2 SD of 0 . 57921 ) representing inhibitory synapses was determined in wild-type mice and used to define the threshold for large puncta ( average size of small puncta + 2x SD ) . Above this threshold ( ≥1 . 80659 µm2 ) number and size of large immunoreactive puncta , representing mossy fiber terminals , were determined for wild-type and ADAM10-deficient mice . Animals were perfused as described for the immunocytochemistry of synaptophysin . The brains were post-fixed 4 to 6 hr at 4°C in the same fixative , cryoprotected overnight in a solution of 30% sucrose in PBS , and frozen . Frontal olfactory bulb cryosections ( 60 µm thick ) were collected in PBS . The free floating sections were incubated 15 min in methanol/PBS 1:1 containing 1% H2O2 , blocked with a 2% BSA solution in PBS for 30 min , and then incubated for 2 hr with the plant lectin DBA conjugated to biotin ( Dolichos Biflorus Agglutinin , Sigma ) at 20 µg/ml in PBS and 0 . 25% Triton X-100 . For the detection of the lectin , sections incubated 45 min with CyTM3-conjugated streptavidin . After washing in PBS , sections were mounted on glass slides with antifade media ( Vectashield , Vector Labs , Burlingame , Ca ) containing 4´ , 6-diamidino-2-phenylindole ( DAPI; 1 . 5 µg/ml ) and sealed . DAPI was used for nuclear staining . An experimenter not aware of the genotype of the mice was able to identify all mutants according to the lectin staining detecting aberrantly projecting axons in the olfactory bulb ( Montag-Sallaz et al . , 2002 ) . On sections of the olfactory bulb from three animals per genotype , lectin-stained glomeruli ( KO n = 110; WT n = 126 ) were classified as compact or diffuse . In WT , 89 . 2 ± 0 . 99% of all glomeruli were compact and 10 . 8 ± 0 . 99% diffuse , whereas in KO only 58 . 2 ± 9 . 1% were compact and 41 . 8 ± 9 . 4% diffuse . P-values for Western blots and ELISA were calculated with log2 transformed relative values using a heteroscedastic , two-sided t-test . Immunohistochemical data were subjected to Analysis of Variance ( ANOVA , factor genotype ) and post hoc analysis ( Scheffe's or Fisher PLSD ) considering p<0 . 05 as significant . | Several neurodegenerative disorders , including Alzheimer’s disease , arise when protein-cutting enzymes process proteins in the wrong way . The resulting protein fragments can accumulate in nerve cells and cause them to die , leading to symptoms such as memory loss . In the case of Alzheimer’s disease the toxic protein fragment – called amyloid beta – can be produced when one enzyme cuts the amyloid precursor protein . However , the amyloid beta fragment is not made when a different enzyme called ADAM10 cuts the amyloid precursor protein first . There has been a lot of interest in finding drugs that activate ADAM10 to treat Alzheimer’s disease . However , ADAM10 also cuts other proteins on the surface of cells and it is important to know about these proteins if ADAM10 is going to be successfully targeted by a drug . To tackle this issue , Kuhn et al . have now searched for new proteins ( or ‘substrates’ ) that are cut by ADAM10 in mouse nerve cells . The experiments identified proteins that were cut in normal nerve cells , but remained unprocessed in cells where the gene for ADAM10 had been deleted . This search uncovered almost 100 new substrates of ADAM10 that were then validated using biochemical techniques . Among these substrates were many proteins that are normally anchored into the membranes of nerve cells and involved in guiding and positioning these cells in the brain so that they can connect and communicate with each other . Kuhn et al . then deleted the gene for ADAM10 only in the frontmost part of the mouse brain . This led to the nerve cells forming abnormal networks in the regions of the brain that process smells and emotions . Overall the experiments proved that ADAM10 is important not only for the prevention of Alzheimer’s disease , but also for the normal development of the brain . Future studies could now explore how stimulating ADAM10 affects the levels of its substrates . Also , a better understanding of the substrates of ADAM10 may be useful both to predict side effects of drugs that activate ADAM10 and to monitor patients who are responding well to these drugs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2016 | Systematic substrate identification indicates a central role for the metalloprotease ADAM10 in axon targeting and synapse function |
Protein concentration gradients pattern developing organisms and single cells . In Schizosaccharomyces pombe rod-shaped cells , Pom1 kinase forms gradients with maxima at cell poles . Pom1 controls the timing of mitotic entry by inhibiting Cdr2 , which forms stable membrane-associated nodes at mid-cell . Pom1 gradients rely on membrane association regulated by a phosphorylation-dephosphorylation cycle and lateral diffusion modulated by clustering . Using quantitative PALM imaging , we find individual Pom1 molecules bind the membrane too transiently to diffuse from pole to mid-cell . Instead , we propose they exchange within longer lived clusters forming the functional gradient unit . An allelic series blocking auto-phosphorylation shows that multi-phosphorylation shapes and buffers the gradient to control mid-cell levels , which represent the critical Cdr2-regulating pool . TIRF imaging of this cortical pool demonstrates more Pom1 overlaps with Cdr2 in short than long cells , consistent with Pom1 inhibition of Cdr2 decreasing with cell growth . Thus , the gradients modulate Pom1 mid-cell levels according to cell size .
In many organisms and cell types , graded protein patterns provide positional information . This is true from the smallest bacteria , where polar gradients of protein activity define the position of the division apparatus ( Kretschmer and Schwille , 2016 ) , to the largest multicellular organisms , where morphogen concentration gradients define regions of gene expression during development ( Briscoe and Small , 2015 ) . Although mechanisms of gradient formation vary , in all systems the graded patterns are thought to convey information at a distance from the source . In fission yeast , concentration gradients formed by the DYRK-family kinase Pom1 have received considerable attention , due to the role of Pom1 in regulating the timing of mitotic entry and thus cell size at division ( Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) . Pom1 gradients are nucleated at cell poles upon dephosphorylation by a type I phosphatase complex whose regulatory subunit Tea4 is delivered by microtubules ( Hachet et al . , 2011; Martin et al . , 2005; Tatebe et al . , 2005 ) . Dephosphorylation of Pom1 reveals a lipid-binding activity that maps to a 200-aa region in its N-terminal half . At the plasma membrane , Pom1 forms small clusters and is thought to laterally diffuse ( Hachet et al . , 2011; Saunders et al . , 2012 ) . It also undergoes autophosphorylation reactions that promote its detachment from the membrane , leading to the graded pattern ( Hachet et al . , 2011 ) . An interesting feature of Pom1 gradients is a noise correction mechanism that compensates for large variations in protein concentration at the cell poles , which can vary up to fourfold ( Hersch et al . , 2015; Saunders et al . , 2012 ) . In a simple diffusive gradient , the decay length ( the distance at which the concentration is reduced to a certain fraction of its amplitude ) is independent of the amplitude at the pole . In contrast , the Pom1 gradient is corrected by varying the slope of the gradient decay: gradients with higher Pom1 concentration have a steeper decay , while those with lower Pom1 concentration have a flatter decay . Two models have been proposed to explain the source of this correction . One model suggests gradient buffering is the consequence of concentration-dependent inter-molecular phosphorylation , which promotes Pom1 detachment from the membrane ( Hersch et al . , 2015 ) . This model also explains the correction for the even larger variations in levels of Tea4 concentration at cell poles . A second model hypothesizes that buffering results from concentration-dependent clustering of Pom1 , with higher concentrations leading to larger , slower diffusive clusters , causing a traffic-jam effect at the cell tips ( Saunders et al . , 2012 ) . However , direct experimental evidence testing these models is scarce . Pom1 has two physiological functions . First , Pom1 provides spatial information for division site positioning: pom1∆ cells divide off-center ( Bähler and Nurse , 2001; Bähler and Pringle , 1998 ) . Second , Pom1 provides temporal information to regulate the timing of mitotic entry: pom1∆ cells divide precociously at an aberrantly small size ( Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) . For both functions , Pom1 directly phosphorylates the SAD-family kinase Cdr2 , but on different residues ( Bhatia et al . , 2014; Rincon et al . , 2014 ) . Pom1 function in division site placement likely also involves additional substrates . To delay mitotic commitment , Pom1 phosphorylates the C-terminus of Cdr2 ( Bhatia et al . , 2014; Deng et al . , 2014 ) , which antagonizes the activating phosphorylation of the Cdr2 kinase domain by the cytosolic CaMKK Ssp1 ( Deng et al . , 2014 ) . Cdr2 localizes at the mid-cell cortex , where it forms large , stable clusters called nodes , which contain many other proteins including a second SAD-family kinase Cdr1 ( Martin and Berthelot-Grosjean , 2009; Morrell et al . , 2004; Moseley et al . , 2009 ) . The signal relay between Cdr2 and Cdr1 is not yet elucidated , but the output is an inhibitory phosphorylation of Wee1 kinase , which itself exerts direct inhibitory activity on the sole cyclin-dependent kinase CDK1 ( Kanoh and Russell , 1998; Young and Fantes , 1987 ) . In contrast to the stable Cdr1 and Cdr2 association to the nodes , Wee1 visits are only transient ( Allard et al . , 2018; Moseley et al . , 2009 ) . These visits increase in frequency and duration as cells grow , consistent with the idea that Wee1 is inactivated in longer cells to permit CDK1 activation and mitotic entry . Although genetic and biochemical evidence have firmly established Cdr2 as Pom1 substrate , there has been much debate on where the Pom1-Cdr2 interaction takes place , and whether the strength of this interaction varies in the course of a single cell cycle . Initial work proposed Pom1 gradients as a means to measure cell size , because total fluorescence measurements of Pom1-GFP along cell length revealed higher medial fluorescence in short than long cells ( Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) . This led to the model that Pom1 inhibits Cdr2 in short cells , but that inhibition is relieved upon attaining sufficient cell size , thus coupling cell size with mitotic entry . Two lines of evidence indicate that Pom1 activity on Cdr2 indeed varies with cell size . First , the levels of Cdr2 phosphorylation by Ssp1 increase during G2 , consistent with a progressive decrease in Pom1-dependent inhibition ( Deng et al . , 2014 ) . Second , the frequency and duration of Wee1 visits to Cdr2 nodes increase as cells grow , with direct evidence showing that Pom1 suppresses Wee1 visits in short cells ( Allard et al . , 2018 ) . In addition , Pom1 re-localization to cell sides , which is prominent upon glucose starvation , leads to strong mitotic delay ( Kelkar and Martin , 2015 ) . However , subsequent analyses of cortical fluorescence profiles on confocal mid-plane images failed to detect significant differences in the levels of Pom1-GFP at mid-cell between short and long cells , raising questions about the previously proposed model ( Bhatia et al . , 2014; Pan et al . , 2014 ) . The apparently invariant Pom1 mid-cell levels in cells of various lengths led to the suggestion that Pom1 may control Cdr2 activity elsewhere ( Bhatia et al . , 2014 ) , or is not involved in cell size homeostasis ( Pan et al . , 2014 ) , in agreement with the observation that pom1∆ cells retain homeostatic capacity ( Wood and Nurse , 2013 ) . Because the number of Cdr2 nodes at mid-cell increases with cell surface growth , this also led to the suggestion that Cdr2 , not Pom1 , may be the critical cell size sensor in the pathway ( Pan et al . , 2014 ) . Thus , there is currently a controversy between the invariant Pom1 levels at mid-cell and the size-dependent effect of Pom1 on Cdr2 function . In this work , we have coupled generation of a systematic pom1 mutant allelic series with a wide range of imaging methods – including single molecule super-resolution PALM ( photo-activated localization microscopy ) imaging and tracking , confocal and TIRF ( total internal reflection fluorescence ) microscopy – to obtain quantitative information on the patterning of Pom1 gradients . This enabled three major findings: first , Pom1 gradients are primarily shaped by phosphorylation-mediated detachment with clusters acting as the relevant membrane-associated unit; second , Pom1 regulates Cdr2 for mitotic entry at the mid-cell cortex; third , TIRF imaging reveals significantly higher levels of Pom1 at the mid-cell cortex in short cells .
We investigated the diffusion and membrane dissociation dynamics of single molecules of Pom1 in S . pombe cells . Cells were prepared either on a flat agarose pad and imaged horizontally along their long axis , or on a micropatterned surface imprinted with holes , where cells oriented vertically for cell pole imaging ( Figure 1A , Figure 1—figure supplement 1 ) . We relied on photoconversion , localization , and tracking of single fluorescent proteins in living cells , in our case the Pom1-mEos3 . 2 fusion expressed from the native genomic locus . Single molecules of Pom1-mEos3 . 2 were observed along the side or pole of the cell , with a higher density of localizations at the poles consistent with the known Pom1 density gradient , and tracked by monitoring their position over time to produce single molecule trajectories ( Figure 1B ) ( Manley et al . , 2008 ) . We used astigmatic imaging to ensure that we tracked only the molecules in the plane of the membrane ( see Materials and methods ) . The Pom1-mEos3 . 2 distribution along cortical profiles was indistinguishable from that of Pom1-GFP ( Figure 1—figure supplement 2 ) . To determine the dissociation rate of single molecules , we performed time-lapse imaging on horizontally oriented cells with several lag times τTL ( Gebhardt et al . , 2013 ) ( Figure 1C ) . This allows the effects of dissociation and photobleaching to be analytically separated , since varying the lag time will vary the contribution of photobleaching and therefore the effective residence time ( teff ) ( Figure 1C ) , while the actual residence time of molecules remains unchanged . Interestingly , we observed that the residence time exhibits a multimodal distribution , which could be fitted with a bi-exponential decay corresponding to short and long residence times , or fast and slow dissociation rates ( Figure 1D , Figure 1—figure supplement 3A ) . The majority of Pom1 molecules ( 76% ) comprise the fast-dissociating part , while the remaining ( 23% ) dissociate more slowly . This last population contained few molecules , but its long tail could be further explained by the presence of two or more slowly dissociating sub-populations . To understand the contribution of Pom1 auto-phosphorylation in membrane detachment , we analyzed the time-lapse data of Pom1 binding using both Pom1WT-mEos3 . 2 and kinase-dead Pom1KD-mEos3 . 2 to extract binding times ( toff ) and dissociation rates ( koff = 1/toff ) ( Figure 1E ) . Both Pom1WT and Pom1KD showed fast and slow-dissociating populations , with a similar fold difference in dissociation rates between the two populations ( 2 . 8x and 2 . 7x , respectively ) . Interestingly , when we compared the fast-dissociating populations in the two strains , the Pom1WT population dissociated 1 . 7x faster than the Pom1KD population ( tofffast= 1 . 1 ± 0 . 7 s and 1 . 9 ± 0 . 4 s for Pom1WT and Pom1KD , respectively ) . The slowly dissociating population showed a similar trend ( toffslow= 3 . 1 ± 0 . 8 s and 5 . 2 ± 1 . 1 s for Pom1WT and Pom1KD , respectively ) ( Figure 1E-F ) . We note that an even slower sub-population may exist , shown as the tail distribution in Figure 1D , but this represents a very small fraction of Pom1 molecules ( <1% ) for which we lack sufficient number of tracks to extract a reliable dissociation rate . Thus , Pom1 activity , which leads to auto-phosphorylation , promotes a faster dissociation rate of Pom1 from the membrane , in agreement with previous biochemical observations ( Hachet et al . , 2011 ) . Furthermore , both Pom1WT and Pom1KD dissociation kinetics exhibit at least two distinct populations perhaps corresponding to different multimerization states . To study the lateral diffusion of Pom1 at the plasma membrane , single fluorescent Pom1 proteins were tracked ( Figure 1B ) and analyzed to extract their diffusion coefficients . We found that the track duration teff and diffusion coefficient Deff were inversely correlated ( Figure 2A ) , with shorter tracks exhibiting faster diffusion and longer tracks exhibiting slower diffusion . Since long residence times imply a slower dissociation time , slowly diffusing Pom1 molecules present slow dissociation dynamics , and vice versa . This suggests that in larger , slowly diffusing clusters , Pom1 remains more stably associated with the membrane . Pom1WT and Pom1KD exhibit a broad distribution of diffusion coefficients ( Figure 2B ) consistent with previous measurements by fluorescence correlation spectroscopy ( Saunders et al . , 2012 ) . The distribution of Pom1WT at cell sides was shifted toward slightly faster diffusion compared with Pom1KD . To better understand this difference , we defined two thresholds at D ≥ 10−1 µm2/s and D ≥ 10−2 µm2/s , which separate molecules into three sub-populations of fast ( D ≥ 10−1 µm2/s ) , intermediate ( 10−2 ≤ D < 10−1 µm2/s ) and slow-diffusing molecules ( D < 10−2 µm2/s ) ( Figure 2B ) . Note that in this analysis , all fast-diffusing molecules are also fast-dissociating , but intermediate and slow-diffusing molecules can exhibit either fast- or slow-dissociating behaviors . Interestingly , there was a substantially higher proportion of fast-diffusing molecules for Pom1WT than Pom1KD , which also on average diffused faster ( Dmean = 0 . 31 ± 0 . 01 µm2/s for Pom1WT and 0 . 21 ± 0 . 02 µm2/s Pom1KD; Figure 2C–E ) . We note that the average diffusion rate of these three populations did not vary along cell length with distance from cell poles ( Figure 2E ) . These analyses were robust to changes in threshold choice ( Figure 2—figure supplement 1 ) . Thus , the fast population of Pom1WT diffuses faster than Pom1KD . We then performed the same analysis on Pom1WT at the cell pole using cells oriented vertically ( Figure 1A ) . Interestingly , the distribution of diffusion coefficients was shifted toward slower values compared to Pom1WT at the cell sides ( Figure 2B ) . Indeed , when using the same thresholds as above , a much smaller proportion of molecules were fast-diffusing compared with the sides ( Figure 2C; Figure 2—figure supplement 1A–C ) , very similar to the proportion observed for Pom1KD . This is consistent with Pom1 being mainly in its dephosphorylated state at the cell pole , in agreement with the presence of the Tea4-PP1 phosphatase at this location ( Hachet et al . , 2011 ) . We then tracked the evolution of the number of events along cell length starting from the cell tip . The overall number of events decreased for Pom1WT but not Pom1KD , consistent with their described localization patterns ( Figure 2F ) . Interestingly , when we considered fast , intermediate and slow populations separately , we found that Pom1WT exhibits a change in the relative proportion of molecules: the proportion of fast diffusive Pom1 remained relatively constant all along the gradient , but the proportion of intermediate and slow diffusive Pom1 decreased , resulting in a relative increase in the fast population ( Figure 2G , Figure 2—figure supplement 1E–G ) . In contrast , the proportions of fast , intermediate and slow diffusive populations of Pom1KD remain balanced all along the gradient , leading to a nearly constant ratio . In summary , these measurements provide two important insights: First , the measured binding times and diffusion rates indicate that individual Pom1 molecules cover on average a small distance before detaching from the membrane . Let’s consider the fast-diffusing molecules . These represent 4 . 5% of the population , diffuse on average at 0 . 31 µm2/s and are all part of the 76% of molecules binding the membrane for an average time of 1 . 1 s . In fact , given the inverse correlation observed between diffusion and binding times , they likely bind the membrane for an even shorter time . From these values , we can estimate a maximum travelled distance of 0 . 8 µm . Slower-diffusing molecules travel an even shorter distance before detaching . Thus , it is unlikely that individual molecules continuously track from cell pole to cell sides . Second , the shorter binding time of fast-diffusive molecules and their progressive increase in proportion at a distance from the cell pole for Pom1WT but not Pom1KD suggests the possibility this may be caused by progressive phosphorylation-dependent Pom1 detachment from larger clusters . This led us to examine more closely the mode of Pom1 attachment to the membrane and the role auto-phosphorylation plays to shape the gradient . Previous work has shown that Pom1 interacts with lipids ( Hachet et al . , 2011 ) and a fragment containing amino acids 305 to 510 can efficiently bind the plasma membrane ( Figure 3A , fragment #1 ) . A BH-search prediction ( Brzeska et al . , 2010 ) performed for this fragment identified two potential membrane binding sites ( Figure 3B ) . These two regions map to the most conserved sequences in the 305–510 fragment: the first ( aa 437–444 ) is within a 22aa-long sequence ( MB1; aa 423–444 ) that is identical in the four Schizosaccharomyces species ( S . pombe , S . octosporus , S . japonicus , and S . cryophilus ) , the second ( aa 480–492 ) falls within a weak amphipathic helix prediction ( MB2; aa 477–494 ) ( Figure 3A–C ) . To test the validity of these predictions , we constructed a series of shorter and/or mutagenized GFP-tagged Pom1 fragments integrated as single copy in wildtype and pom1∆ cells . The localization of all fragments tested was identical in wildtype and pom1∆ cells ( Figure 3D and Figure 3—figure supplement 1 ) . The Pom1 fragment spanning aa 305–510 localizes uniformly at the plasma membrane ( fragment #1 ) . Truncation of the C-terminal 20 aa ( fragment 305-490aa , fragment #2 ) , which cuts into the predicted amphipathic helix , resulted in a strong reduction ( though not complete loss; see below ) of Pom1 cortical localization . Progressive N-terminal truncations showed that a minimal 468-510aa fragment ( fragment #5 ) containing the putative amphipathic helix was sufficient for membrane binding . In this fragment , converting the hydrophobic Ile494 to the polar , uncharged Asp residue disrupted membrane binding ( I494N , fragment #6 ) , indicating that Pom1 binds the membrane through the predicted amphipathic helix . However , the same point mutation in the full 305-510aa fragment ( fragment #7 ) did not disrupt cortical localization , indicating the presence of a second lipid-binding domain . We then mutagenized seven amino acids within the conserved MB1 region to alanine ( generating the mutant allele MB1* , fragment #8 ) , which also on its own did not perturb membrane binding . However , combining both the I494N and MB1* mutations within fragment 305–510 yielded a fully cytosolic localization ( fragment #9 ) . We conclude that Pom1 localization to the plasma membrane relies on two adjacent lipid-binding motifs . To confirm the results from the fragment analysis , we introduced the same mutations in full-length Pom1 at the native genomic locus . Individual I494N and MB1* mutations led to a decrease of fluorescence intensity at the cell tip , which was exacerbated in the double mutant . Nevertheless , in this double mutant small Pom1 clusters were visible at the cell tips , likely due to Pom1 direct binding to the phosphatase regulatory subunit Tea4 through PxxP motifs ( Hachet et al . , 2011 ) . Indeed , additional mutagenesis of the five previously identified PxxP motifs in combination with the MB1* and I494N mutations rendered Pom1 entirely cytosolic . Pom1 autophosphorylation promotes membrane detachment . From over 40 phosphorylation sites identified in silico and by mass-spectrometry analysis , combined mutation of six of these sites was previously shown to abolish the Pom1 gradient ( [Hachet et al . , 2011]; note that each site contains up to three serines or threonines mutated in aggregate ) . Five of these sites are located in the 305-510aa fragment: numbers 1 , 2 , and 3 are within the conserved MB1 membrane-binding region , while 4 and 5 are located in the MB2 amphipathic helix ( Figure 3A , sites indicated by black boxes ) . The sixth one is distal to the kinase domain and was not directly investigated here . To test the contribution of multi-phosphorylation for gradient shape and buffering , we generated a series of endogenously tagged phospho-blocking Pom1 alleles , carrying alanine substitution in 1 , 2 , 3 or 5 phosphorylation sites ( Figure 4A , top row ) , and quantified Pom1 gradients by measuring the fluorescence profile at the cell cortex in medial plane confocal images . A decrease in the number of phosphorylation sites led to a gradual flattening of the gradient shape manifested in a decrease of Pom1 intensity at the cell tip and an increase at the cell middle ( Figure 4B–C , left panels ) . The gradual change of gradient shape indicates the additive nature of multiple autophosphorylation events and reveals that no particular site contributes to Pom1 gradient shape more than another . The gradient shape of the Pom15A mutant was indistinguishable from those of the previously described non-phosphorylatable Pom16A and inactive Pom1KD ( Hachet et al . , 2011 ) , indicating that these five sites represent the principal sites modulating Pom1 localization . To further probe the influence of phosphorylation on Pom1 membrane binding , we used single molecule time-lapse imaging of Pom13A-mEos3 . 2 to extract dissociation and diffusion rates . The fast-detaching population of Pom13A molecules showed membrane binding times intermediate between Pom1WT and Pom1KD ( tofffast= 1 . 7 ± 0 . 1 s; Figure 1 , Figure 1—figure supplement 3B ) . The number of tracks was not sufficient to determine the residence time of any slower-dissociating population with confidence . Thus , consistent with the fact that only some of the auto-phosphorylation sites are mutated in this allele , Pom13A molecules bind the membrane longer than Pom1WT , but not as long as Pom1KD . The distribution of diffusion coefficients was also intermediate between Pom1WT and Pom1KD ( Figure 2 ) . The thresholds defined above ( as in Figure 2B ) similarly showed a lower proportion of fast-diffusing molecules than Pom1WT but a higher proportion than Pom1KD ( Figure 2C-D , Figure 2—figure supplement 1A-C and Figure 2—figure supplement 2 ) . These data are consistent with auto-phosphorylation promoting Pom1 detachment from the plasma membrane . To assess phosphorylation-dependent gradient shape changes in a simplified system containing a single membrane-binding site , we used the Pom1MB1* mutant . This mutant binds the membrane solely with its amphipathic helix , which contains only phospho-sites 4 and 5 . Consistent with poorer membrane binding , Pom1MB1* gradient profiles showed decreased intensity compared to wildtype , at both cell poles and cell sides ( Figure 4D ) . Mutagenesis of phospho-site 5 , generating Pom1MB1*-1A , led to further gradient flattening with decreased Pom1 intensity at cell tips and increase at cell middle ( Figure 4D ) . Thus , autophosphorylation regulates each of the two membrane-binding sites . The progressive increase in medial cortical Pom1 levels in the phospho-site allelic series is consistent with the previously proposed idea that sequential phosphorylation events provide a timer function for Pom1 diffusion from cell poles ( Hachet et al . , 2011 ) . Additionally , non-phosphorylated Pom1 alleles may directly bind the membrane at the cell sides . To test the second scenario , we monitored the localization of the allelic series in tea4∆ cells , which lacks the phosphatase regulatory subunit ( Alvarez-Tabarés et al . , 2007; Hachet et al . , 2011 ) . In tea4Δ cells , all Pom1 phospho-mutants bound the cortex nearly uniformly ( Figure 4A , bottom row ) , with levels that increased with the number of phospho-site mutations ( Figure 4B–C , right panels ) , and a concomitant decrease in Pom1 cytosolic levels ( Figure 4—figure supplement 1C ) . Again , Pom15A cortical levels were indistinguishable from Pom16A , but a little lower then Pom1KD , for unknown reasons . These data are in agreement with the idea that each phosphorylation event progressively lowers membrane affinity to reduce Pom1 binding in the medial region . We note that changes in Pom1 distribution in the phospho-site mutants are not due to a change in Pom1 protein concentration , as verified by western blot analysis ( Figure 4—figure supplement 1B ) . Comparing Pom1 medial cortical levels in WT and tea4∆ cells showed significantly lower amounts for Pom1 and Pom11A in tea4∆ , but higher or similar levels for Pom12A , Pom13A and Pom15A . This suggests that the flatter gradients observed in the Pom1 phospho-mutants are not only due to a reduction of Pom1 detachment from the membrane , allowing lateral diffusion over a longer distance , but also to an increase in Pom1 attachment to the membrane at mid-cell . Thus , Pom1 auto-phosphorylation both favors its detachment from , and prevents its attachment to , the membrane . A key feature of the Pom1 gradient is its robustness to variations within the system . Previous work showed that variability in Pom1 concentration at cell poles is counteracted by varying the gradient decay length , which leads to a strong negative correlation between decay length and Pom1 amplitude at the pole ( Hersch et al . , 2015; Saunders et al . , 2012 ) . Pom1 clustering and Pom1 inter-molecular phosphorylation have both been theoretically proposed as source for this correction , though we so far lack experimental evidence . We assessed the contribution of the multi-phosphorylation reaction in gradient buffering by plotting the correlation between decay length and amplitude at the pole in Pom1 phospho-mutants that retain significant pole enrichment . Pom11A , Pom12A and Pom13A all showed a loss of negative correlation of decay length to Pom1 concentration at the cell tip ( Figure 4E ) . Thus , these Pom1 mutants poorly correct intrinsic variations of Pom1 concentrations at the cell tips . Another evidence for buffering comes from the steep decrease in the coefficient of variation of Pom1WT from the cell tip to the cell middle ( Hersch et al . , 2015 ) . While we confirm this observation with our current data set , the decrease in the coefficient of variation at cell tips and cell sides is much smaller for Pom11A , Pom12A and Pom13A ( from 9 . 6-fold for Pom1WT to 6 . 6 for Pom11A , 5 . 8 for Pom12A , and 4 . 2 for Pom13A; Figure 4F , Figure 4—figure supplement 1D ) . We conclude that the Pom1 phosphorylation cycle directly contributes to Pom1 gradient shape robustness . One important physiological role of Pom1 is to set cell size at division by negatively regulating the SAD-family kinase Cdr2 , which forms stable cortical nodes at mid-cell ( Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) . Consistent with previous reports that Pom16A and Pom1 overexpression lead to cell size increase at division ( Hachet et al . , 2011; Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) , we observed a gradual increase in cell length at division in mutants of the phospho-site allelic series ( Figure 5A ) . Cell length at division was strongly correlated with the medial cortical Pom1 levels ( Figure 5B ) , consistent with the idea that medial Pom1 levels set the cell division size . To evaluate the relative contributions of cytosolic and cortical Pom1 in Cdr2 inhibition , independently of gradient formation , we obtained the same measurements in tea4∆ cells: the progressive increase in cortical Pom1 in Pom1 phospho-mutants also correlated with an increase of cell length in tea4Δ background ( Figure 5A–B ) . We note that tea4∆ cell lengths were slightly longer than WT for pom1 alleles containing up to three phosphosite mutations . By contrast , the cytosolic Pom1 signal in the allelic series was largely invariant in WT and showed a modest decrease in tea4∆ leading to an inverse correlation with cell length at division ( Figure 5—figure supplement 1A ) . The correlations between cell length and Pom1 levels at cell poles ran in opposite directions in WT and tea4∆ cells ( Figure 5—figure supplement 1B ) . We conclude that the medial cortical pool of Pom1 is the relevant pool for cell size regulation . Because Pom1 clusters rather than individual molecules may shape the gradient ( see Figure 1 above ) , and inspired by recent work showing visits of Cdr2 nodes by the downstream Wee1 kinase ( Allard et al . , 2018; Gerganova and Martin , 2018 ) , we turned to live cell TIRF imaging . To test the method’s sensitivity and selectivity for cortical signals , we first compared the fluorescence levels of Pom1WT , Pom11A , Pom13A and Pom15A phospho-mutants , which reproduced the increased Pom11A , Pom13A and Pom15A medial cortical localization seen by confocal microscopy ( Figure 5C ) . Specific measurements of cluster size and intensity also showed higher values for Pom13A and Pom15A ( Figure 5—figure supplement 2 ) . We then imaged two cytosolic proteins . First , cytosolic GFP , expressed under the pom1 promoter , could not be detected by TIRF though it was seen by epifluorescence , confirming that the evanescent field detects only cortical molecules ( Figure 5C , Figure 5—figure supplement 1C ) . By contrast , Pom1-GFP in tea4∆ cells , which by confocal microscopy is not detected at the cortex ( see Figure 4A; Hachet et al . , 2011; Kokkoris et al . , 2014; Padte et al . , 2006 ) , revealed a cortical signal by TIRF imaging ( Figure 5C ) . In tea4∆ , Pom1 formed cortical clusters , though cluster number , size and intensity were lower than in wild type ( Figure 5D ) , which likely explains why the cortical Pom1WT signal in tea4Δ is virtually indistinguishable from the cytosolic signal by confocal microscopy . Thus , very transient encounters of Pom1 at Cdr2 nodes at the cortex , rather than a fully cytosolic Pom1 , could account for the longer size of tea4∆ cells noted above . We conclude that TIRF provides a highly specific and sensitive imaging setup for Pom1 clusters at the yeast cortex . In TIRF timelapse imaging , Cdr2-GFP formed stable nodes at mid-cell , which did not move substantially over 60 s ( Allard et al . , 2018 ) , whereas Pom1-GFP clusters were highly dynamic ( Figure 5E ) . Fast 100 ms acquisition intervals were used to monitor Pom1 clusters , which exhibited an array of dynamics with clusters moving laterally , splitting or merging . Figure 5F and Videos 1–2 provide representative examples . Example one shows an initial large cluster that splits in two at 3 . 0 s , remerges , and splits again at 4 . 4 s . Clusters 4 and 5 provide a second example of a small cluster moving laterally and merging with a larger one . Another common cluster dynamic is exemplified by clusters exhibiting dimming and/or dispersal of signal , followed by an immediate increase in fluorescence . This behavior can be seen in example 2 between times 0 . 0 s and 0 . 3 s , 1 . 8 s and 2 . 1 s , and 2 . 2 s and 2 . 5 s , the latter one dimming to only a few detectable fluorescent pixels . Note that the detection threshold was set at a value higher than the fluorescence levels measured in a strain not expressing GFP ( see Materials and methods ) . Similarly , example three dims and regains fluorescence between 2 . 1 s and 2 . 6 s , and 3 . 2 s and 4 . 0 s . It then completely disappears at 7 . 0 s . There were also many instances where the Pom1 signal was too fluid to unambiguously follow individual clusters over time . These fluctuations in the fluorescence signal of individual clusters indicate that clusters often recombine and can gain and loose individual Pom1 molecules over time . Similar dynamics with oscillatory intensities were observed in the Pom13A and Pom15A mutants ( Videos 3 and 4 ) . To estimate the lifetime of individual clusters , we followed 38 Pom1 clusters in 12 cells . The cluster lifetime from appearance to disappearance or splitting of the cluster ranged to over 8 s , the length of the imaging timeframe ( Figure 5G ) . This value is very likely to be under-estimated as the longest-lived clusters were present from start to end of the time-lapse imaging . These values are higher than previous measurements obtained by confocal microscopy ( Saunders et al . , 2012 ) , probably because of the higher sensitivity of TIRF imaging . They are also substantially longer than the binding time obtained by PALM imaging for individual molecules ( around 1 to 3 s , see Figure 1 ) . These observations further support the idea that individual Pom1 molecules turn over within single clusters . Therefore , we propose that Pom1 clusters are the functional units shaping the gradient , as their longer residence time would permit diffusion from the cell pole all the way to the zone of action at mid-cell . To investigate the Pom1-Cdr2 interaction at the cortex , we acquired Pom1 TIRF images at 1 s interval for 60 s and took snapshot TIRF images of Cdr2 at the start and end of the imaging period in a dual tagged Pom1-GFP Cdr2-tdTomato strain ( Figure 6A ) . The Cdr2 snapshots provided the location of nodes to which we mapped individual regions of interest ( ROIs ) , in which we quantified Pom1 intensity in the GFP channel . Indeed , we were able to observe Pom1 encounters of the Cdr2 nodes of various duration within the 60 s imaging period ( Figure 6A–B ) . To distinguish whether these encounters are targeted visits or due to random collisions of Pom1 clusters , we shifted the same size ROIs away from , but in the immediate vicinity of , Cdr2 nodes . We observed a very similar pattern for Pom1 in the non-node-associated ROIs , which suggests that laterally moving Pom1 clusters randomly collide into Cdr2 nodes ( Figure 6C ) , distinct from the targeted visits from the cytosol reported for Wee1 ( Allard et al . , 2018 ) . Remarkably , when we clustered the data according to cell length , the average Pom1 intensity at all measured Cdr2 nodes was significantly higher in short ( 6 to 8 µm ) than long ( 12 to 14 µm ) cells in three independent experimental repeats ( Figure 6D , Figure 6—figure supplement 1A–B ) . The values for cells of intermediate length ( 9 to 11 µm ) were more variable , probably depending on the average length of these cells . We observed a similar pattern in the duration of Pom1-Cdr2 encounters , which we defined as the length of time the Pom1-GFP value at a Cdr2 node remained above a defined fluorescence threshold: In short cells , a higher proportion of Cdr2 clusters were continuously occupied by Pom1 over the 60-s imaging period ( Figure 6E , Figure 6—figure supplement 1C ) . Consistent with the observation that Pom1 behavior is similar at and in the vicinity of Cdr2 nodes , measuring the total medial Pom1 TIRF signal gave similar results: Pom1 levels were higher in the middle of short than long cells ( Figure 6F–G ) . Thus , the medial cortical Pom1 levels , which are those critical for Cdr2 regulation , decrease as cells grow , consistent with cell length-dependent relief of Pom1 inhibition .
The Pom1 gradient forms upon local dephosphorylation of Pom1 at the cell pole , promoting Pom1 membrane binding ( Hachet et al . , 2011 ) . Previous modeling work demonstrated that the multi-phosphorylation reaction Pom1 undergoes prior to its membrane detachment could provide a timer function for Pom1 lateral diffusion and serve as a buffering mechanism for the gradient ( Hersch et al . , 2015 ) . An alternative model proposed that differential diffusion coefficients of Pom1 clusters of varying sizes underlie the buffering of the gradient with the largest clusters causing a ‘traffic jam’ event at the cell tip , which is relieved by the progressive fractionation and the subsequent increase in diffusion coefficients of smaller clusters away from the gradient source ( Saunders et al . , 2012 ) . The data presented here integrates these two mechanisms in a single model for Pom1 gradient formation . Single molecule measurements by PALM imaging show that the gradient is comprised of molecules with a wide distribution of diffusion coefficients . Their dissociation dynamics also revealed at least two populations with distinct binding times . Note that the fast-dissociating population is not identical to the fast-diffusing one . However , there is an overall inverse correlation between binding times and diffusion rates , such that fast-diffusing molecules are also binding the membrane only for a short time . These may represent individual Pom1 molecules not associated with a cluster or in small clusters , and are a minority of all Pom1 molecules . The slower-diffusing molecules may be part of larger clusters . Many of these molecules detach from the membrane slower , but a large pool also exhibits fast dissociation , perhaps indicating peripheral association with the cluster . Importantly , the diffusion coefficients and binding times reveal that individual molecules ( in a cluster or not ) will only travel a maximum distance of 0 . 8 µm before detaching from the plasma membrane . Thus , a Pom1 molecule binding at the cell tip upon dephosphorylation by the Tea4-PP1 complex travels only a short distance and does not reach the cell middle . By contrast , Pom1 clusters as a whole are longer lived at the plasma membrane . Clusters appeared very dynamic , changing intensity over time , splitting , dispersing and fusing again . Importantly , TIRF measurements easily identified cluster lifetimes of over 8 s . This longer lifetime of clusters at the plasma membrane indicates that individual molecules exchange within a cluster . Specifically , this suggests that existing clusters must be able to bind Pom1 molecules directly from the cytosol . In TIRF imaging , we indeed found examples of isolated clusters whose fluorescence intensity diminished before increasing again , providing support for this idea . These longer lifetimes are consistent with clusters being able to form at the cell pole and diffuse laterally all the way to mid-cell , at least in short 7–8 µm cells . Thus , we propose that Pom1 clusters are the functional units that shape the graded distribution of Pom1 . The phospho-site mutant allelic series shows that Pom1 distribution critically depends on auto-phosphorylation . Indeed , progressive alanine substitution at up to five auto-phosphorylation sites causes progressive flattening of the Pom1-graded distribution , with apparently additive contribution of each phosphorylation event . Because the five phosphorylation sites all map very close to each other within two adjacent membrane-binding motifs , it is likely that they have to be phosphorylated sequentially , extending the time frame between the dephosphorylation event taking place at the cell pole and the full auto-phosphorylation , promoting membrane detachment . Each phosphorylation event reduces the affinity to the plasma membrane , likely affecting both koff and kon . Indeed , our PALM data shows that the koff of single Pom1 molecules is modulated by their phosphorylation status: Pom1KD , which is not phosphorylated binds the plasma membrane longer than the partly dephosphorylated Pom13A , which itself binds longer than Pom1WT . Similarly , alanine-substitutions of phospho-sites led to a progressive increase in the membrane-associated Pom1 fraction in tea4∆ cells , with a fold-change larger than the one measured for the koff . This suggests that , although not directly measured , the kon of Pom1 to the plasma membrane is also modulated by phosphorylation , which decreases the membrane association rate . Phosphorylation may also modulate cluster formation . Clusters formed in all the Pom1 mutant alleles studied here , and exhibited a similar wide range of diffusion rates in Pom1WT , Pom13A and Pom1KD , suggesting – if we make the validated assumption that cluster size influences diffusion rate ( Saunders et al . , 2012 ) – a similar range of cluster sizes . Thus , dephosphorylated Pom1 efficiently forms clusters . Phosphorylated Pom1 may form clusters less efficiently , as Pom1WT forms smaller clusters in tea4∆ cells . We note that it is unclear whether Pom1 is fully phosphorylated in this mutant , or whether it may be inefficiently dephosphorylated by the still present PP1 catalytic subunit lacking the Tea4 regulatory subunit . Previous data had indeed shown that Pom1 dephosphorylation does not strictly require Tea4 ( Kokkoris et al . , 2014 ) . In either case , the data indicate that dephosphorylation at the cell pole also promotes clustering , which may be further favoured by the direct binding with Tea4 ( Hachet et al . , 2011 ) . We note that it may be impossible to fully dissociate cluster formation from membrane binding . The comparison of Pom1WT and Pom1KD reveals three important differences . First , the diffusion rates of Pom1KD at cell sides were similar to those of Pom1WT at cell poles , consistent with local dephosphorylation of Pom1 at this location . Second , on cell sides the fast pool of Pom1WT molecules was more abundant and diffused faster than Pom1KD . This suggests that , by reducing the electrostatic interaction with neighbouring phospholipids , auto-phosphorylation promotes faster mobility of individual Pom1 molecules . Finally , the proportion of single Pom1 molecules increased with distance from cell poles in Pom1WT but not Pom1KD . These observations are consistent with progressive phosphorylation-induced dissociation of Pom1 from the membrane and/or clusters . From these data , we propose a revised model on how the specific gradient shape is achieved . Localized Tea4-PP1 phosphatase activity at cell poles dephosphorylates Pom1 , revealing two membrane-binding domains – an amphipathic helix and an adjacent positively charged region – both of which permit the association of Pom1 with the plasma membrane . Pom1 dephosphorylation also favors the formation of clusters at the membrane , which helps carry Pom1 over a longer distance . Within a cluster , individual Pom1 molecules have a short lifetime of 1 to 3 s on average , but new Pom1 molecules join from the cytosol . As clusters split and merge , this permits the transport of single molecules from cluster to cluster . The clusters may diffuse at different rates , either because they are of different sizes and therefore contain different numbers of membrane-binding sites , or because each membrane binding site may be differently phosphorylated and thus bind the membrane with distinct affinity . Indeed , as Pom1 is active on itself , aging clusters become more phosphorylated , which promotes dissociation from the cluster and from the membrane to single molecules . Even though single molecules are present at the membrane a long distance from the poles , their binding is short-lived , detaching from the membrane within 1 s . Thus , clusters act as a diffusion unit , whose regenerating capacity is reduced via progressive auto-phosphorylation with time/distance from the cell pole . One physiological function of Pom1 is to prevent the activation of Cdr2 kinase . Our data clearly establish that the meaningful levels of Pom1 are those at the mid-cell cortex , where Cdr2 forms nodes . Indeed , we see a strong correlation between cell size at division ( the phenotypic outcome of Cdr2 regulation ) and Pom1 medial levels in both WT and tea4∆ cells . By contrast , the correlations between cytosolic or cell pole levels of Pom1 and cell length are inconsistent in WT and tea4∆ cells , excluding the alternative models that Pom1 may act on Cdr2 at cell poles ( Bhatia et al . , 2014 ) or in the cytosol . These data agree with the extended size of cells overexpressing Pom1 or mis-targeting it to the medial cortex ( Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) . They also agree with the strong cell lengthening effect of naturally re-distributed Pom1 upon glucose starvation ( Kelkar and Martin , 2015 ) . Our new observation that tea4∆ cells have significant cortical Pom1 signal also fits with the idea that Pom1 remains active on Cdr2 in this mutant as manifested by the longer size of tea4∆ than pom1∆ ( this work and Martin and Berthelot-Grosjean , 2009 ) . In this mutant , abundant cytosolic Pom1 likely allows substantial stochastic encounter with , and binding to , the plasma membrane . We note that the phospho-mutant Pom1 alleles did not cause noticeable changes in the position of the division site . This is in agreement with the finding that the two functions of Pom1 in controlling positioning and timing of division are separable ( Bhatia et al . , 2014 ) . This also indicates that a graded Pom1 distribution is less important for division site placement , consistent with observations in S . japonicus ( Kinnaer et al . , 2019 ) . Although Pom1 mid-cell levels clearly correlate with cell size at division in various mutant and environmental conditions , a key unresolved and debated question has been whether Pom1 levels vary during the growth of a single cell , and therefore whether Pom1 may contribute to cell size homeostasis . The higher sensitivity and specificity of TIRF imaging now allows to answer this question unequivocally: there is more Pom1 at the mid-cell cortex of small than long cells . We considered the possibility that this higher concentration may be a simple consequence of cell extension . However , a fission yeast cell doubling in length increases its volume about 1 . 1-fold more than its surface . Thus , considering a constant Pom1 concentration and an invariant average gradient shape ( Bhatia et al . , 2014; Saunders et al . , 2012 ) , a higher number of Pom1 molecules have less membrane space at their disposal , which would lead to higher Pom1 concentration at mid-cell in longer cells . Because we observe the opposite , the increased mid-cell levels of Pom1 in short cells must be due to Pom1 gradients extending into mid-cell of small but not long cells . The diffusion rates and cluster lifetimes we have measured are consistent with this scenario . These may be tuned to allow Pom1 diffusion over 3–4 µm distance , sufficient to reach the middle of short , but not long cell , which underlies the difference in Pom1 mid-cell cortical levels in cells of different sizes . Thus , the graded Pom1 distribution is able to convey information in a manner dependent on distance from the source . These findings poise Pom1 to function as a sensor of cell dimension that provides more inhibition on Cdr2 in short than long cells . This proposition is consistent with biochemical data that Cdr2 activating phosphorylation , which Pom1 counteracts , increases with cell growth ( Deng et al . , 2014 ) and that Cdr2 is more active in long than short cells , as measured by the number of visits by the Wee1 kinase ( Allard et al . , 2018 ) . It is also consistent with Pom1 function being exquisitely dose-dependent , both in terms of global levels ( Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) and specifically at the mid-cell cortex ( this work ) . Pom1 likely exerts its inhibitory action primarily on short cells , when its medial concentration is highest . Indeed , blocking Pom1 activity leads to precocious mitosis in short cells lacking other forms of length control , but not in otherwise wildtype cells ( Martin and Berthelot-Grosjean , 2009 ) . Pom1 may not be the only sensor protein in the pathway: Cdr2 was also proposed to monitor cell surface area through a dynamic exchange of molecules between the cytoplasm and the plasma membrane to form medial nodes whose numbers increase with cell growth ( Facchetti et al . , 2019; Pan et al . , 2014 ) . Thus , cell growth , by both promoting a decrease in Pom1 levels and an increase in the number of Cdr2 nodes at mid-cell , enhances the activation of Cdr2 in long cells . As mutants in both Pom1 and Cdr2 retain cell size homeostasis , cells likely have secondary sizer mechanisms , perhaps monitoring different geometrical quantities ( Facchetti et al . , 2019; Wood and Nurse , 2013 ) .
Standard methods of S . pombe culturing and genetic manipulations were used . For PALM imaging , S . pombe cells were grown in rich yeast extract ( YE ) medium and imaged during the exponential growth at an OD600 comprised between 0 . 4 and 0 . 7 . All live-cell imagings were performed on YE 2%-agarose pads . For all other imaging experiments strains were grown at 25°C in fully supplemented synthetic Edinburgh minimal medium ( EMM ) . A complete list of all used strains is provided in Supplementary file 1 . For the truncation analysis ( Figure 3D ) , Pom1 fragments were amplified by PCR and cloned under the control of the pom1 promoter in single integration vectors ( kind gifts from Dr . Aleksandar Vjestica ) . The vectors were linearized and integrated at the ura4 locus in wild type and pom1Δ strains . A list of plasmids used in this study is provided in Supplementary file 2 . The I494N mutation was introduced by site-directed mutagenesis . To generate the MB1* allele , we made use of a native restriction enzyme site ( BglII ) at aa426 to replace fragment 426–510 with one in which aa429-436 within the conserved region ( MB1 ) were replaced by seven alanines . This mutated fragment was amplified with a forward primer annealing from aa436 onward , and carrying an overhang for the seven alanines and the BglII restriction site . Primer sequences used for mutagenesis are listed in Supplementary file 3 . To generate pom1 alleles at the native genomic locus ( Figure 3E ) , pSM2142 ( containing pom1 promoter , ORF fused in frame to GFP , kanMX and pom1 3’UTR ) was used as a backbone for site-directed mutagenesis , except for the pom1MB1* allele , for which the seven residues were changed to alanine in three rounds of site-directed mutagenesis . To generate the MB1* I494N 5PxxP* mutant , a sequence including the sequence coding for aa81-524 ( BspEI to MluI ) containing all mutations ( as above and in Hachet et al . , 2011 ) except PxxP sites 4 and 5 was ordered as a synthetic gBlocks from LubioScience GmbH and introduced in replacement of the wildtype fragment in plasmid pSM2142 . PxxP sites 4 and 5 were then introduced by site-directed mutagenesis . The plasmids were digested and transformed into a pom1Δ::ura4+ background . The phosphosite mutant allelic series ( Figure 4A ) was generated through site-directed mutagenesis of the sites on a pREP41 plasmid , carrying the full length Pom1 sequence ( pSM738 ) . The mutagenized plasmid was digested and transformed in a strain in which ura4+ replaced the pom1 sequence coding for aa400-1006 ( pom1∆ ( aa400-aa1006 ) ::ura4+ ) . Integrants were selected on 5-FOA . Each pom1 allele was subsequently tagged with GFP-kanMX through transformation with linearized pSM1731 , a plasmid containing the end of pom1 ORF without stop codon fused in frame to GFP , the kanMX selection marker and pom1 3’UTR . For the Pom1MB1*-1A ( 5 ) mutant , phosphosite five was mutagenized through site-directed mutagenesis on pSM2237 ( Pom1MB1*-GFP ) to generate pSM2264 , which was digested and transformed into a pom1Δ::ura4+ strain as above . All generated strains were verified via sequencing . Pom1-mEos3 . 2 , Pom1KD-mEos3 . 2 , and Pom13A-mEos3 . 2 strains were generated for PALM microscopy using a standard PCR-based approach ( Bähler et al . , 1998; Laplante et al . , 2016 ) and verified by PCR . BH-search prediction performed at https://hpcwebapps . cit . nih . gov/bhsearch/ with window size for residue averaging of 15 amino acids and values for amino acids set to standard BH parameters . For cell pole imaging , S . pombe cells were vertically immobilized on a YE-2% agarose pad obtained by imprinting on an epoxy resin mold containing an array of micro-pillars ( Wang and Tran , 2014 ) with diameter of 6 μm and height of 20 μm ( Figure 1—figure supplement 1 ) . SU-8 photolithography and PDMS lithography were performed at the EPFL Center of Microtechnology . SU-8 ( GM1060 Gersteltec Sarl ) was spin-coated onto a silicon wafer with a thickness of 20 μm and then baked at 95°C for 40 min . The wafer was then gradually cooled from 95°C to 30°C during 30 min . SU-8 polymerization was induced by exposure to 350 nm light for 10 s through a quartz mask containing disk patterns of 6 μm diameter . Post-photolithography baking was performed at 95°C for 40 min , followed by gradual cooling for 30 min . SU-8 was then developed with manufacturer-provided SU-8 developer and cleaned with isopropanol and dried with compressed air . The SU-8 substrate was hardened by baking at 150°C for 30 min . PDMS ( Sylgard 184 silicone base , Sylgard curing agent ) was then poured onto the surface of the SU-8 mold , which was pre-treated with trimethylchlorosilane vapor ( TMCS 33014 from Sigma ) to render it non-stick . The PDMS was then polymerized during at least 2 hr at 80°C . The resulting PDMS mold was then used to create the final epoxy resin mold . Epoxy resin R123 bisphenol and epoxy hardener R614 ( Soloplast Vosschemie ) were poured onto the PDMS substrate and polymerized for 24 hr . S . pombe cells were genetically modified to express the photoactivable fluorophore mEos3 . 2 fused to Pom1 protein at the endogenous genomic locus ( Laplante et al . , 2016 ) . To measure diffusion dynamics , imaging was performed on a previously described custom-built microscope ( Holden et al . , 2014 ) . Cells were imaged in two channels: the fluorescence channel for precise tracking of single Pom1 molecules and the phase contrast channel for cell segmentation and determination of the membrane plane . In order to selectively image Pom1 at the membrane , we used astigmatic imaging to encode the axial position of single molecules ( Huang et al . , 2008 ) . This leads to a change in the shape of the point spread function of molecules not in focus ( from circular to elongated ) , allowing to discard them from the analysis . In a post-processing step , we analyzed single molecules which were at axial positions ranging from −200 nm to +200 nm ( with 0 corresponding to the plasma membrane position ) , eliminating signal detected off the membrane . A z-calibration was performed by taking sequential images of 0 . 1 µm fluorescent beads ( Invitrogen TetraSpeck ) , displacing the objective by 20 nm steps over an axial range of 1 µm . This calibration allows us to define the PSF widths in the x and y ( lateral ) directions as a function of the z ( axial ) displacement . The imaging was performed with an NA 1 . 49 oil immersion objective lens ( Nikon ) , and fluorescence was detected using an Evolve 128 EMCCD camera ( Photometrics ) with a 20 ms integration time . Fluorescence was excited with a 560 nm laser ( MPB VFL-P-300–560 ) using an irradiance of 4 kW/ cm2 . Molecules of mEos3 . 2 were photoconverted using a 405 nm laser ( Coherent OBIS ) with an irradiance of ~0–16 W/ cm2 . To measure dissociation , Pom1-mEos3 . 2 molecules were photoconverted by a pulse of 405 nm and then imaged continuously ( no time-lapse +20 ms exposure time ) or with the time-lapse sequences ( time-lapse durations of 100 + 20 ms or 200 + 20 ms , Figure 1c ) with 560 nm light until complete bleaching of the photoconverted molecules before the cycle was repeated . Imaging of cells were performed until no more activation of Pom1-mEos3 . 2 was observed . Brightfield illumination for phase contrast imaging was performed with a white LED ( Thorlabs MCWHL2 ) , passed through a green filter ( Chroma ET525/50mc ) , focused into the back focal plane ( BFP ) of a condenser lens ( Nikon MEL56100 ) . This channel was equipped with a 1024 × 768 CMOS camera ( The Imaging Source DMK 31BU03 ) . Image acquisition was controlled through Micro-manager . The images were processed with a custom-made ImageJ plugin to segment the single molecule data . Briefly , this consisted of generating a maximum intensity projection ( MIP ) of all spots , dilating and binarizing the MIP to make a cellular mask and then multiplying this mask by the raw image data . This ensures signal exclusively within cells is analysed and eliminates spurious noise occasionally detected outside of a cell . 3D molecule localization was then performed with RapidSTORM 3 . 3 ( Wolter et al . , 2012 ) software , with a z calibration ( PSF X and Y width versus Z ) used as an input . Single molecules which had an SNR above 50 were localized . Subsequently , a wobble distortion was corrected using an open source Matlab script ( Carlini et al . , 2015 ) ; this eliminated lateral shifts in localizations above and below the focal plane . Before tracking , single molecules were filtered based on their integrated intensity and axial position to ensure that only molecules on the membrane were tracked . Localized molecules were then tracked using a MATLAB-based routine based on Crocker and Grier ( 1996a ) : molecules belonged to the same track if they were within a 320 nm radius within consecutive frames . No gaps within tracks were permitted . The distance of each track’s first point was calculated relative to the cell pole , which was determined from the phase contrast image . First , the phase contrast channel was aligned to the fluorescence channel from images of 500 nm fluorescent beads ( Invitrogen TetraSpeck ) using a custom-written MATLAB program . Briefly , bead images were localized in 2D and then used to define a rigid transform using a custom MATLAB script . This transformation was then used to map the fluorescence channel onto the phase contrast channel . Second , the center-line of the cell were defined manually using MATLAB’s’ imline’ function , with line extremities corresponding to the poles x , y positions . The tracking , dissociation and diffusion analysis was performed with custom MATLAB script to extract and study the diffusion and dissociation dynamics . The recorded molecule trajectories provide an apparent dissociation rate ( keff ) which is the result of both the molecule photobleaching ( kbleach ) and the actual dissociation rate ( koff ) contributions . By performing time-lapse imaging at time-lapse periods ( τTL ) of 20 , 120 and 220 ms , we extracted the dissociation constant of Pom1 wide type or mutants . First , the apparent dissociation rate ( keff ) is extracted by fitting of the exponential distribution of the track lengths ( teff ) for every time-lapse experiment ( Figure 1D; Figure 1—figure supplement 3A ) ( 1 ) f ( t ) =Ae-keff×teffwhere the track length is defined as the number of sequential localization events ( n ) spatially separated by less than 321nm multiplied by the time-lapse period τTL , teff= ( n-1 ) ×τTL , and τTL is the sum of the camera integration time ( τint– equal to 20ms in our experiment ) and the time delays introduced within a pair of two consecutive images . This distribution is the result of the sum of two independent Poisson processes: the photobleaching that occurs only under laser exposure ( i . e . during τint ) and the dissociation of Pom1 that can occur any time during the τTL period . ( 2 ) keff×τTL=kbleach×τint+koff×τTL Thus , Equation 2 can be rewritten as y ( τTL ) =q+m×τTL where koff=m ands kbleach=q/τint are easily extracted from the slope and the intercept of the linear fit of keff×τTL versus τTL data . We performed a weighted linear fit where we assigned to each keff value a weight proportional to the S . D . extracted from the fit of the exponential distribution . The emerging of deviation from a linear trend in the log plot of the distribution of the track lengths indicated the possible presence of transient binding with different kinetic constants . Assuming the presence of at least two populations , one with a faster off-rate constant and one with a slower one , the model describing the dissociation of Pom1 then becomes:f ( t ) =Io , aexp ( − ( kbleachτintτtl+koff , 1 ) t ) teff<TLf ( t ) =Io , a ( 1−Io , b ) exp ( − ( kbleachτintτtl+koff , 2 ) t ) tmax<teff<TLwhere Io , a is the number of molecules at start , t is real time in seconds and Io , b represents the percentage of Pom1 molecules exhibiting off-rate constant koff , 2 . Diffusion coefficients were calculated for each track essentially as described in Manley et al . ( 2008 ) . The individual protein diffusion coefficient ( D ) is extracted from tracks containing more than five consecutive localizations without any gap between localizations using MSD analyser , a MATLAB-based package ( Tarantino et al . , 2014 ) . D is derived from the MSD distribution of a Brownian particle’s trajectories parameterized through the Einstein–Smoluchowsky equation MSD=2dD∆t where d is the number of dimensions of the trajectory data ( d = 2 in this work , since we consider diffusion at the membrane ) and ∆t is the time lag over which the MSD is measured . D can thus be extracted from the slope of linear fit of the first 25% of the mean MSD curve ( Crocker and Grier , 1996b ) . The equation t=x22D was used to derive diffusion distances from diffusion coefficients . Confocal microscopy ( Figures 3 and 4 ) was performed on an inverted DMI4000B Leica microscope equipped with an HCX Plan Apochromat 100x/1 . 46 NA oil objective and an UltraVIEW system ( Perkin Elmer; including a real-time confocal scanning head CSU22 from Yokagawa Electric Corporation ) , solid state laser lines , and an electron-multiplying charge-coupled device camera ( C9100 , Hamamatsu Photonics ) . Medial section images were obtained at 300 ms exposure time at 100% laser power for five consecutive time points at maximum speed with sum image projections used for quantification and figure preparation . The same settings were used to compare the distributions of Pom1-GFP and Pom1-mEos3 . 2 . For cell length measurements , cells were stained with calcofluor and imaged with a Leica epifluorescence microscope ( 60X magnification ) . TIRF microscopy ( Figures 5 and 6 ) was performed on a DeltaVision OMX SR imaging system , equipped with a 60 × 1 . 49 NA TIRF oil objective ( oil 1 . 514 ) , an illumination pathway for ring-TIRF and a front illuminated sCMOS camera size 2560 × 2160 pxl ( manufacturer PCO ) . Imaging settings were: 512 × 512 pxl field of view , 21 ms exposure time , laser power of 20% with TIRF angles: 488 nm at 86 . 9° and 568 nm 83 . 2° . Samples were placed on a 0 . 17 ± 0 . 01 mm thick glass slide and imaged within 15 min . Imaging was performed in two modes: every second over a 60-s imaging period and every 100 ms over an 8-s imaging period . A widefield image of the medial plane of cells , expressing cytosolic GFP ( kind gift from Dr . Magdalena Marek ) mixed with a Cdr2-GFP strain ( Figure 5—figure supplement 1C ) was taken as a single snapshot on the same imaging system , using light path setting conventional , 21 ms exposure time , 20% laser power , after which the cortical plane of the same field of view was imaged in TIRF . Cortical gradient profiles were quantified from confocal microscopy image sum projections by manually drawing a line along the cortex of the cell from the cell tip to the cell middle , generating four gradient profiles per cell in ImageJ ( NIH ) . For the phosphosite mutant allelic series ( Figure 4B ) , 240 gradient profiles per strain were generated ( n = 3 experiments , 60 cells ) . 160 gradient profiles were generated for Pom1MB1* and Pom1MB1*-1A ( 5 ) ( Figure 4D ) from n = 2 experiments , 40 cells . The profiles were aligned to gradient maximum intensity value at the cell tip , averaged per strain and plotted against distance from the cell pole . The three individual experiments are shown in Figure 4—figure supplement 1 . For the quantification of Pom1 intensity levels at cell middle , the average value over the medial-most 1 . 5 μm of the gradient tails of each individual gradient was calculated and presented as boxplots ( Figure 4C ) . We developed a simple MATLAB script to extract decay length ( Suppl . Info 1 ) . Briefly , all 240 gradient profiles were aligned to the maximum value at the cell pole and smoothed with a Gaussian filter as previously described ( Hersch et al . , 2015 ) . The profiles were binned by 5% and the first 0 . 5 μm of the profile values were deleted to avoid the effect of the gradient plateau at the cell pole . The decay length was obtained as the slope of the linear regression on the log of the binned average profiles and plotted against the log of the Pom1 amplitude at the pole . For the correlation plot between cell pole intensities to cell length ( Figure 5—figure supplement 1B ) , the intensity at cell pole was calculated as the average values from the first 0 . 83 μm of individual gradient profiles aligned to the maximum value at cell poles . For the quantification of cytosolic Pom1 fluorescence intensities ( Figure 4—figure supplement 1C ) , the cytosolic signal was quantified per cell from n = 3 independent experiments , total of 50–60 cells . Subtraction of background was performed for both camera noise and auto-fluorescence of a non-GFP-tagged strain that was imaged on the same day for each experiment . Cell length measurements ( Figure 5A ) were performed on cells from three individual experiments ( total number of quantified cells per strain is labeled on the figure ) , manually drawing a straight line at the detected calcofluor signal from cell tip to cell tip in ImageJ ( NIH ) . Cells imaged at a medial plane . Individual lengths were recorded and presented as boxplots . For correlation plots to Pom1 levels at cell middle or at the cell pole , averages were calculated for each strain . To quantify Pom1 cluster area and intensity ( Figure 5D ) , we drew ROIs around 323 individual clusters from 26 wild type cells and 197 individual clusters from 25 tea4Δ cells in ImageJ ( NIH ) . The quantification was done on the image of the first time point from a 60-s imaging period . The cluster area and intensity for phosphomutants Pom11A , Pom13A , and Pom15A ( Figure 5—figure supplement 2B ) were quantified in the same manner for 37–42 individual cells from three independent experiments . The number of quantified clusters is listed in the figure . For the quantification of cluster duration ( Figure 5G ) , raw TIRF images were corrected for bleaching with the EMBL tool CorrectBleach and further smoothed in ImageJ . Clusters were then tracked individually for the duration of the 8 s imaging period . Fluorescence quantifications from TIRF imaging for global Pom1 levels ( Figure 6G ) were performed by drawing a 45-pixel wide ROI across the entire detectable TIRF signal per cell . To generate two gradient profiles , one from each cell tip , the geometric middle of the cell was used for alignment . To measure intensities at cell middle , the values from the last 1 . 5 μm of each gradient were averaged . The same method was used for the quantification shown in Figure 5C . For quantification of local Pom1 levels at Cdr2 nodes ( Figure 6B ) , a nine-pixel ROI was drawn around Cdr2-tdTomato nodes and used to detect signal in the Pom1-GFP channel . Control ROIs of the same size were shifted to the immediate vicinity of a Cdr2 node ( Figure 6C ) . Data corrections were done for background camera noise for each individual image and for bleaching . To estimate the average encounter duration , we recorded the length of time the Pom1 signal at a given node was above a defined fluorescence threshold . The threshold choice was instructed by the minimum detectable signal in that experiment . In all cases that threshold was higher than the fluorescence levels measured in the GFP channel using a strain that did not express GFP , in which Cdr2-tdTomato was used to identify the TIRF focal plane and the ROIs of interest . The exact subtracted arbitrary value is indicated in figure legends . Note that the difference observed between short and long cells was robust to changes in threshold choice . All boxplots plotted using the BoxPlotR web-tool ( http://shiny . chemgrid . org/boxplotr/ ) with definition of whisker extend by Tukey . Yeast cultures were grown in YE medium at 30°C to OD600 = 0 . 8 , collected by centrifugation at 3000r ≤ m at 4°C for 5 min and washed with 1x CXS buffer ( 50 mM HEPES , pH 7 . 0 , 0 , 20 mM KCl , 1 mM MgCl2 , 2 mM EDTA pH7 . 5 containing an anti-proteolitic tablet ( Roche , Ref 05892791001 ) . Lysates were obtained via mechanical breakage with acid-treated glass beads ( Sigma ) , using a BeadBeater homogenizer for 10 repetitions at 4 . 5V of 30 s on , 30 s off on ice cycles . The samples were centrifuged at 10 , 000 g for 20 min at 4°C and extracts were recovered by pipetting into a new tube . Protein concentration was determined via spectroscopy using Bradford reagent . 270 μg of proteins were loaded per sample on a 10% SDS-PAGE gel and transferred via wet Western blot transfer . 1° antibodies used: α-GFP 1:1000 dilution ( Mouse , Roche , Cat . No . 11814460001 ) , α-Tubulin ( TAT1 , Mouse , 1:5000 dilution ) , 2° antibodies α-mouse ( HRP detection , Promega , W4021 ) . The mean intensity quantification of 4 independent experiments is shown in Figure 4—figure supplement 1B . Before averaging , values were corrected for the corresponding background and presented as a ratio of α-GFP to α-tubulin signal . | All organisms need to know how to arrange different cell types during the development of their organs and tissues . This information is provided by protein concentration patterns , or gradients , that tell cells how to behave based on where they are positioned . The same fundamental principles also work on a smaller scale . For example , although the rod-shaped yeast Schizosaccharomyces pombe is a single-celled organism , it uses protein concentration gradients to control its growth and timing of division . Before S . pombe cells divide , they need to check that they have reached the right size . Several mechanisms contribute to this information . One of them involves a concentration gradient of a protein known as Pom1 , which is found on the cell membrane , with more protein at the cell extremities and less towards the middle . Pom1 serves to block the activity of Cdr2 – an enzyme that localizes to the cell middle and controls cell division . An open question has been whether Pom1 levels at the center drop as the cell grows , coordinating growth and division . One explanation for how the Pom1 gradient could be regulated is by the removal and addition of phosphate groups . At the cell’s tip , an enzyme removes phosphate groups from Pom1 , causing it to bind to the membrane . As Pom1 diffuses along the membrane , it continuously ‘re-phosphorylates’ itself . This promotes Pom1 to gradually detach , restricting it from spreading along the membrane towards the cell middle . Another explanation is that clusters of Pom1 , formed at the membrane , help establish a gradient by moving along the membrane at different rates: larger clusters , formed in high concentration areas , move slower than smaller clusters , causing levels of Pom1 to be higher at the tip , and lower towards the middle . Now , Gerganova et al . set out to find which of these two processes contributes more to shaping the Pom1 gradient , and determine where Pom1 acts on Cdr2 . Gerganova et al . used super resolution microscopy to track individual Pom1 molecules inside yeast cells . This revealed two findings . First , that individual Pom1 molecules do not travel all the way from the cell tip to the center , but ‘hop’ between clusters as they move towards the middle . Second , in longer cells levels of Pom1 on the membrane drop at the center , where Pom1 encounters Cdr2 . As a result , Cdr2 will come across higher levels of Pom1 in short cells , but low levels of Pom1 in long cells . This allows Pom1 to act as a measure of cell size , preventing short cells from dividing too soon . The role of clusters in creating gradients is not only relevant for yeast cell division . It could potentially apply to the gradients that organize cells and tissues in different organisms . Future work could examine whether similar principles apply in more complex systems . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2019 | Multi-phosphorylation reaction and clustering tune Pom1 gradient mid-cell levels according to cell size |
Singlet oxygen is a highly toxic and inevitable byproduct of oxygenic photosynthesis . The unicellular green alga Chlamydomonas reinhardtii is capable of acclimating specifically to singlet oxygen stress , but the retrograde signaling pathway from the chloroplast to the nucleus mediating this response is unknown . Here we describe a mutant , singlet oxygen acclimation knocked-out 1 ( sak1 ) , that lacks the acclimation response to singlet oxygen . Analysis of genome-wide changes in RNA abundance during acclimation to singlet oxygen revealed that SAK1 is a key regulator of the gene expression response during acclimation . The SAK1 gene encodes an uncharacterized protein with a domain conserved among chlorophytes and present in some bZIP transcription factors . The SAK1 protein is located in the cytosol , and it is induced and phosphorylated upon exposure to singlet oxygen , suggesting that it is a critical intermediate component of the retrograde signal transduction pathway leading to singlet oxygen acclimation .
Growth of photosynthetic organisms depends on light energy , which in turn can cause oxidative damage to the cell if not managed properly ( Li et al . , 2009 ) . Light intensity is highly dynamic in terrestrial and aquatic environments , and the cell must constantly control the dissipation of light energy to avoid photo-oxidative stress while maximizing productivity . In addition to being the site of photosynthesis , the chloroplast houses many essential biochemical reactions such as fatty acid and amino acid biosynthesis , but most of its proteins are encoded in the nucleus and must be imported after translation . Therefore the nucleus must monitor the status of the chloroplast and coordinate gene expression and synthesis of proteins to maintain healthy chloroplast functions . It is known that signals originating from a stressed or dysfunctional chloroplast modulate nuclear gene expression , a process that is called retrograde signaling ( Nott et al . , 2006; Chi et al . , 2013 ) . In Arabidopsis thaliana the gun mutants have helped to define the field of chloroplast retrograde signaling , leading to the identification of GUN1 , a pentatricopeptide repeat protein that is a regulator of this process ( Koussevitzky et al . , 2007 ) , and pointing to the involvement of the tetrapyrrole biosynthetic pathway ( Vinti et al . , 2000; Mochizuki et al . , 2001; Larkin et al . , 2003; Strand et al . , 2003; Woodson and Chory , 2008 ) . A role for heme in retrograde signaling has been shown in Chlamydomonas reinhardtii as well ( von Gromoff et al . , 2008 ) . Many of the gun studies were conducted in context of a dysfunctional chloroplast treated with norflurazon , an inhibitor of carotenoid biosynthesis . More recently a number of exciting advances have shed light on small molecules playing roles in retrograde stress signaling , including methylerythritol cyclodiphosphate , an intermediate of isoprenoid biosynthesis in the chloroplast ( Xiao et al . , 2012 ) , 3-phosphoadenosine 5-phosphate ( PAP ) ( Estavillo et al . , 2011 ) , as well as a chloroplast envelope transcription factor PTM ( Sun et al . , 2011 ) . Plastid gene expression involving sigma factors has been implicated in affecting nuclear gene expression , although the mechanism is unknown ( Coll et al . , 2009; Woodson et al . , 2012 ) . Activation of gene expression by reactive oxygen species ( ROS ) has been well documented ( Apel and Hirt , 2004; Mittler et al . , 2004; Gadjev et al . , 2006; Li et al . , 2009 ) . Thus ROS have been proposed as a means for chloroplasts to signal stress to the nucleus and many examples of global gene expression changes in response to ROS have been described ( Desikan et al . , 2001; Vandenabeele et al . , 2004; Vanderauwera et al . , 2005 ) . Singlet oxygen ( 1O2 ) is a highly toxic form of ROS that can be formed in all aerobic organisms through photosensitization reactions in which excitation energy is transferred from a pigment molecule to O2 . For example , porphyria in humans is caused by defects in tetrapyrrole metabolism that can lead to accumulation of photosensitizing intermediates , which generate 1O2 in the light ( Straka et al . , 1990 ) . In oxygenic photosynthetic organisms , 1O2 is mainly generated at the reaction center of photosystem II , when triplet excited chlorophyll transfers energy to O2 ( Krieger-Liszkay , 2005 ) . 1O2 is the predominant cause of lipid oxidation during photo-oxidative stress ( Triantaphylidès et al . , 2008 ) and is associated with damage to the reaction center ( Trebst et al . , 2002 ) . Because of the abundance and proximity of the two elements of 1O2 generation , the photosensitizer chlorophyll and O2 , it was hypothesized that oxygenic photosynthetic organisms must have evolved robust means to cope with this ROS ( Knox and Dodge , 1985 ) . In Arabidopsis , the EX1 and EX2 proteins in the chloroplast are required for the execution of a 1O2-dependent response: growth arrest in plants and programmed cell death in seedlings , that is distinct from cell damage ( op den Camp et al . , 2003; Wagner et al . , 2004; Lee et al . , 2007 ) . Different players in 1O2 signaling have emerged recently , such as β-cyclocitral , an oxidation product of β-carotene in Arabidopsis ( Ramel et al . , 2012 ) , a bZIP transcription factor ( SOR1 ) responding to reactive electrophiles generated by 1O2 ( Fischer et al . , 2012 ) , and a cytosolic zinc finger protein conserved in Arabidopsis and Chlamydomonas , MBS ( Shao et al . , 2013 ) . In the anoxygenic photosynthetic bacterium Rhodobacter sphaeroides , a σE factor is responsible for the elicitation of the gene expression response to 1O2 ( Anthony et al . , 2005 ) . The unicellular green alga Chlamydomonas reinhardtii is an excellent model organism for investigation of retrograde 1O2 signaling . Chlamydomonas exhibits an acclimation response to 1O2 , in which exposure to a sublethal dose of 1O2 leads to changes in nuclear gene expression that enable cells to resist a subsequent challenge with higher levels of 1O2 ( Ledford et al . , 2007 ) . We hypothesized that acclimation mutants should include regulatory mutants that are defective in sensing and responding to 1O2 . Here we describe the isolation of such a mutant and identification of a cytosolic phosphoprotein SAK1 that is critical for the acclimation and transcriptome response to 1O2 .
Chlamydomonas acclimates to singlet oxygen ( 1O2 ) generated by the exogenous photosensitizing dye rose bengal ( RB ) in the light ( Ledford et al . , 2007 ) . As shown in Figure 1A , wild-type ( WT ) cells that were pretreated with RB in the light were able to survive a challenge treatment with much higher concentrations of RB , unlike cells pretreated with RB in the dark . By screening an insertional mutant population ( Dent et al . , 2005 ) for strains that were sensitive to 1O2 , we isolated a mutant called singlet oxygen acclimation knocked-out1 ( sak1 ) that is defective in acclimation to 1O2 ( Figure 1A ) . We have previously shown that Chlamydomonas WT cells can also acclimate to RB following pretreatment with high light ( Ledford et al . , 2007 ) , indicating that high light and RB induce overlapping responses to 1O2 . When subjected to the same conditions ( high light pretreatment followed by challenge with RB ) , sak1 demonstrated less robust cross-acclimation ( Figure 1B ) . We also tested conversely whether pretreatment with RB can acclimate the cells to growth in high light or in the presence of norflurazon . No increase in resistance to high light or norflurazon was induced by pretreatment with RB in either WT or sak1 ( Figure 1—figure supplement 1 ) . The viability phenotypes after RB treatment shown in Figure 1A were paralleled by changes in Fv/Fm values , a chlorophyll fluorescence parameter representing photosystem II efficiency ( Figure 1C ) . In both WT and sak1 , pretreatment did not cause an inhibition of photosystem II , as demonstrated by unchanged Fv/Fm values after 30 min . However , pretreatment increased resistance of photosystem II to the RB challenge only in WT and not in sak1 cells ( Figure 1C ) . The pretreatment protected the cells only transiently , as by 90 min of challenge treatment both genotypes appeared to have experienced similar inhibition of photosystem II ( Figure 1C ) , consistent with the hypothesis that sak1 is disrupted in early sensing and/or initiation of 1O2 response rather than its direct detoxification . 10 . 7554/eLife . 02286 . 003Figure 1 . The sak1 mutant is defective in singlet oxygen acclimation . ( A ) Acclimation phenotype of WT and sak1 . The cells were pretreated in the dark ( − ) or under light ( + ) in the presence of rose bengal ( RB ) , which requires light for generation of 1O2 . Pretreatment was followed by a subsequent higher concentration of RB ( Challenge ) as indicated under light . ( B ) Cells grown in low light were either kept in low light ( − ) or transferred to high light ( + ) for an hour before challenge in the light with increasing RB concentrations . ( C ) Fv/Fm values were measured after each time point indicated . Pretreatment ( PreT ) with 0 . 5 μM RB was applied for 30 min with ( +PreT ) or without ( −PreT ) light . After the pretreatment , RB was added to both dark and light samples to a final concentration of 3 . 75 μM RB ( challenge ) , and Fv/Fm was measured for 90 min at 30 min intervals ( total 120 min ) . First arrow: addition of pretreatment; second arrow: addition of challenge . ( D ) sak1 has wild-type sensitivity to other photo-oxidative stresses . Serial dilutions of WT and sak1 were spotted onto minimal ( HS ) plates at the indicated light intensity or on TAP plates containing the indicated inhibitor . DCMU , 3- ( 3 , 4-dichlorophenyl ) -1 , 1-dimethylurea; low light ( LL ) , 80 µmol photons m−2 s−1; high light ( HL ) , 450 µmol photons m−2 s−1 . ( E ) Gene expression of a known 1O2-responsive gene , GPX5 , is induced during acclimation , while two genes associated with H2O2 response , APX1 and CAT1 , are not . WT cells were mock-pretreated without RB ( white bars ) or pretreated with RB in the light ( black bars ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 00310 . 7554/eLife . 02286 . 004Figure 1—figure supplement 1 . Pretreatment with RB does not increase resistance to high light or norflurazon in cells grown on plates . Cells were pretreated with 1 μM RB with ( + ) or without ( − ) light , then spotted on minimal plates and grown under high light ( HL ) or grown photoheterotrophically on TAP plates containing norflurazon ( NF ) and grown under low light for 4 days . Cells were spotted in serial dilutions . DOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 004 In contrast to its RB sensitivity , sak1 exhibited wild-type resistance to high light , various photosynthetic inhibitors and generators of other ROS , suggesting its defect is specific to 1O2 ( Figure 1D ) . When tested for the gene expression response of the known 1O2-specific gene GPX5 ( Leisinger et al . , 2001 ) during acclimation , WT cells showed a 20- to 30-fold induction , whereas a known H2O2-responsive ascorbate peroxidase gene ( APX1 ) in Chlamydomonas ( Urzica et al . , 2012 ) and a catalase gene ( CAT1 ) , known to be H2O2 responsive in Arabidopsis ( Davletova et al . , 2005; Vanderauwera et al . , 2005 ) , were unchanged . The mutant sak1 showed attenuated GPX5 induction , as expected for a mutant defective in the 1O2 response ( Figure 1E ) . To obtain insight into the cellular processes and the genes involved in 1O2 acclimation , we used RNA-seq to define the transcriptome of WT cells during acclimation . The sequences were mapped to the Chlamydomonas reinhardtii genome version 4 ( v4 ) , and 16476 transcripts corresponding to gene models were detected ( Wakao et al . , 2014 ) . We validated the data by quantitative reverse transcriptase PCR ( qRT-PCR ) for some of the differentially expressed genes during acclimation ( Figure 2 ) . Basal expression of some of the genes was elevated in sak1 compared to WT ( Cre16 . g683400 and GST1 , Figure 2 ) . Comparisons of the fold change ( FC ) values obtained by RNA-seq and qRT-PCR for the genes tested in Figure 2 are shown in Figure 2 . The FC values are comparable between the two methods , although genes with FC greater than 20 ( detected by RNA-seq ) showed FC values ( estimated by qRT-PCR ) that were two to three times higher ( Cre06 . g281250 . t1 . 1 , Cre13 . g566850 . t1 . 1 , Cre06 . g263550 . t1 . 1 , Cre14 . g623650 . t1 . 2 ) . Some of the genes were also induced by a transition from low light to high light , although not as strongly ( Table 1 ) , indicating that the 1O2 response elicited by addition of RB partly overlaps with that caused by increased light intensity . To examine whether the transcriptome changes were specific to 1O2 , we examined the expression of several previously identified H2O2-responsive genes ( Urzica et al . , 2012 ) ( Table 2 ) . Two of the seven genes , VTC2 ( 3 . 4-fold ) and DHAR1 ( twofold ) were induced during 1O2 acclimation , whereas the other five genes were not differentially expressed ( induced more than twofold ) in our data . For these two genes , their magnitude of induction by 1O2 was smaller than that of H2O2-treated cells ( both genes were ∼ninefold induced by 1 mM H2O2 treatment for 60 min ) ( Urzica et al . , 2012 ) . These differences suggest that our treatment with 1O2 did not lead to a large-scale induction of H2O2-responsive genes , and it is likely that the two above-mentioned genes involved in ascorbate metabolism respond to both H2O2 and 1O2 . 10 . 7554/eLife . 02286 . 005Figure 2 . qRT-PCR analysis of genes identified to be 1O2-responsive by RNA-seq . ( A ) The error bars indicate standard deviation of biological triplicates . The locus of the transcript ( v5 ) and gene name if annotated , are indicated . *SOUL1 was named gene in v4 but not in v5 . ( B ) Comparison of fold change values from RNA-seq data and qPCR . Fold change values were calculated for RNA-seq as described in ‘Material and methods’ , and the values for qPCR are averages obtained from biological triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 00510 . 7554/eLife . 02286 . 006Table 1 . Moderate induction of 1O2 genes during high light exposureDOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 006Fold change ( SD ) *Gene name or IDWTsak1GPX52 . 86 ( 1 . 06 ) 1 . 08 ( 0 . 23 ) CFA13 . 75 ( 0 . 99 ) 1 . 78 ( 0 . 52 ) SOUL23 . 45 ( 1 . 25 ) 1 . 82 ( 0 . 22 ) MRP33 . 10 ( 0 . 39 ) 2 . 37 ( 0 . 32 ) Cre14 . g6139501 . 42 ( 0 . 53 ) 1 . 57 ( 0 . 46 ) LHCSR1†14 . 91 ( 4 . 25 ) 2 . 91 ( 1 . 35 ) *Fold change values are the average of biological triplicates and their standard deviations are indicated in parentheses . †Known to have elevated expression in high light grown cells ( Peers et al . , 2009 ) . 10 . 7554/eLife . 02286 . 007Table 2 . Expression of H2O2 response genes during 1O2 acclimationDOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 007Gene IDRPKM*Fold change†Gene namev4v5WT-mockWT-RBsak1-mocksak1-RBWTsak1APX1Cre02 . g087700 . t1 . 1Cre02 . g087700 . t1 . 249 . 7036 . 2279 . 6558 . 830 . 730 . 74MSD3Cre16 . g676150 . t1 . 1Cre16 . g676150 . t1 . 20 . 300 . 180 . 700 . 170 . 600 . 25MDAR1Cre17 . g712100 . t1 . 1Cre17 . g712100 . t1 . 235 . 9538 . 3033 . 5351 . 341 . 071 . 53DHAR1Cre10 . g456750 . t1 . 1Cre10 . g456750 . t1 . 220 . 4040 . 9325 . 6942 . 182 . 011 . 64GSH1Cre02 . g077100 . t1 . 1Cre02 . g077100 . t1 . 228 . 2726 . 9140 . 4249 . 950 . 951 . 24GSHR1Cre06 . g262100 . t1 . 2Cre06 . g262100 . t1 . 319 . 1719 . 0219 . 3922 . 410 . 991 . 16VTC2Cre13 . g588150 . t1 . 1Cre13 . g588150 . t1 . 218 . 1662 . 5335 . 10103 . 123 . 442 . 94*Average of RPKM obtained from two sequencing lanes as described in ‘Material and methods’ . †Calculated as ratio of ( RPKM-RB ) / ( RPKM-mock ) . During acclimation of WT to 1O2 , 515 genes were up-regulated at least twofold with a false discovery rate ( FDR ) smaller than 1% ( Supplementary file 1 , C1 ) , and 33% of these could be categorized into functional classes based on MapMan ( Thimm et al . , 2004 ) using the Algal Functional Annotation Tool ( Lopez et al . , 2011 ) ( Figure 3A , B ) . The enriched classes are marked with asterisks , and the genes within those classes are listed in Table 3 . Genes involved in sterol/squalene/brassinosteroid metabolism ( in the hormone and lipid metabolism functional classes ) were notably enriched ( Table 3 ) . A sterol methyltransferase was also detected to display differential expression in our previous microarray analysis ( Ledford et al . , 2007 ) . Brassinosteroids are not known to exist in Chlamydomonas , and in plants increasing evidence indicates sterols have a signaling role independent of brassinosteroids ( Lindsey et al . , 2003; Boutté and Grebe , 2009 ) . Two cyclopropane fatty acid synthases ( CFAs ) were among the up-regulated lipid metabolism genes ( Table 3 ) . Another function that was notable among up-regulated genes , although they were not grouped to a common functional class by MapMan , were two genes coding for SOUL heme-binding domain proteins that were SAK1-dependent ( SOUL2 and Cre06 . g299700 . t1 . 1 , formerly annotated as SOUL1 ) ( Figure 2 ) . Genes annotated as involved in transport comprised one of the most enriched classes ( Figure 3B ) . These included a number of multidrug-resistant ( MDR ) and pleiotropic drug-resistant ( PDR ) type transporters as well as other various transporters for ions , peptides , and lipids ( Table 3 ) . The former types of transporters may reflect the cells' response to pump RB out . When the responses to the chemical RB and 1O2 were uncoupled by comparing gene expression in cultures kept in the dark with and without RB , all of the tested 1O2-induced genes and ABC transporters identified from our RNA-seq remained unchanged by RB in the dark in both WT and sak1 ( Table 4 ) . This result indicates that the up-regulation of these genes when RB was added in the light was a response to 1O2 rather than to RB itself . Up-regulation of stress genes included those coding for chaperones and some receptor-like proteins ( Figure 3B; Table 3 ) , suggesting that the cells do mount a stress response during acclimation though not visible by gross growth phenotype ( Figure 1A ) or decrease in Fv/Fm ( Figure 1C ) . A smaller number of 219 genes was down-regulated during acclimation in WT ( Supplementary file 1 , C1 ) , only 21% of which had functional annotation . The most enriched classes of down-regulated genes were nucleotide metabolism and transport , the latter including a distinct type of transporter for small metabolites and ions , different from those found among up-regulated genes that included many MDR- and PDR-type transporters ( Figure 3B; Table 3 ) . 10 . 7554/eLife . 02286 . 008Figure 3 . Differentially expressed genes from pair-wise comparisons . ( A ) Venn diagram representing differentially expressed genes in WT and sak1 . Mapman functional classes distribution of differentially expressed genes ( passing criteria of fold change greater than 21 [up] or smaller than 2−1 [down] with FDR <1% ) during acclimation in ( B ) WT and ( C ) sak1 . ( D ) Differentially expressed genes when comparing WT and sak1 in basal conditions ( i . e . , before exposure to 1O2 ) . The functional classes represented by the numbers are listed; asterisks indicate classes that were enriched compared to the genome . DOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 00810 . 7554/eLife . 02286 . 009Table 3 . Enriched functional classes among differentially expressed genes in WT during 1O2 acclimationDOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 009Primary MapMan classSecondary Mapman classGene ID ( v4 ) Gene ID ( v5 ) Gene nameAnnotationUp-regulated genes transportABC transporters and multidrug resistance systemsCre03 . g169300 . t1 . 1Cre03 . g169300 . t2 . 1ABC transporter ( ABC-2 type ) Cre04 . g220850 . t1 . 1Cre04 . g220850 . t1 . 2ABC transporter ( ABC-2 type ) Cre11 . g474600 . t1 . 1§Cre02 . g095151 . t1ABC transporter ( ABC-2 type ) Cre03 . g151400 . t1 . 2Cre03 . g151400 . t1 . 3ABC transporter ( subfamilyA member3 ) Cre14 . g618400 . t1 . 1§Cre14 . g618400 . t1 . 2ABC transporterCre09 . g395750 . t1 . 2Cre09 . g395750 . t1 . 3ABC transporter ( plant PDR pleitropic drug resistance ) Cre14 . g613950 . t1 . 1§Cre14 . g613950 . t2 . 1ABC transporter , Lipid exporter ABCA1 and related proteinsCre17 . g725150 . t1 . 1Cre17 . g725150 . t1 . 2ABC transporterCre04 . g224400 . t1 . 2§Cre04 . g224400 . t1 . 3ABC transporter ( plant PDR pleitropic drug resistance ) Cre13 . g564900 . t1 . 1§Cre13 . g564900 . t1 . 2MRP3ABC transporter , Multidrug resistance associated proteinCre17 . g721000 . t1 . 1Cre17 . g721000 . t1 . 2ABC transporter ( ABCA ) Cre04 . g224500 . t1 . 2Cre04 . g224500 . t1 . 3ABC transporter ( plant PDR pleitropic drug resistance ) Cre01 . g007000 . t1 . 1§Cre01 . g007000 . t1 . 2ABC transporter ( ABC-2 type ) unspecified anionsCre13 . g574000 . t1 . 2Cre13 . g574000 . t1 . 3Chloride channel 7Cre17 . g729450 . t1 . 1Cre17 . g729450 . t1 . 2Chloride channel 7amino acidsCre04 . g226150 . t1 . 2Cre04 . g226150 . t1 . 3AOC1Amino acid carrier 1; belongs to APC ( amino acid polyamine organocation ) familymiscCre16 . g683400 . t1 . 1§Cre16 . g683400 . t1 . 2CRAL/TRIO domain ( Retinaldehyde binding protein-related ) Cre17 . g718100 . t1 . 1Cre17 . g718100 . t1 . 2Phosphatidylinositol transfer protein SEC14 and related proteins ( CRAL/TRIO ) Cre06 . g311000 . t1 . 2Cre06 . g311000 . t1 . 3FBT2Folate transportecalciumCre09 . g410050 . t1 . 1§Cre09 . g410050 . t1 . 2Ca2+ transporting ATPasepotassiumCre07 . g329882 . t1 . 2Cre07 . g329882 . t1 . 3Ca2+-activated K+ channel proteinsphosphateCre16 . g686750 . t1 . 1Cre16 . g686750 . t1 . 2PTA3Proton/phosphate symportermetalCre13 . g570600 . t1 . 1Cre13 . g570600 . t1 . 2CTR1CTR type copper ion transportermetabolite transporters at the mitochondrial membraneCre06 . g267800 . t1 . 2Cre06 . g267800 . t2 . 1Mitochondrial carrier protein hormone metabolism*brassinosteroidCre16 . g663950 . t1 . 1Cre16 . g663950 . t1 . 2Sterol C5-desaturaseCre02 . g076800 . t1 . 1Cre02 . g076800 . t1 . 2delta14-sterol reductaseCre12 . g557900 . t1 . 1Cre12 . g557900 . t1 . 1CDI1C-8 , 7 sterol isomeraseCre02 . g092350 . t1 . 1Cre02 . g092350 . t1 . 2Cytochrome P450 , CYP51 Sterol-demethylaseCre12 . g500500 . t1 . 2Cre12 . g500500 . t2 . 1SAM-dependent methyltransferasesjasmonateCre19 . g756100 . t1 . 1Cre03 . g210513 . t112-oxophytodienoic acid reductaseauxinCre14 . g609900 . t1 . 1Cre14 . g609900 . t1 . 1Predicted membrane protein , contains DoH and Cytochrome b-561/ferric reductase transmembrane domainsCre06 . g276050 . t1 . 1Cre06 . g276050 . t1 . 2Aldo/keto reductaseCre16 . g692800 . t1 . 2Cre16 . g692800 . t1 . 3Aldo/keto reductaseCre03 . g185850 . t1 . 2Cre03 . g185850 . t1 . 2pfkB family , sugar kinase-related minor CHO metabolismothersCre06 . g276050 . t1 . 1Cre06 . g276050 . t1 . 2Aldo/keto reductaseCre16 . g692800 . t1 . 2Cre16 . g692800 . t1 . 3Aldo/keto reductaseCre03 . g185850 . t1 . 2Cre03 . g185850 . t1 . 2pfkB family , sugar kinase-relatedcalloseCre06 . g302050 . t1 . 1Cre06 . g302050 . t1 . 21 , 3-beta-glucan synthasemyo-inositolCre03 . g180250 . t1 . 1Cre03 . g180250 . t1 . 2Myo-inositol-1-phosphate synthase stressbioticCre01 . g057050 . t1 . 1§Cre03 . g144324 . t1Leucine Rich RepeatCre01 . g016200 . t1 . 2Cre01 . g016200 . t1Mlo FamilyCre28 . g776450 . t1 . 1§Cre08 . g358573 . t1PSMD1026S proteasome regulatory complexabioticCre12 . g501500 . t1 . 1NF†Cre02 . g132300 . t1 . 2Cre09 . g395732 . t1DnaJ domainCre07 . g339650 . t1 . 2Cre07 . g339650 . t1 . 3DNJ20DnaJ-like proteinCre01 . g033300 . t1 . 1§Cre01 . g033300 . t2 . 1No annotation‡Cre16 . g677000 . t1 . 1Cre16 . g677000 . t1 . 2HSP70EHeat shock protein 70ECre08 . g372100 . t1 . 1Cre08 . g372100 . t1 . 2HSP70AHeat shock protein 70A lipid metabolismphospholipid synthesisCre13 . g604700 . t1 . 2Cre13 . g604700 . t1 . 3PCT1CDP-alcohol phosphatidyltransferase/Phosphatidylglycerol-phosphate synthaseCre06 . g281250 . t1 . 1§Cre06 . g281250 . t1 . 2CFA1Cyclopropane fatty acid synthaseCre09 . g398700 . t1 . 1§Cre09 . g398700 . t1 . 2CFA2Cyclopropane fatty acid synthase‘exoticsߣ ( steroids , squalene etc ) Cre01 . g061750 . t1 . 1Cre03 . g146507 . t1SPT2Serine palmitoyltransferaseCre83 . g796250 . t1 . 1NF†SPT1Serine palmitoyltransferaseCre02 . g137850 . t1 . 1Cre09 . g400516 . t1TRAM ( translocating chain-associating membrane ) superfamilyFA synthesis and FA elongationCre03 . g182050 . t1 . 1Cre03 . g182050 . t1Long-chain acyl-CoA synthetases ( AMP-forming ) Cre06 . g256750 . t1 . 1Cre06 . g256750 . t1 . 2Acyl-ACP thioesterasemiscshort chain dehydrogenase/reductase ( SDR ) Cre12 . g556750 . t1 . 2Cre12 . g556750 . t1 . 3Short chain dehydrogenaseCre27 . g775000 . t1 . 1Cre12 . g549852 . t1Short chain dehydrogenaseCre17 . g731350 . t1 . 2Cre17 . g731350 . t1 . 2Short chain dehydrogenaseCre08 . g381510 . t1 . 1§NF†Short chain alcohol dehydrogenaseUDP glucosyl and glucoronyl transferasesCre02 . g144050 . t1 . 1Cre02 . g144050 . t2 . 1Acetylglucosaminyltransferase EXT1/exostosin 1Cre16 . g659450 . t1 . 1Cre16 . g659450 . t1 . 2Lactosylceramide 4-alpha-GalactosyltransferaseCre03 . g173300 . t1 . 1Cre03 . g173300 . t1 . 2Lactosylceramide 4-alpha-GalactosyltransferasedynaminCre02 . g079550 . t1 . 1Cre02 . g079550 . t1 . 2Dynamin-related GTPase , involved in circadian rhythmsmisc2Cre06 . g258600 . t1 . 1§Cre06 . g258600 . t2 . 1Predicted hydrolase related to dienelactone hydrolaseacid and other phosphatasesCre06 . g249800 . t1 . 1Cre06 . g249800 . t1 . 2Sphingomyelin synthaseDown-regulated genes nucleotide metabolismsalvageCre13 . g573800 . t1 . 1Cre13 . g573800 . t1 . 2Phosphoribulokinase / Uridine kinase familysynthesisCre12 . g503300 . t1 . 1Cre12 . g503300 . t1 . 2Phosphoribosylamidoimidazole-succinocarboxamide synthaseCre06 . g308500 . t1 . 1Cre06 . g308500 . t1 . 2CMP2Carbamoyl phosphate synthase , small subunitCre14 . g614300 . t1 . 1Cre14 . g614300 . t1 . 2Inosine-5-monophosphate dehydrogenase transportABC transporters and multidrug resistance systemsCre06 . g273750 . t1 . 2Cre06 . g273750 . t1 . 3SUA1Chloroplast sulfate transporterCre02 . g083354 . t1 . 1Cre02 . g083354 . t1ATP-binding cassette , subfamily B ( MDR/TAP ) , member 9calciumCre06 . g263950 . t1 . 2Cre06 . g263950 . t1 . 3Na+/K + ATPase , alpha subunitmetabolite transporters at the envelope membraneCre08 . g363600 . t1 . 1Cre08 . g363600 . t1 . 2Glucose-6-phosphate , PEP/phosphate antiportermetalCre17 . g720400 . t1 . 2Cre17 . g720400 . t1 . 3HMA1Heavy metal transporting ATPaseP- and V-ATPasesCre10 . g459200 . t1 . 1Cre10 . g459200 . t1 . 2ACA4Plasma membrane H + -transporting ATPasephosphateCre02 . g144650 . t1 . 1Cre02 . g144650 . t1 . 2PTB12Na+/Pi symporterpotassiumCre06 . g278700 . t1 . 2Cre06 . g278700 . t1 . 2Myotrophin and similar proteins*Functional terms are inferred by homology to the annotation set of Arabidopsis thaliana ( Lopez et al . , 2011 ) . †Corresponding gene model was not found in v5 . ‡No functional annotations found on v5 but defined by MapMan on Algal Functional Annotation Tool ( Lopez et al . , 2011 ) . §Induction during 1O2 acclimation dependent on SAK1 ( Table 5 ) . 10 . 7554/eLife . 02286 . 010Table 4 . 1O2 response genes are not induced when RB is added in the darkDOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 010Fold change +RB/−RB ( SD ) *Gene name or IDWTsak1GPX51 . 13 ( 0 . 33 ) 0 . 87 ( 0 . 31 ) SAK11 . 38 ( 0 . 08 ) 1 . 29 ( 0 . 19 ) CFA10 . 90 ( 0 . 04 ) 1 . 44 ( 0 . 22 ) SOUL21 . 17 ( 0 . 25 ) 1 . 11 ( 0 . 19 ) MRP3† , ‡1 . 13 ( 0 . 12 ) 1 . 07 ( 0 . 25 ) Cre12 . g503950† , ‡0 . 93 ( 0 . 06 ) 1 . 20 ( 0 . 12 ) Cre14 . g613950† , §0 . 65 ( 0 . 06 ) 0 . 79 ( 0 . 15 ) Cre04 . g220850† , ‡1 . 00 ( 0 . 09 ) 1 . 29 ( 0 . 04 ) Cre09 . g395750† , ‡1 . 05 ( 0 . 10 ) 1 . 29 ( 0 . 12 ) *Average of fold change and standard deviation ( SD ) of biological triplicates . †Annotated as transport function . ‡ABC transporter . §Sec14-like phosphatidylinositol transfer protein . Although only 33% of the up-regulated genes have a functional annotation ( Figure 3B ) , it is interesting that the 1O2 response in Chlamydomonas involves genes and biological processes that appear to be distinct from those that respond specifically to 1O2 in Arabidopsis ( op den Camp et al . , 2003 ) . A total of 70 1O2-response genes have been defined using a microarray with the flu mutant in Arabidopsis ( op den Camp et al . , 2003 ) . These genes include the following classes ( number of genes ) : metabolism ( 11 ) , transcription ( 5 ) , protein fate ( 4 ) , transport ( 2 ) , cellular communication/signal transduction ( 17 ) , cell rescue/defense in virulence ( 4 ) , subcellular localization ( 2 ) , binding function or cofactor requirement ( 1 ) , transport facilitation ( 5 ) and others ( 19 ) . From this list of 70 genes we found four similarly annotated genes within our 515 genes induced by 1O2 in Chlamydomonas: a Myb transcription factor , a mitochondrial carrier protein , an amino acid permease , and an ATPase/aminophospholipid translocase . None of these genes in Chlamydomonas was the closest ortholog of the corresponding Arabidopsis gene . Conversely , genes similar to those strongly up-regulated in a SAK1-dependent manner such as CFAs , SOUL proteins , GPX , and sterol biosynthetic enzymes were not found among the Arabidopsis 1O2-specific genes despite having clear counterparts in Arabidopsis . Taken together , these results suggest that these two organisms may deploy distinct mechanisms in their responses to 1O2 . In the sak1 mutant , 1020 genes were up-regulated , whereas 434 genes were down-regulated during acclimation ( Supplementary file 1 , C2 ) . 350 of the 515 genes up-regulated in WT overlapped with the set of up-regulated genes in the mutant ( Figure 3A ) . Comparing the fold changes of genes in WT and sak1 during acclimation , we defined 104 genes as SAK1-dependent genes that displayed moderate to strong attenuation in their response ( fold change ratio <0 . 5 ) ( Table 5 ) . Some of the genes that belong to enriched biological classes found among WT up-regulated genes are indicated in Table 3 . Interestingly , the most strongly induced genes in WT were found among this group; 37 out of 104 SAK1-dependent genes were among the top 10% most strongly induced genes ( Table 5 ) . 33 out of these 37 most strongly induced SAK1-dependent genes displayed strong disruption in their up-regulation; reduced to 0 . 01–0 . 25 of magnitude of fold change in sak1 as compared to WT ( Table 5 ) . These results indicate SAK1 is required for the induction of the most strongly induced genes during acclimation reflecting its critical role in regulating the cellular acclimation response to 1O2 . 10 . 7554/eLife . 02286 . 011Table 5 . Genes that require SAK1 for induction by 1O2DOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 011Gene ID ( v4 ) Gene ID ( v5 ) Gene nameAnnotationFC WT* ( log2 ) FC sak1 ( log2 ) Attenuation ( FC-sak1/FC-WT ) †Basal repression in sak1 ( log2 ) Cre02 . g137700 . t1 . 1‡Cre09 . g4004046 . 491 . 800 . 04−3 . 35Cre06 . g281250 . t1 . 1‡Cre06 . g281250CFA1Cyclopropane fatty acid synthase5 . 921 . 160 . 04−2 . 10Cre27 . g775950 . t1 . 2Cre12 . g5579285 . 830 . 810 . 03Cre01 . g033300 . t1 . 1Cre01 . g0333005 . 72−0 . 390 . 01Cre13 . g566850 . t1 . 1‡Cre13 . g566850SOUL2SOUL heme-binding protein5 . 531 . 330 . 05−2 . 60Cre14 . g623650 . t1 . 1Cre14 . g623650Alcohol dehydrogenase4 . 891 . 670 . 11Cre13 . g600650 . t1 . 1Cre06 . g278245Rieske 2Fe-2S domain4 . 761 . 640 . 12Cre06 . g263550 . t1 . 1Cre06 . g263550LCI7R53 . 5-related protein4 . 461 . 770 . 15Cre07 . g342100 . t1 . 1Cre07 . g3421004 . 431 . 400 . 12Cre06 . g299700 . t1 . 1‡Cre06 . g299700SOUL1SOUL heme-binding protein4 . 320 . 430 . 07−1 . 13Cre09 . g398700 . t1 . 1‡Cre09 . g398700CFA2Cyclopropane fatty acid synthase4 . 050 . 180 . 07−1 . 00Cre12 . g492650 . t1 . 1‡Cre12 . g492650FAS2Fasciclin-like protein4 . 010 . 070 . 07−1 . 24Cre08 . g381510 . t1 . 1NF3 . 940 . 730 . 11Cre10 . g458450 . t1 . 2Cre10 . g458450GPX5Glutathione peroxidase3 . 912 . 060 . 28Cre11 . g474600 . t1 . 1Cre02 . g095151ABC transporter ( ABC-2 type ) 3 . 900 . 440 . 09Cre13 . g600700 . t1 . 1Cre06 . g2782463 . 781 . 480 . 20Cre14 . g613950 . t1 . 1Cre14 . g6139503 . 651 . 380 . 21Cre06 . g269300 . t1 . 1Cre06 . g269300DUF13653 . 500 . 400 . 12Cre08 . g380300 . t1 . 2Cre08 . g380300MSRA3Peptide methionine sulfoxide reductase3 . 450 . 660 . 14Cre28 . g776450 . t1 . 1Cre08 . g358573TRP7Transient receptor potential ion channel3 . 31−0 . 790 . 06Cre01 . g031650 . t1 . 2Cre01 . g031650CGLD12Potential galactosyl transferase activity3 . 300 . 670 . 16Cre14 . g629061 . t1 . 1NFDUF21773 . 250 . 080 . 11Cre12 . g503950 . t1 . 1Cre12 . g503950CRAL/TRIO domain3 . 240 . 310 . 13Cre13 . g564900 . t1 . 1Cre13 . g564900ABC transporter transmembrane region3 . 220 . 340 . 14Cre02 . g139500 . t1 . 1Cre09 . g401701DUF12953 . 04−0 . 160 . 11Cre14 . g618400 . t1 . 1Cre14 . g6184002 . 971 . 150 . 28Cre17 . g715150 . t1 . 1Cre17 . g7151502 . 890 . 130 . 15Cre17 . g741300 . t1 . 2‡Cre17 . g741300SAK12 . 880 . 660 . 21−2 . 77Cre01 . g007300 . t1 . 1Cre01 . g0073002 . 85−1 . 150 . 06Cre16 . g648700 . t1 . 2‡Cre16 . g648700ABC transporter ( ABC-2 type ) 2 . 790 . 260 . 17−1 . 26Cre13 . g566900 . t1 . 2Cre13 . g5669002 . 76−0 . 380 . 11Cre02 . g137750 . t1 . 2Cre09 . g400441JmjC domain2 . 72−0 . 310 . 12Cre06 . g263500 . t1 . 1Cre06 . g263500Archease protein family ( DUF101 ) 2 . 671 . 020 . 32Cre01 . g016150 . t1 . 1‡Cre01 . g016150ADP-ribosylglycohydrolase2 . 650 . 170 . 18−1 . 26Cre08 . g380000 . t1 . 1Cre08 . g380000Formylglycine-generating sulfatase enzyme2 . 591 . 530 . 48Cre14 . g615600 . t1 . 1Cre14 . g615600Putative serine esterase ( DUF676 ) 2 . 53−0 . 540 . 12Cre11 . g472900 . t1 . 2Cre02 . g095113CAP-Gly domain2 . 45−0 . 050 . 18Cre06 . g269250 . t1 . 1Cre06 . g2692502 . 440 . 550 . 27Cre02 . g120600 . t1 . 1Cre09 . g4030712 . 440 . 940 . 35Cre06 . g261200 . t1 . 1Cre06 . g261200ERG25Sterol desaturase2 . 420 . 640 . 29Cre16 . g683400 . t1 . 1Cre16 . g683400CRAL/TRIO domain2 . 400 . 080 . 20Cre22 . g765150 . t1 . 1Cre11 . g467725hypothetical protein2 . 300 . 460 . 28Cre13 . g571800 . t1 . 2Cre13 . g571800DUF13362 . 270 . 720 . 34Cre13 . g579450 . t1 . 2Cre13 . g579450CST1Membrane transporter2 . 271 . 230 . 49Cre08 . g380350 . t1 . 1Cre08 . g3803502 . 21−0 . 010 . 21Cre16 . g649250 . t1 . 2Cre16 . g6492502 . 080 . 580 . 35Cre11 . g476250 . t1 . 1Cre11 . g4762502 . 080 . 490 . 33Cre02 . g108000 . t1 . 2Cre02 . g1080002 . 081 . 030 . 49Cre13 . g583300 . t1 . 1Cre13 . g5833001 . 98−0 . 480 . 18Cre04 . g215300 . t1 . 2NF1 . 970 . 570 . 38Cre02 . g139450 . t1 . 1Cre09 . g401663DUF9471 . 95−0 . 620 . 17Cre03 . g194750 . t1 . 2Cre03 . g1947501 . 950 . 730 . 43Cre06 . g258600 . t1 . 1Cre06 . g258600Dienelactone hydrolase family1 . 91−0 . 950 . 14Cre10 . g418700 . t1 . 1Cre10 . g418700Probable N6-adenine methyltransferase1 . 87−0 . 030 . 27Cre10 . g444550 . t1 . 1Cre10 . g444550SPP1ASignal peptide peptidase1 . 810 . 510 . 41Cre01 . g060050 . t1 . 2Cre03 . g1458071 . 78−0 . 110 . 27Cre09 . g410050 . t1 . 1Cre09 . g410050Calcium transporting ATPase1 . 760 . 510 . 42Cre03 . g163400 . t1 . 2Cre03 . g1634001 . 76−0 . 170 . 26Cre01 . g008450 . t1 . 1Cre01 . g008450Nuf2 family1 . 73−0 . 540 . 21Cre12 . g536650 . t1 . 1Cre12 . g5366501 . 720 . 350 . 39Cre02 . g114900 . t1 . 2Cre02 . g114900ANK23predicted protein1 . 710 . 080 . 32Cre16 . g661850 . t1 . 2Cre16 . g661850Calcium/calmoduline dependent protein kinase association1 . 690 . 030 . 32Cre14 . g615500 . t1 . 2Cre14 . g615500Glycoprotease family1 . 68−0 . 760 . 18Cre11 . g483100 . t1 . 2Cre11 . g483100Protein kinase1 . 66−0 . 490 . 22Cre28 . g776650 . t1 . 1Cre08 . g3585691 . 640 . 330 . 40Cre07 . g340250 . t1 . 2Cre07 . g340250Protein kinase1 . 63−0 . 410 . 24Cre06 . g296250 . t1 . 2Cre06 . g296250SYK1tRNA synthetase , class II1 . 600 . 540 . 48Cre06 . g310500 . t1 . 1Cre06 . g3105001 . 570 . 180 . 38Cre07 . g342800 . t1 . 2Cre07 . g342800CGL16Predicted protein1 . 490 . 320 . 44Cre03 . g181450 . t1 . 2Cre03 . g181450DUF16191 . 470 . 350 . 46Cre66 . g793601 . t1 . 1Cre35 . g7594971 . 470 . 030 . 37Cre14 . g614050 . t1 . 2Cre14 . g614050MAP65Microtubule associated protein1 . 430 . 060 . 39Cre04 . g217500 . t1 . 1Cre04 . g217500Inosine-uridine preferring nucleoside hydrolase1 . 420 . 190 . 43Cre06 . g292950 . t1 . 1Cre06 . g292950DNA polymerase delta , subunit 41 . 38−0 . 120 . 35Cre16 . g661750 . t1 . 1Cre16 . g661750Calcium/calmoduline dependent protein kinase association1 . 38−0 . 120 . 35Cre01 . g007000 . t1 . 1Cre01 . g007000ABC transporter ( ABC-2 type ) 1 . 350 . 210 . 45Cre04 . g224400 . t1 . 2Cre04 . g224400ABC transporter ( ABC-2 type ) 1 . 34−0 . 130 . 36Cre01 . g068400 . t1 . 2Cre16 . g6807901 . 330 . 160 . 45Cre05 . g237400 . t1 . 1Cre05 . g237400DAE1Diaminopimelate epimerase1 . 320 . 220 . 47Cre14 . g609600 . t1 . 2Cre14 . g6096001 . 32−0 . 580 . 27Cre05 . g234850 . t1 . 2Cre05 . g234850Ubiquitin carboxyl-terminal hydrolase1 . 290 . 160 . 46Cre03 . g179200 . t1 . 1Cre03 . g1792001 . 28−0 . 480 . 30Cre10 . g417730 . t1 . 1Cre10 . g4177301 . 270 . 170 . 47Cre03 . g159700 . t1 . 2Cre03 . g1597001 . 26−0 . 140 . 38Cre12 . g540150 . t1 . 2Cre12 . g5401501 . 19−0 . 240 . 37Cre01 . g006550 . t1 . 2‡Cre01 . g006550No annotation1 . 17−0 . 490 . 32−1 . 60Cre03 . g159950 . t1 . 2Cre03 . g1599501 . 17−0 . 170 . 40Cre27 . g775900 . t1 . 2Cre12 . g5575031 . 14−0 . 700 . 28Cre02 . g121600 . t1 . 1Cre09 . g387208Protein kinase1 . 140 . 000 . 46Cre14 . g609550 . t1 . 1NF1 . 13−0 . 840 . 26Cre07 . g315050 . t1 . 2Cre07 . g3150501 . 12−0 . 030 . 45Cre04 . g218800 . t1 . 2Cre04 . g218800THB3Truncated hemoglobin1 . 11−0 . 500 . 33Cre02 . g133300 . t1 . 1Cre09 . g3966241 . 11−0 . 430 . 34Cre01 . g060650 . t1 . 2Cre03 . g1460671 . 10−0 . 420 . 35Cre01 . g057050 . t1 . 1Cre03 . g1443241 . 100 . 040 . 48Cre06 . g304950 . t1 . 1Cre06 . g3049501 . 07−0 . 650 . 30Cre08 . g358200 . t1 . 2Cre08 . g358200A4Protein kinase1 . 07−0 . 820 . 27Cre16 . g689550 . t1 . 2Cre16 . g689550PTK8Putative tyrosine kinase1 . 06−0 . 170 . 43Cre17 . g720950 . t1 . 1Cre17 . g7209503-oxo-5-alpha-steroid 4-dehydrogenase1 . 05−0 . 260 . 40Cre02 . g090950 . t1 . 2Cre02 . g0909501 . 05−0 . 270 . 40Cre16 . g683350 . t1 . 1Cre16 . g6833501 . 03−0 . 670 . 31Cre02 . g109450 . t1 . 1Cre02 . g1094501 . 01−0 . 030 . 48Cre16 . g652750 . t1 . 1Cre16 . g6527501 . 01−0 . 290 . 41Cre03 . g190000 . t1 . 1Cre03 . g1900001 . 00−0 . 990 . 25*Data were ordered by FC in WT . †Of the 52 most highly induced genes in WT ( the top 10% ) , 37 were SAK1-dependent , and the induction of 33 of these genes was strongly attenuated to only 0 . 01-0 . 25 of magnitude of FC found in the WT . Dashed line indicates cutoff of FC for the top 10% most strongly induced genes . ‡Genes that are repressed at basal level in sak1 . NF , not found in v5 . Classes of up-regulated genes in sak1 were distinct from those of WT and included secondary metabolism of isoprenoids ( Figure 3C; Table 6 ) , precursors to photoprotective pigments such as carotenoids and tocopherols ( Li et al . , 2009 ) . Phenylpropanoids , a group of metabolites associated with defense against stresses such as ultraviolet light and herbivores ( Maeda and Dudareva , 2012 ) , also represented a larger part of the response in sak1 as compared to WT ( Figure 3C ) . Another mutant-specific class of genes was cell vesicular transport , suggesting alteration in cell organization in response to the loss of SAK1 ( Figure 3C; Table 6 ) . There were 434 genes that were down-regulated by 1O2 in the sak1 mutant ( Supplementary file 1 , C2 ) , none of which overlapped with the set of down-regulated genes in WT , in contrast to the overlap of up-regulated genes in the two genotypes ( Figure 3A ) . Enriched classes of genes included those involved in DNA , nucleotide metabolism , hormone metabolism ( not of brassinosteroid ) and tetrapyrrole metabolism ( Figure 3C , Table 6 ) . 10 . 7554/eLife . 02286 . 012Table 6 . Enriched functional classes among differentially expressed genes in sak1 during 1O2 acclimationDOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 012Primary Mapman classSecondary Mapman classGene ID ( v4 ) Gene nameAnnotationUp-regulated genes Secondary metabolismisoprenoidsCre13 . g565650 . t1 . 1Geranylgeranyl pyrophosphate synthase/Polyprenyl synthetaseCre06 . g267600 . t1 . 1Lycopene epsilon cyclaseCre09 . g407200 . t1 . 1Phytoene desaturaseCre06 . g267600 . t1 . 1Lycopene epsilon cyclaseCre01 . g011100 . t1 . 1Prenyltransferase and squalene oxidase repeat , Oxidosqualene-lanosterol cyclase and related proteinsN miscCre08 . g381707 . t1 . 1NF*phenylpropanoidsCre03 . g207800 . t1 . 1Alcohol dehydrogenase , class VCre14 . g623650 . t1 . 1Alcohol dehydrogenase , class V ( Zinc-binding ) Cre01 . g039350 . t1 . 1Cytochrome P450 reductase , possibly CYP505B familysulfur-containingCre06 . g299400 . t1 . 1NF*waxCre17 . g722150 . t1 . 1PKS3Type III polyketide synthaseCre07 . g318500 . t1 . 2FAE1/Type III polyketide synthase-like protein , Chalcone and stilbene synthases Lipid metabolism‘exotics’ ( steroids , squalene etc ) Cre01 . g061750 . t1 . 1serine palmitoyltransferaseCre02 . g137850 . t1 . 1NF*Cre83 . g796250 . t1 . 1NF*Cre01 . g011100 . t1 . 1Prenyltransferase and squalene oxidase repeat , Oxidosqualene-lanosterol cyclase and related proteinsFA synthesis and FA elongationCre06 . g256750 . t1 . 1Acyl carrier protein thioesteraseCre03 . g182050 . t1 . 1Long-chain acyl-CoA synthetases ( AMP-forming ) Cre02 . g074650 . t1 . 1Kelch repeat-containing proteins , Acyl-CoA binding proteiglycerol metabolismCre01 . g053000 . t1 . 1GPD2Glycerol-3-phosphate dehydrogenase/dihydroxyacetone-3-phosphate reductaseglycolipid synthesisCre13 . g583600 . t1 . 1DGD1Digalactosyldiacylglycerol synthaselipid degradationCre01 . g057450 . t1 . 2NF*Cre02 . g126050 . t1 . 1NF*phospholipid synthesisCre06 . g281250 . t1 . 1CFA1Cyclopropane fatty acid synthaseCre01 . g038250 . t1 . 1SDC1Serine decarboxylaseCre11 . g472700 . t1 . 1NF*Cre13 . g604700 . t1 . 2CDP-alcohol phosphatidyltransferase/Phosphatidylglycerol-phosphate synthase Cellvesicle transportCre18 . g744100 . t1 . 1NF*Cre17 . g721900 . t1 . 1COG5Component of oligomeric golgi complexCre01 . g003050 . t1 . 1SEC8Component of the Exocyst ComplexCre04 . g224800 . t1 . 1Endosomal R-SNARE protein , Vamp7/Nyv1-familyCre17 . g728150 . t1 . 1Endosomal R-SNARE protein , Yky6-familyCre12 . g507450 . t1 . 1Trans-Golgi network Qa-SNARE protein , Syntaxin16/Syx16/Tlg2/Syp4-familyCre03 . g210600 . t1 . 1NF*Cre04 . g225900 . t1 . 1Endosomal R-SNARE protein , Vamp7/Nyv1-familyCre02 . g101400 . t1 . 1CHC1Clathrin Heavy ChainCre17 . g709350 . t1 . 1Late endosomal Qc-SNARE protein , Syx8/Syntaxin8-familyCre07 . g342050 . t1 . 1Endosomal Qb-SNARE , Npsn-familyCre16 . g692050 . t1 . 1ER-Golgi Qa-SNARE protein , Syntaxin5/Syx5/Sed5/Syp3-familyCre16 . g676650 . t1 . 1AP1G1Gamma1-AdaptinCre02 . g099000 . t1 . 1Late endosomal Qc-SNARE protein , Syx6/Tlg1/Syp5/6-familyCre12 . g554200 . t1 . 2ER-Golgi Qb-SNARE , Memb/GS35/Bos1-familyCre06 . g310000 . t1 . 1AP4E1Epsilon4-AdaptinCre10 . g421250 . t1 . 1EXO70Hypothetical Conserved Protein . Similar to Exo70 , a subunit of the exocyst complexCre07 . g330950 . t1 . 1AP4S4Sigma4-AdaptinCre12 . g488850 . t1 . 2Adaptin , alpha/gamma/epsilondivisionCre06 . g269950 . t1 . 1CDC48Protein involved in ubiquitin-dependent degradation of ER-bound substratesCre08 . g359200 . t1 . 2Regulator of chromosome condensation ( RCC1 ) organisationCre13 . g588600 . t1 . 2Kinesin ( SMY1 subfamily ) Cre12 . g513450 . t1 . 1TUH1Eta-TubulinCre01 . g010950 . t1 . 226S proteasome regulatory complex , subunit PSMD10 ( Ankyrin repeat ) Cre16 . g679650 . t1 . 2Fimbrin/PlastinCre06 . g261950 . t1 . 1Myotrophin and similar proteins ( Ankyrin repeat ) Cre06 . g291700 . t1 . 1RSP3Radial spoke protein 3Cre10 . g446700 . t1 . 1ANK28Ankyrin repeat and DHHC-type Zn-finger domain containing proteins Hormone metabolism†abscisic acidCre16 . g657800 . t1 . 2CCD3Carotenoid cleavage dioxygenaseauxinCre14 . g609900 . t1 . 1Predicted membrane protein , contains DoH and Cytochrome b-561/ferric reductase transmembrane domainsbrassinosteroidCre16 . g663950 . t1 . 1Sterol C5 desaturaseCre02 . g092350 . t1 . 1Cytochrome P450 , CYP51 superfamily; sterol 14 desaturaseCre12 . g557900 . t1 . 1CDI1C-8 , 7 sterol isomeraseCre02 . g076800 . t1 . 1Delta14-sterol reductase , mitochondrialCre12 . g500500 . t1 . 224-methylenesterol C-methyltransferaseethyleneCre02 . g108450 . t1 . 1FAP280Flagellar Associated Protein , transcriptional coactivator-like , putative transcription factorjasmonateCre19 . g756100 . t1 . 1NF* Miscacid and other phosphatasesCre09 . g396900 . t1 . 1NADH pyrophosphatase I of the Nudix family of hydrolasesCre06 . g259650 . t1 . 1Calcineurin-like phosphoesterase , Acid-phosphatase-relatedCre06 . g249800 . t1 . 1Sphingomyelin synthetase -relatedcytochrome P450Cre05 . g234100 . t1 . 1Cytochrome P450 , CYP197 superfamilydynaminCre02 . g079550 . t1 . 1DRP2Dynamin-related GTPase , involved in circadian rhythmsCre05 . g245950 . t1 . 1DRP1Dynamin-related GTPaseglutathione S transferasesCre03 . g154950 . t1 . 1Glutathione S-transferasemisc2Cre12 . g538450 . t1 . 1EPT1CDP-Etn:DAG Ethanolamine phosphotransferaseshort chain dehydrogenase/reductase ( SDR ) Cre12 . g556750 . t1 . 2Short-chain dehydrogenase/reductaseCre08 . g384864 . t1 . 1SH3 domain , protein bindingCre27 . g775000 . t1 . 1NF*Cre17 . g731350 . t1 . 2Short chain dehydrogenaseUDP glucosyl and glucoronyl transferasesCre02 . g111150 . t1 . 2ELG26Exostosin-like glycosyltransferaseCre02 . g144050 . t1 . 1Acetylglucosaminyltransferase EXT1/exostosin 1Cre03 . g204050 . t1 . 2ELG6Exostosin-like glycosyltransferasesCre11 . g474450 . t1 . 1NF*Cre03 . g173300 . t1 . 1Lactosylceramide 4-alpha-galactosyltransferase ( alpha- 1 , 4-galactosyltransferase ) Cre02 . g116600 . t1 . 1ELG23Exostosin-like glycosyltransferaseDown-regulated genes Hormone metabolism†cytokininCre18 . g744950 . t1 . 2NF*Cre16 . g678900 . t1 . 1Response regulator receiver domainCre01 . g040450 . t1 . 1HDT1Histidine-aspartic acid phosphotransferase 1 ( phosphorylation cascade ) ethyleneCre09 . g403550 . t1 . 1Iron/ascorbate family oxidoreductases Nucleotide metabolismdeoxynucleotide metabolismCre12 . g491050 . t1 . 1RIR2Ribonucleotide reductase ( RNR ) , small subunitCre12 . g492950 . t1 . 1RIR1Ribonucleotide reductase ( RNR ) , large subunit , class ICre16 . g667850 . t1 . 1dUTP pyrophosphatasesynthesisCre14 . g614300 . t1 . 1Inosine-5-monophosphate dehydrogenase/GMP reductaseCre07 . g318750 . t1 . 1Phosphoribosylformylglycinamidine cyclo-ligase Tetrapyrrole synthesisporphobilinogen deaminaseCre16 . g663900 . t1 . 1Porphobilinogen deaminaseprotochlorophyllide reductaseCre01 . g015350 . t1 . 1Light-dependent protochlorophyllide reductaseurogen III methylaseCre02 . g133050 . t1 . 2NF* DNArepairCre16 . g670550 . t1 . 2XP-G/RAD2 DNA repair endonucleasesynthesis/chromatin structureCre07 . g338000 . t1 . 1MCM2Minichromosome maintenance proteinCre07 . g314900 . t1 . 2ATP-dependent RNA helicase , DEAD/DEAH helicaseCre03 . g172950 . t1 . 1CBF5Centromere/microtubule binding proteinCre01 . g015250 . t1 . 1Eukaryotic DNA polymerase deltaCre27 . g774200 . t1 . 2NF*Cre07 . g316850 . t1 . 1MCM4Minichromosome maintenance proteinunspecifiedCre10 . g451250 . t1 . 2Adenylate and guanylate cyclase catalytic domain , 3-5 exonucleaseCre01 . g059950 . t1 . 2NF**Corresponding gene model was not found in v5 . †Functional terms are inferred by homology to the annotation set of Arabidopsis thaliana ( Lopez et al . , 2011 ) . To better understand the physiology of sak1 , including the primary and secondary effects of lacking SAK1 , we also focused on changes in transcript levels at the basal level , that is , without 1O2 treatment . At basal level 699 genes were induced , and 737 genes were repressed in the mutant compared to WT ( Supplementary file 1 , C3 ) , displaying the genome-wide response to the loss of SAK1 function despite the mutant’s wild-type appearance under normal lab growth conditions ( Figure 1D ) . The enriched classes of genes that are differentially expressed are shown in Figure 3D . Genes induced in the mutant at basal level were enriched for those annotated to be involved in nucleotide metabolism , DNA , and RNA ( Figure 3D; Table 7 ) . Interestingly genes involved in tetrapyrrole and photosynthesis were enriched both in elevated and repressed genes at the basal level in sak1 . There was no overall trend of these two pathways being up- or down-regulated , since these genes were at different steps of the pathway or encoded a select isoform of an enzyme or a subunit of a complex ( Figure 3D; Table 7 ) . 10 . 7554/eLife . 02286 . 013Table 7 . Enriched functional classes among differentially expressed genes in sak1 at basal levelDOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 013Primary Mapman classSecondary Mapman classGene ID ( v4 ) Gene nameAnnotationElevated in sak1 nucleotide metabolismdeoxynucleotide metabolismCre12 . g491050 . t1 . 1RIR2Ribonucleotide reductase ( RNR ) , small subunitCre12 . g492950 . t1 . 1RIR1Ribonucleotide reductase ( RNR ) , large subunit , class ICre16 . g667850 . t1 . 1dUTP pyrophosphatasephosphotransfer and pyrophosphatasesCre02 . g122450 . t1 . 1NF*Cre02 . g093950 . t1 . 1PYR5Uridine 5'- monophosphate synthase/orotate phosphoribosyltransferaseCre12 . g519950 . t1 . 1Flagellar Associated Protein similar to adenylate/guanylate kinasesCre26 . g772450 . t1 . 1NF*synthesisCre65 . g793400 . t1 . 1NF*Cre02 . g079700 . t1 . 1PYR2Aspartate carbamoyltransferaseCre01 . g048950 . t1 . 1dUTP pyrophosphataseCre07 . g318750 . t1 . 1Phosphoribosylformylglycinamidine cyclo-ligase . DNArepairCre07 . g314650 . t1 . 1Chloroplast RecA recombination proteinsynthesis/chromatin structureCre04 . g214350 . t1 . 2Eukaryotic DNA polymerase alpha , catalytic subunitCre07 . g314900 . t1 . 2ATP-dependent RNA helicase ( DEAD/DEAH ) Cre04 . g223850 . t1 . 1Cytoplasmic DExD/H-box RNA helicaseCre01 . g015250 . t1 . 1Eukaryotic DNA polymerase delta , catalytic subunit . Cre07 . g342506 . t1 . 1Ubiquitin-protein ligaseCre07 . g338000 . t1 . 1MCM2Minichromosome maintenance proteinCre03 . g178650 . t1 . 1MCM6MCM6 DNA replication proteinCre07 . g312350 . t1 . 2DNA polymerase alpha , primase subunitCre01 . g009250 . t1 . 2TOP2DNA topoisomerase IICre26 . g772150 . t1 . 1NF*Cre07 . g316850 . t1 . 1MCM4Minichromosome maintenance protein 4Cre06 . g263800 . t1 . 2tRNA-splicing endonuclease positive effector ( SEN1 ) Cre06 . g295700 . t1 . 2MCM3Minichromosome maintenance proteinCre06 . g251800 . t1 . 1RFC4DNA replication factor C complex subunit 4unspecifiedCre07 . g322300 . t1 . 2DNA repair helicase of the DEAD superfamilyCre17 . g718100 . t1 . 1Phosphatidylinositol transfer protein SEC14 and related proteins ( CRAL/TRIO ) Tetrapyrrole synthesisGlu-tRNA reductaseCre07 . g342150 . t1 . 1HEM1Glutamyl-tRNA reductaseGlu-tRNA synthetaseCre44 . g788000 . t1 . 1Glutamyl-tRNA reductaseCre06 . g306300 . t1 . 1CHLI1Magnesium chelatase subunit Imagnesium chelataseCre07 . g325500 . t1 . 1Magnesium chelatase subunit Hprotochlorophyllide reductaseCre01 . g015350 . t1 . 1POR1Light-dependent protochlorophyllide reductase PhotosynthesisCalvin-Benson cycleCre05 . g234550 . t1 . 1Fructose-biphosphate aldolaselight reactionCre07 . g330250 . t1 . 1PSAHSubunit H of photosystem ICre07 . g334550 . t1 . 1Photosystem I subunit PsaOCre06 . g261000 . t1 . 1PSBR10 kDa photosystem II polypeptidephotorespirationCre12 . g542300 . t1 . 1GYK1Glycerate kinaseCre06 . g253350 . t1 . 1GCSHGlycine cleavage system , H-proteinCre06 . g293950 . t1 . 1SHMT2Serine hydroxymethyltransferase 2 TransportABC transporters and multidrug resistance systemsCre04 . g222700 . t1 . 1ATPase component of ABC transporters with duplicated ATPase domains/Translation elongation factor EF-3bCre17 . g728400 . t1 . 2ABCtransporter ( ABC-2 type ) Cre05 . g241350 . t1 . 2ABCtransporter ( ABC-2 type ) Cre03 . g169300 . t1 . 1ABCtransporter ( ABC-2 type ) Cre11 . g474600 . t1 . 1NF*amino acidsCre04 . g226150 . t1 . 2AOC1Amino acid carrier 1; belongs to APC ( Amino acid Polyamine organo Cation ) familycalciumCre09 . g388850 . t1 . 1ACA1P-type ATPase/cation transporter , plasma membranemetabolite transporters at the envelope membraneCre06 . g263850 . t1 . 2TPT2Triose phosphate/phosphate translocatormetabolite transporters at the mitochondrial membraneCre10 . g449100 . t1 . 1Mitochondrial oxodicarboxylate carrier proteinCre01 . g069350 . t1 . 1NF*Cre15 . g641200 . t1 . 1Mitochondrial fatty acid anion carrier protein/Uncoupling proteinCre09 . g396350 . t1 . 1Mitochondrial carrier protein PET8miscCre06 . g311000 . t1 . 2FBT2Folate transporteCre17 . g718100 . t1 . 1Phosphatidylinositol transfer protein SEC14 and related proteins ( CRAL/TRIO ) phosphateCre16 . g686750 . t1 . 1PTA3Proton/phosphate symporterCre16 . g675300 . t1 . 2Sodium-dependent phosphate transporter , major facilitator superfamilypotassiumCre12 . g553450 . t1 . 2NF*sulphateCre17 . g723350 . t1 . 1SUL2Sulfate anion transporterunspecified cationsCre13 . g573900 . t1 . 1Na+:iodide/myo-inositol/multivitamin symporterssugarsCre16 . g675300 . t1 . 2Sodium-dependent phosphate transporter , major facilitator superfamily RNAprocessingCre10 . g427700 . t1 . 1ATP-dependent RNA helicase , DEAD/DEAH box helicaseCre12 . g538750 . t1 . 1LSM1U6 snRNA-associated Sm-like protein LSm1 , RNA cap binding; ( SMP6d ) Cre10 . g433750 . t1 . 2PAP1Nuclear poly ( A ) polymeraseCre03 . g182950 . t1 . 1NF*Cre08 . g375128 . t1 . 1NF*regulation of transcriptionCre17 . g728200 . t1 . 2YL-1 protein ( transcription factor-like 1 ) Cre06 . g275500 . t1 . 1AP2 Transcription factorCre28 . g777500 . t1 . 2NF*Cre13 . g572450 . t1 . 1Response regulator receiver domain ( sensor histidine kinase-related , regulation of transcription ) Cre14 . g620500 . t1 . 1AP2 Transcription factorCre16 . g673150 . t1 . 1Histone deacetylase complex , catalytic component RPD3Cre02 . g078700 . t1 . 2DNA damage-responsive repressor GIS1/RPH1 , jumonji superfamilyCre03 . g198800 . t1 . 1Myb-like DNA-binding domainCre04 . g218050 . t1 . 2RWP-RK domainCre07 . g324400 . t1 . 1VPS24Subunit of the ESCRT-III complex , vaculoar sortin proteinCre11 . g481050 . t1 . 1SWI/SNF-related chromatin binding proteinCre02 . g101950 . t1 . 1TMU2tRNA ( uracil-5 ) -methyltransferaseCre10 . g459600 . t1 . 2CAATT-binding transcription factor/60S ribosomal subunit biogenesis proteinCre01 . g018650 . t1 . 2NF*Cre01 . g012200 . t1 . 2NF*Cre02 . g129750 . t1 . 1NF*Cre10 . g461750 . t1 . 2DNA ( cytosine-5- ) -methyltransferaseCre01 . g004600 . t1 . 2RWP12Putative RWP-RK domain transcription factorCre09 . g400100 . t1 . 1Predicted Zn-finger protein , zinc and DNA binding domainsCre07 . g335150 . t1 . 2SBP domainRNA bindingCre16 . g662700 . t1 . 1NF*Cre07 . g330300 . t1 . 1RNA-binding protein musashi/mRNA cleavage and polyadenylation factor I complex , subunit HRP1Cre06 . g275100 . t1 . 1RNA-binding protein musashi/mRNA cleavage and polyadenylation factor I complex , subunit HRP1transcriptionCre07 . g322200 . t1 . 1NF*Repressed in sak1 TransportABC transporters and multidrug resistance systemsCre02 . g097800 . t1 . 2ABC transporter ( MDR ) Cre17 . g725200 . t1 . 1ABC transporter , peptide exporterCre13 . g580300 . t1 . 1ABC transporter family proteinCre10 . g439000 . t1 . 2Long-chain acyl-CoA transporter , ABC superfamily ( involved in peroxisome organization and biogenesis ) amino acidsCre06 . g292350 . t1 . 1AOC4Amino acid carriercalciumCre06 . g263950 . t1 . 2Sodium/potassium-transporting ATPase subunit alphaCre16 . g681750 . t1 . 2Calcium transporting ATPasemetabolite transporters at the mitochondrial membraneCre03 . g172300 . t1 . 1Mitochondrial phosphate carrier proteinCre09 . g394800 . t1 . 2Mitochondrial substrate carrier proteinmetalCre03 . g189550 . t1 . 2ZIP3Zinc transporter , ZIP familyCre11 . g479600 . t1 . 2Sodium/calcium exchanger NCX1 and related proteinsCre06 . g281900 . t1 . 1ZIP7Zinc transporter and related ZIP domain-containing proteinsmiscCre02 . g089900 . t1 . 1Secretory carrier membrane proteinCre10 . g448050 . t1 . 1Retinaldehyde binding protein-related ( CRAL/TRIO domain ) Cre03 . g177750 . t1 . 2Multidrug resistance pumpNDP-sugars at the ERCre02 . g112900 . t1 . 1GDP-fucose transporter ( Triose-phosphate transporter family ) P- and V-ATPasesCre01 . g027800 . t1 . 1ATPvHVacuolar ATP synthase subunit HCre10 . g446550 . t1 . 1ATPvFVacuolar ATP synthase subunit FCre03 . g176250 . t1 . 1ATPvD1Vacuolar ATP synthase subunit DCre06 . g250250 . t1 . 1ATPvCVacuolar ATP synthase subunit CCre10 . g459200 . t1 . 1ACA4P-type ATPase/cation transporter , plasma membrane ( Low CO2 inducible gene ) phosphateCre12 . g515750 . t1 . 2Sodium-dependent phosphate transporter-relatedCre08 . g379550 . t1 . 2Sodium-dependent phosphate transporter , major facilitator superfamilyCre12 . g489400 . t1 . 1PTB7Putative phosphate transporter , sodium/phosphate transporterCre02 . g144650 . t1 . 1PTB12Sodium/phosphate symporterunspecified anionsCre09 . g404100 . t1 . 1Cl- channel CLC-7 and related proteins ( CLC superfamily ) Cre17 . g729450 . t1 . 1Cl- channel CLC-7 and related proteins ( CLC superfamily ) Cre01 . g037150 . t1 . 2Voltage-gated chloride channel activitysugarsCre03 . g206800 . t1 . 2HXT1Hexose transporterP- and V-ATPasesCre03 . g176250 . t1 . 1ATPvD1Vacuolar ATP synthase subunit DCre10 . g446550 . t1 . 1ATPvFVacuolar ATP synthase subunit FCre01 . g027800 . t1 . 1ATPvHVacuolar ATP synthase subunit H Mitochondrial electron transport / ATP synthesiscytochrome c reductaseCre01 . g051900 . t1 . 1RIP1Rieske iron-sulfur protein of mitochondrial ubiquinol-cytochrome c reductase ( complex III ) Cre06 . g262700 . t1 . 2Ubiquinol cytochrome c reductase , subunit 7F1-ATPaseCre02 . g116750 . t1 . 2F0F1-type ATP synthase , alpha subunitCre01 . g018800 . t1 . 1ATP6Mitochondrial F1F0 ATP synthase subunit 6Cre10 . g420700 . t1 . 1Mitochondrial F1F0-ATP synthase , subunit epsilon/ATP15Cre16 . g680000 . t1 . 1ATP5Mitochondrial ATP synthase subunit 5 , OSCP subunitNADH-DHCre10 . g434450 . t1 . 1NUOA9Putative NADH:ubiquinone oxidoreductase ( Complex I ) 39 kDa subunitCre08 . g378900 . t1 . 1NUO3NADH:ubiquinone oxidoreductase ND3 subunitCre10 . g450400 . t1 . 1NUO5NADH:ubiquinone oxidoreductase ( Complex I ) 24 kD subunit Lipid metabolism'exotics' ( steroids , squalene etc ) Cre14 . g615050 . t1 . 13-oxo-5-alpha-steroid 4-dehydrogenase , Steroid reductase required for elongation of the VLCFAs ( enoyl reductase ) Cre12 . g530550 . t1 . 2KDG2Diacylglycerol kinase , sphingosine kinaseCre02 . g137850 . t1 . 1NF*FA desaturationCre17 . g711150 . t1 . 1Omega-6 fatty acid desaturase ( delta-12 desaturase ) glyceral metabolismCre13 . g577450 . t1 . 2Glycerol-3-phosphate dehydrogenaseglycolipid synthesisCre13 . g583600 . t1 . 1DGD1Digalactosyldiacylglycerol synthaseCre16 . g656400 . t1 . 1SQD1UDP-sulfoquinovose synthaselipid degradationCre06 . g252801 . t1 . 2CGI-141-related/lipase containing protein ( TAG lipase ) Cre03 . g164350 . t1 . 2Lysophospholipase , putative drug exporter of the RND superfamilyphospholipid synthesisCre06 . g281250 . t1 . 1CFA1Cyclopropane fatty acid synthaseCre09 . g398700 . t1 . 1CFA2Cyclopropane fatty acid synthaseCre11 . g472700 . t1 . 1NF*Cre06 . g262550 . t1 . 1Zinc finger MYND domain containing protein 10 PhotosynthesisCalvin-Benson cycleCre12 . g511900 . t1 . 1RPE1Ribulose phosphate-3-epimeraseCre02 . g120100 . t1 . 1RBCS1Ribulose-1 , 5-bisphosphate carboxylase/oxygenase small subunit 1light reactionCre05 . g243800 . t1 . 1CPLD45Photosystem II Psb27 proteinCre10 . g420350 . t1 . 1PSAEPhotosystem I reaction center subunit IVCre01 . g071450 . t1 . 2NF*Cre06 . g291650 . t1 . 1FerredoxinCre05 . g242400 . t1 . 1No functional annotationphotorespirationCre09 . g411900 . t1 . 2SHMT3Serine hydroxymethyltransferase 3Cre06 . g295450 . t1 . 1HPR1Hydroxypyruvate reductaseMajor CHO metabolismdegradationCre09 . g415600 . t1 . 2Starch binding domainCre11 . g473500 . t1 . 2NF*Cre09 . g415600 . t1 . 2Starch binding domainsynthesisCre06 . g289850 . t1 . 2SBE1Starch Branching EnzymeCre17 . g721500 . t1 . 1Granule-bound starch synthase I miscacid and other phosphatasesCre13 . g568600 . t1 . 2Multiple inositol polyphosphate phosphatase-related , Acid phosphatase activityalcohol dehydrogenasesCre13 . g569350 . t1 . 1Sterol dehydrogenase-related , Flavonol reductase/cinnamoyl-CoA reductasecytochrome P450Cre07 . g356250 . t1 . 2Cytochrome P450 CYP4/CYP19/CYP26 subfamilies , beta-carotene 15 , 15'-monooxygenaseCre07 . g356250 . t1 . 2Cytochrome P450 CYP4/CYP19/CYP26 subfamilies , beta-carotene 15 , 15'-monooxygenasedynaminCre17 . g724150 . t1 . 1DRP3Dynamin-related GTPaseGCN5-related N-acetyltransferaseCre16 . g657150 . t1 . 2N-acetyltransferase activity ( GNAT ) familygluco- , galacto- and mannosidasesCre03 . g171050 . t1 . 2GHL1Glycosyl hydrolasemisc2Cre14 . g614100 . t1 . 1GTR26Dolichyl-diphosphooligosaccharide-protein glycosyltransferaserhodaneseCre07 . g352550 . t1 . 1RDP3Putative rhodanese domain phosphataseshort chain dehydrogenase/reductase ( SDR ) Cre07 . g352450 . t1 . 1Corticosteroid 11-beta-dehydrogenase and related short chain-type dehydrogenases , 3-hydroxybutyrate dehydrogenaseCre12 . g559350 . t1 . 11-Acyl dihydroxyacetone phosphate reductase and related dehydrogenasesCre03 . g191850 . t1 . 1Short chain dehydrogenaseUDP glucosyl and glucoronyl transferasesCre11 . g474450 . t1 . 1NF*Cre03 . g205250 . t1 . 2ELG4Exostosin-like glycosyltransferaseCre16 . g659500 . t1 . 1Lactosylceramide 4-alpha-galactosyltransferaseCre11 . g483400 . t1 . 2ELG10Exostosin-like glycosyltransferase Tetrapyrrole synthesisGlu-tRNA synthetaseCre12 . g510800 . t1 . 1CHLI2Magnesium-chelatase subunit chlImagnesium protoporphyrin IX methyltransferaseCre12 . g498550 . t1 . 2Magnesium protoporphyrin IX S-adenosyl methionine O-methyl transferase ( Magnesium-protoporphyrin IX methyltransferase ) ( PPMT ) unspecifiedCre12 . g516350 . t1 . 1COX10Cytochrome c oxidase assembly protein Cox10urogen III methylaseCre02 . g133050 . t1 . 2NF**Corresponding gene model was not found in v5 . We observed that some of the genes more strongly dependent on SAK1 had repressed transcript levels ( e . g . , CFA1 and SOUL2 ) , indicating that SAK1 is required for their basal expression , while others had elevated basal levels ( GPX5 ) , suggesting that expression of these genes is controlled also by other pathways . As is discussed in the following section , SAK1 expression monitored by qRT-PCR followed the latter trend as the 5′UTR of the gene was elevated in the mutant ( Figure 4E ) , which may be a result of response to other factors such as a possible oxidization product of 1O2 . The SAK1-dependent genes induced by 1O2 and repressed at basal level in the mutant ( i . e . , those that require SAK1 for basal expression ) are indicated in Table 5 . 10 . 7554/eLife . 02286 . 014Figure 4 . Genetic and molecular analysis of sak1 . ( A ) The insertion of a zeocin resistance gene and the RB sensitivity phenotype are linked . Twelve complete tetrads from a backcross of sak1 to wild type are shown . Numbers indicate independent tetrads , and letters ( a-d ) indicate the individual progeny from tetrads . ( B ) Gene structure of SAK1 and the insertion site . Gray boxes indicate positions of primers used for qPCR . ( C ) Transformation of sak1 with a genomic fragment containing SAK1 rescues the acclimation phenotype . sak1 ( gSAK1 ) -1 and sak1 ( gSAK1 ) -2 are two independent transformants . ( D ) sak1 ( gSAK1 ) -1 and sak1 ( gSAK1 ) -2 show recovery of 1O2 target gene expression . Y-axis indicates fold change during acclimation to 1O2 . ( E ) qRT-PCR of SAK1 in WT and sak1 mutant using primers for 5′- and 3′-UTR shown in panel B . ( F ) SAK1 protein is induced in WT and detected as higher molecular weight bands during acclimation to 1O2 generated by RB . ( G ) SAK1 transcript probed for 5′-UTR in cells transferred from low light to high light for 1 hr . Error bars indicate standard deviation of biological triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 014 The sak1 mutant was generated by insertional mutagenesis using a plasmid that confers resistance to zeocin ( Dent et al . , 2005 ) . Progeny obtained from a backcross of sak1 with WT showed that the mutation causing the RB sensitivity phenotype was linked to zeocin resistance ( Figure 4A ) . The site of insertion was identified by thermal asymmetric interlaced ( TAIL ) -PCR ( Liu et al . , 1995 ) as the second exon of the annotated gene Cre17 . g741300 on chromosome 17 ( Figure 4B ) . To test whether this gene is responsible for the mutant phenotype , a genomic fragment containing the gene with an additional ∼500 bp region upstream of the predicted transcription start site was cloned and introduced into the mutant by co-transformation . Among the approximately 300 transformants screened , two clones appeared to have recovered the RB acclimation phenotype ( Figure 4C ) . Furthermore , induction of genes we found attenuated in sak1 ( Figure 2 ) was restored in these transformants ( Figure 4D ) , confirming that Cre17 . g741300 is the SAK1 gene required for acclimation and the gene expression response to 1O2 . In WT , the SAK1 gene itself was induced by 6- to 10-fold during acclimation when probed for the 5′-and 3′-UTR of the transcript by qRT-PCR ( Figure 4E ) . The mutant displayed elevated basal level and induction of the 5′-UTR during acclimation , whereas the 3′-UTR of the transcript was undetectable , indicating that the full-length transcript was absent in sak1 ( Figure 4E ) . An antibody raised against an epitope of the SAK1 protein detected a single band in basal conditions , whereas the SAK1 protein appeared as multiple bands with higher molecular weight in acclimated WT cells , all of which were absent in the mutant ( Figure 4F ) . SAK1 transcript was induced when probed for the 5′-UTR during high light exposure in both WT and sak1 ( Figure 4G ) similarly to other 1O2-response genes identified by RNA-seq ( Table 1 ) , indicating that SAK1 itself is part of the endogenous response to high light . The predicted SAK1 protein consists of 1141 amino acid residues and has no domains with functional annotation . Only a ∼150-residue region at the C-terminus , designated the SAK1 domain , has similarity to other proteins . Many predicted proteins within chlorophytes ( Volvox carteri [8 proteins] , Coccomyxa subellipsoidea [3 proteins] , Chlamydomonas [14 proteins] , Chlorella variabilis [9 proteins] and Micromonas [3 proteins] ) ( Table 8 ) contain this domain as shown in the alignment in Figure 5—figure supplement 1 . Among the 37 members of the chlorophyte SAK1 domain family , 13 have possible bZIP transcription factor domains ( six were significant Pfam hits and seven were below the threshold for significance but recognizable by Pfam ) ( Figure 5 ) . One protein contained a mitochondrial ( transcription ) termination factor ( mTERF ) domain ( Figure 5 ) , defined by its three leucine zipper domains required for DNA binding ( Fernandez-Silva et al . , 1997 ) . Proteins with more distantly related SAK1 domains were found by PSI-BLAST in plants , many of which were hypothetical or unknown proteins but also included bZIP transcription factors . 10 . 7554/eLife . 02286 . 015Table 8 . SAK1 domain containing proteins in chlorophytesDOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 015Number in alignmentOrganismTranscript/Protein IDaTranscript/Protein IDaTranscript/Protein ID*1Volvox carteriVocar200092352Volvox carteriVocar200024373Volvox carteriVocar200026724Volvox carteriVocar200049235Volvox carteriVocar200123496Volvox carteriVocar200059887Volvox carteriVocar200071588Volvox carteriVocar200078839Coccomyxa subellipsoidea5740510Coccomyxa subellipsoidea5965511Coccomyxa subellipsoidea5769412Chlamydomonas reinhardtiiCre16 . g652650 . t1 . 313Chlamydomonas reinhardtiiCre06 . g271000 . t1 . 214Chlamydomonas reinhardtiiCre06 . g285800 . t1 . 215Chlamydomonas reinhardtiiCre06 . g275600 . t1 . 216Chlamydomonas reinhardtiiCre06 . g285750 . t1 . 317Chlamydomonas reinhardtiiCre06 . g270950 . t1 . 218Chlamydomonas reinhardtiig9774 . t1SAK1Chlamydomonas reinhardtiiKF98524220Chlamydomonas reinhardtiiCre03 . g179150 . t1 . 221Chlamydomonas reinhardtiig3701 . t122Chlamydomonas reinhardtiiCre03 . g179250 . t1 . 223Chlamydomonas reinhardtiiCre03 . g179200 . t1 . 224Chlamydomonas reinhardtiiCre01 . g004800 . t1 . 225Chlamydomonas reinhardtiiCre01 . g048550 . t1 . 326Chlorella variabilisEFN5126027Chlorella variabilisEFN5349628Chlorella variabilisEFN5561829Chlorella variabilisEFN5765230Chlorella variabilisEFN5565831Chlorella variabilisEFN5426232Chlorella variabilisEFN5451033Chlorella variabilisEFN5580634Chlorella variabilisEFN5349235Micromonas sp . RCC299ACO6134736Micromonas pusilla CCMP1545EEH5779137Micromonas sp . RCC299ACO65814*1–25 , as defined on phytozome . net; 26–37 , CrSAK1 , genbank accession numbers . 10 . 7554/eLife . 02286 . 016Figure 5 . SAK1 contains an uncharacterized domain present in some bZIP transcription factors . Schematic of relative positions of SAK1 and bZIP domains . One protein ( Cv28 ) contains a mitochondrial termination factor ( mTERF ) domain . The letters and numbers in the abbreviated names represent initials of the species and numbers listed in Table 8 . Proteins with italicized names contain bZIP domains that were recognized by Pfam but scored below significance . DOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 01610 . 7554/eLife . 02286 . 017Figure 5—figure supplement 1 . Multiple sequence alignment of SAK1 domains . The SAK1 domains of 37 chlorophyte proteins were aligned by MUSCLE ( phylogeny . fr ) . Protein identities are as shown in Table 8 . Star indicates a relatively conserved residue within the SAK1 domain that was predicted to be a possible phosphorylation site ( Figure 5—figure supplement 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 01710 . 7554/eLife . 02286 . 018Figure 5—figure supplement 2 . Secondary structure prediction of SAK1 domain . SAK1 domain modeled against its best-hit nickel cobalt resistance protein cnrr by PHYRE . 44% ( coverage ) of the SAK1 domain was aligned with 73 . 6% confidence . DOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 01810 . 7554/eLife . 02286 . 019Figure 5—figure supplement 3 . Prediction of phosphorylation sites in SAK1 . Prediction of phosphorylation sites by NetPhos 2 . 0 . Orange bar indicates the position of SAK1 domain , star indicates a relatively conserved residue among the 37 members containing the SAK1 domain . DOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 019 Amino acid positions 900 to 1089 of SAK1 , corresponding to the region aligned with other proteins in Figure 5—figure supplement 1 , were searched for secondary structure using PHYRE , and this region was predicted to consist of mostly alpha helices with some disordered intervals . The top hit was a cobalt/nickel-binding resistance protein cnrr , and 44% of the residues were modeled with 73 . 6% confidence ( Figure 5—figure supplement 2 ) . To obtain insight into the function of SAK1 , we isolated subcellular fractions enriched for chloroplast , ER , cytosol , and mitochondria from WT cells . The Chlamydomonas cell contains a single large chloroplast that is physically connected to other organelles such as the ER , making it particularly challenging to fractionate . The patterns of markers specific for chloroplast , ER , cytosol , and mitochondria showed that each target fraction was enriched as expected , although with some cross contamination ( Figure 6A , B ) . The distribution of SAK1 in these fractions resembled most closely that of the cytosolic marker NAB1 ( Mussgnug et al . , 2005 ) , although the SAK1 signal was not as enriched as NAB1 in the cytosolic fraction , possibly due to partial degradation of SAK1 during the fractionation . The localization was the same in cells with and without RB treatment ( Figure 6A ) . Because SAK1 was required for the induction of many genes during acclimation to 1O2 and the list of proteins with similarity to SAK1 included those predicted to be bZIP transcription factors , we tested whether SAK1 protein was dually targeted to the nucleus and cytosol , which would account for the lack of enrichment of SAK1 in the cytosolic fraction ( Figure 6A ) . As shown in Figure 6C although a faint SAK1 signal was detected in nuclear fraction , there was no enrichment as seen for the nuclear marker histone H3 ( H3 ) . The distribution of the cytosolic marker NAB1 indicated the contamination of the nuclear fraction by cytosolic proteins ( Figure 6C ) . Therefore we conclude that the low signal of SAK1 in the nuclear fraction is likely to be due to cytosolic contamination . Attempts to detect the protein by immunofluorescence using anti-SAK1 antibodies as well as anti-FLAG and anti-HA antibodies against tagged proteins in transgenic lines were unsuccessful due to a very low signal-to-noise ratio even in bleached cells . 10 . 7554/eLife . 02286 . 020Figure 6 . SAK1 is a phosphorylated protein that is in the cytosol . ( A and B ) SAK1 is detected in the cytosol and not in other subcellular fractions . ( C ) SAK1 is not enriched in nuclear extracts . Approximately 30 μg of protein was loaded into each well except for mitochondrial fractions that were loaded approximately 7 . 5 μg protein due to low protein yield in isolated fractions . Subcellular markers: Chloroplast ( CP ) , PSAD; Endoplasmic reticulum ( ER ) , KDEL; Cytosol , NAB1; Mitochondria ( mito ) , cytochrome c ( Cyt c ) ; Nuclear , histone 3 ( H3 ) . The arrowhead indicates the band corresponding to Cyt c . ( D ) Protein extracts from cells treated with increasing concentrations of RB were then treated with phosphatase ( + ) or only with buffer ( − ) before detection of SAK1 by immunoblot analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 020 By SDS-PAGE and immunoblot analysis , SAK1 appeared in multiple forms with higher molecular weight during acclimation compared to that observed in control cells ( Figures 4F and 6A , C ) . When the extracted protein samples were treated with phosphatase , the diffuse pattern of multiple forms collapsed into a single band detected by immunoblot analysis that had an even higher mobility that that of untreated cells ( Figure 6D ) . This result indicates that SAK1 is a phosphorylated protein during basal conditions , and it is further phosphorylated upon exposure of cells to 1O2 .
To understand the retrograde signal transduction pathway involved in the cellular response to 1O2 , we focused on the unique ability of Chlamydomonas to acclimate to 1O2 stress ( Ledford et al . , 2007 ) , and we isolated a regulatory mutant that is unable to acclimate . Several previous genetic screens aimed at dissecting the mechanisms of 1O2 signaling have concentrated on the nuclear gene expression response to 1O2 , often relying on the response of a single marker gene ( Baruah et al . , 2009a; Brzezowski et al . , 2012; Fischer et al . , 2012; Shao et al . , 2013 ) . In contrast , our screen exploited a physiological response to sublethal levels of 1O2 , which induces the wild type to survive a subsequent , otherwise lethal treatment with the 1O2 generator RB ( Ledford et al . , 2007 ) . The sak1 mutant completely lacks this ability to acclimate to 1O2 ( Figure 1A ) . An analogous phenotype is exhibited by the yap1Δ mutant of Saccharomyces cerevisiae , which is unable to acclimate to hydrogen peroxide stress ( Stephen et al . , 1995 ) . In contrast to the complete loss of acclimation to RB , sak1 acclimates ( but less effectively than WT ) when pretreated with high light and challenged with RB ( Figure 1B ) . This result suggests that the high light pretreatment induces a broader response than that elicited by RB and that sak1 is still able to respond to other signals besides 1O2 ( e . g . , plastoquinone redox state , H2O2 , and/or superoxide ) that are involved in the response to high light . When tested on TAP agar plates for photoheterotrophic growth in the presence of various photosynthetic inhibitors , the sak1 mutant displayed sensitivity to RB but not to other inhibitors ( Figure 1D ) . In particular , sak1 is not more sensitive than WT to high light or norflurazon ( an inhibitor of the biosynthesis of carotenoids , which function as quenchers of 1O2 ) . We speculate that the lack of 1O2-sensitive phenotype in these plate experiments is attributable to the time-scale of the treatments involved . 1O2 generated by RB or during a transfer to higher light intensity is transient , whereas NF requires longer time to exert its effect because it needs to enter the cell , inhibit biosynthesis , and deplete cells of existing carotenoids . During this time , the cell is likely able to acclimate by detoxifying and reducing the generation of 1O2 by various means such as changing the composition of the photosynthetic apparatus . We have previously shown that acclimation to 1O2 is transient and is dissipated by 24 hr post-treatment ( Ledford et al . , 2007 ) . Consistent with this , pretreatment with RB does not acclimate the cells to stresses such as growth in high light or norflurazon that require a period of days to assess an effect on viability ( Figure 1—figure supplement 1 ) . We have also observed that under our experimental conditions , the induction of target gene expression upon exposure to 1O2 lasts up to 90 min and then declines . We conclude that SAK1 functions mainly during transient perturbations that generate 1O2 . However , during steady-state growth under high light or norflurazon , the cell is able to cope by other means that do not involve SAK1 . A physiological acclimation response that results in such an evident growth phenotype ( Figure 1A ) likely involves large-scale changes in gene expression , and transcriptome analysis of wild-type cells showed that hundreds of nuclear genes are up- or down-regulated during acclimation to 1O2 ( Figure 3A , B; Supplementary file 1 , C1 ) . The sak1 mutant is specifically impaired in regulation of a notable subset of these genes , that is , those that are most strongly induced in the wild type ( Table 5 ) , suggesting that these genes play a key role in the acclimation response to 1O2 . In particular , many genes involved in sterol and lipid metabolism were induced by 1O2 in Chlamydomonas ( Figure 3B; Table 3 ) . For example , two genes encoding putative cyclopropane fatty acid synthase ( CFA1 and CFA2 ) exhibited SAK1-dependent induction ( Figure 2 ) . Cyclopropane fatty acids have been found in large amounts in the seeds of Sterculia foetida ( Bao et al . , 2002 ) , although its biological function is unknown . In bacteria , it has been implicated in oxidative stress responses ( Guerzoni et al . , 2001; Kim et al . , 2005 ) and particularly in the anoxygenic photosynthetic bacterium Rhodobacter sphaeroides , CFA gene expression is induced during 1O2 stress by a σE factor ( Ziegelhoffer and Donohue , 2009 ) . Interestingly CFA mutants of R . sphaeroides are compromised in the induction of genes in response to 1O2 , suggesting a regulatory role of the gene , protein , or the product of its enzymatic function ( cyclopropane fatty acids , Bao et al . , 2002 ) in gene expression rather than solely a biochemical stress response ( Nam et al . , 2013 ) . Another intriguing class of up-regulated genes enriched during 1O2 acclimation in WT and not in sak1 was a group of genes encoding transporters , especially ABC transporters related to the MDR and PDR types . This was not surprising considering that 1O2 exists in aquatic and terrestrial environments , where it is generated by photosensitizing humic substances ( Frimmel et al . , 1987; Steinberg et al . , 2008 ) , which are known to affect microbial populations including phytoplankton ( Glaeser et al . , 2010 , 2014 ) . Assuming that some of these transporters function to export photosensitizing molecules from the cell , our results suggest that removal of photosensitizers is an integral part of the 1O2 response in Chlamydomonas , rather than simply a response to the presence of a xenobiotic compound such as RB ( Table 4 ) . It is likely that Chlamydomonas , a soil-dwelling microalga , needs to respond to 1O2 that is generated not only in the chloroplast , but also in other compartments . In this context , it is noteworthy that a recent study has demonstrated light-independent 1O2 generation in multiple organelles other than the chloroplast under various biotic and abiotic stresses in plants ( Mor et al . , 2014 ) . Two proteins with SOUL heme-binding domains were among SAK1-dependent up-regulated genes ( SOUL2 and Cre06 . g299700 . t1 . 1 , formerly annotated as SOUL1 in v4 ) . Aside from their ability to bind various porphyrins ( Blackmon et al . , 2002; Sato et al . , 2004 ) , SOUL heme-binding proteins have been described in diverse biological functions in mice , such as in apoptosis by interacting with a mitochondrial anti-apoptotic factor Bcl-xL ( Ambrosi et al . , 2011 ) or an isoform-specific role in retina and pineal gland ( Zylka and Reppert , 1999 ) . The latter form is suggested to play a role in transporting heme or by binding free heme to prevent oxidative stress ( Sato et al . , 2004 ) . In Arabidopsis a chloroplast-localized SOUL5 protein has been shown to interact with a heme oxygenase , HY1 , and mutation of the gene encoding SOUL5 causes oxidative stress ( Lee et al . , 2012 ) . Chlamydomonas contains five putative SOUL heme-binding proteins , only one of which contains an amino-terminal chloroplast transit peptide . The two SOUL protein genes induced by 1O2 in our study do not seem to be targeted to the chloroplast , and they may function in the cytosol where SAK1 resides . It would be interesting to test whether these proteins bind porphyrins and are required for 1O2 acclimation . A recent study reported the role of bilins in retrograde signaling in Chlamydomonas through characterization of heme oxygenase mutants disrupted in bilin biosynthesis and transcriptome analyses during dark to light transitions ( Duanmu et al . , 2013 ) . The transcriptome changes indicated that much of the cell’s response during a dark-to-light transition ( DL ) involves photo-oxidative stress . Interestingly , among the 515 genes up-regulated in WT during 1O2 acclimation , 144 genes overlapped with those that are induced during DL ( Table 9 ) . Focusing on the 104 genes that we defined as SAK1-dependent ( Table 5 ) , 31 genes overlapped ( Table 9 ) . CFA1 , CFA2 , and SOUL2 were among these genes , suggesting that a part of the gene expression response to DL in Chlamydomonas is a response to 1O2 . SAK1 itself was also up-regulated during DL as was SOR1 , which encodes a more broadly oxidative stress-responsive bZIP transcription factor ( Fischer et al . , 2012 ) . We found that 64 of the genes induced during acclimation to 1O2 were also up-regulated in the gain-of-function sor1 mutant ( Fischer et al . , 2012 ) . However , the most strongly induced SAK1-dependent genes were not among these genes , except for GPX5 , consistent with the idea that SAK1 and SOR1 function in different pathways . 10 . 7554/eLife . 02286 . 021Table 9 . Genes up-regulated during both 1O2 acclimation and dark to light transitionDOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 021Gene ID ( v4 ) Gene nameAnnotationRB ( log2 ) DL ( log2 ) ( Duanmu et al . , 2013 ) Cre02 . g137700 . t1 . 1*6 . 492 . 34Cre06 . g281250 . t1 . 1*CFA1cyclopropane fatty acid synthase5 . 924 . 49Cre01 . g033300 . t1 . 1*5 . 723 . 62Cre13 . g566850 . t1 . 1*SOUL2SOUL heme-binding protein5 . 532 . 25Cre13 . g600650 . t1 . 1*4 . 763 . 26Cre06 . g263550 . t1 . 1*LCI7R53 . 5-related protein4 . 465 . 27Cre07 . g342100 . t1 . 1*4 . 431 . 84Cre09 . g398700 . t1 . 1*CPLD27coclaurine N-methyltransferase4 . 051 . 36Cre12 . g492650 . t1 . 1*FAS2fasciclin-like protein4 . 019 . 24Cre08 . g381510 . t1 . 1*3 . 943 . 27Cre10 . g458450 . t1 . 2*GPX5glutathione peroxidase3 . 913 . 08Cre11 . g474600 . t1 . 1*3 . 901 . 99Cre13 . g600700 . t1 . 1*3 . 785 . 79Cre14 . g613950 . t1 . 1*3 . 652 . 68Cre06 . g269300 . t1 . 1*3 . 501 . 99Cre08 . g380300 . t1 . 2*MSRA3peptide methionine sulfoxide reductase3 . 451 . 79Cre01 . g031650 . t1 . 2*CGLD12protein with potential galactosyl transferase activity3 . 304 . 90Cre14 . g629061 . t1 . 1*3 . 251 . 88Cre13 . g564900 . t1 . 1*3 . 223 . 38Cre13 . g586450 . t1 . 13 . 213 . 50Cre02 . g139500 . t1 . 1*3 . 042 . 12Cre19 . g756100 . t1 . 13 . 046 . 53Cre01 . g036000 . t1 . 23 . 021 . 16Cre14 . g618400 . t1 . 1*2 . 972 . 16Cre17 . g741300 . t1 . 2*2 . 881 . 92Cre16 . g648700 . t1 . 2*2 . 792 . 35Cre17 . g729950 . t1 . 12 . 772 . 61Cre17 . g721000 . t1 . 12 . 702 . 12Cre06 . g263500 . t1 . 1*2 . 673 . 37Cre01 . g016150 . t1 . 1*2 . 652 . 92Cre08 . g380000 . t1 . 1*2 . 593 . 74Cre04 . g224800 . t1 . 1VAMP74R-SNARE protein , VAMP72-family2 . 583 . 34Cre03 . g210150 . t1 . 12 . 573 . 44Cre14 . g615600 . t1 . 1*2 . 532 . 40Cre06 . g293100 . t1 . 1Qc-SNARE SYP6-like protein2 . 484 . 90Cre08 . g368950 . t1 . 1DHQS3-dehydroquinate synthase2 . 392 . 49Cre10 . g424350 . t1 . 2metalloprotease2 . 373 . 18Cre12 . g537225 . t1 . 12 . 343 . 39Cre07 . g336900 . t1 . 22 . 322 . 31Cre16 . g664050 . t1 . 12 . 311 . 88Cre16 . g677750 . t1 . 12 . 042 . 22Cre12 . g537227 . t1 . 12 . 003 . 46Cre17 . g737050 . t1 . 1RabGAP/TBC protein1 . 992 . 32Cre06 . g297450 . t1 . 11 . 931 . 46Cre06 . g258600 . t1 . 1*1 . 913 . 63Cre16 . g663950 . t1 . 1SC5D , C-5 sterol desaturase1 . 892 . 03Cre13 . g588150 . t1 . 11 . 866 . 21Cre17 . g722150 . t1 . 1PKS3type III polyketide synthase1 . 851 . 61Cre16 . g688550 . t1 . 1GSTS1glutathione-S-transferase1 . 841 . 20Cre03 . g207800 . t1 . 11 . 847 . 09Cre10 . g444550 . t1 . 1*SPP1Asignal peptide peptidase1 . 815 . 33Cre13 . g602500 . t1 . 21 . 761 . 59Cre03 . g163400 . t1 . 2*1 . 762 . 15Cre10 . g450000 . t1 . 11 . 742 . 18Cre01 . g015500 . t1 . 11 . 721 . 55Cre02 . g105750 . t1 . 21 . 713 . 23Cre01 . g061750 . t1 . 1SPT2serine palmitoyltransferase1 . 712 . 29Cre83 . g796250 . t1 . 11 . 681 . 59Cre16 . g656150 . t1 . 11 . 673 . 55Cre01 . g002050 . t1 . 21 . 663 . 15Cre12 . g556750 . t1 . 2Tic32-like 1Short-chain dehydrogenase , classical family , similar to PsTic321 . 663 . 15Cre12 . g559100 . t1 . 11 . 663 . 11Cre09 . g411750 . t1 . 21 . 611 . 96Cre11 . g482650 . t1 . 21 . 573 . 40Cre06 . g310500 . t1 . 1*1 . 576 . 23Cre09 . g397900 . t1 . 1transmembrane protein1 . 562 . 02Cre04 . g215600 . t1 . 11 . 532 . 64Cre02 . g093800 . t1 . 11 . 514 . 99Cre02 . g093750 . t1 . 1NRX2Nucleoredoxin 21 . 506 . 26Cre01 . g004350 . t1 . 11 . 502 . 29Cre01 . g034600 . t1 . 11 . 502 . 22Cre11 . g472600 . t1 . 21 . 482 . 00Cre12 . g500500 . t1 . 2SMT1sterol-C24-methyltransferase1 . 463 . 05Cre13 . g577950 . t1 . 1VPS6subunit of the ESCRT-III complex1 . 452 . 36Cre02 . g118200 . t1 . 11 . 442 . 79Cre01 . g012500 . t1 . 1PRA1prenylated rab acceptor family protein1 . 432 . 46Cre12 . g521600 . t1 . 21 . 422 . 89Cre03 . g179100 . t1 . 1ubiquitin fusion degradation protein1 . 413 . 38Cre09 . g413150 . t1 . 21 . 394 . 31Cre13 . g572200 . t1 . 1tyrosine/tryptophan transporter protein1 . 392 . 57Cre03 . g185850 . t1 . 2PfkB-type carbohydrate kinase1 . 373 . 05Cre18 . g743600 . t1 . 11 . 371 . 65Cre02 . g076800 . t1 . 1sterol reductase1 . 362 . 41Cre06 . g256750 . t1 . 1FAT1acyl carrier protein thioesterase1 . 351 . 67Cre17 . g729450 . t1 . 11 . 341 . 90Cre11 . g471550 . t1 . 11 . 343 . 29Cre09 . g395750 . t1 . 21 . 332 . 87Cre14 . g617100 . t1 . 11 . 333 . 33Cre16 . g691500 . t1 . 1Sec14p-like lipid-binding protein1 . 332 . 28Cre02 . g079550 . t1 . 1DRP2Dynamin-related GTPase1 . 322 . 34Cre02 . g079300 . t1 . 1VPS4AAA-ATPase of VPS4/SKD1 family1 . 321 . 96Cre05 . g231700 . t1 . 21 . 312 . 40Cre02 . g132300 . t1 . 2DNJ12DnaJ-like protein1 . 302 . 24Cre69 . g794101 . t1 . 11 . 302 . 65Cre13 . g565600 . t1 . 21 . 293 . 42Cre13 . g593700 . t1 . 1monooxygenase , DBH-like1 . 291 . 81Cre12 . g498000 . t1 . 21 . 283 . 88Cre06 . g292900 . t1 . 21 . 282 . 16Cre08 . g372100 . t1 . 1HSP70AHeat shock protein 7A1 . 272 . 28Cre01 . g039350 . t1 . 1NCR2NADPH-cytochrome P45 reductase1 . 262 . 19Cre03 . g211100 . t1 . 11 . 262 . 11Cre17 . g731800 . t1 . 11 . 251 . 78Cre17 . g730650 . t1 . 11 . 252 . 28Cre02 . g123000 . t1 . 21 . 241 . 42Cre05 . g247700 . t1 . 21 . 242 . 71Cre08 . g360800 . t1 . 2haloacid dehalogenase-like hydrolase1 . 234 . 39Cre07 . g350750 . t1 . 1PTOX1alternative oxidase1 . 223 . 32Cre17 . g703750 . t1 . 11 . 202 . 21Cre06 . g306041 . t1 . 11 . 202 . 90Cre02 . g116650 . t1 . 11 . 202 . 83Cre08 . g379400 . t1 . 21 . 183 . 04Cre16 . g677000 . t1 . 1HSP70EHeat shock protein 7E1 . 182 . 50Cre06 . g283900 . t1 . 11 . 185 . 24Cre14 . g626750 . t1 . 11 . 174 . 12Cre01 . g010700 . t1 . 11 . 162 . 10Cre01 . g002000 . t1 . 2predicted proteim1 . 151 . 68Cre04 . g213150 . t1 . 11 . 152 . 78Cre16 . g694250 . t1 . 11 . 152 . 92Cre05 . g246400 . t1 . 11 . 152 . 74Cre02 . g128450 . t1 . 11 . 132 . 82Cre03 . g180250 . t1 . 1Myo-inositol-1-phosphate synthase1 . 132 . 05Cre03 . g186150 . t1 . 11 . 131 . 78Cre02 . g137800 . t1 . 11 . 132 . 00Cre11 . g471500 . t1 . 1MFT10predicted protein1 . 111 . 40Cre10 . g435200 . t1 . 11 . 102 . 13Cre13 . g593850 . t1 . 21 . 103 . 91Cre19 . g754000 . t1 . 21 . 102 . 33Cre13 . g593869 . t1 . 11 . 103 . 90Cre08 . g377300 . t1 . 21 . 093 . 27Cre04 . g225050 . t1 . 2predicted protein1 . 093 . 55Cre07 . g330300 . t1 . 11 . 082 . 22Cre12 . g500450 . t1 . 21 . 083 . 00Cre06 . g262000 . t1 . 11 . 081 . 87Cre10 . g441550 . t1 . 2MAM3Bpredicted protein1 . 071 . 54Cre06 . g249800 . t1 . 1unknown conserved protein1 . 072 . 08Cre01 . g038250 . t1 . 1SDC1serine decarboxylase1 . 061 . 92Cre44 . g788200 . t1 . 11 . 062 . 13Cre08 . g359200 . t1 . 21 . 032 . 69Cre05 . g245950 . t1 . 1DRP1Dynamin-related GTPase1 . 032 . 15Cre05 . g234100 . t1 . 1CYP745A1cytochrome P451 . 012 . 61Cre07 . g328700 . t1 . 21 . 011 . 56Cre10 . g440250 . t1 . 21 . 012 . 14Cre17 . g725200 . t1 . 1MDR-like ABC transporter1 . 013 . 30Cre82 . g796100 . t1 . 11 . 012 . 49*Genes defined as SAK1-dependent in Table 4 . Cloning of the SAK1 gene revealed that it encodes a large previously uncharacterized phosphoprotein located primarily in the cytosol ( Figure 6A , D ) , suggesting that it functions as an intermediate in the retrograde signaling pathway from the chloroplast to the nucleus that leads to 1O2 acclimation . Previous genetic screens in Arabidopsis have identified proteins in the chloroplast , such as EX1 and EX2 ( Wagner et al . , 2004; Lee et al . , 2007 ) , and in the nucleus , such as PLEIOTROPIC RESPONSE LOCUS 1 ( Baruah et al . , 2009b ) and topoisomerase VI ( Simková et al . , 2012 ) , that are involved in 1O2 signaling . By screening for mutants that are unable to induce a 1O2-responsive reporter gene ( HPS70A ) in Chlamydomonas , a small zinc finger protein ( Cre09 . g416500 . t1 . 2 ) called MBS was recently identified as having a role in ROS signaling in both Chlamydomonas and Arabidopsis ( Shao et al . , 2013 ) . Like SAK1 , MBS in Chlamydomonas is located in the cytosol , raising a question about the relationship of these two proteins in 1O2 signaling . As expected , we found HSP70A among the genes induced by RB treatment of Chlamydomonas ( Table 3 ) however in sak1 it was not significantly induced above the twofold threshold , suggesting that SAK1 might function in the same signaling pathway as MBS . The MBS gene itself is not induced by 1O2 ( Shao et al . , 2013 ) , and we will investigate the genetic and biochemical relationship of SAK1 and MBS in future research . SAK1 contains a novel domain of ∼150 amino acid residues that is found in several chlorophyte species ( Table 8 ) . The sequence of this domain is not highly conserved ( Figure 5—figure supplement 1 ) , and is even less conserved among land plant proteins , although it is detectable by PSI-BLAST , indicating that it has diverged in sequence in plants and algae . We identified 37 proteins that have the SAK1 domain , 13 of which also contained a bZIP transcription factor domain , consistent with a function in regulating gene expression . Under our standard laboratory growth conditions , SAK1 appears to have a relatively low level of phosphorylation , but it becomes hyperphosphorylated during 1O2 acclimation ( Figure 6D ) . Phosphorylation prediction software NetPhos 2 . 0 ( http://www . cbs . dtu . dk/services/NetPhos/ ) predicted 24 serine , 9 threonine , and one tyrosine residue as possible sites throughout the protein ( Figure 5—figure supplement 3 ) . One of these serine residues is within the conserved SAK1 domain and is relatively conserved for polar amino acids . At this position , 18 SAK1 family members had threonine , and three had serine residues including SAK1 ( Figure 5—figure supplement 1 ) . We speculate that phosphorylation of SAK1 in the cytosol is a necessary intermediate step in 1O2 acclimation . Through further analysis of the transcriptome data , isolation of proteins that physically interact with SAK1 , and characterization of additional , non-allelic sak mutants , we hope to identify the kinase that is responsible for the direct modification of SAK1 as well as other upstream and downstream components of this retrograde signaling pathway in Chlamydomonas .
The sak1 mutant was generated by insertional mutagenesis as described previously ( Dent et al . , 2005 ) from WT strain 4A+ . Cells were grown at 22°C photoheterotrophically in Tris-acetate phosphate media ( TAP ) unless otherwise stated ( Harris , 2009 ) . For systematic screening of large number of strains for increased or decreased resistance to RB , individual strains were inoculated into 180-200 μl TAP medium in 96-well plates , grown for a at least 3 days to saturation under light intensity of 60–80 μmol photons m−2 s−1 , spotted onto TAP plates with 2 . 7 , 3 . 0 , or 3 . 3 μM RB , and scored for their growth compared to WT and sak1 . For more quantitative evaluation of RB sensitivity , the cells were grown to saturation in 1 ml of TAP medium because we have observed rapidly growing cells to have more variable sensitivity to RB ( data not shown ) . The cells were counted and adjusted to equal cell density then dispensed into aliquots in duplicate 96-well plates . One of the duplicates was pretreated in dark while the other was placed in light for 40 min with 1 μM RB . For challenge treatments , 4 . 5 , 5 . 1 , 5 . 7 , 6 . 3 , 6 . 9 , and 7 . 5 μM RB was added to both plates , which were placed under light for 1 hr and then spotted onto TAP agar media with no RB . All treatments were applied under light intensity of 60–80 µmol photons m−2 s−1 , which is the light intensity described as low light unless stated otherwise . Cells were grown under 100 μmol photons m−2 s−1 , adjusted to 2 × 106 cells ml−1 , and treated with RB at a final concentration of 0 . 5 μM for 30 min ( pretreatment ) in light ( + ) or dark ( − ) . After the pretreatment all the cultures were exposed to an additional 3 . 75 μM RB ( challenge ) in low light and collected for measurement of Fv/Fm at 30 , 60 , and 90 min . The cells were dark-acclimated for at least 30 min before applying a saturating light pulse of 2000 μmol photons m−2 s−1 and measuring the chlorophyll fluorescence yield using an FMS2 fluorometer ( Hansatech Instruments , Norfolk , UK ) . Cultures were grown for at least two light–dark cycles ( 12 hr light-12 hr dark ) , and then cell density was adjusted to 2–2 . 5 × 106 cells ml−1 and split into two flasks ( one control and the other for RB treatment ) at least an hour prior to adding RB to a final concentration of 1 μM . An equal volume of H2O was added to the control . RB was added ∼6 hr after the start of the light cycle under light intensity of ∼100 µmol photons m−2 s−1 and the treatment lasted for an hour before harvest . The cells were cooled and harvested by centrifugation at 1200×g for 3 min at 4°C , frozen with liquid nitrogen and stored at −80°C until extraction of RNA . For low light to high light transfer experiment , cultures were grown in continuous light in minimal ( HS ) medium for 3 days to cell density of 3 × 106 cells ml−1 at 45 µmol photons m−2 s−1 . The light intensity was increased to 500 µmol photons m−2 s−1 for 1 hr before harvest . RNA was extracted with TRIzol ( Life Technologies , Carlsbad , CA ) following manufacturer's instructions and treated with DNaseI ( Promega , Madison , WI ) , then cleaned up using Qiagen RNeasy columns ( Qiagen , Germantown , MD ) . cDNA was synthesized using Omniscript ( Qiagen , Germantown , MD ) starting with 2–3 μg DNA-free RNA per 20 μl reaction . qPCR was performed using a 7300 FAST qPCR machine ( Life Technologies , Carlsbad , CA ) . The primers were designed with a Tm of 60°C using Primer3 or PrimerExpress ( Life Technologies , Carlsbad , CA ) ( Table 10 ) . All primer pairs described in this study were confirmed as having 90–105% amplification efficiency and linear amplification within their dynamic range in experimental samples using serial dilutions of cDNA prior to the experiments . Relative transcript levels were calculated by ΔΔCt method ( Livak and Schmittgen , 2001 ) using CβLP as internal reference . 10 . 7554/eLife . 02286 . 022Table 10 . Primers used for qRT-PCR analysesDOI: http://dx . doi . org/10 . 7554/eLife . 02286 . 022v4 IDv5 IDGene nameForwardReverseCre01 . g007300 . t1 . 1Cre01 . g007300 . t1 . 2AGCATGTGCGTGTGGAGTAGCCTTACCATAGGCCTGACCAau5 . g10700_t1aCre03 . g177600 . t1 . 3CTGGACATGTCGGCTATGAAGCTCATGTCGTACTCCAGCAau5 . g13389_t1*Cre06 . g299700 . t1SOUL1†TGCGTATGGGTGTCCACTAATGGGGATCTTCTTCATGTCCCre06 . g263550 . t1 . 1Cre06 . g263550 . t1 . 2LCI7TTTGGTTGCGTTGCATGTATTCAACGCGGTGTCAAACTTACre06 . g281250 . t1 . 1Cre06 . g281250 . t1 . 2CFA1CCTACAACGACAACGACGTGGGAAGTTCCAGGATGACCAGCre06 . g298750 . t1 . 1Cre06 . g298750 . t1 . 2AOT4CCGTGTGCACAGATTCAAAGCACACAGCGCCTCCTACATACre08 . g358200 . t1 . 2Cre08 . g358200 . t2 . 1TGTGGCATCAAGGTGTGTTGTAACCCCACACCCCTCTCTTTCre09 . g398700 . t1 . 1Cre09 . g398700 . t1 . 2CFA2CGACCTGCTGCTCTACTTCCGTGTAGGCGGTGGTCAAGATCre10 . g458450 . t1 . 2Cre10 . g458450 . t1 . 3GPX5AACCAATCGCCTAACACCTGCACTTGCTAGCCACGTTCACCre12 . g503950 . t1 . 1Cre12 . g503950 . t1 . 2GGAGGGAGTACCACGAGACAGATTGCTGTAAGGCCGGATACre13 . g564900 . t1 . 1Cre13 . g564900 . t1 . 2MRP3TCATGACGTACATCTCGATTCTCAAGGGAATGTAGTAGCGCTGAATGau5 . g4402_t1*Cre13 . g566800 . t1 . 2TGCTTGGAAGACCCACTTTTGAGCTGGAGTTGCAGTTGTGCre13 . g566850 . t1 . 1Cre13 . g566850 . t1 . 2SOUL2CCCTCCCCTCCTTCAGACTACGTACCTGAGGCGCATATTTCre14 . g613950 . t1 . 1Cre14 . g613950 . t2 . 1CGCCCAACCCCATGATCCCGCAACGTACCGTGATGCre16 . g683400 . t1 . 1Cre16 . g683400 . t1 . 2CCTGAACAAACACACGATGGGAACGCCGTCAAATCATCTTCre16 . g688550 . t1 . 1Cre16 . g688550 . t1 . 2GST1AGTGCGGAGGAAGTCGTAAAGTAAAAGACGTGCGTGCAAAg6364 . t1CβLP ( RCK1 ) GAGTCCAACTACGGCTACGCGGTGTTCAGGTCCCACAGACCre14 . g623650 . t1 . 1Cre14 . g623650 . t1GACAACGCGGCCTACAAGACCGAGCTGGCGGTGTTAAau5 . g2281_t1*g16723 . t1MKS1GCTTGAGCGCGAGACGAACGCTGAAAGCATTGCAGAAGCre08 . g380300 . t1 . 2Cre08 . g380300 . t1 . 2ACCACCAGCAGTACCTGTCCCGCTCCAATAAAGCCTTCAGau5 . g7871_t1‡ ( Cre17 . g741300 . t1 . 2 ) ‡SAK1 ( 5'UTR ) CAAGTGCTCATGAGAGGCCTTATACGTCATCCAGTTCCACATCCau5 . g7871_t1‡ ( Cre17 . g741300 . t1 . 2 ) ‡SAK1 ( 3'UTR ) TCAAGCGTGTGGGTAAGAGCTAACGCTATCTCCGTCCTAATCCACre08 . g365900 . t1 . 1Cre08 . g365900 . t1 . 2LHCSR1CACACAATTCTGCCAACAGCATCTGCTTCACGGTTTGGTCCre04 . g220850 . t1 . 1Cre04 . g220850 . t1 . 2TAATGGTATGGATGCGGTCAACTGCCAGTTATGGGTCCTGCre09 . g395750 . t1 . 2Cre09 . g395750 . t1 . 3ACCGTCCGTGAACCTTACTGCGCAAACACGTCTCAAAGAA*Was originally mapped and identified as augustus version 5 models within Chlamydomonas genome v4 . †SOUL1 was given the name in v4 but not v5 . ‡Primers were designed against experimentally obtained cDNA ( Genbank accession KF985242 ) and differs from v5 . Closest gene model is au5 . g7871_t1 . RNA was extracted ( Schmollinger et al . , 2014 ) and the quality was determined using a 2100 Bioanalyzer ( Agilent Technologies , Santa Clara , CA ) . The triplicate RNA was pooled and 10 μg total RNA was used to prepare RNA-seq library according to the manufacturer's protocol ( Illumina , San Diego , CA ) . The quality of the library was assessed using a 2100 Bioanalyzer before sequencing with Genome Analyzer ( Illumina , San Diego , CA ) . Each sample was run in replicates on two lanes . RNA-Seq data was analyzed as before ( Duanmu et al . , 2013 ) . On average , 75% of the sequences could be assigned unambiguously to Augustus v10 . 2 gene models to generate the matrix of counts per gene . This matrix was used for differential expression analysis using DESeq ( Anders and Huber , 2010 ) using per-condition dispersion estimates and variance stabilization to compute moderate fold changes . Genes were classified as differentially expressed based on a ( moderate ) twofold regulation and a false discovery rate ( FDR ) <1% . Near full-length cDNA was isolated by RT-PCR ( described in above section; Gene expression analysis by qRT-PCR ) and rapid amplification of cDNA ends ( RACE ) using GeneRACER ( Life Technologies , Carlsbad , CA ) as previously described ( Molnar et al . , 2009 ) . Despite multiple attempts the 5′ end of the transcript could not be amplified by 5′-RACE . Because the experimentally obtained CDS differed from the most current v5 , it has been deposited to genbank ( accession KF985242 ) . Though some differences exist at the nucleotide level , the protein sequence of the resulting CDS was identical to that of au5 . g7871_t1 . Genomic DNA containing SAK1 was amplified using primers 5′-CAGGACCGGGCACTGAGTGAAGGTTA-3′ ( + ) and 5′-ATGATGCACTGTGGGACACGCTGAGT-3′ ( − ) using PrimeStar HS with GC buffer ( Takara/Clontech , Palo Alto , CA ) and cloned into pGEM-Teasy after adding an adenine . The resulting plasmid was co-transformed with pBC1 and selected with 1 μM paromomycin . Transformation of sak1 was performed as described previously ( Kindle et al . , 1989 ) . To raise antibodies against SAK1 , an epitope at the N-terminus of the translated coding sequence of SAK1 ( DTLLTPLREDATAESGGDA ) was designed , synthesized and injected into rabbits , and the resulting crude serum was affinity purified ( Open Biosystems/Thermo Scientific , Waltham , MA ) . For immunoblot detection of SAK1 , proteins were separated with NuPAGE 3–8% Tris Acetate gels ( Life Technologies , Carlsbad , CA ) and transferred to nitrocellulose membranes . All other blots were prepared from running the protein on 10–20% Tris-glycine gels and transferring to a PVDF membrane . The membranes were blocked for several hours in 5% milk in TBS-T , incubated with the primary antibody overnight , then with secondary antibody for several hours in 1% milk TBS-T before washing and developing with a chemiluminescence detection kit . Commercial antibodies were anti-histone H3 ( ab1791; Abcam , Cambridge , UK ) and anti-KDEL ( ab12223; Abcam , Cambridge , UK ) . Other antibodies were generous gifts from Jean-David Rochaix ( anti-PSAD ) , Olaf Kruse ( anti-NAB1 ) , and Patrice Hamel ( anti-cytochrome c ) . Nuclear fractions were prepared from 450 ml of synchronized cultures with ∼2 × 106 cells ml−1 that had been incubated with or without 2 μM RB under light for 40 min . The cells were collected and treated with autolysin for 40 min and examined for the removal of cell walls by addition of 1 volume of 0 . 1% Triton-X . Nuclear extract was prepared as described previously ( Winck et al . , 2011 ) using CelLytic PN kit ( Sigma-Aldrich , St . Louis , MO ) . Because there were bands detected in the nuclear extract close to the size of SAK1 , nuclear extract was prepared from WT ( 4A+ ) and sak1 rather than a cell wall-deficient strain ( cw15 ) . Chloroplasts were isolated from cell wall-less strain cw15 as described previously ( Klein et al . , 1983 ) . Mitochondria were isolated as described ( Eriksson et al . , 1995 ) . After unbroken cells , chloroplasts , and mitochondria were collected , the ER fraction was collected by centrifugation at 100 , 000×g for 90 min at 4°C . The remaining supernatant was enriched for cytosol . Protein was extracted and prepared for SDS-PAGE as described ( Calderon et al . , 2013 ) with minor modifications . Protein was quantified by using BCA1 kit ( Sigma-Aldrich , St . Louis , MO ) after extraction with the methanol-chloroform method ( Wessel and Flügge , 1984 ) . | Plants , algae and some bacteria use photosynthesis to extract energy from sunlight and to convert carbon dioxide into the sugars needed for growth . One by-product of photosynthesis is a highly toxic molecule called singlet oxygen . Typically , organisms deal with stressful events such as the presence of toxic molecules by producing new proteins . However , protein production is generally initiated in the nucleus of the cell , and photosynthesis is carried out in structures called chloroplasts . Cells must therefore be able to alert the nucleus to the presence of toxic levels of singlet oxygen in the chloroplasts . Like some plants that can withstand a gradual decrease in temperature , but not a sudden cold snap , the alga Chlamydomonas reinhardtii is capable of resisting high doses of singlet oxygen if it has previously been exposed to low doses of the molecule . Wakao et al . exploited this ability to hunt for algae that are unable to acclimate to singlet oxygen , and found that these cells are unable to produce a protein called SAK1 . Wakao et al . reveal that many factors involved in the algae's cellular response to singlet oxygen depend on the presence of SAK1 . In addition , the response of the algae cells to singlet oxygen differs to the one seen in the model plant Arabidopsis thaliana , suggesting that the two organisms have found different ways to deal with the same problem . The location of a protein in a cell can give clues to its function . SAK1 is present in the fluid surrounding cellular compartments—the cytosol—which is consistent with it acting as a signaling molecule between the chloroplast and the nucleus . Wakao et al . present further evidence for this hypothesis by demonstrating that the number of phosphate groups attached on SAK1 changes when exposed to singlet oxygen—a feature often seen in signaling proteins . In addition , part of SAK1 resembles proteins that can bind to DNA , which indicates that SAK1 may be directly involved in initiating protein production . The discovery of SAK1 represents a starting point for understanding how the site of photosynthesis , the chloroplast , communicates with the nucleus . It also has implications for developing plants and algae that have a higher tolerance to environmental stress conditions for agriculture and biofuel production . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"plant",
"biology"
] | 2014 | Phosphoprotein SAK1 is a regulator of acclimation to singlet oxygen in Chlamydomonas reinhardtii |
Eukaryotes and many archaea package their DNA with histones . While the four eukaryotic histones wrap ~147 DNA base pairs into nucleosomes , archaeal histones form ‘nucleosome-like’ complexes that continuously wind between 60 and 500 base pairs of DNA ( ‘archaeasomes’ ) , suggested by crystal contacts and analysis of cellular chromatin . Solution structures of large archaeasomes ( >90 DNA base pairs ) have never been directly observed . Here , we utilize molecular dynamics simulations , analytical ultracentrifugation , and cryoEM to structurally characterize the solution state of archaeasomes on longer DNA . Simulations reveal dynamics of increased accessibility without disruption of DNA-binding or tetramerization interfaces . Mg2+ concentration influences compaction , and cryoEM densities illustrate that DNA is wrapped in consecutive substates arranged 90o out-of-plane with one another . Without ATP-dependent remodelers , archaea may leverage these inherent dynamics to balance chromatin packing and accessibility .
Eukaryotic genomes are orders of magnitude larger and more complex than those of archaea or bacteria . They manage their massive genomes through a hierarchical packaging scheme that utilizes histones to form nucleosomes , a complex that contains two H2A-H2B histone heterodimers flanking a central ( H3-H4 ) 2 heterotetramer and stably wraps ~147 base pairs of DNA ( Figure 1A; Luger et al . , 1997 ) . While all eukaryotes invariably contain histones , genes encoding putative ‘minimalist’ histone proteins have been identified in most archaea ( Henneman et al . , 2018 ) , but so far not in bacteria . While still a subject of debate , the presence of histones in the two domains of life has supported theories that eukaryotes evolved directly from an archaeon ( Heinicke et al . , 2004; Malik and Henikoff , 2003; Brunk and Martin , 2019; Watson , 2019; Eme et al . , 2017; Spang et al . , 2018; Sandman and Reeve , 2006 ) . Archaeal histones share many features with their eukaryotic equivalents , such as the three-helix histone fold motif ( α1-L1-α2-L2-α3 ) ( Decanniere et al . , 2000 ) , as well as obligate dimerization and transient tetramer formation ( Marc et al . , 2002 ) . In both archaea and eukaryotes , histones have a preference for occupying particular sites in the genome ( Ammar et al . , 2012; Nalabothula et al . , 2013 ) . Eukaryotic histones utilize highly disordered cationic N-terminal tails to regulate gene transcription and chromosome compaction via post-translational modifications ( PTMs ) ( Luger and Richmond , 1998; Musselman et al . , 2012; Bowman and Poirier , 2015 ) , whereas most archaeal histones do not contain tail sequences , with the exception of some sequences found within the Asgard clade ( Henneman et al . , 2018 ) . Additionally , histones in eukaryotes have evolved histone fold ‘extensions’ , unique secondary structure elements that define the outer surface of the nucleosome as well as stabilize the histone core and the final turn of nucleosomal DNA . No such diversification has yet been identified in any of the archaeal histones . Recently , we determined the crystal structure of Methanothermus fervidus histone HMfB bound to ~90 base pairs of DNA ( Mattiroli et al . , 2017; Bhattacharyya et al . , 2018 ) . HMfB is a typical representative of most archaeal histones , as it contains no tails and no histone fold extensions , and it exists as either a homodimer or heterodimer with the closely related HMfA histone ( Sandman et al . , 1994 ) . Nevertheless , striking similarities to the eukaryotic nucleosome were observed , including a superhelical DNA path around a histone core that is nearly identical to that in nucleosomes , promoted by conserved DNA contacts with histones , and by dimer-dimer interactions through ‘four-helix bundles’ of α-helices . However , unlike nucleosomes , which are defined octameric particles that arrange as 10 nm ‘beads on a string’ on long DNA fragments , crystal contacts in the archaeal system promote an extended superhelical winding of DNA around a histone core that consists of more than the canonical four histone dimers . This arrangement , which is consistent with in vivo footprinting results ( Mattiroli et al . , 2017 ) , places the L1 loops of every nth and ( n+3 ) th dimer in close proximity to one another ( Figure 1C ) . The importance of the L1-L1 interaction in stabilizing this arrangement in vitro and in vivo was established by substituting the highly conserved G17 residue with bulkier amino acids ( e . g . G17D ) , resulting in transcription regulation defects . Here , we show that archaeal histones , when assembled on long segments of DNA ( >200 base pairs ) , experience disparate solution dynamics compared to eukaryotic nucleosomes , and we present cryo-EM structures of two classes of particles arranged on a 207 base pair DNA fragment with deflections of the wrapping pathway . Despite many nucleosome-like properties , we refer to these constructs as ‘archaeasomes’ , rather than ‘archaeal nucleosomes’ , because of their inherent differences in higher order structures and solution behaviors , including the ability to form histone cores with more than four dimers . We classify archaeasomes by the length of bound DNA ( i . e . shorthand of ‘Arc120’ for ‘archaeasome with 120 base pairs of DNA’ ) . We use molecular dynamics ( MD ) simulations to show that archaeasomes can expand , like the stretching of a ‘slinky’ , to increase overall accessibility without sacrificing histone-histone or histone-DNA interactions . Simulations also reinforce the importance of L1-L1 histone stacking interactions in regulating this dynamic equilibrium . Sedimentation velocity analytical ultracentrifugation ( SV-AUC ) and cryoEM highlight two distinct accessibility dynamics at play in archaeasome systems , and cryoEM analysis identifies archaeasomal states that continually wrap DNA but deflect the wrapping pathway 90o out-of-plane . The intrinsic dynamic behavior of archaeasome slinkies stands in contrast to eukaryotic nucleosomes , which are compact and highly stable and require modifications and elaborate machinery to efficiently regulate chromatin accessibility , whereas archaea may utilize this inherent stochastic behavior in order to access their DNA .
We applied molecular dynamics ( MD ) simulations to model the stability and solution dynamics of archaeasomes assembled on DNA of increasing lengths . Three system sizes were studied: an archaeasome with 90 base pairs of DNA ( ‘Arc90’ , Video 1 ) , one with 120 base pairs of DNA ( ‘Arc120’ , Video 2 ) , and one with 180 base pairs of DNA ( ‘Arc180’ , Video 3 ) . In T . kodakarensis , the G17D ( G16D in HMfB ) mutation in the L1 loop was previously shown to abolish the characteristic footprint of histone-based chromatin in the cell ( Mattiroli et al . , 2017 ) , and we simulated this system at the archaeasome level ( ‘Arc120-G17D’ ) to elucidate the atomistic mechanisms for this observed behavior . In each system , the DNA footprint of ~30 base pairs per dimer was maintained on the hundreds of nanoseconds timescale . Nevertheless , the Arc90 system experienced a high degree of global dynamics . RMSD calculations of backbone atom positions showed a maximum deviation of ~6 Å from the initial Arc90 coordinates , with trough-to-peak heights of ~4 Å throughout the time course ( Figure 2A , B ) . RMSF calculations identify regions of highest flexibility where the DNA binds to the terminal dimers ( Figure 2—figure supplement 1 ) , and visual inspection of Arc90 trajectories revealed that the system samples a ‘clamshell’ motion , where terminal dimers and associated DNA fluctuate to create ‘closed’ and ‘open’ accessibility states ( Videos 4 and 5 ) . The equilibrium between closed and open forms was quantitated by measuring the center-of-mass distance between each DNA end and the neighboring superhelical turn of DNA , where separations of ~16 Å and ~30 Å are indicative of closed and open states , respectively ( Figure 2C , Figure 2—figure supplement 2 ) . Over the clamshell motion , the four-helix bundle interfaces that bind consecutive dimers to one another was maintained , and only small rearrangements of the local contact network were observed . Together , these data show that archaeasomes retain protein-protein contacts while still allowing for significant molecular motion . Simulations of Arc120 and Arc180 systems showed that L1-L1 interactions drastically reduced the population of the accessible state . Both the maximum observed RMSD and variance in the timeseries were dampened in these systems ( Figure 2A , B ) , and RMSF calculations showed that DNA bound to the terminal dimers were less dynamic than in Arc90 trajectories ( Figure 2—figure supplement 1 ) . Similarly , the distance between DNA entry and exit sites and the neighboring DNA gyre in Arc120 and Arc180 were unimodal around closed-state values ( Figure 2C ) , with average separations of 17 . 4 ± 0 . 2 Å and 16 . 1 ± 0 . 2 Å , respectively . While the Arc120-G17D system has the proper number of dimers to form L1-L1 stacking interactions , the G17D mutation was designed to disfavor interactions , and this system experienced increased dynamics relative to the wild-type Arc120 trajectories . RMSD measurements showed a unimodal distribution , as was seen in the Arc120 and Arc180 simulations , but the maximum observed RMSD values were more similar to the Arc90 trajectories ( Figure 2 ) . Similarly , local fluctuations in Arc120-G17D DNA positions were increased relative to the Arc120 trajectories ( Figure 2—figure supplement 1 ) , and separation of the superhelical gyres was significantly larger than the Arc120 values ( 21 . 0 ± 0 . 4 Å , p-value<0 . 0001 ) . This also increased the solvent-accessible surface area by over 1 , 000 Å2 ( 57 , 316 ± 34 Å2 vs 58 , 422 ± 73 Å2 , p-value<0 . 001; Figure 3 ) . Similar to the Arc90 system , the four-helix bundle interactions of consecutive dimers were maintained in these systems , despite the increase in DNA separation and surface accessibility . Histone-DNA interactions were largely unchanged by these dynamics . Molecular Mechanics-Generalized Born Surface Area ( MM-GBSA ) calculations were used to estimate histone-DNA binding energies ( Table 1 ) . MM-GBSA values should not be interpreted literally , but can be used to identify qualitative and robust changes in DNA binding interactions in nucleosomes as a result of histone modifications ( Bowerman and Wereszczynski , 2016; Bowerman et al . , 2019; Morrison et al . , 2018 ) . In agreement with previous computational studies ( Rojec et al . , 2019 ) , raw MM-GBSA values suggest that DNA-binding strength increases as additional dimers are added to an archaeasome stack ( ΔGArc90 < ΔGArc120 < ΔGArc180 ) . However , this may be an artifact of molecular mechanics forcefields , which are additive and thereby affected by the total number of interactions in a system . When MM-GBSA values were normalized to the number of dimers in each system , there was no net change in DNA binding ability as a function of archaeosome size . Additionally , the G17D point mutation slightly reduced the dimer-DNA interaction strength , but this difference was not statistically significant when compared to the wild-type Arc120 system ( p-value of 0 . 157 ) . These data , in conjunction with maintained four-helix bundle interactions , show that archaeasomes can sample open and accessible conformations without histone dissociation . Thus , the loss of the 30 base pair ladder pattern in MNase digestion of chromatin isolated from cells carrying the G17D mutation as their only source of histones may be due to increased archaeasome dynamics ( Figure 3 ) , rather than frequent histone dissociation ( Mattiroli et al . , 2017 ) . Our simulations provide in silico confirmation for the notion that archaeasomes indeed wrap several DNA turns , with more than four histone dimers , as a bona fide solution-state . The Arc90 and Arc120-G17D simulations suggest that archaeasomes may sample extended conformations without even partial dissociation of histones from either DNA or their histone partners . To structurally characterize these assemblies in solution , we reconstituted archaeasomes on 207 base pairs of Widom 601 DNA ( Arc207 ) with saturating amounts of recombinant HTkA histones from Thermococcus kodakarensis , which are very similar to HMfB histones ( overall 59% identity , 81% similarity , with higher conservation in sequences involved in histone-histone and histone-DNA interactions ) ( Mattiroli et al . , 2017 ) . The solution behavior of the Arc207 construct was then analyzed by single particle cryoEM and sedimentation velocity analytical ultracentrifugation ( SV-AUC ) . As controls , we also collected SV-AUC traces from eukaryotic ( Xenopus laevis ) nucleosomes reconstituted on 147 base pairs of Widom 601 DNA ( Lowary and Widom , 1998 ) . Gel shift assays showed that full complex saturation occurs when DNA and histones are mixed at the previously reported stoichiometric limit ( ~30 bp per dimer; Figure 4—figure supplement 1; Mattiroli et al . , 2017 ) . SV-AUC traces show that the Arc207 complex sediments homogeneously at ~10 . 6 s ( Figure 4 , circles ) , and the Nuc147 control sediments more rapidly than the Arc207 system at ~11 . 2 s ( Figure 4 , squares ) . As sedimentation is dependent on both molar mass ( higher mass yields faster sedimentation ) and shape anisotropy ( higher anisotropy yields increased drag and slower sedimentation ) , the relatively lower sedimentation rate of the Arc207 sample has one of two interpretations: either its mass is less than that of a nucleosome , or the particle is more extended in solution , exhibiting higher shape anisotropy and drag . The mass and anisotropy of each molecule were modeled from SV-AUC traces using the Monte-Carlo-coupled Genetic Algorithm ( GA-MC ) module of Ultrascan III ( Table 2 ) . The molecular weight for both particles was slightly overestimated , consistent with a systematic error in specific volume estimation for protein-DNA complexes , which can propagate to modest errors in molecular weight estimation ( Demeler et al . , 2014 ) . Nevertheless , these data show that the Arc207 complex is fully saturated with histones , that the total mass is consistent with seven histone dimers assembled on 207 bp of DNA , and that its mass is greater than that of Nuc147 . This suggests that the Arc207 system is more anisotropic than Nuc147 in order to satisfy the decrease in sedimentation rate , which is confirmed by the frictional ratios ( f/fo ) estimated from GA-MC calculations ( 1 . 94 and 1 . 55 for Arc207 and Nuc147 , respectively ) . For reference , the f/fo value of free 207 DNA is 3 . 14 so the Arc207 complex is more compact than free DNA but not as compact as the eukaryotic nucleosome . We next utilized single particle cryoEM to capture the three-dimensional structure of the Arc207 complex . Our simulations of the similarly sized Arc180 system predicted that the complex would be compact , but SV-AUC modeling conversely suggests that Arc207 forms an extended conformation . CryoEM data shows that the Arc207 system is indeed open , as it exhibits nucleosome-like particle dimensions with increased spacing between neighboring DNA gyres ( Figure 4B ) , in agreement with SV-AUC data . This is also apparent in the two-dimensional classifications ( Figure 4C ) . Because of the apparent structural heterogeneity of this assembly , we were unable to extract three-dimensional conformations from the dataset , even at low resolution . Eukaryotic chromatin fibers can be compacted in vitro through the addition of divalent cations such as Mg2+ ( Schwarz et al . , 1996 ) . To test whether this also applies to archaeal chromatin , we analyzed Arc207 samples in the presence of 0 , 1 , 2 , 5 , 7 , 8 , and 10 mM MgCl2 by SV-AUC , and we observed changes in both sedimentation coefficient and frictional ratio ( Figure 5 ) . Arc207 samples exhibit an increase in sedimentation rate with increased Mg2+ concentration from 0 to 5 mM , with no additional increase between 5 and 10 mM MgCl2 ( Figure 5A , top ) . In comparison , the nucleosome system shows little to no change in sedimentation from 0 to 2 mM MgCl2 ( Figure 5A , bottom ) , and start to self-associate ( aggregate ) at concentrations above 2 mM Mg2+ . GA-MC analysis of the Arc207 traces shows the same correlation between f/fo and MgCl2 concentration that is observed for the sedimentation rate , with apparent compaction from 1 to 5 mM MgCl2 and no further reduction beyond 5 mM MgCl2 ( Figure 5B , top ) . A modest reduction in f/fo is observed for the Nuc147 sample , but the most notable differences between the archaeasome and nucleosome systems is the loss of nucleosome sample due to aggregation . For Nuc147 , OD260 values rapidly decline with increased Mg2+ , and no appreciable sample remains at 5 mM MgCl2 and above ( Figure 5B , bottom ) . In contrast , the Arc207 complex displays no significant losses , even at 10 mM MgCl2 . This shows that , unlike eukaryotic chromatin , divalent cations compact archaeasomes without promoting fiber-fiber association . Single particle cryoEM was utilized to determine the three-dimensional structure of archaeasomes in the presence of 5 mM MgCl2 . Particles appeared more defined than without MgCl2 , and individual particles with tightly wrapped DNA conformations are distinguishable from other states where density is visible as perpendicular extensions to the wrapping plane ( Figure 6A , blue and gold boxes ) . A total of 1 , 879 , 294 particles were identified according to a neural network trained on manual particle selections and then classified . The two-dimensional classification of this dataset identify two different populations of Arc207: Class I , which upon first inspection has the classical ‘nucleosome-like’ fold ( Figure 6B ) , and Class II , in which two perpendicular DNA wrappings are clearly observed ( Figure 6C ) . We separated these particles into independent 3D classes and refinement schemes , which yielded 80 , 609 particles in Class I and 5959 particles in Class II . Many more particles could have been included in each class , especially Class II , but we were conservative to ensure that no interactions with neighboring particles affected their configuration . Refinement of these densities yielded maps at 9 . 5 Å and 11 . 5 Å resolution for Class I and Class II , respectively ( Figure 6D , E ) . At 9 . 5 Å resolution , secondary structure and side chain assignment is not possible for Class II , which has the higher resolution and larger particle count of the two classes . However , key features of the archaeasome previously seen in the crystal structure are clearly discernible from the density , such as DNA wrapping around a core of five histone dimers and periodic , overlapping densities that we associate with α2 , the longest of the core α-helices in each histone ( Video 6 ) . Even though Arc207 ( 207 base pairs of DNA bound to seven histone dimers ) were deposited on the grid , refined density of state Class II describes only 150 base pairs of DNA and five histone dimers ( Video 7 ) . As such , we trimmed the final frame from one of our MD simulations of the Arc180 complex down to an Arc150 complex and docked it to the map using the ‘fit in map’ function of Chimera . We simulated density at 8 Å resolution for the MD-derived structure , which yields a correlation coefficient of 0 . 85 with the empirical map . This shows that our MD simulations describe with high fidelity a compact archaeasome composed of five histone dimers wrapping ~150 bp of DNA . The density for Class II has sufficient volume to be fit with the full 207 base pair DNA and seven histone dimers ( Video 8 ) . The larger portion of the volume is best described by four histone dimers binding ~120 base pairs of DNA in a nucleosome-like structure , and the smaller portion can be fit with three histone dimers binding ~90 base pairs . While the connecting density suggests only moderate bends in the helical axis , the wrapping pathways of the subunits result in the two faces being arranged at a near 90° angle . The continuity of the DNA between the two moieties was confirmed by modulating the electron density contours ( Video 9 ) . Structural models were generated by rigid-body docking separate Arc120 and Arc90 subunits into the associated density , and the connecting DNA was energy minimized and subjected to a short MD simulation to confirm that the connecting DNA segment is reconcilable with B-form DNA parameters ( Figure 6—figure supplement 1 ) . To confirm that the structures described by the EM densities also exist in solution , and to correlate these findings with our SV-AUC analyses of particle compaction , we utilized the SoMo plugin of UltraScan to calculate sedimentation properties from our derived models . For the fully resolved Arc207 model , with the Arc90 ‘lid’ bending ~90o from the Arc120 ‘core’ , we calculate a frictional ratio of 1 . 53 , in close agreement with the 1 . 55 value extracted from the experimental traces . For the particles that can be fit with five histone dimers and 150 bp DNA ( Class I ) , we surmised that the remaining 60 base pairs of DNA ( and two histone dimers ) extend from the observed density but assume variable orientations with respect to the main particle resulting in very weak density . Close inspection of the two-dimensional classes in this category indeed shows additional out-of-plane density that is consistent with the missing two dimers and ~60 base pairs of DNA , albeit at low contrast relative to background ( Figure 6B ) . Further three-dimensional classifications of the 9 . 5 Å density reduced the overall resolution but yielded indications of the missing volume , albeit at low fidelity ( Figure 6—figure supplement 2 and Video 10 ) . To further model this particle , we extended the Arc150 structure with an additional Arc60 subunit positioned 90o out-of-plane , similar to the model derived for the fully observed Arc207 density . This model predicted a similar value for the frictional coefficient of 1 . 50 . In contrast , models with continuous wrapping and no deflections in the DNA pathway ( as observed in the crystal structure , but not populated significantly in the cryo EM images ) yield a frictional coefficient of 1 . 36 . These data together show that archaeasomes on an extended DNA fragment can be viewed as a distribution of archaeasome subunits that can open to a ~90o angle .
Using in silico and in vitro approaches , we investigated the structure and dynamics of the ‘slinky-like’ architecture of histone-based archaeal chromatin . SV-AUC and cryoEM both show that the archaeasome , containing more than the characteristic four histone fold dimers observed in nucleosomes , is inherently dynamic and exists in a variety of open states while maintaining DNA-histone interactions . Archaeasomes can be stabilized with divalent cations . In apparent absence of archaeal ATP-dependent chromatin remodeling factors ( large machines that regulate chromatin access in eukaryotes ) , this architecture provides an alternative mechanism for compacting chromatin and maintaining genome accessibility . Our cryoEM structures show that archaeosomes are metastable and extended in the absence of divalent cations . Even in the presence of 5 mM Mg2+ , archaeasomes exhibit at least two configurations . Indeed , our structures contain a maximum of four to five histone dimers arranged in a continuous helical ramp . Beyond that , the helical histone ramp is disrupted at the four-helix bundle interface by an outward rotation of the remaining two to three histone dimers and attached DNA . The existence of these structures in solution is supported by analysis of the hydrodynamic parameters of archaeosomes obtained from SV-AUC . Extending this arrangement to even longer DNA , one could picture particles that wrap anywhere from ~90 to~150 base pairs ( using three to five histone fold dimers ) arranged at ~90 degree angles with respect to each other , connected by minimal linker DNA . This arrangement may leave the major groove of the connecting DNA susceptible to nuclear factors , such as micrococcal nuclease or transcription factors , and explains the ~30 bp MNase digestion ladder observed in native archaeal chromatin . Consistent with this interpretation , molecular dynamics simulations of Arc90 assemblies ( with three histone fold dimers ) displayed dynamic breathing at terminal sites of the complex , while histone-histone and histone-DNA contacts are maintained . Models constructed from repeats of this conformation agree with particle conformations extracted from SV-AUC hydrodynamic parameters , as well as direct images and two-dimensional class averages from cryoEM data gathered in the absence of Mg2+ . There , multiple turns of DNA , fully saturated with histones , were observed but with considerable flexibility and distance between neighboring gyres . In contrast , Arc120 and Arc180 simulations ( containing four and five histone dimers , respectively ) did not exhibit this degree of dynamic behavior . However , these simulations may have been artificially biased for the closed configuration , because the starting structures had intact L1-L1 interactions . Separation of these contacts may require more than several hundred nanoseconds of computational sampling , and simulations of the Arc120-G17D system indeed show appreciable separation between DNA turns by weakening this interface while maintaining four-helix bundle interactions . In eukaryotes , chromatin structure can be further modulated by the incorporation of histone variants ( Bönisch and Hake , 2012 ) . While in vivo studies have identified disparate roles of variant histones in archaea ( Čuboňováa et al . , 2012 ) , very little is known about the structural modifications that they invoke . In a recent study by Stevens et al . , 2020 , molecular dynamics trajectories showed that substituting major type histones in Methanosphaera stadtmanae with a ‘capstone’ paralog , a histone variant with sequence modifications at the four-helix bundle , disrupts the tetramer interface and separates histone dimers on a continuous DNA fragment , thereby limiting archaeasome size . In the context of the capstone model , our EM-derived structures suggest that destabilizing the four-helix bundle interface may yield a subtle increase in DNA accessibility through increased exposure of the major groove , and without the need to fully displace the archaeasome subunit . As we have shown , this dynamic behavior is already sampled by major-type archaeal histones deposited on a continuous DNA fragment , and substitution of terminal histones with capstone-containing dimers will further bias the assembly toward the open configuration . In this way , lengths of DNA that would typically contain a single archaeasome unit of considerable length would instead be segmented in to several sub-archaeasomes configured in repeat perpendicular arrangements that modestly expose bridging DNA segments . Additionally , high rigidity in certain DNA sequences may inherently poise these segments as linker sequences , and the frequency of these less flexible regions would similarly influence the number of 90o-oriented subunits present along the genome . Our SV-AUC traces show that Arc207 assemblies exhibit no signs of self-association upon the addition of up to 10 mM MgCl2 , unlike eukaryotic nucleosomes and nucleosome arrays , where this favors extensive inter-nucleosome and inter-array interactions and aggregation ( Schwarz et al . , 1996 ) . Instead , these conditions bias individual archaeasomes toward more compact states without promoting inter-archaeasome interactions . Interestingly , estimates of Mg2+ concentration within T . kodakarensis cells have been quoted at ~120 mM ( Nagata et al . , 2017 ) , much higher than measurements of Mg2+ in mammalian nuclei ( 16–18 mM ) ( Romani , 2013 ) . Per dimer , archaeal histones provide weaker electrostatic screening of DNA than eukaryotic histones , due to their relatively lower isoelectric points ( pI of ~8 vs~11 ) , and the greater overall negative charge of archaeasomes may encourage rapid sequestering of any free Mg2+ ions in the cell . This would induce a gradient-based mechanism for rapid and excessive uptake of cations to prevent chromatin compaction from monopolizing the Mg2+ necessary for enzyme action . On the other hand , the ability of T . kodakarensis archaeasomes to compact but not aggregate as a result of elevated Mg2+ concentration may be a direct result of evolving in high-salt environments , as regulating between internal and external concentrations may have been more energetically demanding than simply adapting their chromatin structure . More specific measures of local Mg2+ concentration in archaeal chromatin may help differentiate these two mechanisms . Archaeal histones were successfully expressed in E . coli cells and were found to coat bacterial DNA ( Rojec et al . , 2019 ) . Surprisingly , this resulted in only minor perturbations of cell growth , despite E . coli lacking the evolutionary machinery to handle histones . Furthermore , in vitro assays have shown that transcription through archaeasomes is slowed but not altogether stopped ( Xie and Reeve , 2004 ) . In eukaryotes , navigation of polymerases through chromatin is assisted by histone chaperones and ATP-dependent remodeling factors ( Hammond et al . , 2017; Clapier et al . , 2017; Markert and Luger , 2021 ) , but no known homologs to these complexes have been identified as yet in archaea . In absence of remodelers , the inherent dynamics of archaeasome-based chromatin may thereby allow limited access to chromatin by the sporadic ( and possibly stochastic ) appearance of near-zero linker DNA and Arc60 or Arc90 substates . On the other hand , it is possible that sequences encoding chromatin remodelers are yet to be found . On a related note , the ‘SMC-like’ coalescin proteins was recently identified in the histone-less Sulfolobus archaea ( Takemata and Bell , 2021 ) , and the access of these yet unidentified ‘hidden agents’ would similarly be tuned by the dynamics of the archaeasome . The degree to which DNA rigidity , variant histones , or unknown histone- and DNA-binding proteins regulate chromatin accessibility is an exciting future area of research’ .
Archaeasome systems of varying sizes were constructed from the crystal structure containing ~90 bp of DNA bound to histone HMfB ( PDB 5T5K ) . Systems containing ~90 base pairs of DNA and three histone dimers ( Arc90 ) , ~120 base pairs of DNA and four histone dimers ( Arc120 ) , and ~180 base pairs of DNA and six histone dimers ( Arc180 ) were created as prescribed by the crystal lattice . The Arc90 system is intended to represent the ‘fundamental unit’ of archaeasome-based chromatin , as it is the crystallographic unit of the solved structure as well as the smallest DNA protection footprint observed by MNase digestion ( Mattiroli et al . , 2017 ) . The Arc120 and Arc180 complexes provide systems in which to study the contribution of stacking histone-histone interactions , as well as the wrapping of additional DNA superhelical turns , in stabilizing or destabilizing the proposed archaeasome . Additionally , an Arc120 system with the G17D mutation was also generated ( Arc120-G17D ) , in analogy to the destabilizing G17D mutation that was previously studied in vivo . The overhanging DNA bases that formed crystallographic contacts were removed in our simulations . Each system was neutralized and solvated in a TIP3P box of 100 mM NaCl ( Jorgensen et al . , 1983 ) , and masses were repartitioned from heavy atoms to covalently bonded hydrogen atoms to allow for the use of a four fs timestep ( Hopkins et al . , 2015 ) . Parameters for protein atoms were taken from the Amber FF14SB forcefield ( Maier et al . , 2015 ) , DNA parameters were taken from the Amber bsc1 forcefield ( Ivani et al . , 2016 ) , and ions were parameterized according to the modifications of Joung and Cheatham , 2008 . Systems were then energy minimized for 5000 steps while constraining solute heavy atoms with a 10 kcal/mol/Å2 harmonic potential , followed by 5000 steps without restraints . After energy minimization , three independent simulations of each system were conducted according to the following protocol: simulations were heated from 10 K to 300 K over the course of 50 ps in the NVT ensemble with heavy atom restraints applied , system densities were equilibrated and positional restraints were slowly released over the course of 200 ps in the NPT ensemble ( target pressure of 1 atm ) , and simulations were extended for 300 ns without positional restraints in the NPT ensemble . During these production simulations , terminal base pair fraying was removed from the simulation by reinforcing hydrogen bonding in the terminal base pairs . No restraints were applied when participating atoms were within 3 . 5 Å of one another , but a harmonic potential was applied when participating atoms spread beyond this cutoff ( force constant = 5 . 0 kcal/mol/Å2 ) . Minimization and MD simulations were conducted in the pmemd engine ( v18 ) , with CUDA acceleration utilized for the simulations ( Salomon-Ferrer et al . , 2013 ) . System equilibration was monitored through RMSD calculations of backbone atom positions , and local flexibility was measured via the root mean-squared fluctuation ( RMSF ) of the backbone . DNA breathing dynamics were quantified using the center of mass distance between the terminal base pairs and the neighboring superhelical turn . Complex stabilities were assessed using the MM-GBSA method with the igb5 solvent parameters and mbondi2 atomic radii ( Onufriev et al . , 2004 ) . Trajectories and single frames were rendered using VMD , and structural analyses ( RMSF , RMSD , DNA breathing ) were calculated using cpptraj . Statistical significance between populations were determined by unpaired t-test . HTkA histones were provided by the Histone Source ( CSU Fort Collins , CO ) , and the 207 base pair , Widom 601-derived sequence was purified as described ( Mattiroli et al . , 2017; Dyer et al . , 2003 ) . Archaeasome complexes ( Arc207 ) were formed by mixing 207 bp DNA with HTkA histones at a 1:7 molar ratio of DNA to histone dimer in a buffer containing 100 mM KCl and 50 mM Tris ( pH 8 . 0 ) and incubated for 20 min at room temperature . Samples were then dialyzed in buffers containing 0 , 2 , 5 , 7 , or 10 mM MgCl2 at a volume ratio of 1:1000 two times , first for 2 hr and then overnight , as preparation for subsequent measurements . As histone stocks are suspended in high glycerol , the dialysis process simultaneously served to effectively remove glycerol . Sedimentation velocity analytical ultracentifugation ( SV-AUC ) measurements were conducted in absorbance mode ( λ = 260 nm ) . Samples were loaded in to an An60Ti rotor in 400 µL cells with two-channel Epon centerpieces and then spun at 35 , 000 rpm at 20°C in a Beckman XL-A ultracentrifuge . Partial specific volumes of the samples were determined using UltraScan3 ( v4 . 0 ) ( Demeler , 2005 ) . Time and radially invariant noises were subtracted through two-dimensional sediment analysis ( 2DSA ) , and the final 2DSA model parameters were used to initialize genetic algorithm and Monte Carlo analyses . Sedimentation coefficients were determined using van Holde-Weischet analysis ( in Svedberg units , corrected to solvent conditions of water at 20°C ) , and molecular weights and frictional ratios ( f/fo - the degree of elongation/flexibility in comparison to ideal spherical particles ) were determined from genetic algorithm analysis of SV-AUC traces . Theoretical f/fo values for modeled structures were derived using the SoMo plugin of UltraScan , where bead models were created through the ‘SoMo Overlap’ scheme and hydrodynamics were calculated with the ZENO algorithm ( Brookes et al . , 2010; Douglas et al . , 1994 ) . After overnight dialysis ( 100 mM KCl , 50 mM Tris pH 8 , 5 mM MgCl2 ) , Arc207 samples were concentrated to 0 . 9 mg of DNA per mL , and 4 µL of sample was deposited onto glow-discharged grids ( 40 mA for 45 s ) . Samples were screened on copper Cflat ( 1 . 2/1 . 3 ) grids , and formaldehyde-cleaned Quantifoil ( 2/2 ) grids were used for final data collection in order to increase image acquisition speed through a ‘2 × 2’ collection scheme , where a single defocus level is used for a cluster of four grid holes . Each grid was manually plunge frozen in liquid ethane and stored in liquid nitrogen . Datasets were collected on a FEI Tecnai F20 equipped with a Gatan K3 camera at 29 , 000x magnification in counting mode ( yielding 1 . 291 Å per pixel ) , 200 kV accelerating voltage , dosage rate of ~1 e-/Å2 per frame , and 50 frames per micrograph stack ( total dose of 50 e-/Å2 ) . A total of 5388 image stacks were collected for subsequent three-dimensional analysis . An Elsa Cryo-Transfer Holder ( Gatan , Inc ) and defocus range of −1 . 2 to −2 . 6 μm was used . Using the Relion interface ( v3 . 0 ) ( Zivanov et al . , 2018 ) , micrographs were motion corrected using MotionCorr2 ( Zheng et al . , 2017 ) , and CTF parameters were estimated from gCTF ( Zhang , 2016 ) . Ten micrographs were then randomly selected from a collection of motion-corrected micrographs and denoised using the janni_denoise . py function of the SPHIRE-crYOLO particle-picking pipeline ( Wagner , 2019 ) . Particles were manually picked from these denoised micrographs and used to train a neural network through crYOLO for picking particle coordinates across all micrographs , where the final particle predictions utilized the same denoising process ( Wagner et al . , 2019 ) . Densities shown here are the result of refinements to the largest dataset , collected in the 2 × 2 scheme . In total , 1 , 879 , 294 particles were predicted by crYOLO and extracted from the motion-corrected ( non-denoised ) micrographs using Relion and imported to CryoSparc ( v2 . 12 . 4 ) for two-dimesional ( 2D ) classification ( Punjani et al . , 2017 ) . 2D classes were manually filtered to remove particles containing primarily noise or interactions with overlapping neighbor particles . Subsequent 2D classifications showed two predominant particle types: the ‘Class I’ archaeasome state characterized by tightly wound DNA , and the ‘Class II’ archaeasome state showing nucleosome-like DNA arrangements with distinct out-of-plane densities , forming ‘base’ and ‘lid’ subunits . These classes were then separated from one another for three-dimensional ( 3D ) classification and refinement by CryoSparc . Class I density was refined through Bayesian Polishing of the contributing particles in Relion , but the Class II density saw no benefit from polishing and the CryoSparc-derived density is reported . Reconstructed volumes were deposited to the Electron Microscopy Data Bank ( EMD-23403 , EMD-23404 ) . Figures and videos were generated with VMD ( v1 . 9 . 3 ) ( Humphrey et al . , 1996 ) . Class I density ( containing volume of five histone dimers and ~150 bp of DNA ) was modeled by extracting five dimers and ~150 bp of DNA from the end state of a randomly selected Arc180 simulation . Simulation coordinates were fit to the EM-derived density through rigid body docking in Chimera . The Class II density was modeled by first docking Arc90 and Arc117 constructs in the smaller and larger volumes , respectively . Then , the bridging DNA segments were ligated via tleap and energy-minimized via pmemd in implicit solvent ( Amber FF14SB protein forcefield , DNA bsc1 parameters , and igb5 implicit solvent model with mbondi2 atomic radii modifications and a 100 mM monovalent salt environment ) . Calculation of nucleic acid geometry parameters was conducted with cpptraj . | All animals , plants and fungi belong to a group of living organisms called eukaryotes . The two other groups are bacteria and archaea , which include unicellular , microscopic organisms . All three groups have genes , which are typically stored on long strands of DNA . Eukaryotes have so much DNA that they use proteins called histones to help package and organize it inside each cell . Archaea also have simplified histones that help store their DNA , and studying these proteins could reveal how eukaryotic histones first evolved . In eukaryotes , groups of eight histones form a short cylinder that organizes a small section of DNA into a structure called a nucleosome . Each cell needs hundreds of thousands of nucleosomes to arrange its DNA . Eukaryotic cells also contain other proteins that release pieces of DNA from histones so that their genetic information can be used . The histones in Archaea don’t form discrete nucleosomes , instead , they coil DNA into ‘slinky-like’ shapes . It’s still unclear how DNA packing in archaea works and how it differs from eukaryotes . Bowerman , Wereszczynski and Luger used computer simulations , biochemistry and cryo-electron microscopy to study the histones from archaea . The archaeal ‘slinky-like’ histone structures are more flexible than nucleosomes , and can open and close like clamshells . This flexibility allows the information in the genomes of Archaea to be easily accessed , so , unlike in eukaryotes , archaeal cells may not need other proteins to release the DNA from the histones . The ability to package DNA allows cells to contain many more genes , so evolving histones was a vital step in the evolution of eukaryotic life , including the appearance of animals . Archaeal histones may reflect early versions of histones in eukaryotes , and can be used to understand how DNA packing has evolved . Furthermore , a greater understanding of Archaea may help better explain their role in health and global ecosystems , and allow their use in industrial applications . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2021 | Archaeal chromatin ‘slinkies’ are inherently dynamic complexes with deflected DNA wrapping pathways |
How proteins harness mechanical force to control function is a significant biological question . Here we describe a human cell surface receptor that couples ligand binding and force to trigger a chemical event which controls the adhesive properties of the receptor . Our studies of the secreted platelet oxidoreductase , ERp5 , have revealed that it mediates release of fibrinogen from activated platelet αIIbβ3 integrin . Protein chemical studies show that ligand binding to extended αIIbβ3 integrin renders the βI-domain Cys177-Cys184 disulfide bond cleavable by ERp5 . Fluid shear and force spectroscopy assays indicate that disulfide cleavage is enhanced by mechanical force . Cell adhesion assays and molecular dynamics simulations demonstrate that cleavage of the disulfide induces long-range allosteric effects within the βI-domain , mainly affecting the metal-binding sites , that results in release of fibrinogen . This coupling of ligand binding , force and redox events to control cell adhesion may be employed to regulate other protein-protein interactions .
Protein function is controlled by a variety of chemical modifications to amino acid side chains , cleavage or isomerization of peptide bonds and cleavage or formation of disulfide bonds ( Butera et al . , 2014a; Cook and Hogg , 2013 ) . These chemical reactions are usually facilitated by enzymes and their cofactors . Mechanical force is another factor that is increasingly being recognized to control protein chemical reactions ( Garcia-Manyes and Beedle , 2017; Cross , 2016 ) . Mechanical force can markedly reduce the reaction energy barrier . Reactions that are too slow become relevant on a biological time scale and others would not occur at all without the input of force . Low forces can trigger bond rotation and rupture of hydrogen bonds , while high forces can break or form covalent bonds . A number of cell surface receptors have been shown to be regulated by mechanical forces ( Chen et al . , 2017 ) , including the integrins . Vertebrates express 24 different integrins that comprise one of 18 different α-subunits and one of 8 different β-subunits ( Hynes , 2002 ) . Integrin-mediated adhesion and signalling events regulate virtually all cell growth and differentiation , while dysregulation of integrins are involved in the pathogenesis of cancer , auto-immune conditions and vascular thrombosis . The integrin heterodimers recognize overlapping but distinct sets of ligands on other cells , extracellular matrix or on pathogens . The integrins are type one trans-membrane proteins that consist of large extracellular segments characterized by various domains and small transmembrane and cytoplasmic segments . Most integrins exist on the cell surface in an inactive state that does not bind ligand or signal . Integrins are activated by intracellular stimuli such as talin binding through a process termed inside-out signalling , and by ligand occupancy that transduces extracellular signals into the cytoplasm through outside-in signalling ( Tadokoro et al . , 2003; Luo et al . , 2007 ) . Affinity for ligands is mostly controlled by global and local conformational rearrangements of the integrin ectodomains . The extracellular segments exist in at least three conformational states: the bent conformation with closed headpiece ( low affinity for ligand ) , the extended conformation with closed headpiece ( intermediate affinity ) and the extended conformation with open headpiece ( high-affinity ) ( Zhu et al . , 2013 ) . While much is known about the structure and function of resting and activated integrins , little is known about how integrins disengage from their ligands . For instance , when cells migrate , integrins adhere at the leading edge and de-adhere at the trailing edge . Force has been suggested as one of the mechanisms to explain both the engagement and disengagement of ligand ( Zhu et al . , 2008 ) . It was proposed that binding of the integrin β subunit cytoplasmic domain to actin filaments results in lateral translocation of the integrin heterodimer on the cell surface that causes integrin extension . Engagement of immobilised extracellular ligand greatly increases this lateral force that favours the high-affinity , open headpiece conformation . Disassembly of the actin cytoskeleton and dissociation of the β subunit cytoplasmic domain removes the lateral force that results in the closed headpiece conformation and ligand disengagement . In support of this model , cells become highly elongated and stop migrating when integrins are locked in the high-affinity conformation or when actin disassembly is blocked in the uropod ( Smith et al . , 2007 ) . Here we report a chemical modification of an activated integrin that results in disengagement of ligand . Platelet clumping at sites of blood vessel injury is mediated by cross-linking of platelet αIIbβ3 integrin by the bivalent ligand , fibrinogen . This integrin is critical for thrombus formation and is the target of successful anti-thrombotic agents in routine clinical use for acute coronary syndrome . ERp5 is a protein disulfide isomerase ( PDI ) family member oxidoreductase released from platelets upon activation ( Jordan et al . , 2005 ) , and from platelets and endothelial cells at sites of thrombosis in mice ( Passam et al . , 2015 ) . Human platelets contain about 13 , 300 molecules of ERp5 per platelet ( Burkhart et al . , 2012 ) , while mouse platelets contain an estimated 60 , 000 molecules ( Zeiler et al . , 2014 ) . Secreted ERp5 binds to the β3 subunit of platelet surface αIIbβ3 integrin ( Jordan et al . , 2005; Passam et al . , 2015 ) . Systemic inhibition of ERp5 with function-blocking antibodies inhibits thrombosis in mice , which implies an essential role for secreted ERp5 in this biology ( Passam et al . , 2015 ) . Our studies indicate that ERp5 mediates de-adhesion of activated platelet αIIbβ3 integrin from fibrinogen . Ligand binding to extended αIIbβ3 integrin triggers cleavage of the βI-domain Cys177-Cys184 disulfide by ERp5 , which is enhanced by mechanical force . Cleavage of the disulfide results in release of fibrinogen due to allosteric effects at the metal-ion-dependent adhesion ( MIDAS ) site .
Immunoblotting of human platelet lysate and releasate indicates that approximately half of the platelet ERp5 molecules are released into the supernatant upon activation ( Figure 1—figure supplement 1 ) . ERp5 binds to β3 integrin with a dissociation constant in the low micromolar range ( Passam et al . , 2015 ) , and an anti-ERp5 antibody has been reported to inhibit fibrinogen binding to activated platelets and platelet aggregation in vitro ( Jordan et al . , 2005 ) . These findings suggested that ERp5 directly regulates platelet αIIbβ3 integrin function , although the mechanism remains elusive . We examined the effect of soluble ERp5 on platelet αIIbβ3 integrin activation and fibrinogen binding . Incubation of washed platelets with ERp5 did not trigger αIIbβ3 integrin activation , and ERp5 had no effect on integrin activation by ADP ( Figure 1A ) . Integrin activation was measured by binding of PAC-1 , an antibody that recognizes the fully activated integrin with an open headpiece ( Shattil et al . , 1985; Luo et al . , 2003 ) . This result indicated that ERp5 is not directly involved in integrin activation , so we explored a role for ERp5 in post-activation events . Effect of soluble ERp5 on the kinetics of adhesion of washed platelets to fibrinogen as a function of fluid shear force was examined ( Figure 1B ) . There was a biphasic effect of ERp5 on platelet binding to fibrinogen at two wall shear stress ( Figure 1C ) . At 10 dyn/cm2 ( 1000 s−1 ) platelet adhesion in the first 4 min of flow was unaffected by ERp5 and reduced with time thereafter . The negative effect of ERp5 on platelet adhesion was more pronounced at 30 dyn/cm2 ( 3000 s−1 ) . Platelet adhesion in the first 1 min of flow was unaffected by ERp5 and reduced significantly with time thereafter . This finding suggested that ERp5 was triggering dissociation of fibrinogen from activated platelet αIIbβ3 integrin . To better understand this mechanosensitive phenomenon , the effect of ERp5 on binding of fibrinogen to αIIbβ3 integrin was characterized using a force spectroscopy technique - biomembrane force probe ( BFP ) ( Ju et al . , 2016 , Ju et al . , 2013 ) . The BFP brings an αIIbβ3 integrin coated bead into contact with a fibrinogen bearing force probe ( Figure 2A ) . Upon target retraction , it detects the αIIbβ3 bond from the pico-force signal measured ( Figure 2B , red ) , while a zero force indicates a no-bond event ( Figure 2B , black ) . The intermittent ‘touch and retract’ cycles mimic platelet translocation behavior under shear ( Ju et al . , 2013; Yago et al . , 2008 ) . The adhesion frequencies ( the number of ‘bond’ touches ( Figure 2F , red ) divided by the number of total touches ) reflects the binding affinity at zero tensile force . Unexpectedly , we found that soluble ERp5 had no significant effect on the αIIbβ3–fibrinogen adhesion frequencies ( Figure 2C ) . To investigate the force effect , we measured αIIbβ3 integrin–fibrinogen bond lifetimes at multiple clamped forces in the absence or presence of ERp5 . This interaction is characterized by slip-bond behaviors in which force accelerates bond dissociation , consistent with the previous optical tweezer study ( Litvinov et al . , 2011 ) . Surprisingly , at the low force regime of 5–15 pN , ERp5 displayed a similarly non-significant effect on the αIIbβ3–fibrinogen bond lifetimes as the adhesion frequency assay , whereas as force goes beyond 15 pN , it greatly enhanced αIIbβ3–fibrinogen dissociation with reduced bond lifetimes ( Figure 2D ) . In accordance with the perfusion experiments , these findings indicate that ERp5 has a force-dependent de-adhesive effect on the αIIbβ3–fibrinogen interaction . ERp5 is an oxidoreductase that can cleave , form or potentially rearrange disulfide bonds in protein substrates . From crystal structures of the complete ectodomain of αIIbβ3 integrin ( Zhu et al . , 2008 ) , 28 disulfide bonds have been defined in the β3 subunit . No unpaired cysteines within a few Angstoms of each other were identified , which implied that all possible disulfide bonds in the subunit are intact . We hypothesized that the functional effect of ERp5 on fibrinogen binding was a result of its cleavage of one or more of the 28 β3 disulfide bonds . This was tested by measuring the presence of unpaired cysteine thiols in platelet surface β3 before and after activation by ADP . Platelet activation resulted in increased labelling of β3 by a thiol-specific probe ( Figure 3A ) , indicating that one or more disulfide bonds were cleaved in the subunit upon platelet activation . To identity the β3 disulfide ( s ) cleaved by ERp5 , it was critical that we accurately quantify the redox state of the subunits disulfide bonds . This was achieved using a differential cysteine alkylation and mass spectrometry technique ( Pasquarello et al . , 2004; Bekendam et al . , 2016 ) . Briefly , reduced disulfide bond cysteines in purified platelet β3 integrin were alkylated with 2-iodo-N-phenylacetamide ( IPA ) , and the oxidized disulfide bond cysteines with a stable carbon-13 isotope of IPA following reduction with dithiothreitol ( Figure 3B ) . Sixty-eight cysteine containing peptides ( Supplementary file 1 ) reporting on 24 of the 28 β3 integrin disulfides were resolved by mass spectrometry and quantified ( Figure 3—figure supplement 1 ) . The four disulfides we were unable to map are the Cys528-Cys542 and Cys536-Cys547 bonds of the EGF-3 domain , Cys575-Cys586 of the EGF-4 domain and Cys601-Cys604 of the Ankle domain ( Figure 3C ) . Purified platelet αIIbβ3 integrin was incubated with 2- or 10-fold molar excess of ERp5 or PDI and the redox state of the disulfides quantified . PDI was used to test the substrate specificity of ERp5 . PDI is the archetypal member of the PDI family of oxidoreductases , which includes ERp5 , and is also secreted by platelets and endothelial cells at sites of thrombosis in mice and binds to surface β3 integrin ( Cho et al . , 2008; Cho et al . , 2012 ) . Reactions were also performed with redox-inactive ERp5 and PDI to test the redox dependence of their action . Redox-inactive ERp5 and PDI were produced by replacing the active-site cysteines with alanines . In accordance with crystal structures of the integrin ectodomain , all 24 disulfide bonds in untreated β3 integrin were >90% oxidized , with one exception ( Figure 3D and E ) . Approximately 10% of the βI-domain Cys177-Cys184 bond was reduced in the β3 preparations . Incubation of the integrin with ERp5 or PDI did not significantly change the redox sate of any of the 24 disulfide bonds ( Figure 3D and E ) . Our fibrinogen binding studies suggested that ERp5 was having an effect on the extended/activated integrin . The native integrin exists predominantly in a bent conformation with closed headpiece . Soaking of the αIIbβ3 integrin headpiece with RGD peptide ligand results in variably extended configurations ( Zhu et al . , 2013 ) . Six intermediate conformations and fully extended conformation with open headpiece have been described . Incubation of RGD-bound αIIbβ3 integrin with ERp5 resulted in significant cleavage of only one of the 24 β3 disulfides: the βI-domain Cys177-Cys184 bond . There was dose-dependent cleavage of the disulfide by ERp5 ( Figure 3F ) and the bond was not cleaved by PDI ( Figure 3G ) , indicating selectivity for ERp5 . Control RGE peptide that does not bind the integrin did not facilitate cleavage of the disulfide by ERp5 ( Figure 3—figure supplement 2 ) . As anticipated , redox-inactive ERp5 did not cleave the bond ( Figure 3F ) . These results indicate that ERp5 specifically cleaves the βI-domain Cys177-Cys184 disulfide in one or more of the extended/open αIIbβ3 conformations . Approximately 30% of the Cys177-Cys184 disulfide bond in the purified integrin preparation is cleaved by 10-fold molar excess of ERp5 under static conditions , which is a ~20% increase over baseline . The force spectroscopy findings suggest that extent of cleavage is likely to be higher when the integrin is subject to mechanical shearing . ERp5-mediated dissociation of fibrinogen from αIIbβ3 is greatly enhanced when force goes beyond 15 pN and is complete at 40 pN ( Figure 2D ) . The βI-domain Cys177-Cys184 disulfide bond is also cleaved on the platelet surface by platelet ERp5 . Washed human platelets were incubated function-blocking anti-ERp5 antibodies or isotype control antibodies and the redox state of the βI-domain disulfide determined in the integrin immunoprecipitated from lysate . Lysis of platelets releases stored ERp5 and fibrinogen that was predicted to mediate cleavage of the Cys177-Cys184 disulfide bond . The Cys177-Cys184 disulfide bond was reduced in ~10% of the integrin population in untreated or control antibody treated platelet lysate ( Figure 3H ) , which is in accordance the redox state of this bond in purified platelet αIIbβ3 ( Figure 3D–G ) . Incubation of platelets with function-blocking anti-ERp5 antibodies during lysis inhibited reduction of the Cys177-Cys184 bond ( p<0 . 001 ) ( Figure 3H ) . The Cys177-Cys184 disulfide is a –/+RHhook in all crystal structures of the protein , which includes bent , extended , ligand-free and ligand-bound structures ( Supplementary file 2 ) . Disulfide bonds are classified based on the geometry of the five dihedral angles that define the cystine residue ( Schmidt et al . , 2006 ) . Twenty possible disulfide bond configurations are possible using this classification scheme and all 20 are represented in protein structures . Some disulfide bonds are cleaved in the mature protein to control function ( Hogg , 2003 ) , the so-called allosteric bonds ( Schmidt et al . , 2006 ) , and these disulfides are increasingly recognized to have one of three configurations ( Butera et al . , 2014a ) . The –/+RHhook is one of these configurations , along with –RHstaple and –LHhook bonds ( Butera et al . , 2014b ) . The conformational constraints imposed on the –/+RHhook and –RHstaple disulfides by topological features stress the bonds via direct stretching of the sulfur-sulfur bond and neighbouring angles ( Schmidt et al . , 2006; Zhou et al . , 2014 ) . Stretching of sulfur-sulfur bonds increases their susceptibility to cleavage ( Baldus and Gräter , 2012; Li and Gräter , 2010; Wiita et al . , 2006; Wiita et al . , 2007 ) , so this internal stress fine tunes bond cleavage and thus the function of the protein in which the bond resides . The position of the Cys177-Cys184 disulfide relative to the ligand binding pocket and access to the bond by ERp5 was examined in the crystal structure of the extended holo headpiece ( PDB code 2vdo ) . Cys184 of the bond is surface exposed ( Supplementary file 2 ) on a face of the βI-domain that is remote from the ligand binding pocket ( Figure 4A ) . This suggests that the Cys184 sulfur atom of the disulfide is attacked by ERp5 to cleave the bond and explains why ERp5 access is not blocked by RGD ligand binding . The ERp5 N-terminal part consists of two thioredoxin-like domains containing a catalytic dithiol/disulfide in CysGlyHisCys motifs , a and a’ , separated by an x segment ( Figure 4B ) . These domains are followed by a possible substrate binding domain , b . The redox potentials of the a ( Cys55-Cys58 ) and a' ( Cys190-Cys193 ) catalytic disulfides of ERp5 were determined using differential cysteine alkylation and mass spectrometry . The equilibrium data is shown in Figure 4B . The standard redox potentials of the a and a' domain disulfides of ERp5 are −206 mV and −211 mV , respectively . These redox potentials are about mid-way between the potentials of the PDI ( Bekendam et al . , 2016 ) and thioredoxin ( Lundström and Holmgren , 1993 ) catalytic disulfides . N- and C-terminal fragments of ERp5 containing a single active-site were tested for cleavage of the Cys177-Cys184 disulfide . Both fragments cleaved the bond with the same efficiency as full-length protein ( Figure 4C ) . This is in agreement with the equivalent redox potentials of the active-site dithiols/disulfides ( Figure 4B ) . It also indicates that the substrate binding domain of ERp5 is not required for access to and cleavage of the Cys177-Cys184 disulfide . It was possible , though , that separating the two catalytic domains of ERp5 would influence substrate specificity , that is , the disulfide bond or bonds cleaved by ERp5 . This was tested by examining the effect of the ERp5 fragments on adhesion of washed human platelets to fibrinogen in the first 4 min of flow at a shear rate of 1000 s−1 . These conditions were chosen as full-length ERp5 has no effect over this time frame at this shear rate ( Figure 1C ) . As for full-length ERp5 and redox-inactive ERp5 , where the active site cysteines of both thioredoxin-like domains are replaced with serines , the N-terminal catalytic domain of ERp5 had no effect on platelet adhesion to fibrinogen under these conditions . In marked contrast , the C-terminal domain enhanced platelet adhesion to fibrinogen by ~50 fold ( Figure 4D ) . Large platelet aggregates adhered to the fibrinogen-coated slides ( Figure 4D inset ) . The C-terminal domain also enhanced platelet aggregation is response to a PAR-1 agonist , while full-length ERp5 had no effect ( Figure 4E ) . These findings indicate that both catalytic domains in full-length ERp5 are required for specificity of cleavage of the Cys177-Cys184 disulfide . The result implies that separating the domains leads to cleavage of other disulfide bonds in the system and different functional effects . An intact Cys177-Cys184 disulfide bond was found to be required for normal fibrinogen binding in a 2004 structure/function study of the disulfide bonds in β3 integrin ( Kamata et al . , 2004 ) . We confirmed this result by measuring binding of soluble or immobilized fibrinogen to BHK cells expressing wild-type or disulfide mutant ( C177 , 184S ) αIIbβ3 integrin in the absence or presence of the integrin activator , Mn2+ . Expression of wild-type ( 64 . 8% of cells ) and disulfide mutant integrin ( 82 . 9% of cells ) was confirmed by staining with anti-β3 antibody ( Figure 5A ) . Soluble fibrinogen bound to less than 20% of wild-type or disulfide mutant β3 positive cells in the absence of Mn2+ ( Figure 5B ) . In the presence of Mn2+ , binding of soluble fibrinogen to αIIbβ3 with a broken Cys177-Cys184 disulfide bound was impaired ( p<0 . 05 ) compared to wild-type integrin ( Figure 5B ) . Cells expressing wild-type integrin adhered to immobilized fibrinogen in the absence and presence of Mn2+ , whereas adherence of the disulfide mutant integrin was severely impaired in both conditions ( p<0 . 0001 in the absence of Mn2+ and p<0 . 001 in the presence of Mn2+ ) ( Figure 5C ) . The structural changes in the oxidized and reduced states of the αIIbβ3 integrin headpiece that underpin the impaired affinity for fibrinogen were examined by Molecular Dynamics simulations . Three different αIIbβ3 starting structures were analyzed; the bent conformation ( PDB code 3fcs ) , extended apo conformation ( PDB code 3fcu ) , and extended holo conformation that includes the fibrinogen γC peptide ( PDB code 2vdo ) . We find disulfide bond reduction to render the whole headpiece more flexible in all three cases , as reflected by a wider distribution of conformations along conformational modes as obtained from Principal Component Analysis ( Figures 6A and Figure 6—figure supplement 1 ) . The strongest effect is observed for the extended holo structure compared to bent and extended apo structures ( compare Figure 6A with Figure 6—figure supplement 1 ) . The allosteric network originating from disulfide bond reduction , measured by force distribution analysis , emanates from the metal binding sites into the domain periphery , as shown by using decreasing force reduction , whereas the force differences in the β-propeller domain are minor . At lower forces , however , this domain is also allosterically reached . In more detail , the network involves both cysteines , the critical residue for ligand binding D119 , as well as residues involved in ion positioning , such as D217 and N214 ( Figure 6E and Figure 6—figure supplement 1 ) .
ERp5 secreted by activated platelets binds to the β3 subunit of platelet αIIbβ3 integrin ( Jordan et al . , 2005; Passam et al . , 2015 ) . We now report that ERp5 cleaves the βI-domain Cys177-Cys184 disulfide bond nearby the fibrinogen binding pocket of extended activated integrin that results in release of fibrinogen ( Figure 7 ) . Two coupled events control cleavage of the disulfide bond: ligand binding and mechanical force . RGD-ligand binding to the integrin and shear force facilitate ERp5 reduction of the disulfide . Stretching of sulfur-sulfur bonds , either by internal pre-stress or external forces , increases their susceptibility to cleavage ( Baldus and Gräter , 2012; Li and Gräter , 2010; Wiita et al . , 2006; Wiita et al . , 2007 ) . Our data suggest that ERp5 cleavage of the disulfide is enabled by ligand- and force-dependent stretching of the sulfur-sulfur bond . However , the experimental data are also consistent with a force-dependent , ligand-bound conformation that provides enhanced access of the disulfide to ERp5 . This force-coupled ligand binding redox event is an intriguing example of mechano-chemically coupled catalysis ( Neumann and Tittmann , 2014 ) . Our findings are also of significance in understanding how platelets harness force to balance haemostasis and thrombosis functions . The Cys177-Cys184 disulfide bond is exposed to solvent and accessible to ERp5 on a face of the βI-domain that is not involved in ligand binding . Cleavage of the disulfide bond , however , results in long-range allosteric effects within the βI-domain of αIIbβ3 , mainly affecting the metal-binding sites , along with a higher conformational mobility of the whole βI-domain . Interestingly , we do not observe significant changes of α-helix 7 , which has been previously shown to regulate the affinity of the integrin for ( RGD ) ligands ( Luo and Springer , 2006 ) . Our data instead suggest that the increased mobility due to disulfide bond scission leads to local effects around the RGD-binding MIDAS site , which in turn are directly responsible for the observed affinity reduction of αIIbβ3 for fibrinogen upon Cys177-Cys184 reduction by ERp5 . We note that we cannot exclude allosteric effects on the α-helix seven on longer time scales . Both the N- and C-terminal catalytic domains of ERp5 are required for specificity of cleavage of the Cys177-Cys184 disulfide . Although the individual domains , when expressed and tested separately , were found to cleave the Cys177-Cys184 disulfide with comparable efficiency , removal of the N-terminal thioredoxin-like domain from ERp5 resulted in a fragment that markedly promoted platelet clumping on a fibrinogen-coated surface under fluid shear and enhanced PAR-1 mediated platelet aggregation . Interestingly , the C-terminal thioredoxin domains of ERp57 and PDI also potentiate platelet aggregation ( Zhou et al . , 2015; Wang and Essex , 2017 ) . The PDI C-terminal thioredoxin domain mediates P-selectin expression and ATP secretion in a αIIbβ3-independent manner . The effects of the isolated thioredoxin domains are likely the result of cleavage of other disulfide bond ( s ) in platelet proteins . These findings highlight the possibility that proteolytic processing of the oxidoreductases in the circulation may generate fragments with independent activities . They also suggest that care that needs to be taken when targeting oxidoreductases for development of new anti-thrombotics ( Flaumenhaft et al . , 2015 ) . Inhibiting different aspects of the factors may have unintended consequences . Our studies of ERp5’s role in thrombus formation in vivo ( Passam et al . , 2015 ) and ERp5 effects on platelet adhesion under flow shown herein , indicate that ERp5 is involved in propagation of the thrombus and not thrombus initiation . ERp5 accumulates in the developing thrombus ( Passam et al . , 2015 ) , which may be a mechanism to limit thrombus growth by shifting the balance between fibrinogen cross-linking of platelet αIIbβ3 and fibrinogen dissociation from this receptor . Testing this theory in vivo will be complicated by the likely possibility that ERp5 has more than one substrate in the thrombus . For instance , our preliminary results indicate that ERp5 significantly enhances binding of platelets to von Willebrand factor . The βI Cys177-Cys184 disulfide bond is conserved in 7 of the 8 β integrins , or 23 of the 24 vertebrate integrins ( Supplementary file 3 ) . β4 of the α6β4 laminin receptor is the only β integrin that does not contain the bond . Notably , PDI and ERp57 , like ERp5 , have been shown to bind to β3 integrins . PDI binds to β3 integrins in the thrombus ( Cho et al . , 2012; Kim et al . , 2013 ) and αvβ3 integrin on endothelial cells ( Swiatkowska et al . , 2008 ) , while ERp57 binds to platelet surface β3 integrin ( Holbrook et al . , 2012; Wang et al . , 2013 ) . It is possible that different oxidoreductases regulate de-adhesion of different integrins by cleaving the βI-domain disulfide bond . Neutrophil PDI , for instance , modulates ligand binding to αMβ2 integrin and neutrophil recruitment during venous inflammation ( Hahm et al . , 2013 ) . Future studies will show to what extent mechano-redox regulation is a mechanism at play beyond β3/ERp5 to control other integrins and other protein-protein interactions .
All procedures involving collection of human blood from healthy volunteers were in accordance with St George Hospital Human Ethics Committee ( HREC 16/009 ) , Human Research Ethics Committee of the University of Sydney ( HREC 2014/244 ) and the Helsinki Declaration of 1983 . Whole blood was drawn into ACD-A tubes ( BD Vacutainer ) and platelet rich plasma collected by centrifugation at 200 g for 20 min at room temperature . Following addition of 1 µM prostaglandin E1 , platelets were collected by centrifugation at 800 g for 20 min , washed with Hepes Tyrodes glucose buffer ( 20 mM Hepes , 134 mM NaCl , 0 . 34 mM Na2HPO4 , 2 . 9 mM KCl , 12 mM NaHCO3 , 1 mM MgCl2 , 5 mM glucose , pH 7 . 4 ) and resuspended in the same buffer at a concentration of 300 , 000–400 , 000 per µL . Recombinant ERp5 and protein disulfide isomerase ( PDI ) were produced in E . coli as described ( Passam et al . , 2015 ) . Plasmids for N- and C-terminal domains of ERp5 were from Thomas Spies , Fred Hutchinson Cancer Research Centre , USA . Platelets ( 106 in 0 . 1 mL ) were incubated with ERp5 ( 2 µM ) in the absence or presence of ADP ( 20 µM ) for 2 min at room temperature . Platelets were stained with fluorescene-conjugated PAC-1 antibody ( RRID:AB_2230769 ) ( 0 . 5 µg/ml ) for 30 min at 25°C , washed and fixed with 1% paraformaldehyde , and binding measured by flow cytometry on a BD FACS Canto II . The assembly and function of the platelet flow chambers are described in more detail in Bio-protocol ( Dupuy et al . , 2019 ) . Microchannels of Vena8 Fluor Biochip ( Cellix Ltd ) were coated with 10 µL of 20 µg/mL human fibrinogen overnight at 4°C in a humidified box , blocked with 10 µL of 0 . 1% bovine serum albumin in phosphate buffered saline ( PBS ) for 1 hr at room temperature and washed with 40 µL of PBS . Washed platelets were prepared as above , labeled with 1 µg/mL calcein ( Thermo Fisher ) and injected by Mirus NanoPump into the channels at shear rates of 1000 s−1 and 3000 s−1 ( flow rates of 40 and 120 µl/min , respectively ) within 3 hr from blood collection . Adhesion of platelets was monitored in real time with images captured via an ExiBlu CCD camera ( Q imaging , Canada ) connected to an AxioObserver A1 Inverted Epi-Fluorescence microscope ( Zeiss , Germany ) . Images were captured using the accompanying VenaFlux 2 . 3 imaging software . Images were analyzed at positions 2 , 4 and 6 ( located at 6 , 14 and 22 mm from the entry site of blood ) of the microchannels at 1 min intervals ( since initiation of flow ) using the ImagePro Premier 64-bit software . These positions are representative of flow nearest , mid-way and furthest from the entry of blood into the channel . Data were exported into Excel and area coverage by platelets was calculated for each position . Results are from 4 to 9 donors for each experiment . Aggregation studies were performed with washed platelets prepared as described ( Zhou et al . , 2015; Wang et al . , 2013 ) . Washed platelets were resuspended in Hepes Tyrodes glucose buffer at 300 , 000 per µL . Platelets were incubated with ERp5 proteins for 3 min and aggregation initiated with 7 µM TRAP-6 ( Roche ) . Aggregation was measured by light transmission using a ChronoLog Aggregometer ( Model 560-Ca ) . Results are from three healthy donor platelets and expressed as % aggregation over time . Biomembrane Force Probe ( BFP ) experiments were performed as we previously described ( Ju et al . , 2013; Ju et al . , 2015a; Ju et al . , 2015b ) . Briefly , the BFP utilizes two micropipettes , one aspirating a biotinylated human red blood cell ( RBC ) with a glass bead to serve as a force transducer , termed ‘probe’ . The bead is attached on the RBC apex via biotin-SA interaction . The probes spring constant , set to 0 . 25–0 . 3 pN/nm , is determined by the aspiration pressure and the radii of the micropipette and the RBC-bead contact area . The other micropipette aspirates a second bead , termed ‘target’ . The probe and target beads are respectively coupled with purified human fibrinogen and αIIbβ3 with maleimide-PEG3500-NHS ( JenKem , USA ) in carbonate/bicarbonate buffer ( pH 8 . 5 ) . The force spectroscopy traces are obtained by measuring the RBC-bead deflection from the probe beads edge tracking . Bond formation/dissociation and force application are enabled and monitored in controlled BFP touch cycles ( ~2 . 5 s each ) . Other details are found in published protocols ( Chen et al . , 2015; Ju et al . , 2017 ) . RBCs are isolated , biotinylated with Biotin-PEG3500-NHS ( JenKem , TX ) and stored ( up to 2 weeks ) for BFP experiments . In each cycle , the αIIbβ3-bearing target bead is driven to approach and contact the fibrinogen-probe bead with a 20-pN compressive force for a certain contact time ( 0 . 2 s ) that allows for bond formation . The target is then retracted at a constant speed ( 3 . 3 μm/s ) for bond detection . During the retraction phase , a ‘bond’ event is signified by a tensile force . Conversely , no tensile force indicates a ‘no-bond’ event . For the adhesion frequency assay , ‘bond’ and ‘no-bond’ events are enumerated to calculate an adhesion frequency in 50 repeated cycles for each probe–target pair . For a force-clamp assay to measure bond lifetimes , upon detection of ‘bond’ event in a similar BFP cycle , a feedback loop pauses the retraction at a desired clamped force ( 5–60 pN ) until bond dissociation . After that , the target is recoiled to the original position to complete the cycle . Lifetimes are measured from the instant when the force reaches the desired level to the instant of bond dissociation . The redox states of 24 of the 28 β3 integrin disulfide bonds was measured in isolated human platelet β3 integrin ( Abcam ) . Unpaired cysteine thiols in β3 integrin were alkylated with 5 mM 2-iodo-N-phenylacetamide ( 12C-IPA , Cambridge Isotopes ) for 1 hr at room temperature , the protein resolved on SDS-PAGE and stained with colloidal coomassie ( Sigma ) . The β3 band was excised , destained , dried , incubated with 40 mM dithiothreitol and washed . The fully reduced protein was alkylated with 5 mM 2-iodo-N-phenylacetamide where all six carbon atoms of the phenyl ring have a mass of 13 ( 13C-IPA , Cambridge Isotopes ) . The gel slice was washed , dried and deglycosylated using 5 units PNGase F ( Sigma ) , before digestion of β3 integrin with 12 . 5 ng/μl of chymotrypsin ( Roche ) in 25 mM NH4CO2 and 10 mM CaCl2 for 4 hr at 37°C followed by digestion with12 . 5 ng/μl of trypsin overnight at 25°C . Peptides were eluted from the slices with 5% formic acid , 50% acetonitrile . Liquid chromatography , mass spectrometry and data analysis were performed as described ( Chiu et al . , 2014; Cook et al . , 2013 ) . Sixty-eight peptides encompassing disulfide Cys residues were resolved and quantified ( Figure 3—figure supplement 1 ) . The levels of the different redox forms of the cysteines was calculated from the relative ion abundance of peptides labelled with 12C-IPA and/or 13C-IPA . To calculate ion abundance of peptides , extracted ion chromatograms were generated using the XCalibur Qual Browser software ( v2 . 1 . 0; Thermo Scientific ) . The area was calculated using the automated peak detection function built into the software . More detailed protocol for differential cysteine labelling and mass spectrometry quantification is described in Bio-protocol ( Chiu , 2019 ) . The redox potentials of the a ( Cys53-Cys56 ) and a' ( Cys397-Cys400 ) active-site dithiols/disulfides of ERp5 were determined by differential cysteine alkylation and mass spectrometry . Recombinant ERp5 ( 5 µM ) was incubated in argon-flushed phosphate-buffered saline containing 0 . 1 mM EDTA , 0 . 2 mM oxidized glutathione ( GSSG , Sigma ) and various concentrations of reduced glutathione ( GSH , Sigma ) for 18 hr at room temperature to allow equilibrium to be reached . Microcentrifuge tubes were flushed with argon prior to sealing to prevent oxidation by ambient air during the incubation period . Unpaired cysteine thiols in ERp5 were alkylated with 5 mM 12C-IPA for 1 hr at room temperature . The proteins were resolved on SDS-PAGE and stained with SYPRO Ruby . The ERp5 bands were excised , destained , dried , incubated with 100 mM dithiothreitol ( DTT ) and washed . The fully reduced proteins were alkylated with 5 mM 13C-IPA and the gel slices washed and dried before digestion of proteins with 12 ng/μL of chymotrypsin ( Roche ) in 25 mM NH4CO2 overnight at 25°C . Peptides were eluted from the slices with 5% formic acid , 50% acetonitrile . Liquid chromatography , mass spectrometry and data analysis were performed as described ( Cook et al . , 2013 ) . The fraction of reduced active-site disulfide bond was measured from the relative ion abundance of peptides containing 12C-IPA and 13C-IPA . To calculate ion abundance of peptides , extracted ion chromatograms were generated using the XCalibur Qual Browser software ( v2 . 1 . 0; Thermo Scientific ) . The area was calculated using the automated peak detection function built into the software . The ratio of 12C-IPA and 13C-IPA alkylation represents the fraction of the cysteine in the population that is in the reduced state . The results were expressed as the ratio of reduced to oxidized protein and fitted to Equation 1: ( 1 ) R=[GSH]2GSSGKeq+[GSH]2GSSGwhere R is the fraction of reduced protein at equilibrium and Keq is the equilibrium constant . The standard redox potential ( E0′ ) of the ERp5 active-site disulfides were calculated using the Nernst equation ( Equation 2 ) : ( 2 ) E0'=EGSSG0'-RT2FlnKequsing a value of −240 mV for the standard redox potential of the GSSG disulfide bond . Washed platelets ( 106 in 0 . 1 mL ) were prepared as above , labeled with 12C-IPA ( 5 mM ) for 1 hr at room temperature , centrifuged at 2000 g for 10 min and washed with Hepes Tyrodes glucose buffer . Platelets were lysed with 0 . 1 mL of 2% NP40 , 30 mM Hepes , 150 mM NaCl , 2 mM EDTA , pH 7 . 4 buffer containing proteinase inhibitor cocktail , and lysate was collected after centrifugation at 10 , 000 g for 20 min . Lysate ( 2 mg ) and AP3 antibody ( RRID:AB_2056630 ) ( 40 µg ) were mixed in 0 . 5 ml of IP/lysis buffer ( Pierce ) and rotated overnight at 4°C . αIIbβ3 integrin was collected on 80 µl of protein A/G agarose with rotation for 2 hr at 25°C . The beads were washed three times with IP lysis/buffer and three times with PBS . 12C-IPA ( 5 mM ) was added to the beads and incubated with rotation for 1 hr at 25°C . The integrin was eluted from the beads with 0 . 1 M glycine , the pH neutralized with 0 . 1 M Tris , pH 9 . 5 buffer and the β3 subunit resolved on SDS-PAGE and processed as above . cDNA of β3 was cloned in pcDNA3 vector ( carrying the neomycin resistance gene ) and cDNA of αIIb was cloned into pCEP4 vector ( carrying the hygromycin resistance gene ) as previously described ( Mor-Cohen et al . , 2008 ) . The β3 C177 , 184S mutant was created in the pcDNA β3 vector by site-directed mutagenesis using the QuikChange kit from Stratagene . Verification of mutations was confirmed by DNA sequencing . Plasmids were linearized using PVUI for β3/pcDNA3 and AvrII for αIIb/pCEP4 . BHK cells , grown in DMEM supplemented with 2 mg/ml L-glutamine and 5% FCS , were co-transfected with 1 µg of wild type or mutant pcDNA/β3 and 1 µg of pCEP4/αIIb with lipofectamine . Cells were also transfected with empty pcDNA and pCEP4 vectors as negative control . Transfected cells were grown in media containing 0 . 5 mg/ml hygromycin and 0 . 5 mg/ml G418 . Two separate transfections for wild type and mutant construct were performed . Cells were sorted for comparable expression of the integrin by staining with fluorescene-conjugated anti-CD61 antibody ( RRID:AB_929170 ) and flow cytometry . For fibrinogen binding assays , near-confluent transfected cells were washed twice with phosphate-buffered saline ( PBS ) and detached by adding 1 mM EDTA . Cells were suspended in DMEM , pelleted and washed twice before resuspending 0 . 5 × 106 in 0 . 1 mL PBS supplemented with 1 mM MgCl2 , 1 mM CaCl2 without or with 1 mM MnCl2 . Cells were incubated wth 10 µg/mL CD61-APC and 0 . 1 mg/mL fibrinogen-FITC for 30 min at 25ºC , washed with PBS , fixed with 1% paraformaldehyde , washed twice more with PBS and binding measured by flow cytometry on a BD FACS Canto II . Binding is expressed as the % of CD61+ cells that bound fibrinogen . For cell adhesion assays , near confluent cells were washed in PBS and detached by adding 5 mM EDTA . Cells were suspended to 106 cells/mL in 0 . 5 mL of 20 mM HEPES pH 7 . 4 buffer containing 0 . 1 M NaCl2 , 1 mM CaCl2 , 1 mM MgCl2 without or with 1 mM MnCl2 . Flat bottom 96 well MaxiSorp Nunc-Immuno plates ( Thermo Fisher ) were coated overnight with 40 µg/mL fibrinogen and blocked for two hours with 10 mg/mL denatured BSA . Cells , 100 μL of 106 cells/mL , were added to the wells , allowed to adhere for 2 hr at 37°C , and then gently washed three times with 200 µL PBS . The wells were overlayed with 100 µL PBS and imaged at 10x magnification using an inverted light microscope . Experiments were performed in triplicate and three images were taken per well . The number of attached cells were counted using ImageJ and a custom macro written for this purpose . Cell attachment is represented in absolute numbers . The crystal structures of the bent conformation of αIIbβ3 ( PDB code 3fcs [Zhu et al . , 2008] ) , extended apo conformation of αIIbβ3 ( PDB code 3fcu [Zhu et al . , 2008] ) , and extended holo conformation of αIIbβ3 ( PDB code 2vdo [Springer et al . , 2008] ) were prepared by separating the headpiece from the other domains ( β-propeller residues 1–452 , βI-domain residues 105–353 , and 14 residue bound-peptide in the holo ) . The termini were capped in PyMOL version 1 . 8 ( Schrodinger LLC , 2015 ) ( acetyl group on N-terminus , N-methylaminly group on C-terminus ) , sugars were neglected and calcium and magnesium ions were kept in the system . The protonation states of the amino acids were calculated using PROPKA ( Rostkowski et al . , 2011 ) . A dodecahedric box , leaving at least 1 . 5 nm distance between protein and box boundaries , was considered . The box was filled with water and the total system charge was neutralized with sodium and chloride ions ( 150 mM ) . For calcium and magnesium ions the default Amber parameters were used , for sodium and chloride the optimized Joung parameters ( Joung and Cheatham , 2008 ) were considered , and TIP3P ( Jorgensen et al . , 1983 ) was used as water model . MD simulations of each redox state ( C177-C184 in the βI-domain oxidized and reduced ) were performed with GROMACS 2016 ( Abraham et al . , 2015 ) , using the Amber99sb*-ILDN force field for the protein ( Lindorff-Larsen et al . , 2010 ) . Energy minimization was performed using the steepest-descent algorithm , followed by 500 ps of NVT-ensemble equilibration ( Berendsen thermostat ( Berendsen et al . , 1984 ) with τ = 0 . 1 ps ) with position restraints on the protein ( restraint force constant = 1000 kJ mol−1 nm2 ) while random starting velocities were assigned to each trajectory . Next , 1 ns of NPT-ensemble equilibration with position restraints on the protein was performed . The LINCS algorithm ( Hess , 2008 ) was used to constraint bonds involving protein hydrogen atoms , while SETTLE ( Miyamoto and Kollman , 1992 ) was used to constraint both bonds and angles of water molecules , enabling an integration time step of 2 fs . The temperature was kept constant at 300 K by coupling the system to the V-rescale ( Berendsen et al . , 1984; Bussi et al . , 2007 ) thermostat with τ = 0 . 1 ps . The pressure was kept constant at 1 bar coupling isotropically the system to a Parinello-Rahman barostat ( Parrinello and Rahman , 1981 ) , with τ = 5 ps and compressibility 4 . 5 × 10−5 bar−1 . Lennard-Jones interactions were calculated using a cut-off of 1 nm and long-range electrostatics were calculated by particle-mesh Ewald summation ( Darden et al . , 1993 ) . The Verlet-buffer scheme was employed to treat the non-bonded neighbour interactions . For each of the three structures and each redox state , five independent trajectories were calculated , in production runs of 200 ns in length each , while saving system coordinates every 10 ps . From each run the first 50 ns were accounted as equilibration time and discarded for subsequent analysis , resulting in 750 ns of cumulative simulation time per redox state for each of the starting structures . Principal Component Analysis ( Amadei et al . , 1993 ) , consisting in the calculation and diagonalization of the atomic-position covariance matrix , was performed using the GROMACS 2016 analysis tools , on the least-squares fitted trajectories of the backbone atoms . Eigenvectors of the covariance matrix were calculated of all ten concatenated runs ( both oxidized and reduced ) . Each concatenated trajectory of each redox states was then individually projected along the first two eigenvectors , which accounted for approximately 30% of the collective motions . Force distribution analysis ( FDA ) of MD simulations provides insight into mechanical signal propagation within a protein of interest upon an external perturbation ( Stacklies et al . , 2011 ) , such as the reduction of the Cys177-Cys184 bond in the βI-domain of αIIbβ3 by ERp5 investigated here . FDA was performed on the trajectories of both redox states ( Cys177-Cys184 oxidized and reduced ) , using the FDA implementation in GROMACS 5 . 0 . 7 ( Stacklies et al . , 2011 ) and the corresponding FDAtools 1 . 0 . The pairwise inter-residue forces Fij between residues i and j , were calculated from each frame . Pairwise forces within the atoms of the protein were used , whereas forces from water and ions were neglected . The time-average <Fij> was computed for each redox form and pairs of residues for which <Fij ( reduced ) > - <Fij ( oxidized ) > was larger than specified threshold cut-off values were presented . To model the open structure of the complete αIIbβ3 integrin ectodomain , the bent configuration ( PDB code 3fcs [Zhu et al . , 2008] ) was used as the starting structure . In order to open the structure , two peptide bonds were broken: residues 600–601 in chain A to simulate the opening of the hinge between calf1 and the thigh ( ‘α knee’ ) , and residues 475–476 in chain B to simulate the opening of the hinge between EGF-1 and EGF-2 ( ‘β knee’ ) . The opened model was structurally aligned to its appropriate domains in the opened model structure of the αVβ3 integrin ectodomain ( PDB code 3ije [Zucker et al . , 2016; Xiong et al . , 2009] ) . Missing residues that were not solved in 3fcs ( 746–774 and 840–873 in chain A , and 75–78 and 477–482 in chain B ) were modelled using Modeller ( Webb and Sali , 2016 ) with 3ije as a template . Parametric unpaired two-tailed t test was used to evaluate differences between groups . Statistical results are reported as p values < 0 . 05 , <0 . 01 , <0 . 005 or <0 . 001 . | Many proteins embedded in a cell’s surface allow the cell to interact with its surroundings . Integrins are a group of cell surface proteins that have many uses in different cells . Integrins become activated when they come into contact with other specific proteins , which like other molecules that bind to proteins are referred to collectively as “ligands” . Much research has focused on how ligands become attached to integrins and how this activates these cell surface proteins . Yet how integrins release ligands and become inactive has not been studied before . One type of integrin , called αIIbβ3 , is involved in blood clotting . Found on the surface of blood platelets – the fragments of cells in the blood that play a central role in clotting , this integrin binds to a ligand called fibrinogen . Fibrinogen links platelets together to form clots by building bridges between integrins . Passam , Chiu et al . have now studied platelets from donated human blood to understand how the integrin αIIbβ3 disengages from fibrinogen . The investigation showed that an enzyme called ERp5 aids the release of fibrinogen from the integrin . ERp5 can be released by blood vessel walls and by activated platelets . The experiments revealed that ERp5 breaks a chemical link , called a disulfide bond , in the integrin , but only when the protein is already bound to its ligand . Breaking the disulfide bond ( a chemical process known as reduction ) changes the integrin’s structure so that it lets go of fibrinogen . Moreover , when physical forces such as blood flow put the integrin under strain , the ERp5 enzyme becomes more effective . These findings show how ligand binding and mechanical force work together to control the breaking of a chemical bond in a human integrin . This chemical event then in turn controls the release of the integrin's ligand . It is possible that other protein-protein interactions may involve similar mechanisms , but this remains to be explored . Lastly , Passam , Chiu et al . suggest that the release of fibrinogen might help to limit the growth of blood clots so they do not block the blood vessels . Further studies should test this hypothesis . Inappropriate clotting can have severe health effects including heart attacks and strokes . As such , this investigation may hint at a more subtle way to regulate clotting through integrin αIIbβ3 , such as boosting fibrinogen release to see if it helps slow or reduce clotting without stopping it altogether . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2018 | Mechano-redox control of integrin de-adhesion |
Hsp70 participates in a broad spectrum of protein folding processes extending from nascent chain folding to protein disaggregation . This versatility in function is achieved through a diverse family of J-protein cochaperones that select substrates for Hsp70 . Substrate selection is further tuned by transient complexation between different classes of J-proteins , which expands the range of protein aggregates targeted by metazoan Hsp70 for disaggregation . We assessed the prevalence and evolutionary conservation of J-protein complexation and cooperation in disaggregation . We find the emergence of a eukaryote-specific signature for interclass complexation of canonical J-proteins . Consistently , complexes exist in yeast and human cells , but not in bacteria , and correlate with cooperative action in disaggregation in vitro . Signature alterations exclude some J-proteins from networking , which ensures correct J-protein pairing , functional network integrity and J-protein specialization . This fundamental change in J-protein biology during the prokaryote-to-eukaryote transition allows for increased fine-tuning and broadening of Hsp70 function in eukaryotes .
The Hsp70 chaperones are involved in a remarkably broad range of protein folding processes ( Finka et al . , 2015; Mayer and Bukau , 2005 ) , which renders them unique among the cellular chaperone systems . This functional versatility is achieved through the activity of an array of cochaperones that regulates the ATP-dependent substrate binding and release cycle of Hsp70 partner chaperones . The members of the J-protein family target Hsp70 to substrates , thereby starting the functional chaperone cycle ( Kampinga and Craig , 2010 ) . The essential role of J-proteins in diversifying Hsp70 targets and functions is reflected in the expansion of the number of J-protein family members with increasing organismal complexity ( Kampinga and Craig , 2010; Nillegoda and Bukau , 2015 ) . Nucleotide exchange factors ( NEFs ) reset Hsp70 for the next cycle of substrate binding by stimulating the exchange of ADP with ATP . Additionally , the different types of NEFs may co-determine Hsp70 function by regulating substrate release and communication with downstream protein quality control pathways ( Bracher and Verghese , 2015 ) . Traditionally , members of the J-protein family , which is subdivided into classes A , B and C , were viewed as functioning independently , interacting with Hsp70 chaperones in a one-to-one stoichiometry and leading to distinct outcomes in biological processes ( Kampinga and Craig , 2010 ) . This view has changed with the discovery of complex formation between canonical members of class A and class B J-proteins through transient interactions in metazoa ( Nillegoda et al . , 2015 ) . The domain architecture of the canonical J-proteins consists of the characteristic N-terminal J-domain ( JD ) linked to the substrate binding C-terminal domain ( CTD in class B; Zinc finger-like region ( ZFLR ) +CTD in class A ) via a glycine/phenylalanine ( G/F ) -rich flexible region . These canonical members dimerize through the C-terminally located dimerization domain ( DD; Figure 1A ) ( Kampinga and Craig , 2010 ) . The JD is a helical bundle consisting of four α-helices ( I , II , III and IV ) and a loop region containing the highly conserved tripeptide His-Pro-Asp ( HPD ) motif ( Figure 1C ) . Residues located in α-helices II , III and the HPD motif are implicated in the communication with Hsp70 leading to ATPase stimulation ( Genevaux et al . , 2002 ) . Although structurally related , canonical class A and class B J-proteins show distinct substrate-binding preferences at the CTDs ( Fan et al . , 2004; Reidy et al . , 2014 ) . The CTDs also provide interaction sites for the JD of opposite class members during J-protein complex formation ( Nillegoda et al . , 2015 ) . The JD-CTD contact sites display complementary class-specific electrostatic potentials . A negatively charged patch localized mostly on α-helices I and IV of class A and B JDs forms salt bridges with positively charged regions on the CTDs of the opposite class J-proteins ( Nillegoda et al . , 2015 ) ( Figure 1C , D ) . The resulting J-protein complexes have a wider substrate spectrum compared to the individual J-proteins or possible homo J-protein oligomers , because of the amassing of distinct substrate binding modules . This networking strategy is employed by the metazoan Hsp70-based aggregate-solubilizing systems ( disaggregases ) to broaden the substrate specificity spectrum and to increase machine efficacy ( Nillegoda et al . , 2015 ) . The gain in protein disaggregation power through interclass J-protein networking gives the human Hsp70-based disaggregase a level of potency comparable to that of the extremely efficient non-metazoan Hsp100-Hsp70 bichaperone disaggregase systems in bacteria , fungi and plants ( Nillegoda and Bukau , 2015 ) . The Hsp100 ( ClpB , Hsp104 ) component of this bichaperone system , however , disappeared during the evolution of multi-cellular organisms . The discovery of a potent metazoan Hsp70-based disaggregase activity driven by J-protein networking is therefore the missing link in our understanding of efficient amorphous aggregate solubilization in complex organisms . The evolutionary origin of J-protein networking via transient complex formation , however , is unknown . It is also unclear what factors determine and delimit the exact J-protein pairing , particularly within large J-protein families in higher eukaryotes such as humans . In this study , we investigate the molecular basis for J-protein networking and its evolution . By comparison of structural features of JD and CTD domains of canonical J-proteins across kingdoms of life , we observe a high degree of conservation in electrostatic potentials at the proposed JD and CTD contact faces within each of the classes ( A and B ) of the eukaryotic J-proteins . Further , phylogenetic and coevolutionary analysis of canonical class A and B J-proteins highlights a distinct phylogenetic signature between prokaryotes and eukaryotes , compatible with interclass J-protein complex formation . Using cell biology and biochemical approaches , we find a switching in J-protein biology at the prokaryote-to-eukaryote transition where class members begin to form functional networks , allowing for the emergence of powerful , yet regulatable eukaryotic disaggregase systems . We , furthermore , decipher the networking code between the cytosolic J-proteins in human cells and describe a naturally occurring strategy to correctly pair interacting J-proteins . This code , based on electrostatic potential distribution patterns at the JD and CTD interaction faces , possibly ensures functional integrity within J-protein networks that expanded during the rise of complex life .
Although we have previously shown that canonical metazoan J-proteins form interclass complexes with electrostatically complementary interactions between JDs and CTDs , it remains unclear whether these interclass J-protein complexes also occur in non-metazoans , especially given the fact that orthologs of both classes exist in bacteria , protists , fungi , plants and protozoa ( Figure 1B ) ( Nillegoda and Bukau , 2015 ) . We hypothesized that if interclass J-protein cooperation occurs in these organisms , then structural elements promoting complex formation ( Nillegoda et al . , 2015 ) must be preserved among canonical class A and class B J-proteins of bacterial and fungal origins . To investigate our conjecture , we constructed models of the three-dimensional structures of JDs and CTDs of canonical class A and B J-proteins from human to bacteria and investigated the degree of conservation of complementary class-specific electrostatic potentials at intermolecular JD-CTD interfaces . The prokaryotic sample included class A J-protein DnaJ and class B J-protein CbpA from a wide range of bacteria ( Figure 1C , D ) . The eukaryotic sample consisted of non-metazoan ( plants: ATJ3 ( A ) and At5g25530 ( B ) ; yeast: Ydj1 ( A ) and Sis1 ( B ) ) and metazoan ( nematode: DNJ-12 ( A ) and DNJ-13 ( B ) ; human: DNAJA2 ( A ) and DNAJB1 ( B ) ) J-proteins belonging to classes A and B . The structures of the protein domains were determined experimentally or modeled by comparative modeling ( see Materials and methods ) . Analysis of JDs showed a general conservation of protein structure and electrostatic potentials within each of the J-protein classes throughout evolution . Both classes A and B JDs displayed a bipolar charge distribution ( which is more prominent among class B JDs ) , where the positive patch around α-helix II implicated in Hsp70 binding was the most prominent feature of the electrostatic potential ( Figure 1C , D ) . Among the CTDs , however , clear class-dependent differences were observed between prokaryotic and eukaryotic representatives ( Figure 1C , D ) . Qualitatively , the eukaryotic class BCTDs were dominantly positively charged , whereas in prokaryotic structures , we observed a mixture of positively and negatively charged patches ( Figure 1D ) . In eukaryotic class ACTDs , the ZFLR+CTD-I region is peppered with exposed positively and negatively charged patches , whereas CTD-II was predominantly negatively charged . In contrast , there was a switch of these electrostatic potential patterns in prokaryotic class A J-proteins: the ZFLR+CTD-I regions were dominantly negative , while the CTD-II regions showed clusters of both positive and negative patches ( Figure 1C ) . To quantitatively assess the differences in electrostatic potential among prokaryotic and eukaryotic CTDs , we performed Protein Interaction Property Similarity Analysis ( PIPSA ) ( Wade et al . , 2001 ) around the JD interaction interface located at the CTD hinge regions of DNAJA2 and DNAJB1 ( Nillegoda et al . , 2015 ) ( black dotted circles , Figure 1C , D; also see Materials and methods ) . The 25 Å radius spheres encompass residues previously implicated in opposite class JD interaction from crosslinking and Förster resonance energy transfer ( FRET ) experiments and JD docking simulations ( Nillegoda et al . , 2015 ) . Based on the electrostatic similarities around the hinge regions , the PIPSA analysis showed clustering of the J-proteins into two groups separating the prokaryotes from eukaryotes ( Figure 1E , F ) . The J-proteins ATJ3 and At5g25530 from Arabidopsis thaliana showed electrostatic potential patterns that were more eukaryotic-like . The clustered groups of CTDs of both classes A and B J-proteins of yeast , nematode and human reflected highly conserved charge distributions at the JD interaction interface ( Figure 1E , F ) . The same regions in prokaryotic CTDs showed distinct clustering for both classes but indicated a different charge distribution from the eukaryotic CTDs ( Figure 1E , F ) . We conclude that the electrostatically complementary opposite class JD interaction interface is highly conserved among human , worm and yeast J-proteins but not in bacterial counterparts . We extended the set of bacteria analyzed by modeling additional DnaJ and CbpA CTD structures from Gram-negative proteobacteria ( alpha , beta and gamma ) and the Gram-positive firmicute Clostridium ultunense ( Figure 1—figure supplement 1 and Figure 1—figure supplement 2 ) and made similar observations . However , we observed eukaryotic-like features ( e . g . general increase in exposed positive charges at the JD interaction region ) emerging in class BCTDs of some bacteria such as C . ultunense , Acetobacter aceti and Sphingomonas sp ( Figure 1—figure supplement 2A ) , but not in the partnering class A CTD ( Figure 1—figure supplement 1A and Figure 1—figure supplement 3 ) . Taken together , these features suggest J-protein networking via interclass complex formation may occur in both animals and simpler eukaryotic unicellular organisms , such as yeast , but not in bacteria . 10 . 7554/eLife . 24560 . 003Figure 1 . Conservation of class-specific electrostatic potential distributions predict interclass J-protein complex formation in eukaryotes . ( A ) Domain architecture of class A and class B J-proteins ( shown as protomers ) . Class A J-proteins have an N-terminal J-domain ( JD ) , a glycine/phenylalanine-rich flexible region ( G/F ) , C-terminal β-sandwich domains ( CTD-I and II ) and a CTD-I inserted zinc-finger-like region ( ZFLR ) . The Hsp70-interacting HPD motif is indicated in red . Protomer dimerization to form homodimers occurs at the dimerization domain ( DD ) . The ZFLR in CTD-I is absent in the domain architecture of class B J-proteins . ( B ) Evolutionary tree ranking the kingdoms of organisms analyzed in this study . ( C , D ) Electrostatic isopotential contour maps ( cyan + 1 , red −1 kcal/mol/e ) of CTD homodimers and of J-domains of class A ( green cartoon diagrams ) ( C ) and class B ( blue cartoon diagrams ) ( D ) J-proteins . Roman numerals show the four α-helices on class A JD . J-proteins from the following organisms are compared: Homo sapiens ( DNAJA2 , DNAJB1 ) , Caenorhabditis elegans ( DNJ-12 , DNJ-13 ) , Saccharomyces cerevisiae ( Ydj1 , Sis1 ) , Arabidopsis thaliana ( ATJ3 , At5g25530 ) , Pseudomonas oryzihabitans , Bordetella pertussis and Escherichia coli ( DnaJ , CbpA ) . The dashed circles on the CTDs of DNAJA2 and DNAJB1 represent the spherical region used for local PIPSA analysis of electrostatic potential similarity . ( E , F ) Local PIPSA analysis results for class A CTD ( E ) and class B CTD ( F ) electrostatic potentials . The electrostatic potentials in the spherical regions ( radius of 25 Å ) indicated by the dashed black circles in ( C ) and ( D ) were clustered by similarity using Ward’s clustering . The heat maps show clustering of J-proteins by similarity ( higher similarity indicated by a red shift ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 00310 . 7554/eLife . 24560 . 004Figure 1—figure supplement 1 . Electrostatic isopotential contour maps of class A J-proteins from humans , fungi , nematodes and bacteria . ( A ) Class A CTD dimers ( cyan +1 , red −1 kcal/mol/e ) . The protein structure is depicted in green cartoon representation . The J-protein name and corresponding Uniprot ID are given in parentheses for each organism . The human class A J-proteins are represented by DNAJA1 ( P31689 ) and DNAJA2 ( O60884 ) . Saccharomyces cerevisiae ( Ydj1 , P25491 ) and Caenorhabditis elegans ( DNJ-12 , O45502 ) represent fungi and nematodes , respectively . Bacterial DnaJ are represented from the following subgroups: alphaproteobacteria ( A0A063 × 4A7 , Q1NCH5 ) , betaproteobacterium ( Q7VVY3 ) , gammaproteobacteria ( P08622 , P0A1G8 , C4T9C4 , A0A0D7F716 ) and firmicute ( M1ZGL1 ) . ( B ) As in ( A ) Class A JDs . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 00410 . 7554/eLife . 24560 . 005Figure 1—figure supplement 2 . Electrostatic isopotential contour maps of class B J-proteins from humans , fungi , nematodes and bacteria . ( A ) Class B CTD dimers ( cyan + 1 , red −1 kcal/mol/e ) . CTD dimer structure depicted in blue cartoon representation . The J-protein name and corresponding Uniprot ID are given in parentheses for each organism . Human class B J-proteins are represented by DNAJB1 ( P25685 ) and DNAJB4 ( Q9UDY4 ) . Saccharomyces cerevisiae ( Sis1 , P25294 ) and Caenorhabditis elegans ( DNJ-13 , Q20774 ) represent fungi and nematodes , respectively . Bacterial CbpA is represented by the following subgroups: alphaproteobacteria ( A0A063XA16 , Q1NEX3 ) , betaproteobacterium ( J7RE62 ) , gammaproteobacteria ( P36659 , P63262 , Q9BQH2 , A0A0D7FE35 ) and firmicute ( M1ZLZ3 ) . ( B ) As in ( A ) Class B JDs . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 00510 . 7554/eLife . 24560 . 006Figure 1—figure supplement 3 . Evaluation of JD interaction sites on CTDs of the opposite class J-proteins . ( A , B ) Local PIPSA analysis of electrostatic potentials at ( A ) class B JD interaction sites on CTDs of class A J-proteins and ( B ) class A JD interaction sites on CTDs of class B J-proteins . Eukaryotic sequences are colored in black and prokaryotic ones in red . The electrostatic potentials in the spherical region ( radius of 25 Å ) indicated by the dashed black circles ( Figure 1—figure supplement 1A and Figure 1—figure supplement 2A ) were clustered by electrostatic distance using Average ( a ) and Ward’s ( b ) clustering . The heat map shows clustering of J-proteins according to electrostatic distance ( high similarity indicated by a red shift ) . The color key and density plot is depicted on the top left . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 006 To obtain an in-depth understanding of the evolution of the JD-CTD intermolecular interactions between canonical class A and B J-proteins , we generated class-specific phylogenetic trees and separately analyzed the JD and the CTD regions . The class A and class B trees were built from 12 , 215 and 4194 sequences , respectively , encompassing all kingdoms of life . As expected , we found the trees to carry coherent phylogenetic signals: eukaryotes ( highlighted in gray background ) were in general set apart from prokaryotes , and sub-trees were mostly consistent with sub-classifications ( Figure 2A , B and Figure 2—figure supplement 1A , B ) . Furthermore , eukaryotic class A J-protein sequences of organellar origin were found to mix with sub-trees of prokaryotic regions of the trees ( separated by pink lines; Figure 2B and Figure 2—figure supplement 1A ) , consistent with their probable bacterial origins ( Lu et al . , 2006; Deloche et al . , 1997 ) . 10 . 7554/eLife . 24560 . 007Figure 2 . Phylogenetic and coevolutionary analyses of the JD-CTD interaction between class A and B J-proteins . ( A ) Phylogenetic tree of class B J-domains . Color-coding separates different phylogenetic groups . Grey area highlights the separation of eukaryotes ( fungi , viridiplantae and other eukaryotes ) from prokaryotes ( proteobacteria , firmicutes and other bacteria ) and Archaea . ( B ) As in ( A ) , phylogenetic tree of class A CTDs . Pink lines delimit organellar sequences of eukaryotic organisms . ( C ) Structural view of most discriminating positions predicted by PDA ( red ) plotted on JD of DNAJB1 ( blue , five positions ) and CTD of DNAJA2 ( green , nine positions ) . DCA-derived coevolving residue pairs depicted on DNAJB1JD and DNAJA2CTD ( orange ) . Experimentally determined cross-linking residues between DNAJB1 and the DNAJA2 are indicated in purple ( Nillegoda et al . , 2015 ) . Location of the triple charge reversion ( E/D→R ) mutations that disrupts interclass J-protein complex formation between DNAJB1 and DNAJA2 denoted by ( * ) ( Nillegoda et al . , 2015 ) . The HPD motif of DNAJB1JD is shown in grey . Roman numerals show the four α-helices on JD . ( D ) Mapping of sequence clustering derived from PDA ( see Materials and methods ) using the most discriminating positions on to JD and CTD trees of class B and class A J-proteins , respectively . The two identified groups ( green and yellow nodes ) covered 81% in the case of the clustering done on the JDs and 100% when clustering was done on the CTDs . Unclassified sequences are depicted in white . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 00710 . 7554/eLife . 24560 . 008Figure 2—figure supplement 1 . Phylogenetic and coevolutionary analyses between class AJDs and class BCTDs . ( A ) Phylogenetic tree built from the J-domains of class A J-proteins . Color-coding separates different phylogenetic groups . Grey area highlights the separation of eukaryotes ( fungi , viridiplantae and other eukaryotes ) from prokaryotes ( proteobacteria , firmicutes and other bacteria ) and Archaea . ( B ) As in ( A ) , built from the CTDs of class B J-proteins . Pink lines delimit organellar sequences of eukaryotic organisms . ( C ) Structural view of PDA predicted most discriminating positions ( red ) plotted on the JD of DNAJA2 ( green , six positions ) and the CTD of DNAJB1 ( blue , nine positions ) . DCA derived coevolving residue pairs depicted on DNAJB1JD and DNAJA2CTD ( orange ) . Experimentally determined cross-linking residues between the DNAJA2JD and the DNAJB1CTD are indicated in purple ( Nillegoda et al . , 2015 ) . Location of the triple charge reversion ( E/D→R ) mutations that disrupts interclass J-protein complex formation between DNAJB1 and DNAJA2 denoted by ( * ) ( Nillegoda et al . , 2015 ) . The HPD motif of DNAJB1JD is shown in grey . ( D ) Mapping of sequence clustering derived from PDA ( see Materials and methods ) using the most discriminating positions on to JD and CTD trees of class A and class B J-proteins , respectively . The two identified groups ( green and yellow nodes ) covered 78% in the case of the clustering done on the JDs and 100% when clustering was done on the CTDs . Unclassified sequences are depicted in white . ( E ) Graph showing the distribution of predicted DCA pairs between class BJDs and class ACTDs . The threshold for obtaining statistically significant inter-protein coevolving pairs was set at 5% of contact appearance after 300 realizations . DCA-derived coevolving residue pair on DNAJA2JD and DNAJB1CTD ( orange ) . Neighboring charged residues are depicted in black . ( F ) As in ( E ) graph showing the distribution of predicted DCA pairs between class AJDs and class BCTDs . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 00810 . 7554/eLife . 24560 . 009Figure 2—figure supplement 2 . Evaluation of robustness in Phylogenetic Discriminant Analysis . In all right sub-panels , the numbering corresponds to the indexing of the MSAs and not of individual sequences . The red vertical red lines indicate the mixing score of the reference triplet ( D6 , E61 , E64 for DNAJA2; D4 , E69 , E70 for DNAJB1 ) for the J-domains and the 5th percentile for the CTDs , which lack a reference triplet . In all right panels , the dashed red lines mark the mean selection rate of the null model , while the magenta ( resp . green ) dashed lines mark 3 ( resp . 10 ) standard deviations of the mean selection rate ( See Materials and methods ) . ( A ) PDA derived p-value and top positions for class B JDs . Left: Distribution of the mixing scores H ( entropy ) indicative of discriminatory power , the lower H the greater the discriminatory power ( see Materials and methods ) . Right: Histogram of the percentage of appearances of positions in triplets having a lower mixing score H than the reference triplet ( positions indicated in red ) . ( B ) as in ( A ) PDA-derived p-value and top positions for class A CTDs . The vertical red line marks the mixing score limiting the 5th percentile . ( C ) PDA derived p-value and top positions for class B JDs . ( D ) PDA derived p-value and top positions for class B CTDs . ( E ) Robustness analysis of PDA of class B JDs . Included groups: Other prokaryotes , other eukaryotes , fungi , and proteobacteria . ( F ) Robustness analysis of PDA of class B JDs . Included groups: Other prokaryotes , other eukaryotes , fungi , proteobacteria and firmicutes . ( G ) Robustness analysis of PDA of class B JDs . Included groups: Other prokaryotes , other eukaryotes , fungi , proteobacteria , firmicutes and viridiplants . ( H ) Robustness analysis of PDA of class B JDs using Modularity Clustering method . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 009 We next scanned for eukaryote-specific phylogenetic signatures that supported interclass J-protein complex formation . The scanning was performed with a new approach named Phylogenetic Discriminant Analysis ( PDA ) that determines the residues that best separate eukaryotes from prokaryotes ( see Materials and methods ) . The PDA analysis applied to JDs of class B ( class BJDs ) identified a total of five positions ( Figure 2C: residues in red on DNAJB1JD in blue; see also Methods and Figure 2—figure supplement 2A ) that showed a strong phylogenetic separation between prokaryotes and eukaryotes ( eukaryotes highlighted in grey; Figure 2D , left ) . Importantly , of the identified discriminatory hits , two residues mapped onto the E69 and E70 positions on JD of DNAJB1 ( Figure 2C ) . These two residues are part of the negatively charged amino acid triplet ( D4 , E69 and E70 denoted by * ) that was previously experimentally identified as having a strong influence in complex formation between the DNAJB1JD and the DNAJA2CTD in humans ( Nillegoda et al . , 2015 ) . The third residue identified ( I63 ) also localized to the same region on α-helix IV of DNAJB1JD ( Figure 2C ) . The other residue positions ( L29 and K35 ) flanked the HPD motif ( His-Pro-Asp , grey; Figure 2C ) , an essential region for Hsp70 binding ( Suh et al . , 1998; Jiang et al . , 2007 ) . PDA on class ACTDs highlighted nine positions that strongly discriminated phylogeny between prokaryotes and eukaryotes ( Figure 2C , Figure 2D , right and Figure 2—figure supplement 2B ) . Here , the hits occured in three regions ( mapped onto DNAJA2CTD in green cartoon , Figure 2C ) . Of the hits in the CTD , residues Y128 and D222 locate to the CTD-I-CTD-II hinge region . Residue D222 showed the strongest phylogenetic discriminatory value ( Figure 2—figure supplement 2B ) . The neighboring residue of D222 ( purple; K226 , Figure 2C ) was previously shown to cross-link with the JD of DNAJB1 ( Nillegoda et al . , 2015 ) , placing D222 near the interaction site for class B JD . The hits in the third region were mapped onto the dimerization domain ( DD ) of class A J-proteins . The phylogenetic discriminatory signal at the DD is currently not understood . To complement the phylogenetic study , we next performed Direct Coupling Analysis ( DCA ) ( Morcos et al . , 2011 ) to capture coevolving protein contacts in class BJDs and class ACTDs ( see Materials and methods ) . We observed a statistically significant coevolving residue pair corresponding to V221 on DNAJA2CTD and E62 on DNAJB1JD ( orange; Figure 2C and Figure 2—figure supplement 1E ) . These residues also map onto the vicinity of the respective JD and CTD interacting regions of DNAJB1 and DNAJA2 , further confirming our experimental , structural and PDA-derived findings . Although it is surprising to observe a coevolution between a valine and a glutamic acid , V221 is flanked by a strongly charged region formed by H220 , D222 and K223 . Further , we also observed in the sequence alignment that the position of V221 is generally surrounded by several charged residues . Thus , V221 may coordinate the local environment of this charged region to interact with charged residues proximal to E62 on DNAJB1JD . The overlap of PDA hits with the DCA and experimentally implicated residues in interclass J-protein complex formation implies the presence of a eukaryote-specific phylogenic signature for the interaction between class BJDs at the CTD hinge region of class ACTDs . A reciprocal phylogenetic analysis was next performed with class AJDs and class BCTDs , which provided similar observations ( Figure 2—figure supplement 1 and Figure 2—figure supplement 2C , D ) . The positions that best capture prokaryotic to eukaryotic phylogenetic splitting on class BCTDs showed two clusters localized to either CTD or DD regions ( Figure 2—figure supplement 1C , D ) . Importantly , the CTD cluster containing residues I175 and K209 were located at the hinge region ( mapped on the DNAJB1CTD structure ) . K209 residue was previously observed to cross-link with the JD of DNAJA2 ( Nillegoda et al . , 2015 ) . Class A JDs , however , provided less clear-cut results . Nevertheless , DCA captured a coevolving residue at position R63 , which is flanked by E61 and E64 ( Figure 2—figure supplement 1C , F ) . These residues form the class B CTD interaction interface on the JD of DNAJA2 ( Nillegoda et al . , 2015 ) . The coevolving G278 residue localized to the JD interaction interface in the DNAJB1CTD hinge region . In summary , we find phylogenetic and coevolutionary signatures compatible with bi-directional JD-CTD interactions between class A and class B J-proteins of eukaryotic origins . These findings support the evolution of interclass J-protein networks beyond metazoa but not in prokaryota . Our data show the emergence of two protein-protein interaction regions in eukaryotic JDs: One for Hsp70 and the second for interclass J-protein complex formation . We also identified a similar region for partner protein interaction at the CTDs of canonical J-proteins . When combined , PDA and DCA analyses identify the hinge region between CTD-I and II subdomains as the primary interface for JD binding during interclass J-protein complex formation , which agrees with our previous crosslinking , FRET and docking simulation results obtained with human J-proteins ( Nillegoda et al . , 2015 ) . The recently reported interaction between ubiquitin ligase Rsp5 and Ydj1 for targeting aberrant proteins during heat stress also highlights the importance of this hinge region for partner protein binding ( Fang et al . , 2014 ) . This protein-protein interaction interface shows no significant overlap with known substrate-binding regions of canonical J-proteins . Consistently , binding of JD fragments to CTDs does not block substrate association with J-proteins ( Nillegoda et al . , 2015 ) . To validate our structural , phylogenetic and coevolutionary analyses , we employed an in situ antibody-based proximity ligation assay ( PLA ) ( Söderberg et al . , 2006 ) to visualize interclass J-protein complex formation in cells . J-protein complexes containing human class A DNAJA2 and class B DNAJB1 were detected as red puncta after signal amplification ( each punctum representing a complex formation event ) in cultured human HeLa cells ( Figure 3A ) . This reconfirms our original biochemical findings for interclass J-protein complex formation in humans ( Nillegoda et al . , 2015 ) . Addition of either one of the J-protein specific antibodies alone did not generate red puncta ( Figure 3B , C ) . To further validate the specificity of the interactions , we prevented J-protein pairing by separately depleting each member using RNAi knockdowns . As expected , we observed a drastic decrease in the number of complexes per cell in the individual J-protein depletions ( Figure 3D–F ) . The J-protein knockdowns and antibody specificities were confirmed by Western blotting ( Figure 3—figure supplement 1A ) . Next , we performed the same assay in S . cerevisiae cells for the two major yeast cytosolic J-proteins , class A Ydj1 and class B Sis1 . We observed a strong punctated red fluorescence signal in wild type cells , confirming the predicted complex formation between the two yeast J-proteins ( Figure 3G ) . Control cells carrying a deletion of the ydj1 gene showed no red puncta ( Figure 3J and Figure 3—figure supplement 2A ) . Similarly , depletion of the essential Sis1 ( expressed from TetO7 repressible promoter ) using doxycycline ( +dox ) also showed a dramatic decrease in complex formation in cells ( Figure 3K , L and Figure 3—figure supplement 2B ) . Non-specific signal amplification was not observed in controls lacking either one of the J-protein antibodies ( Figure 3H , I and Figure 3—figure supplement 2C ) . 10 . 7554/eLife . 24560 . 010Figure 3 . Interclass J-protein complexes occur in human and yeast cells , but not in bacteria . ( A–F ) Detection of interclass J-protein complexes in human cells . ( A ) DNAJA2 ( class A ) and DNAJB1 ( class B ) form interclass complexes in HeLa cells . Red punctae reflect fluorescent signal amplification from single DNAJA2+DNAJB1 interaction events using an in situ proximity ligation assay . Fluorescently labeled oligonucleotides ( red ) hybridize to amplified the DNA proximity signal ( see cartoon on left ) . Nuclei stained with DAPI ( cyan ) . ( B , C ) Technical controls of PLA for antibody specificity . ( B ) PLA performed with anti-DNAJA2 antibody only . ( C ) PLA performed with anti-DNAJB1 antibody only . ( D–F ) RNAi knockdown of either DNAJA2 ( E ) or DNAJB1 ( F ) disrupts interclass J-protein complex formation in HeLa cells . Control reaction with scrambled RNAi shown in ( D ) . ( G–L ) Interclass J-protein complex formation in yeast ( Saccharomyces cerevisiae ) cells . ( G ) Appearance of red punctae denotes complex formation between Ydj1 ( class A ) and Sis1 ( class B ) . Nuclei stained with DAPI ( cyan ) . ( H ) PLA performed with anti-Ydj1 antibody only . ( I ) PLA performed with anti-Sis1 antibody only . ( J ) Ydj1-Sis1 complex formation is abrogated in cells with Ydj1 gene knocked out . ( K , L ) PLA against Ydj1 and Sis1 in S . cerevisiae cells depleted of Sis1 . Sis1 expression is switched off ( tet-off ) in the absence ( K ) or presence ( L ) of doxycycline ( see Methods ) . ( M ) No complexing events occur between DnaJ-YFP ( class A ) and CbpA-mCherry ( class B ) in Escherichia coli log phase cells after signal amplification . Primary antibodies directed toward YFP and mCherry tags . Bacterial DNA stained with DAPI ( cyan ) . ( N ) Complex formation detected between DnaJ-YFP ( class A ) and DnaK in log phase E . coli cells . Primary antibodies target YFP tag and DnaK . Scale bar = 10 μm . n ( biological repeats ) = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 01010 . 7554/eLife . 24560 . 011Figure 3—figure supplement 1 . J-protein levels after knockdown or overexpression in HeLa cells . ( A ) Western blot of DNAJA2 and DNAJB1 levels remaining after specific RNAi knockdowns . Control treatment performed with non-specific scrambled RNAi . GAPDH levels used as loading control . ( B ) Protein levels of V5-DNAJB1 , V5-DNAJB2 and V5-DNAJB8 after overexpression in HeLa cells . GAPDH , loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 01110 . 7554/eLife . 24560 . 012Figure 3—figure supplement 2 . Yeast and bacterial class A and class B J-protein levels analyzed by western blotting . ( A ) Immunoblotting for Ydj1 and Sis1 in wild type ( WT ) and ∆ydj1 Saccharomyces cerevisiae strains . Sis1 levels are measured as reference J-protein control . Pgk1 , loading control . ( B ) Doxycycline induced knockdown of Sis1 ( see Materials and methods ) . Ydj1 levels are measured as reference J-protein control . Pgk1 , loading control . ( C ) Specificity determination of Ydj1 and Sis1 antibodies used in PLA . ( D ) Immunoblotting for YFP and mCherry using lysates from log and stationary phase Escherichia coli cells expressing CbpA-mCherry and DnaJ-YFP . CbpA-mCherry is expressed under the endogenous promoter . DnaJ-YFP levels shown after 2 hr of IPTG ( 25 µM ) induction . DnaJ-YFP is expressed at moderate levels even in the absence of IPTG due to promoter leakage . S2 protein , loading control . ( E ) PLA for DnaJ-YFP and CbpA-mCherry in stationary phase E . coli cells . Primary antibodies target YFP and mCherry tags of DnaJ and CbpA . Bacterial DNA stained with DAPI ( cyan ) . ( F ) As in ( E ) , PLA for CbpA-mCherry and DnaK in stationary phase E . coli cells . Primary antibodies target mCherry tag and DnaK . Scale bar = 10 μm . n ( biological repeats ) = 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 012 To discern complex formation between DnaJ and CbpA , the only class A and B J-protein members in E . coli , we also reconstituted the proximity assay in bacterial cells . In contrast to the interclass J-protein interactions observed in eukaryotic cells , we did not observe complexes between DnaJ and CbpA after signal amplification in log phase E . coli cells ( Figure 3M ) where both J-proteins are expressed ( Figure 3—figure supplement 2D ) ( Tatsuta et al . , 1998 ) . Cellular CbpA levels reach maximum levels during stationary phase in response to nutrition starvation ( Yamashino et al . , 1994 ) , but no interclass DnaJ-CbpA complexes were observed in E . coli cells grown into stationary phase ( Figure 3—figure supplement 2D , E ) . In contrast , our positive controls captured JD-driven interactions between DnaJ and DnaK ( bacterial Hsp70 ) and CbpA and DnaK ( Figure 3N and Figure 3—figure supplement 2F ) . C-terminal YFP and mCherry tagging of DnaJ and CbpA does not compromise in vivo functions , JD-mediated interactions or localization ( Winkler et al . , 2010; Chintakayala et al . , 2015 ) . These findings corroborate the structural data that showed absence of complementary JD interaction interfaces on prokaryotic CTDs ( Figure 1C , D ) . In essence , we now provide direct evidence for interclass J-protein complex formation in both metazoan and non-metazoan eukaryotic cells , which seems to be absent in bacteria . Previously , we showed that the Hsp70-based protein disaggregases in nematodes and humans rely on complex formation between class A and B J-proteins to expand substrate recognition and potentiate solubilization of amorphous aggregates ( Nillegoda et al . , 2015 ) . As Ydj1 and Sis1 also form interclass J-protein complexes in S . cerevisiae cells , we assessed the impact of this phenomenon on the function of the yeast protein disaggregase system ( Figure 4 ) . As a control , we used the homologous disaggregase system from E . coli , where interclass communication between J-proteins is apparently absent . The aggregate solubilizing bichaperone machines in bacteria and yeast critically depend on the cooperation between the powerful Hsp100 AAA+ ATPases and the Hsp70 system ( Glover and Lindquist , 1998; Goloubinoff et al . , 1999; Kaimal et al . , 2017 ) . Hsp70 recruits and activates Hsp100 while the J-proteins provide the overall substrate selectivity for the disaggregase system ( Mogk et al . , 2015 ) . Despite considerable similarities in machine architecture and function ( Mogk et al . , 2015; Sousa , 2014 ) , the bacterial ClpB-DnaK bichaperone system is separated from the yeast counterpart ( Hsp104-Ssa1 ) by more than 2 billion years of evolution in which the prokaryote-to-eukaryote transition occurred ( Hedges et al . , 2004 ) . 10 . 7554/eLife . 24560 . 013Figure 4 . Interclass J-protein communication has distinct functional consequences for prokaryotic and eukaryotic protein disaggregation systems . ( A ) Scheme for in vitro disaggregation/refolding and refolding-only reactions . Solubilization of preformed heat-aggregated luciferase using the human HSPA8-HSPH2 ( Hsp70-Hsp110 ) system with DNAJA2 ( Class A , green ) , DNAJB1 ( Class B , blue ) or DNAJA2+DNAJB1 ( magenta ) . Control reaction containing aggregates without chaperone mix ( black ) . n = 3 . ( B ) As in ( A ) , reactivation of luciferase aggregates by the yeast Hsp104-Ssa1 ( Hsp100-Hsp70 ) bichaperone disaggregation system with J-proteins Ydj1 ( class A , green ) , Sis1 ( Class B , blue ) and Ydj1+Sis1 ( magenta ) . The NEF is Sse1 , which is homologous to HSPH2 . n = 3 . ( C ) Reactivation of aggregated luciferase by the bacterial ClpB-DnaK ( Hsp100-Hsp70 ) bichaperone system with GrpE ( NEF ) in the presence of DnaJ ( Class A , green ) , CbpA ( Class B , blue ) or DnaJ+CbpA ( magenta ) . ( D ) SEC profile of aggregated 3H-luciferase ( elution fractions F1-F3 , black ) . Soluble luciferase monomers ( ~63 kDa ) elute in fraction F4 ( red ) . MW scale on top in kDa . ( E , F ) Aggregate quantification for F1-F4 in SEC profile obtained with yeast ( E ) and bacterial ( F ) bichaperone systems after 120 min of disaggregation . Profiles show the disappearance of 3H-luciferase from aggregates ( F1-F3 ) with concomitant accumulation of the disaggregated monomers in F4 . Black bars indicate aggregate levels obtained from control reactions lacking chaperones . Values normalized to total counts in each reaction . n = 3 . Error bars reflect mean ± sem . See Materials and methods for protein concentrations used . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 01310 . 7554/eLife . 24560 . 014Figure 4—figure supplement 1 . Yeast class A and B J-proteins form interclass complexes to synergize disaggregation activity of the Hsp104-Ssa1 bichaperone system . ( A ) Reactivation of heat-aggregated luciferase using yeast Hsp104-Ssa1 ( Hsp100-Hsp70 ) bichaperone disaggregation system with J-proteins Ydj1 ( class A ) , Sis1 ( Class B ) and Ydj1+Sis1 under saturating chaperone levels . The NEF is Sse1 . n = 3 . ( B ) Refolding of heat inactivated monomeric luciferase ( see Materials and methods ) by the yeast Hsp70 chaperone system . Class A , class B and class A+B J-protein containing reactions are indicated by green , blue and magenta , respectively . ( C ) As in ( B ) , refolding of heat inactivated luciferase by the bacterial Hsp70 chaperone system . ( D ) Quantification of aggregates remaining in SEC profile after disaggregation/refolding of aggregated 3H-luciferase ( 40 min ) by the yeast Hsp104-Ssa1 bichaperone system with either a class A or a class B J-protein alone ( green ( class A ) or blue ( class B ) ) or class A+B J-proteins combined ( magenta ) . Control reaction without chaperones ( black ) . Elution fractions F1 to F4 ( red lines ) . F1 , void volume , size of the luciferase aggregates eluted ≥5000 kDa; F2 , aggregates ~ 700 to~4000 kDa; F3 , aggregates ~ 200 to~700 kDa , F4 disaggregated monomers ( ~63 kDa ) . Values normalized to total counts in each reaction . n = 3 . ( E ) As in ( D ) , aggregate quantification after treatment with yeast Hsp70 chaperone system ( 120 min disaggregation ) . n = 2 . ( F ) As in ( E ) , aggregate quantification after treatment with yeast Hsp70 chaperone system substituted with human HSPA8 ( 40 min disaggregation ) . n = 2 . ( G ) Quantification of aggregates remaining in SEC profile after disaggregation/refolding of aggregated tritiated luciferase ( 40 min ) by bacterial ClpB-DnaK bichaperone system . n = 3 . ( H ) Aggregate quantification after treatment with bacterial ClpB-DnaK bichaperone system ( 120 min disaggregation ) . A further threefold lower chaperones levels used from non-saturating conditions ( see Materials and methods ) . n = 3 . ( I ) Aggregate quantification after treatment with bacterial Hsp70 chaperone system ( 120 min disaggregation ) . n = 2 . Error bars reflect mean ± sem . See Materials and methods for protein concentrations used . ( J ) Brownian Dynamics docking results for Escherichia coli DnaJCTD with either CbpAJD or DNAJB1JD . Total number of clusters per simulation , denominator; number of selected clusters ( corresponding to 90% of all docked complexes ) , numerator . The range of average energy values for the selected clusters ( in units of kT ) is shown in parentheses . The average distance of the center of geometry ( COG ) of the cluster representatives to the COG of the cluster representatives of the previously docked DNAJB1JD to the DNAJA2CTD is given ( see Materials and methods ) . The selected clusters indicate a defined binding mode and large negative energy values indicate favorable binding . The docked site for the E . coli DnaJ CTD differs from that for human DNAJA2 . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 01410 . 7554/eLife . 24560 . 015Figure 4—figure supplement 2 . Cross species communication between class A and B J-proteins in protein refolding and protein disaggregation . ( A ) Refolding of heat inactivated monomeric luciferase by human , yeast and bacterial class A and B J-proteins in combination with human HSPA8-HSPH2 chaperone system . Class A , class B and class A+B J-protein containing reactions are indicated by green , blue and magenta , respectively . J-proteins in the combinations are indicated below the graph . n = 2 . ( B ) As in ( A ) , disaggregation/refolding of heat aggregated luciferase by human , yeast and bacterial class A and B J-proteins in combination with HSPA8-HSPH2 chaperone system . n = 2 . ( C ) Cross species cooperation of yeast and human J-proteins via interclass J-protein complex formation to synergize protein disaggregation . Dotted and solid lines plot additive and synergistic affects , respectively . Values for the additive curves were obtained by tallying the measurements obtained for single J-protein reactions at the concentrations used in the A+B combination experiments . n = 2 . ( D ) Competition of unlabeled bacterial , yeast and human J-proteins for the JD-CTD interaction formed between labeled DNAJB1 and DNAJA2 analyzed by FRET . Bars represent donor quenching efficiency of JD and CTD intermolecular interactions . Cartoons above graph show fluorophore positions mapped onto DNAJB1 and DNAJA2 protomers . Red dotted lines indicate intermolecular JD-CTD bidirectional interactions . n = 3 . Two-tailed t-test **p<0 . 01 , ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 015 In vitro disaggregation/refolding reactions containing the yeast bichaperone system with either Ydj1 ( A , green ) or Sis1 ( B , blue ) showed considerable reactivation of preformed aggregates of model substrate firefly luciferase at high chaperone to substrate ratios ( Figure 4—figure supplement 1A ) consistent with previous reports ( Seyffer et al . , 2012; Glover and Lindquist , 1998 ) . Of note , the amorphous luciferase aggregates used in these assays were generated in the presence of a small heat shock protein ( Hsp26 ) to mimic in vivo protein aggregation conditions ( Cashikar et al . , 2005; Haslbeck et al . , 2005 ) and allow increased substrate accessibility for disaggregation machineries ( Rampelt et al . , 2012; Nillegoda et al . , 2015 ) . Compared to single J-protein-containing reactions ( class A , green; class B , blue ) , the reactions consisting of both Ydj1 and Sis1 ( magenta ) showed synergistic increase in luciferase reactivation , especially under non-saturating conditions ( Figure 4B ( non-saturating ) and Figure 4—figure supplement 1A ( saturating ) ) similar to the human HSPA8-HSPH2 ( Hsp70-Hsp110 ) disaggregase system containing DNAJA2 ( A ) and DNAJB1 ( B ) under identical experimental conditions . We conclude that the yeast disaggregase system is geared to function with both Ydj1 and Sis1 individually , but requires synergistic action of the two J-proteins to gain high efficiency similar to metazoan Hsp70-based disaggregases . In contrast , the refolding-only reactions performed with soluble misfolded monomeric luciferase ( Figure 4A , see scheme ) using yeast Hsp70 system ( Ssa1-Sse1-Ydj1/Sis1 ) showed no synergy in the presence of class A + class B J-proteins ( Figure 4—figure supplement 1B ) . Therefore , the observed synergy in aggregate resolution occurs at the upstream disaggregation step and not in the subsequent polypeptide refolding step mediated solely by the yeast Hsp70 system . On the contrary , under similar conditions , the E . coli bichaperone disaggregase system reactivated aggregated luciferase very efficiently independent of J-protein class and mixing ( Figure 4C ) . The E . coli Hsp70 ( DnaK ) system also did not require the cooperation between DnaJ and CbpA for efficient refolding of soluble misfolded luciferase ( Figure 4—figure supplement 1C ) . Collectively , our functional assays show eukaryotic disaggregases are cogged to depend on J-protein complex formation for efficacy , which is absent in bacteria . These findings are fully consistent with our biological , phylogenetic , coevolutionary and protein structure based observations . To dissect the disaggregation process further and determine aggregate specificities of bacterial and yeast class A and B J-proteins , we employed a size-exclusion chromatography ( SEC ) -based strategy ( Nillegoda et al . , 2015 ) . SEC in principle separates proteins ( monomeric/oligomeric ) in solution by size . However , this separation may be influenced not only by the size but also by the shape and density ( nature of packing ) of the protein aggregates detected . Tritiated ( 3H ) luciferase after heat denaturation/aggregation elutes in two peaks ( Figure 4D ) ( Nillegoda et al . , 2015 ) . For ease of analysis , we further divided the broad aggregate elution peak into two fractions and defined a total of three aggregate fractions ( F1 , F2 and F3 ) . For simplicity , we define these aggregate fractions by size ( F1 , ≥5000 kDa; F2 , 700–4000 kDa; F3 , 200–700 kDa ) ( Figure 4D ) . Soluble monomeric luciferase ( ~63 kDa ) elutes in the F4 fraction . Aggregate resolution was quantified by the degree of tritium counts remaining in each fraction compared to control ( black ) post disaggregation . Using this aggregate profiling system , we observed that Ydj1 ( A , green ) and Sis1 ( B , blue ) J-proteins alone selectively target the yeast Hsp100-Hsp70 disaggregase to resolve small ( F3 ) and large ( F1 ) luciferase aggregates , respectively ( Figure 4E ) . However , when combined , the resulting interclass J-protein complexes targeted the disaggregase machinery to all aggregate populations , including medium-sized aggregates eluting in F2 fraction ( Figure 4E and Figure 4—figure supplement 1D ) . In accordance , a synergistic increase in disaggregated luciferase monomers appeared in fraction F4 ( magenta , Figure 4E ) . The human disaggregase system , containing DNAJA2 ( A ) and DNAJB1 ( B ) , also showed similar aggregate resolution patterns ( Nillegoda et al . , 2015 ) , indicating interclass J-protein action in broadening aggregate targeting specificity is conserved from yeast to human . In the absence of Hsp104 , the yeast Hsp70 system showed relatively low luciferase aggregate resolubilization at 120 min ( Figure 4—figure supplement 1E ) consistent with in vivo findings ( Hsieh et al . , 2014 ) . However , residual disaggregase activity was observed with detectable drops in F1-F3 aggregate levels in the mixed class J-protein containing reaction , although insufficient for full J-protein cooperation as observed when Ssa1 was substituted with human HSPA8 ( Figure 4—figure supplement 1F ) . These findings possibly reflect an early stage of the evolution of Hsp70-based disaggregase systems that appear powerful in metazoa ( Nillegoda and Bukau , 2015 ) . Our earlier findings showed that the ClpB-DnaK bichaperone system reactivates aggregated luciferase with high efficiency independent of J-protein class ( Figure 4C ) and complex formation ( Figure 3M ) . To gain mechanistic insight , we examined how the E . coli disaggregase system achieves this robustness in the absence of interclass J-protein complex formation . The SEC profiling assays showed that DnaJ and CbpA lacked class-based selective aggregate targeting . Both bacterial J-proteins independently guided the ClpB-DnaK disaggregase to all aggregate sizes in F1-F3 fractions ( Figure 4F ) , consistent with the luciferase activity results in Figure 4C . The shorter reaction time point ( 40 min ) , however , showed the appearance of class B-like specialization in CbpA revealed by a delay in solubilizing smaller aggregates in F2-F3 ( blue , Figure 4—figure supplement 1G ) . These smaller aggregates were , nevertheless , completely resolved at 120 min ( Figure 4F ) . Subtle synergistic variations in activities masked by robustness of the bacterial system may yet hint toward some degree of J-protein cooperation . We , however , did not observe such synergistic activity even when the bacterial disaggregase components were further depleted by three fold to limit activity ( Figure 4—figure supplement 1H ) . Unbiased Brownian dynamics docking simulations ( Martinez et al . , 2015 ) between JD and CTD of CbpA and DnaJ did not capture the preferred binding arrangement described for interclass complex formation between human J-proteins ( Nillegoda et al . , 2015 ) ( Figure 4—figure supplement 1J ) , further corroborating our structural , biochemical and cell biological findings . Reactions containing only the bacterial Hsp70 system showed considerably less aggregate dissolution at 120 min ( Figure 4—figure supplement 1I ) . In essence , based on this example , bacteria use a different strategy to achieve high efficiency in protein disaggregation , evading a requirement for interclass J-protein complex formation . The extremely high overlap between DnaJ and CbpA for luciferase aggregate selection may explain the simple interchangeability of these two J-proteins for disaggregase targeting function in vivo ( Winkler et al . , 2012 ) . Overall , we observe key operational changes between bacterial and yeast bichaperone disaggregase systems mediated by different J-protein configurations . Our data show that class-specific specialization among eukaryotic J-proteins restrict the aggregate sizes that can be solubilized by the Hsp70 system . We reasoned that the less specialized class A or class B bacterial J-proteins maybe able to override this selectivity when combined with the human Hsp70-based disaggregase , leading to increase of disaggregation activity . We first assessed the cross-species communication between bacterial J-proteins and human Hsp70 system . Both bacterial DnaJ and CbpA cooperated with human HSPA8 and HSPH2 to refold heat-denatured , but predominantly monomeric luciferase ( Nillegoda et al . , 2015 ) ( Figure 4A scheme , Figure 4—figure supplement 2A ) . This is consistent with stimulation of ATP hydrolysis in HSPA8 by bacterial J-proteins ( Minami et al . , 1996 ) . Of note , the DnaJ-containing reaction showed a considerably increased luciferase activity compared to CbpA-containing reactions , consistent with the higher refolding capacity of class A J-proteins with Hsp70 ( Nillegoda et al . , 2015 ) . However , deferring from the cross-species collaboration observed for protein refolding and contrary to our expectations , DnaJ and CbpA ( single or mixed ) completely failed to cooperate with the human Hsp70 system for protein disaggregation ( Figure 4—figure supplement 2B ) . In contrast , yeast Ydj1 and Sis1 fully complemented the human orthologs in both protein refolding and disaggregation . Further , when paired with human DNAJA2 and DNAJB1 to form A+B cross-species J-protein combinations , only yeast ( and not E . coli ) members could cooperate and synergize protein disaggregation ( Figure 4—figure supplement 2B ) well above additive effects ( Figure 4—figure supplement 2C ) . To further validate these observations , we next tested for cross-species physical interaction between human and yeast class A and B J-proteins in vitro using a previously established FRET-based competition assay , which captures the intermolecular JD-CTD bidirectional cross interaction ( indicated by red dotted lines , Figure 4—figure supplement 2D ) between human DNAJA2 and DNAJB1 ( Nillegoda et al . , 2015 ) . Addition of five-fold excess unlabeled DNAJA2 or DNAJB1 decreased energy transfer ( measured as donor quenching ) between fluorophores located at the JD of DNAJA2 ( labeled with acceptor fluorophore ReAsH attached to the CCGPCC motif at the N-terminus of the JD ) and the hinge region of the CTD of DNAJB1 ( labeled with donor fluorophore Alexa Fluor 488 at residue Cys278 ) ( Figure 4—figure supplement 2D ) consistent with our previous findings ( Nillegoda et al . , 2015 ) . Similarly , addition of excess unlabeled yeast , but not bacterial J-proteins competed with complex formation between the labeled human class A and B J-proteins ( Figure 4—figure supplement 2D ) . The observed cross-species interactions among yeast and human J-proteins correlate with the general conservation of electrostatic potentials at the JD-CTD interaction sites on eukaryotic class A and B J-proteins ( Figure 1C , D ) . The level of donor quenching by the yeast J-proteins was however less compared to the human homologs . This is likely due to the composite nature of J-protein complex formation and small structural variations at the JD-CTD interaction interfaces ( Figure 1E , F ) . Together , these findings suggest that the communication between class A and class B J-proteins is not only playing a role in aggregate selection , but also important for Hsp70-based disaggregase machine assembly and/or architecture . The number of class A and B members in the cytosol have multiplied during evolution from prokaryotes to eukaryotes ( Nillegoda and Bukau , 2015; Kampinga and Craig , 2010 ) . The interclass J-protein networking function we describe provides additional flexibility to eukaryotic Hsp70 systems in extending the range of targeted substrates and protein quality control processes . This raises the question how prevalently interclass J-protein networking is employed for distinct functions such as disaggregation within expanded J-protein families of eukaryotes . The human cytosol contains four canonical class A members ( DNAJA1 , DNAJA2 , DNAJA4 and an isoform of the mitochrondrial DNAJA3 ) and nine class B J-proteins of which DNAJB1 , DNAJB4 and DNAJB5 constitute the canonical members . The rest of class B members form a set of non-canonical J-proteins that possess an N-terminal JD and a G/F rich region , but lack the two β-sheet rich barrel topology CTDs and the DD ( Lu et al . , 2006; Kampinga and Craig , 2010 ) . Instead , these non-canonical class B J-proteins , such as DNAJB2 and DNAJB8 , have very diverse C-terminal domains with distinct functions ( Kampinga and Craig , 2010 ) . DNAJB2 and DNAJB8 , but not DNAJB1 , specialize in preventing aggregation of amyloidogenic proteins via the C-terminally located ubiquitin-interacting motifs ( UIMs ) and serine-rich stretches ( SSF-SST ) ( Figure 5A ) , respectively ( Labbadia et al . , 2012; Hageman et al . , 2010; Gillis et al . , 2013 ) . Moreover , DNAJB2 seems to compete with the folding/holding chaperone activities of DNAJB1 and to enhance the degradation of misfolded proteins prior to their aggregation in cells ( Howarth et al . , 2007; Westhoff et al . , 2005 ) . DNAJB8 functions as a poly-dispersed oligomeric complex instead of forming dimers as observed with DNAJB1 . We further checked for functional divergence within these class B J-protein family members by assessing the cooperation of the two non-canonical members with DNAJA2 ( class A , canonical member ) in protein disaggregation . Compared to DNAJB1 , DNAJB2 and DNAJB8 were incapable of reactivating aggregated luciferase even when mixed with DNAJA2 ( Figure 5B ) , showing a clear separation in function . Moreover , FRET competition assays revealed that these specialized J-proteins were unable to complex with DNAJA2 and DNAJB1 ( Figure 5C ) . 10 . 7554/eLife . 24560 . 016Figure 5 . A naturally occurring discriminatory strategy correctly pairs J-proteins for specialized functions in eukaryotes . ( A ) Structural organization of canonical ( DNAJB1 ) and non-canonical ( DNAJB2 and DNAJB8 ) members of the class B J-protein subfamily . ( B ) Reactivation of aggregated luciferase by the human Hsp70-based disaggregase system with canonical and non-canonical J-proteins under saturating chaperone levels . n = 3 . See Materials and methods for protein concentrations used . ( C ) Competition of unlabeled DNAJB1 , DNAJB2 and DNAJB8 for the JD-CTD interaction between DNAJB1 and DNAJA2 analyzed by FRET . Bars represent donor quenching efficiency of JD and CTD intermolecular interactions . Cartoon shows fluorophore positions mapped onto DNAJB1 and DNAJA2 protomers . Red dotted lines indicate intermolecular JD-CTD bidirectional interactions . The N-terminus of DNAJA2JD and the C-terminus of DNAJB1CTD ( at residue Cys278 ) were labeled with the acceptor fluorophore ReAsH and the donor fluorophore Alexa Fluor 488 , respectively . n = 3 . Two-tailed t-test **p<0 . 01 . ( D ) Electrostatic isopotential contour maps ( cyan + 1 , red −1 kcal/mol/e ) of the J-domains of DNAJB1 , DNAJB2 and DNAJB8 . DNAJB1RRR is the triple mutant of DNAJB1 ( D4R , E69R , E70R ) that fails to interact with opposite class CTDs ( Nillegoda et al . , 2015 ) . ( E–L ) Interclass interactions between V5-DNAJB1 ( control ) , V5-DNAJB2 and V5-DNAJB8 with either DNAJA2 ( E–H ) or Hsp70 ( I–L ) captured by PLA in HeLa cells . Red punctae reflect a single complex formation event between the indicated J-proteins . PLA controls for antibody specificities in mock transfected cells ( E , I ) . Nuclei stained with DAPI ( cyan ) . Scale bar = 10 μm . n ( biological repeats ) = 2 . The Hsp70 antibody recognizes both human Hsp70 and Hsc70 variants . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 01610 . 7554/eLife . 24560 . 017Figure 5—figure supplement 1 . The J-domain of DNAJB8 is insufficient for interclass J-protein cooperation and disaggregation synergy . ( A ) Local PIPSA analysis results comparing the electrostatic potentials of eukaryotic class B JD in the region around α-helices I and IV and the mutated residues in DNAJB1RRR ( inset , black circle indicating the spherical region analyzed in PIPSA ) . The JD from Escherichia coli ( P36659 , CbpA ) was included as a control . The electrostatic potentials were clustered by similarity using Ward’s clustering . The heat maps show clustering of J-proteins by similarity ( higher similarity indicated by a red shift ) . ( B ) FRET assays measuring the competition of wild-type DNAJB1 or J-domain chimeras of DNAJB1 with the JD-CTD interaction between DNAJB1 and DNAJA2 . Bars represent donor quenching efficiency of JD and CTD intermolecular interactions . Cartoons above graph show fluorophore positions mapped onto DNAJB1 and DNAJA2 protomers . Red dotted lines indicate intermolecular JD-CTD bidirectional interactions . n = 3 . Two-tailed t-test *p<0 . 05 , **p<0 . 01 , ns = not significant . ( C ) As in ( B ) , FRET assays measuring the competition of wild type DNAJB1 , Sis1 , DNAJB1RRR or J-domain chimeras of DNAJB1 with the JD-CTD interaction between DNAJB1 and DNAJA2RRR . RRR denotes J-domain charge reversal mutations D4R , E69R , E70R for DNAJB1 and D6R , E61R , E64R for DNAJA2 . n = 3 . ( D–F ) Disaggregation/refolding of heat aggregated luciferase by human HSPA8-HSPH2 chaperone system containing DNAJB1 , CbpAJD ( E . coli ) -DNAJB1CTD and DNAJB8JD-DNAJB1CTD . Interclass J-protein combinations are generated with class A J-protein DNAJA2 . Inset , electrostatic isopotential contour maps of JDs of DNAJB1 , E coli CbpA and DNAJB8 ( cyan + 1 , red −1 kcal/mol/e ) . n = 3 . Error bars reflect mean ± sem . See Methods for protein concentrations used . ( G ) Electrostatic potential of the J-domain of human DNAJB6 . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 017 How is this functional and physical separation , which seems crucial to avoid futile interactions among different J-protein class members effectively established ? The minimal structural element required for interclass J-protein communication is the JD . Isolated JDs effectively bind to opposite class CTDs and compete with complex formation between wild-type J-proteins ( Nillegoda et al . , 2015 ) . Therefore , we hypothesized that JDs of DNAJB2 and DNAJB8 must contain inherent signals to prevent interclass interactions . Since JD-CTD interactions are sensitive to high-salt concentrations ( Nillegoda et al . , 2015 ) , we examined the electrostatic potentials at the JDs of these non-canonical members . In JDs of DNAJB2 and DNAJB8 , the negatively charged patch critical for opposite class CTD interaction was replaced by a highly positively charged region ( Figure 5D ) . This configuration was similar to that of the previously analyzed complex-forming defective , triple charge-reversal variants ( RRR ) of class A and B J-proteins ( Nillegoda et al . , 2015 ) ( e . g . DNAJB1RRR; D4R , E69R and E70R ( RRR ) , Figure 5D ) . We could also quantitatively show that the electrostatic potentials in the region around helices I and IV of the JDs of DNAJB2 , DNAJB8 and DNAJB1RRR clearly differ from the of the electrostatic potentials of the same region in canonical DNAJB1/DNAJB4 JDs ( Figure 5—figure supplement 1A ) . We presumed that DNAJB2 and DNAJB8 avoided interaction with DNAJA2 using this naturally occurring charge reversal ( negative to positive ) at the JDs as a repulsive signal . To test our presumption , we generated a DNAJB1 chimera containing either the JD of DNAJB8 or the JD of CbpA ( control ) . The E . coli CbpAJD has a charge bipolarity character comparable to eukaryotic class BJDs ( Figure 1D and Figure 5—figure supplement 1A ) . FRET competition assays with excess unlabeled DNAJB8JD-DNAJB1CTD and CbpAJD-DNAJB1CTD chimeras showed equal reduction in donor quenching , but to a lesser degree compared to wild-type DNAJB1 ( Figure 5—figure supplement 1B ) . The partial FRET donor quenching by the DNAJB8JD-DNAJB1CTD chimera most likely resulted from a unidirectional JD–CTD interaction in which the JD of DNAJA2 interacts with the CTD of DNAJB1 in the DNAJB8JD-DNAJB1CTD chimera . Consistently , this partial intermolecular tethering was completely abolished when the FRET experiment was repeated with an N-terminally ReAsH labeled DNAJA2RRR ( Nillegoda et al . , 2015 ) ( Figure 5—figure supplement 1C ) . DNAJA2RRR carries a J-domain charge reversal triple mutation ( D6R , E61R and E64R ) , which diminishes its JD from binding to the CTD of the DNAJB8JD-DNAJB1CTD chimera . As a negative control , we used unlabeled DNAJB1RRR as a competitor , and as expected , this complex forming defective mutant behaved similar to the DNAJB8JD-DNAJB1CTD chimera ( Figure 5D and Figure 5—figure supplement 1C ) . Importantly , the degree of intermolecular tethering by a unidirectional JD–CTD interaction was insufficient for full J-protein cooperation and disaggregation efficiency ( Figure 5—figure supplement 1D–F ) , consistent with previous observations ( Nillegoda et al . , 2015 ) . We surmise that a bidirection JD–CTD tethering may stabilize and/or contribute to a specific architecture for J-protein oligomerization that facilitates the assembly of the Hsp70-based disaggregase on the surface of an aggregate . In agreement , Sis1 and the CbpAJD-DNAJB1CTD chimera that could establish bidirectional JD–CTD cross tethering with DNAJA2 ( Figure 4—figure supplement 2D and Figure 5—figure supplement 1B , C ) due to compatible JD-CTD contact sites , were able to synergize protein disaggregation ( Figure 4—figure supplement 2C and Figure 5—figure supplement 1E ) . The strength of the JD-driven intermolecular bidirectional tethering correlates with the degree of disaggregation synergy observed with the respective J-protein pairs ( Figure 5—figure supplement 1D , E ) . These findings indicate that further refinements have occurred in the electrostatic potentials of eukaryotic JDs to maximize cooperation with opposite class members . Finally , we provide proximity ligation assay derived biological evidence to confirm our biochemical findings ( Figure 5E–L ) . In cultured HeLa cells , V5 tagged DNAJB2 and DNAJB8 did not form red puncta when tested for complex formation with DNAJA2 , even under overexpression conditions ( Figure 5G , H ) , but readily interacted with Hsp70 via JDs ( Figure 5K , L ) . Controls with V5-DNAJB1 displayed complex formation with both DNAJA2 and Hsp70 indicating that the N-terminal tag does not interfere with JD mediated interactions ( Figure 5F , J ) . In essence , non-native interactions among members within the J-protein network are simply abolished by an ensuing charge reversion ( negative to positive ) at the CTD interaction region of J-domains .
In this study , we provide structural , phylogenetic , biochemical and cell biological evidence to support a eukaryote-specific occurrence of interclass complexes between the canonical J-proteins of classes A and B . The distinct change in J-protein biology at the prokaryotic-to-eukaryotic split , where class members form cooperative networks ( Figure 6 ) , may have triggered functional consequences linked to specific changes in organismal physiology . Habitat wise , bacteria and yeast are exposed to constantly changing harsh environmental stresses such as extreme heat , and are particularly dependent on potent protein disaggregases for stress recovery and survival ( Sanchez and Lindquist , 1990; Mogk et al . , 1999 ) . The bacterial ClpB-DnaK system may rely on the broad ( size-based ) aggregate targeting ability of DnaJ and CbpA for high disaggregation efficiency , which excludes the need for interclass J-protein complex formation . There is also a biological pertinence to the absence of interclass J-protein cooperation in bacteria . DnaJ and CbpA show a clear temporal and spatial separation in expression patterns and intracellular localization in E . coli ( Azam et al . , 2000; Cosgriff et al . , 2010; Li et al . , 2014; Yamashino et al . , 1994 ) . This conceivably prevents considerable encountering events between the two molecules in a biological setting hence reducing the chance to evolve a cooperative function . Additionally , the operon linked , J-protein inhibitor CbpM may further contribute to this by stably binding to the JD of CbpA ( Sarraf et al . , 2014 ) and blocking the evolution of JD-mediated interactions ( Chenoweth et al . , 2007 ) . Moreover , as opposed to eukaryotes , bacteria contain only single copies of the canonical J-protein class members optimized for essential biological functions , which may disfavor positive selection for new features ( Ohta , 2000; Richard and Yvert , 2014 ) such as sites for opposite class JD interaction . Together , the negative selective pressure , the distinct cytosolic localization of DnaJ and CbpA , and the presence of JD blocking CbpM explain why the two J-protein types may not have evolved to interact in E . coli . 10 . 7554/eLife . 24560 . 018Figure 6 . Canonical class A and class B members form interclass J-protein networks in eukaryotic cells . Cytosolic yeast and human canonical class A and class B members ( e . g . Ydj1 and Sis1 , Saccharomyces cerevisiae; DNAJA2 and DNAJB1 , Homo sapiens ) form interclass J-protein complexes via complementary binding interfaces at the JDs and the hinge regions of CTDs . E . coli cytosol contains only a single pair of canonical class A ( DnaJ ) and class B ( CbpA ) J-proteins . These bacterial J-proteins however fail to form interclass J-protein complexes due to the lack of complementary structural features required to establish intermolecular JD-CTD contacts . DOI: http://dx . doi . org/10 . 7554/eLife . 24560 . 018 In contrast , further diversification of eukaryotic class A and B members has resulted in narrowing the substrate selection for the yeast disaggregase system . To compensate for this loss in functional plasticity , we observe the emergence of interclass J-protein complexes that combine different substrate-binding modules of both classes A and B for broader recognition of certain aggregate types . This substrate targeting modulation of eukaryotic disaggregases mediated by J-proteins may have helped evolve new beneficial functions that are absent in prokaryotes; For example , a specialized role of the Hsp104-Ssa1 system in regulating prion-like conformational behavior of naturally occurring cell signaling proteins in unicellular yeast ( Newby and Lindquist , 2013 ) . In multicellular organisms , additional evolutionary constraints and fitness costs linked to maintaining the Hsp104 system may have triggered the loss of Hsp104 and selected for Hsp70-based disaggregases ( Nillegoda and Bukau , 2015 ) . How the different classes of J-proteins ( single and in complex ) recognize distinct aggregate populations is unclear . The packing of the aggregate , the amino acid composition of the exposed protein segments of trapped molecules , and perhaps even the shape of the binding surface could influence aggregate selection by J-proteins . The J-protein class-specific differences in the substrate-binding domains ( Fan et al . , 2005 , 2004; Reidy et al . , 2014 ) , interdomain communication ( Reidy et al . , 2014 ) and binding modes ( Terada and Mori , 2000 ) could add further layers of complexity for this function . Our molecular understanding of how J-proteins interact with protein substrates remains largely enigmatic . In addition to broadening the substrate targeting , interclass J-protein complex formation could also help nucleate Hsp70 oligomerization . This is envisioned to facilitate the assembly of the metazoan Hsp70-based disaggregases on aggregate surfaces ( Nillegoda et al . , 2015; Nillegoda and Bukau , 2015 ) . In yeast , such J-protein-mediated conglomerations of Hsp70 may aid in both efficient recruitment and activation of Hsp100 hexamers ( Seyffer et al . , 2012; Carroni et al . , 2014 ) on different aggregate types . To achieve a similar outcome , the bacterial disaggregase system seems to employ J-proteins with broad aggregate recognition and is proposed to rely on homo J-protein oligomerization ( Celaya et al . , 2016 ) . The precise basis of J-proteins binding to aggregates has not been defined , but presumably J-proteins nucleate where looped-out polypeptide stretches are available for interaction . The discriminatory electrostatic potential signals for partner protein selection in J-protein networks , which fine-tune the entire Hsp70-based protein folding system , are communicated by a remarkably simple domain topology of four α-helices in J-domains . The distinct positively charged patch on α-helix II proximal to the HPD motif ( Figure 1C ) facilitates Hsp70 binding ( Lu and Cyr , 1998; Hennessy et al . , 2005 ) . This interaction may be further modulated in some Hsp70-J-protein pairs for specific functions through the dominantly negatively charged electrostatic cloud next to the HPD motif ( e . g . see JDs of DnaJ ( A . aceti , Sphingomonas sp . and C . ultunense ) , Ydj1 ( S . cerevisiae ) and DNAJB8 ( H . sapiens ) ; Figure 5D , Figure 1—figure supplement 1B and Figure 1—figure supplement 2B ) . Similarly , the electrostatic patch formed at the opposite end ( α-helices I and IV ) discriminates J-protein partnering for specialized functions in eukaryotic networks . For the interaction with canonical class A members ( e . g . DNAJA2 ) , a negatively charged patch proximal to α-helices I and IV of class B J-domains is required . A charge reversion at this site deters JD-CTD contacts with DNAJA2’s CTD , providing specificity to interclass J-protein pairing . It is very likely that this rule of interaction for complex formation is conserved between canonical class A ( e . g . DNAJA1 , DNAJA2 , DNJ-12 , Ydj1 ) and class B members ( e . g . DNAJB1 , DNAJB4 , DNJ-13 , Sis1 ) that show a high degree of electrostatic potential conservation at the JD-CTD contact interfaces ( Figure 1—figure supplement 1 and Figure 1—figure supplement 2 ) . On the contrary , a charge reversion at the CTD-binding interface of JDs helps DNAJB2 and DNAJB8 to avoid interacting with DNAJA2 and possibly other class A members . DNAJB6 , similar to its homolog DNAJB8 , has also lost the bipolar electrostatic potential in the JD suggesting that this non-canonical J-protein may also fail to complex with DNAJA2-like members ( Figure 5—figure supplement 1G ) . Our predictions , however , do not fully exclude the possibility that these non-canonical J-proteins are unable to interact with other canonical J-proteins under different growth conditions for joint chaperone actions in eukaryotic cells . In yeast , the J-domain of Tim14 ( class C ) forms a complex with the pseudo J-domain of Tim16 ( class C ) during mitochondrial protein import ( Mokranjac et al . , 2006 ) and such JD-JD driven interactions may also exist among some J- or J-like proteins for specialized functions . Extensive biochemical and functional genomic approaches are now needed to fully understand the extent and regulation of this intricate J-protein network , especially in humans where J-protein targeted chaperone machineries are implicated in a wide range of pathologies including cancer , neurodegeneration , muscular atrophies , metabolic disorders and infectious diseases ( Koutras and Braun , 2014; Gibbs and Braun , 2008; Knox et al . , 2011; Maier et al . , 2008; Synofzik et al . , 2014 ) . In summary , our broad survey of J-proteins across kingdoms of life captures a eukaryote-specific , adaptive evolution in canonical class A and class B J-proteins to allow for interclass complex formation that modulates Hsp70 machinery targeting . Our data shows how J-proteins , individually or in complex , are employed to regulate the operational efficacy of bacterial , yeast and human disaggregation systems . We finally explain a naturally occurring elegant strategy to correctly pair J-proteins for specialized tasks ( e . g . protein disaggregation ) , especially in humans where over 50 isoforms of J-proteins exist . In effect , the diverse members of eukaryotic class A and B J-proteins deriving from multiple gene duplications are reconnected via selective complex formation to ensure fine-tuning of distinct biological functions .
Multiple sequence alignments ( MSA ) for class A and B J-proteins were built separately as follows: for each class , we collected a seed with curated and manually annotated sequences and aligned them using MAFFT ( Katoh et al . , 2002 ) . Sequences in the class B J-protein seed comprised the J-domain , the GF region and both CTDs . Sequences in the class A J-protein seed additionally contained the characteristic zinc-finger domain . We then used HMMER ( Finn et al . , 2011 ) to build a hidden Markov model and scan the union of the Uniprot and Swissprot databases to extract homologues for both sub-families . The retrieved sequences were then filtered by removing all hits containing more than 20% gaps . To ensure that no class A sequences were present in the class B alignment ( and vice-versa ) , we further filtered the two datasets as follows: all complete and unaligned sequences from the two MSAs were retrieved; from the class B MSA , we discarded all instances whose complete sequences contained the characteristic zinc-finger ( ZF ) CxxCxGxG motif . While the canonical ZF typically has four of these motifs , we observed that some ZF domains had variations at one of the two glycines . As a consequence , we only kept sequences in the class A MSA whose complete sequences contained at least two of these characteristic motifs . This procedure resulted in 12215 class A J-protein sequences and 4194 class B J-protein sequences . Direct Coupling Analysis ( DCA ) was performed using the asymmetric version of the pseudo-likelihood method ( Morcos et al . , 2011; Ekeberg et al . , 2013 ) . Sequences were reweighted using a maximum of 90% identity threshold ( Ekeberg et al . , 2013; Hopf et al . , 2014; Balakrishnan et al . , 2011 ) . For the prediction of inter-class interactions , the interacting sequences of the two classes should be concatenated for each organism . The canonical approach for prediction of protein-protein interactions using DCA consists in using information on the genomic location of the sequences to predict which pairs of sequences are most likely to be paired in an organism . This strategy has been effective in the case of bacterial sequences organized in operons ( Hopf et al . , 2014; Ovchinnikov et al . , 2014; Feinauer et al . , 2016 ) , but fails when applied to eukaryotes or bacterial sequences that are distant on the genome . In this work , we faced six issues with the pairing problem: ( 1 ) DnaJs are present as multiple paralogs in all organisms . ( 2 ) The number of DnaJs strongly varies between organisms . ( 3 ) There is no systematic knowledge of interacting partners . ( 4 ) DnaJs are not located in operons or on nearby positions in the genome . ( 5 ) The systematic matching of all DnaJAs with all DnaJBs in an organism leads to a very large number of possible combinations , most of which probably do not interact . ( 6 ) Matching too many non-interacting pairs dilutes the coevolutionary inter-protein signal . Because of the these difficulties , we adopted the following matching strategy: for each class A J-protein sequence in a given organism , we matched it with a single randomly chosen class B J-protein sequence of the same organism . We also enforced that any class B J-protein sequence was only matched with a single class A J-protein sequence . These class A/B matched sequences are then collected for all species and form a randomly matched MSA . This process was repeated 300 times , resulting in an ensemble of different MSAs . DCA was then performed for each alignment . Finally , we considered those predicted DCA pairs which appeared in at least 5% of the 300 MSAs , as selected using a threshold of 0 . 8 ( Hopf et al . , 2014 ) on the normalized DCA score above which inter-protein residue pairs were considered statistically significant . Phylogenetic trees were built using the RaxML software suite ( Stamatakis , 2014 ) . To decrease their size , MSAs for class A and class B proteins were first pruned , retaining only sequences with maximum 90% identity . Phylogenetic trees were then computed with a standard protocol ( 20 maximum likelihood searches , 100 bootstraps ) and the best tree was returned . We observed that support values ( Yang and Rannala , 2012 ) were generally low for the interior branches , but the overall phylogenetic separation was robust , reproducing a coherent phylogeny at large scale . We developed a methodology to assess the residues responsible for the phylogenetic and functional differences observed in J-proteins , that we call Phylogenetic Discriminant Analysis ( PDA ) . This method bears some resemblance with the critical variable selection methodology introduced in Grigolon et al . , 2016 ) but relies on phylogenetic annotations and thus falls into the category of supervised learning . Our approach was as follows: For each of the ( N3 ) possible position triplets ( N is the MSA width ) , we built a reduced MSA consisting only of these three positions . We then performed principal component analysis ( PCA ) on this reduced MSA to project the sequences on a maximum-variance subspace ( Casari et al . , 1995 ) . The sequences were then clustered together in this subspace using hierarchical clustering ( Murtagh and Contreras , 2012 ) . To define the number of clusters , we set a cut-off on the distance between clusters equal to the average distance between all sequences . By doing so , the number of clusters does not need to be explicitly chosen for each position triplet . The homogeneity of each cluster c with respect to phylogeny was then measured by means of the Shannon entropyh ( c ) =−∑icP ( ic ) logP ( ic ) where P ( ic ) is the fraction of sequences in cluster c belonging to the phylogenetic group i . The distributions of phylogenetic groups were then measured for each cluster , and their entropies computed . The average entropy over all clusters is thenH ( C ) =−∑c∈Cw ( c ) h ( c ) where w ( c ) is the fraction of sequences in cluster c and C is the set of all clusters found by the algorithm . The average entropy over the clusters is then a measure of how well the position triplet discriminates the different sub-classes . If the average entropy is low , then the clusters built on a given triplet contain a lower mixing of phylogenetic classes , and the position triplet is hence a good discriminator of phylogeny . As the entropy is a monotonic function of the ‘mixing’ , the entropy H is thus a good scale to score the discriminative power of triplets: The lower H is , the less mixed are the clusters based on a given triplet of positions . By computing the distribution of entropies over all triplets ( Figure 2—figure supplement 2 , left sub-panels ) , we could evaluate an empirical p-value assessing the statistical significance level of a triplet in discriminating the phylogenetic groups . The p-value of a triplet was simply defined as the fraction of triplets having equal or lower entropy score H . The choice of considering triplets was based on experimental evidence showing that a triple RRR mutant in the J-domain abolished cooperation between class A and B J-proteins ( Nillegoda et al . , 2015 ) . We therefore wanted to examine if the choice of these three positions was statistically found from sequence analysis . In the case of the analysis of the J-domains , we could systematically test all possible 54740 triplets , whereas for the CTD analysis , we randomly chose a subset of 50000 triplets to limit the computational burden . We performed sixfold cross-validation and verified that this under-sampling was a good approximation . This method could in principle be extended to the analysis of k-mers ( k > 3 ) . However , the number of combinations grows exponentially with k ( for the JD with 70 positions , this results in ~900 , 000 4-mers and 12 millions 4-mers , while for the CTD there are already ~ 160 millions 4-mers ) . An alternative strategy is to look at all 3-mers , and retain positions that appear most often in the strongly discriminating triplets ( Figure 2—figure supplement 2 , right sub-panels ) . Here , we have followed this strategy . The most discriminating positions ( top five in the main text ) were selected as the positions that appeared most often in all triplets having lower entropy than the reference RRR triplet at residues 4 , 69 and 70 in DNAJB1 ( or 6 , 61 , and 64 in DNAJA2 ) . The reference RRR triplet was here used as a threshold for considering strongly discriminant positions , as we wanted to test whether this particular triplet had any statistical significance from a phylogenetic discrimination point of view . In the case of the J-domains , the reference triplet had a p-value of 4 . 6% . For the CTD analysis , where no reference triplets were available , we set the reference p-value to 5% . In order to set a threshold to select the most frequently appearing residues in the high ranked triplets , we considered a uniform prior null model: If m denotes the number of selected high ranked triplets , the null model for the average probability of each residue to be selected is simply given by pNull ( i ) =3/Npos ( where Npos = 70 for the JD , 254 ( resp . 275 ) for CTDB ( resp . CTDA ) ( Figure 2—figure supplement 2 , right sub-panels , dashed red lines ) . Given a finite sampling of m , the standard error of the mean of the null model is given by σp=pNull ( 1−pNull ) m . This allows estimating a p-value for the outliers of the distributions of selection probability in terms of standard deviations from the mean ( Figure 2—figure supplement 2 , righ sub-panels . Dashed magenta ( resp . green ) lines denote 3 ( resp . 10 ) standard errors of the mean . We note that the strong assumptions of the null model ( completely uniformly distributed triplets ) , results in outliers having high deviations from the mean ( or alternatively very low p-values ) . In the main analysis , we have used a conservative choice , considering outliers above 10-sigma deviations as significant ( p-value<10−23 ) . We observe that taking a less restrictive selection ( three sigma , p-value<1 . 5 10−3 ) results in the selection of a small number of additional PDA residues , which lie in close vicinity to the ones selected by the more stringent 10-sigma criterion . To assess the robustness of the PDA analysis , we tested this methodology with different separations of phylogenetic classes ( from ‘Bacteria-Eukaryotes’ up to ‘Fungi-Proteobacteria-Firmicutes-Viridiplantae-Other Bacteria-Other Eukaryotes’ and found the results to be robust ( Figure 2—figure supplement 2A–G ) . Furthermore , we verified that when using another clustering method ( modularity based clustering [Granell et al . , 2011] ) , the results did not change ( Figure 2—figure supplement 2H ) . The same structures and models for human ( DNAJA1 , DNAJA2 , DNAJB1 , DNAJB4 ) and C . elegans ( DNJ-12 , DNJ-13 ) proteins as in Nillegoda et al . ( 2015 ) were used . For the other proteins studied , three dimensional structures were available and used for the following: J-domains of Sis1 ( PDB ID: 2o37 ) , E . coli class A ( PDB ID: 1xbl ) , E . coli class B ( PDB ID: 3ucs ) , human DNAJB8 ( PDB ID: 2dmx ) , human DNAJB2 ( PDB ID: 2lgw ) and CTD of Sis1 ( PDB ID: 1c3g ) and Ydj1 ( PDB ID: 1nlt , 1xao ) . The structure of the CTD dimer of Ydj1 was built using the structure of the CTD monomer ( PDB ID: 1nlt ) , which was superimposed twice on a crystal structure containing the dimerization site ( PDB ID: 1xao ) by using PyMOL ( http://www . pymol . org ) . For the remainder of the proteins studied , no crystal or NMR structure was available . Therefore , three-dimensional structures of the domains of these proteins were built by comparative modeling using the Swiss-Model webserver ( http://swissmodel . expasy . org ) ( Biasini et al . , 2014 ) . For all class A CTD models , the CTD dimer model of Ydj1 was used as a template structure and the two Zn2+ ions were transferred afterwards . For the class A CTD of Pseudomonas oryzihabitans ( Uniprot accession: A0A0D7F716 ) , a less conserved loop close to the Zn2+ binding region was modeled with different backbone coordinates from the template structure but these prohibited realistic Zn2+ binding because of a too large binding distance . Therefore , three residues were changed in the sequence to force the Swiss-Model algorithm to model the same backbone coordinates as in the template structure ( 'KIIPEP' → 'DIIKDP' ) . Afterwards , the three residues in the model structure were back-mutated using the mutagenesis tool in the PyMOL software . This model was then used as a template structure in the Swiss-Model webserver to slightly adapt the side chains in the mutated region . Only in the case of the Sphingomonas sp ( strain SKA58 ) DnaJ ( UniProt accession: Q1NCH5 ) was the model of Acetobacter aceti 1023 DnaJ ( UniProt accession: A0A063 × 4A7 ) , which was built using the Ydj1 model , used as a template because the less conserved loop around the Zn2+ binding region was modeled better for Zn2+ ion binding than when the Ydj1 model was used . In the case of the Bordetella pertussis ( UniProt accession: Q7VVY3 ) class A CTD , the sequence alignment was manually adapted to enable the modeling of the C-terminal region . For this purpose , a multiple sequence alignment of the four gamma and the beta bacterial sequences and the yeast sequence ( template structure ) was considered using the software DeepView ( Guex et al . , 2009 ) . A DeepView project with the adapted alignment was uploaded to the Swiss-Model webserver . For the class B CTDs , the Swiss-Model webserver was used to find a template structure and , if multiple templates were found , the one with the highest sequence identity to the target structure was chosen and then , in the case of more than one structure for this sequence , the corresponding structure with the highest QMEAN4 score . For the following class B CTDs ( UniProt accession numbers ) , the template structure 3lz8 . B was used: P36659 , P63262 , W9BQH2 , J7RE62 , F4JY55 . The PDB ID 3lz8 . A was used as a template structure for the following class B CTDs ( UniProt accession numbers ) : A0A0D7FE35 , Q1NEX3 , M1ZLZ3 , O75953 . For the Type B CTD of A0A063XA16 , the structure with the PDB ID 4j80 . A was used as a template . The dimer structure of Sis1 was built by superimposing the crystal structure of the monomer ( PDB ID: 1c3g ) twice on the 19 C-terminal residues of the crystal structure of the JB1 dimer ( PDB ID: 3agz ) . For the class A J-domain of Acetobacter aceti 1023 DnaJ ( UniProt accession: A0A063 × 4A7 ) , the structure with the PDB id 4j80 was chosen , and for the Sphingomonas sp ( strain SKA58 ) DnaJ ( UniProt accession: Q1NCH5 ) and the ATJ3 ( UniProt accession: Q94AW8 ) , the structure with PDB ID 4rwu was chosen as the template . In the case of DNAJA4 ( UniProt accession: Q8WW22 ) , the structure with PDB ID 2lo1 was taken as the template . For all other class A J-domains , the structure with the PDB ID 1xbl from E . coli was used as the template . For the class B J-domains , the following templates were used ( UniProt accession: PDB ID of template structure ) : A0A063XA16:4j7z , Q1NEX3:2dmx , M1ZLZ3:2yua , F4JY55 ( At5g25530 ) :2m6y , O75953 ( DNAJB5 ) :4wb7 , O75190 ( DNAJB6 ) :4j7z . For all other class B J-domains , the E . coli structure with the PDB id 3ucs was used . The structures were prepared by adding polar hydrogen atoms to the protein structures with WHATIF5 ( Vriend , 1990 ) . The electrostatic potential of each protein was calculated by numerically solving the linearized Poisson–Boltzmann equation with UHBD ( Madura et al . , 1995 ) . Electrostatic potential grids with 2503 grid points with 1 Å spacing were used for all proteins . The relative dielectric constants of the solvent and the protein were set to 78 . 0 and 4 . 0 , respectively , and the dielectric boundary was defined by the protein’s van der Waals surface . The ionic strength was set to 50 mM at a temperature of 300 K , with an ion exclusion radius ( Stern layer ) of 1 . 5 Å . The protein atoms were assigned OPLS atomic partial charges and radii ( Jorgensen et al . , 1996 ) . All class A J-domain structures were superimposed on the DNAJA2 J-domain . The class A CTDs were superimposed on the lower CTD-II domain of DNAJA2 . All class B J-domain structures were superimposed on the DNAJB1 J-domain . The class B CTDs were superimposed on the upper CTD-I domain of DNAJB1 . All structures were superimposed with the alignment tool of the PyMOL software . The similarity of the calculated electrostatic potentials of the superimposed structures was computed using the PIPSA ( Protein Interaction Property Similarity Analysis ) software ( Wade et al . , 2001 ) . The resulting distance matrix was used for a Ward’s clustering . Only for Type A CTDs was an average-clustering used , but this yielded similar results to the Ward’s clustering . For the local PIPSA analysis , a center and a radius were defined as follows . For the local PIPSA analysis of the class A CTD , the midpoint between the residue K226 in the DNAJA2 CTD and K21 in the DNAJB1 J-domain was chosen . This pair of residues was found in a lysine-specific cross-linking experiment ( Nillegoda et al . , 2015 ) . A docking simulation of the DNAJA2 CTD and the DNAJB1 J-domain supported the domain interaction and the coordinate of the midpoint was taken from the representative complexed structure ( see [Nillegoda et al . , 2015] for more information ) . The radius of the sphere was set to 25 Å to include the whole predicted interaction site . The same procedure was applied for the DNAJB1 CTD and the DNAJA2 J-domain , for which two cross-linking residues , K209 in the DNAJB1 CTD and the K46 in the DNAJA2 J-domain , were identified . The radius of the sphere was also set to 25 Å . For the PIPSA analysis of the metazoan JDs , average-clustering was applied . All metazoan JD structures were superimposed on the DNAJB1 JD and a sphere with a radius of 25 Å was set to cover the region around α-helix I and IV and the RRR mutation site of DNAJB1RRR . For the Brownian Dynamics simulations , the SDA software ( Martinez et al . , 2015 ) was used with the same conditions and the same clustering procedure for the docked J-domain as described in our previous study ( Nillegoda et al . , 2015 ) . The docked cluster representatives were used to calculate the average Euclidean distance between their center of geometry and the center of geometry of the previously docked JDDNAJB1 ( cluster one and two ) to the CTDDNAJA2 ( Nillegoda et al . , 2015 ) . The CTD of E . coli used for the docking simulations was superimposed on the CTDDNAJA2 before carrying out the docking simulations . Because of the dimeric structure , the distance to both cluster representatives 1 and 2 was calculated and the smaller distance was used for calculating the average distance . Bacterial , yeast and human recombinant proteins were expressed and purified as described previously ( Rampelt et al . , 2012; Nillegoda et al . , 2015; Westhoff et al . , 2005; Haslberger et al . , 2008 ) . The plasmid for His-tagged DNAJB2a purification was obtained from Dr . M . E . Cheetham ( University College London , UK ) . E . coli strain K-12 MG1655 encoding CbpA-mCherry ( Chintakayala et al . , 2015 ) was a kind gift from David Grainger ( University of Birmingham , UK ) . Strain NA01 was generated by transforming MG1655 encoding CbpA-mCherry with plasmid pDK194 carrying DnaJ-YFP under T7 promoter . IPTG induction of DnaJ-YFP was carried out as described previously ( Winkler et al . , 2010 ) . Strains were grown in Luria-Bertani ( LB ) medium at 30°C with appropriate antibiotic selections . S . cerevisiae strains were grown in yeast extract-peptone-dextrose ( YPD ) media at indicated temperatures using standard methods . Log phase cultures in YPD media obtained at 30°C . Sis1 depletion strain tet07-sis1 ( MATa; his3-1; leu2-0; met15-0; pSIS1::kanR-tet07-TATA URA3::CMV-tTA ) was obtained as kind gifts from Dr . D . Cyr ( University of North Carolina , USA ) . For Sis1 depletion , tet07-sis1 cells ( control: tet-off cells ) were grown overnight in YPD media , diluted back to OD600 = 0 . 05 in YPD containing 10 μg/ml doxycycline and grown for 20 hr . Log phase cells for experiments were obtained by diluting cells back to OD600 = 0 . 05 and allowing three cells divisions in fresh doxycycline containing media . ydj1 deletion strain ( VCY010 , MATα; his3Δ1; leu2Δ0; lys2Δ0; ura3Δ0; ydj1Δ::kanMX4 ) was derived from BY4742 . HeLa cells were obtained from American Type Culture Collection ( ATCC-CCL2; RRID:CVCL_0030 ) were grown in DMEM ( Gibco , Thermo Fisher Scientific , UK ) supplemented with 10% ( v/v ) FBS at 37°C in 5% CO2 . Mycoplasma contamination of the HeLa cell culture was tested negative with LookOut Mycoplasm Detection kit ( Sigma-Aldrich; MP0035 , MO , USA ) . Plasmids pcDNA5/FRT/TO V5-DNAJA2 , V5-DNAJB1 , V5-DNAJB2a and V5-DNAJB8 were kind gifts from Harm Kampinga ( University of Groningen ) . Plasmid transfections were carried according to standard protocols using Lipofectamine 2000 ( Thermo Fisher Scientific , UK ) . siRNA transfections were performed according to standard protocols using DharmaFECT one transfection reagent ( Dharmacon , CO , USA ) . HeLa cells were transfected with 50 μM siRNA smartpools ( Dharmacon onTarget Plus ) against DNAJA2 ( Dharmacon , L-012104–01 ) , DNAJB1 ( Dharmacon , L-012735–01 ) or scrambled non-targeting siRNA ( Dharmacon , D-001810–10 ) for 72 hr . Proximity ligation assay in bacteria: Log and stationary phase ( grown for 18 hr ) E . coli cells grown in LB medium at 30°C were fixed by adding ice cold 99% methanol and incubating at −20°C for 30 min . The fixed cells were attached to Poly-L-Lysine coated slides and treated with Lysozyme solution ( 2 mg/ml Lysozyme , 25 mM Tris HCl pH 8 . 0 , 50 mM glucose , 10 mM EDTA ) at room temperature for 30 min . The cells were then washed 3x in 100 ml PBST ( 140 mM NaCl , 2 mM KCl , 8 mM K2HPO4 , 1 . 5 mM KH2PO4 , 0 . 05% tween20 ) . The cells were treated with 99% methanol followed by an acetone wash . Methanol fixation quenched YFP and mCherry fluorescence allowing us to use 561 nm solid-state laser to specifically image signal from DUOLINK fluorophore ( orange kit ) that hybridize to amplified PCR product . The air-dried cells were then subjected to DUOLINK blocking , antibody treatment ( 1° antibody dilution , 1:300 ) , ligation , DNA amplification and mounting according to manufacturer’s guidelines ( Sigma-Aldrich ) . Proximity ligation assay in yeast: Log phase S . cerevisiae cells grown in YPD medium at 30°C were fixed with 4% para-formaldehyde for 15 min at room temperature and washed 2x with 100 mM KPO4 pH 6 . 5 buffer and 1x with wash buffer ( 1 . 2 M Sorbitol in 100 mM KPO4 pH 6 . 5 ) . Cell walls were digested with Zymolase solution ( 500 µg/ml Zymolase 100T , 1 . 2 M Sorbitol , 100 mM KPO4 pH 6 . 5 , 20 mM 2-Mercaptoethanol ) at 30°C for 20 min . The resulting spheroplasts were washed 3X with wash buffer and attached to Poly-L-Lysine coated slides . The attached spheroplasts were then washed 3x with permeabilizing solution ( 1% TritonX100 in 100 mM KPO4 pH 6 . 5 ) . DUOLINK blocking , antibody treatment ( 1° antibody dilution , 1:300 ) , ligation , DNA amplification and mounting steps were carried out according to manufacturer’s guidelines ( Sigma-Aldrich ) . Proximity ligation assay in Hela cells: HeLa cells were grown in DMEM ( Gibco ) supplemented with 10% ( v/v ) FBS at 37°C in 5% CO2 . Cells were plated at a density of 2 × 104 cells/well in poly-lysine coated 10-well diagnostic slides ( Thermo scientific , MA , USA ) and incubated for 24 hr . Cells were fixed with 4% para-formaldehyde in PBS and the proximity ligation assay was carried out according to DUOLINK manufacturer’s guideline for mammalian cells ( Sigma-Aldrich ) . A 1:150 1° antibody dilution was used . Confocal microscopy was performed on a LSM 780 system ( Carl Zeiss , Germany ) . Images of HeLa cells were taken with 20x/0 . 8 NA Plan Apochromat objective ( Carl Zeiss ) and identical acquisition settings with a pinhole of approx . one airy unit . A 63x/1 . 4 NA Plan Apochromat objective ( Carl Zeiss ) was used for yeast and bacterial cell imaging . DNA-stained DAPI was excited with a 405 nm pulsed diode laser and the DUOLINK signal was excited with a 561 nm solid-state laser . Commercially available antibodies against DNAJA2 ( rabbit monoclonal ) , DNAJB1 ( mouse monoclonal ) , V5 tag ( mouse monoclonal ) , GAPDH ( mouse monoclonal ) , Pgk1 ( mouse monoclonal ) and mCherry ( mouse monoclonal ) were obtained from Abcam ( UK ) ( ab157216; RRID:AB_2650527 , Enzo life sciences ( NY , USA ) ( ADI-SPA-450-E; RRID:AB_10621843 ) , Invitrogen ( CA , USA ) ( R960-25; RRID:AB_2556564 ) , Sigma ( G8795; RRID:AB_1078991 ) , Invitrogen ( 459250; RRID:AB_2532235 ) and Abcam ( ab125096; RRID:AB_11133266 ) , respectively . Anti-mouse Ydj1 ( SMC-150; RRID:AB_2570364 ) and anti-rabbit Sis1 ( COP-080051; RRID:AB_10709957 ) were obtained from StressMarq Biosciences Inc . ( Canada ) , and Cosmo Bio Co . ( Japan ) , respectively . Antibody against YFP ( rabbit polyclonal; RRID: AB_2650530 ) was generated in the laboratory . Anti-DnaK ( rabbit polyclonal; RRID:AB_2650528 ) and anti-human Hsp/Hsc70 ( rabbit polyclonal; RRID:AB_2650529 ) antibodies were a kind gift from Matthias Mayer ( University of Heidelberg ) . Western blot analysis was carried out using standard methodologies . Alkaline phosphatase conjugated anti-rabbit ( AP-1000; RRID:AB_2336194 ) and anti-mouse ( AP-2000; RRID:AB_2336173 ) IgG ( H+L ) antibodies from Vector Laboratories ( CA , USA ) were used as secondary antibodies . The detection was carried out with ECF ( GE Healthcare , IL , USA ) . Luciferase refolding-only , luciferase disaggregation/refolding and size exclusion chromatography were performed as previously described ( Nillegoda et al . , 2015 ) . In brief , protein aggregates were generated by heating 25 nM luciferase ( final concentration set to 20 nM ) with 125 nM sHSP26 at 45°C for 15 min in HKM buffer ( 50 mM Hepes-KOH pH 7 . 5 , 50 mM KCl , 5 mM MgCl2 , 2 mM DTT , 2 mM ATP pH 7 . 0 , 10 μM BSA ) . Denatured monomeric luciferase was obtained by heating 20 nM luciferase with 100 nM sHSP26 at 42°C for 10 min in HKM buffer containing chaperones of the indicated disaggregase system . Following concentrations of disaggregase components were used in disaggregation/refolding experiments . Bacterial: 750 nM ClpB , 750 nM DnaK , 250 nM J-protein ( total ) and 75 nM GrpE . Yeast ( non-saturating ) : 750 nM Hsp104 , 750 nM Ssa1 , 250 nM J-protein ( total ) and 38 nM Sse1 . Yeast ( saturating ) : 2 µM Hsp104 , 2 µM Ssa1 , 667 nM J-protein ( total ) and 100 nM Sse1 . Human: 750 nM HSPA8 , 250 nM J-protein ( total ) and 38 nM HSPH2 . DNAJA2/ DNAJA2RRR was labeled with ReAsH and DNAJB1 with Alexa Fluor 488 as described before ( Nillegoda et al . , 2015 ) . FRET measurements were performed for the following FRET pairs: CTD-labeled DNAJB1 together with J-domain-labeled DNAJA2/ DNAJA2RRR . Emission spectra were recorded on Jasco FP750 spectrofluorimeter at 30°C and quenching of donor fluorescence ( Alexa Fluor 488 ) was quantified at 517 nm and expressed as percentage of donor fluorescence in the absence of acceptor . Human J-proteins were mixed at 0 . 1 µM DNAJB1 and 1 µM DNAJA2/ DNAJA2RRR in 25 mM HEPES pH 7 . 5 , 50 mM KCl , MgCl2 , equilibrated for 15 min at 30°C . For competition measurements , 5 µM ( 5-fold excess relative to acceptor protein ) of unlabeled bacterial ( DnaJ , CbpA ) , yeast ( Ydj1 , Sis1 ) , human ( DNAJB1 , DNAJB2a , DNAJB8 , DNAJB1RRR ) and chimeric J-proteins were added to the aforementioned labeled J-protein pairs and equilibrated for 15 min at 30°C before fluorescence measurements . Experiments were performed in triplicate . | All cells must maintain their proteins in a correctly folded shape to survive . The task of sustaining a healthy set of proteins has increased with the rise of complex life from prokaryotes ( such as bacteria ) that form simple single-celled organisms to eukaryotes ( such as yeast , plants and multicellular animals ) . As a result of organisms ageing or acquiring genetic mutations , or under stressful conditions such as high temperature , proteins can lose their normal shape and clump together to form “aggregates” . These aggregates are potentially toxic to cells and have been linked to many human diseases including neurodegeneration and cancer . Cells contain molecular machines that help break down aggregates and subsequently recycle or rescue trapped proteins . Some of these machines are based around a protein called Hsp70 , which can perform a wide range of protein folding processes . So-called J-proteins help Hsp70 to select aggregates to be targeted for break down . It used to be thought that different classes of J-proteins interacted with Hsp70 separately . However , in 2015 , researchers showed that in humans , two different classes of J-proteins can bind to each other to form a “complex” , which has distinct aggregate selection properties . Now , Nillegoda et al . – including several of the researchers involved in the 2015 study – have examined the evolutionary history of these J-protein complexes . This revealed that different classes ( A and B ) of J-proteins first cooperated after prokaryotes and eukaryotes diverged from each other . In particular , the molecular machinery that breaks down aggregates in yeast cells – but not the machinery found in bacteria – depends on complexes formed from the two classes of J-proteins . Further investigation revealed that in humans , J-proteins have structural features that ensure they pair up correctly to perform unique activities . Furthermore , Nillegoda et al . suggest that cooperation between J-proteins may have enabled organisms such as humans – which contain over 40 distinct J-proteins – to carry out further specialized protein-folding tasks that do not occur in prokaryotes . Overall , the findings presented by Nillegoda et al . reveal another important layer to protein quality control in eukaryotic cells . The next step is to understand the possible roles of different J-protein complexes play in J-protein associated cellular protein quality control processes such as preventing protein aggregation , refolding or recycling abnormal proteins . This knowledge could ultimately be used to develop treatments for diseases and disorders in which protein aggregates form . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2017 | Evolution of an intricate J-protein network driving protein disaggregation in eukaryotes |
The assembly of the preinitiation complex ( PIC ) occurs upstream of the +1 nucleosome which , in yeast , obstructs the transcription start site and is frequently assembled with the histone variant H2A . Z . To understand the contribution of the transcription machinery in the disassembly of the +1 H2A . Z nucleosome , conditional mutants were used to block PIC assembly . A quantitative ChIP-seq approach , which allows detection of global occupancy change , was employed to measure H2A . Z occupancy . Blocking PIC assembly resulted in promoter-specific H2A . Z accumulation , indicating that the PIC is required to evict H2A . Z . By contrast , H2A . Z eviction was unaffected upon depletion of INO80 , a remodeler previously reported to displace nucleosomal H2A . Z . Robust PIC-dependent H2A . Z eviction was observed at active and infrequently transcribed genes , indicating that constitutive H2A . Z turnover is a general phenomenon . Finally , sites with strong H2A . Z turnover precisely mark transcript starts , providing a new metric for identifying cryptic and alternative sites of initiation .
The regulation of chromatin structure and its dynamics is integral to the control of gene expression in eukaryotes . The protein core of a canonical nucleosome , modular in nature , consists of a tetramer of histones H3 and H4 [indicated as ( H3-H4 ) 2] and two dimers of histones H2A and H2B ( indicated as H2A-H2B ) ( Arents et al . , 1991 ) . Nucleosomal DNA , 147-bp in length , wraps around the octameric histone core in 1 . 7 left-handed turns making 3 minor groove contacts with each histone pair and additional contacts near the entry-exit sites with H3 ( Luger et al . , 1997 ) . Repeating units of nucleosomes are organized along the genomic DNA in a non-random fashion with nucleosome-depleted regions ( NDRs ) overlapping key regulatory elements , such as promoters and replication origins ( Eaton et al . , 2010; Yuan et al . , 2005 ) . Nucleosome-repelling DNA sequences and ATP-dependent remodeling activities contribute to NDR formation ( Kaplan et al . , 2009; Zhang et al . , 2011 ) . Like a molecular sieve , chromatin blocks non-specific protein-DNA interactions and allows localized assembly of DNA binding factors , such as the general transcription factors ( GTFs ) and RNA polymerase ( Pol ) II , at the NDRs ( Rhee and Pugh , 2012 ) . Mutants that perturb the native nucleosome organization can lead to transcriptional derepression and initiation from aberrant start sites ( Han and Grunstein , 1988; Kaplan et al . , 2003; Whitehouse et al . , 2007 ) . How the transcription machinery engages the nucleosome in and around a promoter and how these nucleosomes are mobilized at different stages of transcription are important questions related to the mechanism of transcriptional control . The GTFs and Pol II assemble on an 80-to-200-basepair NDR to form the ‘closed’ preinitiation complex ( PIC ) ( Rhee and Pugh , 2012 ) . The nucleosome immediately downstream of the NDR is termed the +1 nucleosome ( Albert et al . , 2007 ) . In Saccharomyces cerevisiae , the +1 nucleosome covers the transcription start site ( TSS ) of most genes . Therefore , it is expected that at some point during the transcription process the +1 nucleosome must be disassembled . It is currently unclear at what stage of transcription disassembly of the +1 nucleosome occurs . It remains possible that chromatin remodeling enzymes are required to remove the +1 nucleosome before Pol II can engage the TSS . For other promoters where the +1 nucleosome covers the TATA element and for those with an untraditional nucleosome structure called the 'fragile nucleosome' within the NDR , histone eviction likely precedes and regulates PIC assembly ( Kubik et al . , 2015; Rhee and Pugh , 2012 ) . In metazoans , where the TSS is further upstream of the +1 nucleosome than in yeast ( Schones et al . , 2008 ) , the +1 nucleosome stalls elongation ( Weber et al . , 2014 ) . In all cases , the disassembly of these promoter-proximal nucleosomes is likely a regulatory barrier for full-length transcript synthesis ( Churchman and Weissman , 2011; Weber et al . , 2014 ) . The promoter-proximal nucleosomes at the +1 position and to a lesser extent the -1 position ( upstream of the NDR ) are enriched for the histone variant H2A . Z ( Albert et al . , 2007; Raisner et al . , 2005 ) . Together with the histone variant H3 . 3 , H2A . Z forms nucleosomes that are labile in high salt in vitro ( Jin and Felsenfeld , 2007; Zhang et al . , 2005 ) . In yeast , where the major H3 is similar to the H3 . 3 isoform in metazoans , H2A . Z is preferentially evicted from promoters during gene activation over H2A ( Santisteban et al . , 2000; Zhang et al . , 2005; Venters et al . , 2011 ) . Although mutants of HTZ1 ( the gene that encodes H2A . Z in yeast ) are viable and exhibited only minor defects in steady-state mRNA levels , H2A . Z is required for rapid transcriptional activation ( Dhillon et al . , 2006; Halley et al . , 2010; Mizuguchi et al . , 2004; Santisteban et al . , 2000; Zhang et al . , 2005 ) . These findings suggest that H2A . Z nucleosomes are predisposed for disassembly to allow for a rapid transcriptional response . What drives H2A . Z nucleosome disassembly in vivo will be the focus of this study . The incorporation of H2A . Z into nucleosomes is catalyzed by the ATP-dependent chromatin remodeling complex SWR1 ( Mizuguchi et al . , 2004 ) . The ~1 megadalton SWR1 complex comprises the catalytic core protein Swr1 , a member of the Swi2/Snf2-related ATPase , plus 13 other polypeptides ( Kobor et al . , 2004; Krogan et al . , 2003; Mizuguchi et al . , 2004 ) . SWR1 is targeted to promoters by its intrinsic affinity for the NDR and promoter-specific histone acetylation ( Raisner et al . , 2005; Ranjan et al . , 2013 ) . It catalyzes a histone replacement reaction that involves the coupled removal of a nucleosomal H2A-H2B dimer with the insertion of an H2A . Z-H2B dimer that is delivered to the enzyme by one of several histone chaperones , including Nap1 , Chz1 , and FACT ( Luk et al . , 2007 , 2010; Mizuguchi et al . , 2004 ) . The two H2A-H2B dimers in a homotypic H2A ( AA ) nucleosome are replaced sequentially , first generating the heterotypic H2A/H2A . Z ( AZ ) nucleosome as an intermediate ( Figure 1—figure supplement 1A , step I-a ) and the homotypic H2A . Z ( ZZ ) nucleosome as the final product ( Figure 1—figure supplement 1A , step I-b ) ( Luk et al . , 2010 ) . The ( H3-H4 ) 2 tetramer remains associated with the DNA before and after each step of the replacement reaction as no net loss of nucleosomal species occurs during the histone replacement reaction in vitro ( Luk et al . , 2010 ) . While it is well established that H2A . Z is enriched at the promoter-proximal nucleosomes , it is underappreciated that substantial amount of H2A is also present at these sites ( Luk et al . , 2010 ) . Experiments using sequential ChIP and tiling microarray analysis have demonstrated that in a population of G1-arrested cells , nucleosomes in the AA , AZ and ZZ configurations can all be detected at the +1 positions of most promoters ( Luk et al . , 2010 ) . This observation suggests that the SWR1 reaction that generates ZZ nucleosomes is opposed by a pathway ( s ) that converts ZZ nucleosomes back to the AA state in a replication-independent manner . Consistent with this dynamic model , rapid , constitutive H3 turnover is observed at most +1 nucleosomes ( Dion et al . , 2007 ) . Since the ( H3-H4 ) 2 tetramer is at the center of the histone core ( Luger et al . , 1997 ) , H3 turnover implies complete disassembly of the ZZ nucleosome ( Figure 1—figure supplement 1A , step II ) . Reassembly likely leads to the formation of the canonical AA nucleosomes as H2A is ~10 times more abundant than H2A . Z and SWR1 does not assemble H2A . Z nucleosomes de novo on DNA ( Luk et al . , 2010; West and Bonner , 1980 ) ( Figure 1—figure supplement 1A , step III ) . The ATP-dependent remodeling complex INO80 has been reported to mediate the reverse replacement reaction ( in which a nucleosomal H2A . Z-H2B dimer is replaced by a free H2A-H2B dimer ) ( Figure 1—figure supplement 1A , steps I-c and I-d ) ( Papamichos-Chronakis et al . , 2011 ) . Analysis of H2A . Z ChIP followed by microarray ( ChIP-chip ) showed that H2A . Z in ino80∆ cells redistributes from promoters to gene body regions as compared to wild-type cells ( Papamichos-Chronakis et al . , 2011 ) . Deletion mutant of ARP5 , a gene encoding a critical component of the INO80 complex , exhibited global H2A . Z accumulation especially around the promoters as demonstrated by the ChIP-exo technique ( Yen et al . , 2013 ) . However , the ChIP-chip data of a more recent study disagreed , showing that the genome-wide H2A . Z occupancy was similar in ino80∆ and wild-type cells ( Jeronimo et al . , 2015 ) . Therefore , what contributes to the conversion of ZZ nucleosomes to the AA state remains controversial . This study addresses the hypothesis that the transcription machinery is a major driving force of the disassembly of the +1 H2A . Z nucleosomes . We used the anchor away approach to deplete components of the transcription machinery ( Haruki et al . , 2008 ) and a quantitative ChIP-seq approach to probe changes in H2A . Z occupancy genome-wide at single-basepair resolution . We observed reciprocal increase of H2A . Z and decrease of H2A genome-wide after depletion of the PIC . By contrast , nuclear depletion of Ino80 did not cause global H2A . Z accumulation . These findings suggest that the assembly of the Pol II transcription machinery and/or its activity contribute ( s ) to the constitutive turnover of H2A . Z at yeast promoters .
The genomic H2A . Z level at any given promoter in a cell population is in a steady state that is maintained by the deposition mediated by SWR1 and the eviction mediated by putative chromatin remodeling pathway ( s ) ( Luk et al . , 2010 ) . To test the contribution of the transcription machinery in H2A . Z eviction genome-wide , conditional yeast mutants were used to block the assembly of the PIC . If eviction of H2A . Z-containing nucleosomes is blocked , H2A . Z levels are expected to increase and H2A levels to decrease , as the SWR1 complex continues to replace nucleosomal H2A-H2B with H2A . Z-H2B dimers ( Figure 1—figure supplement 1B ) . To block PIC assembly , TBP was conditionally depleted from the nucleus using the anchor-away approach with SPT15 , the gene that encodes TBP ( Haruki et al . , 2008 ) . When fused to the FKBP12-rapamycin-binding domain ( FRB ) , TBP-FRB can be dragged out of the nucleus in a rapamycin-dependent manner by the FKBP12 tag on the pre-ribosome ( Haruki et al . , 2008 ) . Previous studies have shown yeast cells expressing SPT15-FRB ( hereafter referred to as TBP-FRB ) rapidly depleted TBP from Pol II promoters , blocked Pol II recruitment and shut off transcription within 1 hr of rapamycin treatment ( Grimaldi et al . , 2014; Haruki et al . , 2008; Wong et al . , 2014 ) . In our experiments , TBP-FRB relocalized to the cytoplasm with similar kinetics ( Figure 1—figure supplement 2 ) . To measure the relative occupancy of H2A . Z and H2A genome-wide , deep sequencing was combined with quantitative ChIP of H2A . Z ( Luk et al . , 2010 ) . Specifically , chromatin from fixed haploid yeast cells expressing a 2xFLAG-epitope-tagged HTZ1 gene was digested with micrococcal nuclease ( MNase ) to generate mononucleosomes . H2A . Z-containing nucleosomes were separated from the canonical AA nucleosomes by binding to anti-FLAG affinity gel followed by elution using FLAG peptides . Given that H2A . Z is the only H2A variant in budding yeast and the IP efficiency was consistently over 80% ( Figure 1—figure supplement 3 ) , the flow-through ( FT ) of the IP reaction was highly enriched for the homotypic AA nucleosomes ( referred to as the H2A nucleosomes hereafter ) and the IP fraction was enriched for the heterotypic AZ and homotypic ZZ nucleosomes ( referred collectively to as the H2A . Z nucleosomes hereafter ) ( Luk et al . , 2010 ) . The DNA extracted from both the FT and IP fractions , as well as the DNA from the input nucleosomes used for the IP , was mapped by deep sequencing . Forty-four reference regions ( called no-Z-zones , covering 152 , 021 bp and 1 , 161 nucleosomes ) with very low H2A . Z but high H2A occupancy were manually chosen and were used to normalize the FT fraction data to the input ( Figure 1—source data 1 and Figure 1—figure supplement 4A ) . Depletion of signal in the normalized FT fraction data relative to the input represents the immuno-depleted H2A . Z associated DNA . The amplitude of the H2A . Z data was adjusted using a curve fitting algorithm such that the sum of the normalized profiles of the H2A . Z fraction and FT fraction surrounding the +1 nucleosome region ( N = 4 , 738 ) equals , to a first approximation , the input profile ( Luk et al . , 2010 ) ( Figure 1—figure supplement 4B ) . This approach , called quantitative ChIP-seq or qChIP-seq , is similar to ChIP-coupled quantitative PCR ( ChIP-qPCR ) in that occupancy is reported in relation to the input of each ChIP reaction but is genome-wide . Unlike standard ChIP-seq , which typically involves normalization by equalizing the read counts of ChIP samples and reports ChIP signals in relation to some background or threshold ( e . g . mean of genomic ChIP signal ) that may vary among samples , the qChIP-seq method allows more quantitative comparison between samples , especially when a global change in H2A . Z occupancy is expected . Using qChIP-seq , the coverage of nucleosomal DNA was determined for the H2A . Z ( green ) , the FT ( indicated as H2A in red ) , and the input ( gray ) fractions in the TBP-FRB haploid strain and the isogenic untagged control strain ( no FRB ) . Under permissive conditions ( no RAP ) , H2A . Z was most prevalent at the +1 nucleosomal positions ( marked by ‘+’ ) , less at the -1 positions and progressively less towards the 3’ end of genes in agreement with published results ( Albert et al . , 2007 ) ( Figures 1A and B , top , no RAP ) . This phenomenon is further demonstrated when the H2A . Z and input profiles were compiled at the dyads of 4 , 738 +1 nucleosomes ( Figures 1C and D , top left ) . The FT fraction is a good representation of nucleosomal H2A levels , at least qualitatively , for two reasons . First , although the FT fraction may contain unstable histone-DNA complexes that are devoid of H2A , e . g . tetrasomes , or non-histone complexes ( Reja et al . , 2015 ) , these structures are highly depleted in these experiments as the chromatin was subjected to extensive MNase digestion ( unless indicated otherwise , below ) . Second , an earlier study that used an anti-H2A antibody to ChIP the FT fraction followed by high-resolution tiling microarray analysis produced an AA nucleosome profile that is highly similar to the FT profiles of our current experiments ( Luk et al . , 2010 ) ( Figures 1C and D , bottom left ) . Therefore , the qChIP-seq data are consistent with previous conclusions that nucleosomes enriched for H2A dominate in coding regions and that , while depleted at the +1 positions , a substantial amount of H2A nucleosomes remains ( Figures 1C and D , bottom left ) ( Luk et al . , 2010 ) . The FT profile is referred to as the H2A profile hereafter . 10 . 7554/eLife . 14243 . 003Figure 1 . H2A . Z nucleosome occupancy determined by qChIP-seq in the TBP-FRB and the untagged control ( no FRB ) strains with and without rapamycin treatment . ( A , B ) Sequencing tag coverage of H2A . Z ( in green ) , FT ( indicated as H2A in red ) , and input ( in gray ) at a representative genomic region on chromosome III . Blue traces indicate Δ ( Z/A ) , the H2A . Z-to-H2A ( Z/A ) ratio with rapamycin treatment ( +RAP ) minus that without treatment ( no RAP ) . Plus signs and arrowheads mark +1 nucleosomes and transcription start sites , respectively . ( C , D ) Compiled read counts ( midpoints ) of H2A . Z ( green ) , H2A ( red ) , and input ( gray ) nucleosomes were centered around the dyad of 4 , 738 +1 nucleosomes . ( E ) Verification of the qChIP-seq data by qPCR using primer pairs covering the +1 nucleosomes of the indicated promoters and regions within the indicated open reading frames ( ORF ) . ( F ) Scatter plots and histograms showing the change in ( H2A . Z/input ) of the +1 and reference nucleosomes as a function of endogenous H2A . Z level before rapamycin treatment . The ( H2A . Z/input ) value represents the ratio of H2A . Z tag coverage over input tag coverage within a 120 bp region around the nucleosome dyad . Open black circles mark the +1 nucleosomes ( Rhee et al . , 2014 ) . Orange dots mark the reference nucleosomes used for normalization . Dotted lines represent the upper and lower thresholds for significant change in H2A . Z levels , which are defined as two standard deviations from the median of the no FRB control . The percentages of data points within and outside the threshold regions are indicated . The qChIP-seq and qPCR data represent averages of >3 independent ChIP reactions ( technical replicates ) of two independent cultures ( biological replicates ) . The error bars in ( E ) represent standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 00310 . 7554/eLife . 14243 . 004Figure 1—source data 1 . Reference regions or 'no-Z-zones' used for normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 00410 . 7554/eLife . 14243 . 005Figure 1—source data 2 . Normalization factors . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 00510 . 7554/eLife . 14243 . 006Figure 1—source data 3 . Nucleosome tag profiles of TBP-FRB around the +1 dyads before normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 00610 . 7554/eLife . 14243 . 007Figure 1—source data 4 . Nucleosome tag profiles of no FRB around the +1 dyads before normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 00710 . 7554/eLife . 14243 . 008Figure 1—source data 5 . Nucleosome tag profiles of RPB1-FRB around the +1 dyads before normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 00810 . 7554/eLife . 14243 . 009Figure 1—source data 6 . Average tag coverage around the +1 dyads of TBP-FRB . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 00910 . 7554/eLife . 14243 . 010Figure 1—source data 7 . Average tag coverage around the +1 dyads of no FRB . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 01010 . 7554/eLife . 14243 . 011Figure 1—source data 8 . Average tag coverage around the +1 dyads of RPB1-FRB . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 01110 . 7554/eLife . 14243 . 012Figure 1—source data 9 . Source data for Figure 1E . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 01210 . 7554/eLife . 14243 . 013Figure 1—figure supplement 1 . A proposed model to account for the constitutive histone turnover at yeast promoters . ( A ) The conversion of a +1 H2A nucleosome to an H2A . Z nucleosome by SWR1 is followed by nucleosome disassembly . Histone chaperones are omitted for simplicity . NDR: nucleosome-depleted region . ‘+1’: +1 nucleosome . TSS: transcription start site . H2A . Z-H2B dimers are in green , H2A-H2B dimers in red , and ( H3-H4 ) 2 tetramers in gray . ( Step I-a ) SWR1 replaces one nucleosomal H2A-H2B dimer in a homotypic 'AA' +1 nucleosome with a dimer of H2A . Z-H2B to generate a heterotypic 'AZ' nucleosome . ( Step I-b ) SWR1 replaces the second H2A-H2B dimer with H2A . Z-H2B generating a homotypic 'ZZ' nucleosome . ( Steps I-c and I-d ) The reported reverse replacement reaction by the INO80 complex ( Papamichos-Chronakis et al . , 2011 ) . ( Step II ) The ZZ nucleosome is disassembled by an unknown mechanism indicated by a question mark . ( Step III ) A canonical AA nucleosome is reassembled . ( B ) Same as ( A ) except step II is blocked . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 01310 . 7554/eLife . 14243 . 014Figure 1—figure supplement 2 . Fluorescence microscopy of yeast expressing TBP-FRB-GFP with and without rapamycin treatment . At the indicated times after the addition of 1 µg/mL rapamycin ( RAP ) , cells were fixed with 4% formaldehyde for 5 min , washed with PBS and stained with DAPI before imaged under a Zeiss Axio Observer Z1 microscope . DAPI: red; GFP: green; DIC: differential interference contrast optics . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 01410 . 7554/eLife . 14243 . 015Figure 1—figure supplement 3 . Immunoblot analysis to control for anti-FLAG IP efficiency . The nucleosomes used in the IP reactions were prepared from formaldehyde fixed haploid cells expressing the 2xFLAG-epitope-tagged Htz1 ( H2A . ZFL ) . Nucleosomes were released from crude chromatin by MNase digestion . After centrifugation , the supernatant was filtered through a low-binding PVDF membrane . A sample of the soluble nucleosome was collected as the input ( IN ) of the IP reaction . After incubation with anti-FLAG agarose , an equivalent amount of the flow-through fraction ( FT ) was collected . Both the IN and the FT were heated at 95˚C in SDS-PAGE sample buffer for 30 min to allow decrosslinking before analysis with SDS-PAGE and anti-H2A . Z immunoblotting . Serially diluted samples of the IN fraction were loaded next to the FT lane for comparison . The number above each lane indicates the relative sample volume applied to the gel . Western signals were developed by the ECL Prime reagent ( GE Healthcare , Pittsburgh , PA ) and imaged by the ImageQuant FLA4010 imaging system ( GE Healthcare ) . Representative western blots for individual IP reactions of both no RAP and + RAP samples are shown for the no FRB control in ( A ) , TBP-FRB in ( B ) , and RPB1-FRB in ( C ) , and INO80-FRB in ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 01510 . 7554/eLife . 14243 . 016Figure 1—figure supplement 4 . The strategy used to normalize relative H2A and H2A . Z occupancy . ( A ) Examples of the no-Z-zones used in the normalization of the H2A profile . ( B ) An overview of the ‘TAZ’ normalization approach used to rescale the H2A . Z and H2A nucleosomal profiles . The compiled tag counts ( midpoints ) in relation to the distance from the +1 nucleosomal dyads ( n = 4 , 738 ) were plotted for H2A ( A ) , H2A . Z ( Z ) and input ( T ) nucleosomes , which are in red , green , and black , respectively . To normalize the raw data , the scaling factors , m and n , were applied to the H2A and H2A . Z profiles respectively . The scaling factor m was determined for each IP reaction by dividing the tag count within the no-Z-zones of the input fraction by that of the H2A fraction . The scaling factor n was subsequently determined by a curve-fitting algorithm such that the sum of the resulting profiles ( m × H2A + n × H2A . Z ) equals , to a first approximation , the input nucleosomal profile . ( C ) The informatics pipeline of TAZ normalization . The sequencing tags of the H2A , H2A . Z , and input fractions ( in FASTQ file format ) were mapped to the yeast genome by bowtie ( Langmead et al . , 2009 ) . Mapped reads were either presented as tag coverage ( density covered by paired-end reads ) or tag counts ( density of mid-points of paired-end reads ) along the yeast genome . Bowtie was also used to determine the tag counts within the no-Z-zones in the input and the H2A fractions . The TAZ curve-fitting algorithm is provided in a Python script , which requires four inputs: the scaling factor m and the compiled nucleosome profiles of H2A , H2A . Z and input around the +1 dyads . The algorithm generates the scaling factor n for normalization of the H2A . Z profile . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 01610 . 7554/eLife . 14243 . 017Figure 1—figure supplement 5 . Heatmaps of H2A . Z , H2A and input nucleosomes for the TBP-FRB , no FRB , and RPB1-FRB strains . Nucleosome tag coverage of H2A . Z ( in green ) , H2A ( in red ) and input ( in white ) were plotted around the +1 dyads ( n = 4 , 182 ) for TBP-FRB in ( A ) , no FRB in ( B ) and RPB1-FRB in ( C ) before and after rapamycin treatment . Genes were sorted by TFIIB/Sua7 occupancy ( Rhee and Pugh , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 01710 . 7554/eLife . 14243 . 018Figure 1—figure supplement 6 . Concordance of relative H2A . Z occupancy between biological replicates . The ( H2A . Z/input ) signals of the biological replicates at the +1 nucleosomal positions ( Rhee et al . , 2014 , n = 4 , 738 ) were plotted against each other for TBP-FRB , no RAP in ( A ) , TBP-FRB , + RAP in ( B ) , no FRB , no RAP in ( C ) , and no FRB , + RAP in ( D ) . The biological replicates represent independent yeast cultures . Dotted red lines: arbitrary reference perimeter set at -0 . 25 on both axes . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 01810 . 7554/eLife . 14243 . 019Figure 1—figure supplement 7 . Using ∆ ( Z/A ) as a parameter to identify +1 nucleosomes with PIC-dependent H2A . Z eviction . ( A , B ) Scatter plots and histograms showing the ∆ ( Z/A ) values of +1 and reference nucleosomes as a function of endogenous H2A . Z level ( before rapamycin treatment ) for the TBP-FRB and no FRB strains , respectively . Open black circles mark the +1 nucleosomes . Orange dots mark the reference nucleosomes used for normalization . Dotted horizontal lines represent the upper and lower thresholds for significant change in ∆ ( Z/A ) , which are two standard deviations from the median of the no FRB control . The percentages of data points within and outside the threshold regions are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 01910 . 7554/eLife . 14243 . 020Figure 1—figure supplement 8 . Relative H2A . Z occupancy before and after Rpb1 depletion . ( A ) Sequencing tag coverage of H2A . Z ( in green ) , H2A ( in red ) , and input ( in gray ) at a representative genomic region on chromosome III . Blue traces indicate ∆ ( Z/A ) , the H2A . Z-to-H2A ( Z/A ) ratio with rapamycin treatment minus that without . Plus signs and arrowheads mark +1 nucleosomes and transcription start sites , respectively . ( B ) Compiled read counts ( midpoints ) of H2A . Z ( green ) , H2A ( red ) , and input ( gray ) nucleosomes were centered around the dyad of 4 , 738 +1 nucleosomes . ( C ) Scatter plots and histograms showing the change in ( H2A . Z/input ) of the +1 or reference nucleosomes as a function of endogenous H2A . Z level . Open black circles mark the +1 nucleosomes . Orange dots mark the reference nucleosomes . Dotted black lines are the upper and lower thresholds for significant change in H2A . Z levels . The data in ( A–C ) represent averages of two biological replicates . ( D ) Scatter plots showing the concordance of log ( H2A . Z/input ) at +1 nucleosomes of the biological replicates . Dotted red lines: arbitrary reference perimeter set at –0 . 25 on both axes . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 02010 . 7554/eLife . 14243 . 021Figure 1—figure supplement 9 . H2A . Z accumulation is not due to aberrant accumulation of the SWR1 complex . ( A ) ∆ ( Z/A ) around the +1 dyads after TBP depletion . Genes were sorted by endogenous SWR1 occupancy ( Venters and Pugh , 2009 ) . ( B ) Sequencing tag coverage of H2A . Z , H2A and input nucleosomes surrounding the SWR1 and FUN12 genes . ( C ) ChIP-qPCR analysis was performed using an antibody directed against Swr1 and primers targeting the promoters of SWR1 and FUN12 relative to a control region near TEL6R ( Ranjan et al . , 2013 ) . Swr1 enrichment represents the mean of a total of four independent ChIP reactions from two biological replicates . Error bars: standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 02110 . 7554/eLife . 14243 . 022Figure 1—figure supplement 10 . Input nucleosome occupancy at the +1 positions before and after TBP depletion . Left panel: The occupancy of +1 nucleosomes represents the read coverage within a 120 bp region around the +1 dyad ( n = 4 , 738 ) . The +1 occupancy associated with the top 3% most actively transcribing genes are in red ( Lipson et al . , 2009 ) . The red line and dotted black line represent the linear regression for the highlighted genes and all the genes , respectively . Right panel . Same as left except that the +1 nucleosomes associated with the bottom 3% least active genes are highlighted . Blue line is the corresponding linear regression . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 02210 . 7554/eLife . 14243 . 023Figure 1—figure supplement 11 . qPCR and H2A western analyses of the flow-through fractions . ( A ) qChIP-seq and qPCR quantification of the nucleosomal DNA in the flow-through ( FT ) relative to the input of the anti-FLAG ( H2A . Z ) IP reactions . The primer pairs for qPCR cover +1 nucleosomes of the indicated promoters and regions within the indicated open reading frames ( ORF ) that are known to be depleted for H2A . Z . Error bars for the qPCR represent the range of two biological replicates , whereas those for the qChIP-seq analysis represent the standard deviation of independent ChIP reactions of two biological samples . ( B ) Immunoblotting analysis of H2A in the FT and input fractions showing that the bulk of H2A remains in the FT after the anti-FLAG IP reaction . The protein samples were decrosslinked by heating before analyzed by SDS-PAGE and anti-H2A western blotting as described in Figure 1—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 023 Rapamycin treatment of TBP-FRB cells for 1 hr resulted in H2A . Z accumulation at the promoter-proximal nucleosomes of most genes , with a corresponding depletion of H2A signal ( compare + RAP and no RAP in Figures 1A and C and in Figure 1—figure supplement 5A ) . Rapamycin treatment alone did not cause significant change in histone dynamics as the untagged wild-type cells exhibited similar promoter-specific H2A . Z levels before and after rapamycin treatment ( Figures 1B and D , Figure 1—figure supplement 5B ) . To confirm that the increase of H2A . Z levels in the TBP-FRB strain was not an artifact of normalization , qPCR was employed to measure the immunoprecipitated DNA in the H2A . Z fraction relative to the input at three +1 nucleosome regions and two coding regions . Similar to the sequencing analysis , the qPCR experiments indicated an increase in H2A . Z nucleosomal DNA at +1 nucleosomes after depletion of TBP , with no change observed for DNA located within the open reading frames ( ORFs ) , confirming the robustness of the qChIP-seq approach ( Figure 1E ) . To evaluate the H2A . Z change at the +1 positions genome-wide , the change in ( H2A . Z/input ) ratios at 4 , 738 +1 nucleosomes after TBP depletion {i . e . ( H2A . Z/input ) [RAP] – ( H2A . Z/input ) [no RAP]} were plotted against the endogenous H2A . Z levels represented by the logarithmically transformed H2A . Z tag counts before rapamycin treatment ( Figure 1F ) . A threshold for significant change in ( H2A . Z/input ) was defined by two standard deviations above and below the median of the untagged control ( Figure 1F , dotted lines ) . The change in ( H2A . Z/input ) values of the reference nucleosomes used in normalization was plotted for comparison ( Figure 1F , orange dots ) . Fifty-six percent of +1 nucleosomes exhibited a significant increase in relative H2A . Z signal upon TBP depletion ( Figure 1F ) . The ( H2A . Z/input ) signals of the two biological replicates were highly reproducible ( Figure 1—figure supplement 6A–B ) . Note the global shift of data points towards the upper right quadrant in the +RAP data as compared to the no RAP sample , indicating a global increase of relative H2A . Z in both biological replicates ( compare Figure 1—figure supplement 6A–B ) . By contrast , similar shift of data points towards the upper right quadrant was not observed for the no FRB control ( Figure 1—figure supplement 6C–D ) . It is noteworthy that under the permissive condition , the TBP-FRB strain exhibited a higher endogenous H2A . Z occupancy compared to the untagged strain ( compare no RAP in Figures 1C and D ) . The difference likely reflects a partial defect of the TBP-FRB allele that has predisposed the cells to H2A . Z accumulation . But importantly , the measurement of change in the ( H2A . Z/input ) ratio ( Figure 1F ) normalizes any differences in the ground state ( no RAP ) and highlights the functional consequence of the depletion of TBP or other factors in question ( below ) . The change in ( H2A . Z/H2A ) ratios {i . e . ( H2A . Z/H2A ) [RAP] – ( H2A . Z/H2A ) [no RAP]} , which is referred to as ∆ ( Z/A ) hereafter , provides an even more sensitive , nonetheless non-linear , indicator of H2A . Z dynamics because of the antagonistic change in H2A . Z and H2A occupancy . As shown in Figure 1—figure supplement 7A , 72% +1 nucleosomes exhibited an increase in ∆ ( Z/A ) signal upon TBP depletion . When the ∆ ( Z/A ) values were plotted along the chromosome coordinates , this parameter clearly identified sites with strong H2A . Z dynamics ( Figure 1A , blue traces ) . The ∆ ( Z/A ) parameter will later be used to identify novel transcription start sites . To test if depletion of another component of the PIC could also lead to H2A . Z accumulation , we targeted Rpb1 , the largest subunit of Pol II , for nuclear depletion by anchor-away . The RBP1-FRB construct has previously been shown to effectively shut off transcription ( Haruki et al . , 2008 ) . Similar to the results of TBP-FRB removal , nuclear depletion of Rpb1-FRB led to strong H2A . Z accumulation and H2A depletion at most promoters ( Figure 1—figure supplement 8 and Figure 1—figure supplement 5C ) . One explanation for the increase in H2A . Z levels at the +1 promoter upon TBP depletion is that the balance between H2A . Z eviction and deposition of H2A . Z nucleosomes by SWR1 has been disrupted . SWR1 continues to convert H2A nucleosomes to the H2A . Z containing forms but with no eviction to restore these nucleosomes back to the AA state in the absence of the PIC . Consistent with this idea , sites with strong H2A . Z dynamics are more enriched for endogenous SWR1 ( Figure 1—figure supplement 9A ) . An alternative explanation for this phenomenon , however , is that TBP depletion does not block H2A . Z eviction but instead leads to recruitment of aberrantly high levels of SWR1 to promoters as the PICs dissociate . This model predicts that SWR1 levels should increase at promoters upon TBP depletion . To distinguish between these two possibilities , ChIP-qPCR was used to monitor the occupancy of the Swr1 subunit in the TBP-FRB strain at different times after rapamycin treatment . The promoter regions of SWR1 itself and FUN12 were chosen because these sites are known to be enriched for SWR1 ( Ranjan et al . , 2013; Yoshida et al . , 2010 ) and exhibited strong changes in H2A . Z dynamics upon TBP depletion ( Figure 1—figure supplement 9B ) . Rather than accumulating at the promoters of these two genes , SWR1 dissociated shortly after TBP-FRB depletion as demonstrated by the decrease in Swr1 ChIP signal ( Figure 1—figure supplement 9C ) . This observation supports the idea that the PIC is required to actively evict H2A . Z nucleosomes . In addition , the coincidental accumulation of H2A . Z and depletion of SWR1 suggest that endogenous SWR1 dissociates from promoters after depositing H2A . Z . This is in agreement with the in vitro observation that SWR1 has a lower affinity for the homotypic H2A . Z nucleosome product than the homotypic H2A nucleosome substrate ( Ranjan et al . , 2015 ) . Yeast promoters that are TFIID-enriched or -depleted are apparently regulated by different mechanisms ( Rhee and Pugh , 2012 ) . Although both types of promoters require TBP for transcription , TFIID-enriched genes are often TATA-less and associated with a more open NDR , whereas TFIID-depleted promoters generally contain a consensus TATA element and are more covered with nucleosomes and more enriched with the SAGA complex ( Basehoar et al . , 2004; Rhee and Pugh , 2012 ) . To see if the +1 nucleosomes at these two types of promoters exhibit differential H2A . Z dynamics , the occupancy of H2A , H2A . Z and input nucleosomes , and the corresponding ∆ ( Z/A ) values , before and after rapamycin treatment were sorted by TFIID enrichment and transcript abundance ( Lipson et al . , 2009; Rhee and Pugh , 2012 ) . As seen in Figure 2A , the +1 nucleosomes that exhibited stronger H2A . Z accumulation after TBP depletion are generally more enriched for TFIID . 10 . 7554/eLife . 14243 . 024Figure 2 . Change in nucleosomal H2A . Z and H2A levels near promoters before and after depletion of TBP . ( A ) Heatmaps showing the average normalized tag coverage of H2A . Z ( green ) , H2A ( red ) and input ( white ) , and the corresponding ∆ ( Z/A ) values ( yellow ) around the +1 dyads of the TBP-FRB strain with and without rapamycin treatment . Promoters enriched for the PIC ( based on Sua7 occupancy ) were grouped into high and low TFIID ( based on Taf1 occupancy ) and were then sorted by mRNA abundance ( n = 3 , 919 ) ( Lipson et al . , 2009; Rhee and Pugh , 2012 ) . The ribosomal protein ( RP ) genes ( n = 128 ) are 'zoomed in 4x' meaning that the line thickness is 4 times of the other genes . ( B ) Compiled nucleosome tag counts of H2A . Z ( green ) , H2A ( red ) , and input ( gray ) around the +1 dyads of the RP genes . ( C ) The +1 nucleosomes were grouped according to transcriptional frequency ( Holstege et al . , 1998 ) and the ( H2A . Z/input ) values were presented as box plots . Box: interquartile range ( IQR ) ; line in box: median; whiskers: range . The most active genes ( >50 mRNA/hr ) were sub-divided into two groups that are SAGA- or TFIID-enriched . The number of +1 nucleosomes in the transcriptional frequency groups <1 , 1–4 , 4–16 , 16–50 , and >50 mRNA/hr are 937 , 1932 , 1018 , 219 and 161 , respectively . There are 22 promoters in the SAGA >50 group and 130 in the TFIID >50 group . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 02410 . 7554/eLife . 14243 . 025Figure 2—figure supplement 1 . Normalized nucleosome tag coverage for H2A . Z , H2A , and input before and after the depletion of Rpb1 . ( A ) Heatmaps were plotted and sorted by TFIID/Taf1 levels as described in Figure 2A . ( B ) Compiled tag counts of the RP genes . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 025 The ribosomal protein ( RP ) genes are regulated by TFIID and are among the most highly transcribed ( Rhee and Pugh , 2012; Warner , 1999 ) . Their promoters are unusual in that the endogenous +1 nucleosomes are generally depleted for H2A . Z but relatively enriched for H2A ( Figure 2B , n = 128 ) . Upon depletion of TBP , relative H2A . Z accumulates dramatically at the +1 position ( Figure 2B , green ) . The data suggest that H2A . Z deposition occurs at the RP gene promoters but that H2A . Z nucleosomes are quickly removed by a mechanism that is dependent on TBP , leading to a low steady-state H2A . Z occupancy . Interestingly , Rpb1 depletion led to almost no change in H2A . Z occupancy at the +1 nucleosomes of RP genes ( Figure 2—figure supplement 1 ) . Since RP genes promoters are occupied by high level of PIC components ( indicated by the TFIIB marker Sua7 ) , partial PIC may remain bound after Rpb1 depletion . This suggests that H2A . Z removal at the RP genes requires TBP but not Pol II . To further understand the link between transcriptional activity and H2A . Z dynamics , the +1 H2A . Z occupancy of 4 , 267 promoters before and after TBP depletion was sorted and grouped by the transcriptional frequency of their downstream genes and compared by box plot analysis ( Figure 2C ) ( Holstege et al . , 1998 ) . Before TBP-depletion , the +1 nucleosomes associated with moderately or infrequently transcribing genes exhibited >40% ( median ) relative H2A . Z occupancy for genes with <16 mRNA/hr ( n = 3 , 887 ) and ~30% ( median ) for those with 16–50 mRNA/hr ( n = 219 ) . For the top 3% of most highly transcribing genes ( >50 mRNA/hr , n = 161 ) the steady-state H2A . Z occupancy is ~11% ( median ) before TBP-depletion . Upon rapamycin treatment , substantial increase of relative H2A . Z occupancy is observed in all groups ( Figure 2C ) . The increase is more dramatic in the most highly transcribing genes , suggesting that the low steady-state H2A . Z occupancy is due to strong transcriptional activity ( Figure 2C , >50 mRNA/hr ) . Interestingly , when these highly transcribing genes were sorted based on the enrichment of SAGA or TFIID ( Basehoar et al . , 2004 ) , H2A . Z accumulated to a smaller extent for the SAGA-enriched genes than for the TFIID-enriched genes ( Figure 2C , right ) . Therefore , unlike the strong TFIID-enriched promoters where low steady-state +1 H2A . Z level is due to robust PIC-dependent H2A . Z eviction activity , low H2A . Z at the strong SAGA-enriched promoters is due to either strong PIC-independent eviction or weak SWR1-mediated H2A . Z deposition or both . An alternative approach to monitor H2A . Z eviction is to conditionally block H2A . Z deposition and follow the depletion of H2A . Z . Since SWR1-mediated H2A . Z deposition requires Swc5 , a subunit of the SWR1 complex ( Wu et al . , 2005 ) , anchor-away was utilized to deplete Swc5 so as to conditionally block H2A . Z deposition . SWR1 activity of the SWC5-FRB strain was strongly inhibited after 30 min of rapamycin treatment as demonstrated by the robust loss of H2A . Z occupancy ( Figures 3A–B and Figure 3—figure supplement 1A–B ) . When the relative H2A . Z levels ( H2A . Z/input ) of the +1 nucleosomes were sorted and grouped by the transcriptional frequency of the downstream genes , almost background level of H2A . Z was observed in all groups , indicating that robust , constitutive H2A . Z eviction at the +1 position occurs at both active and infrequently transcribed genes ( Figure 3C ) . To further understand how fast is the eviction of H2A . Z , relative H2A . Z levels were measured at various time points after Swc5 depletion ( Figure 3—figure supplement 2 ) . At most +1 positions , relative H2A . Z levels dropped below 50% of endogenous levels after 15 min of rapamycin treatment , indicating that the occupancy half-life of H2A . Z is less than 15 min . Since virtually baseline level of H2A . Z remains after 60 min of rapamycin treatment ( Figure 3—figure supplement 2 , red ) , the data also suggests that SWR1 is the sole deposition pathway of H2A . Z at budding yeast promoters . 10 . 7554/eLife . 14243 . 026Figure 3 . Depletion of Swc5 revealed rapid PIC-dependent eviction of H2A . Z at the +1 nucleosome of active and infrequently transcribed genes . ( A ) Compiled tag counts of H2A . Z nucleosomes around the +1 dyads in the SWC5-FRB strain before and after 30 min of rapamycin treatment . ( B ) Scatter plot analysis showing the change in relative H2A . Z occupancy against endogenous H2A . Z level in the SWC5-FRB strain at the +1 nucleosomes ( n = 4 , 738 ) after 30 min of rapamycin treatment . Gray: +1 nucleosomes . Orange: reference nucleosomes depleted for H2A . Z . ( C ) Same as Figure 2C , except the SWC5-FRB strain was used . ( D–F ) Same as ( A–C ) , except the SWC5-FRB TBP-FRB strain was used . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 02610 . 7554/eLife . 14243 . 027Figure 3—source data 1 . Compiled nucleosome tag profiles of SWC5-FRB and SWC5-FRB TBP-FRB around the +1 dyads before normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 02710 . 7554/eLife . 14243 . 028Figure 3—source data 2 . Average tag coverage around the +1 dyads of SWC5-FRB . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 02810 . 7554/eLife . 14243 . 029Figure 3—source data 3 . Average tag coverage around the +1 dyads of SWC5-FRB TBP-FRB . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 02910 . 7554/eLife . 14243 . 030Figure 3—figure supplement 1 . Concordance of relative H2A . Z occupancy between technical replicates ( independent IP reactions ) . The +1 nucleosome log ( H2A . Z/input ) values for the replicates of SWC5-FRB and SWC5-FRB TBP-FRB were plotted as described in Figure 1—figure supplement 6 . +RAP: Rapamycin treatment for 30 min . Dotted red lines: arbitrary reference perimeter set at -0 . 25 on both axes . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 03010 . 7554/eLife . 14243 . 031Figure 3—figure supplement 2 . Relative H2A . Z occupancy at the +1 nucleosomes of the SWC5-FRB strain at different times after rapamycin treatment . Same as Figure 3C except additional time points after rapamycin treatment were included . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 031 Next , we tested whether TBP-FRB depletion can slow H2A . Z eviction while the H2A . Z deposition activity of SWR1 is inhibited . Consistent with the idea that the PIC is required for H2A . Z eviction , depleting both TBP and Swc5 in the double mutant resulted in a less dramatic decrease in relative H2A . Z occupancy at most +1 nucleosomes as compared to the single SWC5-FRB mutant for the same duration ( i . e . 30 min ) of rapamycin treatment ( Figures 3D–F and Figure 3—figure supplement 1C–D ) . It is noteworthy that TBP depletion did not completely prevent the depletion of H2A . Z caused by Swc5 depletion . One explanation is that the kinetics of TBP depletion by anchor away is slower than that of Swc5 . Alternatively , a PIC-independent H2A . Z eviction pathway might be operating . The INO80 complex has been reported to catalyze the reverse H2A . Z replacement reaction in which nucleosomal H2A . Z-H2B dimers are replaced with free H2A-H2B dimers ( Papamichos-Chronakis et al . , 2011 ) . Therefore , inactivation of the INO80 gene ( which encodes the catalytic core subunit of the INO80 complex ) in cells with intact SWR1 activity is expected to accumulate H2A . Z . Endogenous INO80 was fused to a tandem FRB-GFP tag to allow conditional depletion by anchor-away and visualization by fluorescence microscopy . Mutants defective for INO80 function are hypersensitive to hydroxyurea ( Shen et al . , 2003 ) . Cells with INO80-FRB-GFP ( referred to as INO80-FRB hereafter ) exhibited slow growth in medium containing both hydroxyurea and rapamycin but not hydroxyurea alone , confirming that INO80 function can be abolished in a rapamycin-dependent manner ( Figure 4—figure supplement 1A ) . As expected , in the absence of rapamycin , Ino80 was found in the nucleus ( Huh et al . , 2003 ) ( Figure 4—figure supplement 1B–C at 0 min +RAP ) . After 90 min of rapamycin treatment , Ino80 was largely dispersed from the nuclei ( Figure 4—figure supplement 1C ) . These cells were then fixed by formaldehyde crosslinking and the H2A . Z and H2A levels were measured by qChIP-seq . No significant global increase of H2A . Z was observed after Ino80 depletion ( Figures 4A–C and Figure 4—figure supplement 1D ) . Instead , nucleosomal arrays became 'fuzzier' as demonstrated by the general decrease of nucleosomal peak height and the decrease of valley depth at the linker regions ( Figure 4D , top ) . By contrast , in the untagged control ( no FRB ) , the density and positions of the nucleosomal arrays before and after rapamycin treatment were unchanged ( Figure 4D , middle ) . Using the difference in nucleosomal density within the linker region between +1 and +2 as a criterion for fuzziness , the chromatin arrays of 983 genes that require Ino80 for positioning were identified by clustering analysis ( k = 3 ) . The compiled chromatin arrays of these promoters showed , upon Ino80 depletion , the -1 and +1 nucleosomes shifted away from the NDR ( Figure 4D , bottom ) . Importantly , the direction of shift at these nucleosomal positions are consistent with the published results of ino80∆ and are reproducible in both biological replicates ( Yen et al . , 2012 ) ( Figure 4—figure supplement 1E ) . Overall , our data is in agreement with the INO80 remodeler functioning as a histone octamer slider but not an evictor of H2A . Z nucleosomes in vivo ( Jeronimo et al . , 2015; Shen et al . , 2003; Yao et al . , 2016; Yen et al . , 2012 ) . 10 . 7554/eLife . 14243 . 032Figure 4 . The effects of Ino80 depletion on H2A . Z occupancy and nucleosomal positions . ( A ) Sequencing tag coverage of H2A . Z ( green ) , H2A ( red ) , and input ( gray ) nucleosomes , as well as the corresponding ∆ ( Z/A ) values were plotted as described in Figure 1A . ( B ) Compiled read counts around the +1 nucleosome dyads . ( C ) Scatter plot and histogram showing the change in ( H2A . Z/input ) for individual +1 nucleosomes after Ino80 depletion . The thresholds ( dotted lines ) were determined as described in Figure 1F . Open black circles: +1 nucleosomes . Orange: reference nucleosomes . ( D ) The compiled input nucleosome profiles before ( black ) and after ( red ) rapamycin treatment in INO80-FRB ( top ) and no FRB , ( middle ) were re-plotted using the data from ( B ) and Figure 1D to highlight any difference in the nucleosomal arrays . Bottom: The compiled tag counts of 983 genes with fuzzier nucleosomal organization upon Ino80 depletion . The integers above the nucleosomal peaks indicate shift distance in base pairs . Positive values indicate right-shift after rapamycin treatment and vice versa . Peak center positions were determined by curve fitting with a Gaussian model followed by local maxima calculation . All qChIP-seq data of Ino80-FRB represent the mean of two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 03210 . 7554/eLife . 14243 . 033Figure 4—source data 1 . Nucleosome tag profiles of INO80-FRB around the +1 dyads before normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 03310 . 7554/eLife . 14243 . 034Figure 4—source data 2 . Average tag coverage around the +1 dyads of INO80-FRB . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 03410 . 7554/eLife . 14243 . 035Figure 4—source data 3 . The list of 983 genes with fuzzier nucleosomal organization after Ino80 depletion . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 03510 . 7554/eLife . 14243 . 036Figure 4—figure supplement 1 . Verification of the conditional depletion of Ino80-FRB and concordance of biological replicates . ( A ) Spotting assay of the anchor-away strain bearing the INO80-FRB allele fused in-frame with GFP and the isogenic untagged ( no FRB ) control in the presence or absence of 100 mM hydroxyurea ( HU ) and/or 1 µg/mL rapamycin . ( B ) Fluorescence microscopy of fixed yeast cells expressing the INO80-FRB- ( GFP ) construct . The cells were fixed with 4% formaldehyde for 5 min before they were washed with PBS and stained with DAPI ( pseudo-colored in red ) . ( C ) Fluorescence microscopy of live cells expressing INO80-FRB- ( GFP ) before and after various times of rapamycin treatment . DIC: differential interference contrast optics . ( D ) Scatter plots showing the concordance of log ( H2A . Z/input ) at +1 nucleosomes for the biological replicates . ( E ) Compiled input nucleosome profiles in INO80-FRB before ( black ) and after ( red ) rapamycin treatment of the biological replicates . The integers above the nucleosomal peaks indicate the shift distance of the peaks after rapamycin treatment . Peak center positions were determined as described in Figure 4D . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 036 The nuclear depletion experiments of TBP and Rpb1 also provide insights into the role of the transcription machinery in nucleosomal spacing organization . Both TBP and Rpb1 depletion caused histone octamers to reposition away from the NDR ( Figure 5A ) . By contrast , no significant positional shift was observed in the nucleosomal arrays of the control cells ( Figure 4D , no FRB ) . Similar results were previously observed by inactivating the transcription machinery with the rpb1-1 mutant , which contains a temperature sensitive allele of RPB1 ( Weiner et al . , 2010 ) . These data are in agreement with the in vitro data that showed an elongating Pol II can disassemble the nucleosome in its path , while reassembling the nucleosome at a position slightly more upstream ( Clark and Felsenfeld , 1992 ) . But unlike the rpb1-1 data , which exhibited strong downstream shift in the +1 nucleosomes under the non-permissive condition , the +1 shift upon TBP and Rpb1 depletion was comparatively minor ( Weiner et al . , 2010 ) ( Figure 5A ) . In fact , the nucleosomal positional shift is progressively more dramatic towards the 3’ end of the genes in the TBP and Rpb1 depletion experiments ( Figure 5A ) . Interestingly , the positional shift caused by Rpb1 depletion is greater than that caused by TBP depletion ( Figure 5A ) . One explanation for the difference is that TBP depletion blocks the Pol II molecules that are initiating but not those that have already engaged in elongation , whereas Rpb1 depletion removes all Pol II from the genome . But in both cases , the size of the NDR remains largely unchanged before and after the depletion of TBP or Rpb1 indicating that the formation of the NDR can be established in the absence of the PIC . 10 . 7554/eLife . 14243 . 037Figure 5 . Nucleosomes shift away from the NDR in response to TBP and Rpb1 depletion . ( A ) The compiled tag counts of the input nucleosomal fraction ( n = 4 , 738 ) before ( black ) and after ( red ) nuclear depletion of TBP-FRB ( top ) and Rpb1-FRB ( bottom ) were re-plotted using the data from Figure 1 and Figure 1C—figure supplement 8B . The integers above the nucleosomal peaks indicate shift distance after rapamycin treatment . Peak center positions were determined as described in Figure 4D . ( B ) Same as ( A ) except that the compiled data of the ribosomal protein genes ( n = 128 ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 037 The phenomenon of nucleosome downshift in gene body is exaggerated at the RP genes where Pol II-mediated transcriptional activity is very high ( Figure 5B ) ( Warner , 1999 ) . But again , the positional shift of the +1 nucleosomes is comparatively insignificant ( Figure 5B ) . This observation is important as a recent study showed repression of the RP genes by heat shock can cause their +1 nucleosomes to shift tens of basepairs in the upstream direction , indicating that these +1 nucleosomes are normally pushed downstream when the RP genes are actively transcribed ( Reja et al . , 2015 ) . Our data showed that PIC is not responsible for the downstream push of the +1 nucleosomes and suggests that a chromatin remodeling enzyme ( s ) might be responsible for setting up the NDR before PIC assembly . Robust H2A . Z dynamics is generally associated with the +1 nucleosomal position and , to lesser extent , the -1 position [see ∆ ( Z/A ) in Figure 2A , right] . However , since many yeast genes are oriented divergently , the H2A . Z dynamics observed at the -1 nucleosome of some promoters could be the +1 nucleosome of a divergent promoter . Indeed , when promoters were sorted based on the orientation of the upstream gene , positive ∆ ( Z/A ) values were more often seen in the upstream region of a promoter with a divergent-oriented gene ( Head-Head ) than with a tandem-oriented gene ( Head-Tail ) ( Venters and Pugh , 2009 ) ( Figure 6A ) . The difference in H2A . Z dynamics at the +1 and -1 positions is exemplified by 44 divergent promoters that are separated by a bona fide -1 nucleosome ( Figure 6—figure supplement 1 ) . At these sites , H2A . Z accumulation was restricted to the +1 positions but not the -1 positions . 10 . 7554/eLife . 14243 . 038Figure 6 . Rapid H2A . Z dynamics is restricted to the +1 nucleosomes of Pol II promoters , not to -1 nucleosomes or fragile nucleosomes . ( A ) A heatmap showing the ∆ ( Z/A ) values of TBP-FRB around the +1 dyads of 3 , 143 promoters sorted by the orientation of the upstream gene . Head-Head ( H-H ) : promoters with an upstream gene oriented divergently . Head-to-tail ( H-T ) : oriented in tandem . ( B ) Normalized H2A . Z , H2A , and input nucleosome tag coverage around the repetitive RDN1 locus . A red arrow marks the TSS of the Pol I-controlled RDN37-1 promoter . ( C ) Compiled tag coverage of 275 tRNA genes with and without rapamycin treatment . ( D ) Heat maps of input nucleosomes around the center of tRNA genes . ( E ) Nucleosome profiles of under-digested chromatin from the TBP-FRB strain centered at the dyad of the +1 nucleosomes . Cyan triangles mark the peak of fragile nucleosomes . ( F ) Same as ( E ) except the profiles were aligned at the dyad of fragile nucleosomes ( Kubik et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 03810 . 7554/eLife . 14243 . 039Figure 6—source data 1 . Nucleosome tag profiles around the +1 dyads for the under-digested TBP-FRB chromatin sample in ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 03910 . 7554/eLife . 14243 . 040Figure 6—source data 2 . A list of -1 nucleosomes that are flanked by two +1 nucleosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 04010 . 7554/eLife . 14243 . 041Figure 6—figure supplement 1 . Compiled nucleosome tag coverage of 44 divergent promoters in TBP-FRB with a bona fide -1 nucleosome flanked by +1 nucleosomes . Tag coverage instead of tag mid-point was plotted to smooth the data . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 041 In addition to Pol II-transcribed genes , TBP is also required for transcription by Pol I and Pol III ( Cormack and Struhl , 1992; Schultz et al . , 1992 ) . An earlier study showed that TBP-FRB can be depleted from Pol I and Pol III promoters to ~60% and ~10% of the original levels , respectively , using the same duration of rapamycin treatment ( Grimaldi et al . , 2014 ) . Contrastingly , at the Pol I-dependent RDN37 promoter , depletion of TBP-FRB actually caused modest H2A . Z depletion and H2A accumulation ( Figure 6B ) . The Pol III-dependent tRNA genes have intragenic promoters ( Schramm and Hernandez , 2002 ) . These genes are typically flanked by nucleosomes with low H2A . Z content ( Albert et al . , 2007 ) . After TBP depletion , the flanking nucleosomes showed only a subtle increase in H2A . Z ( Figure 6C ) . However , there is a substantial encroachment of nucleosome density ( presumably H2A-containing ) within the tRNA gene-associated NDRs ( Figures 6C–D ) . These results suggest that the assembly or activity of the Pol III transcription machinery is important for creating the NDRs at these sites . The NDR region is not completely histone free but is occasionally associated with unstable , histone-containing structures , also known as 'fragile nucleosomes' , that are hypersensitive to MNase digestion ( Kubik et al . , 2015; Weiner et al . , 2010; Xi et al . , 2011 ) . An outstanding question is whether these fragile nucleosomes contain H2A . Z ( Pradhan et al . , 2015 ) . In some of our qChIP-seq experiments , where the chromatin was under-digested , fragile nucleosomes were observed ( Figure 6E , cyan arrowhead ) . Strikingly , fragile nucleosomes are completely devoid of H2A . Z . The phenomenon is more obvious when sequencing tags around the dyads of 1 , 953 fragile nucleosomes were compiled ( Figure 6F ) ( Kubik et al . , 2015 ) . When H2A . Z eviction is blocked by TBP depletion , no H2A . Z accumulation was observed at the fragile nucleosome positions ( Figures 6E–F ) . This indicates H2A . Z deposition does not occur at these sites . The lower signal in input of the + RAP sample over the fragile nucleosome was due to slight over-digestion . Overall , our data suggest that H2A . Z deposition and eviction are highly localized at the +1 nucleosomes of Pol II-transcribed genes . The finding that H2A . Z dynamics revealed by the TBP-FRB mutant is linked to Pol II transcription start sites , raises the possibility that the ∆ ( Z/A ) parameter can be used to identify cryptic , alternative , or previously unassigned transcription start sites . Indeed , the ∆ ( Z/A ) parameter correctly marked the start sites of many previously annotated cryptic unstable transcripts ( CUTs ) ( Figure 7A , red arrows , CUT531 and CUT445 ) and noncoding RNAs ( Figure 7A , cyan arrows , SUT013 and ICR1 ) ( Bumgarner et al . , 2012; Xu et al . , 2009 ) . To identify novel start sites , the ∆ ( Z/A ) profiles around the dyads of all 61 , 568 annotated nucleosomes of the yeast genome ( Kubik et al . , 2015; Jiang and Pugh , 2009 ) were sorted by k-means clustering ( Figure 7—figure supplement 1A ) . Since the +1 nucleosome is expected to have a maximum of ∆ ( Z/A ) signal centered around the dyad as opposed to a +2 or -1 nucleosome , which should have an off-centered ∆ ( Z/A ) maximum contributed by the +1 neighbor , nucleosomes with symmetrically distributed ∆ ( Z/A ) signal were selected ( Figure 7—figure supplement 1A ) . The process was reiterated again with the remaining nucleosomes ( n = 56 , 193 ) to identify +1 nucleosomes with weaker ∆ ( Z/A ) signals ( Figure 7—figure supplement 1A ) . This approach identified 4 , 576 potential +1 nucleosomes , of which 3 , 684 were known +1 nucleosomes ( n = 6 , 427 ) of protein-coding and noncoding genes ( Jiang and Pugh , 2009; Kubik et al . , 2015; Rhee et al . , 2014 ) ( Figure 7—figure supplement 1A ) . The remaining 892 were not previously identified as +1 and therefore could represent novel +1 nucleosomes ( Figure 7—figure supplement 1B and Figure 7—source data 1 ) . CUTs are normally degraded by the RNA surveillance machinery but accumulate in the exosome subunit mutant rrp6∆ ( Xu et al . , 2009 ) . Many of the new +1 nucleosomes are associated with subtle , unannotated CUTs revealed by the rrp6∆ strain ( Figure 7B , grey bars ) . Therefore global comparison of H2A depletion and H2A . Z accumulation in the absence of the PIC provides a new parameter for transcription start site determination . 10 . 7554/eLife . 14243 . 042Figure 7 . Using H2A . Z dynamics before and after TBP depletion to identify cryptic and alternative transcription start sites . ( A ) Genomic regions highlighting representative CUTs and noncoding RNAs , e . g . CUT531 , SUT013 , CUT445 , and ICR1 . Transcript data of wild-type ( black ) and rrp6∆ ( purple ) strains are from ( Xu et al . , 2009 ) . ( B ) H2A . Z dynamics around subtle cryptic transcription start sites . Gray bars: unannotated CUTs . ( C ) Start sites masked by upstream transcripts were revealed by strong H2A . Z dynamics ( blue arrows ) . ( D ) The cryptic divergent promoters within the coding region of a Ty3 element . Brown traces: Nascent Pol II RNA tag coverage . Orange dots: previously annotated +1 nucleosomes of protein coding genes ( Rhee et al . , 2014; Kubik et al . , 2015 ) . Gray dots: +1 nucleosomes of CUTs and SUTs . Green ‘+’ signs: new +1 nucleosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 04210 . 7554/eLife . 14243 . 043Figure 7—source data 1 . Novel and previously annotated +1 nucleosomes identified by PIC-dependent H2A . Z dynamics . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 04310 . 7554/eLife . 14243 . 044Figure 7—figure supplement 1 . Identification of novel +1 nucleosomes using the ∆ ( Z/A ) parameter associated with TBP depletion . The ∆ ( Z/A ) profiles around the dyads of 61 , 568 annotated nucleosomes ( Jiang and Pugh , 2009; Kubik et al . , 2015 ) were sorted by k-means clustering ( k = 4 ) . Nucleosomes with robust , symmetrically distributed ∆ ( Z/A ) signals were selected . The procedure was reiterated once with the remaining nucleosomes ( n = 56 , 193 ) with weaker ∆ ( Z/A ) signals . ( B ) Venn diagram showing the overlap of +1 nucleosomes identified by the clustering analysis and known +1 nucleosomes of protein coding and noncoding gene ( Jiang and Pugh , 2009; Kubik et al . , 2015 ) . ( C ) Pol II nascent RNA-seq data highlighting the genomic region around the HSE1 gene . ( D ) Standard RNA-seq analysis of mRNA of a wild-type strain in the S288C ( BY4741 ) background . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 044 The ∆ ( Z/A ) parameter also identifies alternative and novel start sites masked by other transcripts ( Figure 7C ) . The transcriptional start site of GDH2 , which encodes the NAD-linked glutamate dehydrogenase , has previously been mapped to a position on chromosome IV around 74 , 109 bp ( Figure 7C , left , green arrow ) ( Miller and Magasanik , 1991; Xu et al . , 2009 ) . The identification of a new +1 nucleosome within the coding region of GDH2 implies that there is a second start site ~1 kb downstream of the first ( Figure 7C , left , blue arrow ) . Concordantly , microarray mRNA profiling data of wild-type cells ( Figure 7C , black dots ) showed an elevated mRNA level after the second start site indicating that a population of transcripts was indeed initiated from the internal start site ( Xu et al . , 2009 ) . In the case of HSE1 , the novel start site identified by H2A . Z accumulation is likely the bona fide start site masked by an upstream SUT ( Figure 7C , right , blue arrow ) . Further high resolution RNA-seq analysis of nascent Pol II transcripts showed that the originally annotated HSE1 transcribed region ( Xu et al . , 2009 ) consists of two transcripts organized in close tandem ( Figure 7—figure supplement 1C ) . Therefore , the originally annotated TSS of HSE1 belongs to the upstream SUT that terminates immediately before the newly identified TSS of HSE1 ( blue arrow ) , although read-through transcripts emanating from the upstream TSS cannot be excluded ( Figure 7—figure supplement 1C ) . In another example , H2A . Z-enriched nucleosomes that flank an NDR within the Ty3 retrotransposon accumulated H2A . Z ( and lost H2A ) after TBP-FRB depletion , implying the presence of a pair of divergent promoters ( Figure 7D ) . Indeed , our nascent Pol II RNA data , as well as standard RNA-seq analysis of mRNA , confirmed the presence of transcripts initiated from these sites ( Figure 7D and Figure 7—figure supplement 1D ) . Cryptic antisense transcription has been reported for the Ty1 retrotransposon and the antisense Ty1 CUT is important for gene silencing and suppression of Ty1 mobility ( Berretta et al . , 2008 ) . Therefore , the antisense Ty3 CUT identified here may play a similar role ( Figure 7D , gray bar ) .
The promoter-proximal NDR serves not only as a platform for the assembly of the transcription machinery , but also as a hub for the recruitment of an array of chromatin remodeling factors ( Rhee and Pugh , 2012; Venters et al . , 2011 ) . For any given promoter , these proteins apparently do not co-exist but transiently bind and dissociate in a regulated fashion to set up a chromatin environment in and around the NDR that promotes an accurate transcriptional response . The challenge is to understand how these factors are recruited and dislodged , how they regulate the dynamics of chromatin structure , and in what order these steps occur . This work demonstrates that rapid , constitutive turnover of H2A . Z occurs genome-wide , even at infrequently transcribing genes . At the nucleosomes covering or immediately downstream of the start sites of Pol II transcripts , the relative enrichment of H2A . Z is a steady state maintained by two major opposing forces imposed by the SWR1 complex and the transcription machinery . SWR1 converts H2A nucleosomes to the H2A . Z-containing forms ( Figure 8 , step I ) and the transcription machinery then actively disassembles them . 10 . 7554/eLife . 14243 . 045Figure 8 . An updated histone cycle model . Step I: The SWR1 complex is recruited to the NDR and replaces the two nucleosomal H2A-H2B dimers with H2A . Z-H2B dimers . The H2A . Z-H2B dimers are always bound by histone chaperones in vivo but is omitted here for simplicity ( Luk et al . , 2007 ) . Step II-a: Assembly of the PIC at NDR adjacent to the +1 H2A . Z nucleosome . General transcription factors are identified with single letters for simplicity . The model of PIC was adapted from ( Sainsbury et al . , 2015 ) Step II-b: The PIC engages the H2A . Z nucleosome . Dotted black lines depict the unstructured ssDNA , which is not drawn to scale . Red line depicts the nascent RNA . Black arrow indicates the transcription start site ( TSS ) . ( H3-H4 ) 2 tetramers are in gray , H2A . Z-H2B dimers in green , and H2A-H2B dimers in red . Step III-a: After the dissociation of Pol II and GTFs from the promoter , a canonical H2A nucleosome reassembles over the TSS but at an imprecise position that is likely too upstream of the TSS . Step III-b: Chromatin remodelers , such as RSC , slide the octamer downstream to the stereotypic +1 position completing the histone cycle . DOI: http://dx . doi . org/10 . 7554/eLife . 14243 . 045 Given that the SWR1 complex is recruited to the promoter in part through its affinity for long , exposed linker DNA ( Ranjan et al . , 2013 ) , preventing PIC assembly by TBP or Pol II removal could , in theory , allow more SWR1 to bind the NDR leading to H2A . Z accumulation . As for the infrequently transcribed genes , one could argue that the NDRs of these promoters might be occupied with partial PICs that may become accessible to SWR1 binding when TBP or Pol II is removed . Our data argue against this hypothesis in that Swr1 occupancy decreased , rather than increased , in response to TBP depletion . Therefore , it is the SWR1 complex already residing at the NDR before TBP depletion that is responsible for the increased H2A . Z deposition . However , since the Swr1 occupancy data are based on two promoters ( with reliable Swr1 ChIP signal ) , aberrant SWR1 recruitment elsewhere in the genome cannot be ruled out . The observation that Swr1 occupancy decreased upon TBP depletion is interesting as it most likely reflects the lower affinity of SWR1 for the H2A . Z nucleosome relative to the H2A nucleosome ( Ranjan et al . , 2015 ) . As such , after SWR1 has generated the H2A . Z nucleosome product , the enzyme dissociates from the NDR . Alternatively but not exclusively , since optimal SWR1 recruitment requires H3 and H4 tail acetylation , inhibition of transcription could decrease histone acetylation , and thus indirectly impair SWR1 recruitment ( Liu et al . , 2005; Raisner et al . , 2005; Ranjan et al . , 2013 ) . Perturbation of PIC assembly causes promoter-specific H2A . Z , but not H2A , to accumulate , suggesting that the constitutive disassembly occurs preferentially with H2A . Z-containing nucleosomes ( Figure 8 , steps II-a and II-b ) . Currently , it is not known whether the heterotypic AZ or the homotypic ZZ form is the preferred nucleosomal state for disassembly as our experiments did not distinguish between these forms . One explanation for the preference of H2A . Z eviction is that the PIC preferentially assembles on promoters that have a +1 H2A . Z nucleosome . Consistent with this idea , optimal TBP recruitment during oleate induction requires H2A . Z and the C-terminal domain of H2A . Z has been reported to bind RNA Pol II ( Adam et al . , 2001; Wan et al . , 2009 ) . Another nonexclusive possibility is that H2A . Z-containing nucleosomes are preferentially recognized by the PIC for disassembly . As such , when the PIC is fully engaged with the +1 H2A . Z nucleosome , some component ( s ) of the PIC actively disassembles the nucleosome , thereby allowing rapid , robust transcriptional activation . In support of this idea , rapid induction of GAL1-10 , PHO5 and the heat-responsive genes requires H2A . Z and is accompanied by preferential eviction of H2A . Z over H2A ( Santisteban et al . , 2000; Venters et al . , 2011; Zhang et al . , 2005 ) . But regardless of the basis of how the preference for H2A . Z removal is achieved , our data suggest that the eviction of H2A . Z is , at least in part , coupled to some steps during transcription , rather than an independent process upstream of PIC assembly ( Figure 8 , step II-b ) . A recent report that utilized ChIP and qPCR to monitor histone H3 occupancy showed that inactivation of transcription by the temperature sensitive alleles rpb1-1 and tbp ts-1 , led to promoter-proximal H3 accumulation , suggesting that the PIC is required to maintain an open chromatin state at the +1 position that is depleted of all histones ( Ansari et al . , 2014 ) . In the current study , the input nucleosomal levels at the +1 positions were largely unchanged for most genes upon TBP ( or Rpb1 ) depletion . This discrepancy may be an artifact caused by a normalization step in our study that equalizes the input tag counts of all samples . Therefore , if inhibition of transcription had indeed caused a global increase in nucleosomal levels , the input nucleosome of the TBP depleted sample would be undercounted . But because the qChIP-seq approach reports H2A . Z IP efficiency , even if the input nucleosome occupancy were indeed higher in the TBP depleted sample , the absolute increase of H2A . Z would in fact be greater in response to TBP depletion , further supporting our conclusion . Furthermore , the TBP depleted sample does not appear to be severely undercounted , as the least transcribed genes ( bottom 3% ) exhibited similar input nucleosome densities after normalization ( Figure 1—figure supplement 10 ) . Interestingly , another group has previously used MNase-seq to probe the change of chromatin organization in the rpb1-1 mutant and showed that the +1 nucleosomes also did not accumulate when transcription was blocked ( Weiner et al . , 2010 ) . We speculate that PIC-dependent disassembly of the +1 H2A . Z nucleosomes involves some metastable sub-nucleosomal intermediates ( Ramachandran et al . , 2015; Rhee et al . , 2014 ) . Since the DNA associated with the sub-nucleosomal species is likely hypersensitive to MNase digestion , it would be underrepresented in our input fraction . By contrast , standard H3 ChIP is likely less biased against the sub-nucleosomal species and therefore revealed H3 accumulation when transcription was blocked ( Ansari et al . , 2014 ) . Earlier electron microscopy and genome-wide mapping studies place the PIC immediately upstream of the TSS on the DNA , suggesting that the +1 H2A . Z nucleosome and the PIC ( in the closed conformation ) can coexist in many promoters ( Murakami et al . , 2013; Rhee and Pugh , 2012 ) . It is currently unknown which step of the transcription process causes H2A . Z eviction . However , we speculate that promoter scanning mediated by TFIIH is the driving force ( Figure 8 , step II-b ) . A recent single-molecule study showed that yeast PIC repeatedly scans the promoter over an extended region ( 85 base pairs on average ) , leading to cycles of transcription bubble formation and collapse before committing to promoter escape ( Fazal et al . , 2015 ) . Formation of such extended transcription bubbles in the NDR will overwind the DNA downstream , leading to a propensity to form positive supercoils in the +1 nucleosomal region . Given that DNA wraps the histone octamer in a left-handed turn ( i . e . negative writhe ) and that the promoter-distal end of the nucleosome is restricted to rotate freely , positive supercoiling will likely destabilize the +1 nucleosome . Repeated cycles of futile scanning without promoter escape may also occur at infrequently transcribing genes explaining the constitutive H2A . Z dynamics at those promoters . Although we favor a model in which H2A . Z eviction is a direct consequence of the initiation process , we cannot exclude the possibility that a subsequent step of transcription , such as promoter escape or elongation , is the driving force as TBP depletion blocks all steps of transcription . We also cannot exclude TBP or the integrity of the PIC may be required to recruit another factor that functions to disassemble H2A . Z nucleosomes . Since robust PIC-dependent H2A . Z eviction occurs even at promoters that infrequently fire , this provides an opportunity to use H2A . Z dynamics as a parameter to locate initiation sites that were previously missed by the conventional transcript mapping approach ( Xu et al . , 2009 ) . Sites with strong H2A . Z dynamics were able to identify not only initiation sites of protein coding genes , SUTs and CUTs , but also sites that are masked by upstream transcripts . Although this approach cannot pinpoint the location of a TSS at base pair resolution , the sites are generally restricted within the identified +1 nucleosome near the edge adjacent to the NDR . Importantly , since PIC-dependent H2A . Z eviction is linked to an NDR , the direction of the predicted transcripts can be inferred based on the location of the NDR and is almost always pointing away from the NDR . Another limitation of this approach is that it may not identify the TSSs of promoters that are associated with low endogenous H2A . Z deposition activity and/or have low PIC-dependent H2A . Z eviction activity as in the case of the SAGA-dominating promoters . Whether the INO80 complex plays a role in evicting H2A . Z at promoters has been controversial . Evidence supporting this idea comes from in vitro experiments showing that the INO80 complex can mediate a reaction that replaces the nucleosomal H2A . Z-H2B dimer with H2A-H2B , which supposedly opposes the reaction catalyzed by SWR1 ( Papamichos-Chronakis et al . , 2011 ) , as well as in vivo experiments showing a global increase of H2A . Z occupancy in ino80∆ or arp5∆ ( both components of the INO80 complex ) strains ( Papamichos-Chronakis et al . , 2006; Yen et al . , 2013 ) . However , a more recent study showed ino80∆ mutants exhibited normal H2A . Z occupancy at promoters ( Jeronimo et al . , 2015 ) . Our experiments agree with the latter study arguing that INO80 does not contribute to global eviction of H2A . Z , as no accumulation of H2A . Z was observed when Ino80 was depleted . One explanation for the discrepancy between these various experiments is that ino80∆ and arp5∆ are both associated with genome instability , resulting in aneuploidy that complicates the interpretation of genomic data ( Chambers et al . , 2012 and Ashby Morrison personal communication ) . To circumvent this problem , we assayed H2A . Z occupancy within two hours of conditionally depleting Ino80 from the nucleus , thereby minimizing aneuploidy . Furthermore , the previous studies used ChIP-chip and ChIP-exo to measure H2A . Z occupancy ( Jeronimo et al . , 2015; Papamichos-Chronakis et al . , 2011; Yen et al . , 2013 ) . However , these approaches may not be ideal for comparing H2A . Z occupancy in different strains . Standard microarray normalization involves scaling with data averages that could mask any global occupancy change ( Quackenbush , 2002 ) . ChIP-exo measures the distribution of crosslinking points between the immunoprecipitated protein and DNA that could be influenced by variation of crosslinking efficiency in mutants with chromatin dynamics defects ( Yen et al . , 2013 ) . The qChIP-seq approach conducted here takes into account the input , flow-through , and IP fractions of each IP reaction and uses the immuno-depleted signal in the flow-through to scale the IP signal . Therefore , occupancy data reflects IP efficiency relative to input at individual nucleosomal sites and is not influenced by global occupancy change . It is also less sensitive to variation in crosslinking efficiency as formaldehyde crosslinking serves only to preserve the histone-DNA interaction during the pull down . Our data showed that Ino80 depletion did not cause H2A . Z accumulation , arguing against the INO80 remodeler being the main evictor of H2A . Z in the constitutive turnover process . This observation however is not inconsistent with INO80 being recruited to remodel specific promoters in response to induction as seen in the PHO5 and GAL genes ( Barbaric et al . , 2007; Rosonina et al . , 2014 ) . Although INO80 is not involved in the constitutive eviction of H2A . Z at the +1 positions , our data suggest that INO80 does contribute to the positioning of nucleosomes near the promoters . While the sliding defect associated with Ino80 depletion is reproducible and consistent with previous studies ( Yao et al . , 2016; Yen et al . , 2012 ) , the defect is small compared to that associated with mutants in genes encoding other chromatin remodelers , such as RSC , CHD1 , ISW1 , and ISW2 ( Gkikopoulos et al . , 2011; Hartley and Madhani , 2009; Ocampo et al . , 2016; Whitehouse et al . , 2007 ) . Therefore , the role of INO80 in nucleosome positioning could be indirect or redundant with another remodeling factor . We propose an updated 'histone cycle' model to unify the molecular events that lead to the promoter-specific H2A . Z dynamics ( Figure 8 ) . First , ATP-dependent remodelers , such as RSC , are recruited to the promoter region by sequence specific DNA-binding factors to establish the NDR ( Badis et al . , 2008; Hartley and Madhani , 2009 ) ( Figure 8 , step III-b ) . SWR1 is recruited to the NDR by its affinity for extended stretches of naked DNA and , to lesser extent , promoter-specific histone acetylation ( Hartley and Madhani , 2009; Ranjan et al . , 2013 ) . SWR1 then converts the canonical AA nucleosome to the AZ and ZZ chromatin states sequentially ( Luk et al . , 2010 ) ( Figure 8 , step I ) . For promoters that contain a fragile nucleosome within the NDR region ( omitted in Figure 8 for simplicity ) , this structure is somehow refractory to the SWR1-mediated histone replacement reaction . SWR1 dissociates from the promoters after depositing H2A . Z at the +1 position due to its lower affinity for the H2A . Z nucleosomal product ( Ranjan et al . , 2015 ) . The PIC assembles at the NDR adjacent to the newly formed +1 H2A . Z nucleosome ( Rhee and Pugh , 2012 ) . At some point when the PIC engages the TSS , the H2A . Z nucleosome disassembles . This may involves sub-nucleosomal intermediates before all histones are completely evicted from the DNA . Following promoter escape or unproductive dissociation of the PIC , a nucleosome with the major histones , H2A , H2B , H3 , and H4 , reassembles completing the histone cycle ( Figure 8 , step III-a ) . The promoter-specific histone dynamics described by the histone cycle appears to be a general phenomenon in yeast . We show that the Pol II transcription machinery has a major contribution to the eviction step of H2A . Z in the histone cycle . Together , chromatin remodelers and the transcription machinery choreograph the movement of histones within the promoter leading to a dynamic chromatin architecture conducive for transcription . How these enzymes are coordinated to set the histone cycle in motion will be the next important question .
The genotypes of the yeast strains used in this study are listed in Supplementary file 1 . The anchor-away strains HHY221 , HHY170 and HHY209 were obtained from Euroscarf ( Haruki et al . , 2008 ) . The H2A . Z FLAG tagged strains yEL152 ( no FRB tag control ) , yEL154 ( SPT15-FRB-GFP ) , yEL170 ( RPB1-FRB ) , and yEL189 ( INO80-FRB-GFP ) were constructed by integrating a 2xFLAG-URA3 fragment at the 3’ end of the HTZ1 gene in the precursor strains HHY221 , yEL098 , yEL090 , and yEL123 , respectively . yEL098 is a tetrad segregant of the diploid HHY221 x HHY209 strain and yEL090 is a segregant of the diploid HHY221 x HHY170 . The 2xFLAG-URA3 fragment was obtained by PCR amplification of the pRS416-HTZ1-2xFLAG plasmid described in ref . ( Mizuguchi et al . , 2004 ) . Successful integration was confirmed by colony PCR and sequencing . The SWC5-FRB ( yEL219 ) and SWC5-FRB TBP-FRB ( yEL220 ) strains were generated by integrating the FRB-HIS3MX6 fragment at the 3’ end of the SWC5 coding region in yEL152 and yEL154 , respectively , using the one-step gene replacement procedure ( Longtine et al . , 1998 ) . The FRB-HIS3MX6 fragment was amplified by PCR using the plasmid pFA6a-FRB-HIS3MX6 ( P30579 , Euroscarf ) as template ( Haruki et al . , 2008 ) . Similarly , the INO80-FRB-GFP precursor strain ( yEL123 ) was generated by transforming HHY221 ( yEL044 ) with a FRB-GFP-HIS3MX6 PCR product targeting the 3’ end of INO80 . The FRB-GFP-HIS3MX6 fragment was amplified from the plasmid pFA6a-FRB-GFP-HIS3MX6 ( P30581 , Euroscarf ) ( Haruki et al . , 2008 ) . For microscopy , yeast cells were grown in YPD at 30˚C to an OD600 of 1 before rapamycin was added to a final concentration of 1 µg/mL . Cells before and after the addition of rapamycin were removed at the indicate times and were fixed with 4% formaldehyde for 5 min followed by washing with 1x PBS and staining with 1 µg/mL DAPI . For the microscopy analysis of the INO80-FRB-GFP cells in Figure 4—figure supplement 1C , live cells were used in order to shorten the lag time between imaging and fixation before the qChIP-seq analysis . The yeast cells used in quantitative ChIP were cultured at 30˚C in 2 L of yeast extract-peptone-dextrose ( YPD ) to an optical density 600 ( OD600 ) of 0 . 8 before 1 µg/mL rapamycin was added and incubated for the indicated duration . The rapamycin-treated culture , as well as the untreated control , were fixed with 3% paraformaldehyde for 7 . 5 min and quenched by 0 . 12 M glycine as previously described ( Luk et al . , 2010 ) . Cell pellets equivalent to 400 mL culture were aliquoted and washed with 1x PBS before flash frozen and stored at –80˚C . Cells equivalent to 400-mL culture volume were spheroplasted and lysed with a 7-mL dounce homogenizer ( piston B , Wheaton , Millville , NJ ) as previously described ( Luk et al . , 2010 ) . After centrifuging the lysate ( ~1000 µL ) at 13 , 000 x g for 10 min at 4˚C , the chromatin-enriched pellet was washed 3 times with buffer A [50 mM HEPES ( pH 7 . 5 ) , 80 mM NaCl , 0 . 25% Triton X-100 , 2x protease inhibitor cocktail ( cOmplete EDTA-free , Roche , Switzerland ) ] . The pellet was resuspended in 1/2 x lysate volume of buffer A plus 1 mM of CaCl2 . The suspension was pre-incubated to 37˚C for 2 min before digestion with 1 . 5–1 . 8 U/μL MNase ( Worthington , Lakewood , NJ ) at 37˚C for 10–20 min . The concentration of MNase and duration of digestion were empirically determined for each sample to achieve ~90% mono-nucleosomal form . After stopping the digestion reaction by the addition of 10 mM EGTA , the sample was centrifuged at 20 , 400 x g for 15 min at 4˚C . The supernatant , which contained the soluble nucleosomes , was cleared by passage through a low-binding PVDF filter ( Ultrafree GV , EMD Millipore , Billerica , MA ) . Before the soluble nucleosomes were subjected to quantitative ChIP , a 50 µL aliquot was saved . The DNA extracted from this sample represents the input nucleosome fraction . The remaining sample ( ~500 µL ) was diluted 10-fold in buffer B [25 mM HEPES-KOH pH 7 . 6 , 80 mM KCl , 1 mM EDTA , 1x protease inhibitors ( PI , which consists of 1 . 7 mg/mL PMSF , 3 . 3 mg/mL benzamidine hydrochloride , 0 . 1 mg/mL pepstatin A , 0 . 1 mg/mL leupeptin , 0 . 1 mg/mL chymostatin ) ] followed by incubation with 1/20x volume ( ~250 µL resin volume ) of anti-FLAG M2 affinity gel ( A2220 , Sigma-Aldrich , St . Louis , MO ) at 4˚C for 4 hr . The flow-through of the IP reaction would represent the H2A nucleosome fraction . The bead-bound H2A . Z nucleosomes were washed three times with buffer C ( 25 mM HEPES-KOH pH 7 . 6 , 0 . 3 M KCl , 1 mM EDTA , 0 . 01% NP-40 , 1x PI ) and eluted with 2x resin volume ( ~500 µL ) of buffer D ( 25 mM HEPES-KOH pH 7 . 6 , 0 . 3 M KCl , 1 mM EDTA , 0 . 1% NP-40 , 0 . 1% DOC , 1x PI , 0 . 5 µg/µL 3xFLAG peptides ) at 4˚C overnight . The eluate was cleared by passage through a GV 0 . 22 µm filter ( EMD Millipore ) . The FLAG eluate would represent the H2A . Z nucleosome fraction . To extract the nucleosomal DNA , the sample was adjusted to 400 µL with molecular grade water followed by the addition of 32 µL 5 M NaCl , 8 µL 0 . 5 M EDTA , 20 µL 10% SDS , 2 . 5 µL 20 µg/µL glycogen ( Thermo Fisher , Waltham , MA ) , and 8 µL of 20 mg/mL proteinase K ( Thermo Fisher ) . The mixture was incubated at 55˚C for one hour to facilitate protein digestion and 65˚C for > 15 hr to reverse the crosslinks . The DNA was purified by standard phenol-chloroform extraction followed by ethanol-NaOAc precipitation . The pellet was resuspended in 100 µL of 5 µg/mL RNase ( Roche ) in 1x TE pH 8 . 0 and incubated at 37˚C for > 1 hr . The resulting nucleosomal DNA was purified using the QIAquick spin column ( Qiagen , Germany ) and quantified by the Qubit HS assay ( Thermo Fisher ) . To prepare the DNA libraries for sequencing , 30 ng of nucleosomal DNA was treated with 5 U of alkaline phosphatase ( CIP , NEB , Ipswich , MA ) for 1 hr at 37˚C followed by the addition of 20 µg glycogen ( Thermo Fisher ) . The mixture was then purified by standard phenol-chloroform extraction and ethanol-NaOAc precipitation . The DNA was resuspended in 12 µL of the TruSeq Resuspension Buffer ( Illumina , San Diego , CA ) and quantified by the Qubit HS assay . Ten nanograms of CIP-treated DNA was applied to the TruSeq ChIP workflow ( Illumina ) with the following modifications . After the end-repair step , instead of using the AMPure beads for purification , the sample was purified by standard phenol-chloroform extraction and ethanol-NaOAc precipitation . The adapter-ligated DNA was amplified on a PCR machine by 15 thermal cycles . The PCR product was quantified by the Qubit assay ( Thermo Fisher ) and the quality was verified by agarose electrophoresis and SYBR green staining ( Thermo Fisher ) . Densitometry was performed on a Typhoon FLA9500 scanner station installed with the ImageQuant TL software ( GE Healthcare , Pittsburgh , PA ) . Paired-end sequencing ( 33 cycles ) was performed on a MiSeq sequencer ( Illuimna ) . Sequencing reads were aligned to the S . cerevisiae genome ( SGD version R64-1-1 ) using Bowtie 1 . 1 . 2 ( Langmead et al . , 2009 ) . Aligned data of the H2A . Z , H2A and input nucleosome fractions were processed without smoothing using a combination of custom Python scripts and BEDTools programs ( Quinlan , 2014 ) . They were plotted along the genome either as tag coverage ( density covered by paired-end reads ) or tag counts ( density of mid-points of paired-end reads ) . The amplitude of the H2A . Z ( Z ) , H2A ( A ) , and input ( T or total ) nucleosome profiles was scaled using an approach called ‘TAZ normalization’ ( Figure 1—figure supplement 4 ) . First , the tag counts within 44 reference regions ( called the no-Z-zones ) , which are enriched for H2A but depleted for H2A . Z ( Figure 1—figure supplement 4A , Figure 1—source data 1 ) , were determined by Bowtie . The normalization factor m , representing the ratio of no-Z-zones tag count of the input fraction over that of the H2A fraction , was used to scale the H2A nucleosome profile ( Figure 1—figure supplement 4B , Figure 1—source data 2 ) . To normalize the H2A . Z nucleosome profile for each IP reaction ( technical replicate ) , the tag counts around 4 , 738 promoters were compiled around the +1 dyads for the H2A , H2A . Z and input fractions . The normalization factor n , was determined by a curve fitting algorithm such that the sum of the compiled profiles of H2A and H2A . Z ( m × H2A + n × H2A . Z ) equals , to a first approximation , the compiled profile of the input ( Figure 1—figure supplement 4B ) . The Python script used to determine the m and n values can be found in Source code 1 . Normalization accuracy was verified using a visualization tool , which runs on Datagraph ( Visual Data Tools Inc . ) ( Supplementary file 3 ) . A more detailed procedure on how to use the Python script and the Datagraph program can be found in Supplementary file 4 . Finally , the input profiles for each IP reaction were converted to counts per million ( CPM ) by the normalization factors c . The products of c × m and c × n represent the final normalization factors for the H2A and H2A . Z nucleosome profiles , respectively . The compiled nucleosomal profiles for each qChIP-seq experiment can be found in Figure 1—source data 3 , 4 and 5 , Figure 3—source data 1 , Figure 4—source data 1 and Figure 6—source data 1 , and the normalization factors in Figure 1—source data 2 . All qChIP-seq experiments with the exception of the experiment in Figure 3 represent the average of two biological replicates ( i . e from independent cultures ) . In Figure 3 , the qChIP-seq data of SWC5-FRB and SWC5-FRB TBP-FRB strains represent the averages of two independent IP reactions ( technical replicates ) of the same biological sample . The biological replicate for SWC5-FRB is shown in Figure 3—figure supplement 2 , which included additional time points after rapamycin treatment . Data averaging was performed on the normalized tag coverage or tag counts after the TAZ normalization step; therefore , sequencing reads were not pooled . Antibodies against the yeast Swr1 and H2A . Z were gifts of Carl Wu . The anti-FLAG ( F1804 ) and the anti-H2A ( 39235 ) antibodies were purchased from Sigma-Aldrich and Active Motif ( Carlsbad , CA ) , respectively . For H2A and H2A . Z western , both antibodies were used at a dilution of 1:2 , 000 . ChIP-qPCR analysis was performed as previously described with the following modifications ( Aparicio et al . , 2005 ) . ChIP reactions were performed using 1 . 25 mg dynabeads conjugated to Protein A or G . Five microliter of anti-Swr1 was used in each ChIP reaction . qPCR was performed on a LightCycler 96 Real-Time PCR system ( Roche ) . The primers used in qPCR are listed in Supplementary file 2 . Nascent Pol II transcript sequeuncing was performed essentially according to ( Churchman and Weissman , 2012 ) but with the following modifications . Briefly , yeast cells expressing RPB3-3xFLAG ( yEL297 ) were cultured in 6 L YPD to a final OD of 2 . Washed cells were pulverized in Buffer E ( 20 mM HEPES-KOH pH 7 . 6 , 20% Glycerol , 50 mM KOAc , 1 mM EDTA , 1x PI ) for 15 min using the Freezer Mill ( SPEX , Metuchen , NJ ) . The resulting whole cell extract was incubated with 40 U/mL of DNase I ( DPRF , Worthington ) , 25 Units/mL of SUPERase RNase Inhibitor ( Thermo Fisher ) and 10 mM MnCl2 at 4˚C for 30 min before centrifugation at 48 , 000 x g for 30 min . Immunopurification was performed using 12 mL ( 1-L culture equivalent ) of cleared lysate and 400 µL ( bed-volume ) of anti-FLAG agarose ( Sigma-Aldrich ) on a rotator at 4˚C for 2 . 5 hr . After washing with Buffer F ( 20 mM HEPES-KOH pH 7 . 6 , 10% glycerol , 110 mM KOAc , 0 . 1% NP-40 , 1x PI ) , elution was performed with 400 µL of 0 . 5 µg/µL 3xFLAG peptide in Buffer G ( 20 mM HEPES-KOH pH 7 . 6 , 10% glycerol , 110 mM KOAc , 0 . 1% NP-40 , 1 mM DTT , 1x PI ) on a rotator at 4ºC for 2 hr . RNA co-eluted with the Rpb3-3xFLAG Pol II complex was purified on a miRNeasy column ( Qiagen ) according to the manufacturer protocol . To further enrich for the nascent Pol II RNA , ribosomal RNA was depleted using the Yeast Ribo-Zero kit ( Illumina ) . cDNA library was prepared using the TruSeq stranded mRNA library prep kit ( Illumina ) . Sequencing reads were aligned to the yeast genome with Bowtie as described above . A combination of custom Python scripts and BEDTools programs ( Quinlan , 2014 ) were used to generate the tag coverage data for Figure 7D and Figure 7—figure supplement 1C . The Pol II nascent RNA data represents the average of two biological replicates . Sequencing data are available at the National Center for Biotechnology Information Sequence Read Archive ( http://www . ncbi . nlm . nih . gov/sra ) under accession number SRP051897 and are also accessible through BioProject ID PRJNA271808 . Processed genome-wide qChIP-seq data are available through Dryad under the Digital Object Identifier: doi:10 . 5061/dryad . dj782 ( Tramantano et al . , 2016 ) . | To fit the genetic information of an animal , yeast or other eukaryote into cells , DNA is tightly wound around proteins called histones to form repeating units known as nucleosomes . However , this tight winding prevents proteins from accessing the DNA , and so prevents gene transcription – the first stage of producing the molecules encoded by a gene . For transcription to take place , nucleosomes at DNA sequences called promoters must be reorganized and disassembled , thereby allowing proteins to bind to and engage these sequences and to turn nearby genes on . H2A is a histone protein that is found in the majority of nucleosomes in yeast cells . A different form of this histone – called H2A . Z – is found in nucleosomes near the promoter of almost every gene . It is thought that nucleosomes that contain H2A . Z are recognized and disassembled as the gene turns on , but it is unclear how this happens . To investigate how H2A . Z nucleosomes are disassembled , Tramantano et al . depleted yeast cells of various proteins thought to play a role in the disassembly process . This indicated that the proteins that transcribe genes play crucial roles in the process of disassembling the H2A . Z nucleosomes , because H2A . Z accumulated at promoters in cells that are depleted of these proteins . Further investigation revealed that disassembled H2A . Z nucleosomes are reassembled with H2A histones , before being converted back to the H2A . Z form by an enzyme called SWR1 . This turnover of H2A . Z was seen at active genes and those that are infrequently transcribed , suggesting that it is a general phenomenon . Tramantano et al . also found that the turnover rate of H2A . Z can be used to accurately predict the sites in the DNA where transcription starts . This observation could therefore help to identify previously unknown transcription start sites . Future work could address further questions about how H2A . Z nucleosomes are disassembled . For example , what is the mechanical force that drives this process ? And at what step of the transcription process does it occur ? | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"biochemistry",
"and",
"chemical",
"biology"
] | 2016 | Constitutive turnover of histone H2A.Z at yeast promoters requires the preinitiation complex |
Responses of sensory neurons represent stimulus information , but are also influenced by internal state . For example , when monkeys direct their attention to a visual stimulus , the response gain of specific subsets of neurons in visual cortex changes . Here , we develop a functional model of population activity to investigate the structure of this effect . We fit the model to the spiking activity of bilateral neural populations in area V4 , recorded while the animal performed a stimulus discrimination task under spatial attention . The model reveals four separate time-varying shared modulatory signals , the dominant two of which each target task-relevant neurons in one hemisphere . In attention-directed conditions , the associated shared modulatory signal decreases in variance . This finding provides an interpretable and parsimonious explanation for previous observations that attention reduces variability and noise correlations of sensory neurons . Finally , the recovered modulatory signals reflect previous reward , and are predictive of subsequent choice behavior .
Sensory information is represented in the activity of populations of neurons , but the responses of individual neurons within these populations are not uniquely determined by external stimuli: repeated presentations of the same stimulus elicit different spike trains . Although some of this variability presumably arises from noise in local circuits , a substantial portion appears to be due to fluctuations in neurons’ excitability , i . e . their gain ( Goris et al . , 2014 ) . Since this may result from changes in internal states that modulate sensory responses , such as wakefulness , reward , and expectations , or from changes in top-down signals arriving from other cortical areas , it is likely that a significant component of the gain fluctuations in individual neurons are not “private” , but shared across populations ( Ecker et al . , 2014; Schölvinck et al . , 2015; Lin et al . , 2015 ) . Shared gain fluctuations could have a crucial impact on sensory computations , depending on their structure ( Moreno-Bote et al . , 2014 ) . Yet this structure remains largely unknown . Visual attention provides a well-known example of a process that affects neural gain . When monkeys direct their attention to a visual stimulus , the mean firing rates of specific subsets of visual neurons in striate ( McAdams and Maunsell , 1999; Herrero et al . , 2008 ) and extrastriate cortex ( Moran and Desimone , 1985; Treue and Maunsell , 1996; Treue and Maunsell , 1999; Martínez-Trujillo and Treue , 2002; Williford and Maunsell , 2006; Cohen and Maunsell , 2009; Mitchell et al . , 2009 ) have been found to increase . This motivates the hypothesis that , at the level of the population , attention acts to increase the gain of selected neurons , thereby increasing the signal-to-noise ratio of sensory representations , and hence improving performance on perceptual tasks ( McAdams and Maunsell , 1999; Treue and Maunsell , 1996; Cohen and Maunsell , 2009; Lee and Maunsell , 2009; Reynolds and Heeger , 2009 ) . This “classical” view of attention has been augmented by recent observations that spatial attention affects more than mean response: in particular , attention reduces normalized measures of spike count variance , as well as stimulus-conditioned spike count correlations ( i . e . “noise correlations” ) across pairs of neurons ( Cohen and Maunsell , 2009; Mitchell et al . , 2009; Herrero et al . , 2013 ) . These measurements ( and similar findings in the context of perceptual learning ( Gu et al . , 2011; Jeanne et al . , 2013 ) , cognitive challenge ( Ruff and Cohen , 2014 ) , task engagement ( Downer et al . , 2015 ) , or wakefulness ( Poulet and Petersen , 2008 ) seem to imply that attention does more than just increase neural gain; for instance , it might change the underlying connectivity of the network . But the origin of these effects is difficult to interpret , as changes in pairwise correlations can arise from changes in direct or indirect couplings between neurons , or changes in common modulatory input ( Goris et al . , 2014; Ecker et al . , 2014; Brody , 1999; Yatsenko et al . , 2015 ) . In particular , several groups have suggested that variability in the attentional signal itself might contribute to spike count variance and correlation ( Goris et al . , 2014; Cohen and Maunsell , 2010; Harris and Thiele , 2011; Ecker et al . , 2012 ) . Here , we develop a functional model for the population activity that accounts for all the above observations . The model includes stochastic modulatory signals that alter the gains of targeted subsets of neurons . We fit the model to spiking data from populations of ∼100 neurons in visual area V4 , simultaneously recorded from both hemispheres of a macaque monkey that was performing a change-detection task under directed spatial attention ( Cohen and Maunsell , 2009 ) . The model is fit without specification of the hemispheric origin of the neurons , or the extent to which they were influenced by the various gain factors . The resulting fitted model reveals that the population was predominantly influenced by two independent shared modulatory signals , each operating primarily within one hemisphere , and each targeting the neurons most relevant for the task . The statistics of each of these signals changed significantly depending on the attentional condition , with each modulator exhibiting a decrease in variance when the monkey was cued to attend to the stimuli in the corresponding hemifield . Together with the ( classical ) increases in mean response , these changes in the statistics of the shared modulatory signals account for the previously-reported decrease in neural variability and noise correlations . Finally , we show that the inferred modulatory signals are correlated with the monkey’s behavioral performance on each trial , and are influenced by the reward received on the previous trial . The structure and statistics of shared modulatory fluctuations thus provide a parsimonious account of attentional effects on population coding and behavior . A preliminary account of these results was presented in Rabinowitz et al . ( 2015a ) .
We used a computational model to explore the structure of neural population activity in the presence of spatial attention ( Figure 1 ) . We describe the stimulus-driven instantaneous firing rate of each neuron n over time as fn ( s ( t ) ) , where the function fn ( · ) describes the mapping from stimulus , s ( t ) , to that neuron’s firing rate . In addition , we allow each neuron’s gain to be affected by three signals: ( 1 ) the current attentional cue , c ( t ) , ( a binary signal ) ; ( 2 ) a slowly-varying global drift , d ( t ) ; and ( 3 ) a set of shared time-varying modulators , mk ( t ) . We assume that there are K such modulatory signals , indexed by k . The degree to which each of these three signals affects the gain of each neuron is specified by a set of coupling weights , {un , vn , wn , k} , and the net instantaneous firing rate of the neuron is written as: ( 1 ) rn ( t ) =fn ( s ( t ) ) ·expun·c ( t ) +vn·d ( t ) +∑k=1Kwn , k·mk ( t ) 10 . 7554/eLife . 08998 . 003Figure 1 . Diagram of the modulated population model . Shown are three neurons ( yellow boxes ) , each with a firing rate that is a function of the stimulus multiplied by three time-varying gain signals: the binary attentional cue; a slow global drift; and a set of shared modulators . The influence of each of these signals on each neuron is determined by a coupling weight , indicated by the thickness of the blue and black lines . Only one shared modulator is shown in this schematic , but the model allows for more , each with its own coupling weights . The spike counts of each neuron are conditionally Poisson , given the firing rate . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 00310 . 7554/eLife . 08998 . 004Figure 1—figure supplement 1 . Example slow drifts in spike counts of four simultaneously-recorded units ( two from each hemisphere ) , taken from a recorded population of 77 units . Each point is the average spike count observed over ten consecutive stimulus presentations . The blocked structure of the task ( i . e . the alternating cue directions ) is indicated with alternating colors . Thick lines indicate the portion of model-estimated firing rate due to the combination of stimulus drive , cue , and the slow drift signal ( without the shared modulators ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 00410 . 7554/eLife . 08998 . 005Figure 1—figure supplement 2 . Signatures of modulatory ( multiplicative ) effects in the neural responses . The shared modulator model is based on the assumption that the stimulus-driven firing rates of sensory neurons are modulated ( multiplied ) by a set of additional inputs . Evidence for fluctuating modulation of sensory responses was recently provided in ( Goris et al . , 2014 ) . The statistical argument that these inputs are multiplicative ( rather than additive ) is based on a consideration of their effects as a function of firing rate: multiplicative noise has its greatest effects on response variance ( and covariance ) for stimuli evoking high firing rates , while additive noise has its greatest effects on response variance ( and covariance ) for stimuli evoking low firing rates . This is difficult to assess in the context of the V4 attentional dataset , since the standard stimuli were all identical , and responses to the targets were often interrupted by saccades . Restricting the analysis to target responses that were not interrupted is possible , but such conditioning on a behavioral state ( and likely a modulator state ) would complicate any interpretation . There was , however , some variability in evoked firing rates arising from adaptation . Many neurons showed a trend of slightly decreasing response to the sequence of standard stimuli , with an average total decrease of ∼5% in firing rate over the ten presentations after the first . This decrease was sufficiently small that including it in the analysis of the main text only marginally improved predictive log-likelihoods ( as in Figure 2a ) , and did not qualitatively change any of the main results or conclusions . These small changes in firing rate over the stimulus sequence were nevertheless sufficient to examine the hypothesis that neural response variance was due a multiplicative noise source . Following the logic of ( Goris et al . , 2014 ) , if we assume a multiplicative noise source with variance σ2 , the Fano factor ( variance divided by the mean ) should increase with firing rate , μ:FFmult ( μ ) =μ σ2μ2μ=1+σ2μOn the other hand , for an additive noise source of variance σ2 , the Fano factor should decrease with firing rate:FFadd ( μ ) =μ σ2μ=1+σ2μThese expected trends are illustrated in panel ( a ) . Panel ( b ) shows the mean value of these quantities , estimated for the cue-towards condition across all cells . The data are clearly consistent with a multiplicative noise source . A similar trend is observed in the cue-away condition , albeit with an overall lower mean rate , and higher Fano factor . This analysis assumes that the noise source has constant variance across stimulus conditions . There were some small , non-monotonic changes in the estimated shared modulators’ variance over the stimulus sequence . Factoring these in does not change the direction of the predictions or the data shown here . A similar analysis can be performed on pairwise response statistics ( as stimuli evoke higher mean rates , a multiplicative model predicts correlations will increase , while an additive model predicts correlations will decrease ) . But these predictions prove more sensitive to the assumption of stability in σ2 , so we omit them here . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 005 The exponential acts to convert the weighted sum of the three signals , which may take on positive or negative values , into a product of three positive-valued modulatory quantities . Importantly , only the stimulus s ( t ) and cue signal c ( t ) are known . The drift , d ( t ) , and modulators , mk ( t ) , as well as all of the coupling weights must be fit to the experimental data . To accomplish this , we assumed that the firing rate of each neuron , rn ( t ) , was constant over the duration of each stimulus presentation ( 200 ms ) , and that the observed spike counts arose from a Poisson distribution with that rate . We then fit this probabilistic model to the spike count data by maximizing the posterior probability of the model parameters ( see Materials and methods ) . We describe each of the component signals in turn . The stimulus and cue signals encapsulate the external factors that are set by the experimenter and available to the monkey . Since we only analyzed responses to the standard stimuli— which were identical for all trials of a given day—the stimulus-dependent drive in this experiment , fn ( s ( t ) ) , is captured by a single mean firing rate per neuron . For clarity of presentation , we assume that this is the firing rate when the monkey was cued away from this stimulus ( i . e . to the opposite side ) . The response of each neuron is then affected by the cue signal to a different degree , which is captured by an additional free parameter per neuron ( the “coupling weight” to the cue signal ) , un . Together , these two factors comprise the “classical” model of attention , wherein the attentional cue results in a change in the mean firing rate of each neuron . The second modulatory signal , d ( t ) , is motivated by recent reports of coordinated , global slow fluctuations in neural response . These have been hypothesized to arise from fluctuations in global state variables such as arousal , with different neurons potentially affected by these signals to different degrees ( Goris et al . , 2014; Ecker et al . , 2014; Schölvinck et al . , 2015; Okun et al . , 2015; McGinley et al . , 2015 ) . The V4 data analyzed here also exhibit slow global drifts in neural gain in each recording , with typical time constants on the order of minutes to tens of minutes ( Figure 1—figure supplement 1 ) . We therefore allowed a single global slow signal to affect the population . Its timescale and time course were inferred from the population responses , as well as its coupling weights to each neuron ( see Materials and methods ) . While this substantially improved model predictions ( see below ) , global state fluctuations have been described elsewhere and are not the main focus of this work . Importantly , the inclusion or exclusion of this global drift signal ( even allowing more than one ) does not qualitatively change the results or interpretation presented here . Our primary interest was in the K fast modulatory signals , mk ( t ) , which we introduced to explain any shared structure in the population activity that remains after the structure of the task and global slow drifts have been accounted for . We assumed that these consisted of a small number of independent sources . The number of sources was itself a free parameter of the model , which we explore below . To evaluate the model fits , we held aside 20% of the spike counts from the full set of stimulus presentations , randomly chosen across units and time . The accuracy of model predictions for these held-out data improved substantially with the inclusion of each model component ( Figure 2a ) . Since all scores are cross-validated , these improvements indicate that the models are explaining structure in the data rather than overfitting to random response fluctuations . The figure shows that the shared modulators were twice as important as the experimentally-controlled attentional cue signal in making predictions on held-out data . This improvement is remarkable given that the stimulus presented was identical on every trial: these model components are capturing a substantial portion of the trial-to-trial variability in the population responses . 10 . 7554/eLife . 08998 . 006Figure 2 . The fitted model explains the observed spiking responses , with estimated modulators that are both anatomically and functionally targeted . ( a ) Performance comparison of various submodels , measured as log-likelihood ( LL ) of predictions on held-out data . Values are expressed relative to performance of a stimulus-drive-only model ( leftmost point ) , and increase as each model component ( cue , slow drift , and different numbers of shared modulators ) is incorporated . The grey square shows the predictive LL for a two-modulator model , with each modulator constrained to affect only one hemisphere ( i . e . with coupling weights set to zero for neurons in the other hemisphere ) . This restricted model is used for all results from Figure 2d onwards , excepting the fine temporal analysis of Figure 6c . ( b ) Modulators are anatomically selective . Inferred coupling weights for a two-modulator model , fit to a population of units recorded on one day . Each point corresponds to one unit . As the model does not uniquely define the coordinate system ( i . e . there is an equivalent model for any rotation of the coordinate system ) , we align the mean weight for LHS units to lie along the positive x-axis ( see Materials and methods ) . ( c ) Distribution of inferred coupling weights aggregated over all recording days indicates that each shared modulator provides input primarily to cells in one hemisphere . ( d ) Hemispheric modulators are functionally selective . Units which are better able to discriminate standard and target stimuli in the cue-away condition have larger coupling weights ( blue line ) . Discriminability is estimated as the difference in mean spike count between standard and target stimuli , divided by the square root of their average variance ( d′ ) . Values are averaged over units recorded on all days , subdivided into five groups based on their coupling weights . Shaded area denotes ±1 standard error . Pearson correlation over all units is r = 0 . 42 . This relationship is not seen for the weights that couple neurons to the slow global drift signal ( gray line , Pearson correlation r = 0 . 00 ) . The relationship between d′ and cue weight is significant , but weaker than for modulator weight ( r = 0 . 24 ) ; this is not shown here as the cue weights are differently scaled . ( e ) Same as in ( d ) , but with units subdivided into subgroups according to mean firing rate . Each line represents a subpopulation of ∼500 units with similar firing rates ( from red to blue: 0–7; 7–12; 12–17; 17–25; 25–35; 35–107 spikes/s ) . Within each group , the Pearson correlations between d′ and coupling weight are between 0 . 2–0 . 3 , but the correlations between mean rate and coupling weight are weak or negligible . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 00610 . 7554/eLife . 08998 . 007Figure 2—figure supplement 1 . The dataset is sufficient to support the estimation of up to 8 shared modulators . To test whether the number of recovered modulators is limited by insufficiency of the dataset , we simulated data from model neural populations that were under the influence of different numbers of modulators . All simulated datasets were matched in size to the physiological data , and simulated shared gain fluctuations were adjusted in amplitude to produce the same pairwise spike count statistics as the actual data . We then fit the model to these synthetic datasets , and measured how well the model fits recovered the true modulatory structure . More specifically , we extracted default model parameters by fitting each neural population: the stimulus-driven mean firing rates F , the cue-dependent gains C , and the slow global drifts D . Then , for a given number of simulated modulators K ( from 1 to 12 ) , we sampled a ( T × K ) matrix of time-varying modulator values ( each value i . i . d . Gaussian ) , and a ( K × N ) matrix of random weights ( each weight i . i . d . Gaussian ) , producing a net modulator matrix M as the matrix product of these two . We then sampled spike counts Y from the generative model , Y~Poiss ( F1T⊙exp ( C+D+λM ) ) . For each population and K , we chose the scaling parameter λ such that the median noise correlation between the simulated neurons matched the median noise correlation between the actual neurons . We next fitted the models to these simulated datasets to see how well they recovered the underlying structure . As for the actual data ( Figure 2 ) , we evaluated the model fits via the predictive log-likelihood on held-out data . Each colored line shows the predictive LLs of the fitted models for a given “ground truth” number of modulators . In comparison , the grey squares show the model performance for the actual data . There are two important patterns here . First , for simulated models containing up to 8 modulators , the predictive LLs are greatest when we fit a model having the same number of modulators as the ground truth number used to simulate the data . This demonstrates that the model is in principle able to recover more modulators than the 4 we fit to the actual data . Second , as the number of simulated modulators increases , the ability of the fitted models to make predictions on held-out data declines . This is because the total energy of the shared gain fluctuations is constrained by the measured noise correlations , and is spread amongst the simulated modulators . In this respect , the model predictions on the actual data are most consistent with simulations of 3 or 4 modulators . Finally , it is worth noting that , in simulation , when fitting more modulators than the ground truth , the predictive performance suffers . This reflects overfitting to the noise in the training set . We do not see as pronounced a decline for the model fits to the actual data: instead , the predictive LLs appear to saturate with the number of modulators . This difference between the actual and synthetic data likely reflects our assumption in the simulations that the modulators were all of equal magnitude . A saturation of predictive LL may arise when there is a small set of dominant modulators , and a number of weaker ones . We do not have the statistical power to explore such a long tail of modulatory influences within this dataset , and focus instead on the strongest components . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 00710 . 7554/eLife . 08998 . 008Figure 2—figure supplement 2 . The structure of the modulators in higher-dimensional modulator models . In the main text , we identify three striking properties of the two dominant shared modulators: ( 1 ) they each target one of the two V4 hemispheres; ( 2 ) they preferentially target the task-specific neurons within these hemispheres; and ( 3 ) their variance decreases under cued attention . Here we show that these features are present within higher-dimensional modulator models . In Figure 2—figure supplement 3 , we show that these additional modulatory components do not convey any additional structure in these three domains . It is useful to first view the modulator model as a form of exponential-family Principal Component Analysis ( PCA ) ( Collins et al . , 2001; Solo and Pasha , 2013;Pfau et al . , 2013 ) . Standard PCA , like the shared modulator model , uncovers directions in signal space of maximal variation . However , PCA suffers from an identifiability problem: it can uniquely recover the subspace in which a small set of signals lie , but not the coordinate axes . PCA does select a particular orthogonal coordinate system to represent this subspace , but this solution is not unique , is sensitive to noise , and typically reveals little about the underlying generative process . This same identifiability problem is present with the shared modulator model . In the two-modulator case , we are able to resolve the ambiguity in the coordinate system by exploiting anatomical information ( Figure 2b–c; see Materials and methods ) . However , the problem of identifiability becomes more acute in higher dimensions . Here , we show that the results presented in the main text for the 1-modulator/hemisphere model also hold for the unconstrained 2-modulator model . We also extend the 2-modulator results to the 3- and 4-modulator cases . This is necessary as , unlike standard PCA , the solutions to our equations in lower dimensions do not necessarily lie within subspaces of higher-dimensional solutions . This is because the regularization scheme and algorithm we use create biases that disrupt any strict nesting . We therefore need to explicitly test whether the structures we uncover in the 2-modulator model are also present in the 3- and 4-modulator models . And this needs to be done under the limitations of the identifiability problem , i . e . without choosing a particular coordinate system for the modulation subspace . First column: In Figure 2b–c , we showed that the vectors of modulator coupling weights for LHS units and RHS units in the 2D modulator model were typically orthogonal . Here we show that this holds in higher dimensions . For each recording day , we measured the angle between the average weight vector for LHS units ( w̄L ) , and the average weight vector for RHS units ( w̄R ) , i . e . the arc cosine of their inner product . The 2-modulator hemisphere-constrained model used in most of the main text has this orthogonality enforced by constraint ( top row ) . For the unconstrained 2- , 3- , and 4-modulator models ( remaining rows ) , the blue histograms show the distribution of these angles across recording days . For comparison , we shuffle the anatomical labels on each unit and repeat the analysis to obtain the red histograms . The clustering of the actual data around π/2 indicates near orthogonality of the hemispheric weights . Second column: In Figure 2d , we showed that neurons which were task-relevant ( i . e . had larger d′ values ) were more strongly coupled to the ( 1D ) hemispheric shared modulators . Here , we show that this holds in higher dimensions . For each recording day , we measured the magnitudes of all units’ coupling weight vectors , w2 . Green histograms show the distribution of magnitudes for the quartile of units with largest d′ values; brown histograms show the distribution of magnitudes for the quartile with the smallest d′ values . Third column: In Figure 3b , we showed that the variance of the ( 1D ) hemispheric shared modulators changed according to the attentional cue: specifically , when the cue switched , one hemispheric modulator decreased in variance , while the other increased in variance . To show that this holds in higher dimensions , it is necessary to construct an appropriate metric for this change in second-order statistics that generalizes to higher dimensions , and that also does not depend on a choice of coordinate system . To accomplish this , we measure the effect of the attentional cue as a change in the covariance of the ( multivariate ) modulator . Considering the change from the cue-right to the cue-left condition , we can measure the effect on the modulator’s second-order statistics via the ratio of the two modulator covariances , CcueLCcueR-1 . The eigenvalues of this matrix then provide a coordinate-system-free measure of how the modulator statistics change . If the largest eigenvalue , λmax , is significantly greater than 1 , then there is a direction in modulation space that became more variable due to the switch in cue . If the smallest eigenvalue , λmin , is significantly less than 1 , then there is a direction in modulation space that became less variable due to the switch in cue . Eigenvalues close to 1 indicate that the variance of modulation in that direction was unchanged by the cue . Thus these two values , λmax and λmin , play an analogous role to the ratios of modulator variance examined in Figure 3b . The scatter plots show the distribution of λmax and λmin for the higher-dimensional modulator models . Blue points show these eigenvalues from each recording day; red points show the distributions obtained if we shuffle the cue labels for each trial . Importantly , when λmax exceeds 1 and λmin is less than 1 ( i . e . when the points lie in the lower right quadrant ) , then the change in attentional cue is causing an increase in modulator variance in one direction , and a decrease in modulator variance in an orthogonal direction . These effects are clear ( and significant , compared with the null distribution in red ) in all cases . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 00810 . 7554/eLife . 08998 . 009Figure 2—figure supplement 3 . The modulators' anatomical , functional , and attentional structure manifests primarily within the dominant two dimensions of modulation . In the main text , we recover a set of anatomical , functional , and attentional properties within a 2-dimensional modulator model . In Figure 2—figure supplement 2 , we demonstrate that these properties also manifest within 3- and 4-modulator models . We wondered whether any additional structure can be seen in the extra two dimensions of modulation that we can include beyond the 2-modulator model . To answer this , we partitioned the 4-dimensional modulator space into two 2-dimensional halves . For each recording day , we define a particular 2D subspace of modulation along anatomical grounds . We measured the mean weight vectors for LHS and RHS units , w̄L and w̄R respectively . As shown in Figure 2—figure supplement 2 , these two vectors were always near-orthogonal . We define their 2D span as the “hemispheric subspace of modulation” , H=span ( w¯L , w¯R ) . This is a 2D subspace of the 4D weight vectors , capturing the largest component of hemispheric-specificity in the modulator weights . What remains in the 4D modulation space is the hemispheric subspace’s orthogonal complement , H⊥ . This divides the 4D space into two 2D subspaces , and thus amounts to a partial choice of a coordinate system . We can therefore study the anatomical and functional properties of the coupling weights in H and H⊥ , and also the attention-dependent statistics within the corresponding 2D spaces of time-varying modulator values . This panel shows that all three properties described in Figure 2—figure supplement 2 manifest predominantly in H , but not in H⊥ . In summary , each V4 hemisphere is being driven by a shared modulatory signal , that preferentially affects task-specific neurons , and has statistics that depend on the attentional cue provided to the animal . In addition , there is some evidence that other shared modulatory factors are affecting the population of V4 neurons . However , these latter signals do not share the same properties: their net effects are weaker , they do not appear to have the same anatomical or functional specificity , and they do not appear to be affected by the attentional cue . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 00910 . 7554/eLife . 08998 . 010Figure 2—figure supplement 4 . Units with higher mean firing rates typically had stronger coupling to their respective population modulator ( r2 = 0 . 21 ) . This observation motivates the control analyses shown in Figure 2e , Figure 5—figure supplement 1 and Figure 8—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 010 We next asked how many independent shared modulators were needed to explain the recorded responses . In principle , the activity of the recorded populations could be influenced by a very large number of shared modulators , each with its own unique connectivity and temporal patterns . We found that the spike count predictions on held-out data improved substantially with the inclusion of two modulators , and showed a modest additional increase for up to four modulators . Including additional modulators beyond these did not improve predictions . To verify that this outcome did not reflect insufficiency of the dataset in constraining the model , we generated synthetic datasets by simulating responses of the model with different numbers of modulators , each of equal strength ( Figure 2—figure supplement 1 ) . These synthetic datasets had the same number of neurons and trials as the true data , and we scaled the magnitude of the simulated modulators’ fluctuations to produce the same average noise correlations as found in the true data . We found that we could accurately recover up to 8 independent modulators from these synthetic datasets – more than the number recovered from the true data . We next asked how the estimated modulators , and their associated weights , were structured with respect to the neural populations and the task . These structures are most easily visualized for the two-modulator model , and we therefore restrict our analysis to this case for the remainder of this article . Results for three- and four-modulator models are given in Figure 2—figure supplement 2 . Several striking patterns emerge in the fitted model . First , although the model was not given any information regarding anatomical location or connectivity of the neurons , the estimated coupling weights for each modulator clearly identify the hemisphere in which the corresponding neurons reside ( Figure 2b–c ) . In a given recording , each modulator had largely positive weights for neurons in one hemisphere ( indicating that these neurons’ gains were being co-modulated by this signal ) , and small weights for neurons in the opposite hemisphere . Thus , the weight structure of the estimated model suggests that the two hemispheric V4 subpopulations are modulated by two independent signals . Based on this observation , we examined a restricted model , in which we explicitly enforce the hemispheric assignment of each modulator ( from here on , referred to as the LHS and RHS modulators ) , by setting the weights for neurons in the opposite hemisphere to zero . This enforced assignment reduces the number of weight parameters by a factor of two , and results in a modest improvement in the quality of model predictions on held-out data , reaching the level of the 4-modulator model ( Figure 2a , grey square ) . For the remainder of this article , we retain this enforced assignment of the modulators to the hemispheres . In addition to the distinct anatomical connectivity of the two modulators , we found that they also exhibited specific functional connectivity . We characterized the task-relevance of each neuron in terms of its ability to discriminate the standard and target stimuli in the cue-away ( inattentive ) condition . Specifically , for each neuron , we computed the difference in mean spike count for the two stimuli , relative to the standard deviation ( known in the perceptual psychology literature as d′ ) . Comparison of these values to the modulator coupling weights indicates that the modulators preferentially targeted the most task-informative neurons ( r = 0 . 42; Figure 2d ) . One might suspect that some portion of this effect is simply due to firing rate—units with higher mean firing rates typically had stronger coupling weights to the modulators , paralleling previous observations ( Cohen and Maunsell , 2009; Mitchell et al . , 2009; de la Rocha et al . , 2007; Ecker et al . , 2010 ) ( Figure 2—figure supplement 4 ) . But the relationship between task informativeness and modulator coupling remains robust even when conditioned on firing rate ( Figure 2e ) . Finally , this correlation with functional specificity was weaker for the coupling weights to the cue signal ( r = 0 . 24 ) , and entirely absent from the coupling weights to the slow global drift signal ( r < 0 . 01; Figure 2d ) . The model thus far reveals the action of two structured modulatory signals , each providing input to task-informative V4 neurons in one of the two hemispheres . Given that these signals were recovered from population activity during a cued attentional task , we next ask whether they exhibit systematic changes across differently cued blocks . The model recovers estimates of the slow drift signal and each hemisphere’s modulator for every stimulus presentation ( Figure 3a ) . Here , a clear pattern is evident: when the monkey was cued to attend to one visual hemifield , the shared modulator of the corresponding ( contralateral ) V4 population had a smaller variance ( Figure 3b ) . The modulators’ mean values were unchanged across blocks , as any changes in mean rate are captured by the coupling to the cue signal: 75% of neurons had a positive coupling weight to the cue signal , capturing an increase in their firing rate under cued attention . Thus , in general , attention both increases and stabilizes the time-varying gain of the corresponding neural population . 10 . 7554/eLife . 08998 . 011Figure 3 . Time-varying model signals that determine the gain of units in one V4 hemisphere . ( a ) Example values of the cue signal ( imposed by experiment ) , the slow drift ( inferred ) , and a single hemispheric modulator ( inferred ) across stimulus presentations for one day and hemisphere . In the model , the gain of each neuron is obtained by exponentiating a weighted sum of these three signals ( see Equation ( 1 ) ) . Histogram in the bottom left shows the distribution of modulator values when the monkey was cued towards the contralateral side ( blue ) , and away from it ( red ) . ( b ) Modulator variance decreases under cued attention . Histogram shows the ratio of modulator variances estimated in the two cue conditions . Averaged across days and hemispheres , cued attention reduces modulator variance by 23% . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 011 These two changes in shared gain , in turn , provide a simple explanation for the observed changes in the statistics of individual and paired neural spike counts . Consider first a simulation of two conditionally-Poisson neurons . In the classical model of attention , these neurons’ gains increase when the cue is directed to the appropriate hemifield . This produces two major effects on each neuron’s marginal spike count statistics ( Figure 4a ) : the mean increases , and the variance goes up as well ( due to the Poisson mean-variance relationship ) . The Fano factors ( ratio of variance to mean ) remain unchanged . 10 . 7554/eLife . 08998 . 012Figure 4 . Changes in the statistics of the inferred modulator under cued attention explain the observed changes in spike count statistics . ( a ) The classical model of attention . Simulation of two neurons with positive coupling weights , u1 and u2 , to the cue signal . When the cue is directed to the corresponding spatial location ( top ) , both the mean and variance of the simulated neurons’ spike counts increase ( bottom ) . Shaded areas demarcate analytic iso-density contours , i . e . the shape of the joint spike count distributions . ( b ) The effect of the shared modulator . Simulation of two simulated neurons with positive coupling weights , w1 and w2 , to a shared modulator . A decrease in modulator variance leads to a decrease in both the variance and correlation of spike counts ( bottom ) . ( c ) Effects on an example pair of units within the same hemisphere , on one day of recording . The cue increases the gain of both cells ( numbers indicate cue coupling weights ) , and the inferred modulator exhibits a decreased variance in cued trials ( again numbers indicate coupling weights; top ) . The spiking responses of the cells exhibit a combination of the effects simulated in ( a ) in ( b ) : increased mean , decreased Fano factor , and decreased correlation ( bottom; means from 7 . 0 to 8 . 1 and 6 . 0 to 7 . 2 spikes/stim respectively; Fano factors from 1 . 9 to 1 . 6 and 1 . 7 to 1 . 6 respectively; correlation from 0 . 19 to 0 . 10 ) . The shaded areas demarcate smoothed iso-density contours estimated from the data . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 012 Next , consider what happens if these simulated neurons are coupled to the shared modulator ( Figure 4b ) . A decrease in the variance of this shared signal leads to a reduction in the spike count variance of each neuron , without a large change in their mean firing rates ( Goris et al . , 2014 ) . Consequently , their respective Fano factors decrease . Moreover , since both neurons are coupled to the same modulatory signal , the decrease in modulator variance also causes a decrease in the spike-count correlation of the pair . The data recorded in the experiment exhibit a combination of these effects . An example is shown in Figure 4c . Under cued attention , the variance of this V4 hemisphere’s shared modulator decreases . When we consider two neurons within this population with strong coupling weights to both the cue signals and the modulators , their marginal and joint spike count statistics are changed by attention as predicted by the simulation: their mean firing rates go up , their Fano factors go down , and their noise correlations decrease . The behavior shown in Figure 4 crucially depends on the two neurons being coupled to the shared modulator . The model thus makes a very specific prediction: the magnitudes of the Fano factors and noise correlations should increase with the magnitude of the weights with which the neurons are coupled to the hemispheric modulators . We find that this prediction is clearly borne out by the data ( Figure 5a ) , and is robust when controlled for firing rate ( Figure 5—figure supplement 1a ) . In addition , since the modulator variance decreases under cued attention , the model predicts that those neurons that are more strongly coupled should exhibit a larger attention-induced reduction in these measures . This effect is also apparent in the data ( Figure 5a ) . Ultimately , the model accounts for the majority ( but not all ) of the attention-induced changes in Fano factor and noise correlation ( Figure 5b ) . Thus , our population-level model accounts not only for the single neuron statistics and pairwise correlations , but also for the diversity of effects seen across the population . 10 . 7554/eLife . 08998 . 013Figure 5 . Attention-induced changes in neural response statistics are larger for neurons that are more strongly coupled to the shared modulator . ( a ) Observed Fano factor and noise correlations , as a function of model coupling weight . Units from all days are divided into five groups , based on their fitted coupling weight to their respective population modulator ( model-based quantities ) , as in Figure 2d . Points indicate the average Fano factors and noise correlations ( model-free quantities ) within each group , when attention was cued towards the associated visual hemifield ( blue ) and away from it ( red ) . Shaded area denotes ± 1 standard error . Unitwise Spearman correlations: ρ = 0 . 31/0 . 44/0 . 40/0 . 51 ( fano cue twds/fano cue away/ncorr cue twds/ncorr cue away ) . ( b ) Comparison of model-predicted vs . measured decrease in Fano factor and noise correlation . Units are divided into ten groups , based on coupling weights ( darker points indicate larger weight ) . The model accounts for 62% of the cue-induced reduction in Fano factor , and 71% of the reduction in noise correlation . ( c ) Comparison of cue weights and modulator weights . Units that are strongly coupled to the cue signal are typically strongly coupled to the modulator signal , though the relationship is only partial ( unitwise Spearman correlation: ρ = 0 . 26 ) . These results are robust when controlled for firing rate ( Figure 5—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 01310 . 7554/eLife . 08998 . 014Figure 5—figure supplement 1 . Differences in coupling weight explain the observed statistics of single and pairwise firing rates , even when controlled for mean firing rate . The relationship shown in Figure 5 contains a potential confound in that units with higher mean firing rates are typically more strongly coupled to their corresponding shared modulators ( Figure 2—figure supplement 4 ) . Here , we perform the same analyses on subpopulations with similar mean firing rate . ( a ) We repeated the analysis of Figure 5a , but subdivided the total population of units in two ways: first , by mean firing rate into six groups ( rows ) , and then by coupling weight into five subgroups ( points on each plot ) . Each row thus replicates Figure 5a for a controlled subpopulation of approximately 500 units with similar firing rates . Within each group , the correlation between mean rate and modulator coupling weight was weak or negligible . Nevertheless , the relationships of Fano factor and noise correlation to modulator weight remain . ( b ) We also repeated the analysis of Figure 5b , subdividing the total population of units by mean firing rate into six groups , as in the rows of ( a ) . Again , the relationship between cue and modulator weights remains . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 014 The model does not assume any relationship between the coupling weights for cue and shared modulators . We did observe that neurons which were most strongly coupled to the modulators were also most strongly coupled to the cue signal ( Figure 5c; even when controlled for firing rate Figure 5—figure supplement 1b ) . But overall , the correlation between the coupling weights to the cue signal and the coupling weights to the hemispheric modulator was modest ( Spearman ρ = 0 . 26 ) . We fit a restricted model in which these two signals had identical coupling weights , and found that the predictive ( log-likelihood ) performance was reduced by 20% . Moreover , while the modulator weights were predictive of the Fano factor and noise correlation effects ( as shown in Figure 5a; correlations ρ > 0 . 3 ) , the cue weights were much less so ( ρ < 0 . 1 ) . Thus , although the effects of the cue and of the inferred hemispheric modulator overlap , they are not identical , suggesting that they arise ( at least in part ) from distinct sources . We made three additional observations regarding the shared modulators . First , the two hemispheres’ time-varying modulator values were almost completely uncorrelated ( Figure 6a; r = 0 . 03 ) . The very small positive correlation might result from the influence of a global signal spanning both hemispheres . Second , the modulators exhibited correlations over successive stimulus presentations ( Figure 6b ) , indicating that gain fluctuations can persist over time scales of seconds . Third , while the data were insufficient to infer the value of the modulators at a sub-trial resolution ( i . e . at time scales shorter than 200 ms ) for individual trials , we were able to estimate their average sub-trial time course . We found that the shared modulators had only weak effects during response onsets , and exerted their influence on neural gain primarily during the sustained response period ( Figure 6c ) . This property of the modulators is consistent with previous reports that attentional effects on firing rates ( and Fano factors ) are greater during sustained periods ( Cohen and Maunsell , 2009;Mitchell et al . , 2009 , and with more general observations that network behavior is less affected by context or state during onset transients ( Churchland , 2010 ) . 10 . 7554/eLife . 08998 . 015Figure 6 . Statistical properties of the hemispheric modulators . ( a ) Joint statistics of the two hemispheric modulators . Blue points: simultaneous values of the two modulators aggregated over all days . Thick black ellipse: iso-density contour at one standard deviation of the Gaussian density matching the empirical covariance . Thinner black ellipse: two standard deviations . Dashed lines: principal axes ( eigenvectors ) of this covariance , with the thicker dashed line indicating the axis with the larger eigenvalue . The vertical elongation of the ellipse shows that the variance of the modulator for the cued side is smaller than the variance of the modulator for the opposite side . The slight clockwise orientation shows that the two modulators have a very small positive correlation ( r=0 . 03 , p negligible ) . ( b ) Autocorrelation of modulators across successive stimulus conditions . Individual lines show the within-block autocorrelation of each estimated modulator; the thick lines shows the average across days and hemispheres . For simplicity of presentation , the targets and the gaps between trials have been ignored . The time constant of this process is on the order of several seconds . ( c ) Average time course of shared modulation within each stimulus presentation . We extended the population response model by allowing the value of the modulator to change over the course of each stimulus presentation . Given limitations of the data at fine temporal resolutions , we assumed that the temporal evolution of the modulator within each stimulus presentation followed some stereotyped pattern ( up to a scale factor that could change from one stimulus presentation to the next; see Materials and methods ) . Fine blue lines: modulators’ ( normalized ) temporal structure extracted for each recording day . Heavy black line: average across days . Grey shaded area: normalized peri-stimulus time histogram ( arbitrary units ) of spiking responses during presentations of the standard stimuli , averaged across all units , days , and cue conditions , with zero denoting spontaneous rate . Shared modulation predominantly occurs during the sustained period and is nearly absent during the onset transient . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 015 Finally , we note that the relationships between attention and modulator statistics were not only evident in the hemispheric-modulator model , but were also present in models with more modulators ( Figure 2—figure supplement 2 ) . Nevertheless , these additional modulators did not reveal any additional anatomical or functional specificity , or attentional dependencies ( Figure 2—figure supplement 3 ) . We have thus far used the model to expose a set of internal modulatory signals that selectively affect the gains of neurons in the population . We wondered whether these modulatory signals had any effect on behavior . For clarity in describing these results , we use the terms “cued” and “opposite” rather than “left” and “right” to describe the visual hemifields , V4 populations , modulators , and targets . For example , if the monkey was cued to the left during a block , then the left hemifield is cued , the monkey’s right ( i . e . contralateral ) V4 is cued , and the right V4 population’s modulator is the cued modulator . If , during a trial in this block , a target is presented on the right , we refer to it as an opposite target . First , we asked whether the shared modulators had any influence on the monkey’s trial-by-trial performance . We found that the values of both the cued and the opposite modulators , during presentations of the standard stimuli , were predictive of whether the monkey would detect the upcoming target stimulus . As may be expected , changes in each hemispheric modulator predicted performance changes for their associated targets: an increase in the cued modulator preceded an increased detection probability for cued targets; while an increase in the opposite modulator preceded higher detection probabilities for opposite targets ( top left and bottom right of Figure 7a; see Figure 7—figure supplement 1a for full psychometrics ) . Suprisingly , this effect is substantially stronger for the opposite side . This result is similar to a previous study on this dataset aimed at directly decoding the trial-by-trial attentional state of the animal ( Cohen and Maunsell , 2010 ) , though the two results rely on different readouts of the population activity . 10 . 7554/eLife . 08998 . 016Figure 7 . Inferred modulatory signals are predictive of behavioral performance , and are influenced by previous reward . ( a ) Average effect of modulator values on subsequent behavioral performance , averaged across all days and difficulty levels . Values show the average change in hit probability for targets on the cued side ( left column ) and the opposite side ( right column ) following a unit increase in the cued ( top row ) and opposite ( bottom row ) modulators . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 . Full psychometric curves are shown in Figure 7—figure supplement 1a . ( b ) Average effects of previous trial reward on current trial performance . Note that this is a direct comparison of the behavioral data , and does not involve the modulator model . ( c ) Average effects of previous trial reward on the value of the two hemispheric shared modulators . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 01610 . 7554/eLife . 08998 . 017Figure 7—figure supplement 1 . Relationship between modulators and behavior: additional details . ( a ) Psychometric performance , averaged across all days . These plots expand the results of Figure 7 showing the interacting effects of trial difficulty and the two hemispheric modulators on task performance . The hit probability is shown as a function of the orientation change in degrees for trials where the target was on the cued side ( black points ) or the opposite side ( single gray point; opposite side targets were only presented at 12 deg ) . For each condition , the color gradient shows the effects that values of the cued modulator ( left panel ) and opposite modulator ( right panel ) have on the hit probability . We fit a family of psychometric curves to the cued-target conditions , with the two modulator values as regressors; the colored lines in each panel show two of these curves , indicating the biasing effect of mcued= ± 3σcued ( left panel ) , and mopp= ± 3σopp ( right panel ) on performance . ( b ) Left: average effects of previous trial reward on current trial performance , from Figure 7b . Note that this is a direct comparison of the behavioral data , and is not dependent on the modulator model . Right: effect of reward on performance predicted by chaining together the effects of reward on modulator ( Figure 7a ) and modulator on performance ( Figure 7c ) . The biasing effect of reward on behavior , as mediated by the V4 modulators , is consistent with the observed data ( left ) , but captures only a relatively small proportion of the total reward bias ( ∼5–10% ) . To estimate the total behavioral reward bias , we fitted Bernoulli-GLMs ( i . e . GLMs with a Bernoulli observation process ) which predict the response ( hit/miss ) , given the previous trial’s reward ( hit for target on cued side/hit for target on opposite side/other ) as regressors . When the current trial’s target was cued , we treated the orientation change as an additional regressor , and we included a lapse parameter as behavioral performance typically saturated below 100% correct ( Wichmann and Hill , 2001 ) . The effect of previous reward in this model manifests as a bias term within the sigmoid ( logistic ) nonlinearity . To estimate the V4-mediated reward bias , we measured how large these total behavioral reward biases were if they had to pass through the “bottleneck” of the V4 modulators . We thus fitted three GLMs: a Gaussian-GLM which predicts the cued modulator on a trial , given the previous reward ( and the previous modulator values ) ; a second Gaussian-GLM which predicts the opposite modulator on a trial in the same way; and a Bernoulli-GLM which predicts the response ( hit/miss ) , with the two modulators on that trial as a regressor . By multiplying these two effects together ( the average change in modulators due to each previous reward state in the first and second GLMs , with the modulator coefficients in the third GLM ) , we obtained the desired quantities . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 017 In addition to this modulator-driven improvement in detection probabilities , we also uncovered a striking deficit in performance on the opposite side . Specifically , an increase in one side’s modulator predicted a decrease in detection probability on the other side ( top right and bottom left of Figure 7a ) . This is not due to anti-correlation of the two modulators ( they are nearly uncorrelated; Figure 6a ) . Rather , this pattern likely reflects the competitive structure of the task , which requires the animal to make comparisons between the represented stimuli in both hemispheres . The influence of V4 activity on detection probability in this task thus cannot be fully explained by a 1D “axis of attention” ( Cohen and Maunsell , 2010 ) , but depends on the joint ( 2D ) gain modulation across the two hemispheres . Second , we examined the relationship of the modulator values to the outcome of the previous trial . It is well known that both humans and animals show serial dependence in behavioral tasks: our choices in a task are biased by the percepts , actions and outcomes of previous experience ( Senders and Sowards , 1952; Green , 1964; Lau and Glimcher , 2005; Busse et al . , 2011; Fischer and Whitney , 2014 ) . The monkeys in this task also exhibit a serial dependence: after receiving a reward for correctly identifying a cued target , they were more likely to score a hit for a cued target , and a miss for an opposite target , on the next trial . Conversely , after receiving a reward for correctly identifying an opposite target , on the next trial they are substantially more likely to score a hit for an opposite target , and a miss for a cued target ( Figure 7b ) . This sequential bias is explicitly captured in the modulatory signals controlling the gains of the V4 populations . A hit ( and thus , a reward ) on a cued target biased the two modulators on the subsequent trial to shift in favor of the cued population , while a hit on the opposite target biased the two modulators on the subsequent trial to shift in favor of the opposite population ( Figure 7c ) . Again , the effect is more substantial for the modulator opposite the cue . This effect is consistent ( in sign ) with the biasing effects of previous reward on the animal’s behavior . However , we note that only 5–10% of the total serial dependence can be explained through the V4 modulators ( Figure 7—figure supplement 1b ) . We conclude that reward must also induce additional biases in the activity of downstream neurons involved in the animals’ decisions . In summary , we found that the inferred signals that modulate neural gain in V4 populations are intimately connected with the monkey’s previous and subsequent behaviors .
Our findings are consistent with previous reports that , under attention , the activity of single V4 neurons changes in relation to aggregate activity . Specifically , attention reduces the correlation between spiking and concurrent slow fluctuations in local field potential ( LFP ) ( Fries et al . , 2001 ) , which themselves are reduced in power ( Fries et al . , 2001; Siegel et al . , 2008 ) . These behaviors are predicted by our model as a simple consequence of the attention-driven changes in modulator variance . Since attention reduces the variance of the shared modulator , it would also be expected to reduce the variance of summed activity across the population , such as that measured in LFPs . In addition , since attention reduces the proportion of neural variance due to shared modulation , the correlation between single neuron spike counts and the aggregate LFP activity would also be reduced ( Goris et al . , 2014 ) . A primary assumption of our model is that the patterns of shared response variability arise from stochastic signals that modulate the gain of sensory neurons . The multiplicative nature of this interaction is broadly consistent with the patterns of response variability seen throughout visual cortex , which show signatures of multiplicative , rather than additive noise ( Goris et al . , 2014 ) . Nevertheless , other modeling efforts have found evidence for shared additive noise ( Lin et al . , 2015 ) , and we cannot fully rule out the possibility that additive noise also contributes to the variability in our data . A strong test of the multiplicative assumption requires analysis of neural responses over a range of stimulus drive levels . In the dataset analyzed here , reliable responses to only one stimulus per neuron are available for analysis ( target responses have not been included , since they are corrupted by co-occurring saccades ) . Nevertheless , the modest reduction in firing rates arising from adaptation to repeated presentations of the standard stimuli provides some opportunity to examine this question . We find that the patterns of response variance along this axis are indeed consistent with the multiplicative hypothesis ( Figure 1—figure supplement 2 ) . A more definitive comparison of additive vs . multiplicative interactions could be achieved with more extensive experimental manipulations of stimulus parameters . One of us recently reported that pairs of neurons within the same hemisphere whose responses provide evidence for opposite behavioral choices can exhibit increased noise correlations under spatial attention ( Ruff and Cohen , 2014 ) . At first glance , this result seems at odds with the attention-induced reduction of variance in the inferred modulators of our model , which generally leads to a decrease in noise correlations . However , a plausible explanation may arise from considering that populations are not just affected by attention-dependent sources of modulation , but also by additional sources that are attention-independent . In the datasets we study here , we identify two examples: the global slow drifts ( Figure 1—figure supplement 1 ) , and the extra , non-hemispheric , fast modulators ( Figure 2—figure supplement 3 ) . If the two subpopulations responsible for encoding the two visual stimuli each have their own , separate , attention-dependent modulator , while both populations are subject to an additional set of common , attention-independent modulations , then attention might serve to “unmask” the cross-population correlations arising from these common modulatory signals . Preliminary simulations suggest that such a model could explain these observations . Finally , the model fits indicate that slow , global drifts and structured , rapid fluctuations in shared gain make substantial contributions to the super-Poisson variability of these neurons . However , the model does not account for all of the observed cue-dependent changes in variability ( Figure 5b ) , suggesting that the model structure ( e . g . Poisson spiking , with rate modulated by an exponentiated sum of shared signals ) is too restricted to capture the full extent of the effects . Moreover , the dimensionality and structure of the estimated modulators are limited by the experiment itself ( Gao and Ganguli , 2015 ) , and thus cannot be taken as a complete account of modulatory activity in V4 . The behavioral task is designed to engage two attentional conditions , each associated with stimuli presented over extensive blocks of trials in one hemifield of the visual world . Similarly , the stimuli in the task provide only a minimal characterization of the selectivity of individual neurons ( e . g . receptive field locations , tuning properties ) . We expect that a more complex task , in which attention is precisely focused in terms of visual location , stimulus properties , and/or time , could potentially reveal modulatory signals targeting more specific subpopulations of neurons . Our model is functional , and like other functional models of neural responses , such as those based on Generalized Linear Models ( Truccolo et al . , 2005; Pillow , 2008 ) , its components are phenomenological rather than biophysical . The principal value of these models is in providing a parsimonious quantitative framework that explains the relationship between stimuli , neural responses , and behavior . In this respect , a primary contribution of our work is to show that the patterns of neural variability in a visual cortical population during this attentional task are dominated by a few internal signals . This need not have been the case: the model could have required many more modulators to explain the structure of observed population activity , or the imposed structure of the model might have proven inappropriate or insufficient to provide an account of the data . Notwithstanding the abstraction of the functional model , the inferred properties of the low-dimensional modulatory signal provide constraints on potential underlying mechanisms . We can envision three broad scenarios for the mechanistic source of shared gain fluctuations: bottom-up ( stimulus-driven ) input , recurrent activity within the population , and top-down state signals . Each of these mechanisms has previously been proposed as an explanation for excess neural response variance or noise correlations ( Goris et al . , 2014; Ecker et al . , 2014; Ecker et al . , 2012; Ecker et al . , 2010; Zohary et al . , 1994; Shadlen et al . , 1996; Shadlen and Newsome , 1998; Nienborg and Cumming , 2009; Yang and Shadlen , 2007; Arieli et al . , 1996; Litwin-Kumar and Doiron , 2012 ) . The latter two mechanisms seem compatible with the low-dimensional nature ( Figure 2a ) , spatial scale ( Figure 2b–c ) , and temporal scale ( Figure 6 ) of the modulatory signals we have identified . Top-down and recurrent mechanisms need not operate independently: recent theoretical work proposes a combination of these two mechanisms , in which top-down modulation induces a change in the balance of excitability in a recurrent E-I network ( Wiese et al . , 2014 ) . One potential top-down biological substrate is cholinergic input , which has a multiplicative ( Disney et al . , 2007 ) and hemispherically-specific ( Mesulam et al . , 1983 ) effect on neurons in macaque visual cortex , and is known to be enhanced under attention ( Herrero et al . , 2008 ) . However , it has recently been shown that in area V1 , cholinergic input mediates attention-dependent increases in mean firing rate , but not changes in response variance or noise correlations , which appear to depend on NMDA receptors ( Herrero et al . , 2013 ) . Consistent with these findings , our model suggests that the attention-mediated increase and stabilization of gain in area V4 might arise from two separate mechanisms: the estimated coupling weights associated with the cue and modulator signals are similar but not identical ( Figure 5c ) , and when they are forced to be identical , model predictions worsen . A full mechanistic account of the modulators could be developed in future experiments by comparing the estimated modulator signals against other neural or physiological signals ( e . g . neurotransmitter concentrations , or pupil dilation ( McGinley et al . , 2015; Reimer et al . , 2014 ) , and by comparing the weights with detailed anatomical and functional properties of the neurons in the population . We have used a model that extracts patterns of shared gain variability within a neural population , encapsulating them as “modulators” . In turn , we have examined the modulators’ statistical structure , and found that it is consistent with a variety of externally measured or controlled quantities . But we are left with an outstanding question: should we think of these modulators as signal or noise ? That is , do they reflect a controlled endogenous process , or are they random fluctuations that arise because of a lack of control , and are thus detrimental to the encoding of incoming stimuli ? The prevailing view in the neural coding literature is the latter . There has been extensive debate about how correlated fluctuations in neural activity can confound sensory information , and can be difficult to remove once introduced . From this perspective , any process that reduces such fluctuations would improve neural coding ( Cohen and Maunsell , 2009; Mitchell et al . , 2009; Gu et al . , 2011; Jeanne et al . , 2013; Downer et al . , 2015; Zohary et al . , 1994; Abbott and Dayan , 1999; Sompolinsky et al . , 2001; Wilke and Eurich , 2002 ) . The benefits of attention would thus be two-fold ( Figure 4 ) : by increasing the mean gain of a relevant neural population , attention would increase the signal-to-noise ratio ( since , for Poisson spike counts , the response standard deviation grows only with the square root of the mean ) ; and by simultaneously reducing the variance of the gain , attention would reduce the overall response variance ( Goris et al . , 2014 ) , and thus the deleterious effects of this nuisance variable . If this hypothesis is correct , we would expect that gain variability in V4 should follow a general pattern . In the absence of directed attention , all neurons in V4 should be subject to some baseline level of gain variability , due to activity in locally shared circuits . This baseline should presumably be relatively independent of which stimulus features are encoded by which neurons . When attention is directed to a particular neural subpopulation , the shared variability should decrease , improving the coding precision of that subpopulation ( Figure 8a ) . 10 . 7554/eLife . 08998 . 018Figure 8 . Interpreting the role of shared modulation . ( a ) Illustration of how shared gain fluctuations would behave if they were noise , i . e . undesirable random fluctuations . In baseline conditions ( red ) , gain fluctuations would be expected to have similar variance for all neurons in the V4 population . The action of attention would be expected to reduce the variance of gain fluctuations in task-relevant neurons , so as to mitigate their adverse effect on coding precision ( see Figure 4 ) . ( b ) Contrary to this simple “noise” interpretation , the variance of shared gain fluctuations are markedly larger for task-relevant neurons than task-irrelevant neurons in baseline ( cued away ) conditions . Moreover , although this variance decreases under attentional cueing ( cued toward ) , it remains larger for the task-relevant neurons . Functional relevance for each unit is measured as d′ ( as in Figure 2d–e ) ; shared gain variability , σg2 , is measured as the total variance of model-estimated gain fluctuations ( from slow drift and modulators combined ) . These results are robust when controlled for firing rate ( Figure 8—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 01810 . 7554/eLife . 08998 . 019Figure 8—figure supplement 1 . Variance of shared gain fluctuations is larger in task-relevant neurons , even when controlled for firing rate . The relationship shown in Figure 8 contains a potential confound in that units with higher mean firing rates are typically more task-relevant , and also exhibit larger gain fluctuations . Here , we perform the same analyses on subpopulations with similar mean firing rate ( subpopulations as in Figure 5—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08998 . 019 This , however , is not the pattern we observe . The gain fluctuations we uncover do not affect every neuron in the population , but specifically target those neurons relevant to the task ( Figure 8b ) . And while their variance certainly decreases under attention , it remains a puzzle why these gain fluctuations are there at all , given that they are nearly absent from the task-irrelevant neurons ( which , presumably , would be relevant for different tasks ) . This finding suggests that the fluctuating modulatory signals are not noise , but rather reflect meaningful intrinsic signals that play a role in some ongoing computation in the brain . For instance , one tempting possibility is to identify them as fluctuations in the attentional signal itself ( e . g . changes in the spatial locus of attention ) ( Goris et al . , 2014; Cohen and Maunsell , 2010; Harris and Thiele , 2011; Ecker et al . , 2012 ) , or , more generally , as the local manifestation of a dynamic resource allocation strategy . This might reflect a shifting belief-state of the animal in the likelihood or utility of each of the two targets , as proposed by a recent model ( Haefner et al . , 2014 ) . Such an account might explain the modulators’ targeting of task-relevant neurons , their low-dimensional structure , and their connection to behavior . Regardless of what we call this endogenous signal , it is worthwhile considering whether the gain fluctuations it produces pose a problem for downstream interpretation of the neural code . We can envisage three scenarios . First , the downstream neurons that decode the V4 activity may be invariant to the shared gain fluctuations [for example , in a linear decoding framework , the gain fluctuations might lie parallel to the decision boundaries ( Moreno-Bote et al . , 2014 ) ] . However , this does not appear to be the case: modulator fluctuations have a direct and immediate effect on the perceptual decisions made by the animal ( Figure 7a ) . A second scenario is that downstream neurons might have access to the fluctuating shared gain values , and could thus compensate for their biasing effect on perceptual decisions ( e . g . Stevenson et al . , 2010 ) . This does not imply that the gain fluctuations would have no effect: they would still alter the instantaneous signal-to-noise ratio of the neural representation . This would predict that when a modulator increases , detection probability on the corresponding side would improve ( as we see in the top left and bottom right of Figure 7a ) . However , the observation that the value of a modulator also affects performance on the other side ( top right and bottom left of Figure 7a ) is inconsistent with this hypothesis . Any downstream decoder must compare the responses of the two hemispheres in order to perform the task . Thus , the unexpected impact of each hemisphere’s modulator on ipsilateral performance demonstrates that the decoder is unable to fully discount the fluctuations in gain . We are left with a third scenario , in which the downstream neurons cannot perfectly compensate for the shared gain fluctuations in V4 . Considered in isolation , shared modulation would thus be detrimental to coding precision . We might speculate that this negative would be offset by some other ( as of yet , unknown ) benefit of a flexible , varying gain signal . Overall , we find that the effects of attention on the activity of V4 neurons , while seemingly complex at the level of observed responses , reflect population-level patterns that are simple and low-dimensional . This contributes to an emerging trend in systems neuroscience , wherein the dynamics and statistics of large populations have been found to follow structured , coordinated patterns ( Yatsenko et al . , 2015; Okun et al . , 2015; Mazor and Laurent , 2005; Broome et al . , 2006; Mante et al . , 2013; Stokes , 2013; Kaufman et al . , 2014; Cunningham and Yu , 2014 ) . These findings have been driven both by experimental advances in simultaneous acquisition of responses over populations , as well as the development of statistical modeling tools for isolating low-dimensional latent components of population activity ( Kulkarni and Paninski , 2007; Paninski et al . , 2010; Macke , 2011; Vidne et al . , 2012; Archer et al . , 2014 ) . Here , we have shown that these tools may be used to expose and characterize signals that underlie modulatory processes arising under attention .
We fitted probabilistic models to population neural responses . For each recording day , we collect the spike counts from the N neurons over the T stimuli as a ( T × N ) matrix , Y . We assume that each count was drawn from a Poisson distribution , with the ( T × N ) rates obtained from the model response rn ( t ) as defined in Equation ( 1 ) . It is convenient here to express the rates in matrix form , R , such that Yt , n~ Poiss ( Rt , n ) . We describe the components and fitting procedures below . Each neuron n has a mean firing rate for the stimulus , fn ( st ) . Since we only analyze responses to the standards , this is a single scalar value per unit ( though see section “Fine temporal analysis of shared modulation” below ) . We define this to be the mean rate in the cue-away condition ( with the exception of Figure 2a , where the stimulus-drive-only model uses a mean rate across all conditions ) . This mean rate is modulated by a set of multiplicative factors . Since these factors have to be positive , we adopt the convention that they are derived from signals that can be positive or negative , which are then transformed elementwise by a nonlinear function h that is monotonic and has positive range . Choosing h as convex and log-concave simplifies inference further ( Paninski , 2004 ) . The results presented in the main paper use the exponential function exp ( · ) , but we found that using the soft-threshold function log ( 1 + exp ( · ) ) produces qualitatively consistent results . When using the exponential function , the notation simplifies , and the product of multiplicative factors may be written as a single exponential of a sum , as in Equation ( 1 ) . We now describe the gain factors introduced in Equation ( 1 ) . We define the cue signal during the experiment as a length-T binary vector , c , indicating the cue direction for each trial . Then we can write the elements of the rate matrix as Rt , n=fn ( st ) ·exp ( ct un ) , where un is a coupling weight of neuron n to the cue signal . Here , it is useful to define the cue signal differently for units in the two hemispheres , with ct = 1 indicating that the cue is directed to a hemisphere’s corresponding ( contralateral ) hemifield , and ct = 0 otherwise . Thus a positive coupling weight un means that a neuron increases its gain under cued attention . In matrix notation , we write C = cuT , and R=F⊙exp ( C ) , with ⊙ the elementwise product , and F as the matrix of values Ft , n=fn ( st ) . Since the joint activity on any day tended to drift slowly over time , independent of the task blocks , we introduced for each recording a slow drift signal , d . These were fitted after discounting the cue-dependent gains for each neuron . We constrained each recording’s drift signal d to vary slowly over time by placing a Gaussian process prior on it . The time constants of the priors were learned for each recording via evidence optimization ( Park and Pillow , 2011; Park and Pillow , 2013 ) , and were typically on the order of minutes to tens of minutes ( Figure 1—figure supplement 1 ) . The method is described in detail in ( Rabinowitz et al . , 2015b ) . We jointly fitted the slow drift signal d , and the coupling weights of each neuron to this signal , v , via an expectation-maximization algorithm . Arranging the effects of these slow , global drifts into a ( T × N ) matrix D = dvT , this model amounts to R=F⊙exp ( C+D ) . We introduced fast , shared modulators by fitting a ( T × N ) gain matrix M . The rank of this matrix is a measure of the number of shared modulators: we constrain rank ( M ) = K for K = 1 , 2 , 3 , … . In this way , M encapsulates the net effect of K time-varying modulators m ( k ) , with coupling weights w ( k ) , such that Mt , n= ∑k=1Kmt ( k ) wn ( k ) . The matrix M thus expresses a time-varying latent state of the system ( Kulkarni and Paninski , 2007; Paninski et al . , 2010; Macke , 2011; Vidne et al . , 2012; Archer et al . , 2014 ) . We include a zero-mean Gaussian prior on the elements of M , with p ( M ) ∝-τ2MF2 to prevent overfitting . We choose the hyperparameter τ for each population and rank by cross-validation . We find the MAP estimate of M from the model with R=F⊙exp ( C+D+M ) , under the low-rank constraint , after fitting C and D above . Since the low-rank constraint limits the feasible region for the inference problem to a non-convex set , we maximized the log posterior using the Alternating Direction Method of Multipliers ( ADMM ) , in a manner similar to ( Pfau et al . , 2013 ) . We then iterate between solving for D and for C and M ( including their respective hyperparameters ) , though in practice , one or two passes are sufficient for convergence . We cross-validated model fits as follows . For each stimulus presentation , a random subset of 20% of the spike count observations were set aside as a test set . Thus , during the inference on the training data , the values of the latent variables ( in D and M ) for the omitted observations did not contribute to the log likelihood . However , these values were automatically imputed by the priors/constraints on the structure of the latent signals ( the drifts d to vary slowly in time; the shared modulators M to be low rank ) . After convergence on the training set , the imputed values of these signals were used as predictions on the test set . The procedure above recovers a matrix , M , of the shared modulators’ effects of neural gain , but it is not identifiable: the results do not uniquely constrain the modulator time series ( m ) and the weights ( w ) . We resolve these ambiguities as follows . The rank-1 case arises when fitting separate modulators to each V4 hemisphere ( Figures 2d–e , 3–5 , 6a–b , 7 ) . Here there is a scale and sign ambiguity: we can write M=mwT= ( αm ) ( 1αwT ) for any scalar α ≠ 0 . We therefore fix Var ( m ) = 1 , and resolve the scale factor into the weights . Since almost all weights for a given hemisphere had the same polarity , we resolve the sign ambiguity so that the mean weight is positive . The rank-2 case arises when fitting two modulators to a whole population ( Figure 2b–c only ) . In addition to the scale/sign ambiguities , there is also a rotation ambiguity: we can write M=NWT= ( NQ ) ( QTWT ) for any orthogonal matrix Q . For each dataset we choose the rotation Q such that the mean weight on the LHS units is ( w̄ , 0 ) T for some value w̄ . This aligns the brown squares shown in Figure 2b and brown points in Figure 2c along the x-axis . We resolve the reflection ( sign ) ambiguity by requiring that the mean weight across both populations lies in the upper right quadrant . This resolution preserves angles , and thus has no bearing on the observed orthogonality of the weight distributions across paired hemispheres shown in Figure 2b–c ( see Figure 2—figure supplement 2 ) . Our analyses of higher-dimensional models proceed without resolving these ambiguities ( Figure 2—figure supplements 2–3 ) . To quantify the relationship between modulators and behavior , we aggregate the data across all days , and fit a number of generalized linear models ( GLMs ) to this ensemble . We estimated how correct detections for cued targets depend on the modulators ( Figure 7a ) by fitting a psychometric curve , with the cued and opposite modulators as regressors . We parameterized the hit probability on trial t as: hit ( t ) | cued target ∼ Bern ( δ·σ ( α+βTm ( t ) +λlog ( ∆θ ( t ) ) ) ) where the superscript ( t ) indicates trial t , m is a vector of the cued and opposite modulators ( averaged across the standard stimuli on that trial ) , σ is the logistic inverse-link function , Δθ is the target orientation change , δ is a lapse parameter , and the Greek characters α ( bias ) , β ( dependence on the modulators ) , and λ ( dependence on task difficulty ) are free parameters . Other parameterizations of the dependence on target orientation did not change the main result , nor did omission of the lapse term . We estimated how correct detections for the opposite targets depend on the modulators by: hit ( t ) |opposite target∼Bern ( σ ( α' + β'Tm ( t ) ) ) The values reported in Figure 7 reflect the average change in hit probability from a unit increase in mcued or mopp via these two models . We also quantify these effects individually for each cued-target condition in Figure 7—figure supplement 1; this figure also shows the full psychometrics . We estimated how correct detections for the cued and opposite targets depend on previous rewards ( Figure 7b ) by replacing the regressors m ( t ) in the above equations with categorical variables for previous reward ( hit for target on cued side / hit for target on opposite side / other ) , as below . We estimated how previous reward affects the modulator values ( Figure 7c ) by fitting a Gaussian-GLM: m ( t ) ∼N ( a+Br ( t-1 ) +Cm ( t-1 ) , σ2I ) where r ( t−1 ) is the reward from the previous trial , being ( 1 , 0 ) T for a previous hit on a cued target , ( 0 , 1 ) T for a previous hit on an opposite target , and ( 0 , 0 ) T for any other outcome ( miss/catch trial/false alarm/invalid trial ) , and a , B and C are free parameters . In all cases , we assessed the significance of parameter estimates by approximating the posterior ( both through a Laplace approximation and MCMC ) and estimating its integral above/below zero . With the exception of Figure 6c , all analyses in the main text assume that the shared modulators are constant in value ( and thus have uniform effects ) over the course of the response to each stimulus presentation . In Figure 6c , we sought to quantify the dynamics of the shared modulator at finer time scales . We extended the population response model presented in Figure 1 and Equation ( 1 ) by allowing the value of the modulator to change over the course of a stimulus presentation . As the data are very limited at fine temporal resolutions , we could not reasonably estimate the modulators’ values in small time bins for every stimulus presentation . Instead , we assumed that the temporal evolution of the modulator within each stimulus presentation followed some stereotyped pattern ( up to a scale factor that could change from one stimulus presentation to the next ) . We extend the response model to capture spike counts within bins of shorter duration ( here , 10 ms ) . We assume that the spike count Yt , b , n for neuron n within bin b of stimulus presentation t is given by: Yt , b , n∼Poiss ( Fb , n·exp ( Ct , n+Dt , n+Mt , b , n ) ) We thus assume that each neuron has a stimulus-driven mean firing rate that changes from bin to bin ( Fb , n ) , but is identical across repeated stimulus presentations; the cue-dependent gains ( Ct , n ) and slow global drift ( Dt , n ) are constant over all bins within each stimulus presentation; and the shared modulators ( Mt , b , n ) are now free to have structure across stimulus presentations ( t ) , neurons ( n ) , and bins within each stimulus presentation ( b ) . This model is now extremely high-dimensional ( the tensor M having TNB free parameters ) . To overcome this , we must impose structure on M , which we do by assuming that it has low tensor rank . We model the tensor M as an outer product of a rank-2 matrix ( with components Mt , n , as in the two-modulator model presented in the remainder of the text ) , and a vector ω indexed over bins , i . e . Mt , n , b=∑k=12mt ( k ) wn ( k ) ωb . The weight vector ω thus represents the common temporal evolution of the modulators’ effects on neural gain within each stimulus presentation , which is identical ( up to scale factors ) across neurons and successive stimulus presentations . We learn the whole low-rank tensor of modulator values , M , by maximizing the data likelihood . We perform coordinate descent on its components: we iterate between solving for the matrix with elements Mt , n ( via ADMM ) , then solving for the vector ω ( via gradient descent ) . We restrict the fit to the main response period ( 60 ms to 260 ms after stimulus onset ) as the shared fluctuations in spontaneous activity were typically large; the time series ω is thus shown in Figure 6c limited to this period . | Our brains receive an enormous amount of information from our senses . However , we can’t deal with it all at once; the brain must selectively focus on a portion of this information . This process of selective focus is generally called “attention” . In the visual system , this is believed to operate as a kind of amplifier that selectively boosts the signals of a particular subset of nerve cells ( also known as “neurons” ) . Rabinowitz et al . built a model to study the activity of large populations of neurons in an area of the visual cortex known as V4 . This model made it possible to detect hidden signals that control the attentional boosting of these neurons . Rabinowitz et al . show that when a monkey carries out a visual task , the neurons in V4 are under the influence of a small number of shared amplification signals that fluctuate in strength . These amplification signals selectively affect V4 neurons that process different parts of the visual scene . Furthermore , when the monkey directs their attention to a part of the visual scene , the associated amplifier reduces its fluctuations . This has the side effect of both boosting and stabilizing the responses of the affected V4 neurons , as well as increasing their independence . Rabinowitz et al . ’s findings suggest that when we focus our attention on incoming information , we make the responses of particular neurons larger and reduce unwanted variability to improve the quality of the represented information . The next challenge is to understand what causes these fluctuations in the amplification signals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"neuroscience"
] | 2015 | Attention stabilizes the shared gain of V4 populations |
The thiamine pyrophosphate ( TPP ) riboswitch is a cis-regulatory element in mRNA that modifies gene expression in response to TPP concentration . Its specificity is dependent upon conformational changes that take place within its aptamer domain . Here , the role of tertiary interactions in ligand binding was studied at the single-molecule level by combined force spectroscopy and Förster resonance energy transfer ( smFRET ) , using an optical trap equipped for simultaneous smFRET . The ‘Force-FRET’ approach directly probes secondary and tertiary structural changes during folding , including events associated with binding . Concurrent transitions observed in smFRET signals and RNA extension revealed differences in helix-arm orientation between two previously-identified ligand-binding states that had been undetectable by spectroscopy alone . Our results show that the weaker binding state is able to bind to TPP , but is unable to form a tertiary docking interaction that completes the binding process . Long-range tertiary interactions stabilize global riboswitch structure and confer increased ligand specificity .
The gene-regulatory activity of a riboswitch is mediated by its ability to form a substructure , called the aptamer domain , that binds—and thereby senses—a ligand , which is generally a small metabolite ( Roth and Breaker , 2009; Serganov and Nudler , 2013; Serganov and Patel , 2012 ) . Riboswitch aptamers fold and then bind their cognate ligands in a highly specific manner , and as they do so , they compete with alternative RNA structures that can form in conjunction with other domains of the riboswitch , which function either to permit , or to prevent , downstream gene expression . The aptamer of the TPP riboswitch senses the abundance of the coenzyme thiamine pyrophosphate , and in response , reduces the uptake of the vitamin thiamine . Versions of the TPP riboswitch have been discovered in all three kingdoms of life ( Cheah et al . , 2007; Sudarsan et al . , 2003; Winkler et al . , 2002 ) . In Arabidopsis thaliana , the TPP aptamer is located upstream of the 3′ untranslated intron region of the thiC gene , and it contains a complementary sequence that , in the absence of TPP , acts to sequester a downstream splice site: it therefore functions by modulating alternative gene splicing ( Wachter et al . , 2007 ) . Crystal studies have identified striking similarities between the structures of the ligand-bound forms of a truncated TPP aptamer from A . thaliana and the thiM aptamer from Escherichia coli ( Edwards and Ferre-D'Amare , 2006; Serganov et al . , 2006; Thore et al . , 2006 ) . Specifically , both eukaryotic and prokaryotic aptamers share a common binding conformation ( Figure 1 ) , characterized by contacts formed by ( 1 ) the TPP pyrimidine ring and the ‘sensor’ helix arm P2/3 , ( 2 ) the TPP pyrophosphate and ‘sensor’ helix arm P4/5 , and ( 3 ) a long-range tertiary rearrangement that brings together , and stabilizes , an interaction between the two arms carrying the L5 loop and the P3 stem ( Noeske et al . , 2006 ) . 10 . 7554/eLife . 12362 . 003Figure 1 . Single-molecule assay , crystal structure and schematic of the TPP aptamer . ( A ) Experimental geometry of the dumbbell optical-trapping assay with simultaneous FRET monitoring , with key components labeled ( not to scale ) . ( B ) Crystal structure of the truncated TPP riboswitch aptamer with a shortened P3 stem ( PDB entry 3D2G ) . Dye-labeling sites for ATTO550 and ATTO647N are indicated , at nucleotides G88 ( green ) and U25 ( red ) , respectively . ( C ) Secondary structure of the TPP riboswitch , with structural components indicated in color . Sequences surrounding the amino-modified uracil bases in the sensor helix arms ( 5-N-U and 5-LC-N-U; Integrated DNA Technologies , Coralville , IA ) are shown , with the modified bases colored ( P3 , red; L5 , green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12362 . 003 Results from single-molecule force spectroscopy experiments have suggested that TPP binding is modeled kinetically by two sequential states: initial weak binding , followed by strong binding ( Anthony et al . , 2012 ) , ( 1 ) UF→F⇄-TPP+TPP F'·TPP ⇌F''·TPP where UF refers to the unfolded form , and F , F′ and F″ refer to folded forms . The existence of the folded states was inferred from measurements of the molecular extension of the aptamer as a function of the external force applied ( force-extension curves , or FECs ) during mechanically-induced unfolding . FECs obtained for the binding of TPP in the weakly bound , intermediate F′ state were strikingly similar to FECs measured for the complete binding of two TPP analogs , thiamine monophosphate ( TMP ) or thiamine ( T ) . These analogs lack one or both phosphate groups , respectively , resulting in incomplete binding to the bipartite ligand-binding site , which normally forms contacts with both the thiamine and phosphate moieties . Similar FECs were also measured for TPP binding by a mutant aptamer ( Anthony et al . , 2012 ) that is unable to form tertiary contacts between the two arms carrying L5 and P3 . Taken together , these observations are consistent with two nonexclusive possibilities for the structure of the F′•TPP intermediate: ( 1 ) TPP may be incompletely coordinated at the binding site in this aptamer form , or ( 2 ) the aptamer structure may not have brought together its L5 and P3 arms , whose docking may be required for strong binding and the transition to the F″•TPP state . Here , we address the structural ambiguity posed by the F′•TPP state by simultaneously monitoring fluorescent probes attached to the arms carrying L5 and P3 , and aptamer folding , using an optical trap augmented with smFRET capabilities ( Figure 2 ) . By studying folding in the presence of TPP and its analogs , we can explore the roles of long-range tertiary rearrangements and phosphate binding in ligand recognition . 10 . 7554/eLife . 12362 . 004Figure 2 . Schematic optical layout of the dual-beam optical trap and FRET . Solid lines indicate lasers and light sources: 830 nm detection laser ( orange ) , 1 , 064 nm trapping laser ( red ) , 532 nm excitation laser ( green ) , and 470 nm illuminating LED ( blue ) . Filled bars are emissions received by these detectors: position sensitive detector ( PSD ) , electron multiplying charge coupled device ( EMCCD ) camera , and a video camera . Lasers are maneuvered and modified in intensity using acousto-optic deflectors ( AOD ) and acousto-optic modulators ( AOM ) , respectively . The dual beam trapping laser is expanded using a beam expander ( BE ) and split with a beam splitter . The excitation laser emission goes through a dichroic mirror ( D5 , T640LP; Chroma , Brattleboro , VT ) to split the acceptor and donor emissions . Filters ( F ) , Wollaston prisms ( W ) , pinholes ( P ) , dichroic mirrors ( D ) , mirrors ( M ) , and telescopes ( T ) are labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 12362 . 004 Instruments that combine the capabilities of single-molecule fluorescence and optical trapping represent a comparatively new development ( Candelli et al . , 2011; Comstock et al . , 2011; Harada et al . , 1999; Lang et al . , 2004; Mameren et al . , 2006; Sirinakis et al . , 2012; van Mameren et al . , 2009 ) , and these continue to present significant technical challenges ( Lang et al . , 2004; van Mameren et al . , 2008 ) . Recently , novel findings have begun to emerge using optical trapping and smFRET ( Comstock et al . , 2015; Ngo et al . , 2015; Suksombat et al . , 2015 ) .
Using a dual-beam optical trap , we applied controlled loads to each end of an individual TPP aptamer RNA ( 3′ and 5′ ) , and measured the end-to-end extension during mechanical unfolding ( Figure 1A ) . Tertiary interactions were simultaneously scored by a FRET readout between donor and acceptor fluorescent dyes , ATTO550 and ATTO647N ( Figure 1B ) , which were covalently attached to the L5 and P3 helix arms , respectively . Aptamers were subjected to repeated cycles of unfolding and refolding . After a preliminary unfolding of the RNA , the optical traps were brought into close proximity , which reduced the applied load below 5 . 0 pN for a time of 0 . 5–30 s , allowing the aptamer to refold under low force ( the refolding step ) . The force was subsequently ramped up by increasing the trap separation linearly at 200 nm/s ( Figure 3 , Figure 3—figure supplement 1 , 2 ) , as both molecular extension and fluorescence signals were acquired . After full extension , the refolding/unfolding cycle was repeated , until either a dye photobleached or the RNA tether broke . As reported previously , FECs could be categorized into one of three distinct forms ( Figures 3A , B , C ) ( Anthony et al . , 2012 ) , each corresponding to a different folded conformation , as follows . In the absence of TPP , FECs for the folded aptamer ( F ) displayed three opening transitions , or rips , corresponding to the sequential rupture of the secondary structural elements ( P1 + P2 ) , P4/5 and the P3 stem , respectively . In the presence of TPP , two additional FEC forms were observed: ( 1 ) an FEC with a single opening transition at low force ( <20 pN ) , sometimes followed by an additional opening transition , that corresponds to the weakly bound state , F′•TPP , and ( 2 ) an FEC with a single , high-force ( >20 pN ) transition , that corresponds to the tightly bound , fully folded state , F″•TPP . FRET signals acquired during force ramps could be categorized into three groups , with characteristics that were attributable to each of the three established aptamer conformations , F , F′•TPP , and F″•TPP . In the absence of TPP ( Figure 3A ) , the FRET efficiency of the F state remained stable near a low value under reduced loads , and then dropped abruptly to near zero upon the first opening transition , as the duplex elements P1 and P2 unfolded . In the presence of TPP ( Figure 3B ) , the FRET efficiency of F′•TPP started out with a similar low value , but then transitioned to an intermediate level which was not observed in the absence of ligand , until the elements P1 , P2 and P4/5 unfolded together , resulting in an abrupt drop to near zero . This transition was followed by a very tiny rip , barely detectable in the FEC , but not producing any further change in FRET , that corresponded to the unfolding of P3 ( Anthony et al . , 2012 ) . By contrast , FRET signals for F″•TPP ( Figure 3C ) first jumped from a low to an intermediate FRET level ( the same two levels , on average , displayed by F′•TPP ) , but later transitioned to a high FRET level , characterized by a comparatively low variance , until the final opening transition , corresponding to the unfolding of the entire aptamer , which returned the FRET level to the baseline . 10 . 7554/eLife . 12362 . 005Figure 3 . Representative FEC and FRET traces . Representative traces of unfolding for the aptamer conformations ( A ) F , ( B ) F′•TPP , and ( C ) F″•TPP . Simultaneous force-extension curves ( FECs; black lines ) and FRET trajectories ( black circles , gray lines ) are parametrized by time . Colored boxes indicate the APO ( blue ) , WB ( green ) , and SB ( yellow ) FRET states . Open arrowheads mark the end of the refolding period and the start of the force ramp . Small , filled arrowheads mark the location of opening transitions . DOI: http://dx . doi . org/10 . 7554/eLife . 12362 . 00510 . 7554/eLife . 12362 . 006Figure 3—figure supplement 1 . Representative traces of various secondary state conformations and TPP concentrations . Force curves ( black lines ) and FRET trajectories ( black circles , gray lines ) are shown parametrized by time for the following aptamer conformations: ( A ) F″•TPP in the presence of 5 μM TPP , ( B ) F′•TPP in the presence of 5 μM TPP , ( C ) F′•TPP in the presence of 2 mM TMP , ( D ) F′•TPP in the presence of 2 mM T , ( E ) F in the presence of 50 μM TPP , ( F-H ) in the absence of TPP ( panel ( E ) is the same as Figure 3 , panel A ) , ( I ) F″•TPP in the presence of 5 μM TPP . Here , we observe loss and rebinding of ligand . DOI: http://dx . doi . org/10 . 7554/eLife . 12362 . 00610 . 7554/eLife . 12362 . 007Figure 3—figure supplement 2 . FRET changes occur throughout the refolding period and the force ramp , up until the first rip . Representative traces show FRET signatures indicative of a particular helix arm configuration at the end of a refolding period that is maintained throughout the force ramp . Colored boxes , lines , open and filled arrowheads are as described in Figure 3 . ( A ) The aptamer transitions from an unbound configuration to a weakly bound configuration , and then to a strongly bound configuration , which is held up until the catastrophic rip . ( B ) The aptamer transitions from an unbound configuration to a weakly bound configuration prior to the end of the refolding period , and remains in the weakly bound configuration until the catastrophic rip . DOI: http://dx . doi . org/10 . 7554/eLife . 12362 . 007 In both force and FRET channels , unfolding events produced abrupt drops in signal levels . We recorded the FRET efficiencies prior to such events ( defined as the opening FRET values ) along with the associated forces ( defined as the opening force values ) . A plot of opening FRET versus opening force revealed sub-populations that were binned by k-means cluster analysis ( Figure 4A ) ; setting k = 3 selected distinct clusters and established threshold values for each coordinate axis ( Figure 4B ) . Vertical ( force ) thresholds demark the three conformational forms , based chiefly on their secondary structures: FECs with an opening force lower than 11 . 5 pN were assigned to aptamers unfolded when not bound to TPP; opening forces between 11 . 5 pN and 21 . 0 pN were assigned to aptamers that unfolded in the F′•TPP state; and opening forces above 21 . 0 pN were assigned to aptamers that unfolded from the F″•TPP state . The mean opening forces computed for each of the three clusters ( C1 , C2 , and C3 ) matched , within error , the force values that had been estimated in a previous study using other methods ( Anthony et al . , 2012 ) , validating the choice of k . 10 . 7554/eLife . 12362 . 008Figure 4 . Clustering analysis of opening FRET values , and global analysis of full-length FRET traces . ( A ) k-means clustering for opening FRET and opening force values . Filled diamonds mark the mean opening force and opening FRET values for each of the k = 3 populations; dashed vertical and horizontal lines indicate the thresholds for force and FRET states , respectively . ( B ) Table summarizing opening force and FRET centroids , and thresholds for the clusters C1 , C2 and C3 . ( C ) Global fit ( red ) of all data ( black histogram , with error bars ) to a sum of four Gaussians . Individual Gaussian fits are shown as colored lines , with the mean values indicated ( vertical dashed lines ) . ( D ) Table of fit parameters for G1 , G2 and G3 ( N = 435 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12362 . 00810 . 7554/eLife . 12362 . 009Figure 4—figure supplement 1 . k-means cluster analysis of opening force and FRET . k-means clustering results and summary tables shown for ( A ) k = 2 , ( B ) k = 3 , and ( C ) k = 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 12362 . 009 Horizontal ( FRET ) thresholds , 0 . 57 and 0 . 74 , demark the FRET cutoffs for the clusters . The corresponding FRET sub-populations were then compared to a global histogram of all single-molecule records fitted to a sum of four Gaussians ( Figure 4C ) . Mean values returned from the Gaussian fits , G1 , G2 and G3 , were statistically consistent with mean values for opening FRET computed for the C1 , C2 and C3 clusters . By matching the FRET sub-populations against conformations previously characterized by force assignment , we could infer the P3-L5 arm configurations , independent of force . The lowest-valued FRET sub-population , corresponding to the APO state , reflects a significant physical separation between the P3 stem and L5 loop , which remain consistently apart . The intermediate-value FRET sub-population , corresponding to the WB ( weakly bound ) state , reflects more proximal stem-loop interactions . The highest-value FRET population , corresponding to the SB ( strongly bound ) state , reflects the close apposition of P3 and L5 . To characterize further P3-L5 arm behavior , we used the cluster thresholds established for refolding to categorize segments of FEC records obtained at forces up to 10 pN ( i . e . , prior to the first opening transition ) , including the refolding portion , as follows . Segments of FRET trajectories below 0 . 57 , or above 0 . 74 , which tended to exhibit comparatively lower variance , were assigned to the APO or SB states , respectively . FRET trajectories between these values were assigned to the higher-variance , WB state ( such trajectories included short intervals where the FRET value made brief excursions above or below the threshold levels , but never remained consistently out-of-range for more than one or two consecutive time points ) . Thus separated , the individual FRET data were binned into histograms for the corresponding APO , WB , and SB states ( Figure 5A ) . 10 . 7554/eLife . 12362 . 010Figure 5 . Distributions of segmented FRET data in the presence and absence of TPP and its analogs . ( A ) Segmented FRET data in the presence of saturating TPP ( 2 mM ) . After segmenting , FRET records were categorized into APO ( blue ) , WB ( green ) , and SB ( yellow ) FRET states . ( B ) Table summarizing FRET efficiency statistics . ( C ) FRET data from aptamers in the absence of TPP ( purple ) , and in the presence of saturating ( 2 mM ) T ( pink ) and TMP ( orange ) , displayed together with data for the SB state in the presence of TPP ( yellow ) for comparison . ( D ) Table summarizing FRET efficiency statistics for the TPP , T , and TMP conditions indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 12362 . 010 TPP analog ligands that are missing the terminal phosphate group , such as TMP ( thiamine monophosphate ) and T ( thiamine ) , are thought to be able to form contacts with only one half of the bipartite binding site of the aptamer . In support of this notion , force spectroscopy has previously established that FECs of the aptamer bound by such analogs closely resemble those corresponding to the weakly bound , F′•TPP conformation ( Anthony et al . , 2012 ) . We collected data for both FECs and FRET trajectories for aptamer binding in the presence of TMP , T , and in the absence of ligand , and applied the same clustering formalism as for aptamer binding in the presence of TPP ( Figure 5C ) . FRET trajectories acquired in the absence of ligand exhibited a low efficiency ( mean = 0 . 46 ) , and closely matched those assigned earlier to the APO state , which had been observed in the presence of TPP , but prior to presumed ligand binding , lending additional experimental support to that assignment . Aptamers bound to either TMP or T generated FECs similar to those previously observed , and closely resembled the weakly bound conformation in the presence of TPP , as anticipated . The FRET efficiencies were similar for both T and TMP ligands ( means = 0 . 64 and 0 . 69 , respectively ) . However , both these values are significantly lower than the efficiency of the WB state obtained in the presence of TPP ( mean = 0 . 73 ) . Because the opening transitions in FECs tend to be dominated by secondary structure , whereas the FRET signal reports a tertiary rearrangement , we conclude that aptamers bound to TMP or T share a similar secondary structure with the aptamer weakly bound to TPP , but differ in tertiary structure: specifically , the analogs TMP and T fail to bring the helix arms together as closely as TPP . To quantify the helix arm dynamics , we fitted a sequential , four-state hidden Markov model ( HMM ) to refolding portions of the FRET trajectories ( obtained under negligible loads that permit structure formation , <5 pN ) ( Figure 6A ) . FRET values and directional transition rates were obtained at TPP concentrations of both 5 µM and 50 µM ( Figures 6B , C ) . The corresponding values for FRET states identified by hidden Markov modeling were identical , within statistical error , at both TPP concentrations , and notably , also within statistical error for the corresponding states identified using k-means cluster analysis ( Figure 6B and Figure 5B ) . We were therefore able to make a direct correspondence between the HMM-identified states ( called A , B , and C ) and the clustered FRET states ( Apo , WB , and SB , respectively ) . At the higher TPP concentration , all transition rates except for the A → B rate remained the same within error , implying that only this transition , corresponding to the Apo → WB transition , is TPP-dependent ( Figure 6C ) . The inferred dependence is also consistent with the kinetic model proposed on the basis of force spectroscopy , where only the transition between states F′ and F′′•TPP is TPP-dependent ( Figure 7A and 7B ) ( Anthony et al . , 2012 ) . Therefore , both FRET and force data support a sequential , four-state kinetic model . However , the transition rates obtained using the HMM approach are consistently faster than the rates previously derived from force spectroscopy alone . Specifically , at the high TPP concentration , the lifetime in the A state was vanishingly short , and only a single dwell interval was scored by the HMM procedure . Instead , the vast majority of records ( N = 52 of 53 ) started out in state B , and thereafter made subsequent transitions between B and C states . We conclude that the A → B transition rate at the higher TPP concentration is likely faster than our detection capability , and that the apparent 4-fold increase in this transition rate should be taken as a lower bound . 10 . 7554/eLife . 12362 . 011Figure 6 . HMM modeling of refolding FRET trajectories . ( A ) Four-state HMM fit to a 30-second portion of concatenated refolding FRET trajectories in the presence of 5 µM TPP . ( B ) Top: Reaction diagram for the four-state , sequential HMM model , with states UF and A ( donor-fluorescing ) , and B and C ( acceptor-fluorescing ) . Bottom: table summarizing FRET values obtained 5 µM and 50 µM TPP concentrations for each state . ( C ) Table summarizing directional rates for transitions between HMM states A , B , and C . The yellow icon indicates the TPP-dependent transition rate . DOI: http://dx . doi . org/10 . 7554/eLife . 12362 . 01110 . 7554/eLife . 12362 . 012Figure 7 . Correspondences among single-molecule state identifications , and model of aptamer binding to TPP . ( A ) Cartoon summarizing the correspondence between analysis and data collection methodologies . The corresponding states and reaction schemes obtained from HMM analysis ( top line; pink circles ) , helix-arm configuration states obtained from FRET signals ( middle line; blue circles ) , and secondary structural states inferred from force spectroscopy ( lower line; green ellipses ) are shown . ( B ) Model for ligand binding and associated conformational changes . Colored labels indicate states based on aptamer secondary structure ( green circles and ellipses; derived from force data ) and sensor-helix arm configuration ( blue circles; derived from FRET data ) . Prior to TPP binding , the sensor arms are apart . Subsequent to TPP binding , the arms remain mobile but begin to move closer together in the weakly-bound , liganded state . This mobility may reflect a type of conformational heterogeneity that is either dynamic ( top ) , with flexible arms , or static ( bottom ) , with rapid interconversions between transient states ( square brackets; see text ) . The system subsequently transitions , on a timescale of around a second , to a strongly bound state , with the sensor arms fully docked and largely immobilized . DOI: http://dx . doi . org/10 . 7554/eLife . 12362 . 012
The concurrent use of force spectroscopy , which primarily monitors secondary structural states , and FRET , which can probe tertiary structural rearrangements , has helped to reveal new details of the interaction between ligand binding and conformational change in the TPP riboswitch ( Anthony et al . , 2012 ) . The eukaryotic aptamer folds quickly into a stable form , where the P3-L5 arm configuration remains open , as indicated by the stable , low FRET trajectories recorded in the absence of ligand ( Figure 3A , Figure 3—figure supplement 2F–H ) . By contrast , Haller et al . ( 2013 ) reported that , in the absence of ligand , the arms of the bacterial form of the TPP aptamer interchange rapidly among three FRET states , as fit by a hidden Markov model , with dwell times around 100 ms ( Haller et al . , 2013 ) . The slightly slower imaging rate employed here ( 20 frames/s , vs . ~50–100 frames/s ) should still have been adequate to detect any comparable flickering , in principle , but none was evident in our records . The eukaryotic aptamer is therefore structurally more stable in its unbound form ( or conceivably , it switches among configurations , but at a significantly faster rate ) . We note that the two TPP aptamers differ primarily in their P3 stem regions: the bacterial form lacks a distal P3 extension that is present in the eukaryotic form , and this longer element may exert a stabilizing role on the unbound aptamer . Upon ligand binding , the P3 and L5 arms underwent a rapid conformational change , as manifested by the FRET signal , corresponding to the transition from the APO to the WB state ( Figures 3B , C; blue to green boxes ) . The FRET levels in the WB state exhibited comparatively high variance ( Figure 5B ) , and appeared to flicker between intermediate and high FRET values . Transitions out of the WB state were generally irreversible , and these led directly to the SB state , which was characterized by a stable , high FRET level with significantly lower variance . We therefore interpret the SB configuration as one where the two helix arms are fully docked , forming a tight , stable P3-L5 tertiary interaction . Based upon the data , the WB state can be interpreted as one where the P3 and L5 arms are partially docked , that is , they remain mobile but closer , on average , than in the APO state . Alternatively , the WB state might represent a superposition of ( less individually flexible ) docked and undocked conformations that interchange rapidly on the timescale of our data acquisition , at rates >>20 s-1 ( Figure 6C ) . Put another way , the WB state must be mobile , and consistent with either dynamic conformational flexibility or with static conformational heterogeneity ( Torella et al . , 2011 ) . Helix arm behavior was examined further by exploring the effects of aptamer binding by the TPP analogs , thiamine ( T ) and thiamine monophosphate ( TMP ) . The binding site of the aptamer , located at the central junction of the arms , is bipartite , and is able to bind to the thiamine moiety of TPP on one side , and to phosphate groups on the other . One previous study of T and TMP , which lack one and two phosphate groups , respectively , and therefore bind to only one side of the site , concluded that the lack of phosphates prevents proper alignment of the P3-L5 arms and ultimately , docking ( Noeske et al . , 2006 ) . For both bound analogs , the mean FRET efficiencies were smaller than the corresponding FRET efficiency of the SB state ( Figure 5D ) . These results imply that the sensor helix arms spend most of their time farther apart when bound to TMP or T than when bound to TPP in the F″•TPP conformation . We surmise that each additional phosphate group serves to draw the arms closer together during ligand binding , and is responsible for the observed increase in FRET . Perhaps more surprisingly , the FRET trajectories for the aptamer after binding onto TMP and T are stable and low in variance , suggesting that although the P3-L5 structures are unable to dock , they remain in a stable configuration ( Figures 5C and 5D ) . This trend is reflected in the FRET variance obtained during force ramp , which is low at 0 . 005 and 0 . 003 for T and TMP , respectively , but significantly larger at 0 . 010 for the WB FRET state ( Figures 5B and 5D ) . This result implies that when the aptamer is bound to T or TMP , the arms are less dynamic than when the aptamer is bound to TPP in the F′•TPP conformation . Taken together , these data suggest that aptamer binding of TPP in the F′•TPP conformation is a state in which the helix arms are in close proximity due to the presence of two phosphate groups , yet exhibit significant flexibility . This flexibility appears to be essential for proper orientation of the helix arms , which primes the arms to dock and the aptamer to adopt the F″•TPP conformation . Our studies of the complete eukaryotic TPP riboswitch reveal a ligand binding process distinct from that of the bacterial TPP riboswitch . We observed a rather rigid helix arm configuration during the fully folded , APO state , in contrast to the dynamic helix arm behavior reported for the E . coli TPP riboswitch in the absence of TPP . Upon TPP binding , we observed an increase in helix arm dynamics in the WB state , which is also in contrast to bacterial studies that observed a reduction of arm dynamics upon ligand binding . Our observations of FRET trajectories suggest that helix-arm fluctuations occur at significantly faster rates than any secondary structural changes . Based on our data , we conclude that both secondary and tertiary interactions are vital to tight TPP binding . We speculate that in the eukaryotic aptamer , ligand binding first anchors the lower regions of the helix arms , inducing strains that lead to fluctuations in the more distal regions . Such fluctuations increase the frequency of contact between the P3-L5 arms , and thereby increase the probability of docking , lowering the binding energy further . Future studies using additional FRET probes , and conducted at even higher time resolution , should be able to probe independently the motions of the lower aptamer regions . The bacterial and eukaryotic forms of the TPP riboswitch share a similar core structure , bind identical ligands , and regulate thiamine pyrophosphate synthesis ( albeit in rather different ways ) . However , it has become clear that even minor structural differences between forms can yield distinct properties . For example , the comparatively longer and more stable P3 stem found in the eukaryotic riboswitch leads to slower arm dynamics , as expected , with wide-ranging consequences . But piecing together a more detailed , quantitative picture of the folding landscape , which includes strain-dependent ligand binding and the formation of tertiary contacts during helix-arm docking , requires special methodology , such as the combination of fluorescence and force spectroscopy pursued here . It is still early days , but we anticipate that the combined force-FRET technique will be a powerful tool for elucidating riboswitch function across all kingdoms of life .
An instrument characterization/calibration was performed using DNA hairpins with short tetraloops as controls . We examined two hairpins: one with a 15-bp stem ( 15R50/T4 ) and one with a 20-bp stem ( 20R25/T4 ) . The nanomechanical properties of both hairpins have been extensively characterized in a previous force-spectroscopy study ( Woodside et al . , 2006 ) . Hairpins were constructed by ligating separately prepared 5′ and 3′ fragments ( Figure 8A ) ; in each of the fragments , a single amino-modified thymidine residue was introduced near the base of the hairpin ( Integrated DNA Technologies ) . The modified bases were labeled with the fluorescent dyes , ATTO550 and ATTO647N ( ATTO-TEC ) , using an amine-labeling chemistry ( Kricka , 2002; Solomatin and Herschlag , 2009 ) . The dye-labeled fragments were annealed using a temperature ramp protocol , from 95°C to 4°C in 40 min , and ligated with T4 DNA ligase ( Invitrogen , Carlsbad , CA ) for 30 min at 37°C ( Akiyama and Stone , 2009 ) . Proteins were removed by phenol-chloroform extraction , and the ligated hairpins were ethanol-extracted . Doubly labeled hairpins were separated from incomplete hairpin species on a 10% PAGE gel , then electro-eluted , followed by ethanol precipitation . 10 . 7554/eLife . 12362 . 013Figure 8 . Dye characterization using DNA hairpins . ( A ) Opening distances ( F1/2 ) and forces ( Δx ) of DNA hairpins SJ1 and SJ2 were measured at non-equilibrium and compared to previous measurements at equilibrium . ( B ) FRET trajectories and FECs are shown . ( C ) k-means clustering for SJ1 and SJ2 FRET trajectories . Cluster means are indicated with filled diamonds . ( D ) Table summarizing opening force and FRET statistics , force ramp FRET statistics , plus predicted FRET efficiency . ( E ) Donor and acceptor traces ( top ) in the presence of 20 and 30% TQ ( left and right , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12362 . 013 The TPP riboswitch was split into three fragments: wtAtRs1 , wtAtRs2 , and wtAtRs3 ( Dharmacon , Lafayette , CO ) . Fluorescent dyes were attached via amino-modified uracil bases at G88 ( 5-N-U ) and U25 ( 5-LC-N-U ) . G88 is found in the P3 stem structure of fragment wtAtRs1 , and was labeled with ATTO550 dye; U25 is found in the L5 loop of fragment wtAtRs3 and was labeled with ATTO647N dye . Labeled fragments were splint-ligated using T4 RNA ligase 2 , assisted by complimentary DNA oligomers that bridged the fragments . Labeled and ligated riboswitches were gel extracted and eluted , as described ( Anthony et al . , 2012 ) , for the DNA hairpins . Trolox ( TX ) , a soluble analog of vitamin E , has been shown to reduce triplet-state quenching and blinking , especially when combined with Trolox-quinone ( TQ ) ( Cordes et al . , 2009; 2011 ) . TX was dissolved in DMSO to 100 mM and exposed to UV light , yielding 20–28% TQ in 3 mM TX . The TQ concentration was determined by diluting UV-exposed TX to 3 mM in PHC buffer ( 50 mM HEPES pH 7 . 5 , 130 mM KCl , 4 mM Mg2+ , 1 mM EDTA ) and measuring the absorbance at 255 nm , according to: [TQ]=Aγ/d−εγTX[TX]0εγTQ−εγTX where γ is the wavelength ( 255 nm ) , d is the distance ( 1 cm ) , and the extinction coefficients for TX ( εγTX ) and TQ ( εγTQ ) are 400 and 11 , 600 l·mol-1·cm-1 , respectively . Lower percentages of TQ were obtained by diluting TQ in 100 mM unexposed TX . A stock was stored at 4°C , and was diluted to 3 mM in SCAV prior to experiments . DNA hairpins held under low load showed reduced blinking in the presence of TQ ( Figure 8E ) . We incorporated fluorescence excitation and detection capabilities into an existing dual-beam optical trapping apparatus , as described ( Greenleaf et al . , 2008; Lang et al . , 2002 ) . Briefly , the setup ( schematic; Figure 2 ) is based on a Nikon TE2000 inverted light microscope ( Nikon Instruments , Melville , NY ) , and utilizes three lasers: one for optical trapping ( 10W; 1 , 064 nm; diode-pumped Nd:YVO4 laser; Spectra-Physics Lasers , Mountain View , CA ) , one for position detection ( 20 mW; 830 nm; single-mode diode laser; Qioptiq , Waltham , MA ) , and one for fluorescence excitation ( 1 W; 532 nm; diode-pumped , frequency-doubled YAG; CrystaLaser , Reno , NV ) . Fluorescence emission was detected by an electron-multiplying CCD ( EMCCD ) camera used for imaging ( iXon 897; Andor , United Kingdom ) . Fluorescence excitation light was coupled into , and focused through , the microscope objective to a ~1 μm spot in the specimen plane , confocal with the fluorescently-labeled RNA construct . Dichroic filters were installed upstream of the EMCCD to split the emission spectra of the donor and acceptor fluorophores into spatially separate channels at the image plane; the emission image was passed through a rectangular slit to allow simultaneous collection from both acceptor and the donor fluorophores . The fluorescence sub-image size was ~43 x 283 pixels ( pixel size , 0 . 16 x 0 . 16 μm ) . Each region of interest ( ROI ) subtended 7 x 10 pixels for both the donor and acceptor channels . The EMCCD output counts for ATTO550 and -647N dyes were typically around 16 , 000 and 7 , 000 photons per second , respectively . Successive frame exposure times were 50 ms , and about 800 frames were collected per single-molecule record . Polystyrene beads , 0 . 6 and 0 . 8 μm in diameter , were functionalized with streptavidin and anti-digoxigenin and coupled to 1 kb and 2 kb DNA handles , respectively , via single-stranded overhangs , as described ( Anthony et al . , 2012; Greenleaf et al . , 2008 ) . Handle lengths were chosen to provide adequate spatial separation ( approximately 1 µm ) between the optical trapping beams and the fluorophore-labeled constructs . Constructs for study ( DNA hairpins or RNA aptamers ) were first annealed to the handles , and then conjugated to beads by incubation for 1 hr at room temperature . The bead incubation was diluted in oxygen-scavenging buffer ( SCAV; 50 mM HEPES pH 7 . 5 , 130 mM KCl , 4 mM Mg2+ , 1 mM EDTA , 160 Units/ml glucose oxidase , 100 Units/ml catalase , 0 . 8% beta-glucose , and 3 mM UV-treated Trolox ( Sigma-Aldrich , Germany ) with 10–14% Trolox-quinone ) . Both glucose oxidase and catalase were further purified from manufacturer’s stocks by ultrafiltration . Samples were then introduced into flow cells for optical trapping experiments . FECs were obtained by servoing the steerable optical trap at a constant rate ( ~90 nm s-1; stiffness 0 . 25–0 . 3 pN nm-1 ) relative to the fixed trap , using an acousto-optic deflector . Opening forces and extensions were measured by first identifying transitions in the FECs , then fitting the regions before and after these to a double wormlike chain model ( WLC ) , as described ( Greenleaf et al . , 2008 ) . For smFRET traces , correction factors for the leakage of donor emission into the acceptor channel ( β ) and differences between the donor and acceptor in efficiency and quantum yield ( γ ) were calculated as described ( McCann et al . , 2010 ) . The β value was calculated to be 0 . 13 for the ATTO550-ATTO647N dye pair . The γ corrections varied among traces , and were determined by subtracting counts before and after the FRET transition for both donor and acceptor signals . After suitable β , γ , and background corrections , the FRET efficiencies were calculated from E = IA / ( IA + ID ) . k-means clustering for opening forces and FRET values was performed using Python libraries . The clustering parameter , k , was varied from 2–4 with default settings ( Figure 4—figure supplement 1 ) , with optimal results obtained for k = 3 . FRET data were binned using a bin size of 0 . 025 . The global histogram of FRET data was fit to a sum of four Gaussians using a conventional nonlinear least-squares algorithm . Refolding FRET traces were concatenated and analyzed using the MSMBuilder HMM Python package ( Beauchamp et al . , 2011 ) . FRET values and inter-state transition times were calculated for HMM states A , B and C using averaging . Transition rates were calculated from the reciprocals of the directional transition times . | When a gene is switched on , its DNA is first copied to make a molecule of messenger ribonucleic acid ( mRNA ) . The genetic code in the mRNA is then translated into a protein . There are also untranslated regions within mRNAs that do not code for protein themselves , but serve to regulate whether or not a protein is produced from the rest of the mRNA . For example , many mRNAs contain a motif in their untranslated region called a 'riboswitch' . These motifs selectively bind to molecules that are the products of metabolic processes . One riboswitch found in bacteria , animals and plants binds to a molecule known as thiamine pyrophosphate ( TPP ) and regulates genes that control the uptake of a vitamin called thiamine into cells . Newly made mRNA molecules are linear strands that then fold into three-dimensional structures . The TPP riboswitch can adopt distinct shapes depending on whether it is bound to TPP or not . Knowledge of these structures is crucial for understanding how riboswitches regulate protein production . Previous research reported the folding of a TPP riboswitch from bacteria . Here , Duesterberg et al . used a combination of two techniques known as force spectroscopy and Förster resonance energy transfer ( FRET ) to study the folding of the TPP riboswitch from a plant called Arabidopsis thaliana . The experiments show that in the presence of TPP , structural changes occur in two arm-like appendages – known as helix arms – that extend out of the core of the riboswitch . The riboswitch adopts a particular shape when TPP is strongly bound to it , and in this shape the riboswitch can regulate the activity of certain genes . However , if the riboswitch is only weakly associated with TPP , it takes on a shape in which the two helix arms are further apart and the riboswitch is unable to form the interactions required to complete the process of binding to TPP . Duesterberg et al . ’s findings reveal that the way in which the A . thaliana riboswitch changes shape when it is bound to TPP is different to that of its bacterial counterpart . The next challenge will be to observe these shape changes in even more detail , and to use these new techniques to study other riboswitches in various organisms . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | Observation of long-range tertiary interactions during ligand binding by the TPP riboswitch aptamer |
Many believe that humans can ‘perceive unconsciously’ – that for weak stimuli , briefly presented and masked , above-chance discrimination is possible without awareness . Interestingly , an online survey reveals that most experts in the field recognize the lack of convincing evidence for this phenomenon , and yet they persist in this belief . Using a recently developed bias-free experimental procedure for measuring subjective introspection ( confidence ) , we found no evidence for unconscious perception; participants’ behavior matched that of a Bayesian ideal observer , even though the stimuli were visually masked . This surprising finding suggests that the thresholds for subjective awareness and objective discrimination are effectively the same: if objective task performance is above chance , there is likely conscious experience . These findings shed new light on decades-old methodological issues regarding what it takes to consider a neurobiological or behavioral effect to be 'unconscious , ' and provide a platform for rigorously investigating unconscious perception in future studies .
Above-chance performance without awareness in perceptual discrimination tasks is a strong form of unconscious perception . In these demonstrations ( e . g . , blindsight: Weiskrantz , 1986 ) the subjective threshold for awareness ( when a stimulus is consciously ‘seen’ ) seems well above the objective threshold for forced-choice discrimination ( when a stimulus can be correctly identified ) : subjects can discriminate a target above chance performance , yet report no awareness of the target . Many researchers believe normal , healthy subjects can also directly discriminate near-threshold , low-intensity targets without subjective awareness ( e . g . , Boyer et al . , 2005; Charles et al . , 2013; Merikle et al . , 2001; but see Snodgrass et al . , 2004 for an opposing view ) . We conducted an informal survey to confirm this popular belief , which also revealed that many believe convincing evidence for this phenomenon is lacking . We asked survey participants three key questions: ( 1 ) "Do you believe in subliminal perception ? " ( 2 ) "Do you believe that the subjective threshold for awareness is above the objective discrimination threshold ? " and ( 3 ) "If ‘yes’ , do you believe this has been convincingly demonstrated in the literature ? " Most respondents reported believing that subliminal processing exists ( 94% ) , but also that they did not believe it had been convincingly demonstrated in the literature ( 64% ) . These belief patterns were shown even among those who reported having published on subliminal or unconscious perception ( 94% and 61% , respectively ) . See Appendix 1 for full text of questions and detailed survey results . A primary culprit in this controversy is the problem of criterion bias: an observer’s report of ‘unseen’ doesn’t necessarily imply complete lack of awareness , only that the stimulus’ strength fell below some arbitrary boundary for reporting ‘seen’ ( Eriksen , 1960; Hannula et al . , 2005; Lloyd et al . , 2013; Merikle et al . , 2001 ) . Unfortunately , most methods of studying unconscious perception suffer from this ‘criterion problem’ ( e . g . , Charles et al . , 2013; Jachs et al . , 2015; Ramsøy and Overgaard , 2004 ) . With such methods , one could argue that reports of ‘unawareness’ may only mean some stimuli are relatively hard to perceive compared to those that are clearly visible . To avoid this criterion problem , several groups ( Kolb and Braun , 1995; Kunimoto et al . , 2001 ) sought to identify conditions in which confidence was uncorrelated with accuracy , which they argued would indicate no subjective awareness of the target . Unfortunately , some of these efforts were not replicable ( Morgan et al . , 1997; Robichaud and Stelmach , 2003 ) . Others revealed that estimating the correspondence between confidence and accuracy requires mathematical considerations more complicated than originally envisaged ( Evans and Azzopardi , 2007; Galvin et al . , 2003; Maniscalco and Lau , 2012 ) . Importantly , the conceptual link between metacognitive sensitivity ( i . e . , correlation between confidence and accuracy ) and conscious awareness is itself controversial ( Charles et al . , 2013; Fleming and Lau , 2014; Jachs et al . , 2015 ) . Here , we employ a recently-developed confidence-rating method to address this problem ( Barthelmé et al . , 2009; de Gardelle and Mamassian , 2014 ) . Subjects discriminated two stimulus intervals , only one of which contained a target , and indicated confidence in their decisions using a 2-interval forced-choice procedure ( 2IFC ) , that is , indicating which of the two discrimination decisions they felt more confident in . This approach has several advantages . First , 2IFC tasks depend little on response bias compared to multi-point confidence-rating scales . Maintaining the criteria for extensive confidence scales may also be demanding , leading subjects to respond somewhat randomly in conditions of vague awareness and thereby producing the negative result Kolb and Braun ( 1995 ) observed ( Morgan et al . , 1997 ) . Second , the interpretation of 2IFC confidence-rating in this context is straightforward: ‘Performance without Awareness’ would mean subjects can perform the target discrimination yet fail to place bets appropriately to distinguish this performance from discrimination of a blank stimulus ( which guarantees chance performance ) . That is , following psychophysics traditions ( Kolb and Braun , 1995; Peirce and Jastrow , 1884 ) , if a certain above-chance discrimination seems introspectively no different from a random guess based on no stimulus at all ( as reflected by betting behavior ) , we interpret the discrimination to be unconscious . Here , we explored whether such Performance without Awareness occurs in normal observers in two behavioral experiments , and compared these results to predictions of a Bayesian ideal observer .
Nine human observers participated in two experiments of our 2IFC confidence-rating paradigm ( Figure 1 ) . In both experiments , participants viewed two intervals in which they were required to discriminate the orientation ( right or left tilt ) of a Gabor patch target embedded in forward- and backward-masks ( Figure 1A , B ) , and judged which of the discrimination choices they felt more confident in . Crucially , in one of the intervals the target was absent ( Figure 1B ) , such that above-chance discrimination performance was impossible . We performed two experiments to assess the potential contributions of question order , receipt of feedback , and a priori knowledge of the presence of a target-absent interval ( Figure 1C ) . In Experiment 1 , participants judged which decision they felt more confident in and then indicated their orientation decisions for both intervals , while in Experiment 2 they indicated their orientation discrimination decisions before selecting the more-confident interval . In Experiment 2 , we also provided feedback on the confidence decision , and told participants that one interval contained no target; this information was withheld from participants in Experiment 1 . Stimuli , timing details , and order of question prompts in the two experiments are also discussed in greater detail in the Methods section . 10 . 7554/eLife . 09651 . 003Figure 1 . Stimuli and procedures for the 2IFC confidence-rating task . ( A ) Targets consisted of oriented ( 45° left- or right-tilted from vertical ) Gabor patches presented at multiple near-threshold contrast levels; masks consisted of bandpass-noise filtered random RGB values ( see Materials and methods ) . ( B ) Each trial consists of two intervals of discrimination in which the target stimulus ( T ) was forward- and backward-masked ( M ) . Gabor patch targets were presented only in target-present ( TP ) intervals; in target-absent ( TA ) intervals , the target was replaced with blank frames . Otherwise timings of stimuli were matched between the two intervals . ( C ) Experimental tasks . Experiment 1 required subjects to bet on which discrimination they felt more confident before they indicated their orientation discrimination choices ( left or right tilt of the Gabor ) sequentially for both intervals . Shown is an example trial in which TP is presented before TA; in the experiment this order varied randomly from trial to trial . In Experiment 2 , subjects bet on the more confident interval after the discriminations , and feedback was given . ( See Materials and methods for more details . ) DOI: http://dx . doi . org/10 . 7554/eLife . 09651 . 003 For both experiments , we evaluated whether participants exhibited Performance without Awareness ( Figure 2A ) or Performance > Awareness ( Figure 2B ) . In both cases , the response pattern of interest can be visualized as percent of time betting on the target-present interval as a function of percent correct orientation discrimination in the target-present interval . ‘Performance without Awareness’ ( Figure 2A ) would be supported if observers can discriminate the target above chance ( >50% accuracy ) while being unable to bet on their choices more often than betting on the target-absent interval ( which necessarily yields chance-level performance ) . That is , observers correctly discriminate the target’s orientation more than 50% of the time , but bet on the target-present interval 50% of the time ( i . e . , they bet randomly on the target-present versus target-absent interval ) , indicating they are not aware of the information that contributed to their discrimination decision . If this were to occur , it would most likely happen at low discrimination performance levels , yielding a pattern of behavior similar to that presented in Figure 2A . 10 . 7554/eLife . 09651 . 004Figure 2 . Schematic explanation of predictions of the experiments . ( A ) A ‘Performance without Awareness’ pattern of behavior , in which subjects are able to discriminate the target above chance while betting on the target-present interval at chance . ( B ) A ‘Performance > Awareness’ pattern of behavior , in which subjects are less able to bet on their discrimination decisions than they are able to correctly discriminate the target . In both ( A ) and ( B ) , the diagonal dashed line indicates where rate of betting on the target-present interval equals objective discrimination performance . DOI: http://dx . doi . org/10 . 7554/eLife . 09651 . 004 However , in psychophysics , thresholds can also be defined as midway between ceiling and floor performance ( Macmillan and Creelman , 2004 ) , such that threshold discrimination performance is defined as 75% accuracy rather than >50% ( chance level ) . This concept can also be applied to subjective betting data in the sense that betting on the target-present interval could be considered ‘correct’ or ‘advantageous’ betting . In this sense ( threshold = 75% correct performance ) , the subjective threshold for confidence might be above the objective threshold for discrimination . In other words , observers may bet on the target-present interval less often than they get the discrimination correct , but still above chance . This would occur because the orientation discrimination choice requires evaluation of only one interval ( the one with the target in it ) and therefore is subject to only one source of uncertainty , but the ‘betting’ choice requires evaluation of both intervals , and therefore has two potential sources of uncertainty . This pattern of behavior ( Figure 2B ) may occur even if subjects do not display Performance without Awareness , and would be characterized by a pattern of responses that fall below the identity line ( diagonal dashed line ) . We call this possibility ‘Performance > Awareness’ . We discuss the results of both experiments together for ease of interpretation , and because the results are very similar ( Figure 3A–F ) . To anticipate , we found no evidence of Performance without Awareness . Although we found strong evidence of Performance > Awareness across the experiments ( Figure 3A , D ) , subsequent computational modeling ( Bayesian Ideal Observer Model section ) suggests that this is somewhat trivial: even an ideal observer is expected to show Performance > Awareness ( Figure 3G; see Bayesian Ideal Observer Model section for further explanation ) . 10 . 7554/eLife . 09651 . 005Figure 3 . Group-level results of behavioral experiments ( rows 1 and 2 ) , presented in comparison to the predictions of the Bayesian ideal observer model ( row 3; see Materials and methods - Computational Model ) . In both experiments , human observers displayed no evidence of Performance without Awareness , but appeared to demonstrate Performance > Awareness ( panels A and D ) . However , the ideal observer model also demonstrated such behavior ( panel G ) , indicating that it is not suboptimal at all but arises from the 2IFC nature of the confidence task ( see Bayesian Ideal Observer Model results section and Figure 2 caption for explanation ) . Horizontal gray lines in panels A , D , and G indicate chance-level betting ( 50% ) on the target-present ( TP ) interval . Panels B , E , and H show rising Type 2 hit rate ( ‘HR’; when subjects bet on a correct orientation discrimination choice ) but relatively flat Type 2 false alarm rate ( ‘FAR’; when subjects bet on an incorrect orientation discrimination choice ) , and panels C , F , and I show higher orientation discrimination accuracy when the target-present ( TP ) interval is bet on; these patterns suggest that human subjects and the Bayesian ideal observer were rating confidence via assessing their probability of correctly discriminating orientation , rather than target presence versus absence only . The model demonstrates good explanatory power for the data across all participants ( mean proportion of variance accounted for by the model , R2 = 0 . 565 ) . Error bars for behavioral data indicate the standard error of the mean across subjects with data in each bin . DOI: http://dx . doi . org/10 . 7554/eLife . 09651 . 005 To look for evidence of Performance without Awareness , we first plotted percent of trials in which observers bet on the target-present interval against orientation discrimination accuracy for both experiments ( Figure 3A , D ) . In contrast to what might have been suggested based on previous results ( e . g . , Boyer et al . , 2005; Charles et al . , 2013; Merikle et al . , 2001; but see Snodgrass et al . , 2004 ) , visual inspection alone clearly reveals no evidence for Performance without Awareness in either experiment: it looks as though observers could bet on the target-present interval above chance as soon as they were able to discriminate the target above chance , and there is no hint of the Performance without Awareness pattern . We quantitatively assessed the possibility of Performance without Awareness using a Bayesian observer model ( see Modeling Results , below ) , but found no evidence that a Performance without Awareness pattern could capture human behavior . Individual subjects’ performance closely resembles group data and averages ( Appendix 2 ) . Because thresholds can be defined in psychophysical terms ( 75% performance ) rather than absolute terms ( >50% ) , we also evaluated the possibility of Performance > Awareness . We used kernel smoothing regression ( see Materials and methods ) to interpolate each individual subject’s data in order to estimate how often subjects bet on the target-present interval when they were performing at 75% correct on orientation discrimination . Because results are very similar across the two experiments , we combined results from both and performed a two-tailed one-sample t-test to assess whether this predicted percentage betting on the target-present interval significantly diverged from 75% . This analysis revealed that observers bet on the target-present interval significantly less than 75% of the time at 75% correct orientation discrimination accuracy ( Figure 3A , D , Table 1 ) . Thus , observers exhibited Performance > Awareness ( but see also Modeling Results , below ) . 10 . 7554/eLife . 09651 . 006Table 1 . Individual values , means , standard deviations , and p-values for t-tests showing that Performance > Awareness occurs across both experiments . Results from Experiment 2 show that the pattern does not change with different question order or feedback . DOI: http://dx . doi . org/10 . 7554/eLife . 09651 . 006ExptSubjectp ( choose TP interval ) at p ( correct ) = 0 . 7511AVT0 . 6762AM0 . 7143JDK0 . 7164SH0 . 6825MM0 . 6846AC0 . 6857MR0 . 6748MK0 . 6589RA0 . 61921AVT0 . 6662AM0 . 7133JDK0 . 746Mean ( σ ) 0 . 686 ( 0 . 033 ) t ( 11 ) 6 . 718p0 . 00003 We developed a Bayesian ideal observer model utilizing a similar representation space as standard 2-dimensional signal detection theory ( Figure 4 ) ( King and Dehaene , 2014; Macmillan and Creelman , 2004 ) . The primary finding is that even an ideal observer model exhibits Performance > Awareness , as depicted in Figure 1B . Intuitively , this effect occurs because the orientation discrimination choice requires evaluation of only one interval ( the one with the target in it ) and therefore is corrupted by only one source of noise , but the ‘betting” choice requires evaluation of both intervals , and therefore has two potential sources of noise . 10 . 7554/eLife . 09651 . 007Figure 4 . Illustration of the Bayesian ideal observer’s 2-dimensional representation space , following standard 2-dimensional signal detection theory ( King and Dehaene , 2014; Macmillan and Creelman , 2004 ) . ( a ) Distributions Sleft and Sright lie on orthogonal axes cleft and cright representing left- and right-tilted targets , respectively , and the noise distribution lies at the origin . On each simulated trial , the model ‘sees’ two samples , one drawn from a source distribution Si to represent the target-present interval ( dTP ) and the other from the noise distribution to represent the target-absent interval ( dTA ) . It marginalizes across all contrast evidence levels to guess the orientations of both samples according to the posterior probabilities of left- and right-tilted sources . Then , it compares the posterior probabilities of the chosen orientations in each interval to select the interval with higher confidence ( p ( correct ) ) ( see Materials and methods - Bayesian ideal observer model ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09651 . 007 The model ‘performs’ a 2IFC confidence discrimination by comparing the posterior probability of left- or right-tilted source distributions given the data to perform the orientation discrimination task on each of the two intervals on each trial . Then , it uses the posterior probability of the choice it made on each interval as a measure of confidence ( i . e . , p ( correct ) ) , and compares this measure between the two intervals to select the choice it is more confident in ( see Figure 4 and Materials and methods – Bayesian ideal observer model ) . We also explored several model variants to establish the robustness of the model’s performance; see Appendix 4 for details on model variants . Unsurprisingly , the Bayesian ideal observer did not display signs of Performance without Awareness . We next evaluated whether causing the model to exhibit Performance without Awareness ( Figure 2A ) by degrading the 2IFC confidence judgment could produce better fit to participants’ data . We tested three levels of increasing decisional noise ( σd; see Materials and methods ) to cause the model to exhibit increasing Performance without Awareness as described in Figure 2A , and assessed the goodness of fit ( R2 ) for each subject for each decisional noise value . We found that causing the model to exhibit increasing Performance without Awareness behavior resulted in increasingly worse R2 values ( Table 2 ) . To confirm this trend , we conducted a 12 ( subjects; subjects 1–3 who completed both experiments are treated independently ) x 4 ( decisional noise magnitude ) repeated measures ANOVA on the R2 values . This analysis revealed a main effect of decisional noise ( F ( 3 , 33 ) = 19 . 301 , p <0 . 001 ) , indicating that the ideal observer model ( σd = 0 ) best captures human performance , and that any suboptimal Performance without Awareness ( σd >0 ) pattern fits human data more poorly than the ideal observer behavior – even without punishing the decisional noise model for having an additional parameter . 10 . 7554/eLife . 09651 . 008Table 2 . R2 values quantifying goodness of fit for ideal observer ( σd = 0 ) and three alternative decisional noise magnitudes ( σd >0 ) which cause increasing degrees of Performance without Awareness . Decisional noise greater than 0 – i . e . , increased level of Performance without Awareness – causes a drop in goodness of fit between model and human data . See Methods and Appendix 4 for more details . DOI: http://dx . doi . org/10 . 7554/eLife . 09651 . 008ExptSubjectDecisional noise σd0 ( Ideal observer ) 0 . 10 . 20 . 3110 . 4650 . 4590 . 4560 . 44720 . 5800 . 5780 . 5650 . 54430 . 4700 . 4640 . 4480 . 42840 . 3960 . 3920 . 3810 . 36350 . 6490 . 6550 . 6450 . 62860 . 4800 . 4730 . 4580 . 43470 . 4530 . 4520 . 4440 . 42780 . 6020 . 5950 . 5830 . 56390 . 5030 . 5090 . 5120 . 504210 . 6240 . 6240 . 6220 . 61220 . 7830 . 7800 . 7750 . 76630 . 7770 . 7780 . 7670 . 753Mean R2 ( σ ) 0 . 565 ( 0 . 126 ) 0 . 563 ( 0 . 128 ) 0 . 555 ( 0 . 129 ) 0 . 539 ( 0 . 131 ) Crucially , however , the ideal observer does exhibit Performance > Awareness ( Figure 3G ) , and to a similar extent as our human participants ( R2 = 0 . 565; see Appendix 4 for details of goodness of fit metrics ) ; trends for Type 2 hit and false alarm rates ( Figure 3H ) , and percent correct conditional upon having bet on the target-present versus target-absent interval ( Figure 3I ) , also match human data . That the ideal observer exhibits behavior that may seem suboptimal , and in the same pattern as human observers , confirms that this perhaps counterintuitive but optimal behavior arises from the confidence-comparison nature of the 2IFC confidence-rating task: the decision about orientation in the target-present interval is limited by one source of noise ( the single target-present interval ) , but the comparison of confidence is limited by the system’s noise in both intervals . So even if confidence monotonically increases with accuracy for the target-present interval , there will be trials in which – by chance – the discrimination choice for the blank ( target-absent ) interval happens to seem more confident , that is , its posterior probability is larger . This will happen sometimes even on trials in which the observer gets the target-present orientation discrimination correct . In these trials , the observer ( human or simulated ) will select the target-absent interval . This process will lead to the appearance of what we called Performance > Awareness , as displayed by our human participants and ideal observer ( refer also to Figure 2 for additional explanation ) . Thus , the subjective ratings by human participants are already close to ideal , as if the actual effective threshold for subjective awareness is no different from the objective threshold for discrimination . Importantly , this is true despite the apparent measured differences in psychophysically defined thresholds ( 75% ) .
Blindsight ( Weiskrantz , 1986 ) is the intriguing demonstration of Performance without Awareness in neurological patients . Despite widely held beliefs by experts , here we found no evidence that it occurs in normal observers . Importantly , although the measured psychophysical threshold ( 75% ) for awareness seemed to be above the objective discrimination threshold , computational analysis revealed that the actual effective thresholds are essentially the same; people’s subjective ratings are close to ideal , given their objective performance levels . This challenges longstanding beliefs regarding the nature of subjective versus objective thresholds in perceptual studies ( Merikle et al . , 2001; see survey results in Appendix 1 ) . Our findings cannot rule out all forms of unconscious perception , such as subliminal priming , in which the evidence for unconscious processing is typically indirect benefits in reaction times ( Hannula et al . , 2005 ) . However , our findings bear upon those studies , too . Traditionally , interpreting such effects as unconscious required that the relevant stimuli yield zero sensitivity in a direct task ( d’ = 0 ) . Recently , many have relaxed this requirement and considered subjectively reported lack of awareness as sufficient ( Pessiglione et al . , 2009; Soto et al . , 2011 ) , presumably because we ( wrongly ) believed that certain stimuli might surpass the objective threshold while still being below the subjective one . One may also argue that while objective threshold requirements are rigorous , the valid and meaningful measure is the subjective threshold ( Charles et al . , 2013; Merikle et al . , 2001 ) . Our results suggest this reasoning is flawed . If a stimulus surpasses the objective threshold , there is likely conscious experience; subjects likely report lack of awareness because they interpret the response options in relative terms in the context of stimuli of various strengths . This undermines claims that higher-cognitive phenomena – e . g . working memory , error detection , or motivation – can really operate unconsciously , if assessed with reference to subjective rather than objective thresholds ( Charles et al . , 2013; Pessiglione et al . , 2009; Soto et al . , 2011 ) . Although the 2IFC confidence-rating procedure bypasses the response bias problem , interpreting the subjective vs . objective function is non-trivial: to determine whether participants’ Performance > Awareness behavior was optimal required detailed computational analysis . An alternative approach , which may be simpler , would be to compare the objective and subjective functions between task conditions , in a rationale similar to Lau and Passingham ( 2006 ) . Although we found no evidence of ‘blindsight’ in normal observers , our study lays out the logic of what would be required to demonstrate it unequivocally . For example , it has recently been argued that TMS-induced ‘blindsight’ ( Boyer et al . , 2005 ) is contaminated by criterion bias ( Lloyd et al . , 2013 ) . 2IFC confidence-rating may help resolve such issues without invoking theoretically complicated problems concerning signal detection theory ( e . g . , Heeks and Azzopardi , 2015 ) . Thus , despite their negative nature , our findings may beget fruitful lines of inquiry to address which stimuli , procedures , or brain stimulation techniques can selectively impair subjective conscious experience , beyond impacting sheer objective processing sensitivity .
For each subject in each experiment , data were collapsed across tilt ( left/right ) , interval presentation order ( first/second ) , and session for each contrast level . At each contrast level for each subject , we next calculated ( a ) percent correct orientation discrimination , ( b ) percent of trials in which the target-present interval was chosen , ( c ) Type 2 hit rate and Type 2 false alarm rate according to standard Type 2 signal detection theoretic definitions ( Type 2 hit: correct orientation discrimination and bet on target-present interval; Type 2 false alarm: incorrect orientation discrimination and bet on target-present interval ) ( Fleming and Lau , 2014; Maniscalco and Lau , 2012 ) , and ( d ) percent correct orientation discrimination conditional on having chosen the target-present versus target-absent interval . Group-level analyses and graphical presentation were conducted by binning subjects’ data into ten equally-spaced bins of percent correct orientation discrimination performance in the range 0 . 5 – 1 and calculating the mean and standard deviation of each of the above statistics for each bin . To interpolate between discrete data points , we fitted a kernel smoothing regression function to each observer’s data , which is a non-parametric approach to estimate the conditional expectation of a random variable , EYX = fX where f is a non-parametric function . This approach is based on kernel density estimation , implementing Nadaraya-Watson kernel regression ( Nadaraya , 1964; Watson , 1964 ) via ( 1 ) f^x;K , h = ∑i=1nKhx-xiyi∑i=1nKhx-xi where K is a Gaussian kernel with bandwidth h . All analyses were carried out in Matlab R2013a ( Natuck , MA ) and SPSS Version 22 ( IBM Corporation; Armonk , NY ) . We examined the relative agreement between our model’s predictions and collected behavioral data by calculating the multinomial likelihood of the model given the observed data , which has previously been used within a signal detection framework . Details of goodness of fit calculations are described in Appendix 4 . To evaluate whether human participants exhibited Performance without Awareness , we needed to cause the model to also exhibit Performance without Awareness . We therefore degraded the 2IFC confidence judgment process in the following way: On each trial , after the orientation decision had been reached , we programmed an added decisional noise parameter , σd , such that the decision variable D calculated as in Equation 5 was corrupted by additive Gaussian noise with mean 0 and standard deviation σD , such that ( 6 ) D = log ( p ( Schosen , TP|dTP ) p ( Schosen , TA|dTA ) ) + σd This causes the model to perform closer to chance at higher levels of orientation discrimination performance , i . e . to exhibit Performance without Awareness at increasing objective performance levels ( Figure 5 ) . We tested three decisional noise magnitudes – 0 . 1 , 0 . 2 , and 0 . 3 – and calculated the goodness of fit ( see Appendix 4 ) for each σd for each subject . We also examine three other possible contributing factors: correlated noise/non-orthogonal source distributions , signal-dependent ( multiplicative ) , and signal-independent ( additive ) noise ( see Appendix 4 ) . These factors do not affect the qualitative trend of the model’s performance . For completeness , we also examine two other decision rules , detailed in Appendix 5: a heuristic observer which does not ignore contrast evidence as above , but explicitly estimates the most likely contrast level via hierarchical Bayesian inference ( Yuille and Bülthoff , 1996 ) ; and a heuristic likelihood comparison observer ( similar to Barthelmé et al . , 2009 ) . Importantly , the hierarchical model produced behavior similar to the ideal observer , indicating that such behavior is not idiosyncratic or specific only to the ideal observer presented above . The likelihood-only model , on the other hand , failed to produce predictions that matched collected behavioral data , either qualitatively or quantitatively . | In the 1980s , psychologists made an unexpected discovery while working with individuals who had become blind after sustaining damage to areas of the brain required for vision . These individuals could respond correctly to questions about the shape and location of objects in their visual field , even though they could no longer see the objects . This phenomenon became known as 'blindsight' , and it is regarded as a classic example of perception in the absence of conscious awareness . Many researchers who study consciousness believe that everyone is capable of subliminal or unconscious perception: that is , of detecting and processing stimuli without being consciously aware of them . However , studies investigating this phenomenon have produced contradictory results . Peters and Lau have now tested unconscious perception directly , using a recently developed method that overcomes some of the problems faced by previous studies . Human volunteers took part in several trials , in which they were shown two images . Each image was ‘masked’ to prevent the volunteers from consciously registering them . After each image was shown , the volunteers had to state whether a patch of gray and white stripes in the masked image was tilted to the left or to the right . However , one of the two images did not include a gray and white patch . After seeing both images in a trial , the volunteers also had to indicate which of their answers they were most confident about . If the volunteers could perceive the patches without being consciously aware of doing so , their response should show two features . The volunteers should correctly state the tilt direction of the stripes more often than would be expected if they were guessing at random . However , they should also feel no more confident in their responses for the images that did feature a striped patch than for the ‘no patch’ ones . Peters and Lau found no such evidence of unconscious perception . Nevertheless , the volunteers were consistently better at correctly stating the direction the stripes were tilted in than their confidence ratings would suggest . Does this indicate some degree of perception without awareness ? Peters and Lau argue that it does not , because a computer model designed to perform the task showed a similar level of performance to the volunteers . These findings suggest that previous reports of unconscious perception may have been contaminated by the problems that Peters and Lau controlled for , and that perhaps unconscious perception doesn’t occur in people without brain damage . Researchers will now need to do more studies using similar approaches to determine whether observers without brain damage can truly experience unconscious perception , and how such unconscious perception might be represented in the brain . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Human observers have optimal introspective access to perceptual processes even for visually masked stimuli |
Multidrug-resistant bacteria pose a serious health threat , especially in hospitals . Horizontal gene transfer ( HGT ) of mobile genetic elements ( MGEs ) facilitates the spread of antibiotic resistance , virulence , and environmental persistence genes between nosocomial pathogens . We screened the genomes of 2173 bacterial isolates from healthcare-associated infections from a single hospital over 18 months , and identified identical nucleotide regions in bacteria belonging to distinct genera . To further resolve these shared sequences , we performed long-read sequencing on a subset of isolates and generated highly contiguous genomes . We then tracked the appearance of ten different plasmids in all 2173 genomes , and found evidence of plasmid transfer independent from bacterial transmission . Finally , we identified two instances of likely plasmid transfer within individual patients , including one plasmid that likely transferred to a second patient . This work expands our understanding of HGT in healthcare settings , and can inform efforts to limit the spread of drug-resistant pathogens in hospitals .
Horizontal gene transfer ( HGT ) is a driving force behind the multidrug-resistance and heightened virulence of healthcare-associated bacterial infections ( Lerminiaux and Cameron , 2019 ) . Genes conferring antibiotic resistance , heightened virulence , and environmental persistence are often encoded on mobile genetic elements ( MGEs ) , which can be readily shared between bacterial pathogens via HGT ( Juhas , 2015 ) . While rates of HGT are not well quantified in clinical settings , prior studies have shown that MGEs can mediate and/or exacerbate nosocomial outbreaks ( Bosch et al . , 2017; Jamrozy et al . , 2017; Martin et al . , 2017; Sheppard et al . , 2016 ) . Recent studies have also demonstrated that multidrug-resistant healthcare-associated bacteria share MGEs across large phylogenetic distances ( Cerqueira et al . , 2017; Hazen et al . , 2018; Kwong et al . , 2018 ) . Understanding the dynamics of MGE transfer in clinical settings can uncover important epidemiologic links that are not currently identified by traditional infection control methodologies ( Lerminiaux and Cameron , 2019; Schmithausen et al . , 2019; Stadler et al . , 2018 ) . Methods to identify and track the movement of MGEs among bacterial populations on short timescales are limited . Bacterial whole-genome sequencing has transformed infectious disease epidemiology over the last decade ( Ladner et al . , 2019 ) , providing powerful new tools to identify and intervene against outbreaks ( Sundermann et al . , 2019b ) . Despite these advances , efforts to track MGE movement have focused almost exclusively on drug resistance and virulence genes ( Cerqueira et al . , 2017; Hardiman et al . , 2016; Martin et al . , 2017; Stadler et al . , 2018 ) , often ignoring the broader genomic context of the mobile elements themselves . Many studies rely on the identification of plasmid replicons , transposases , and other ‘marker genes’ ( Orlek et al . , 2017 ) , an approach that oversimplifies the diversity of MGEs and may lead to incomplete or erroneous conclusions about their epidemiology . While querying databases containing curated MGE-associated sequences is useful for the rapid screening of clinical isolates for known MGEs , it will not capture novel MGEs . Additionally , whole-genome sequencing using short-read technologies generates genome assemblies that usually do not resolve MGE sequences , due to the abundance of repetitive elements that MGEs often contain ( Arredondo-Alonso et al . , 2017 ) . Advances in long-read sequencing can mitigate this problem; hybrid assembly of short- and long-read sequence data allows the genomic context of chromosomal and extrachromosomal MGEs to be precisely visualized ( Cerqueira et al . , 2017; Conlan et al . , 2014; George et al . , 2017 ) . Finally , studying the epidemiology of MGEs in clinical settings requires detailed individual-level patient clinical data , without which HGT occurrence in the hospital cannot be identified ( Conlan et al . , 2014 ) . Here , we performed an alignment-based screen for shared nucleotide sequences in a large and diverse collection of bacterial genomes sampled from infections within a single hospital over an 18-month time period . With this approach , we identified shared sequences that occurred in the genomes of bacteria belonging to different genera . Because they were identical , we suspect that these sequences recently transferred between bacteria within the hospital setting . Further analysis using long-read sequencing and reference-based resolution of distinct MGEs enabled us to precisely characterize MGE architecture and cargo , and to track MGE occurrence over time . Cross-referencing our results with available patient metadata allowed us to follow these elements as they emerged and were maintained among nosocomial bacterial populations .
Our experimental workflow is depicted in Figure 1A . To identify genetic material shared between distantly related bacteria in the hospital setting , we screened a dataset containing 2173 whole-genome sequences of clinical isolates of high-priority Gram-positive and Gram-negative bacteria collected from a single hospital over an 18-month period beginning in November 2016 as part of the Enhanced Detection System for Hospital-Acquired Transmission ( EDS-HAT ) project at the University of Pittsburgh ( Sundermann et al . , 2019a ) ( Methods and Supplementary file 1 ) . To have maximal contrast , we focused on identical sequences found in the genomes of bacteria belonging to different genera . We performed an all-by-all alignment of the 2173 genomes in the dataset using nucmer ( Marçais et al . , 2018 ) , and filtered the results to retain alignments of at least 5 kb that shared 100% identity between bacteria of different genera . The resulting sequences were extracted and clustered using Cytoscape ( Figure 1B ) . We also explored alignments > 3 kb and >10 kb , and found that the number of clusters identified was highly dependent upon the alignment length cut-off used ( Figure 1—figure supplement 1 ) . We chose to use 5 kb for our analysis because of the intermediate number of resulting clusters . This approach identified shared sequences in 196 genomes belonging to 11 genera , which were grouped into 51 clusters of related sequences ( Supplementary file 2 ) . We compared the patient demographics and clinical features of the subset of patients from whom the 196 isolates encoding shared sequence clusters were derived with the other patients in the dataset ( Table 1 ) . While patient demographics were similar between groups , isolates encoding shared sequence clusters were cultured from patients with more co-morbidities ( as measured by Charlson co-morbidity index , p=0 . 03 ) , and with higher rates of solid organ transplant ( p=0 . 02 ) ( Table 1 ) . The shared sequence clusters we identified ranged in size from two to 52 genomes and comprised two , three , or four different genera ( Figure 1B ) . Shared sequences were found predominantly among Gram-negative Enterobacteriaceae , particularly Klebsiella spp . , Escherichia coli , and Citrobacter spp . ( Figure 1C ) . Annotation of clustered sequences confirmed that more than 80% of clusters encoded one or more genes involved in DNA mobilization , such as plasmid replication , integration , or other mobile functions presumably involved in HGT ( Figure 1D and Supplementary file 2 ) . Approximately one-quarter of the clusters contained antimicrobial resistance genes , including genes encoding resistance to aminoglycosides , antifolates , beta-lactams , macrolides , quinolones , sulphonamides , and tetracyclines ( Figure 1D and E ) . Finally , 8 of 51 clusters encoded genes and operons whose products were predicted to interact with metals , including arsenic , copper , mercury , nickel , and silver ( Figure 1D ) . Collectively , these results indicate that our systematic , alignment-based method successfully identified sequences associated with MGEs , particularly in pathogens known to engage in HGT ( Huddleston , 2014; Juhas , 2015 ) . To assess the phylogenetic distribution of the shared sequence clusters we identified , we constructed a core gene-based phylogeny of the 196 genomes encoding one or more clusters using the Genome Taxonomy Database Tool Kit ( GTDBTK ) ( Parks et al . , 2018; Figure 2 ) . Shared sequence clusters were often found among bacteria in related genera , in particular the Enterobacteriaceae . We did not observe any shared sequences that were present in both Gram-positive and Gram-negative isolate genomes , but we did find shared sequences in the genomes of distantly related bacteria . For example , we identified a shared sequence cluster comprised of three aminoglycoside resistance genes that was identical between a vancomycin resistance-encoding plasmid carried by Enterococcus faecium and the Clostridioides difficile chromosome ( cluster C9 , Figure 3A ) . The C . difficile strain carrying this element was previously found to also harbor an npmA aminoglycoside resistance gene ( Marsh et al . , 2019b ) . Separately , we found a section of an integrative conjugative element that was identical between two Pseudomonas aeruginosa isolates and one Serratia marcescens isolate ( cluster C30 , Figure 3B ) . Identical regions of this element included formaldehyde resistance genes and Uvr endonucleases . Finally , we detected complete and identical Tn7 transposons in the genomes of Acinetobacter baumannii , E . coli , and Proteus mirabilis isolates ( cluster C17 , Figure 3C ) . The Tn7 sequence we detected was also identical to the Tn7 sequence of pR721 , an E . coli plasmid that was first described in 1990 and was sequenced in 2014 ( Komano et al . , 1990 ) . To further investigate the genomic context of the shared sequence clusters we identified , we selected the isolate containing the longest sequence in each cluster from C1-C5 for long-read sequencing using Oxford Nanopore technology . Hybrid assembly combining short Illumina reads and long Nanopore reads generated highly contiguous chromosomal and plasmid sequences , which allowed us to resolve MGEs carrying one or more of the most prevalent shared sequence clusters ( Table 2 ) . We found that several of the shorter and more prevalent shared sequences were carried on a variety of different plasmid and chromosomal MGEs , and furthermore , the sequences co-occurred in different orders , orientations , and combinations ( Table 2 , Figure 4A ) . This kind of ‘nesting’ of mobilizable sequences within larger MGEs has been previously observed ( Sheppard et al . , 2016 ) , and our findings further support the mosaic , mix-and-match nature of the shorter shared sequences we identified . We also confirmed that these shared sequences were indeed mobilizable , since they were found independently within multiple distinct , larger MGEs . A closer examination of the three largest shared sequence clusters ( C1 , C2 , C3 ) showed that C1 sequences did not all share a common ‘core’ nucleotide sequence , but rather could be aligned in a pairwise fashion to generate a contiguous ‘chain’ of sequences ( Figure 4B ) . Clusters C2 and C3 , on the other hand , did contain ‘core’ sequences that were present in all genomes containing the cluster ( Figure 4C and D ) . More than half ( 104/196 ) of the genomes encoding shared sequence clusters contained one or more of the five most prevalent clusters ( C1-C5 , Figure 1B ) . In all five cases , the shared sequences were short ( usually less than 10 kb ) , and they were predicted to be carried on plasmids shared between Enterobacteriaceae . We set out to resolve the genomic context of each of these five clusters in all isolates containing them . We used an iterative approach that started with long-read sequencing and hybrid assembly of the earliest isolate in each cluster to generate reference sequences of cluster-containing MGEs ( chromosomal or plasmid ) ( Supplementary file 3 ) . Then we mapped contigs from Illumina-only assemblies to the MGE reference sequences to assess their coverage in other genomes , using a cutoff of >90% coverage to define an MGE as potentially transferred between isolates ( Materials and methods ) . This approach allowed us to query the presence of MGEs from genomes sequenced with Illumina technology alone , without requiring long-read sequencing of all isolates or relying on external references . We found that 11 of the 104 isolates ( all E . coli ) carried cluster C1 and C3 sequences on their chromosome , while the remaining 93 isolates carried cluster C1-C5 sequences on 17 distinct plasmids . Seven of these plasmids were present in only one isolate in the dataset , but ten plasmids appeared to be shared between more than one isolate ( Table 2 , Figure 5 ) . We also conducted the same reference-based coverage analysis for all 2173 genomes in the original dataset , and identified an additional 16 isolates with >90% coverage of an MGE encoding C1-C5 sequences ( Supplementary file 4 ) . While all the shared sequences we originally identified were present in the genomes of bacteria belonging to different genera , the plasmids that we resolved were variable in how widely they were shared . For example , two plasmids were only found among isolates belonging to a single species and multilocus sequence type ( ST ) , suggesting that they were likely transmitted between patients along with the bacteria that were carrying them ( Figure 5A ) . These included an IncF blaKPC-3 carbapenemase-encoding plasmid ( pKLP00149_2 ) found in 17 K . pneumoniae isolates belonging to ST258 , a multidrug-resistant and highly virulent hospital-adapted bacterial lineage that has recently undergone clonal expansion in our hospital ( Marsh et al . , 2019a ) . All isolates carrying this plasmid belonged to Clade II of ST258 , which has caused multiple outbreaks at our center ( Figure 5—figure supplement 1; Marsh et al . , 2019a ) . We also found an IncF blaOXA-1 extended spectrum beta-lactamase-encoding plasmid in eight E . coli isolates belonging to ST131 , another multidrug-resistant and hypervirulent clone ( Manges et al . , 2019 ) . As above , this plasmid was found in closely related ST131 isolates ( Figure 5—figure supplement 1 ) , suggesting that it was vertically transmitted along with the bacteria carrying it . In addition to plasmids that occurred in bacteria belonging to the same ST , we also identified plasmids that were present in isolates belonging to different STs of the same species , or in different species of the same genus ( Figure 5B ) . All isolates in this case were K . pneumoniae or K . oxytoca , suggesting widespread sharing of plasmids between distinct Klebsiella species and STs . The plasmids all carried antibiotic resistance genes , and many also carried metal interaction genes ( Table 2 ) . Finally , we identified three different plasmids that were shared between different bacterial genera all belonging to the Enterobacteriaceae ( Figure 5C ) . One 9 . 5 kb ColRNAI plasmid ( pKLP00155_6 ) carrying the colicin bacterial toxin was found in 26 isolates belonging to 10 different STs and 4 different genera . Taken together , these results indicate that some plasmids carrying putative MGEs were likely inherited vertically as bacteria were transmitted between patients in the hospital , while others appear to have transferred independently of bacterial transmission . By cross-referencing the isolates containing shared plasmids with de-identified patient data , we found two instances of identical plasmids present in pairs of isolates of different genera that were collected from the same patient , on the same date , and from the same sample source ( Figure 6 ) . A K . pneumoniae ST405 isolate ( KLP00215 ) and an E . coli ST69 isolate ( EC00678 ) collected from a tissue infection from Patient A each harbored a 113 . 6 kb IncF plasmid carrying blaKPC-2 , blaOXA-9 , and blaTEM-1A enzymes , as well as a mercury detoxification operon ( Figure 6A , B ) . An isolate from a second patient ( Patient B , EC00701 , E . coli ST131 ) , which was cultured 109 days after the isolates from Patient A , also encoded a nearly identical plasmid . A systematic chart review for Patients A and B revealed that they occupied adjacent hospital rooms for four days during a time period after Patient A’s isolates were collected but before Patient B’s isolate was collected . During this time , the two patients were treated by the same healthcare staff , who might have transferred bacteria between them . In the second case of putative within-patient HGT , a K . pneumoniae ST231 isolate ( KLP00187 ) and a Citrobacter braakii ST356 isolate ( CB00017 ) were both collected from the same urine sample of Patient C ( Figure 6C ) . Both isolates carried nearly identical 196 . 8 kb IncF plasmids conferring resistance to aminoglycosides , beta-lactams , chloramphenicol , fluoroquinolones , sulfonamides , tetracyclines , and trimethoprim , as well as operons encoding copper and arsenic resistance ( Table 2 ) . Furthermore , isolates from two subsequent patients ( Patient D and Patient E ) also carried plasmids that were similar to the plasmid shared between KLP00187 and CB00017 . Alignment of the sequences of all four plasmids showed that the plasmids isolated from Patient C were nearly identical , while the plasmids from Patients D and E had small differences in their gene content and organization ( Figure 6C ) . A systematic chart review did not identify any strong epidemiologic links between the three patients , suggesting that this plasmid was not passed directly between these patients and might instead have transferred via additional bacterial populations that were not sampled .
Through this study , we have produced a high-resolution view of shared sequence and MGE dynamics among clinical bacterial isolates collected over an 18-month period from a single hospital . We identified , clustered , and characterized identical sequences found in multiple distinct genera , and in the process uncovered both expected and unexpected cases of shared sequence occurrence . We confirmed that some of the most common shared sequences identified were fragments of larger MGEs . We performed long-read sequencing to resolve these larger elements , and in doing so we characterized a large diversity of drug resistance-encoding plasmids . When we traced the presence of various plasmids over time , we found some that were likely transferred vertically along with the bacteria carrying them , and others that appeared to be transferred horizontally between unrelated bacteria . Our study adds to the body of knowledge of HGT in hospital settings in new and important ways . We analyzed a large set of clinical isolates collected from a single health system , and used a systematic approach to identify shared nucleotide sequences , regardless of their type or gene content . While prior studies have used genomic epidemiology to study how HGT contributes to the transmission , persistence , and virulence of bacterial pathogens ( Bosch et al . , 2017; Martin et al . , 2017; Schweizer et al . , 2019; Valenzuela et al . , 2007 ) , the technical challenges of resolving MGEs from whole-genome sequencing data have limited the scope of these findings ( Arredondo-Alonso et al . , 2017 ) . Furthermore , while rates of HGT between pathogenic bacteria have been quantified in vitro , very little information is currently available to assess rates of HGT in vivo or in clinical settings ( Leclerc et al . , 2019 ) . Other studies have deliberately tracked HGT in healthcare settings by focusing either on mobile genes of interest , such as those encoding drug resistance ( Cerqueira et al . , 2017; Hardiman et al . , 2016; Hazen et al . , 2018 ) , or on specific classes of MGEs ( Savinova et al . , 2019 ) . Both of these approaches likely generate incomplete accounts of the extent of HGT in clinical settings . For this reason , we selected a pairwise alignment-based approach , whereby we only looked for identical sequences in the genomes of very distantly related bacteria . In doing so , we did not limit ourselves to only focusing on ‘known’ MGEs , and thus obtained a more accurate and comprehensive overview of the dynamics of HGT between bacterial genera in our hospital . What might cause horizontally-transferred nucleotide sequences to be found at very high identity within phylogenetically distinct bacteria ? Among many possible causes , we could consider the following: ( 1 ) the sequences we identified could have been recently transferred and not have had time to diverge from one another; ( 2 ) they could already be well adapted to optimally perform their functions; or ( 3 ) they could represent genetic elements that are highly intolerant to mutation . We suspect that our dataset contains all three cases . First , in the instances of likely within-patient HGT , both plasmids isolated from the same patient were nearly identical to one another . This suggests that if mutation rates of plasmids are similar to bacterial chromosomes , these plasmids would have transferred shortly before the bacteria were isolated . In both cases of likely within-patient HGT we also observed similar plasmids in the genomes of isolates from other patients , but we identified a likely route of transfer between patients only in the case where the subsequent plasmid was also nearly identical . This finding supports our theory that high plasmid identity is evidence of recent transfer . Second , the plasmids that we identified only in ST258 K . pneumoniae or in ST131 E . coli are likely well adapted to these lineages , perhaps because plasmid-imposed fitness costs have already been resolved through compensatory adaptations ( San Millan , 2018 ) . Third , the Tn7 transposon sequence we uncovered , which was identical in bacterial isolates from three different genera , was also identical to over two dozen publicly available genome sequences queried through a standard NCBI BLAST search . The insertion of the Tn7 transposon downstream of glmS in all of our isolates suggests TnsD-mediated transposition ( Parks and Peters , 2009 , p . 7 ) , but the reason why the entire transposon sequence remains so highly conserved remains unclear . The vast majority of shared sequences identified through our approach contained signatures of mobile elements , and our follow-up work demonstrated that these sequences could very likely move independently and assemble in a mosaic fashion on larger mobile elements like plasmids , integrative conjugative elements , and other genomic islands . Antibiotic resistance genes were present in only a subset of the shared sequence clusters we identified , which was somewhat surprising given how many resistance genes are known to be MGE-associated . Our follow-up analysis showed , however , that resistance genes were indeed highly prevalent among many of the MGEs that we resolved . This finding is consistent with a recent study of clinical K . pneumoniae genomes , which showed that while antibiotic resistance genes were largely maintained at the population level , they were variably present on different MGEs that fluctuated in their prevalence over time ( Ellington et al . , 2019 ) . Finally , we were surprised by the large number of metal-interacting genes and operons within the shared sequences that we identified . While metal-interacting genes and operons have been hypothesized to confer disinfectant tolerance and increased virulence ( Chandrangsu et al . , 2017; McDonnell and Russell , 1999 ) , precisely how these elements might increase bacterial survival in the hospital environment and/or contribute to infection requires further study . Identification of risk factors and common exposures for HGT has previously been proposed ( Conlan et al . , 2014; Hardiman et al . , 2016; Lerminiaux and Cameron , 2019; Pecora et al . , 2015 ) , but the results of prior efforts have been limited because large genomic datasets from single health systems with corresponding epidemiologic data have not been widely available ( Struelens , 1998 ) . The use of routine whole-genome sequencing for outbreak surveillance in our hospital has allowed us to begin to study how the horizontal transfer of MGEs might be similar or different from bacterial transmission . In addition to finding evidence of vertical transfer of plasmids accompanying bacterial transmission , we also identified several cases in which the same MGE was identified in two or more isolates of different sequence types , species , or genera . In some cases , these isolates were collected within days or weeks of one another . This finding highlights the frequent movement of MGEs between bacterial populations , particularly in hospitalized patients ( Huddleston , 2014; Lerminiaux and Cameron , 2019 ) , and points to the importance of pairing genome sequencing with epidemiologic data to uncover routes of MGE transmission . There are several limitations to our study . First , the dataset that we used only contained genomes of isolates from clinical infections from a pre-selected list of species , and did not include environmental samples or isolates from patient colonization . In the case of between-patient plasmid transfer that we identified , we do not know exactly how the plasmid was transferred from Patient A to Patient B because we did not collect these intermediaries . Second , our method to screen for shared sequences based on cross-genus alignment was based on arbitrary alignment length and identity cutoffs . As expected , we detected more clusters at shorter alignment cut-offs , and we suspect that decreasing the identity threshold would also result in the identification of more and bigger clusters . Additionally , we did not consider sequences found in different bacteria within a single genus for the purposes of cluster identification . The cross-genus parameter we employed may have also artificially enriched the number of MGEs identified among Enterobacteriaceae , which are known to readily undergo HGT with one another ( Cerqueira et al . , 2017 ) . Third , we assigned MGE presence relative to single reference sequences , and based our analysis on reference sequence coverage; subsequent MGEs that either gained additional sequence or rearranged their contents would still be assigned the same MGE , even though they may have diverged substantially from the reference MGE ( Sheppard et al . , 2016 ) . Finally , this study was based exclusively on comparative genome analyses , and the MGEs we resolved from clinical isolate genomes were not tested for their capacity to undergo HGT in vitro . In conclusion , we have shown how bacterial whole genome sequence data , which is increasingly being generated in clinical settings , can be leveraged to study the dynamics of HGT between drug-resistant bacterial pathogens within a single hospital . Our future work will include further characterization of the shared sequences and MGEs we resolved , assessment of sequence sharing across closer genetic distances ( such as within-genus transfer ) , exploration of MGE and host co-evolution , and incorporation of additional epidemiologic information to identify shared exposures and possible routes for MGE transfer independent from bacterial transmission . Ultimately , we aim to develop this analysis into a reliable method that can generate actionable information and enhance traditional approaches to prevent and control multidrug-resistant bacterial infections .
Isolates were collected through the Enhanced Detection System for Hospital-Acquired Transmission ( EDS-HAT ) project at the University of Pittsburgh ( Sundermann et al . , 2019a ) . Eligibility of bacterial isolates for genome sequencing under EDS-HAT required positive clinical culture for high-priority and multidrug-resistant bacterial pathogens with either of the following criteria: >3 hospital days after admission , and/or any procedure or prior inpatient stay in the 30 days prior to isolate collection . Bacterial isolates were collected between November 2016 and May 2018 . Pathogens collected included: Acinetobacter spp . , Burkholderia spp . , Citrobacter spp . , Clostridioides difficile , vancomycin-resistant Enterococcus spp . , extended-spectrum beta-lactamase ( ESBL ) -producing E . coli , ESBL-producing Klebsiella spp . , Proteus spp . , Providencia spp . , Pseudomonas spp . , Serratia spp . , Stenotrophomonas spp . , and methicillin-resistant S . aureus . Eligible isolates were identified using TheraDoc software ( Version 4 . 6 , Premier , Inc , Charlotte , NC ) . The EDS-HAT project involves no contact with human subjects; the project was approved by the University of Pittsburgh Institutional Review Board and was classified as being exempt from informed consent . To assess patient demographics and co-morbidities , information was collected from available patient records and was summarized by an honest broker . In order to define the severity of illness and morbidity for patients included in the study , the Charlson Comorbidity Index score was calculated using ICD-9 and ICD-10 visit diagnoses from inpatient and outpatient encounters in the one year prior to each patient’s admission , including the admission during which a study isolate was collected ( Quan et al . , 2005 ) . For patients that had multiple isolates , demographic and clinical information was reported from the date of the first isolate collected . Differences in demographic and clinical factors between patient groups were assessed using Fisher’s Exact test for categorical variables and Wilcoxon rank-sum test for continuous variables . Genomic DNA was extracted from pure overnight cultures of single bacterial colonies using a Qiagen DNeasy Tissue Kit according to manufacturer’s instructions ( Qiagen , Germantown , MD ) . Illumina library construction and sequencing were conducted using the Illumina Nextera DNA Sample Prep Kit with 150 bp paired-end reads , and libraries were sequenced on the NextSeq sequencing platform ( Illumina , San Diego , CA ) . Selected isolates were also sequenced with long-read technology on a MinION device ( Oxford Nanopore Technologies , Oxford , United Kingdom ) . Long-read sequencing libraries were prepared and multiplexed using a rapid multiplex barcoding kit ( catalog SQK-RBK004 ) and were sequenced on R9 . 4 . 1 flow cells . Base-calling on raw reads was performed using Albacore v2 . 3 . 3 or Guppy v2 . 3 . 1 ( Oxford Nanopore Technologies , Oxford , UK ) . Illumina sequencing data were processed with Trim Galore v0 . 6 . 1 to remove sequencing adaptors , low-quality bases , and poor-quality reads . Bacterial species were assigned by k-mer clustering with Kraken v1 . 0 ( Wood and Salzberg , 2014 ) and RefSeq ( Pruitt et al . , 2007 ) databases . Genomes were assembled with SPAdes v3 . 11 ( Bankevich et al . , 2012 ) , and assembly quality was verified using QUAST ( Gurevich et al . , 2013 ) . All genomes generated by the EDS-HAT project during the 18-month time period from November , 2016 through May , 2018 were included in this study , as long as the genome assemblies had: ( a ) coverage ( read depth ) >40X , ( b ) genome length within 20% of the expected size for the genus ( c ) a total number of contigs less than 400 and , ( d ) an N50 greater than 50 kb . Genomes were annotated with Prokka v1 . 13 ( Seemann , 2014 ) . Multi-locus sequence types ( STs ) were assigned using PubMLST typing schemes with mlst v2 . 16 . 1 ( Jolley and Maiden , 2010; Seemann , 2014 ) , and ribosomal sequence types ( rMLSTs ) for isolates not assigned an ST were approximated by alignment to rMLST reference sequences . Long-read sequence data was combined with Illumina data for the same isolate , and hybrid assembly was conducted using Unicycler v0 . 4 . 7 or v0 . 4 . 8-beta ( Wick et al . , 2017 ) . Illumina genome assemblies were screened all-by-all against one another using nucmer v4 . 0 . 0beta2 ( Marçais et al . , 2018 ) . The nucmer output was filtered to only include alignments between isolates of different bacterial genera of at least 5 , 000 bp at 100% identity . Nucleotide sequences from the resulting alignments were then extracted and compared against one another by all-by-all BLASTn v2 . 7 . 1 ( Altschul et al . , 1990 ) . Results were filtered to only include nucleotide sequences having 100% identity over at least 5000 bp to at least one sequence from another genus . The resulting comparisons were clustered and visualized using Cytoscape v3 . 7 . 1 ( Shannon et al . , 2003 ) . A phylogeny of shared sequence cluster-encoding genomes was constructed using the Genome Taxonomy Database Tool Kit ( GTDBTK ) ( Parks et al . , 2018 ) . Briefly , translated amino acid sequences of 120 ubiquitous bacterial genes were generated , concatenated , and aligned using GTDBTK’s identify pipeline . The resulting multiple sequence alignment was masked for gaps and uncertainties , then a phylogenetic tree was generated using RAxML v8 . 0 . 26 with the PROTGAMMA substitution model ( Stamatakis , 2014 ) and 1000 iterations . Additional core genome phylogenies were generated for ST258 K . pneumoniae and ST131 E . coli genomes using snippy ( v4 . 4 . 5; https://github . com/tseemann/snippy ) and RAxML ( Stamatakis , 2014 ) . The longest nucleotide sequence in each shared sequence cluster was considered representative of that cluster , and was annotated with Prokka v1 . 13 . Representative sequences were compared to publicly available genomes by BLASTn v2 . 7 . 1 against the NCBI Nucleotide database . Antibiotic resistance genes were identified by a BLASTn-based search against the CARD v3 . 0 . 1 ( Jia et al . , 2017 ) and ResFinder v3 . 2 ( Zankari et al . , 2012 ) databases . Plasmid replicons and MOB families were identified by a BLASTn-based search against the PlasmidFinder database v2 . 0 . 2 ( Carattoli et al . , 2014 ) , the plasmid MLST website ( https://pubmlst . org/plasmid; Jolley et al . , 2018 ) , and MOBscan ( Garcillán-Barcia et al . , 2020 ) . Additional features of each shared sequence cluster were identified by consulting annotations assigned by Prokka . Sequences were aligned to one another using Geneious v11 . 1 . 5 ( Biomatters Ltd . , Auckland , New Zealand ) and EasyFig v2 . 2 . 2 ( Sullivan et al . , 2011 ) , and circular plots were generated with Circos ( Krzywinski et al . , 2009 ) . To resolve the MGEs encoding shared sequence clusters C1-C5 , we first selected the earliest isolate containing each cluster for long-read sequencing and hybrid assembly . The closed , cluster-encoding mobile element ( plasmid or chromosomal ) from this earliest isolate was used as a reference for mapping contigs from Illumina assemblies from all other isolates using BLASTn . Briefly , contigs from Illumina-only assemblies were aligned to each reference MGE , and MGEs were called present in isolates having at least 90% coverage of a reference MGE . Among isolates having less than 90% coverage , a representative was again selected for long-read sequencing and hybrid assembly , and the process was repeated until all 104 isolates had been assigned to a MGE . Names of MGEs include the MGE type ( c = chromosomal , p=plasmid ) , the reference isolate , and the hybrid assembly contig number , denoted with an underscore at the end of the name . Plasmids resolved through hybrid assembly were also used as reference sequences to query their presence in the entire 2173 genome data set using the same BLASTn coverage-based analysis as above , using a 90% coverage cut-off based on mapping of contigs from Illumina assemblies . When isolate genomes showed high coverage of multiple reference plasmids , the longest plasmid having at least 90% coverage was recorded . For the coverage-based analysis , we considered all isolates , regardless of whether or not their MGEs were shared across genera . Patients whose isolates carried the two plasmids found to putatively transfer within individual patients were reviewed using a systematic approach modified from previously published methodologies examining patient locations and procedures for potential similarities ( Eyre et al . , 2013; Ward et al . , 2019 ) . Patients were considered infected/colonized with the recovered plasmid on the day of the patients’ culture and all subsequent days . Potential transfer events were considered significant for locations if an uninfected/uncolonized patient was housed on the same unit location or service line location ( units with shared staff ) at the same time or different time as a patient infected/colonized with the plasmid , using a 60-day window prior to the newly infected/colonized patient’s culture date . Additionally , procedures ( e . g . operating room procedures , bedside invasive procedures ) were evaluated for commonalities among all patients 60 days prior to infection/colonization , as well as potential procedures contaminated by prior infected/colonized patients that could have transferred to newly infected/colonized patients , again using a 60-day window prior to the culture date . Procedures were deemed significant if >1 patient had a similar procedure , or if there was a shared procedure within the 60-day window . | Bacteria are able to pass each other genes that make them invulnerable to antibiotics . This exchange of genetic material , also called horizontal gene transfer , can turn otherwise harmless bacteria into drug-resistant ‘superbugs’ . This is particularly problematic in hospitals , where bacteria use horizontal gene transfer to become resistant to several antibiotics and disinfectants at once , leading to serious infections that are difficult to treat . How can scientists stop bacteria from sharing genes with one another ? To answer this question , first it is important to understand how horizontal gene transfer happens in the bacteria that cause infections in hospitals . To this end , Evans et al . examined the genomes of over 2000 different bacteria , collected from a hospital over 18 months , for signs of horizontal transfer . First the experiments identified the genetic material that had potentially been transferred between bacteria , also known as ‘mobile genetic elements’ . Next , Evans et al . examined the data of patients who had been infected with the bacteria carrying these mobile genetic elements to see whether horizontal transfer might have happened in the hospital . By combining genomics with patient data , it was determined that many of the mobile genetic elements identified were likely being shared among hospital bacteria . One of the mobile genetic elements identified was able to provide resistance to several drugs , and appeared to have been horizontally transferred between bacteria infecting two separate patients . The findings of Evans et al . show that the horizontal transfer of mobile genetic elements in hospital settings is likely frequent , but complex and difficult to study with current methods . The results of this study show how these events can now be tracked and analyzed , which may lead to new strategies for controlling the spread of antibiotic resistance . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"epidemiology",
"and",
"global",
"health",
"microbiology",
"and",
"infectious",
"disease"
] | 2020 | Systematic detection of horizontal gene transfer across genera among multidrug-resistant bacteria in a single hospital |
Synaptic transmission from Drosophila photoreceptors to lamina neurons requires recycling of histamine neurotransmitter . Synaptic histamine is cleared by uptake into glia and conversion into carcinine , which functions as transport metabolite . How carcinine is transported from glia to photoreceptor neurons remains unclear . In a targeted RNAi screen for genes involved in this pathway , we identified carT , which encodes a member of the SLC22A transporter family . CarT expression in photoreceptors is necessary and sufficient for fly vision and behavior . Carcinine accumulates in the lamina of carT flies . Wild-type levels are restored by photoreceptor-specific expression of CarT , and endogenous tagging suggests CarT localizes to synaptic endings . Heterologous expression of CarT in S2 cells is sufficient for carcinine uptake , demonstrating the ability of CarT to utilize carcinine as a transport substrate . Together , our results demonstrate that CarT transports the histamine metabolite carcinine into photoreceptor neurons , thus contributing an essential step to the histamine–carcinine cycle .
Histaminergic neurotransmission plays an important role in a variety of mammalian and invertebrate behavioral processes and contributes to the regulation of arousal , sleep and circadian rhythms ( Nall and Sehgal , 2014; Panula and Nuutinen , 2013 ) . Defects in histaminergic signaling are linked to multiple neurodegenerative diseases , depression and Tourette’s syndrome ( Panula and Nuutinen , 2013; Shan et al . , 2015; Castellan Baldan et al . , 2014 ) . Synaptic clearance and recycling of monoamine neurotransmitters , including serotonin and dopamine , depend on transporters that promote uptake into the presynaptic terminals or the surrounding glia ( Torres and Amara , 2007 ) . For histamine , such transporter activities have been documented in mammalian and invertebrate glia ( Yoshikawa et al . , 2013; Edwards and Meinertzhagen , 2010 ) . In the Drosophila visual system , histidine decarboxylase ( Hdc ) generates histamine in photoreceptor neurons ( Burg et al . , 1993 ) . Light-induced depolarization of these neurons ( Wang and Montell , 2007 ) promotes synaptic release of histamine , which opens histamine-gated chloride channels on postsynaptic L1 and L2 lamina neurons , triggering their hyperpolarization ( Pantazis et al . , 2008 ) . These postsynaptic voltage changes can be followed with electroretinograms ( ERGs ) that reveal ON and OFF transient peaks coinciding with initiation and cessation of the light source ( Alawi and Pak , 1971; Heisenberg , 1971 ) . In addition to identifying mutants that interfere with presynaptic release ( Kim et al . , 2012 ) , this phenotype has also contributed to the genetic dissection of histamine recycling ( Edwards and Meinertzhagen , 2010 ) . Synaptic histamine is taken up into epithelial glia that completely envelope photoreceptor synapses ( Meinertzhagen and O'Neil , 1991 ) . Within these glia , histamine is modified by the Ebony-catalyzed condensation with β-alanine ( Richardt et al . , 2002; Borycz et al . , 2002; Hartwig et al . , 2014; Richardt et al . , 2003 ) . The reaction product carcinine is transferred to photoreceptors , which recover histamine by Tan-catalyzed hydrolysis ( Borycz et al . , 2002; True et al . , 2005; Wagner et al . , 2007 ) . This histamine–carcinine cycle ( see below: Figure 4—figure supplement 1 ) , reminiscent of the glutamate–glutamine cycle in the mammalian brain ( Bröer and Brookes , 2001 ) , is necessary to maintain visual neurotransmission ( Borycz et al . , 2005 ) and depends on the compartmentalization of glial Ebony and neuronal Tan ( Stuart et al . , 2007 ) . Mutants for ebony or tan lack ON and OFF transients ( Heisenberg , 1971; Hotta and Benzer , 1969; Chaturvedi et al . , 2014 ) , indicating that neither histamine synthesis in photoreceptor neurons by Hdc nor a putative direct re-uptake mechanism is sufficient to sustain neurotransmitter release at photoreceptor synapses ( Ziegler et al . , 2013 ) . The identity of the transporters that facilitate glial uptake of histamine from the synaptic cleft and transport of its metabolite , carcinine , out of glia and into photoreceptor neurons remains a long-standing mystery , ( Edwards and Meinertzhagen , 2010; Stuart et al . , 2007; Romero-Calderón et al . , 2007; 2008 ) . Here , we show that the previously uncharacterized CG9317 gene , that we named carT , encodes a transporter responsible for uptake of carcinine into photoreceptor neurons . We demonstrate that CarT activity is a prerequisite for visual neurotransduction .
The Drosophila genome is estimated to contain 603 transmembrane transporters ( Ren et al . , 2007; Featherstone , 2011 ) . To concentrate on those transporters with a higher probability of participating in histamine neurotransmitter recycling , we focused on families that contain individual members previously associated with neurotransmitter transport . This includes known serotonin , dopamine , gamma-aminobutyric acid , and glutamate neurotransmitter transporters in the solute-linked carrier ( SLC ) families SLC1 and SLC6 , and other transporters in the SLC17 , SLC18 and SLC22A families ( César-Razquin et al . , 2015 ) . In addition , ATP-binding cassette transporters have been implicated in altered neurotransmitter distribution ( Borycz et al . , 2008 ) . In Drosophila , these transporter families contain 137 members that we considered as candidates potentially involved in the histamine–carcinine cycle . To test their possible roles in photoreceptor neurons , we knocked down each candidate gene individually using Glass Multiple Response element ( GMR ) -Gal4 to drive double-stranded RNA expression in photoreceptors and evaluated visual signal transduction by ERG recordings . The screen yielded a single transporter gene , CG9317 , whose knockdown caused severely reduced ON and OFF transients ( Figure 1A ) . To test the consistency of this result in multiple genetic backgrounds and confirm its specific requirement in photoreceptor neurons , we expressed the CG9317 RNAi transgene with two additional photoreceptor-specific Gal4 lines , longGMR-Gal4 ( Wernet et al . , 2003 ) and otd1 . 6-Gal4 ( McDonald et al . , 2010 ) and the pan-glial driver , repo-Gal4 ( Sepp et al . , 2001 ) . ERG recordings indicated that CG9317 is necessary in photoreceptor neurons , but not glia , for proper histaminergic transduction in the visual system ( Figure 1A–C ) . Of note , compared with wild type , none of these genotypes displayed significantly reduced sustained negative potentials ( Figure 1D ) , indicating that CG9317 knockdown did not affect overall photoreceptor health ( Williamson et al . , 2010 ) . Based on these results and the findings presented below , we will refer to CG9317 as carcinine transporter , abbreviated carT . 10 . 7554/eLife . 10972 . 003Figure 1 . Photoreceptor specific knockdown of CG9317 blocks visual neurotransduction . ( A ) ERGs recorded from female flies expressing a UAS-CG9317 RNAi transgene ( VDRC 101145 ) targeting CG9317 under the control of the indicated Gal4 driver specific for photoreceptors ( GMR-Gal4 , longGMR-gal4 and otd1 . 6-Gal4 ) or glia ( repo-Gal4 ) . Quantifications of ( B ) ON transients , ( C ) OFF transients , and ( D ) sustained negative photoreceptor potentials were averaged from three replicate experiments , including at least 45 traces from 15 flies . Arrows indicate SNP , ON and OFF transient of the control recording . Graphs report upper and lower quartiles ( box ) and minimum and maximum values ( whiskers ) . ns , not significant; *p < 0 . 05 , ****p < 0 . 0001 compared to OreR;GMR-Gal4 control . ERGs , electroretinograms; SNPs , sustained negative potentials . DOI: http://dx . doi . org/10 . 7554/eLife . 10972 . 00310 . 7554/eLife . 10972 . 004Figure 1—source data 1 . List of transporter genes tested by GMR-Gal4 driven knockdown . DOI: http://dx . doi . org/10 . 7554/eLife . 10972 . 004 To test whether carT is the causal gene , the knockdown of which is responsible for loss of ERG ON and OFF transient components , we generated null alleles by clustered regularly-interspaced short palindromic repeats ( CRISPR ) /Cas9-mediated mutagenesis ( Gratz et al . , 2013; Bassett et al . , 2013; Yu et al . , 2013 ) . We isolated three individual CRISPR-induced deletions: 11 bp for carT16A , 26 bp for carT16B and 56 bp for carT43 ( Figure 2A ) . Each of these mutations resulted in a frameshift within the first transmembrane domain ( Figure 2B ) followed by a premature stop codon within 17 , 19 and 1 bp , respectively . All three carT alleles were homozygous viable and fertile , but lacked the ON and OFF transients , indicating that carT is required for visual signal transduction ( Figure 2C , D ) . For further analysis , we focused on the allele with the largest deletion , carT43 . To prove that loss of CarT function causes the ERG-phenotype in carT43 mutants , we tested whether it could be rescued by cell type-specific expression of the CarT transcript . When driven in carT43 mutant photoreceptor neurons by longGMR-Gal4 , CarT restored ON and OFF transient components of ERG recordings ( Figure 2C , D ) . By contrast , glial-specific expression of CarT failed to rescue ON and OFF transient defects ( Figure 2C , D ) . To further probe CarT function , we generated a Myc-CarT transgene . This N-terminally tagged transporter was fully functional , as its expression in carT43 also rescued defects in ON and OFF transients ( Figure 2C , D ) . Taken together , these data indicate that expression of CarT in photoreceptor neurons is necessary and sufficient for visual signal transduction . 10 . 7554/eLife . 10972 . 005Figure 2 . CarT is required in photoreceptors but not glia for normal cellular and behavioral visual responses . ( A ) CRISPR targeted site in the carT gene and resulting mutations . ( B ) Primary structure and predicted transmembrane domains ( yellow ) of CarT . Red arrows indicate location of premature stops within mutant alleles . Gray and black letters indicate conserved and highly conserved amino acids among SLC22 transporters ( Eraly et al . , 2004; Koepsell , 2013 ) , respectively . The SLC22 family is characterized by a large extracellular loop containing four cysteines ( C ) and a glycosylation site ( blue ) at conserved positions . ( C ) ERGs recorded from female flies carrying wild type ( + ) or the indicated CarT alleles ( 16A , 16B , 43 , or tdTomato-CarT ) with or without expression of a wild-type or a Myc-tagged UAS-CarT transgene driven by longGMR-Gal4 or repo-Gal4 as indicated . ( D ) Quantifications of ON and OFF transients , and SNPs of all genotypes in panel C averaged from three replicate experiments , including at least 45 traces from 15 flies ( G: longGMR-Gal4; R: repo-Gal4; -: not present ) . ( E ) Phototactic behavior of OreR and carT43 mutant flies compared with other mutants that disrupt the histamine–carcinine cycle ( HdcMB07212 , ebony1 , or tan1 ) presented as a light preference index . ( F ) Phototactic behavior of flies expressing a UAS-CarT transgene in photoreceptor neurons under control of the longGMR-Gal4 driver in carT43 mutants compared with controls shows the restoration of wild type behavior . For each graph , the box outlines the upper and lower quartiles , and the whiskers show minimum and maximum recorded values . ns , not significant; ****p < 0 . 0001 compared with carT43 ( D ) or OreR ( E and F ) . CRISPR , clustered regularly-interspaced short palindromic repeats; ERG , electroretinograms; SNPs , sustained negative potentials . DOI: http://dx . doi . org/10 . 7554/eLife . 10972 . 00510 . 7554/eLife . 10972 . 006Figure 2—figure supplement 1 . Phylogeny of the Drosophila SLC22 transporters . Drosophila and bee SLC22 transporters were identified using the Basic Local Alignement Search Tool Protein ( BLASTp ) provided by the National Center for Biotechnology Information ( NCBI ) based on similarity to known mammalian and insect family members ( Koepsell , 2013; Zhu et al . , 2015 ) . A phylogeny tree was obtained using Clustal W2 and visualized using iToL ( Letunic and Bork , 2011 ) . Numbers represent simulated branch lengths . DOI: http://dx . doi . org/10 . 7554/eLife . 10972 . 006 To examine whether the disruption of visual neurotransmission in carT mutants alters visual behavior , we used a phototaxis assay ( Benzer , 1967 ) . Wild-type flies were naturally phototactic ( Figure 2E ) . However , flies with defects in the synthesis of histamine ( HdcMB07212 ) or its recycling ( ebony1 or tan1 ) were randomly distributed between the lit and dark arms of a T-maze , consistent with a lack of phototactic preference due to their inability to detect light ( Figure 2E ) . Similarly , phototactic behavior was severely reduced in carT43 flies ( Figure 2E ) . Light preference of carT43 flies was restored by expression of wild-type CarT in photoreceptor neurons under control of longGMR-Gal4 ( Figure 2F ) . Together , these results indicate that CarT expression in photoreceptor neurons is necessary and sufficient for behavioral aspects of normal vision and electrophysiological properties of visual transduction . Interference with aspects of the histamine–carcinine cycle by different mutations causes distinct changes in the distribution of these two metabolites in the retina and lamina , although overall architecture of these tissues is not altered in carT , ebony , tan and Hdc mutants ( Figure 3A–D ) and references ( Chaturvedi et al . , 2014; Borycz et al . , 2008 ) . Consistent with their function in the histamine–carcinine cycle , characteristic increases in the levels of histamine in ebony1 ( Figure 3C ) and carcinine in tan1 heads ( Figure 3D ) were detected by antibody staining ( Chaturvedi et al . , 2014 ) . When compared with wild type , carT43 heads displayed no discernable alterations in histamine staining intensity or distribution ( Figure 3C ) . However , carcinine levels were elevated in carT43 lamina ( Figure 3D , arrow ) and medulla ( arrowhead ) . Staining for carcinine in carT43 lamina partially localized to the glia visualized by repo-Gal4 driven UAS-mCD8::RFP ( Figure 3E ) suggested a build up of carcinine in the glia and the extracellular space in the lamina , consistent with a requirement of CarT in carcinine transport into photoreceptors . Notably , carcinine accumulation in carT43 brains was restored to wild-type levels by photoreceptor-specific expression of a Myc-CarT transgene ( Figure 3D ) . 10 . 7554/eLife . 10972 . 007Figure 3 . Loss of CarT in photoreceptors increases laminal carcinine . ( A ) Confocal sections of photoreceptor axonal endings within the lamina stained for the synapse marker Bruchpilot ( Brp ) and the glia-specific Ebony . ( B ) Micrographs of cryo-sections from control ( OreR ) and flies expressing the UAS-Myc-CarT transgene driven by longGMR-Gal4 in the carT43 background . Retina ( R ) , lamina ( L ) , and medulla ( M ) neuropiles are stained for DNA ( magenta ) and Myc-tagged CarT ( green ) . Arrow points to Myc staining in photoreceptor axonal endings and arrowhead to distal retina . ( C , D ) Micrographs of cryo-sections from control ( OreR ) and mutant flies affecting the histamine–carcinine cycle: carT43 , carT43;GMR-Gal4/UAS-Myc-CarT , ebony1 , tan1 , and HdcMB07212 stained for histamine ( C ) or carcinine ( D ) . Arrow and arrowhead in D point to carcinine accumulations in the lamina and medulla , respectively . ( E ) Micrographs from control ( carT43/+ ) or carT43 flies expressing a UAS-mCD8::RFP transgene driven by repo-Gal4 and stained for carcinine . ( F ) Micrographs showing tdTomato fluorescence of the in-frame tdTomato-CarT allele compared with wild type control . DNA staining is shown in green . Arrowheads point to tdTomato signal at R7 and R8 photoreceptor terminals within the medulla . ( G ) Confocal section of laminal region of a tdTomato-CarT fly stained for glia-specific Ebony . ( H ) Confocal sections of flies heterozygous for tdTomato-CarT ( red ) and photoreceptor-specific 3xPax3-eGFP ( blue ) stained for Brp ( green ) . Sections are parallel to and across photoreceptor axons , respectively . Scale bars are 20 µm in A , E , G and H and 50 µm in B–D and F . DOI: http://dx . doi . org/10 . 7554/eLife . 10972 . 007 Within photoreceptors , the longGMR-driven Myc-CarT transporter was enriched in their axonal endings in the lamina ( arrow in Figure 3B ) and the distal retina ( arrowhead in Figure 3B ) indicating CarT may function near synaptic terminals and in neuronal cell bodies to facilitate perisynaptic and long-distance recycling , two pathways previously suggested ( Chaturvedi et al . , 2014; Borycz et al . , 2012; Rahman et al . , 2012 ) . Endogenous CarT RNA expression is highly enriched in the eye ( http://flybase . org/reports/FBlc0000157 . html ) . To further examine endogenous CarT expression , we utilized CRISPR-induced deletions coupled with homology-directed repair to incorporate an in-frame insertion of tdTomato N-terminally to the CarT coding sequence ( Figure 2B ) . ERG analysis of flies harboring one copy of this tdTomato-CarT over the nonfunctional carT43 allele displayed normal ON and OFF transient responses , indicating that the endogenous tdTomato-tagged CarT was fully functional ( Figure 2C , D ) . This endogenous tdTomato-CarT was highly enriched within photoreceptor projections in the lamina and medulla ( Figure 3F ) . Co-labeling with the glial Ebony ( Figure 3G ) , or synaptic Bruchpilot and a 3xPax3-eGFP , which labels photoreceptor neurons ( Figure 3H ) , indicates that CarT is present in photoreceptors near synaptic terminals ( Figure 3H ) . To determine whether the CarT transporter utilizes carcinine as a substrate , we developed an immunofluorescence-based carcinine transport assay . Full-length or internally truncated Myc-CarT were expressed in S2 cells that were probed by antibody staining for carcinine uptake . When cultured in standard Schneider’s medium , transfected S2 cells displayed no appreciable carcinine staining , indicating a lack of endogenous carcinine and sufficient specificity of the carcinine antibody in this context ( Figure 4A and A’; quantified in D ) . When culture media was supplemented with carcinine ( 0 . 5 mM ) , cells expressing full-length Myc-CarT stained positively for carcinine , indicating its uptake in these cells . Importantly , neighboring untransfected cells ( negative for the Myc epitope ) lacked carcinine staining , despite being maintained in the same carcinine-supplemented medium as Myc-CarT positive cells ( Figure 4B and B’ ) . Furthermore , an internal deletion of 10 of the 12 transmembrane domains within CarT abolished carcinine uptake ( Figure 4C and C’; quantified in D ) . Taken together , these data demonstrate CarT functions as a carcinine transporter . 10 . 7554/eLife . 10972 . 008Figure 4 . CarT is a carcinine transporter . ( A ) Micrographs of S2 cells transfected with the Myc-CarT transgene ( A , A’ and B , B’ ) or the Myc-CarTdel transgene deleted for 10 internal transmembrane domains ( C , C’ ) cultured in media lacking carcinine ( A , A’ ) or supplemented with 0 . 5 mM carcinine ( B , B’ and C , C’ ) and stained for DNA and the Myc epitope ( A–C ) or carcinine ( A’–C’ ) . ( D ) Quantification of carcinine signals normalized to signals for the Myc epitope . ****p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 10972 . 00810 . 7554/eLife . 10972 . 009Figure 4—figure supplement 1 . Diagram summarizing the histamine–carcinine cycle and the proposed role of the CarT transporter in photoreceptor cells . DOI: http://dx . doi . org/10 . 7554/eLife . 10972 . 009 Sequence comparison indicates that CarT is a member of the SLC22 family of transporters ( Eraly et al . , 2004 ) and together with other CarT-like transporters in invertebrates constitutes a distinct subfamily ( Figure 2—figure supplement 1 ) . Interestingly , OCT2 and OCT3 , members of the closely related organic cation subfamily of SLC22A transporters , have been implicated in histamine uptake into human astrocytes ( Yoshikawa et al . , 2013 ) and the transport of histamine receptor antagonists ( Koepsell , 2013 ) . Many of the biochemically characterized members of the extended SLC22 family that are involved in the transport of neurotransmitters , exhibit overlapping substrate specificity providing transport redundancies ( Koepsell , 2013 ) . However , our data argue against such redundancy as carT null alleles efficiently blocks synaptic transmission dependent on the histamine–carcinine cycle ( Figure 4—figure supplement 1 ) . A role in this process has previously been suggested for the putative Na/Cl– dependent neurotransmitter/osmolyte transporter Inebriated ( Stuart et al . , 2007 ) , possibly by directly transporting carcinine into photoreceptors ( Gavin et al . , 2007 ) . No direct evidence for such a function has been reported , however , suggesting that Inebriated may indirectly support the histamine–carcinine cycle by promoting the long-distance recycling of β-alanine from photoreceptors to glia ( Stuart et al . , 2007; Chaturvedi et al . , 2014; Borycz et al . , 2012 ) . An indirect role of Inebriated is also more consistent with its role in water homeostasis in the hindgut ( Luan et al . , 2015 ) . Here , we presented several lines of evidence that CG9317 , which is highly expressed in heads , but not bodies ( Eraly et al . , 2004 ) , encodes the carcinine transporter CarT . Heterologous expression of CarT in cultured cells was sufficient to facilitate carcinine uptake . This biochemical activity and the accumulation of carcinine in CarT lamina could be consistent with CarT facilitating export of carcinine from glia , or its import into photoreceptors . Strong support for the second possibility is provided by genetic experiments that reveal a requirement of CarT in photoreceptors , but not glia and the accumulation of carcinine in the lamina . CarT expression in photoreceptors is necessary and sufficient to sustain visual transduction , as indicated by ON and OFF transients as signatures of synaptic activity and by visual behavior . Together , these findings support a role of CarT as the transmembrane transporter responsible for the uptake of carcinine into photoreceptor neurons , a critical step in the histamine–carcinine cycle . Histamine , similar to its impact on multiple behaviors in the mammalian brain ( Nall and Sehgal , 2014; Panula and Nuutinen , 2013; Shan et al . , 2015; Castellan Baldan et al . , 2014 ) , has been implicated in several circuits in the insect brain as well , including those controlling sleep , circadian rhythms and thermosensation ( Nall and Sehgal , 2014; Buchner et al . , 1993; Oh et al . , 2013; Hong et al . , 2006; Suh and Jackson , 2007 ) . The identification of CarT’s role in the histamine–carcinine cycle will provide additional tools to further address the role of histaminergic circuits in these different behaviors . ERGs were recorded as previously described ( Williamson et al . , 2010 ) . In brief , voltage measurements of immobilized female flies were recorded with electrodes containing 2M NaCl placed on the corneal surface and inserted into the thorax . Measurements were filtered through an electrometer ( IE-210; Warner Instruments , Hamden , CT ) , digitized with a Digidata 1440A and MiniDigi 1B system , and recorded using Clampex 10 . 2 and quantified with Clampfit software ( Molecular Devices , Sunnyvale , CA ) . Light pulses ( 1 s ) were computer controlled ( MC1500; Schott , Mainz , Germany ) . For light-choice assays ( Benzer , 1967 ) , male flies were collected within 12 hr of eclosion and aged for 3–4 d . Flies were anesthetized briefly with CO2 , transferred to empty culture tubes , and left to recover for 1 hr . For testing , a group of 20 flies was introduced into a T-maze apparatus and allowed to distribute for 20 s between a dark tube and a tube exposed to incandescent light . Each group was tested in triplicate . A light preference index was calculated using the equation , PI = ( #Light- #Dark ) / #total . For all genotypes , at least 100 flies were tested . Fly heads were dissected in HL3 hemolymph-like solution ( Stewart et al . , 1994 ) to remove the proboscis and posterior cuticle , fixed for 4 hr in ice cold 4% 1-ethyl-3- ( -3-dimethylaminopropyl ) carbodiimide ( wt/vol , Sigma , St Louis , MO ) in 0 . 1 M phosphate buffer solution , washed overnight in 25% ( wt/vol ) sucrose in phosphate buffer ( pH 7 . 4 ) , embedded in optimal cutting temperature compound , frozen in dry ice , and sectioned at 20 μm thickness on a cryostat microtome ( Hacker-Bright , Winnsboro , SC ) . Sections were incubated overnight with antibodies to histamine ( 1:1000 , Sigma , St Louis , MO , cat# H7403 preabsorbed with 200 µM carcinine ) or carcinine ( 1:1000 , Immunostar , Hudson , WI , cat# 22939 preabsorbed with 200 µM histamine [Chaturvedi et al . , 2014] ) . Other antibodies used include anti-Ebony ( gift from Bernhard Hovemann [Richardt et al . , 2002] ) , anti-Bruchpilot ( nc82 , Hybridoma Bank , Iowa City ) , anti-Myc ( 9E10 , BAbCO , Richmond , CA ) , and anti-RFP ( Rockland , Limerick , PA , cat# 600-401-379 ) . Secondary antibodies were labeled with Alexa488 ( 1:500 , Molecular Probes , Pittsburg , PA , cat# A-11008 ) , Alexa568 ( 1:500 , Molecular Probes , cat# A-11011 ) , or Alexa647 ( 1:250 , Molecular Probes , cat#A-21235 ) . Where indicated , Topro-3 Iodide ( Molecular Probes , T3605 ) was used to stain DNA . Images were captured with 20× NA 0 . 75 or 63× NA 1 . 4 lenses on an inverted confocal microscope ( LSM510 Meta; Carl Zeiss , Oberkochen , Germany ) at 21°C–23°C . Because radioactive carcinine is not commercially available , we developed an immunofluorescence-based transport assay . Drosophila S2 cells were cultured using standard methods and transfected with plasmids containing Myc-tagged constructs as noted using the TransIT-2020 manufacturer’s protocol ( Mirus , Madison , WI ) . Transfected cells were incubated with 0 . 5 mM carcinine for 24 hr prior to a 1 hr fixation in ice cold 4% 1-ethyl-3- ( -3-dimethylaminopropyl ) carbodiimide ( wt/vol , Sigma ) in 0 . 1 M phosphate buffer solution . Cells were then permeabilized in phosphate-buffered saline containing 0 . 3% Saponin , blocked in 5% normal goat serum , and incubated overnight with carcinine and Myc antibodies . Quantification was performed in ImageJ by normalizing the integrated density of the carcinine signal by that of the Myc signal . CarT was identified as a SLC22 family member using BLASTpsi alignments against the NCBI non-redundant protein sequence database . Drosophila SLC22-related family members were identified from existing annotations on FlyBase and by running a BLASTp search with the CarT Isoform C peptide sequence against only Drosophila melanogaster ( taxid:7227 ) . Searches for individual organisms were performed using the following taxa identifiers on NCBI BLASTp suite: Homo sapiens ( taxid:9606 ) , Mus musculus ( taxid:10090 ) , Apis mellifera ( taxid:7460 ) . Additionally , previously identified mammalian SLC22 transporters were included ( Koepsell , 2013; Zhu et al . , 2015; Martin and Krantz , 2014 ) . For preparation of multiple alignments , CLUSTAL Omega was used ( http://www . ebi . ac . uk/Tools/msa/clustalo/ ) with default parameters ( Sievers et al . , 2011 ) . Multiple sequence alignments were entered into Clustal W2 and phylogenies were generated using the Neighbor-joining clustering method with default parameters ( Larkin et al . , 2007 ) . The resulting phylogenies were analyzed using Interactive Tree of Life ( iToL; http://itol . embl . de ) , an online phylogeny tree viewer and editor ( Letunic and Bork , 2011 ) . Statistical significance was determined using one-way analysis of variance ( ANOVA ) followed by Tukey’s or Bonferroni’s multiple comparisons using GraphPad Prism 6 . | Photoreceptors are light-sensitive neurons in the eyes of the fruit fly Drosophila that form connections with other neurons in the fly’s brain . At these connections , which are called synapses , the photoreceptors continuously release a chemical called histamine . Photoreceptors will release more or less histamine depending on changes in light intensity , but always tend to release more histamine than they can produce themselves from scratch . This means that the visual system in Drosophila relies on a pathway that recycles histamine . That is to say , glial cells ( which support the activity of the neurons ) remove the chemical from synapses and return it to the photoreceptor neurons in a slightly modified form called “carcinine” . The photoreceptors then quickly convert the chemical back into histamine , ready to be released . Stenesen et al . set out to identify the proteins that support this recycling pathway , and started by screening around 130 genes that encode transporter proteins for potential roles in histamine recycling . This screen identified a gene encoding a protein that was named CarT . This protein transports carcinine , the modified version of the histamine neurotransmitter . Stenesen et al . show that the photoreceptor neurons make the CarT protein and need this protein to take up the carcinine released by the supporting glial cells . Without CarT , photoreceptor neurons cannot transmit visual information , and so mutant flies in which the gene for CarT is deleted are blind . Follow-up studies related to this work could involve identifying the transporters that move histamine and carcinine in and out of the glia cells , and exploring what other neurons and behaviors in fruit flies rely on CarT’s activity . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion"
] | [
"short",
"report",
"neuroscience"
] | 2015 | The carcinine transporter CarT is required in Drosophila photoreceptor neurons to sustain histamine recycling |
Loss of Foxc1 is associated with Dandy-Walker malformation , the most common human cerebellar malformation characterized by cerebellar hypoplasia and an enlarged posterior fossa and fourth ventricle . Although expressed in the mouse posterior fossa mesenchyme , loss of Foxc1 non-autonomously induces a rapid and devastating decrease in embryonic cerebellar ventricular zone radial glial proliferation and concurrent increase in cerebellar neuronal differentiation . Subsequent migration of cerebellar neurons is disrupted , associated with disordered radial glial morphology . In vitro , SDF1α , a direct Foxc1 target also expressed in the head mesenchyme , acts as a cerebellar radial glial mitogen and a chemoattractant for nascent Purkinje cells . Its receptor , Cxcr4 , is expressed in cerebellar radial glial cells and conditional Cxcr4 ablation with Nes-Cre mimics the Foxc1−/− cerebellar phenotype . SDF1α also rescues the Foxc1−/− phenotype . Our data emphasizes that the head mesenchyme exerts a considerable influence on early embryonic brain development and its disruption contributes to neurodevelopmental disorders in humans .
Dandy-Walker malformation , the most common congenital human cerebellar malformation , is defined by cerebellar vermis hypoplasia , an enlarged fourth ventricle and an enlarged posterior fossa ( Parisi and Dobyns , 2003 ) . Heterozygous loss of the Foxc1 gene contributes to DWM . Although loss of this transcription factor causes significant developmental cerebellar pathology in humans and mice , in mice , its expression is limited to vascular pericytes within the developing cerebellum . In contrast , Foxc1 is widely expressed in the posterior fossa mesenchyme surrounding the cerebellar anlage beginning after e12 . 5 in mice . We therefore hypothesized that disrupted Foxc1-dependent signalling from the posterior fossa head mesenchyme to the adjacent developing cerebellum is key to the DWM phenotype . We previously reported reduced mesenchymal expression of several secreted factors in the mouse Foxc1 e12 . 5 mutant posterior fossa including Bmp 2 , 4 and SDF1α ( Aldinger et al . , 2009 ) . However , Foxc1 mutant cerebellar developmental abnormalities were not fully investigated and the relevance of these mesenchymal expression changes remained unexplored . The cerebellar anlage derived from dorsal rhombomere 1 is initially established by Fgf8 and other signalling molecules originating from the isthmic organizer , a transient embryonic mid/hindbrain junction at neural tube closure ( Nakamura et al . , 2005; Basson and Wingate , 2013 ) . Initial neurogenesis in the cerebellar anlage occurs in two distinct germinal zones . The cerebellar ventricular zone gives rise to waves of GABAergic neurons including cerebellar Purkinje cells which migrate radially away from the ventricular zone into the developing cerebellar cortex ( Seto et al . , 2014 ) . The cerebellar rhombic lip gives rise to cerebellar glutamatergic neurons including cerebellar granule neuron progenitors ( GNPs ) , which migrate tangentially over the developing anlage to form the external granule layer ( EGL ) . From the EGL , newly born granule cells migrate inward radially to form the mature internal granule layer ( Millen and Gleeson , 2008 ) . Extrinsic signalling has previously been implicated in some aspects of mouse cerebellar development . Transventricular Shh from the choroid plexus regulates cerebellar ventricular zone proliferation from e14 . 5 ( Huang et al . , 2010 ) . Mesenchymal Bmp signals induce the cerebellar rhombic lip around e10 . 5 in mice ( Alder et al . , 1996 , 1999; Fernandes et al . , 2012; Tong and Kwan , 2013 ) . Additionally , meningeal SDF1α has a chemoattractive role in both the tangential migration of GNPs away from the rhombic lip to form the EGL from e12 . 5 , as well as a later role in maintaining their proliferative niche adjacent at the pial surface within the EGL ( Hartmann et al . , 1998; Ma et al . , 1998; Zou et al . , 1998; Klein et al . , 2001; Reiss et al . , 2002; Zhu et al . , 2002; Vilz et al . , 2005; Tiveron and Cremer , 2008; Yu et al . , 2010 ) . To investigate which events of cerebellar development are disrupted by loss of Foxc1–dependent secreted factors , we first undertook an extensive phenotypic analysis of Foxc1−/− embryonic cerebellar development . This analysis uncovered a dramatic deficit in cerebellar ventricular zone radial glial cell proliferation and differentiation during early cerebellar anlage development . Using a combination of in vitro and in vivo assays , we next determined that SDF1α , a direct transcriptional target of Foxc1 ( Zarbalis et al . , 2012 ) is a major effector of Foxc1 cerebellar pathogenesis . SDF1α secreted by the posterior fossa mesenchyme acts through Cxcr4 in cerebellar radial glial cells to drive their proliferation and hence embryonic cerebellar anlage growth . We also show that SDF1α−Cxcr4 signalling is required to maintain the radial glial scaffold for ventricular zone derived neuronal migration . These studies dramatically expand the known roles of SDF1α-Cxcr4 signalling to encompass almost every major developmental program of cerebellar development .
Foxc1 expression is initiated in the head mesenchyme adjacent to the developing cerebellum at e12 . 5 . Although the Foxc1−/− mutant cerebellum is not dramatically affected at e12 . 5 , by e18 . 5 the mutant cerebellum is highly dysmorphic ( Aldinger et al . , 2009 ) . To better define the effect of Foxc1 deletion on pre-natal cerebellar development , we assessed cerebellar morphology at e13 . 5 , e15 . 5 and e17 . 5 ( Figure 1A–H ) . Cresyl violet staining confirmed dramatic rhombic lip and EGL abnormalities ( Aldinger et al . , 2009 ) at all stages , however we also noted extensive additional abnormalities of the cerebellar ventricular zone and its derivatives in mutant animals as early at e13 . 5 , when cerebellar anlage size and shape differences were readily apparent ( Figure 1D ) . Unlike the forebrain mesenchyme , the posterior fossa mesenchyme in the Foxc1−/− mutant was structurally normal , as indicated by Laminin , Pdgfr1 and Raldh2 staining ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 03962 . 003Figure 1 . Foxc1 deletion leads to reduced proliferation and increased differentiation in the cerebellar ventricular zone . ( A ) Schematic of a dorsal whole mount view of the embryonic mouse brain and ( B ) Sagittal section of the mid-hindbrain region . Abbreviations include the four axes; A—Anterior , P—Posterior , D—Dorsal , and V—Ventral; CB—Cerebellum and EGL—External Granule Layer . ( C–H ) Sagittal sections of e13 . 5 ( C , D ) , e15 . 5 ( E , F ) and e17 . 5 ( G , H ) cerebellum from WT ( C , E , G ) and Foxc1−/− ( D , F , H ) mice . ( I–L ) Foxc1−/− mice showed a reduction in proliferation at e13 . 5 ( J ) and e15 . 5 ( L ) , compared to WT ( I , K ) . ( M ) Graph showing the percentage of BrdU positive cells in the cerebellar ventricular zone . The percentage of BrdU positive cells in the ventricular zone of the Foxc1−/− cerebellum was significantly lower than WT at e13 . 5 and e15 . 5 . Data represented as mean percentage of BrdU positive cells ± s . e . m . *** indicates significance with respect to the corresponding WT Control ( p < 0 . 05 ) . ( N–Q′ ) β-III Tubulin staining of the e13 . 5 ( N–O ) and e15 . 5 ( P–Q′ ) cerebellum showed by increased differentiation in the ventricular zone of Foxc1 mutants at both e13 . 5 ( O ) and e15 . 5 ( Q , Q′ ) compared to WT ( N , P–P′ ) . Abbreviations used; Mes—Mesenchyme , VZ—Ventricular Zone . Scale bar = 100 µm for all images , except P′ and Q′ where the scale bar = 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03962 . 00310 . 7554/eLife . 03962 . 004Figure 1—figure supplement 1 . Posterior fossa meningeal formation is not compromised in the Foxc1−/− mutants . ( A–F ) Sagittal sections of e13 . 5 ( A–D ) , e15 . 5 ( E–J ) cerebellum from WT ( A , C , E , G , I ) and Foxc1−/− ( B , D , F , H , J ) mice stained for Laminin ( A–B; E–F ) , Pdgfr1 ( C–D; G–H ) , and Raldh2 ( E–F; I–J ) , clearly indicate that posterior fossa mesenchymal formation in Foxc1−/− mutants is not compromised . Abbreviations used include Mes—Mesenchyme . Scale bar = 100 µm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 03962 . 00410 . 7554/eLife . 03962 . 005Figure 1—figure supplement 2 . Ki67 expression in the Foxc1−/− and Cxcr4 conditional knockout cerebellum . ( A–F ) Sagittal sections of e13 . 5 ( A–C ) , e15 . 5 ( D–F ) cerebellum from WT ( A , D ) , Foxc1−/− ( B , E ) and Cxcr4 CKO ( C , F ) mice . Foxc1−/− and Cxcr4 CKO mice showed a reduction in VZ proliferation as evinced by Ki67 staining . Abbreviations used include VZ—Ventricular Zone . Scale bar = 100 µm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 03962 . 005 Since cerebellar anlage size during early development is directly related to cerebellar ventricular zone proliferation , we assessed proliferation by sacrificing animals immediately following a 1 hr BrdU pulse at e13 . 5 and e15 . 5 . The VZ was identified as a layer of cells directly lining the fourth ventricle . Ki67 staining confirmed the boundary of the VZ ( Figure 1—figure supplement 2 ) . In wild-type animals , proliferation is normally downregulated across these time points as the ventricular zone becomes depleted ( Sudarov et al . , 2011 ) ( Compare Figure 1I to Figure 1K ) . Dramatically , at e13 . 5 , Foxc1−/− mutants demonstrated a twofold reduction of ventricular zone proliferation , compared to e13 . 5 wild-type animals . At e15 . 5 , ventricular zone proliferation was also significantly lower in Foxc1−/− mutants compared to control animals ( Figure 1I–M ) . Reduced proliferation in Foxc1−/− mutants was accompanied by a concomitant substantial increase in differentiation as visualized by β-III Tubulin expression , an early marker of differentiating neurons ( Brazelton et al . , 2000 ) . This was readily apparent in Foxc1−/− mutants at e13 . 5 with extensive differentiating neurons distributed throughout the mutant anlage at this early stage ( Figure 1O ) . In contrast , wild-type animals at a comparable stage exhibited β-III Tubulin antibody staining almost entirely restricted to the nuclear transitory zone containing differentiating RL derived neurons ( Figure 1N , NTZ ) . In e15 . 5 wild-type animals , the β-III Tubulin positive cells were widely distributed within the developing anlage , however a β-III Tubulin negative region consisting of two to three layers of cells defined the ventricular zone at this stage ( Figure 1P , P′ ) . In striking contrast , the mutant ventricular zone was occupied by numerous β-III Tubulin–positive cells in e15 . 5 Foxc1−/− mutant animals ( Figure 1Q , Q′ ) . Together , these data demonstrate that loss of Foxc1 does not just disrupt the cerebellar rhombic lip which is adjacent to the head mesenchyme where Foxc1 is normally expressed . Rather , loss of Foxc1 profoundly disrupts all early cerebellar progenitor zones . Cerebellar Purkinje cells are generated between e10 . 5 and e13 . 5 ( Sudarov et al . , 2011 ) and migrate out of the ventricular zone beginning at e14 . 5 . At e15 . 5 in wild-type animals , we observed Purkinje cells defined by Calbindin-expression in a domain overlying the ventricular zone ( Figure 2A , A′ , asterisk ) . By e19 . 5 these wild-type cells had migrated outwards to form a multilayered Purkinje cell plate underneath the external granule layer in the developing cerebellar cortex ( Figure 2C , C′ , asterisk ) . As expected at e15 . 5 in Foxc1−/− mutants , fewer Purkinje cells were present since proliferation of the ventricular zone was compromised during the peak of Purkinje cell production . Purkinje cell migration however was also affected . While some cells remained in the ventricular zone at this stage , most remaining calbindin expressing cells were scattered throughout the anlage ( Figure 2B , B′ , asterisk ) . By e19 . 5 the majority of the mutant Purkinje cells remained within the core of the cerebellar anlage with just a few present in a residual dysmorphic Purkinje cell plate located in the anterior cerebellum ( Figure 2D , D′ , asterisk ) . 10 . 7554/eLife . 03962 . 006Figure 2 . Abnormally positioned Purkinje cells are present in the cerebellum of Foxc1−/− mice . ( A–D ) Sagittal sections of the WT ( A , A′; C , C′ ) and Foxc1−/− cerebellum ( B , B′ , D , D′ ) stained for Calbindin . In the e15 . 5 WT cerebellum ( A–A′ , asterisk ) , Purkinje cells were present in a band overlying the ventricular zone . However , in the Foxc1 null mutant , ( B–B′ , asterisk ) fewer ectopically located PCs were present . By e19 . 5 , PCs in the WT ( C–C′ ) formed a layer of cells directly underneath the ML in the cerebellar cortex ( C′ , asterisk ) . However , in the Foxc1 null mutant ( D , D′ ) , the PCs were arranged as clusters within the anlage ( D′ , asterisk ) . Abbreviations used; PL—Purkinje Layer and VZ—Ventricular Zone . The white dashed line indicates the outer boundary of the EGL and the mesenchyme , while the yellow dotted line represents the cerebellar ventricular surface . The red and yellow asterisks represent the PL . A′–D′ are grey scale images of A-D respectively . Scale bar = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03962 . 006 Since we observed dramatic Purkinje cell migration defects , we next examined the status of radial glial cells ( Figure 3A–H ) and Bergmann glia ( Figure 3J–M ) . RG cells serve as both neuronal progenitors ( Feng et al . , 1994 ) and a scaffold for Purkinje cell neurons as they migrate towards the cerebellar cortex ( Yuasa et al . , 1996; Hatten , 1999 ) while BG serve as a scaffold for migrating granule neurons from the EGL . In wild-type animals at e13 . 5 and e15 . 5 , the nestin-positive soma of RG were located in the ventricular zone with radial fibers extending to the NTZ or pial surface ( Figure 3A , arrows ) . By e15 . 5 , plentiful nestin-positive radial glial fibers extending from the ventricular surface to the pial surface were apparent in the wild-type anterior cerebellar anlage ( Figure 3C–C′ , arrows ) . However , by e17 . 5 , radial glial cells in the posterior wild-type anlage transitioned into Bergmann Glial cells with radial fibers spanning just the presumptive molecular layer in the developing cerebellar cortex ( Li et al . , 2014 ) ( Figure 3E , G; arrows ) . In Foxc1−/− mutants at e13 . 5 , nestin-positive staining was evident; however extended radial fibers were absent ( Figure 3B ) . By e15 . 5 nestin-positive fibers were discontinuous with very few extending to the pial surface ( Figure 3D , D′ , arrows ) . A similar absence of long radial fibers was evident in the e17 . 5 mutant ( Figure 3F , arrows ) . Additionally , Bergmann glial morphology and arrangement were found to be disrupted in the Foxc1−/− mutant ( Figure 3F–H , J–M , Figure 3K , M; arrows ) . In the WT , BG were found to overlap with the PL ( Figure 3E , G , J , L ) , while in the Foxc1−/− mutant cells were found to be present ectopically in the EGL and ML indicating abnormal migration ( Figure 3K , M ) . These changes were accompanied by increased cell death in the ventricular zone of Foxc1−/− mutants as measured by TUNEL labeling ( Figure 3I ) . Together , our results suggest that the disruption of radial glial migratory scaffold in Foxc1−/− mutants contributes to aberrant PC migratory phenotypes in Foxc1−/− mutant mice , which in turn also contribute to an abnormal BG phenotype . 10 . 7554/eLife . 03962 . 007Figure 3 . Radial glial and Bergmann glial morphology is severely disrupted in Foxc1−/− mice . ( A–H ) Sagittal sections of the embryonic mouse cerebellum in WT ( A , C , C′ , E , G ) and Foxc1 null mutants ( B , D , D′ , F , H ) at e13 . 5 ( A , B ) , e15 . 5 ( C–D′ ) , e17 . 5 ( E , F ) and 19 . 5 ( G , H ) stained for Nestin . While radial glial fibers extended from the ventricular zone to the pial surface in the WT cerebellum ( C , C′ , arrows ) , in the Foxc1 null mutant , fibers were discontinuous and did not extend all the way to the pial surface ( D , D′ , arrows ) . In the WT cerebellum , Bergmann glial fibers extended from the EGL to the PL at e17 . 5 ( E ) and e19 . 5 ( G , L; arrows ) . The white straight line in ( E ) demarcates the anlage into two regions—an anterior ( left ) region where fibers extend from the VZ to the pia , and the other ( right ) where Bergmann glial fibers extend from the EGL to the IGL . In the Foxc1 null mutant , these two zones were not apparent and fiber morphology was severely disrupted ( F , H; arrows ) . ( I ) Graph showing the number of TUNEL positive cells that span the length of the cerebellar ventricular zone . The number of TUNEL positive cells in the ventricular zone of the Foxc1−/− cerebellum was significantly higher than WT at e13 . 5 and e15 . 5 . Data represented as average number of TUNEL positive cells per section analysed ± s . e . m . *** indicates significance with respect to corresponding WT Control ( p < 0 . 05 ) . ( J–M ) Sagittal sections of the WT ( J , L ) and Foxc1−/− cerebellum ( K , M ) stained for BLBP . In the e19 . 5 WT cerebellum ( J , box; L , arrow ) , BG were present as a layer overlapping with the PL . However , in the Foxc1 null mutant , ( K , box; M , arrow ) BG were ectopically located in the EGL . Abbreviations used; NTZ—Nuclear Transitory Zone and VZ—Ventricular Zone . The white dashed line indicates the outer boundary of the EGL and the mesenchyme , while the yellow dotted line represents the cerebellar ventricular surface . The white straight line in Image E represents the boundary between two zones—one where fibers extend from the VZ to the pia , and the other where BG fibers extend from the EGL to the IGL . C′–D′ are high magnification images of C–D respectively . Scale bar = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03962 . 007 We hypothesized that Foxc1 regulates the expression of secreted factors in the head mesenchyme which then non-autonomously influence the developing cerebellum . Based on candidate gene analysis , we previously reported that Foxc1−/− mutant posterior fossa mesenchyme has reduced expression of SDF1α ( Cxcl12 ) , Bmp2 and 4 ( Aldinger et al . , 2009 ) . We tested the ability of these secreted factors to influence cerebellar ventricular zone proliferation and migration of ventricular zone-derived neurons using various assays ( Figure 4 ) . The embryonic cerebellar ventricular zone does not express Bmp-receptors although they are expressed in the rhombic lip ( Machold et al . , 2007 ) . In contrast , the SDF1α receptor , Cxcr4 is highly expressed in the embryonic ventricular zone ( Figure 5 and data not shown ) . Addition of SDF1α significantly increased cell division in primary dissociated cultures of the wild-type e13 . 5 cerebellar anlage . Blocking SDF1α function using the Cxcr4 receptor antagonist AMD 3100 ( Rosenkilde et al . , 2004 ) , significantly inhibited BrdU uptake in these cultures ( Figure 4B ) . The majority of cerebellar progenitors at e13 . 5 derive from the cerebellar ventricular zone , although both cerebellar rhombic lip progenitors and a small number of granule cell progenitors are also present at this time . To determine if the anti-proliferative AMD3100 could specifically alter cerebellar ventricular zone proliferation , we cultured whole e13 . 5 cerebellar anlage explants in serum containing media with or without AMD 3100 for 24 hr . Subsequent sections of these whole mount cerebellar cultures revealed that blockage of the SDF1α-Cxcr4 signalling pathway caused a significant increase in ventricular zone neuronal differentiation and reduced proliferation . ( Figure 4A , C–F ) . 10 . 7554/eLife . 03962 . 008Figure 4 . SDF1α induces cell division in cerebellar ventricular zone progenitors and also functions as a chemoattractant . ( A ) Schematic of dorsal whole mount view of the embryonic mouse head and a brief description of experiments included in the figure . ( B ) Graph showing the percentage of BrdU+ cells in primary dissociated cerebellar culture . Addition of SDF1α significantly increased BrdU uptake by cells . Addition of AMD3100 , an antagonist of SDF1α significantly reduced BrdU incorporation . ( C , D; C′ , D′ ) Sagittal sections of e13 . 5 cerebellum cultured whole mount for 1 day and stained for Ki67 ( inset ) and β-III Tubulin . Addition of AMD3100 to the culture significantly reduced cell proliferation and increased differentiation in the ventricular zone ( D ) . ( E ) Graph showing the percentage of Ki67 and ( F ) β-III Tubulin+ cells in the cerebellar ventricular zone . ( G ) Schematic describing the transwell migration assay . Addition of SDF1α significantly increased the number of cells migrating through the membrane of the insert . Addition of AMD3100 to the upper well significantly lowered the number of migrating cells ( H ) . ( H ) Graph quantifying the results of the experiment described in ( G ) ( I ) Schematic of a sagittal section of the e13 . 5 cerebellum . Ventricular zone progenitors are labelled in yellow , while Ptf1a positive cells that exit the ventricular zone are marked red . RL progenitors are labelled light green while granule cell progenitors exiting the RL are labelled dark green . ( J–K′′ ) Matrigel assay to study the effect of SDF1α on the neuronal migration . Ptf1a positive cells from an e13 . 5 cerebellar slice , when incubated with SDF1α coated acrylic beads were seen to move towards the source of the chemokine ( K–K′′ ) . Saline coated beads had no effect on migration ( J , J′ ) . Abbreviations used; VZ—Ventricular Zone , RL—Rhombic Lip , EGL—External Granular Layer and Mes—Mesenchyme . The white dotted line in ( C–D′ ) represents the cerebellar ventricular surface . Scale bar = 100 µm . In Graphs ( B ) , ( E , F ) and ( H ) , data is represented as mean percentage of BrdU or β-III Tubulin positive cells or migrating cells ± s . e . m . In ( B ) and ( H ) ** indicates significance with respect to Control , while *** indicates significance with respect to SDF1α treatment . In ( E , F ) *** indicates significance with respect to Control . ( p < 0 . 05 ) , for all data . DOI: http://dx . doi . org/10 . 7554/eLife . 03962 . 00810 . 7554/eLife . 03962 . 009Figure 5 . Cxcr4 , a receptor of SDF1α , is strongly expressed in the soma and fibers of cerebellar radial glia . ( A–F ) Sagittal sections of the developing mouse cerebellum at e13 . 5 ( A , B , C , D ) , e14 . 5 ( E ) and e17 . 5 ( F ) . RFP-tagged SDF1α is secreted by the mesenchyme ( A , arrow ) . In contract , the receptor of SDF1α , GFP-tagged Cxcr4 is strongly expressed in the ventricular zone , NTZ , EGL and RL at e13 . 5 ( B–C ) . Cxcr4 expression was also seen in the fibers that extended out of the ventricular zone ( C , D , E arrow ) . These fibers colabelled with Nestin ( D , arrow ) indicating they were radial glial fibers . Cxcr4 expression persisted in radial glial fibers at e17 . 5 , but only confined to the anterior region containing radial glial fibers ( F , left of the yellow line ) . Abbreviations used; VZ—Ventricular Zone , NTZ—Nuclear Transitory Zone , RL—Rhombic Lip , EGL—External Granule Layer and Mes—Mesenchyme . The white dashed line in ( A ) indicates the outer boundary of the EGL and the mesenchyme , while the yellow dotted line represents the cerebellar ventricular surface . Scale bar = 100 µm for all images except C , D , where the scale bar = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03962 . 009 SDF1α has chemoattractant properties on cerebellar EGL cells where it is required in the post-natal pia to maintain the adjacent EGL cells in their proliferative niche ( Ma et al . , 1998; Zou et al . , 1998; Klein et al . , 2001; Reiss et al . , 2002 ) . Earlier , it also acts as a chemoattractant for cells migrating from the RL to form the EGL ( Yu et al . , 2010 ) . Since we observed Purkinje cell migration deficits in Foxc1−/− mutants , we assessed if SDF1α has yet another role , acting as a pial surface chemoattractant to differentiating ventricular zone derived cells . Indeed , SDF1α significantly increased the number of e13 . 5 dissociated cerebellar cells that migrated through a membrane in a standard transwell migration assay . Further , addition of AMD 3100 to cells in the upper chamber significantly reduced the number of migrating cells . ( Figure 4G , H ) . To confirm that the cells that were responsive to SDF1α responsive cells were indeed cells arising from the ventricular zone and not just the rhombic lip , we performed a matrigel migration assay using explanted e13 . 5 cerebellar anlage where ventricular zone derived cells were lineally labelled by a Td-tomato reporter ( Ptf1a cre/+; Ai14 ) ( Figure 4I ) . Labelled e13 . 5 explants were embedded in matrigel adjacent to acrylic beads coated with saline or SDF1α ( Figure 4J , K ) . After 48 hr , Td-tomato positive cells were specifically directed towards the source of SDF1α ( Figure 4K , K′ , K′′ ) . Thus , not only can SDF1α promote cerebellar ventricular zone progenitor proliferation , it can also function as a chemoattractant to cerebellar ventricular zone derived cells . Cxcl12 , the gene encoding SDF1α is a direct target of Foxc1 ( Zarbalis et al . , 2012 ) . It is expressed in embryonic posterior fossa mesenchyme based on RT-PCR data and its expression is reduced in Foxc1−/− mutants ( Aldinger et al . , 2009 ) . Using a BAC transgenic mouse encoding an SDF1α:mRFP fusion protein ( Jung et al . , 2009; Mithal et al . , 2013 ) , we confirmed that SDF1α protein is expressed in mesenchymal cells adjacent to the cerebellar anlage at e13 . 5 . In addition , we observed scattered RFP-positive cells , likely pericytes , within the cerebellar anlage at this stage ( Figure 5A , arrow ) . To extend previous reports of cerebellar ventricular zone expression of the SDF1α receptor ( Zou et al . , 1998; Tissir et al . , 2004 ) , we assessed Cxcr4 expression using both a BAC-transgenic Cxcr4-GFP transcriptional reporter mouse ( Mithal et al . , 2013 ) ( Figure 5B , C , arrow ) , and Cxcr4 immunohistochemistry ( Figure 5E , F , arrow ) . At e13 . 5 , Cxcr4 was expressed around the soma of cells in the ventricular zone , the rhombic lip , the nascent EGL , the NTZ and the pia ( Figure 5B ) . Strikingly , Cxcr4 expression was also seen in fibers extending out of the ventricular zone towards the pial surface . These fibers were nestin-positive and thus are radial glial fibers ( Figure 5D , arrow ) . As expected of radial glial fibers , they persisted at e14 . 5 and became restricted to the anterior cerebellum by e16 . 5 ( compare Figure 5F with 3E ) . Unlike nestin however , Cxcr4 expression was not retained in Bergmann Glial cells of the developing cerebellar cortex . Both our in vitro and in vivo expression data strongly suggested that SDF1α-Cxcr4 signalling could influence both the proliferation of cerebellar ventricular zone progenitors and the migration of its derivatives , making this pathway a highly attractive downstream effector of Foxc1 during embryonic cerebellar development . Previous phenotypic descriptions of Cxcr4 and SDF1α KO cerebellar phenotypes have focused on the role of this pathway in EGL development and maintenance ( Ma et al . , 1998; Zou et al . , 1998 ) . Potential earlier roles in cerebellar ventricular zone development have never been addressed . Thus , we examined the embryonic phenotypes of Cxcr4fl/fl; nestin-cre ( Cxcr4 CKO ) mice , where Cxcr4 function was specifically removed from radial glial cells and their derivatives . As seen in Foxc1−/− mutant cerebella , numbers of BrdU-positive ventricular zone cells were also dramatically reduced in the Cxcr4 CKO at e13 . 5 and e15 . 5 ( Figure 6A–D , I , Graph ) . Concurrently , there was an extensive increase in numbers of differentiated β-III tubulin positive cells at e13 . 5 and e15 . 5 . Similar to Foxc1 null mutants , numerous β-III tubulin -positive cells were also ectopically interspersed throughout the diminished Cxcr4 CKO mutant ventricular zone at both stages ( Figure 6F , H , Boxed area ) . Again , similar to Foxc1−/− mutant few calbindin-positive Purkinje cells were present at e15 . 5 in the Cxcr4 CKO and those remaining were scattered throughout the anlage and located adjacent to the pia ( Figure 6L , arrow ) , instead of remaining in a domain more closely associated with the ventricular zone as in wild-type cerebella ( Figure 6K , arrow ) . Although many nestin-positive radial glial cells were present in the Cxcr4-CKO mutant at e13 . 5 and e15 . 5 , long radial glial fibers were not present at either stage ( Figure 6M–P ) . Notably , radial glial fibers were still readily detected in the midbrain ( Figure 6N , arrows ) . As seen in the Foxc1−/− mutant , increased cell death was also observed in the Cxcr4 CKO ventricular zone compared to control at e13 . 5 and e15 . 5 ( Figure 6J ) . In contrast to the Foxc1−/− mutant cerebellum however , nests of ectopic proliferating granule cell progenitors were present within the Cxcr4-Nes-Cre mutant cerebellar anlage at this stage , as others have previously reported ( Figure 6D , asterisk ) . Bergmann glia were found ectopically located in the EGL similar to the Foxc1−/− mutant ( Figure 6R , arrows ) . We conclude that loss of Cxcr4 in the nestin-expressing radial glial cells recapitulates multiple aspects of the Foxc1−/− mutant cerebellar phenotype . The striking similarities of these mutant cerebellar phenotypes strongly argues that Foxc1-dependent SDF1α secretion by the posterior fossa head mesenchyme and its reception by Cxcr4 in adjacent early cerebellar anlage radial glial cells , represent a major downstream effector pathway mediating Foxc1 Dandy-Walker cerebellar pathogenesis10 . 7554/eLife . 03962 . 010Figure 6 . Aberrations in cell division , differentiation and migration in Cxcr4 radial glial conditional knockout mice are similar to those in Foxc1 null mutants . ( A–H ) Sagittal sections of e13 . 5 ( A , B , E , F ) and e15 . 5 ( C , D , G , H ) cerebellum from WT ( A , C , E , G ) and Cxcr4 Conditional Knockout ( B , D , F , H ) mice . ( A–D ) Cxcr4 conditional knockout mice showed a reduction in proliferation at e13 . 5 ( B ) and e15 . 5 ( D ) , compared to WT ( A , C ) . Reduced proliferation was accompanied by increased differentiation in the cerebellar ventricular zone of Cxcr4 CKO mice at both e13 . 5 ( F ) and e15 . 5 ( H ) . The ventricular zone in WT cerebellum at both e13 . 5 ( E ) and e15 . 5 ( H ) largely consisted of β-III Tubulin negative cells . The insets in images ( E–H ) represent the magnified image of boxed regions within the cerebellar ventricular zone . ( I ) Graph showing a significant reduction in the percentage of BrdU positive cells in the cerebellar ventricular zone of Cxcr4 CKO compared to WT cerebellar ventricular zone . Data are represented as mean percentage of BrdU positive cells ± s . e . m . *** indicates significance with respect to corresponding WT Control . ( p < 0 . 05 ) . ( J ) Graph showing a significant increase in the number of TUNEL positive cells in the cerebellar ventricular zone of Cxcr4 CKO compared to WT animals . Data are represented as mean number of TUNEL positive cells ± s . e . m . *** indicates significance with respect to corresponding WT Control . ( p < 0 . 05 ) . ( K–L ) Sagittal sections of e15 . 5 cerebellum from WT ( K ) and Cxcr4 CKO ( L ) mice stained for Calbindin showing aberrant number and position of PCs in the CKO . ( K–L ) Sagittal sections of e13 . 5 cerebellum ( M , N ) and e15 . 5 ( O , P ) from WT ( M , O ) and Cxcr4 CKO ( N , P ) mice stained for Nestin . The morphology of radial glial fibers in the Cxcr4 CKO greatly resembled that in the Foxc1 mutant . ( Q–R ) Sagittal sections of e17 . 5 cerebellum from WT ( Q ) and Cxcr4 CKO ( R ) mice stained for BLBP . The morphology and positioning of BG fibers in the Cxcr4 CKO ( R , arrows ) is similar to the Foxc1 mutant . Abbreviations used include VZ—Ventricular Zone , and Mes—Mesenchyme . Scale bar = 100 µm for all images , except for Q–R where the scale bar = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03962 . 010 If SDF1α is indeed one of the major effectors of the Foxc1 dependent signalling pathway , application of SDF1α in Foxc1−/− mutants should be able to rescue the Foxc1−/− phenotype . Since there are no commercially available SDF1α or Cxcr4 agonists , or posterior fossa specific cre drivers , we carried out an in vitro rescue experiment with subdissected whole cerebellar tissue from e13 . 5 WT and Foxc1−/− embryos . The cerebellar anlage was incubated with or without SDF1α ( 1 µg/ml ) in serum free media for 12 hr . To quantify SDF1α dependent VZ proliferation rescue , the tissue was then fixed , sectioned and stained for Ki67 and β-III Tubulin ( Figure 7A–H ) . SDF1α treatment did not have any effect on the WT cerebellum ( Figure 7A , E , D , H ) , but led to a significant increase in proliferation in the VZ of the Foxc1−/− cerebellum ( Figure 7 , Compare 7B with 7A and 7C; 7D Graph ) . Also , while there were no differences in differentiation levels in the WT cerebellar tissue treated with and without SDF1α ( Figure 7E , H ) , SDF1α significantly inhibited the early differentiation phenotype in the Foxc1−/− cerebellum ( Figure 7 , Compare 7F with 7G and 7E; 7H Graph ) . These data reinforce our conclusion that SDF1α is the major downstream effector of posterior fossa Foxc1 . 10 . 7554/eLife . 03962 . 011Figure 7 . Exogenously applied SDF1α rescues the Foxc1−/− mutant phenotype . ( A–C ) Sagittal sections of the e13 . 5 whole cerebellar tissue from WT ( A ) and Foxc1−/− ( B–C ) embryos , stained for Ki67 . Addition of SDF1α to the culture significantly increased proliferation in the Foxc1−/− cerebellar VZ ( C ) as compared to its mutant negative control not treated with SDF1α ( B ) . Addition of SDF1α did not significantly increase proliferation in the WT ventricular zone ( A , D ) . ( D ) Graph showing the percentage of Ki67 cells in the cerebellar ventricular zone . ( E-–G ) Sagittal sections of the e13 . 5 whole cerebellar tissue from WT ( E ) and Foxc1−/− ( F–G ) embryos stained for β-III Tubulin . SDF1α significantly reduced the number of Tuj1+ cells in the VZ of the Foxc1−/− mutant ( G ) as compared to its mutant negative control ( F ) . Addition of SDF1α did not significantly alter differentiation levels in WT embryos ( E , H ) . ( H ) Graph showing the percentage of β-III Tubulin+ cells in the cerebellar ventricular zone . Data for both graphs ( D ) and ( H ) are represented as mean number of Ki67 or Tuj1 positive cells ± s . e . m . *** indicates significance with respect to corresponding WT Controls; ### indicates significance with respect to the corresponding Foxc1−/− mutant—SDF1α negative control . ( I ) Schematic of a sagittal section of the embryonic mouse cerebellum . In the WT cerebellum , mesenchymal Foxc1 controls the expression of several secreted factors such as SDF1α and Bmp2 , 4 . SDF1α binds to its receptor Cxcr4 which is widely expressed in the RL , ventricular zone and EGL . SDF1α induces proliferation in the cerebellar ventricular zone . Not only is it required to maintain radial glia and their processes , it also functions as a chemoattractant to neurons migrating away from the ventricular zone . In the RL , Bmps maintain the progenitor pool , and SDF1α acts to attract the GC progenitors from the RL to form the EGL , thereby maintaining these progenitors at the pial surface . ( J ) In the Foxc1−/− cerebellum , deletion of Foxc1 leads to a significant downregulation of SDF1α and Bmps . A reduction in SDF1α expression causes reduced ventricular zone proliferation , increased differentiation and abnormal migration . Increased cell death and dysmorphic radial glial fibers are also observed . GCs migrating out of the RL into the EGL follow abnormal paths away from the pial surface into the interior of the cerebellar anlage . Absence of this chemoattractant leads to a buildup of cells along the ventricular zone and RL . DOI: http://dx . doi . org/10 . 7554/eLife . 03962 . 011
Loss of Foxc1 is associated with Dandy-Walker malformation , the most common congenital malformation of the human cerebellum . Although extensive cerebellar pathology results from the loss of Foxc1 , in the mouse , Foxc1 is not highly expressed in the cerebellar anlage , but rather is expressed in the posterior fossa mesenchyme overlying the developing cerebellar anlage beginning at e12 . 5 . This finding leads to the hypothesis that disrupted mesenchyme to brain signalling caused the Foxc1 mutant cerebellar phenotype . We previously reported disorganization of the cerebellar RL and nascent EGL located directly adjacent to the mesenchyme by e14 . 5 ( Aldinger et al . , 2009 ) . We now find that the cerebellar ventricular zone is not insulated from loss of Foxc1 despite its distance from the posterior fossa mesenchyme . Indeed , Foxc1 null mutants have an early and devastating reduction in ventricular zone radial glial cell proliferation , increased cell death and increased differentiation at e13 . 5 , 1 day after failure of normal initiation of expression of Foxc1 in the posterior fossa mesenchyme . At later stages , residual ventricular zone derived Purkinje cells in the mutant display migratory deficits which correlate with loss of radial glial fibers which normally serve as a scaffold for neuronal migration . Using in vitro and in vivo assays , we demonstrated that SDF1α , a direct target of Foxc1 in head mesenchyme ( Zarbalis et al . , 2012 ) , via its receptor , Cxcr4 is a major effector of Foxc1-dependent mesenchymal signalling . Cxcr4 is expressed in ventricular zone progenitors , including the radial glial fibers and their endfeet which associate with the developing pia . Loss of cerebellar Cxcr4 largely phenocopies the Foxc1 null phenotype and the Foxc1−/− phenotype can be rescued by the addition of SDF1α . Together , our data demonstrates SDF1α-Cxcr4 signalling is a major downstream effector of mesenchymal Foxc1 function and is required to maintain both radial glial proliferative and scaffolding functions in the early cerebellar anlage . Head mesenchyme is a mixture of head mesoderm and head neural crest cells which surrounds the developing brain ( Yoshida et al . , 2008 ) . It differentiates to form all structures between the brain and the epidermis , including the meninges which secretes the pial basement membrane the skull and the skin dermis . Signalling from the head mesenchyme and its derivatives to the developing brain has been implicated in a number of important developmental processes from brain regionalization during early neural plate and neural tube stages ( LaMantia et al . , 2000; Le Douarin et al . , 2012; Andoniadou and Martinez-Barbera , 2013 ) through the regulation of tangential migration of Cajal Retzius cells and cortical interneurons ( Kwon et al . , 2011; Zarbalis et al . , 2012 ) . Physical interactions between the pial surface and the end-feet of radial glial cells are also known to be required for radial glial survival , proliferation as well as maintenance of radial fiber morphology to direct the radial migration of neurons ( Radakovits et al . , 2009; Siegenthaler and Pleasure , 2011 ) . The molecular identities of these mesenchymally-derived signals are beginning to be elucidated . For example , Bmp , Fgf and Wnts are coordinately required to induce and then caudalize the neural plate and neural tube ( Andoniadou and Martinez-Barbera , 2013 ) . SDF1α , a chemokine has been shown to regulate tangential migration of cortical neurons . Intact integrins and laminins are required in the pial basement maintain contact with radial glial endfeet ( Radakovits et al . , 2009 ) . To begin to elucidate the nature of Foxc1-dependent posterior fossa head mesenchymal signalling to the developing cerebellum , we previously reported downregulation of a number of genes encoding secreted molecules in Foxc1 null posterior mesenchyme at e12 . 5 including Bmps 2 and 4 and SDF1α ( Aldinger et al . , 2009 ) . SDF1α is a direct target of Foxc1 in head mesenchyme ( Zarbalis et al . , 2012 ) . We have now demonstrated that in vitro , SDF1α could induce cerebellar progenitor proliferation and the migration of neurons; an effect which could be blocked by AMD3100 , an antagonist of the Cxcr4 receptor ( Rosenkilde et al . , 2004 ) . In vivo , published in situ data together with new protein expression analysis demonstrated the presence of Cxcr4 in the cerebellar ventricular zone throughout embryogenesis . In contrast , extensive in situ analysis failed to detect expression of any Bmp receptor activity in the cerebellar ventricular zone at relevant stages ( Machold et al . , 2007 ) ( data not shown ) . Further , Bmp2 and 4 could not alter migration of ventricular zone derived neurons in in vitro assays at e13 . 5 . Thus , SDF1α became a high priority candidate for further analysis . Notably , SDF1α and Cxcr4 have previously been implicated in cerebellar granule cell development ( Ma et al . , 1998; Zou et al . , 1998; Klein et al . , 2001; Reiss et al . , 2002; Zhu et al . , 2002; Vilz et al . , 2005; Yu et al . , 2010 ) . SDF1α expressed in the meninges functions as chemoattractant to direct granule neuron progenitors away from the RL to the surface of the anlage to form the EGL . In the early postnatal cerebellum , SDF1α also exerts an influence on GNPs of the outer EGL which lie directly underneath the pial surface . Loss of SDF1α or Cxcr4 in the cerebellum causes GNPs to prematurely leave their EGL proliferative niche , resulting in ectopic proliferation and dramatic disruptions of the laminar structure of the resultant cerebellum ( Ma et al . , 1998; Zou et al . , 1998 ) . Our results now considerably expand the role of SDF1α-Cxcr4 signalling in the developing cerebellum to include almost all aspects of radial glial function including proliferation , survival and maintenance of radial morphology ( Figure 7 ) . Although the focus of previous studies of SDF1α-Cxcr4 signalling has been on neuronal migration , our new analysis now demonstrates a more fundamental role for this signalling pathway . Loss of either Foxc1 or Cxcr4 causes dramatic 25% and 30% respective reductions in ventricular zone proliferation by e13 . 5 which rapidly depletes the early cerebellar progenitor pool . This difference may reflect the residual SDF1α-Cxcr4 signalling in the Foxc1 mutants ( Aldinger et al . , 2009 ) , vs complete loss in the Cxcr4 CKO . Taken together with other published studies , we conclude that embryonic cerebellar anlage size is regulated by a number of extra-cerebellar signalling systems . Wnt signals derived from both the isthmic organizer ( Sato et al . , 2004; Sato and Joyner , 2009 ) and the early dorsal roof plate ( Chizhikov et al . , 2006 ) determine the initial size of the anlage from ∼e9 . 0 . We have now demonstrated that mesenchymal SDF1α signalling is a major driver of early anlage growth from ∼e12 . 5 . Late embryonic anlage ventricular zone proliferation is influenced by Shh derived from the differentiating choroid plexus from e14 . 5 ( Huang et al . , 2010 ) . An outstanding issue remains how each of these signalling systems are coordinately regulated and integrated across embryogenesis to orchestrate normal anlage development . In addition to serving as neuronal and glial progenitors in the ventricular zone , radial glial cells extend fibers that act as migratory scaffolds for neurons migrating out of the ventricular zone . We observed aberrant migration of Purkinje cells in both Foxc1 null and Cxcr4; nestin-cre mutants , in addition to dysmorphic radial glial fibers . Abnormal migration of cells out of the ventricular zone could be due to the absence or downregulation of SDF1α chemoattraction from the mesenchyme and derived meninges . This is buttressed by the fact that blockage of SDF1α led to a significant reduction in migration of single cells in the in vitro transwell migration assay . Also , in the matrigel assay , when embryonic cerebellar slices were placed in close proximity with SDF1α coated acrylic beads , explant cells and fibers migrated in the direction of the SDF1α source . However , our protein expression data using Cxcr4 antibodies , in addition to Cxcr4-GFP transgenic mice indicate that Cxcr4 is also strongly expressed in radial glial fibers themselves . As early as e13 . 5 and more prominently by e15 . 5 , radial glial fibers were discontinuous and did not extend all the way to the basement membrane of the pia . We conclude that SDF1α secreted by the mesenchyme binds to the receptor on the radial fibers to relay signals that are directly required to maintain these radial fibers as a substrate for migration . Indeed in the spinal cord , SDF1α protein is detectable in radial end feet and transcytosis across the radial glial length is also observed ( Mithal et al . , 2013 ) . Radial glial cells in the cerebellum can differentiate into Bergmann glial cells ( also known as Golgi Epithelial cells ) in the cerebellar cortex ( Li et al . , 2014 ) . In both Foxc1−/− and Cxcr4 CKO mutants , Bergmann glial cells were ectopically placed in the EGL and ML , while in wild-type animals Bergman glial nestin+ and BLBP+ cells extended processes from the EGL to the presumptive IGL and were found overlapping with the PL . Studies have also shown that the association of BG with PCs greatly contributes to their structural and functional maintenance . Sonic hedgehog secretion by PCs also helps maintain BG in the cerebellum . It is hence plausible that the BG phenotype may be secondary to abnormal RG and PCs . Fate mapping studies have shown that Bergmann glial cells are normally born between e13 . 5 and e14 . 5 in the mouse cerebellum ( Sudarov et al . , 2011 ) . It is also possible that continued SDF1α-Cxcr4 signalling is actively required for the differentiation of BG from RG . Further analysis is required to distinguish these two possibilities . Regardless , it is certain that a lack of normal Bergmann glial fibers further contributes to the abnormal lamination in neonatal SDF1α and Cxcr4 mutant animals as these fibers are normally required for the inward migration of EGL cells to form the internal granule layer . Our analysis indicates that radial glial cells in the developing cerebellum are more sensitive to loss of SDF1α signalling compared to other regions of the CNS . In the cerebellum , loss of SDF1α signalling causes immediate and catastrophic cell cycle exit of radial glial cells and corresponding increased neuronal differentiation . Rapid loss of radial glial fiber morphology is also observed . Loss of SDF1α signalling in the forebrain has not been reported to cause microcephaly in mice , and we have shown here that midbrain radial glial cells are relatively unaffected ( Figure 6N ) . A recent study ( Mithal et al . , 2013 ) reported that loss of radial glial Cxcr4 in spinal cord radial glial cells causes disruption of radial glial morphology and reduced mitosis in the ventral ventricular zone by e14 . 5 . However , in the spinal cord by this stage a large number of spinal cord neuronal populations have already been generated ( Tanabe and Jessell , 1996 ) . Although late born populations such as oligodendrocyte lineages are altered , there are not dramatic morphological consequences . The disproportionate mouse cerebellar phenotypes we report here with loss of Foxc1 and Cxcr4 beautifully model the CNS phenotype of Dandy-Walker malformation patients , where the cerebellum is disproportionately diminished in size and altered in morphology relative to other CNS structures . The reasons for the region-specific sensitivities to loss of the SDF1α-Cxcr4 signalling remain unknown and may simply reflect regional differences in the timing of proliferative epochs relative to the onset of SDF1α and Cxcr4 expression . Alternatively , these differences may more fundamentally reflect region specific mesenchymal signalling mechanisms that further analyses will define . Our data contribute to a growing body of evidence that the final form and function of the brain is a product of both intrinsic and extrinsic signalling mechanisms . Here we have demonstrated that the pathogenesis underlying Foxc1-dependent Dandy-Walker malformation , a major human neurodevelopmental disorder , is caused by loss of SDF1α-Cxcr4 mediated cerebellar radial glial proliferative and migrational scaffold function . These functions represent a previously unrecognized role for this major signalling pathway . Thus , SDF1α function is central to almost every major developmental program of the cerebellum . Notably , SDF1α is also a chemokine , central to the body's inflammatory response ( Dotan et al . , 2010; Werner et al . , 2011 , 2013 ) . Maternal infection has long been recognized as a risk factor for neurodevelopmental disorders , including autism which has been associated with cerebellar hypoplasia ( Fatemi et al . , 2008; Buehler , 2011; Burd et al . , 2012; Garbett et al . , 2012 ) . It is intriguing to speculate that neuroinflammation , including misregulated SDF1α signalling during early fetal development may underlie cases of human cerebellar hypoplasia for which genetic mechanisms have been difficult to assign ( Sajan et al . , 2010 ) .
All animal experimentation done in this study was done in accordance with the guidelines laid down by the Institutional Animal Care and Use Committee ( IACUC ) , of Seattle Children's Research Institute , Seattle , WA , USA . Foxc1lacZ/+ mice were generated by Kume et al . ( 1998 ) ( Northwestern University ) . Foxc1lacZ/+ mice were crossed and the day of plug was taken as embryonic day ( e ) 0 . 5 . We refer to this allele as the Foxc1 null allele . Foxc1 homozygous null mutants are referred to as Foxc1−/− mice . Mutant and WT Control embryos were dissected out at days 13 . 5 , 14 . 5 , 15 . 5 , e17 . 5 and e19 . 5 after plugging . Nes-Cre mice ( The Jackson laboratory Stock number −003771 ) were crossed with Cxcr4loxP mice ( The Jackson laboratory Stock number −008767 ) to generate Nes-Cre;Cxcr4 loxP/+ , which were in turn crossed with Cxcr4loxP mice to generate Cxcr4 CKO mice . Cxcr4 GFP , SDF1α RFP mice were generated as described previously ( Jung et al . , 2009; Mithal et al . , 2013 ) . Embryos were fixed in 4% paraformaldeyde ( PFA ) overnight , washed in PBS and sunk in 30% sucrose . Embryos were subsequently embedded in optimum cutting temperature ( OCT ) compound . Mid-sagittal cryo-sections of 11 microns were taken . For BrdU studies , pregnant mice were injected with BrdU ( 100 µg/g body weight ) and sacrificed 1 hr later . In order to label cerebellar ventricular zone lineage , Ptf1a-Cre ( Hoshino et al , 2005 ) mice generated by C . V . Wright ( Vanderbilt University Medical Center ) were crossed with Ai14 reporter mice ( Madisen et al . , 2010 ) ( The Jackson Laboratory Stock number—007908 ) . Immunohistochemistry was performed as described previously ( Haldipur et al . , 2011 ) . Sections were subjected to antigen retrieval prior to staining . All sections were blocked using 5% serum with 0 . 1% triton X , and then incubated with the primary antibody , overnight . The primary antibodies used in this study were BrdU ( Abcam , Cambridge , UK; 1:50 ) , β-III Tubulin ( Promega , Madison , WI , USA; 1:1000 ) , Calbindin ( Swant , Switzerland; 1:3000 ) , Cxcr4 ( Abcam—1:100 ) , BLBP ( Abcam—1:100 ) , Ki67 ( Vector—1:300 ) , Laminin ( Sigma—1:25 ) , Raldh2 ( Sigma—1:400 ) , Pdfgr1 ( BD Biosciences , San Jose , CA—1:200 ) and Nestin ( Millipore , Germany; 1:200 ) . The following day , the following species and subtype appropriate secondary antibodies were used—biotinylated ( 1:250 , Vector laboratories , Burlingame , CA , USA ) or fluorescent dye labelled ( Alexa fluors 488 and 594 , 1:1000 , Molecular probes , Grand Island , NY , USA ) . Sections were counter stained with DAPI using in Vectashield mounting media with DAPI ( 4′ , 6-diamidino-2-phenylindole ) ( Vector laboratories ) or DPX ( Distyrene . Plasticizer . Xylene ) mounting medium . For each antibody , one section was used as a negative control where in the section was incubated with all the above solutions except the primary antibody . For BrdU immunohistochemistry , sections were treated with 1N HCl for 10 min on ice , 2 N HCl for 30 min at RT , and washed with Borate buffer prior to blocking . TUNEL assay was carried out In Situ Cell Death Detection Kit , TMR red ( Roche , Germany ) . Briefly , cerebellar sections were incubated in the TUNEL mix ( terminal deoxynucleotidyl transferase in storage buffer and TMR red labeled-nucleotide mixture in reaction buffer ) for 1 hr at RT . The slides were washed with PBS and mounted with Vectashield mounting media containing DAPI or Propidium Iodide . The cerebellar anlage from e13 . 5 embryos was dissected aseptically in CMF-Tyrode solution . Meninges were removed , tissue were chopped into smaller pieces and collected in CMF-Tyrode . The tissue was treated with trypsin-DNAse and then dissociated in the same solution by triturating to make a single cell suspension , pelleted and resuspended in Dulbecco's modified Eagle's medium-F-12 ( DMEM-F-12 ) containing 15 mM HEPES , L-glutamine , pyroxidine hydrochloride ( Invitrogen , Grand Island , NY , USA ) , N2 supplement ( Invitrogen ) , 10% fetal calf serum , 25 mM KCl , and penicillin-streptomycin . Cells were plated onto poly-D-Lysine -coated Labtek chamber ( Nunc , Roskilde , Denmark ) or Poly-D Lysine coated transwell migration inserts . The cells were then cultured in serum free medium for another 24 hr , during which they were treated with the following factors—SDF1α ( R and D systems , USA , 1 µg/ml ) , and/or AMD 3100 ( 5 µg/ml ) . The treatment continued for the next 24 hr after which cells were washed with 1× PBS and then fixed in 4% PFA . For BrdU labeling experiments , BrdU was added to serum free media at 10 μM . For whole mount cerebellar explants , the cerebellar anlage from e13 . 5 embryos was collected and incubated in serum containing media with or without AMD3100 for 24 hr . For the rescue experiments , the cerebellar anlage was incubated in serum free media with or without SDF1α ( 1 μg/ml ) , for 12 hr . The tissue was then fixed and processed for IHC as stated previously . Cells from an e13 . 5 cerebellum were dissociated and seeded into a cell culture insert . The insert was then placed into a well of a 24 well plate supplemented with Serum Free Media and SDF1α ( 1 µg/ml ) was added to the lower well . To test the effect of SDF1α blockage on migration , AMD3100 ( 5 µg/ml ) was added into the insert while SDF1α was added to the media in the lower well . Cells added to the upper chamber were allowed to migrate for 72 hr . Migrated cells attached to the bottom surface of the insert were quantified by DAPI staining . For matrigel migration experiments , e13 . 5 embryos from Ptf1a Cre; Ai14 mice were collected . The cerebellar anlage was subdissected and sliced . Slices were embedded in matrigel and co-incubated with SDF1α ( 1 µg/ml ) coated acrylic beads for 48–72 hr . Control slices were incubated with saline coated beads . All images were captured at room temperature . Apart from minor adjustment of contrast and brightness no additional image alteration was done . All images were captured using the Zeiss Axioimager Z1 Microscope equipped with an Axiocam MRC camera and Axiovision Rel 4 . 8 software ( Zeiss ) . All cell counts were performed using Image J software ( National Institutes of Health , Bethesda , MD , USA ) . Our analyses include data from n > 3 and N > 5 , where n refers to the total number of embryos used and N represents the total number of sections per animal . All embryos were sectioned at 11 microns sagittally . Sections between WT and mutant were carefully chosen to represent the same orientation along the mediolateral axis . To minimize bias , blind counts were performed . To evaluate proliferation and differentiation , the total number of Ki67 , BrdU and Tuj1 positive cells in the ventricular zone was counted . This was followed by a total DAPI count that represented the total cell count in the ventricular zone . The percentage of proliferating and differentiating cells in the ventricular zone was represented in the graph . For cell death , the total number of TUNEL positive cells that spanned the entire length of the ventricular zone was counted . Data are represented as mean ± s . e . m . Statistical significance was determined by two tailed t-test . p < 0 . 05 was considered statistically significant . | The part of the brain responsible for coordinating and fine-tuning movement , sensory processing and some cognitive functions—the cerebellum—is found tucked away at the back of the brain , where it sits in a hollow in the skull called the posterior fossa . In a relatively common neurological disorder called Dandy-Walker malformation , part of the cerebellum doesn't develop and the posterior fossa is abnormally large . One contributing factor to Dandy-Walker malformation is the loss of a protein called Foxc1 . This protein is a so-called transcription factor , meaning it activates other genes , and so it has various important roles in helping an embryo to develop . In mouse embryos , the gene that produces Foxc1 is not activated in the developing cerebellum itself , but rather in the adjacent mesenchyme , a primitive embryonic tissue that will develop into the membranes that cover the brain and the skull bones that define the posterior fossa . This led Haldipur et al . to propose that the mesenchyme and the cerebellum communicate with each other as they develop . To investigate this idea , Haldipur et al . carefully analysed how the development of the mouse cerebellum goes awry when Foxc1 is absent . This revealed that Foxc1-deficient mice have lower numbers of a type of cell called radial glial cells in their cerebellum . These are ‘progenitor’ cells that develop into the various types of cell found in the cerebellum , and also act as a scaffold for other neurons to migrate across . Therefore , the loss of radial glial cells in Foxc1-deficient mice substantially disrupts how the cerebellum develops , and how the neurons in the cerebellum work . One gene activated by the Foxc1 protein encodes another protein called SDF1-alpha . This protein is released from the tissue that will develop into the posterior fossa , and binds to a receptor protein that is present on radial glial cells in the cerebellum . When this binding occurs , the radial glial cells grow and divide , and so the embryo's cerebellum also grows . Haldipur et al . found that mouse embryos specifically missing this receptor develop many of the abnormalities seen in Foxc1-deficient mice and further , when SDF1-alpha was provided back into Foxc1-deficient cerebella , the defects were rescued . This suggests that the cerebellar defects caused by the loss of Foxc1 stem from disrupting the signalling pathways that are triggered by the interaction between SDF1-alpha and its receptor . These studies highlight that the brain does not develop in isolation . It is strongly dependent on the signals it receives from the embryonic mesenchyme that surrounds it . Identifying these signals and understanding how they can be disrupted by both genetic and non-genetic causes , such as inflammation , may be key to understanding this important class of brain birth defects . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2014 | Foxc1 dependent mesenchymal signalling drives embryonic cerebellar growth |
p27Kip1 ( p27 ) is a cyclin-CDK inhibitor and negative regulator of cell proliferation . p27 also controls other cellular processes including migration and cytoplasmic p27 can act as an oncogene . Furthermore , cytoplasmic p27 promotes invasion and metastasis , in part by promoting epithelial to mesenchymal transition . Herein , we find that p27 promotes cell invasion by binding to and regulating the activity of Cortactin , a critical regulator of invadopodia formation . p27 localizes to invadopodia and limits their number and activity . p27 promotes the interaction of Cortactin with PAK1 . In turn , PAK1 promotes invadopodia turnover by phosphorylating Cortactin , and expression of Cortactin mutants for PAK-targeted sites abolishes p27’s effect on invadopodia dynamics . Thus , in absence of p27 , cells exhibit increased invadopodia stability due to impaired PAK1-Cortactin interaction , but their invasive capacity is reduced compared to wild-type cells . Overall , we find that p27 directly promotes cell invasion by facilitating invadopodia turnover via the Rac1/PAK1/Cortactin pathway .
p27Kip1 is a cell cycle inhibitor that binds to a broad range of cyclin-CDK complexes ( Besson et al . , 2008 ) . p27-mediated cyclin-CDK inhibition involves the N-terminal extremity ( aa 28–89 ) of the protein that allows interaction with both cyclin and CDK subunits ( Ou et al . , 2012; Besson et al . , 2008 ) . The critical role of p27 as a negative regulator of cell proliferation is underscored by the phenotype of p27 knockout mice , which exhibit a hyperproliferative phenotype in multiple tissues and are more susceptible to tumor development than wild-type animals ( Fero et al . , 1996; Kiyokawa et al . , 1996; Nakayama et al . , 1996; Besson et al . , 2006; Fero et al . , 1998 ) . While the tumor suppressive role of p27 is confirmed by the frequent loss of p27 expression in many types of cancers , p27 is also mislocalized in a fraction of tumors due to activation of various signaling pathways that cause its nuclear export and/or cytoplasmic retention ( Chu et al . , 2008 ) . Cytoplasmic p27 has been associated with decreased patient survival , high tumor grade and metastasis in several tumor types including breast carcinomas , acute myeloid leukemia , glioblastoma , melanoma and non-small cell lung carcinomas ( Lin et al . , 2016; Chu et al . , 2008; Liang et al . , 2002; Yang et al . , 2011; Min et al . , 2004; Cheng et al . , 2015 ) . The possibility that cytoplasmic p27 may confer a pro-tumorigenic advantage was confirmed in a mouse model expressing a mutant form of p27 that cannot bind cyclin-CDK complexes ( p27CK- ) ( Besson et al . , 2007; Serres et al . , 2011 ) . These mice are more susceptible than p27-null and wild-type mice to both spontaneous and induced tumor development , thus uncovering an oncogenic role for p27 in vivo ( Serres et al . , 2011; Besson et al . , 2007 ) . It now appears that p27 is a multifunctional protein involved in the regulation of multiple cellular processes , some of which are regulated by p27 in a CDK-independent manner ( Besson et al . , 2008; Sharma and Pledger , 2016 ) . Indeed , p27 has been implicated in the control of cell migration , transcriptional repression , autophagy , stem cell specification and differentiation , cytokinesis , and apoptosis ( Besson et al . , 2008; Sharma and Pledger , 2016; Baldassarre et al . , 2005; Besson et al . , 2004b , 2007; Pippa et al . , 2012; Serres et al . , 2012; Nickeleit et al . , 2008; Liang et al . , 2007; Li et al . , 2012; Nguyen et al . , 2006; Jeannot et al . , 2015 ) . This functional versatility may stem from the intrinsically disordered structure of p27 , which folds upon binding to other proteins , allowing p27 to interact with a wide variety of partners ( Lacy et al . , 2004; Galea et al . , 2008 ) . While these various functions participate in homeostasis in normal cells and tissues , they may be co-opted by tumor cells to promote oncogenesis . Metastasis , the process whereby cells from a primary tumor disseminate throughout the body and form secondary tumors at distant sites , is the leading cause of cancer related death ( Sethi and Kang , 2011 ) . It requires tumor cells to migrate and invade through neighboring tissues , enter the blood flow and then extravasate to disseminate and form new tumors in host tissues ( Sethi and Kang , 2011 ) . Regulation of cell motility was the first CDK-independent role ascribed to p27 ( Nagahara et al . , 1998; Denicourt et al . , 2007; Besson et al . , 2004b ) . In the cytosol , p27 can interact with RhoA , preventing RhoA activation by guanine-nucleotide exchange factors ( GEFs ) and thereby modulating actin cytoskeleton dynamics and migration ( Besson et al . , 2004a , 2004b ) . Consequently , mouse embryo fibroblasts ( MEFs ) lacking p27 have more RhoA activity , increased numbers of stress fibers and focal adhesions and exhibit a defect in migration ( Besson et al . , 2004b ) . This pathway is important for proper migration of bone marrow macrophages and developing cortical neurons and for cancer cell migration and invasion in vivo ( Godin et al . , 2012; Nguyen et al . , 2006; Papakonstanti et al . , 2007; Gui et al . , 2014; Wu et al . , 2006; See et al . , 2010; Larrea et al . , 2009; Jin et al . , 2013 ) . p27 can also regulate cell migration by controlling microtubule stability through Stathmin or directly by binding to microtubules and promoting microtubule polymerization ( Baldassarre et al . , 2005; Godin et al . , 2012 ) . In a 3D environment , cells adopt different strategies to migrate and invade through surrounding matrix ( Petrie and Yamada , 2016; Sethi and Kang , 2011 ) . The mode of migration is influenced , at least in part , by the activities of Rho GTPases: RhoA activity favors an amoeboid migration , whereas Rac1 activity promotes mesenchymal migration ( Sahai and Marshall , 2003; Vial et al . , 2003 ) . Accordingly , cytoplasmic p27 promotes mesenchymal migration via the inhibition of RhoA activity , while cells lacking p27 tend to adopt an amoeboid mode of migration ( Gui et al . , 2014; Belletti et al . , 2010 ) . More recently , p27 was reported to promote epithelial to mesenchymal transition by binding to JAK2 , promoting STAT3 activation and the upregulation of Twist1 ( Zhao et al . , 2015 ) . Here , we describe a novel mechanism by which p27 directly regulates invadopodia formation and cell invasion through extracellular matrix ( ECM ) via Cortactin . Invadosomes ( designating both invadopodia and podosomes ) are thought to allow cells to coordinate ECM degradation with migration within the tissue microenvironment ( Murphy and Courtneidge , 2011; Linder et al . , 2011; Di Martino et al . , 2016 ) . Cortactin plays a key role in the formation of actin protrusive structures such as lamellipodia and invadosomes ( Murphy and Courtneidge , 2011; Kirkbride et al . , 2011; MacGrath and Koleske , 2012 ) . Cortactin has been involved in all steps of the invadosome lifecycle , from assembly , maturation , proteolytic activity and disassembly ( MacGrath and Koleske , 2012; Murphy and Courtneidge , 2011; Moshfegh et al . , 2014 ) . All these steps appear to be regulated by phosphorylation events on Cortactin ( Oser et al . , 2009; Moshfegh et al . , 2014; Murphy and Courtneidge , 2011 ) . Overall , Cortactin is a scaffold protein composed of an N-terminal acidic domain , followed by several actin-binding repeats allowing its interaction with F-actin , a Pro-rich region and a SH3 domain ( MacGrath and Koleske , 2012; Kirkbride et al . , 2011 ) . Phosphorylation of Cortactin by Src and Abl family tyrosine kinases has two effects: first , this releases the actin severing protein Cofilin from an inhibitory interaction with Cortactin and generates actin barbed ends; second , Arp2/3 , N-WASP and Nck1 are recruited onto Cortactin , promoting the polymerization of branched actin ( Oser et al . , 2009; Weaver et al . , 2002; Oser et al . , 2010 ) . Dephosphorylation of Cortactin then allows the inhibition of Cofilin activity and stabilizes the invadopodia ( Oser et al . , 2009 ) . Cortactin also plays a key role in matrix metalloproteinase secretion in mature invadopodia ( Clark et al . , 2007 ) . Finally , invadopodia disassembly is induced by sequential activation of the Rac-GEF Trio , Rac1 and p21-Activated Kinase-1 ( PAK1 ) pathway and presumably by PAK1-mediated phosphorylation of Cortactin on Ser113 , since a Ser113 to Ala mutant blocked invadopodia disassembly ( Moshfegh et al . , 2014 ) . How PAK-phosphorylated Cortactin mediates invadopodia disassembly is still unclear but could be due , at least in part , to a decreased affinity of Cortactin phosphorylated within its F-actin binding repeats ( on S113 , S150 and/or S282 ) for F-actin , potentially destabilizing the structure ( Webb et al . , 2006 , 2005 ) . We found that p27 binds to Cortactin and localizes to invadopodia following serum or growth factor stimulation and promotes cell invasion . Paradoxically , p27-null cells have more invadopodia and degrade gelatin more efficiently , but exhibit impaired invasive capacity . In fact , we found that p27 promotes the recruitment of PAK1 to Cortactin and invadopodia turnover and this is dependent on phosphorylation of S113/S150/S282 of Cortactin , the sites targeted by PAK kinases . Thus , in absence of p27 , the dynamics of invadopodia is altered and prevents efficient invasion through ECM . Altogether , we have identified a novel mechanism by which p27 directly controls cell invasion that could contribute to the increased invasive and metastatic capacity of tumors where p27 is mislocalized in the cytoplasm .
While the role of p27 in the regulation of cell migration is firmly established , how p27 is targeted to specific locations in the cytoplasm to control motility and invasion remains unclear ( Gui et al . , 2014; Belletti et al . , 2010; Besson et al . , 2004a , 2004b; Godin et al . , 2012; Nguyen et al . , 2006; Papakonstanti et al . , 2007; Wu et al . , 2006; See et al . , 2010; Larrea et al . , 2009; Jin et al . , 2013 ) . We recently identified Cortactin in a proteomic screen in which protein arrays ( Protoarray , ThermoFisher Scientific ) were probed with recombinant human p27 , indicating that the two proteins interact directly . We confirmed the binding of p27 to Cortactin in HEK 293 cells overexpressing p27 and Myc-tagged Cortactin ( Figure 1A ) , as well as on endogenous proteins in Hela cells and MEFs immortalized with the human papilloma virus E6 protein ( Figure 1B ) ( Serres et al . , 2011 ) . The interaction domain of Cortactin on p27 was mapped by pull-down assays using various GST-p27 fusion proteins and Myc-Cortactin in HEK 293 cells . Cortactin bound to the p27CK- mutant , that cannot interact with cyclins and CDKs ( Besson et al . , 2006 , 2007 ) , and to the C-terminal half ( aa 88–198 ) of p27 but not the N-terminal half ( aa 1–87 ) of the protein that contains the cyclin and CDK interaction domains ( Figure 1C and D ) . The Cortactin interaction domain on p27 was narrowed down to the last 8 aa of p27 , as a p27CK- 1–190 mutant did not interact with Cortactin any longer ( Figure 1E ) . A p27CK- 1–197 mutant , lacking only the C-terminal threonine residue , still bound to Cortactin ( Figure 1E ) , suggesting that phosphorylation on T198 is not needed for binding to Cortactin . 10 . 7554/eLife . 22207 . 003Figure 1 . p27 binds to Cortactin . ( A–B ) Co-immunoprecipitation of p27 and Cortactin: ( A ) p27 was immunoprecipitated using rabbit anti-p27 ( C19 ) antibodies from HEK293 lysates transfected with plasmids encoding p27 , Myc-Cortactin ( Myc-Cort ) or both . ( B ) Immunoprecipitation of endogenous Cortactin using rabbit anti-Cortactin ( H191 ) antibodies from HeLa or E6 MEF lysates , beads alone were used as control . ( A–B ) Co-immunoprecipitated proteins were detected by immunoblot with mouse anti-c-Myc ( 9E10 ) ( A ) or mouse anti-p27 ( F-8 ) antibodies ( B ) . Immunoprecipitated proteins were visualized by reprobing the membrane with mouse anti-p27 ( F-8 ) ( A ) or with rabbit anti-Cortactin ( H-191 ) antibodies and anti-rabbit Ig light-chain secondary antibodies ( B ) . Immunoblots of extracts show the level of proteins in each condition . ( C–F ) Pull-down assays: HEK293 cells were transfected with Myc-Cortactin ( C–E ) or various deletion mutants of Myc-Cortactin ( F ) ( ΔABR6 , ΔABR5-6 , ΔABR4-6 , ΔABR3-6 , ΔABR , ΔABR/NTA or ΔSH3 , described in the schematic representation of Cortactin , bottom panel ) . NTA: N-Terminal acidic domain; ABR: actin binding repeat; Helix: helical domain; SH3: Src-homology three domain . Lysates were subjected to pull-down assays using GST , GST-p27 or GST-p27CK- ( C ) , or GST , GST-p27 , GST-p27 NT ( 1-87 ) and GST-p27 CT ( 88-198 ) ( D ) , or GST , GST-p27CK- , GST-p27CK- ( 1-197 ) and GST-p27CK- ( 1-190 ) ( E ) or GST and GST-p27 ( F ) . The amounts of Myc-Cortactin bound to the beads and of transfected protein present in the extracts were detected by immunoblot using mouse anti c-Myc ( 9E10 ) antibodies . The amounts of GST or GST p27 and mutants used in the assays were visualized by Coomassie staining of the gels . ( A–F ) All panels show representative results of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 003 Similarly , pull-down assays using full-length GST-p27 were performed on HEK 293 cell lysates expressing various Myc-tagged Cortactin truncation mutants to map the p27 interaction domain on Cortactin ( Katsube et al . , 2004 ) . These experiments revealed that p27 interacts with the N-terminal half of Cortactin , within the actin binding repeats , as a mutant Cortactin lacking the actin binding repeats 3 to 6 ( ΔABR3-6 ) did not bind p27 ( Figure 1F ) . Since Cortactin is a key component of invadopodia , we determined if p27 could colocalize with Cortactin in these structures . To remove most of soluble p27 prior to fixation and immunostaining , p27+/+E6 MEFs were permeabilized with digitonin , as described previously ( Serres et al . , 2012 ) . In these conditions , p27 could be readily observed in invadopodia where it colocalized with Tks5 , a commonly used invadopodia marker ( Seals et al . , 2005 ) , or Cortactin ( Figure 2A and B , respectively ) . The same approach was used to confirm p27 localization to invadopodia in human A549 lung adenocarcinoma and A375 melanoma cell lines ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 22207 . 004Figure 2 . p27 colocalizes with Cortactin in invadopodia . ( A–B ) p27+/+ MEFs were seeded on gelatin-coated coverslips for 24 hr . Cells were permeabilized with digitonin prior to fixation . Cells were labeled with mouse anti p27 ( SX53G8 . 5 ) in red ( A–B ) and rabbit anti-Tks5 ( M-300 ) ( A ) or rabbit anti-Cortactin ( H-191 ) ( B ) in green . Images were acquired using a 60x objective and images displayed are cropped areas . Graphs displaying the fluorescence intensity ( arbitrary unit ) under the arrows in the enlarged panels were generated with the NIS Element software . Scale bars: 20 μm . ( C ) p27+/+ E6 MEFs were starved overnight in DMEM-0 . 1% FCS and then stimulated with growth medium for the indicated times . Cell lysates were subjected to immunoprecipitation using rabbit anti-p27 ( C–19 ) . Immunoprecipitated proteins and corresponding cell extracts were immunoblotted with rabbit anti-Cortactin ( H–191 ) and mouse anti-p27 ( F–8 ) antibodies . β-actin was used as loading control . The graph represents the mean fold change in amounts of Cortactin co-precipitated with p27 in each condition compared to time zero from five independent experiments . These differences were not statistically significant . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 00410 . 7554/eLife . 22207 . 005Figure 2—source data 1 . Quantification of co-immunoprecipitation between p27 and Cortactin in MEF E6 ( Figure 2C ) and HeLa cells ( Figure 2—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 00510 . 7554/eLife . 22207 . 006Figure 2—source data 2 . Statistical analyses for Figure 2C and Figure 2—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 00610 . 7554/eLife . 22207 . 007Figure 2—figure supplement 1 . p27 colocalizes with Cortactin and Tks5 in different tumor cell lines . ( A–C ) Twenty-four hr after seeding , A549 [A-B] or A375 cells [C] were permeabilized with digitonin prior to paraformaldehyde fixation . Cells were immunostained with mouse anti p27 ( SX53G8 . 5 ) ( red ) [A-C] and with rabbit anti-Cortactin ( H-191 ) [A] or rabbit anti-Tks5 ( M-300 ) [B-C] ( green ) antibodies . Graphs display the fluorescence intensity ( arbitrary unit ) in the green and red channels over the distance depicted by the arrow in each enlarged area . Images were acquired using a 60x objective and images displayed are cropped areas . Measurements were made with the NIS Element software ( Nikon ) . Scale bars: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 00710 . 7554/eLife . 22207 . 008Figure 2—figure supplement 2 . p27/Cortactin interaction in HeLa cells after EGF stimulation . ( A ) HeLa cells were starved with DMEM 0 . 1% FCS overnight , stimulated with 100 ng/ml EGF ( Peprotech ) for the indicated times and cell lysates were submitted to co-immunoprecipitation using rabbit anti-p27 ( C-19 ) antibodies . Co-immunoprecipitated proteins and total proteins in lysates were detected by immunoblot with anti-Cortactin and mouse anti-p27 ( F-8 ) antibodies . β-actin was used as loading control . ( B ) The graph represents the mean fold change of Cortactin co-precipitated with p27 compared to time zero in three independent experiments . These differences were not statistically significant . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 008 Invadopodium and podosome formation is stimulated by growth factors ( Murphy and Courtneidge , 2011 ) . Growth factor stimulation causes the degradation of p27 as cells enter the cell cycle but also the relocalization of a fraction of p27 in the cytoplasm between 1 hr and 6 hr post-stimulation , dependent on the phoshorylation of p27 on Ser10 ( Besson et al . , 2006 , 2004b; Boehm et al . , 2002; Ishida et al . , 2002; Rodier et al . , 2001; Connor et al . , 2003 ) . Consistent with the translocation of p27 in the cytoplasm and the formation of invadopodia after serum or growth factor stimulation , an increased association between p27 and Cortactin by co-immunoprecipitation at 1 hr and 3 hr post-stimulation in p27+/+E6 MEFs ( Figure 2C ) and in Hela cells ( Figure 2—figure supplement 2 ) was observed in five and three independent experiments , respectively . However , due to variability in signal intensities among independent experiments , these differences were not statistically significant . Together , our data indicate that a fraction of cytoplasmic p27 localizes to invadopodia after growth factor stimulation , where it binds to Cortactin . Invadopodia and podosomes are capable of both adhering to and degrading ECM ( Murphy and Courtneidge , 2011; Di Martino et al . , 2016 ) . Fibroblasts do not normally form invadosomes unless transformed by Src ( Murphy and Courtneidge , 2011 ) . We first verified that the E6 immortalized MEF lines used in our study were forming functional , matrix degrading invadopodia . For this , p27+/+ , p27−/− and p27CK-/CK- E6 MEFs were seeded on fluorescent gelatin and Tks5 immunostaining indicated that these cells could all form Tks5-containing structures that efficiently degraded gelatin ( Figure 3—figure supplement 1 ) . p27 colocalized with Tks5 at sites of gelatin degradation in MEFs and A549 cells , suggesting that p27 can be present at functional invadopodia ( Figure 3—figure supplement 2 ) . Since p27 binds to Cortactin and can localize to invadopodia , we determined if p27 status influenced the ability of cells to form invadopodia . Counting of the number of cells forming invadopodia using Tks5 immunostaining ( Figure 3A ) revealed that while cells expressing p27 or p27CK- rarely formed invadopodia ( 5 . 83% and 8 . 68% , respectively ) , nearly half of MEFs lacking p27 had invadopodia ( 44 . 66% ) ( Figure 3B ) . In keeping with the number of invadopodia forming cells , measurement of the area of fluorescent gelatin degraded per cell showed that p27−/− cells had a dramatically increased capacity to degrade ECM compared to p27+/+ and p27CK-/CK- MEFs ( Figure 3C and D ) . 10 . 7554/eLife . 22207 . 009Figure 3 . p27 regulates invadopodia formation and matrix degradation . ( A ) p27+/+ , p27CK−/CK− and p27−/− immortalized MEFs were seeded on Oregon green-gelatin ( gelatin-A488 ) for 16 hr . Cells were stained with rabbit anti-Tks5 ( M-300 ) to visualize invadopodia . ( B ) The percentage of cells forming invadopodia was determined in a minimum of 15 fields , representing a minimum of 330 cells per genotype , for each experiment . The graph shows the mean of 3 independent experiments . ( C ) Cells were seeded as in ( A ) . Tks5 staining shows invadopodia ( red ) and areas of degraded fluorescent gelatin indicate invadopodia activity ( green ) . ( D ) The areas of degraded gelatin were measured in at least 15 fields per genotype in each experiment . The graph shows the mean of 3 independent experiments . ( E–G ) p27−/− E6 MEFs were infected with either empty vector or p27 , p27CK− or p27CK− 1–190 vectors and then seeded on Gelatin-A488 for 48 hr . ( E ) p27 levels after retroviral infection were determined by immunoblot with rabbit anti-p27 ( N-20 ) antibodies; Grb2 was used as loading control . ( F–G ) After Tks5 staining , invadopodia forming cells ( F ) were quantified as in B and area of degraded gelatin ( G ) as in D . Scale bars: 50 μm; ‘ns’ not significant; ****p<0 . 0001 . In A and C , images were acquired using a 40x objective and images displayed are cropped areas . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 00910 . 7554/eLife . 22207 . 010Figure 3—source data 1 . Quantification of cells with invadopodia ( Figure 3B ) ; quantification of degraded gelatin area per cell ( Figure 3C ) ; quantification of cells with invadopodia after p27 re-expression ( Figure 3F ) and quantification of degraded gelatin area per cell after p27 re-expression ( Figure 3G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 01010 . 7554/eLife . 22207 . 011Figure 3—source data 2 . Statistical analyses for Figure 3B , C , F and G . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 01110 . 7554/eLife . 22207 . 012Figure 3—source data 3 . Immunoblot scans of Figure 3E . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 01210 . 7554/eLife . 22207 . 013Figure 3—figure supplement 1 . p27+/+ , p27CK−/CK− and p27−/− MEFs form functional invadopodia . HPV E6 immortalized p27+/+ , p27CK−/CK− and p27−/− MEFs were seeded on Gelatin-A488 for 16 hr . Cells were fixed in paraformaldehyde and stained with rabbit anti-Tks5 ( M-300; red ) antibodies to visualize invadopodia . Images were acquired using a 60x objective and images displayed are cropped areas . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 01310 . 7554/eLife . 22207 . 014Figure 3—figure supplement 2 . p27 colocalizes with Tks5 at sites of gelatin degradation . HPV E6 immortalized p27+/+ MEFs ( A ) or A549 lung adenocarcionma cells ( B ) were seeded on Gelatin-A488 and treated with 1 μM FRAX597 to stabilize invadopodia after 2 hr . After 72 hr , cells were permeabilized with digitonin , fixed in paraformaldehyde and stained with mouse anti-p27 ( SX53G8 . 5 , red ) and rabbit anti-Tks5 antibodies ( M-300; purple ) to visualize p27/Tks5 colocalization and sites of gelatin degradation . Images were acquired using a 60x objective and images displayed are cropped areas . Scale bars: 10 μm . Graphs display the fluorescence intensity ( arbitrary unit ) in each channel over the distance depicted by the arrows ( NIS Element software , Nikon ) . ( C–D ) Profiles of Cortactin and p27 localization outside of gelatin degradation area in p27+/+ MEFs ( C ) and A549 cells ( D ) . The graphs were generated by moving the arrows from panels A and B to regions without gelatin degradation . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 014 To confirm the involvement of p27 in regulating invadopodia number and proteolytic activity , we retrovirally infected p27−/− E6 MEFs with either empty vector , wild-type ( WT ) p27 , p27CK- or a p27CK- 1–190 truncation mutant ( Besson et al . , 2004a; Serres et al . , 2012 ) that cannot interact with Cortactin ( Figure 1E ) . Expression levels of p27 in these infected cells were determined by immunoblot ( Figure 3E ) . Quantification of the number of cells forming invadopodia and of the area of gelatin degraded per cell indicated that expression of WT p27 and p27CK- significantly decreased both the number of p27−/− cells forming invadopodia and their capacity to degrade gelatin , while the p27CK- 1–190 mutant had no effect ( Figure 3F and G ) . Thus , our data suggests that p27 limits invadopodia formation and that the domain mediating its interaction with Cortactin is required for this function . The finding that p27 knockout cells more frequently form invadopodia and exhibit increased ECM degradation activity was surprising given the previous reports that p27 promotes migration and invasion ( Denicourt et al . , 2007; Godin et al . , 2012; Nguyen et al . , 2006; Papakonstanti et al . , 2007; Gui et al . , 2014; Wu et al . , 2006; See et al . , 2010; Larrea et al . , 2009; Jin et al . , 2013; Zhao et al . , 2015; Besson et al . , 2004b ) . We compared the motility of p27+/+ , p27−/− and p27CK-/CK- E6 MEFs in 2D scratch wound assays and p27−/− cells had a migration defect ( Figure 4A and B ) , in agreement with our previous findings ( Besson et al . , 2004b ) . We next measured the capacity of these MEFs to invade through a layer of Collagen I in transwell inserts . Similar to 2D migration , 3D invasion was reduced in p27−/− MEFs compared to either WT p27 or p27CK- expressing cells ( Figure 4C and D ) , consistent with previous reports ( Denicourt et al . , 2007; Gui et al . , 2014; Wu et al . , 2006; See et al . , 2010; Jin et al . , 2013; Zhao et al . , 2015 ) . To confirm the involvement of p27 in regulating invasion , p27−/− E6 MEFs retrovirally infected with either empty vector , p27CK- or p27CK- 1–190 ( Figure 4E , left panel ) were subjected to transwell invasion assays . While p27CK- restored invasion of p27−/− cells , the p27CK- 1–190 mutant had no effect ( Figure 4E , right panel ) . Taken together our results indicate that although p27 limits the number of invadopodia in MEFs , it also promotes cellular invasion through ECM . On the other hand , while p27-null cells more frequently form invadopodia and exhibit an increased capacity to degrade ECM , their ability to invade through matrix was significantly impaired . 10 . 7554/eLife . 22207 . 015Figure 4 . p27 promotes cell migration and invasion . ( A ) Representative images of scratch wound migration assays with p27+/+ , p27CK−/CK− and p27−/− immortalized MEFs . Dark grey areas show the initial wound masks and dotted lines the migration fronts . Scale bars: 300 μm . ( B ) Mean cell migration at 12 hr and 24 hr post wounding for each genotype of five independent experiments . Percent of area in which the cells migrated , or wound closing ( relative wound density ) was calculated with the Incucyte software . ‘ns’: not significant; ****p<0 . 0001 . ( C ) Representative images of p27+/+ , p27CK−/CK− and p27−/− immortalized MEFs that invaded through a layer of Collagen I in transwell invasion assays and migrated to the bottom side of the transwell membrane after 48 hr . Scale bars: 100 μm . ( D ) The graph shows the mean number of cells that invaded through Collagen I quantified by XTT staining , expressed relative to p27+/+ cells , of three independent experiments . **p<0 . 01 . ( E ) p27−/− E6 MEFs were infected with either empty vector , p27CK- or p27CK- 1–190 vectors and used in transwell invasion assays as in ( C–D ) . p27 levels after retroviral infection were determined by immunoblot with mouse anti-p27 ( SX53G8 . 5 ) ; β-actin was used as loading control . The graph shows the mean number of cells that invaded through Collagen I quantified by XTT staining , expressed relative to p27−/− cells , of four independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 01510 . 7554/eLife . 22207 . 016Figure 4—source data 1 . quantification of relative wound density ( Figure 4B ) ; quantification of invasion ( Figure 4D ) ; and quantification of invasion rescue by p27 re-expression ( Figure 4E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 01610 . 7554/eLife . 22207 . 017Figure 4—source data 2 . Statistical analyses for Figure 4B , D and E . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 017 Interestingly , a similar , seemingly paradoxical finding was recently reported , in which inhibition of Rac1 , Trio or PAK1 with siRNAs caused a dramatic increase in invadopodia lifetime and in their capacity to degrade matrix , but this was accompanied by a sharp decrease in cell invasion ( Moshfegh et al . , 2014 ) . In this study , activation of the Trio/Rac1/PAK1 pathway induced a putative PAK1-mediated phosphorylation of Cortactin and invadopodia disassembly , as expression of a S113A Cortactin mutant blocked this pathway ( Moshfegh et al . , 2014 ) . In this study , altering the dynamics of invadopodia turnover inhibited invasion ( Moshfegh et al . , 2014 ) . To find out whether p27-null cells more frequently form invadopodia than p27+/+ cells ( Figure 3A and B ) due to an increase in their lifetime , we infected p27+/+ and p27−/− MEFs with Tks5-GFP to visualize invadopodia in live cells . Videomicroscopy analyses revealed that while invadopodia lifetime was on average 16 . 7 min in p27+/+ cells , their mean duration was 65 . 9 min in p27−/− cells ( Figure 5A ) . Thus , invadopodia appear more stable in cells lacking p27 . 10 . 7554/eLife . 22207 . 018Figure 5 . p27 promotes binding of Cortactin to PAK1 . ( A ) Live p27+/+ and p27−/− immortalized MEFs expressing eGFP-Tks5 were imaged by videomicroscopy to measure invadopodia lifetime , using Tks5 as an invadopodia marker . The graph represents the average invadopodia lifetime of 20 invadopodia per genotype per experiment from three independent experiments . ( B–C ) Co-immunoprecipitations using rabbit anti-PAK1 ( N-20 ) of HeLa cell lysates transfected with empty vector or p27 encoding vector ( B ) or p27+/+ and p27−/− E6 MEF lysates ( C ) . Co-immunoprecipitated Cortactin was detected with mouse ( B ) or rabbit ( C ) anti-Cortactin antibodies . Immunoprecipitated PAK1 was visualized by reprobing the membranes with rabbit anti-PAK1 ( N-20 ) and anti-Rabbit Ig light-chain secondary antibodies . Immunoblots of extracts show the level of proteins in each condition . In ( C ) , the graph shows the mean fold change in the ratio of Cortactin to PAK1 co-immunoprecipitated , expressed relative to p27+/+ cells , in three independent experiments . **p<0 . 01 . ( D ) p27+/+E6 MEFs were starved overnight in DMEM 0 . 1% FCS and then stimulated with growth medium for the indicated times . PAK1 was immunoprecipitated from cell lysates using rabbit anti-PAK1 ( N-20 ) . Immunoblots of immunoprecipitates ( left panels ) and extracts ( right panels ) were probed successively with rabbit anti-Cortactin ( H-191 ) and anti-rabbit Ig light chain secondary antibodies and then with mouse anti-PAK1 ( A6 ) antibodies . β-actin was used as loading control . ( E ) The graph shows the mean amount of Cortactin bound to PAK1 at each time-point from four independent experiments , normalized to time zero . These differences were not statistically significant . ( F ) p27+/+ MEFs were seeded on coverslips and permeabilized with digitonin prior to fixation . Cells were labeled with rabbit anti-PAK1 ( N-20 , green ) and mouse anti p27 ( SX53G8 . 5 , red ) antibodies . Images were acquired using a 60x objective and images displayed are cropped areas . The graph displaying the fluorescence intensity ( arbitrary unit ) under the arrow in the enlarged panel was generated with NIS Element software . Scale bars: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 01810 . 7554/eLife . 22207 . 019Figure 5—source data 1 . Quantification of invadopodia lifetime ( Figure 5A ) ; quantification of co-immunoprecipitation between Cortactin and PAK1 in MEFs ( Figure 5C ) ; and quantification of co-immunoprecipitation between Cortactin and PAK1 after serum stimulation ( Figure 5E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 01910 . 7554/eLife . 22207 . 020Figure 5—source data 2 . Statistical analyses for Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 02010 . 7554/eLife . 22207 . 021Figure 5—source data 3 . Statistical analyses for Figure 5C . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 02110 . 7554/eLife . 22207 . 022Figure 5—source data 4 . Statistical analyses for Figure 5E . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 02210 . 7554/eLife . 22207 . 023Figure 5—figure supplement 1 . p27 does not affect Cortactin acetylation or the recruitment of c-Src , Arp2 and ERK1 to Cortactin . ( A ) HEK 293 cells were transfected with either empty vector , p27 or p27CK- coding vectors . Lysates were used for immunoblot with rabbit anti-Acetyl-Cortactin ( BD-Transduction Laboratories ) , anti-Cortactin ( H-191 ) and anti-p27 ( C-19 ) antibodies . ( B–D ) Co-immunoprecipitations were performed using rabbit anti-Cortactin ( H-191 ) in HeLa cell lysates transfected with either empty vector or p27 coding vector . Co-immunoprecipitated proteins were detected with rabbit anti-Arp2 ( H-84 ) ( B ) , anti-c-Src ( SRC2 ) ( C ) or anti-ERK1 ( K-23 ) ( D ) . The immunoprecipitated proteins were visualized by reprobing the membrane with rabbit anti-Cortactin ( H-191 ) and anti Rabbit Ig light-chain secondary antibodies . Immunoblots of extracts show the levels of proteins of interest in each condition . Experiments were performed at least three times , except only twice for [D] . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 023 To determine whether p27 affects invadopodia formation through the regulation of Cortactin , we tested if binding of p27 to Cortactin regulated the interaction of Cortactin with some of its partners . p27 overexpression in Hela cells had no effect on the interaction of Cortactin with c-Src , ERK1 or Arp2 ( Figure 5—figure supplement 1 ) ( Kirkbride et al . , 2011 ) . Also , Cortactin acetylation levels did not change when p27 was overexpressed ( Figure 5—figure supplement 1 ) , suggesting that p27 did not affect the interaction of Cortactin with acetyltransferases such as p300 , Tip60 and ATAT1 or the deacetylases HDAC6 and Sirt1 ( Sun et al . , 2015; Zhang et al . , 2007 , 2009; Castro-Castro et al . , 2012 ) . In contrast , the presence of p27 increased the amount of Cortactin co-immunoprecipitated with PAK1 in Hela cells ( Figure 5B ) . Conversely , in MEFs lacking p27 , the amount of Cortactin co-precipitated with PAK1 was sharply decreased compared to p27+/+ cells , despite the fact that PAK1 levels were elevated in p27−/− cells ( Figure 5C ) . Monitoring the kinetics of association of PAK1 with Cortactin in MEFs revealed that some PAK1 is already associated with Cortactin in serum starved cells , consistent with a previous report ( Vidal et al . , 2002 ) and progressively increase after serum stimulation , reaching a maximal level after 3 hr ( Figure 5D and E ) ; mirroring the kinetics of association of p27 with Cortactin ( Figure 2C and Figure 2—figure supplement 2 ) . These results suggest that p27 binding to Cortactin promotes the association of PAK1 with Cortactin . Immunostaining of p27+/+ MEFs showed a colocalization of PAK1 and p27 at structures resembling invadopodia ( Figure 5F ) . However , despite repeated attempts on both endogenous and overexpressed proteins , we were unable to obtain evidence supporting an interaction between p27 and PAK1 ( data not shown ) . p27 promotes the interaction of PAK1 with Cortactin ( Figure 5A–D ) , which in turn may promote invadopodia disassembly by phosphorylating Cortactin on Ser113 ( Moshfegh et al . , 2014 ) . To test whether PAK1 activity is responsible for the low number of invadopodia observed in p27+/+ MEFs , we monitored the effect of either silencing PAK1 with siRNA or of the PAK1-3 specific inhibitors FRAX597 ( Licciulli et al . , 2013 ) , FRAX1036 ( Ong et al . , 2015 ) and G-5555 ( Ndubaku et al . , 2015 ) on invadopodia formation and activity . While PAK1 silencing or the PAK1-3 inhibitors dramatically increased both the number of cells forming invadopodia and the area of matrix degraded per cell in p27+/+ MEFs , as reported previously ( Moshfegh et al . , 2014 ) , it had no effect in p27−/− cells ( Figure 6A–B , D–F , and Figure 6—figure supplement 1A and B ) . PAK1 silencing was verified by immunoblot for PAK1 in Control or PAK1 siRNA treated cells ( Figure 6C ) . The efficacy of FRAX597 , FRAX1036 and G-5555 in inhibiting PAK1 was evaluated by immunoblot on either vehicle or inhibitor treated cells using phospho-Ser144 PAK1/Ser141-PAK2 antibodies ( Figure 6G and Figure 6—figure supplement 1C ) ( Chong et al . , 2001 ) . These results suggest that p27 acts downstream of PAK1 in the regulation of invadopodia formation and activity , and that this pathway is impaired in p27−/− cells , possibly due to decreased PAK1/Cortactin interaction . 10 . 7554/eLife . 22207 . 024Figure 6 . p27 regulates invadopodia formation downstream of PAK1 . ( A–B ) p27+/+ and p27−/− E6 MEFs were seeded on gelatin-A488 and transfected with either control siRNAs or PAK1 siRNAs for 48 hr . Cells were stained with rabbit anti-Tks5 ( M-300 ) to visualize invadopodia . At least ten fields per conditions were used to count cells forming invadopodia ( A ) , representing a minimum of 212 cells per genotype for each experiment , or to measure the area of degraded gelatin , expressed in fold-change compared to vehicle treated conditions ( B ) . ( C ) PAK1 siRNA efficacy at 48 hr was evaluated by immunoblot with rabbit anti-PAK1 ( N-20 ) antibodies . β-actin was used as loading control . ( D–F ) p27+/+ and p27−/− E6 MEFs were seeded for 48 hr on gelatin-A488 and after 1 hr were treated with DMSO or 1 μM FRAX597 , a PAK1-3 inhibitor . Quantification of cells forming invadopodia ( D ) and gelatin degradation ( E ) was performed as in ( A–B ) , with a minimum of 215 cells counted per genotype per experiment . ( F ) Images were acquired using a 40x objective and images displayed are cropped areas . Scale bars: 50 μm . ( G ) FRAX597 inhibitor efficacy was evaluated by immunoblot with rabbit anti-phospho-Ser144-PAK1/phospho-Ser141-PAK2 and rabbit anti-PAK1 ( N-20 ) antibodies . ( A–B; D–E ) The graphs show the mean of at least three independent experiments . ****p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 02410 . 7554/eLife . 22207 . 025Figure 6—source data 1 . Quantification of invadopodia forming cells ( Figure 6A ) and degraded gelatin area ( Figure 6B ) after PAK1 silencing; quantification of invadopodia forming cells ( Figure 6D ) and degraded gelatin area ( Figure 6E ) after FRAX597 treatment; quantification of invadopodia forming cells ( Figure 6—figure supplement 1A ) and degraded gelatin area ( Figure 6—figure supplement 1B ) after FRAX1036 and G-5555 treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 02510 . 7554/eLife . 22207 . 026Figure 6—source data 2 . statistical analyses for Figure 6A , B , D and E and Figure 6—figure supplement 1A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 02610 . 7554/eLife . 22207 . 027Figure 6—figure supplement 1 . p27 regulates invadopodia formation in a PAK1 dependent manner . ( A–B ) p27+/+ or p27−/− E6 MEFs were seeded for 48 hr on Gelatin-A488 and treated 1 hr after seeding with either DMSO , 1 μM FRAX1036 or 1 μM G-5555 , two PAK1-3 inhibitors . Cells were stained with rabbit anti-Tks5 ( M-300 ) to visualize invadopodia . The graphs show the mean of 3 independent experiments . For each experiment , at least ten fields per conditions were used to count cells forming invadopodia , representing a minimum of 168 cells per genotype ( A ) , or to measure the area of degraded gelatin , expressed in fold-change compared to vehicle treated conditions ( B ) . ( C ) The efficacy of PAK inhibitors was evaluated by immunoblot with rabbit anti-phospho-Ser144-PAK1/phospho-Ser141-PAK2 and Rabbit anti-PAK1 antibodies . Reprobing of membranes with a mouse anti-β actin antibody was used to control protein loading . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 027 If p27 acts on the invadopodia disassembly pathway at the level of PAK1/Cortactin , one would not expect to see any difference in Rac1 activation levels in p27+/+ and p27−/− cells . This is what was observed by GTP-Rac1 pull-down assays ( Figure 7A ) , in agreement with our previous observations ( Besson et al . , 2004b ) . Nevertheless , to further characterize how the Rac1/PAK1/Cortactin signaling pathway is affected in function of p27 status , we inhibited Rac1 itself using either Rac1 siRNAs ( Figure 7B–D ) or the Rac specific inhibitor NSC23766 ( Gao et al . , 2004 ) ( Figure 7E and F ) . Rac1 silencing by siRNAs was verified by immunoblot for Rac1 ( Figure 7D ) . When Rac1 was silenced or inhibited , more p27+/+ cells formed invadopodia and there was a dramatic increase in the area of matrix degraded ( Figure 7B , E and C , F , respectively ) , as reported previously ( Moshfegh et al . , 2014 ) . On the other hand , in p27−/− cells , Rac1 silencing or inhibition had no effect on invadopodia formation or gelatin degradation ( Figure 7B , E and C , F , respectively ) ; supporting the idea that p27 regulates invadopodia disassembly downstream of Rac1 and PAK1 . 10 . 7554/eLife . 22207 . 028Figure 7 . p27 regulates invadopodia formation downstream of Rac1 . ( A ) Cells were seeded for 24 hr and Rac1-GTP levels measured by GTP pull-downs assays using GST-PAK1-CD beads . The amounts of Rac1-GTP bound to the beads and of total Rac1 in the extracts were detected by immunoblot with mouse anti-Rac1 . The graph shows the mean ratio of GTP-Rac1/total Rac1 from six independent experiments . ( B–D ) Cells were transfected with control ( ctl ) or Rac1 #1 or #2 siRNAs for 3 days . Cells were then seeded on Gelatin-A488 for 48 hr and for monitoring siRNA efficiency . Cells were stained with rabbit anti-Tks5 ( M-300 ) to visualize invadopodia . At least ten fields per condition were used to count cells forming invadopodia , representing a minimum of 197 cells per genotype for each experiment ( B ) or to measure the area of degraded gelatin , expressed in fold-change compared to control siRNA treated conditions ( C ) . ( D ) Rac1 silencing was evaluated by immunoblot with mouse anti-Rac1 . β-actin was used as loading control . ( E–F ) Cells were processed as described in Figure 6D and E except that the Rac1 inhibitor NSC23766 at 100 μM was used instead of FRAX597 . A minimum of 153 cells were counted per genotype for each experiment . ( B–C; E–F ) The graphs show the mean of at least three independent experiments . ****p<0 . 0001; **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 02810 . 7554/eLife . 22207 . 029Figure 7—source data 1 . Quantification of Rac1 GTP/Rac1 levels ( Figure 7A ) ; quantification of invadopodia forming cells ( Figure 7B ) and degraded gelatin area ( Figure 7C ) after silencing of Rac1; quantification of invadopodia forming cells ( Figure 7E ) and degraded gelatin area ( Figure 7F ) after NSC23766 treatment; quantification of invadopodia forming cells ( Figure 7—figure supplement 1A ) and degraded gelatin area ( Figure 7—figure supplement 1B ) after RhoA silencing; and quantification of invasion after Y27632 treatment ( Figure 7—figure supplement 1D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 02910 . 7554/eLife . 22207 . 030Figure 7—source data 2 . Statistical analyses for Figure 7A , B , C , E , F , and Figure 7—figure supplement 1A , B and D . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 03010 . 7554/eLife . 22207 . 031Figure 7—figure supplement 1 . RhoA regulation by p27 is not involved in invadopodia formation . ( A–C ) p27+/+ and p27−/− immortalized MEFs were transfected with siRNA control ( ctl ) or siRNA RhoA #1 or #2 . After 3 days , cells were seeded on Gelatin-A488 for matrix degradation and for controlling siRNA efficiencies , for 48 hr . Cells were stained with rabbit anti-Tks5 ( M-300 ) to visualize invadopodia . For each experiment , ten fields per condition , representing a minimum of 256 cells per genotype , were used to count invadopodia forming cells ( A ) or the area of degraded gelatin per cell ( B ) . The graphs show the means of at least three independent experiments . ( C ) siRNA efficiency was evaluated by immunoblot with mouse anti-RhoA ( 26C4 ) antibodies . β-actin was used as loading control . ( D ) p27+/+ and p27−/− immortalized MEFs were subjected to invasion assays as in Figure 4C–D in presence or absence of the ROCK inhibitor Y27632 ( 10 μM ) . The graph shows the fold change in invasion relative to vehicle treated cells ( - ) from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 031 We previously reported that p27 binds RhoA and prevents its activation by its GEFs; consequently , cells lacking p27 have elevated RhoA activity ( Besson et al . , 2004b ) . Since RhoA plays a critical role in cell motility and the control of actin cytoskeleton dynamics , we determined whether RhoA was involved in p27-mediated regulation of invadopodia formation and function . Transfection of RhoA siRNAs in p27+/+ and p27−/− MEFs did not alter the proportion of cells forming invadopodia ( Figure 7—figure supplement 1A ) and only mildly increased the ability of both p27+/+ and p27−/− cells to degrade ECM , although this was not statistically significant ( Figure 7—figure supplement 1B ) . RhoA siRNA efficacy was checked by immunoblot ( Figure 7—figure supplement 1C ) . Furthermore , inhibition of the Rho effectors ROCK1/2 with a pharmacological inhibitor ( Y27632 ) did not affect the ability of p27+/+ and p27−/− cells to invade through Collagen I in transwell invasion assays ( Figure 7—figure supplement 1D ) . Thus , in MEFs , depletion of RhoA did not affect the ability of cells to form invadopodia and slightly increased invadopodia proteolytic activity independently of p27 status . Evidence suggests that invadopodia disassembly by the Trio/Rac1/PAK1 signaling pathway is mediated by phosphorylation of Cortactin on Ser113 ( Webb et al . , 2005; Moshfegh et al . , 2014 ) . In the latter study , expression of an unphosphorylatable mutant of Cortactin ( Cortactin S113A ) increased both invadopodia lifetime and matrix degradation , as observed upon Rac1 or PAK1 silencing ( Moshfegh et al . , 2014 ) . We retrovirally infected p27+/+ and p27−/− E6 MEFs with either empty vector , WT Cortactin , Cortactin S113A or the phosphomimetic mutant Cortactin S113D ( Figure 8A ) and monitored their effect on invadopodia formation and activity . In p27+/+ cells , Cortactin S113A increased the number of cells forming invadopodia , as observed previously ( Moshfegh et al . , 2014 ) , while Cortactin S113D had no effect ( Figure 8B and D ) . In contrast , Cortactin S113A had no effect in p27−/− cells , confirming that the Rac1/PAK1/Cortactin pathway is defective in absence of p27 and that p27 acts upstream of Cortactin . On the other hand , expression of Cortactin S113D in p27−/− cells decreased the number of invadopodia forming cells and the area of degraded gelatin ( Figure 8C and E ) , indicating that mimicking Cortactin phosphorylation rescues the phenotype caused by the absence of p27 . 10 . 7554/eLife . 22207 . 032Figure 8 . Mimicking phosphorylation of Cortactin restores invadopodia dynamics in p27−/− cells . ( A–E ) p27−/− E6 MEFs were infected with empty vector or with vectors encoding wild type Cortactin ( WT ) , S113A-Cortactin ( S113A ) or S113D-Cortactin ( S113D ) . ( A ) Cortactin levels after retroviral infection were determined by immunoblot with rabbit anti-Cortactin ( H-191 ) antibodies . β-actin was used as loading control . ( B–E ) Cells were seeded on gelatin-A488 for 48 hr . After Tks5 staining , cells forming invadopodia ( B–C ) , or the area of degraded gelatin , expressed in fold-change compared to WT Cortactin transfected cells ( D–E ) , were quantified in at least ten fields per condition in each experiment , representing a minimum of 179 cells per genotype . The graphs show the means of at least three independent experiments . ( F–J ) p27−/− E6 MEFs were infected with empty vector or with vectors encoding WT Cortactin , Cortactin TA ( S113A/S150A/S282A ) or Cortactin TD ( S113D/S150D/S282D ) . ( J ) Cortactin levels after retroviral infection were determined as in ( A ) . ( G–J ) Cells were processed as in ( B–E ) to quantify cells forming invadopodia ( G–H ) , or the area of degraded gelatin ( I–J ) , with a minimum of 222 cells counted per genotype per experiment . The graphs show the means of 3 independent experiments . ( K ) Schematic representation of the Rac1/PAK1/phospho-Cortactin pathway involved in invadopodia turnover and matrix degradation and its proposed regulation by p27 . ****p<0 . 0001; ***p<0 . 001; **p<0 . 01; *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 03210 . 7554/eLife . 22207 . 033Figure 8—source data 1 . Quantification of cells forming invadopodia ( Figure 8B–C ) and degraded gelatin area ( Figure 8D–E ) after infection with S113 phospho-mutants of Cortactin; quantification of cells forming invadopodia ( Figure 8G–H ) and degraded gelatin area ( Figure 8I–J ) after infection with triple phospho-mutants of Cortactin; quantification of P-Ser Cortactin/Cortactin ratio ( Figure 8—figure supplement 1B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 03310 . 7554/eLife . 22207 . 034Figure 8—source data 2 . Statistical analyses for Figure 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 03410 . 7554/eLife . 22207 . 035Figure 8—source data 3 . Statistical analyses for Figure 8—figure supplement 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 03510 . 7554/eLife . 22207 . 036Figure 8—source data 4 . Mascot search results for Cortactin for Figure 8—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 03610 . 7554/eLife . 22207 . 037Figure 8—figure supplement 1 . Cortactin is phosphorylated on S113/S150 and/or S282 in vivo . ( A ) HEK 293 cells were tranfected with Myc-tagged Cortactin for 24 hr or left untransfected ( U ) and 3 μM FRAX597 was added for 12 hr where indicated . Cortactin was immunoprecipitated with anti-Myc antibodies ( 9E10 ) and resolved on SDS-PAGE and transferred on PVDF membranes . The amount of Ser-phosphorylated Cortactin was detected by immunoblot with a monoclonal antibody against phospho-serine ( BD ) and the total amount of immunoprecipitated Cortactin by reprobing with anti-Cortactin ( H191 ) antibody . P-Ser and Cortactin band intensities were quantified with Image J to calculate ratios , normalized to one for the control condition . Extracts were probed for Cortactin and Grb2 was used as loading control . ( B ) Average ratio of P-Ser signal/Total Cortactin signals from three independent experiments . Error bars are s . e . m . **p<0 . 01 . ( C ) HEK 293 cells were tranfected with either empty vector , wild-type Cortactin ( WT ) , S113A Cortactin or S113A/S150A/S282A ( TA ) Cortactin for 24 hr . Cortactin was immunoprecipitated with anti-Cortactin ( H191 ) antibodies and processed as in ( A ) . Serine phosphorylation of Cortactin was evaluated with anti-Phospho-Ser antibodies ( BD ) . This data is representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 03710 . 7554/eLife . 22207 . 038Figure 8—figure supplement 2 . Cortactin is phosphorylated on S150 in vivo upon PAK activation . MS/MS analysis of Cortactin immunoprecipitated from HEK 293 cells transfected either with Myc-Cortactin or Myc-Cortactin and Myr-SH3-2 , the second SH3 domain of Nck1 fused to the myristoylation signal of Src , which activate PAK kinase . S150 phosphorylation was only detected from lysates in which PAK activtiy was stimulated by the presence of the Src-SH3-2 domain . ( A ) MS/MS spectrum annotation of the residues 148–161 from Uniprot database reference SRC8_HUMAN or Q14247 ( Src substrate Cortactin ) protein . MS-Fragmented peptide correspond to the m/z selection of 516 . 888 , matching [HASQKDYSSGFGGK +80 Da ( HPO3 ) ]3+ among 691 spectrum matching 98 distinct sequences of the same protein . Measured fragment matching masses bearing the modified serine residue showing a loss of neutral mass ( H3PO4 ) are indicated in yellow . Not all the annotations could fit in the figure . The table below gathers all the matching fragments . ( B ) Table of theoretical fragment masses for the residues 148–161 of Uniprot database reference SRC8_HUMAN or Q14247 ( Src substrate Cortactin ) protein . In red are experimental masses of m/z detection of 516 . 888 matching [HApSQKDYSSGFGGK]3+ from the above spectrum . ( C ) Table of possible matches of the same set of experimental data with the four possible and closest theoretical sequences . The best score matches the phosphorylation of the serine 150 among the residues 148–161 of Uniprot database reference SRC8_HUMAN or Q14247 ( Src substrate Cortactin ) protein . Site analysis percentage corresponds to the probability output from the PhosphoRS algorythm . DOI: http://dx . doi . org/10 . 7554/eLife . 22207 . 038 Importantly , PAK3 was reported to phosphorylate Cortactin on several serine residues ( S113 , S150 and S282 ) within the actin binding repeats ( Webb et al . , 2006 ) suggesting that multiple phosphorylation events may synergize to regulate Cortactin function . To test this possibility , we generated triple alanine mutant ( TA ) ( S113A/S150A/S282A ) and triple aspartic acid mutant ( TD ) ( S113D/S150D/S282D ) of Cortactin and expressed these constructs in p27+/+ and p27−/− E6 MEFs ( Figure 8F ) . Cortactin TA and TD mutants gave essentially similar results as the Cortactin S113A and S113D mutants , respectively ( Figure 8G–J ) . Nevertheless , in p27+/+ MEFs , the Cortactin TA mutant was more potent than Cortactin S113A to promote invadopodia formation ( 13 . 5% p27+/+ cells forming invadopodia with Cortactin S113A versus 23 . 3% with Cortactin TA ) and activity ( 6 . 17 fold increase in area of degraded gelatin with Cortactin S113A versus 10 . 02 fold with Cortactin TA ) ( Figure 8B versus Figure 8G and Figure 8D versus Figure 8I , respectively ) . This data suggests that Cortactin phosphorylation on S150 and/or S282 also contribute with S113 to control invadopodia disassembly . To further confirm that PAK-mediated phosphorylation events on Cortactin are involved in this pathway , we evaluated Cortactin phosphorylation levels in HEK 293 cells expressing Myc-tagged Cortactin treated or not with the PAK1-3 inhibitor FRAX597 . In presence of FRAX597 , there was a 36% decrease in P-Ser Cortactin level ( Figure 8—figure supplement 1A and B ) . This mild reduction in P-Ser levels is consistent with Cortactin also being a substrate for other Ser/Thr kinases such as ERK , Akt and PKC . The contribution of S113 , S150 and S282 phosphorylation in the P-Ser signal observed in Cortactin immunoprecipitates was estimated in HEK 293 cells expressing either wild-type Cortactin , Cortactin S113A or Cortactin TA . P-Ser levels were reduced in Cortactin S113A immunoprecipitations compared to WT Cortactin and further decreased in Cortactin TA immunoprecipitates , confirming that phosphorylation of these sites contribute to the Cortactin P-Ser signal in vivo ( Figure 8—figure supplement 1C ) . Finally , to show that Cortactin was phosphorylated in vivo on PAK-targeted sites ( S113 , S150 and/or S282 ) , we performed mass spectrometry analyses on Cortactin . For this , HEK 293 cells were transfected with Myc-Cortactin alone or Myc-Cortactin and the myristoylated second SH3 domain of Nck1 ( Myr-SH3-2 ) , which was previously shown to activate PAK1 ( Lu et al . , 1997; Lu and Mayer , 1999 ) . MS/MS analyses from Cortactin immunoprecipitates indicated that Cortactin was phosphorylated on S150 ( Figure 8—figure supplement 2 ) . This phosphorylation was detected only in samples from cells co-expressing Cortactin and Myr-SH3-2 suggesting that PAK1 activation is required for this event . Although no phosphorylation was detected on S113 or S282 in these experiments , phosphoproteome analyses from other studies have previously shown that Cortactin is phosphorylated in vivo on S113 , S150 and/or S282: S113 phosphorylation was detected in prostate cancer cells ( Chen et al . , 2010 ) , mouse brain ( Wiśniewski et al . , 2010 ) and breast cancers ( Mertins et al . , 2016 ) . Phosphorylation on S150 was identified in mouse renal cells and human liver ( Bian et al . , 2014; Rinschen et al . , 2010 ) and in the present study . S282 phosphorylation was detected in mitotic cells and breast cancers ( Mertins et al . , 2016; Olsen et al . , 2010; Klammer et al . , 2012 ) . Phosphorylation on all three sites have been detected in ovarian and breast cancer xenogratfs ( Mertins et al . , 2014 ) . Altogether , our results indicate that p27 promotes invadopodia turnover and cell invasion via the Rac1/PAK1/Phospho-Cortactin pathway . In contrast , in absence of p27 , this pathway is largely defective , resulting in increased invadopodia stability and elevated matrix degradation , but impaired invasion .
Our results indicate that p27 regulates invadopodia formation and activity by binding to Cortactin , probably via the regulation of invadopodia turnover mediated by a signaling pathway recently characterized whereby signaling through Trio/Rac1/PAK1/Cortactin induces the disassembly of invadopodia ( Moshfegh et al . , 2014 ) . The main finding presented here is that p27 directly controls cellular invasion by impinging on this pathway through the promotion of the Cortactin/PAK1 interaction and the phosphorylation of Cortactin on Ser113 , S150 and/or S282 by PAK . Indeed , inhibiting the pathway upstream of Cortactin , either with Rac1 and PAK1 inhibitors or siRNAs converted the phenotype of p27+/+ cells to that of p27−/− cells . Conversely , artificially activating this pathway at the level of Cortactin using phosphomimetic mutants for the PAK-targeted sites efficiently rescued the phenotype of p27-null cells . All three phosphorylation sites on Cortactin appear to act synergistically as the phenotype caused by the triple mutants ( Cortactin TA and TD ) was more profound than that of mutating S113 alone . While mimicking the phosphorylation of Cortactin on PAK1-targeted sites clearly causes the disassembly of invadopodia ( Moshfegh et al . , 2014; Webb et al . , 2005 ) , the mechanism involved is still unclear but may be due to decreased affinity of Cortactin phosphorylated within its Actin-binding repeats for F-Actin , which may destabilize invadopodia ( Webb et al . , 2006 ) . Results from Moshfegh et al . ( 2014 ) and the present study show that inhibition of the Trio/Rac1/PAK1/Phospho-Cortactin pathway stabilizes invadopodia and is associated with a dramatic increase in proteolytic degradation of ECM . Surprisingly , this phenotype is accompanied by a sharp decrease in the ability of cells to invade through matrix in 3D invasion assays . Thus , it appears that the turnover of invadopodia rather than elevated ECM degradation is required for efficient invasion . Affecting the dynamics of invadopodia formation , maturation and disassembly is therefore detrimental to invasion , as observed here when this last step is inhibited by interfering with Rac1/PAK1 signaling or when p27 is absent . Interestingly , p27 also regulates cell migration in a similar manner by interfering with RhoA activation by its GEFs ( Besson et al . , 2004a , 2004b ) . In absence of p27 , cells accumulate actins stress fibers and focal adhesions and exhibit a migration defect due to an imbalance in the dynamic cycles of RhoA activation and inactivation ( and concomitant Rac1 inactivation and activation ) required for F-actin turnover , focal adhesion disassembly and efficient migration ( Besson et al . , 2004a , 2004b; Ren et al . , 2000; Arthur and Burridge , 2001; Cox et al . , 2001; Sahai et al . , 2001; Vial et al . , 2003 ) . Migration and invasion are two intimately linked processes ( Friedl and Wolf , 2003; Petrie and Yamada , 2016 ) . It is quite remarkable that p27 appears to regulate both of these processes , although by distinct mechanisms . Indeed , p27 binds to RhoA and Cortactin via the same domain located within the last 8 aa of the protein ( Figure 1E ) ( Larrea et al . , 2009; Godin et al . , 2012 ) , but while RhoA preferentially binds p27 phosphorylated on T198 ( Larrea et al . , 2009 ) , the interaction of p27 with Cortactin did not require the presence of T198 ( Figure 1E ) , suggesting that different pools of the protein are involved in their regulation . In addition , the regulation of invadopodia by p27 did not appear to involve RhoA , as RhoA silencing affected both p27−/− and p27+/+ cells in a similar manner ( Figure 7—figure supplement 1 ) . p27 also regulates invasion by promoting EMT , notably through JAK2/STAT3 signaling and induction of the transcription factor Twist ( Zhao et al . , 2015 ) . We found that p27 binds to Cortactin on its actin-binding repeats and this interaction promoted the association of Cortactin with PAK1 . Since we were unable to detect an interaction between p27 and PAK1 using either endogenous or overexpressed proteins , it remains unknown how p27 promotes the PAK1/Cortactin interaction . Evidence suggest that Cortactin is an intrinsically disordered protein ( IDP ) , especially its actin-binding repeats ( Shvetsov et al . , 2009 ) and p27 is a well-known IDP ( Galea et al . , 2008; Lacy et al . , 2004; Ou et al . , 2012 ) . Since these proteins typically fold upon binding to their partners , an attractive possibility is that p27 binding causes a conformation change in Cortactin that would favor its interaction with PAK1 . p27 is a cyclin-CDK inhibitor and although p27 regulates migration and invasion in a CDK-independent manner ( Besson et al . , 2004b; Larrea et al . , 2009; Zhao et al . , 2015; Gui et al . , 2014 ) , there is evidence to suggest that some cell cycle independent roles of CDKs and cyclins may be mediated via CDK inhibitors such as p27 ( Hydbring et al . , 2016 ) . For instance , the regulation of cell migration by Cyclin D1 requires p27 ( Li et al . , 2006 ) and the role of CDK5 in promoting neuronal progenitor migration in vivo is mediated via the phosphorylation of p27 on Ser10 which regulates the RhoA/ROCK/LIMK/Cofilin pathway ( Nguyen et al . , 2006; Kawauchi et al . , 2006 ) . Interestingly , CDK5 was recently found to play an important role in invadopodia formation and cancer cell invasion , and CDK5 inhibition blocked both these processes ( Quintavalle et al . , 2011; Bisht et al . , 2015 ) . CDK5 phosphorylated Caldesmon , causing its degradation by the proteasome and promoting invadopodia formation and invasion ( Quintavalle et al . , 2011 ) . Since p27 is a poor inhibitor of CDK5 complexes ( Lacy et al . , 2005; Lee et al . , 1996 ) , it would be interesting to test whether this role of CDK5 may involve p27 . Indeed , an attractive hypothesis is that p27 could simultaneously interact with CDK5 complexes , via its N-terminal cyclin-CDK binding domain , and with Cortactin via its C-terminus , thereby recruiting CDK5 in proximity to Caldesmon bound to Cortactin and allowing Caldesmon phosphorylation . Our study further defines the complex role played by p27 in the regulation of cytoskeletal dynamics , migration , EMT and invasion and provides another element to explain why the Cdkn1b gene is rarely mutated in cancer ( Chu et al . , 2008; Besson et al . , 2008; Kandoth et al . , 2013 ) . Indeed , p27 is either downregulated , mostly via increased proteasomal degradation , or excluded from the nuclei of cancer cells . Given that upon cytoplasmic relocalization , p27 promotes both migration and invasion and may serve to coordinately regulate these processes , it appears likely that this feature may be selected for during tumor progression and could participate in the acquisition of a metastatic behavior by cancer cells .
Mouse anti c-Myc ( 9E10 , sc-40 , RRID:AB_627268 ) , p27 ( F8 , sc-1641 , RRID:AB_628074 ) , p27 ( SX53G8 . 5 , sc-53871 , RRID:AB_785029 ) , αPAK ( A6 , sc-166887 , RRID:AB_10609226 ) , RhoA ( 26C4 , sc-418 , RRID:AB_628218 ) and rabbit anti p27 ( C19 , sc-528 , RRID:AB_632129 ) , Myc ( A14 , sc-789 , RRID:AB_631274 ) , Cortactin ( H191 , sc-11408 , RRID:AB_2088281 ) , Tks5 ( M-300 , sc-30122 , RRID:AB_2254551 ) , αPAK ( N-20 , sc-882 , RRID:AB_672249 ) , Arp2 ( H-84 , sc-15389 , RRID:AB_2221848 ) , c-Src ( SRC2 , sc-18 , RRID:AB_631324 ) and ERK1 ( K-23 , sc-94 , RRID:AB_2140110 ) antibodies were purchased from Santa Cruz Biotechnology . Mouse anti p27 ( 610242 ) , Grb2 ( 610112 , RRID:AB_397518 ) , Cortactin ( 610050 , RRID:AB_397462 ) , phopsho-Ser ( 612547 , RRID:AB_399842 ) and Rac1 ( 610650 , RRID:AB_397977 ) antibodies were purchased from BD-Transduction Laboratories . Mouse anti β-actin ( A2228 , RRID:AB_476697 ) antibody was purchased from Sigma-Aldrich . Rabbit anti phospho-Ser144-PAK1/phospho-Ser141-PAK2 ( #2606 , RRID:AB_2299279 ) antibody was purchased from Cell Signalling Technology . Rabbit anti acetyl-Cortactin ( #09–881 , RRID:AB_10584980 ) antibody was purchased from Millipore . Secondary antibodies against whole Ig or Ig light-chain conjugated to horseradish peroxydase or Cyanine-2–3 and −5 were from Jackson ImmunoResearch ( RRID:AB_10015282 , RRID:AB_2340612 , RRID:AB_2307443 , RRID:AB_2340607 , RRID:AB_2340770 , RRID:AB_2340826 , RRID:AB_2340813 , RRID:AB_2340819 , RRID:AB_2339149 , RRID:AB_2338512 ) . siRNA control ( D-001810-10-05 ) , ON-TARGETplus Mouse Rac1 ( 19353 ) siRNA - SMARTpool ( L041170000005 ) ( #2 ) and ON-TARGETplus Mouse PAK1 ( 18479 ) siRNA - SMARTpool ( L048101000005 ) were from Dharmacon . mRac1 siRNA ( #1 ) ( sc-36352 ) , mRhoA siRNA ( #1 ) ( sc-29471 ) and mRhoA siRNA ( #2 ) ( sc-36414 ) were purchased from Santa Cruz Biotechnologies . FRAX597 and Y-27632 were purchased from Selleckchem . NSC23766 was purchased from Tocris Biosciences . FRAX1036 and G-5555 were purchased from MedChemExpress . p27 constructs and p27 point mutants and deletion mutants in pCS2+ , pGEX4T1 ( Pharmacia ) , pET16b ( Novagen ) , pcDNA3 . 1 Hygro ( Invitrogen ) or pQCXIP ( Clontech ) were described previously ( Besson et al . , 2004b; Serres et al . , 2012 ) . The Myc-tagged full-length and deletion mutants of mouse Cortactin ( WT , ΔABR6 , ΔABR5-6 , ΔABR4-6 , ΔABR3-6 , ΔABR , ΔABR/NTA and ΔSH3 ) in pME18S vector were described previously ( Katsube et al . , 2004 ) . Full length human Cortactin was cloned by RT-PCR from IMR90 mRNA and inserted into the pcDNA3 . 1 Hygro Myc vector . Full-length mouse Cortactin , Cortactin S113A and Cortactin S113D were kindly provided by Dr . Alan Mak ( Queen’s University , Kingston , Canada ) ( Webb et al . , 2006 ) and subcloned into pQCXIP . Cortactin TA ( S113A , S150A and S282A ) and TD ( S113D , S150D and S282D ) were generated by site-directed mutagenesis from the pQCXIP-Cortactin-S113A and pQCXIP-Cortactin-S113D vectors , respectively . GFP-Tks5 , kindly provided by Dr . Frederic Saltel ( INSERM UMR1053 , Bordeaux ) , was subcloned into the pQCXIP vector . pCMV6-Myc-hPAK1 was kindly provided by Dr Jonathan Chernoff ( Fox Chase Cancer Center , Temple Health , Philadelphia; Addgene #12209 ) . pGEX2TK-PAK-CD was kindly provided by Dr . John Collard ( The Netherlands Cancer Institute , Amsterdam , The Netherlands ) . pEBB-Src-SH3-2 encoding the second SH3 domain of Nck1 fused to the myristoylation sequence of Src ( myr-SH3-2 ) was a gift from Dr Bruce Mayer ( University of Connecticut , Farmington ) . All plasmids used in this study were verified by sequencing . Primary MEFs were prepared as described previously ( Spector , 1997; Besson et al . , 2004b ) from p27+/+ , p27CK−/CK− or p27−/− embryos . MEFs were immortalized by infection with retroviruses encoding the human papilloma virus E6 protein and hygromycin selection . HeLa ( RRID:CVCL_0030 ) , HEK 293 ( RRID:CVCL_0045 ) , A-375 ( RRID:CVCL_0132 ) and A549 ( RRID:CVCL_0023 ) cells were obtained from Cell Lines Services . These cells were authenticated by short tandem repeat profiling . All cells were routinely tested to be free of mycoplasma contamination by DAPI staining . All cells were grown at 37°C and 5% CO2 in DMEM ( Sigma ) , 4 . 5 g/l glucose supplemented with 10% fetal calf serum , 0 . 1 mM nonessential amino acids and 2 µg /ml penicillin-streptomycin . When indicated , tissue culture vessels were coated with 0 . 2% gelatin from bovine skin ( Sigma G1393 ) for 1 hr . MEFs were infected with different plasmids using retroviruses produced in Phoenix ecotropic cells transfected by the calcium phosphate method . HEK293 cells were also transfected by the calcium phosphate method for 24 hr . HeLa cells were transfected using JetPrime ( Polyplus transfection ) according to manufacturer’s instructions . For siRNA transfections in MEF E6 cells , we used Interferin ( Polyplus transfection ) according to manufacturer’s instructions . Cells were scraped and lysed in IP buffer containing 50 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 2 . 5 mM EGTA , 0 . 1% Tween20 , 10% glycerol and 1% NP-40 complemented with 1 mM DTT , 10 mM β-glycerophosphate , 10 mM NaF , 10 mM sodium orthovanadate , 10 µg/ml Aprotinin , 10 µg/ml Bestatin , 10 µg/ml Leupeptin and 10 µg/ml Pepstatin A . After sonication for 10 s , cells extracts were centrifuged for 5 min at 12 , 500 g and supernatants were collected . Lysates ( 300 µg for MEF E6 and HeLa , 500 µg for HEK293 ) were incubated with 3 µg of indicated antibodies and 12 µl protein-A sepharose beads ( IPA300 , Repligen ) ( co-immunoprecipitation ) or with recombinant GST proteins and glutathione sepharose beads ( Pharmacia ) ( GST pull-down ) at 4°C for 4 hr . Beads were then washed four times in lysis buffer and resuspended in 10 µl 4X sample buffer , boiled , and subjected to western blotting . Pull-down assays for Rac1 activity were performed as described previously ( Besson et al . , 2004b; Sander et al . , 1998; Malliri et al . , 2000 ) . Briefly , E6 MEFs in 100 mm plates were washed with ice-cold PBS , scraped off the plates in 400 μl cell-lysis buffer ( 50 mM Tris-HCl at pH 7 . 4 , 2 mM MgCl2 , 1% NP-40 , 10% glycerol , 100 mM NaCl , complemented with 1 mM dithiothreitol , 10 μg/ml leupeptin , 10 μg/ml aprotinin , and 10 μg/ml pepstatin-A ) on ice . Lysates were centrifuged for 5 min at 14 , 000 rpm at 4°C . Three hundred μl of cleared lysates were incubated for 30 min at 4°C with 8 μl of GST–PAK-CD bound to glutathione-coupled Sepharose beads ( at ∼1 μg GST fusion protein per μl of beads ) . Bead pellets were washed three times with ice-cold cell lysis buffer , resuspended in 4X sample buffer , and subjected to SDS-PAGE as described below . Ten μl of cell lysates were used for protein loading . Cells were lysed in IP buffer as described above . Lysates and immunoprecipitates were mixed with 4X sample buffer and boiled . Proteins were resolved on 8–15% SDS-PAGE ( depending on protein size ) and transferred to polyvinylidene difluoride membrane ( Immobilon-P , Millipore ) . Membranes were blocked with PBS-T ( PBS , 0 . 1% Tween-20 ) , 5% non-fat dry milk and probed with indicated primary antibodies overnight at 4°C with gentle agitation . Membranes were washed three times in PBS-T then incubated with corresponding HRP-conjugated secondary antibody ( 1/10000 ) for 4 hr at room temperature . Bands were visualized using enhanced chemiluminescence detection reagents ( Millipore , BioRad , Ozyme ) and autoradiographic film ( Blue Devil ) . Coverslips were cleaned overnight in 1 M HCl , washed four times in ddH2O and then coated successively with 50 μg/ml Poly L-lysine , 0 . 5% glutaraldehyde , fluorescent gelatin ( 1:10 mix of gelatin from pig skin Oregon green 488 conjugate [G13186 , Molecular Probes] and 0 . 2% gelatin from bovine skin [Sigma G1393] ) , and 5 mg/ml sodium borohydride . Between each coating , coverslips were washed three times with PBS . Coverslips were then sterilized with 70% ethanol and cells were seeded and incubated either overnight or 48 hr before fixation and staining . Cells were seeded on coverslips coated as described above with only non fluorescent gelatin and grown overnight . When indicated , cells were permeabilized with 20 μg/ml digitonin in PBS for 2 min before fixation . Cells were fixed with 2% PFA in PBS for 20 min at 37°C . For immunostaining , cells were permeabilized for 3 min with PBS 0 . 2% Triton X-100 , rinsed three times in PBS and incubated for 20 min in blocking solution ( PBS , 3% BSA , 0 , 05% Tween20 and 0 , 08% sodium azide ) and with primary antibodies diluted in blocking solution for 1 hr . After three washes of 5 min in PBS , cells were incubated for 30 min at 37°C with Cy-2 Cy-3 or Cy-5 conjugated secondary antibodies at 1:500 dilutions . Coverslips were washed 3 times for 5 min in PBS with the first wash containing 0 . 1 μg/ml Hoechst H33342 and mounted on glass slides with gelvatol ( 20% glycerol ( v/v ) , 10% polyvinyl alcohol ( w/v ) , 70 mM Tris pH 8 ) . Images were captured on a Nikon 90i Eclipse microscope using a DS-Qi2 HQ ( Nikon ) camera and the NIS Element software . For live cell imaging , MEFs infected with pQCXIP-Tks5-GFP were seeded in gelatin-coated wells overnight and placed in a controlled atmosphere chamber ( 37°C and 5% CO2 ) for imaging . Images were acquired every 2 min with a Zeiss Cell Observer microscope for 8 hr . Image analysis was performed with Image J/Fiji . Cells were seeded in gelatin coated 96-well plates ( Essen ImageLock , Essen Bioscience ) at 80% confluence and incubated overnight to allow them to reach confluence . Cells were treated for 2 hr before wounding with 2 μg/ml mitomycin C ( M4287 , Sigma ) to block cell proliferation and scratches were performed in the cell monolayer using the wound maker ( Essen Bioscience ) . Cells were washed immediately three times with PBS and re-fed with growth medium . Cell migration was monitored with an Incucyte FLR Live-Cell imaging system equipped with a 20X objective ( Essen Bioscience ) . Images were acquired every 3 hr for 48 hr . Migration was quantified by measuring the relative wound density of at least three biological replicates in each experiment , using the Incucyte software ( Essen Bioscience ) as recommended by the manufacturer . Fifty μl of 1 mg/ml of rat tail Collagen I ( 354236 , Corning ) solution were incubated for 1 hr at room temperature in 8 μm pores transwell inserts in 24-well plates ( #353097 , Corning ) to allow gelling of collagen . The gel was hydrated with DMEM-50% FCS for 3 hr at 37°C . Cells were incubated for 3 hr with 2 μg/ml Mitomycin C , washed twice with DMEM 0 . 1% FCS , trypsinized , counted and 30 , 000 cells were seeded in the transwell upper chamber in 100 μl of DMEM 0 . 1% FCS , with 600 μl of DMEM 10% FCS in the bottom chamber . Plates were incubated for 48 hr at 37°C and 5% CO2 . At the end of the assay , the content of the top chamber was removed with a cotton swab and cells that invaded onto the bottom membrane of the transwell were quantified by XTT staining with 200 μg/ml XTT ( Santa-Cruz Biotechnology , sc-258336 ) previously activated by adding 25 μM phenazine methosulfate ( Sigma , P9625 ) in DMEM 10% FCS for 4 hr at 37°C . Absorbance was read at 450 nm . Cells were then fixed for 5 min in 95% ethanol/5% acetic acid and stained overnight with hematoxylin to capture images of invaded cells . Statistical analyses were performed using Graphpad Prism 6 . 0 software . Differences between groups were evaluated using 1-way ANOVA followed by Bonferroni test for multiple comparison and considered significant when p<0 . 05 . Data are presented as mean ± SEM . HEK293 cells were transfected either with human Myc-tagged Cortactin or Myc-Cortactin and Myr-SH3-2 ( that activates PAK1 ) for 24 hr . Cortactin was immunoprecipitated from lysates from four 100 mm plates per IP using 12 μg per IP of Mouse anti-Myc 9E10 antibody . Samples were separated on 12% SDS-PAGE . Gel was stained with instant blue for 1 hr . Amounts of immunoprecipitated Cortactin was estimated to 4 ug for each IP by comparison with a BSA standard on the same gel . Bands of interest were cut and placed in eppendorf tubes in 30 μl of 1% acetic acid solution and stored at −20C until MS/MS analysis . | When animals develop from embryos to adults , or try to heal wounds later in life , their cells have to move . Moving means that the cells must invade into their surroundings , a dense network of proteins called the extracellular matrix . The cell first attaches to the extracellular matrix; degrades it; and then moves into the newly opened space . Cells have developed specialized structures called invadosomes to enable all these steps . Invadosomes are never static , they first assemble where cells interact with extracellular matrix , they then release proteins that loosen the matrix , and finally disassemble again to allow cells to move . Invadosomes in cancer cells often become overactive , and can allow the tumor cells to spread throughout the body . A lot of different proteins are involved in controlling how and when cells move . p27 is a well-known protein usually found in a cell’s nucleus along with the cell’s DNA . Inside the nucleus , p27 suppresses tumor growth by stopping cells from dividing . However , often in cancer cells p27 moves outside of the cell’s nucleus where it contributes to cell movement via an unknown mechanism . To answer how p27 controls cell invasion , Jeannot et al . used a biochemical technique to uncover which proteins p27 binds to when it is outside of the nucleus . One of its interaction partners was called Cortactin . This protein is known to be an important building block of invadosomes , and is involved in both the assembly and disassembly of these structures . In further experiments , Jeannot studied mouse cells with or without p27 and human cancer cells that can be grown in the laboratory . The experiments revealed that p27 promotes an enzyme called PAK1 to also bind to Cortactin . PAK1 then modified Cortactin , causing whole invadosomes to disassemble , which in turn allowed cells to de-attach from the matrix and move forward . In contrast , cells lacking p27 had more stable invadosomes , attached more strongly to the matrix and were better at degrading it , but could not invade as well as cells with p27 . Overall these experiments showed a new way that p27 promotes cell invasion . The next steps will include finding out exactly how the modification of Cortactin causes the invadosomes to disassemble . Furthermore , it will be important to study whether forcing p27 back into the nucleus can reduce the spread of cancer cells in the body . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"cancer",
"biology"
] | 2017 | p27Kip1 promotes invadopodia turnover and invasion through the regulation of the PAK1/Cortactin pathway |
The impact of phage predation on bacterial pathogens in the context of human disease is not currently appreciated . Here , we show that predatory interactions of a phage with an important environmentally transmitted pathogen , Vibrio cholerae , can modulate the evolutionary trajectory of this pathogen during the natural course of infection within individual patients . We analyzed geographically and temporally disparate cholera patient stool samples from Haiti and Bangladesh and found that phage predation can drive the genomic diversity of intra-patient V . cholerae populations . Intra-patient phage-sensitive and phage-resistant isolates were isogenic except for mutations conferring phage resistance , and moreover , phage-resistant V . cholerae populations were composed of a heterogeneous mix of many unique mutants . We also observed that phage predation can significantly alter the virulence potential of V . cholerae shed from cholera patients . We provide the first molecular evidence for predatory phage shaping microbial community structure during the natural course of infection in humans .
Traditional views of host–pathogen interactions are of a battle between two opposing organisms . This perspective is being challenged with greater appreciation of the influence of the host's microbial ecosystem on these interactions ( Lozupone et al . , 2012 ) . Bacteriophages influence bacterial populations in many ecosystems and specifically temperate phages play a role in many diseases through lysogenic conversion ( Brüssow et al . , 2004 ) . In contrast , the impact of lytic phage predation on bacterial pathogens in the context of human disease is under-appreciated . Vibrio cholerae is a water-borne pathogen responsible for the diarrheal disease cholera . In order for V . cholerae to colonize the human small intestine and elicit diarrhea , it produces a set of virulence determinants including cholera toxin ( Herrington et al . , 1988 ) . The ToxR signaling cascade activates expression of these virulence factors in response to host environmental stimuli ( Miller et al . , 1987; Taylor et al . , 1987; DiRita et al . , 1991; Childers and Klose , 2007 ) . Cholera epidemics appear with regular seasonality in regions like Bangladesh ( Faruque et al . , 2005 ) and can arise unpredictably in vulnerable regions as exemplified by the epidemic that recently began in Haiti following the single-source introduction of a pandemic V . cholerae O1 strain from another continent ( Chin et al . , 2011; Cravioto et al . , 2011; Frerichs et al . , 2012; Katz et al . , 2013 ) . Lytic phages are hypothesized to impact cholera disease burden in Bangladesh ( Faruque et al . , 2005 ) . The predation of V . cholerae by phages following their co-ingestion from the environment is central to this hypothesis; however , molecular evidence in support of this hypothesis is lacking . We recently described the ICP2 species of V . cholerae-specific , virulent podoviruses that are found sporadically in cholera patient stool samples collected since 2001 in Bangladesh ( Seed et al . , 2011 and present study ) . To begin to address the geographic diversity of cholera phages in patient stools , we tested by plaque assay for the presence of phages within nine Haitian cholera patient stool samples collected in 2013 . We identified one sample that had a high titer of phage ( 108 PFU/ml ) relative to V . cholerae ( 105 CFU/ml ) . Whole genome sequencing and comparative analysis of this phage , named ICP2_2013_A_Haiti , revealed 84% identity over 93% of its genome to ICP2_2011_A , a phage isolated from Bangladesh in 2011 . An alignment of the annotated ICP2 genome ( Seed et al . , 2011 ) with the three available ICP2-related isolates shows that synteny is conserved between these geographically distinct phages , with no significant genome rearrangements observed ( Figure 1—figure supplement 1 ) . The Haitian ICP2 isolate is strikingly similar , although clearly distinct , from Bangladeshi ICP2 isolates , and is the first lytic phage reported to be associated with epidemic V . cholerae in Haiti . Remarkably , we observed that most V . cholerae isolates recovered from the same stool sample as ICP2_2013_A_Haiti were resistant to infection by this phage . We investigated phenotypic heterogeneity within this stool sample by testing 269 single colony isolates for sensitivity to ICP2_2013_A_Haiti and found that 267 ( >99% ) were phage-resistant . Eight phage-resistant isolates and the two phage-sensitive isolates from this sample were subjected to whole genome sequencing . All 10 isolates were isogenic except for mutations within ompU encoding the major outer membrane porin ( Supplementary file 1 ) . Furthermore , the ompU mutations in the eight phage-resistant isolates were heterogeneous . We sequenced ompU from an additional 11 phage-resistant isolates from this stool sample and found a total of six unique ompU alleles ( Figure 1A ) . When each of these alleles was used to replace ompU in a clean genetic background , all conferred resistance to ICP2 , yet produced normal amounts of OmpU ( Figure 2 and data not shown ) , suggesting that ICP2 uses OmpU as a receptor to initiate infection and that the mutations disrupt this interaction . These data showing that intra-host V . cholerae isolates are isogenic , only differing in ICP2 resistance mutations , indicates that they diverged within the patient . Furthermore , the presence of multiple different ICP2 resistance mutations within this single host suggests that selection of phage resistance occurred multiple , independent times during the course of infection . We also found evidence for ICP2-mediated selection of OmpU mutants in isolates from Bangladesh . Analysis of the ompU sequence from 54 clinical isolates collected between 2001 and 2011 showed that 15% had non-synonomous mutations ( Figure 1A , Figure 1—figure supplement 2 ) and these alleles were sufficient for ICP2 resistance ( Figure 2 ) . One of the OmpU mutants , G325D , was observed in clinical isolates from Haiti and Bangladesh ( Figure 1A ) . Interestingly , seven of the eight mutant OmpU proteins detected in the Haitian and Bangladeshi clinical isolates had alterations in two predicted extracellular loops , while the eighth had an altered transmembrane segment adjacent to the first of these extracellular loops ( Figure 1B ) . This suggests that ICP2 interacts specifically with these exposed loops to initiate infection . 10 . 7554/eLife . 03497 . 003Figure 1 . The presence of V . cholerae OmpU and ToxR mutants present within and between cholera patients . ( A ) Graphical depiction and frequency of OmpU mutants found within a stool sample containing ICP2_2013_A_Haiti phage ( 108 PFU/ml ) from a single Haitian patient ( top ) and from different patients in Bangladesh ( n = 54 ) ( bottom ) . ( B ) Predicted membrane topology of mature OmpU generated using Pred-TMBB ( Bagos et al . , 2004 ) . Locations of amino acid substitutions or insertions carried by V . cholerae clinical isolates are indicated . ( C ) Graphical depiction and frequency of ToxR mutants found within a stool sample containing ICP2_2011_A ( 109 PFU/ml ) from a single Bangladeshi patient . Amino acid substitutions or nonsense mutations ( asterisks ) are in orange and duplications are in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 03497 . 00310 . 7554/eLife . 03497 . 004Figure 1—figure supplement 1 . ICP2_2013_A_Haiti is closely related to ICP2 bacteriophages from Bangladesh . Comparison of ICP2 genomes collected from cholera patients in Dhaka , Bangladesh in 2004 ( ICP2 ) , 2006 and 2011 , and in Haiti in 2013 using progressiveMauve software . The degree of nucleotide similarity between aligned regions is indicated by the height of the similarity profile ( colored blocks ) , where mauve represents the highly conserved backbone genome and other colors represent segments whose presence varies between isolates . Annotated genes in the ICP2 genome are shown as white boxes , with genes transcribed from the negative strand displaced downward . The numbers above the ICP2 genome show distance in kilobases . DOI: http://dx . doi . org/10 . 7554/eLife . 03497 . 00410 . 7554/eLife . 03497 . 005Figure 1—figure supplement 2 . Identification of OmpU mutants in samples collected at the International Centre for Diarrheal Disease Research , Bangladesh between 2001–2011 . ( a ) Single V . cholerae O1 El Tor isolates from different stool samples collected within a given year are indicated as closed circles . ( b ) The number of isolates with the noted mutation is indicated in a given year . If left blank , OmpU was wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 03497 . 00510 . 7554/eLife . 03497 . 006Figure 2 . OmpU expression of OmpU and ToxR mutants and their sensitivity to ICP2 . Outer membrane fractions were prepared from samples matched by equivalent OD600 units . Samples were separated by SDS-PAGE and subjected to Western blot analysis using rabbit polyclonal antisera against OmpU . The sensitivity of each strain to ICP2_2013_A_Haiti is represented as a histogram of the efficiency of plaquing , which is the plaque count ratio of a mutant V . cholerae strain to that of the wild-type strain . The limit of detection for plaque assays was 10−7 . DOI: http://dx . doi . org/10 . 7554/eLife . 03497 . 006 We also tested for V . cholerae population heterogeneity within an ICP2-positive stool sample isolated in Bangladesh in 2011 . Of 46 single colony isolates tested for sensitivity to ICP2_2011_A , 22% were resistant to the phage . Whole genome sequencing of two phage-sensitive and four phage-resistant isolates from this sample revealed that the strains were isogenic , except that phage-resistant isolates had nonsense mutations in toxR , the direct transcriptional activator of ompU ( Crawford et al . , 1998; Supplementary file 2 ) . We sequenced toxR from an additional eight phage-resistant isolates from this stool sample and found that , analogous to what was observed in the Haitian patient sample , there were multiple unique toxR alleles , which included two non-synonomous and two nonsense mutations ( Figure 1C ) . Each of the four mutant toxR alleles conferred ICP2 resistance and abrogated expression of OmpU when used to replace wild-type toxR in a clean genetic background ( Figure 2 ) . Both ICP2-sensitivity and OmpU expression were restored to the clinical toxR mutants by expressing ompU in trans or by reverting the toxR allele to wild-type ( data not shown ) , indicating that these mutant toxR alleles are necessary and sufficient for ICP2 resistance and that this resistance is mediated through loss of OmpU expression . To address the potential consequences of the phage-resistance mutations on V . cholerae fitness , we tested the four most frequently isolated ompU alleles and all four toxR alleles in survival and growth assays . Recent Tn-seq analyses of V . cholerae showed that OmpU is critical for fitness in an infant rabbit infection model ( Fu et al . , 2013; Kamp et al . , 2013 ) and for dissemination from the host into pond water ( Kamp et al . , 2013 ) . OmpU has also been shown to be important for protection against the bactericidal effect of bile salts ( Provenzano et al . , 2001 ) , cationic peptides ( Mathur and Waldor , 2004 ) , and intestinal organic acids ( Merrell et al . , 2001 ) . When we tested for survival in bile and for competitive fitness in pond water we found that the four OmpU mutants are fully fit ( Figure 3A , B ) , consistent with their wild-type levels of OmpU expression ( Figure 2 ) . Moreover , the OmpU ( G325D ) mutant was fully virulent in a single round , competitive infection in infant rabbits ( Figure 4 , first column ) . However , the OmpU mutants , particularly the A195T and G325D mutants , showed a mild competitive defect after multiple passaging in growth medium ( Figure 3C ) . The mild growth defect of the OmpU mutants in conjunction with the low prevalence of ICP2 ( Seed et al . , 2011 ) may explain why the ICP2-resistant ompU variants do not become fixed in the population ( Figure 1—figure supplement 2 ) . In sharp contrast , in the presence of ICP2 we observed a 10 , 000-fold enrichment of the OmpU ( G325D ) mutant over the wild-type after infection of infant rabbits ( Figure 4 ) , indicating that strong selective pressure is imposed by phage predation during V . cholerae infection . This emulates what we hypothesize happens in human infections in the presence of ICP2 and is the first demonstration of phage predation of V . cholerae in the context of a diarrheal disease model . 10 . 7554/eLife . 03497 . 007Figure 3 . The fitness cost of clinically relevant OmpU and ToxR mutations . ( A ) Clinically relevant OmpU mutants retain fitness in the presence of bile . *p < 0 . 05 significantly different means for the compared data sets ( Mann–Whitney U Test ) . ( B ) OmpU mutants retain competitive fitness in pond water . *p < 0 . 05 significantly different from wild-type control ( Kruskal–Wallis and post hoc Dunn's multiple comparison tests ) . ( C ) OmpU mutants have slight competitive fitness defects when serially passaged in Luria–Bertani broth ( for ca . 58 generations ) . *p < 0 . 05 or ****p < 0 . 0001 significantly different from wild-type control ( Kruskal–Wallis and post hoc Dunn's multiple comparison tests ) . ( D ) ToxR mutants are attenuated in vivo using the infant mouse colonization model . **p < 0 . 01 or ***p < 0 . 001 significantly different from the in vitro median ( Mann–Whitney U Tests ) . The horizontal bars indicate the median of each data set . Open symbols represent data below the limit of detection . DOI: http://dx . doi . org/10 . 7554/eLife . 03497 . 00710 . 7554/eLife . 03497 . 008Figure 3—figure supplement 1 . Clinical isolates harboring ToxR mutations are severely attenuated for infection . Competitions indices ( CI ) in infant mice after 24 hr of infection were determined between each clinical ToxR mutant strain and its isogenic ToxR wild-type revertant strain carrying a lacZ deletion . The lacZ deletion allowed for differentiation of the competing strains upon plating on LB agar plates supplemented with X-gal . Each symbol represents the CI for an individual animal . The horizontal bars indicate the median of each data set . The open symbols represent data below the limit of detection for the ToxR mutant strain . DOI: http://dx . doi . org/10 . 7554/eLife . 03497 . 00810 . 7554/eLife . 03497 . 009Figure 4 . Phage predation leads to enrichment of OmpU mutant over wild-type in vivo . Competitive indices ( CI ) were determined between wild-type ΔlacZ and OmpU G325D in the absence or presence of ICP2_2013_A_Haiti at the multiplicity of infection ( MOI ) indicated in infant rabbits 12 hr post-infection . Each symbol represents the CI for an individual rabbit and the horizontal lines indicate the median for each condition . The open symbols represent data below the limit of detection for the wild-type strain . **p < 0 . 01 , significantly different from no phage control ( Kruskal–Wallis and post hoc Dunn's multiple comparison tests ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03497 . 009 V . cholerae strains harboring any of the four ToxR mutations were avirulent in the infant mouse model of infection and were indistinguishable from a ΔtoxR mutant ( Figure 3D ) . The clinical isolates containing these mutations were also at least 100-fold attenuated ( Figure 3D—figure supplement 1 ) , indicating that these isolates do not harbor compensatory mutations . The cytoplasmic amino terminus ( D17-E112 ) of ToxR where the two non-synonomous mutations mapped has homology to the winged helix-turn-helix family of transcription activators and is involved in DNA binding and transcriptional activation ( Miller et al . , 1987; Ottemann et al . , 1992 ) . A number of point mutations in this domain were previously shown to abrogate activation of ompU and virulence genes ( Ottemann et al . , 1992; Morgan et al . , 2011 ) . The severe virulence defects of the two non-synonomous toxR mutants from the Bangladeshi patient sample ( Figure 3D ) are consistent with an inability of the encoded mutant ToxR proteins to activate these genes . These toxR mutants would not colonize the human small intestine and be recovered in the secretory diarrhea if ingested ( Herrington et al . , 1988 ) , which further supports our conclusion that these mutants arose due to ICP2 predation during cholera infection in this individual . This also indicates that , in this patient , predation and selection of phage-resistant mutants likely occurred late in infection when colonization and virulence gene expression were no longer required for the development of acute symptoms ( Merrell et al . , 2002 ) . We have shown that lytic phages are an unexpected ‘third party’ that impose significant bactericidal pressure on an environmentally transmitted pathogen during the natural course of infection in humans . We observed that adaptations to phage predation involve tradeoffs in evolutionary fitness and provide a molecular mechanism for phage predation impacting V . cholerae transmission and seeding of environmental reservoirs . Our results highlight that host–pathogen interactions are embedded within and strongly affected by a complex microbial ecosystem . Predator-prey dynamics are notably absent between microbiotal phage and their bacterial hosts in the intestinal ecosystem in healthy humans ( Reyes et al . , 2010 ) . In contrast , our findings indicate that diseased states , which are often accompanied by significant bacterial proliferation , likely facilitate these predatory interactions .
Strains utilized in this study are listed in Supplementary file 3 . Stool specimens and V . cholerae isolates from the International Centre for Diarrheal Disease Research , Bangladesh were collected and stored from previous studies ( Nelson et al . , 2008; Seed et al . , 2011 , 2012 , 2013 ) . Phage ( ICP2_2011_A and ICP2_2013_A_Haiti ) were isolated from cholera rice-water stool samples as described ( Seed et al . , 2011 ) . V . cholerae isolates were isolated from stool samples on Luria–Bertani agar with sulfamethoxazole ( 24 µg/ml ) , trimethoprim ( 32 μg/ml ) , nalidixic acid ( 20 µg/ml ) and streptomycin ( Sm ) ( 100 µg/ml ) . Mutations were introduced using pCVD442-lac as previously described ( Seed et al . , 2012 ) . Strains containing the pMMB67EH vector were grown in the presence of 50 μg/ml ampicillin and 1 mM isopropyl-ß-D-thiogalactopyranoside ( IPTG ) . V . cholerae and phage genomic libraries were generated ( Lazinski and Camilli , 2013 ) and sequenced using an Illumina HiSeq2000 ( Tufts University Core Facility ) . Sequence reads from each V . cholerae isolate were mapped to the reference genome V . cholerae O1 2010EL-1786 ( for Haitian isolates , accession numbers NC_016445 . 1 and NC_016446 . 1 ) or to V . cholerae O1 N16961 ( for Bangladeshi isolates , accession numbers NC_002505 . 1 and NC_002506 . 1 ) and mutations were identified using Varscan v2 . 3 . 6 ( Koboldt et al . , 2012 ) and CLC Genomic Workbench software ( Version 6 . 8; CLC Bio , Denmark ) . Variants were called as being present ( different allele than reference ) , absent ( the same allele as reference ) or low-coverage ( <15×; in which case no call was made ) . Cultures were grown to mid-exponential growth phase in Luria–Bertani ( LB ) broth at 37°C with aeration . Bacteria were re-suspended in 200 mM Tris–HCl , pH 8 . 0 and 2 mM EDTA . Sucrose was added to a final concentration of 20% followed by lysozyme treatment for 10 min at 37°C . Using a dry ice/ethanol bath , samples were freeze-thawed twice followed by DNAse I treament for 20 min at room temperature . Samples were spun for 5 min at 16 , 100×g to pellet the membrane fraction . The pellets were re-suspended in 1% Triton X-100 , 10 mM MgCl2 , and 50 mM Tris–HCl , pH 8 . 0 and incubated at 37°C for 20 min . Samples were spun again for 5 min at 16 , 100×g to separate the inner membrane from the outer membrane . The pellets , which contain the outer membrane proteins , were re-suspended in 200 mM Tris–HCl , pH 8 . 0 . Outer membrane fractions were boiled for 10 min in sample buffer containing sodium dodecyl sulfate and β-mercaptoethanol and separated on a NuPAGE 4–12% Bis-Tris polyacrylamide gel ( Life Technologies , Carlsbad , CA ) . Protein gels were transferred to a nitrocellulose membrane for Western blotting . Membranes were probed with rabbit polyclonal antisera against V . cholerae OmpU ( gift of James Kaper ) . A Cy5 goat anti-rabbit antibody was used to develop the blot . Strains ( grown to OD600 = 0 . 5 ) were assessed for their ability to survive 0 . 2% porcine bile ( Sigma , St . Louis , MO ) in 0 . 85% NaCl for 1 hr at room temperature . Pond water survival assays were done as described previously ( Kamp et al . , 2013 ) at 30°C for 48 hr . Competition experiments were performed between the strain of interest ( lacZ+ ) and the appropriate control strain ( ΔlacZ ) , and outputs were plated on LB agar plates containing 100 µg/ml Sm and 40 µg/ml 5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside ( X-gal ) . All animal experiments were in accordance with the rules of the Department of Laboratory Animal Medicine at Tufts Medical Center . Infant mouse colonization assays and in vitro controls were done as described previously ( Seed et al . , 2012 ) . 3-day old infant rabbits were pre-treated with Cimetidine-HCL ( Morton Grove Pharmaceuticals , Morton Grove , IL ) 3 hr prior to infection ( Kamp et al . , 2013 ) and infected with ∼5 × 108 CFU in 2 . 5% sodium bicarbonate buffer ( pH 9 ) . A 1:1 mixture of wild-type to mutant ( for experiments without phage ) or 10:1 wild-type to mutant ( for experiments including phage ) was used . For experiments involving the addition of phage , phage were added immediately before intragastric inoculation to each rabbit inoculum to limit phage adsorption ex vivo ( phage were in contact with the bacterial inoculum for ∼30–60 s prior to gavage of each animal ) . Rabbits were euthanized ∼12 hr post-inoculation and cecal and/or small intestinal fluid was collected by puncture . As with the in vitro assays , the competing strains were enumerated on LB agar plates containing 100 µg/ml Sm and 40 µg/ml X-gal . | Cholera epidemics occur seasonally in areas such as Bangladesh , and outbreaks can also strike in vulnerable regions , as has occurred recently in Haiti . The disease is caused by Vibrio cholerae , a water-borne bacterium that colonizes the small intestine , and its symptoms include severe diarrhea and vomiting which can lead to death if the patient is not treated promptly . Lytic phages are viruses that specifically attack and kill bacteria . After replicating many times inside the bacterial cell , the phages break open and destroy the cell . Over time a bacterial population can evolve to resist this phage ‘predation’; however , it is not known if bacterial pathogens need to defend themselves against phage attack when they infect humans . It had been suggested that phages might affect the progress of cholera infections in people , but molecular evidence that supports this hypothesis was lacking . When testing stool samples from Haitian cholera patients , Seed et al . found one sample contained a lot of lytic phage relative to the amount of V . cholerae present . This phage was very similar to—but distinct from—a phage found in Bangladeshi patients . The V . cholerae bacteria isolated from the stool sample were resistant to attack by the phage . Sequencing the genome of individual bacteria from this sample revealed that each had a mutation that made them resistant to the phage; and while many types of these mutations were found , these were the only differences between all the V . cholerae bacteria in this patient sample . This suggests that this resistance developed independently many different times within the patient due to strong selective pressure from phage predation . When Seed et al . looked at a phage-positive stool sample from a Bangladeshi patient , more mutations that made the bacteria resistant to this phage were found; however , these mutations were different again from the ones in the Haitian bacteria . Because of the nature of these mutations the bacteria from this patient were rendered unable to cause disease and non-transmissible . This work shows that phages can indeed have access to pathogenic bacteria during human infection . It also indicates that the pressure imposed by phage predation can , in some cases , be so strong that the bacteria lose their virulence and ability to spread to other humans in order to become resistant to the phage . | [
"Abstract",
"Main",
"Text",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"microbiology",
"and",
"infectious",
"disease"
] | 2014 | Evolutionary consequences of intra-patient phage predation on microbial populations |
In bacteria , multicellular behaviors are regulated by cell–cell signaling through the exchange of both diffusible and contact-dependent signals . In a multicellular context , Myxococcus cells can share outer membrane ( OM ) materials by an unknown mechanism involving the traAB genes and gliding motility . Using live imaging , we show for the first time that transient contacts between two cells are sufficient to transfer OM materials , proteins and lipids , at high efficiency . Transfer was associated with the formation of dynamic OM tubes , strongly suggesting that transfer results from the local fusion of the OMs of two transferring cells . Last , large amounts of OM materials were released in slime trails deposited by gliding cells . Since cells tend to follow trails laid by other cells , slime-driven OM material exchange may be an important stigmergic regulation of Myxococcus social behaviors .
Myxococcus xanthus , a gram negative deltaproteobacterium , displays complex multicellular behaviors in response to environmental cues such as the presence of prey bacteria or starvation ( Zhang et al . , 2012 ) . In particular , starvation triggers a developmental program where thousands of cells coordinate their motility , moving into aggregation centers to build multicellular fruiting bodies where the cells form metabolically-inert spores . This multicellular response requires an arsenal of intercellular signals , including diffusible long-range signals as well as contact-dependent signals ( Konovalova et al . , 2010; Mauriello et al . , 2009 ) . One intriguing cell–cell communication mechanism involves the cell-to-cell transfer of outer membrane ( OM ) proteins between Myxococcus cells . This phenomenon was originally unmasked by mixing experiments where certain motility mutants were shown to rescue other motility mutants in a process called stimulation ( Nudleman et al . , 2005 ) . Stimulatable mutants all carried mutations in genes encoding predicted OM proteins ( termed cgl or tgl ) . Experiments with the Tgl and the CglB OM lipoproteins suggested that stimulation is transient and does not involve the exchange of genetic material , but results from the physical transfer of Tgl/Cgl proteins from donor Tgl+/Cgl+ cells to recipient Tgl−/Cgl− cells ( Nudleman et al . , 2005 ) . Remarkably , OM protein exchange is not restricted to motility proteins and virtually any OM protein and even lipid can be exchanged between cells ( Wei et al . , 2011; Pathak et al . , 2012 ) . Gliding ( A− ) motility has been shown to be important for transfer , but rather indirectly by promoting the formation of dense regions of aligned cells and favoring intimate cell–cell contacts ( Nudleman et al . , 2005; Pathak et al . , 2012 ) . The transfer process itself depends on two specific proteins , TraA and TraB ( Pathak et al . , 2012 ) . TraA is a protein with hallmarks of yeast floculins , a class of cell surface adhesins that mediate cell–cell interactions leading to flocculation ( Smukalla et al . , 2008 ) and TraB is a secreted protein of unknown function with a possible peptidoglycan-binding domain . TraA and TraB must be expressed both by donor and recipient cells for transfer to occur . Consequently , Wall and colleagues proposed that when adjacent cells engage Tra-dependent surface interactions ( i . e . , homotypic interactions or interactions with other surface ligands ) , the OMs fuse locally and OM materials are exchanged ( Wei et al . , 2011; Pathak et al . , 2012 ) . However , because transfer was studied in bulk assays this hypothesis could not be tested directly . Therefore , other mechanisms remained possible , for example long-range exchange of OM vesicles or even local cell lysis . In this study , we investigated the transfer mechanism at the single cell level to gain more insights into the transfer mechanism .
In a previous study , Wei et al . ( 2011 ) measured the transfer efficiency in agar plate mixing assays ( ‘Materials and methods’ ) , monitoring the appearance of fluorescent recipient cells over time with mCherry fluorescent probes ( OMmCherry and IMmCherry ) , which when fused to type II or type I signal sequences localize to the OM or the inner membrane ( IM ) , respectively . However , no information was obtained about the increase in fluorescence intensity in the recipient cells . Thus , in a prelude to this study , we repeated the Wei et al . ( 2011 ) experiment and further measured fluorescence fluctuations in recipient cells . For completion and to test the transfer of soluble periplasmic proteins , we also constructed a periplasmic probe , fusing mCherry to the Escherichia coli phoA signal sequence ( PERImCherry ) ( ‘Materials and methods’ and Figure 1—figure supplement 1 ) . Consistent with previous works and OM specific protein transfer , only OMmCherry was transferred significantly between cells . As observed by Wei et al . ( 2011 ) , transfer was highly efficient and ∼80% of the total recipient cells were already labeled after 12 hr of co-incubation ( Figure 1A ) . Transfer remained active for the next 36 hr because although the total number of recipient cells became stable after 24 hr , the fluorescence intensity of recipient cells increased regularly until it reached a plateau at 36 hr ( Figure 1B ) . After 36 hr of co-incubation , 20% of the recipient cells displayed a high level of fluorescence , showing that some cells acquire exogenous OM content with very high efficiency ( Figure 1C ) . A low amount of PERImCherry transfer was detected after 48 hr ( Figure 1A ) , suggesting that periplasmic proteins may also be exchanged but with a near background level efficiency . These findings confirm results from previous studies that transfer is a highly efficient OM-specific process . 10 . 7554/eLife . 00868 . 003Figure 1 . Transfer is a highly efficient OM-specific process . ( A ) Percentage of mCherry+ recipient cells as a function of time . For each strain and time point , at least 3000 cells were analyzed in triplicate . Error bars = SD . ( B ) Fluorescence intensity of mCherry+ recipient cells as a function of time . For each time point , the fluorescence numbers are expressed as a percentage of the mean fluorescence intensity of the donor cells population . For each time point , fluorescence intensities were measured for ∼3000 cells per strain . ( C ) Distribution of fluorescence intensities measured in the positive recipient cells after 12 hr ( green bars ) and 36 hr ( orange bars ) of co-incubation . Note the logarithmic scale log ( Fluorescence Intensity ) . The black arrow highlights the appearance of a highly-stained cell sub-population of mCherry+ cells at 36 hr . For each time point , fluorescence intensities were measured for ∼3000 cells per strain . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 00310 . 7554/eLife . 00868 . 004Figure 1—figure supplement 1 . Subcellular localization of indicated fluorescent probes before and after a plasmolysis treatment . ( A and B ) Sub-cellular localization of the OMmcherry , OMsfGFP , IMmcherry , PERImcherry fusions before ( − ) and after ( + ) plasmolysis treatment ( 0 . 5 M NaCl ) . For each fusion , cells were immobilized in a hybrid flow chamber and imaged before and after injection of the plasmolysis solution . Note that fluorescent cytoplasmic aggregates are observed for the IMmcherry fusion after the plasmolysis treatment ( white arrow ) . Scale bar = 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 004 We next tested whether OM transfer between two cells could be captured at the single cell level . Although most of the recipient cells are stained after 12 hr , the staining is generally weak and both brilliance and the fast-bleaching of mCherry prevented single cell transfer analysis with the OMmCherry probe . Therefore , to maximize our chances to observe a transfer event , we constructed a new probe where super-folder GFP ( sfGFP ) , a fast folding bright variant of GFP ( Pédelacq et al . , 2006 ) , is fused to the type II signal sequence ( OMsfGFP ) . In a bulk transfer assay , OMsfGFP and OMmCherry were transferred with similar efficiencies , showing that OMsfGFP could be used in a single cell assay ( Figure 2—figure supplement 1A ) . To this aim , Myxococcus donor cells expressing OMsfGFP were mixed with recipient cells expressing IMmCherry , and the emergence of dual color cells was monitored over time by time-lapse fluorescence microscopy . As observed in Figure 2A , B and Figure 2—figure supplement 1D , unlabeled recipient cells became fluorescent when they came in contact with OMsfGFP donor cells ( Videos 1 , 2 ) . Several lines of evidence argue that the observed fluorescence increase results from the physical transfer of OMsfGFP: ( i ) , Fluorescence transfer was very rapid , approximately a third of the total donor fluorescence appeared in the recipient strain after 12 min of contact ( Figure 2—figure supplement 1B ) . ( ii ) , Fluorescence initially appeared at the contact zone and subsequently diffused throughout the cell body ( Figure 2B ) . Additionally , green fluorescence was enhanced at the recipient cell periphery , reflecting a membrane localization ( Figure 2—figure supplement 1C ) . ( iii ) , IMmCherry was not exchanged between the two cells ( Figure 2A ) . ( iv ) , Green fluorescence transfer was not detected in recipient cells that were not in contact with donor cells ( Figure 2—figure supplement 1C ) , or in a negative control experiment , when they were mixed with a traA mutant ( Figure 2—figure supplement 1A ) . 10 . 7554/eLife . 00868 . 005Figure 2 . Cell-contact-dependent transfer of OMsfGFP/DiO between single cells . ( A ) sfGFP transfer from a donor OMsfGFP+ ( white contour in lower panel ) cell to a recipient OMsfGFP− IMmCherry+ cell ( orange contour in upper panel ) . Scale bar=1 µm . ( B ) Kymographs of green fluorescence intensities in the positive recipient cell ( top ) and the donor cell ( bottom ) shown in ( A ) . Note that in the recipient cell , green fluorescence diffuses from one half ( t12min to t16min ) to the entire cell body . The Y-axis of each kymograph represents the relative position along the cell body , where 0 represents mid-cell and 1 or −1 the cell poles . The −1 pole is the pole closer to the bottom of the frames for each cell shown in panel ( A ) . ( C ) A DiO+ cell ( white cell contour ) transfers DiO to two unlabeled cells ( orange and green contours ) . Fluorescence and corresponding phase contrast images are shown . Fluorescence fluctuations are shown in pseudo colors where high fluorescence levels appear yellow-green and low fluorescence levels appear blue . Note that the green cell is not immediately in contact with the DiO+ cell . A cell that comes in contact with the DiO+ cell but does not become labeled is shown by a red contour . Scale bar = 1 µm . ( D ) Mean DiO fluorescence intensity over time in the donor cell ( gray square ) , the first positive recipient cell ( green circle ) and the second positive recipient cell ( orange circle ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 00510 . 7554/eLife . 00868 . 006Figure 2—figure supplement 1 . Cell-contact-dependent transfer of OMsfGFP . ( A ) OMsfGFP is transferred in a traA dependent-manner . The transfer kinetics of OMmcherry are also shown for comparison . The percentage of fluorescent recipient cells is plotted as a function of time . For each condition and time point , at least 3000 cells were analyzed in triplicate . ( B ) Mean green fluorescence intensities of distinct cell types over time , donor cell ( gray circle ) , the positive recipient cell ( green circle ) and a negative and isolated recipient cell ( orange circle ) observed in Figure 1D . The dashed line represents the time when the donor and recipient cells establish contact . ( C ) Peripheral membrane staining of the recipient cell shown in Figure 1D ( white Arrow ) . For comparison , the red triangles indicate OMsfGFP− recipient cells ( red triangle ) showing no significant level of green fluorescence during the time course . The inset shows a trans-section fluorescence scan . The Y-axis represents the fluorescence intensity and the X-axis represents the relative position along the scan represented by the dashed line . ( D ) OMsfGFP transfer upon transient cell contact between a donor OMsfGFP+ and a recipient IMmCherry+ cell . Note that OMsfGFP transfer is only observed during the second contact between the donor and the recipient cells ( white arrow ) . Scale bar = 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 00610 . 7554/eLife . 00868 . 007Video 1 . Live observations of cell–contact dependent transfer of OMsfGFP between single cells . Corresponding green fluorescence and red fluorescence are shown . For details see Figure 2 . Pictures were taken every 30 s . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 00710 . 7554/eLife . 00868 . 008Video 2 . Live observations of cell–contact dependent transfer of OMsfGFP between single cells . Corresponding phase contrast , green fluorescence and red fluorescence are shown . Pictures were taken every 30 s . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 008 Because transfer seems highly efficient , physical transfer of OMsfGFP would be expected to lead to a decrease in OMsfGFP levels in the donor cells . While indeed a moderate decrease of sfGFP fluorescence is observed in the donor cell ( Figure 2—figure supplement 1B ) , the steepness of this decrease is likely compensated by the high level of newly synthesized OMsfGFP expressed from the strong pilin ( pilA ) promoter . To circumvent this limitation , we made use of the observation that lipids are also exchanged during transfer ( Pathak et al . , 2012 ) and tested the transfer of DiO , a small C18 backbone hydrophobic lipid dye that intercalates in lipid membranes . DiO-labelled cells contain a finite amount of DiO and given that it is highly diffusible , its dilution upon transfer should be obvious . Importantly , DiO-transfer is Tra-dependent ( Pathak et al . , 2012 ) and thus its exchange between cells would also reflect the transfer dynamics . In a cell mixing experiment , DiO-stained cells were observed to transfer DiO to unlabeled cells upon physical contact ( Figure 2C , D; Video 3 ) . In the example shown in Figure 2C , D , DiO-transfer is also observed to a third cell that is not immediately in contact with the DiO-donor cell but is adjacent to the first transferred cell . Remarkably , the fluorescence of the DiO-labeled cell decreased very rapidly , concomitant with the gradual increase of fluorescence in the adjacent unlabeled cells as if all three cells were connected like communicating vessels ( Figure 2C , D ) . Transfer must require specific contacts ( i . e . , collision of TraA proteins ) because unlabeled cells do not systematically acquire fluorescence when they establish a direct contact with a DiO-donor ( Figure 2C , D; red contoured cell ) . In total , the OMsfGFP and the DiO-staining experiments strongly suggest that we were able to capture transfer events at the single cell level . Transfer can occur between more than two cells , potentially explaining why it is facilitated by cell–cell alignment . 10 . 7554/eLife . 00868 . 009Video 3 . Live observations of cell–contact dependent transfer of DiO between single cells . Corresponding phase contrast and green fluorescence which are displayed in pseudo colors , are shown . For details see Figure 3A . Pictures were taken every 30 s . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 009 The DiO experiment suggests that transferring cells are connected like communicating vessels , which would be explained by the formation of transfer sites where the lipid bilayers of each OM fuse locally , giving rise to a single continuous OM between connected cells . What is the evidence for such connections ? While imaging OMsfGFP expressing cells or DiO stained cells , we frequently observed tubular structures that appeared when two connected cells moved apart ( Figure 3A , Video 4 ) . These tubes were exclusively derived from the OM because they were only stained by sfGFP when observed in two-color cells expressing both OMsfGFP and IMmCherry ( Figure 3B and Figure 1—figure supplement 1B ) . The tubes were also observed by Electron Microscopy ( EM ) , appearing as flexible structures characterized by a diameter of 51 . 4 ± 15 nm ( Figure 3C and Figure 3—figure supplement 1A ) . The structures observed by EM were not type-IV pili because ( i ) , polar pili have a much thinner diameter ( Figure 3—figure supplement 1B ) and ( ii ) , they were observed in a pilA mutant ( Figure 3—figure supplement 1C ) . Interestingly , numerous tubes and vesicles were also observed in large amounts around the cells ( Figure 3C ) , suggesting that lipid materials are also released by the cells ( see below ) . 10 . 7554/eLife . 00868 . 010Figure 3 . Lipid tubes are OM-derived and are observed when cells move apart . ( A ) A lipid tubes formed between two cells expressing OMsfGFP . ( B ) Lipid tubes formed by OMsfGFP IMmCherry-expressing cells ( white arrow ) . Scale bar = 1 µm . ( C ) TEM images of lipid tubes . Tubes appear as continuous and flexible structures emerging from the cell surface ( white arrow ) . Note the presence of vesicles in close proximity with the cell body ( black arrows , left panel ) or around the cells ( right panel ) . Scale bar=250 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 01010 . 7554/eLife . 00868 . 011Figure 3—figure supplement 1 . The tubular extensions are not Type-IV pili . ( A ) Measured diameters of the tubes observed by TEM . The diameters distribution is shown as a boxplot ( n=100 ) . ( B ) Polar Type-IV pili observed in wt cells by Transmission Electron Microscopy . ( Scale bar = 100 nm ) . ( C ) Tubes formed by a pilA mutant cell observed by TEM is shown for comparison . ( Scale bar = 100 nm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 01110 . 7554/eLife . 00868 . 012Video 4 . Formation of OMsfGFP tubes between two cells . Corresponding phase contrast and green fluorescence are shown . Pictures were taken every 30 s . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 012 Motile transferring cells may fuse their OMs locally , forming an ‘OM synapse’ . If such synapses are not resolved when the cells physically separate due to motility , OM tubes would appear because of the tight physical connection . This would predict that tube formation is linked to the transfer mechanism . We first tested whether a tube and the cell OM are continuous . For this , we took advantage of the rapid diffusion of DiO and performed fluorescence recovery after photobleaching ( FRAP ) experiments targeting a tube connected to a single DiO+ cell . DiO fluorescence showed a quick recovery , implying rapid exchange between the tube and the cell DiO pool ( Figure 4A , B ) . We then aimed to capture tube formation between transferring cells . Since the tubes are relatively short lived , we also used DiO staining for these experiments . In the example shown in Figure 4C and Video 5 , DiO is exchanged upon contact between two cells , a tube becomes apparent when the cells move apart , strongly suggesting that tubes are formed between transferring cells . Last , if two cells linked by a tube have continuous OMs , they should exchange DiO , even if they are not immediately in contact . Figure 4D and Video 6 , show a DiO-stained tube formed between a brightly fluorescent cell and a weakly fluorescent cell within a larger group of cells . Remarkably , in the cell with weak fluorescence , the level of fluorescence increased steadily as long as the tube connection was maintained , even though the two cells were not in immediate contact ( Figure 4E ) . When the tube was ruptured the fluorescence decreased due to photo-bleaching ( Figure 4E ) . Fluorescence transfer was strictly confined to the tube-connected cells and no fluorescent fluctuations were observed in the other cells of the group ( Figure 4C , D ) . Thus , tubes allow the rapid exchange of DiO and must be continuous between two connected cells . 10 . 7554/eLife . 00868 . 013Figure 4 . Transfer is driven by transient OM fusion between donor and recipient cells . ( A and B ) Fluorescence recovery after photobleaching ( FRAP ) experiments targeting a tube connected to a single DiO+ cell . Rapid DiO exchange is observed between the tube and the cell body . The cell body is positioned at +1 in ( A ) . ( C ) DiO transfer and formation of DiO+ tubes between two cells . An unstained recipient cell ( orange cell contour ) becomes stained in contact with a DiO donor cell ( white cell contour ) . The grey arrow points to a tube formed between the two cells . Note that transfer only occurs between the two cells although other cells are also in contact with the donor cell . Scale bar = 1 µm . ( D ) DiO is exchanged by tubes connecting two cells . Fluorescence and corresponding phase contrast images between two transferring cells ( green and orange contours ) are shown . Scale bar = 1 µm . ( E ) Mean DiO fluorescence intensity over time in the donor cell ( gray square ) and the recipient cell ( orange circle ) . The vertical dashed line represents the time where the tube connection was ruptured . The horizontal dashed line represents the maximal value of fluorescence intensity observed in the recipient cell . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 01310 . 7554/eLife . 00868 . 014Figure 4—figure supplement 1 . The OMsfGFP or OMmCherry fluorescent probes are not significantly exchanged through the lipid tubes . OMsfGFP-stained tubes formed between an OMsfGFP+ IMmcherry− cell and an OMsfGFP− IMmcherry+ recipient cell . No significant exchange of green fluorescence can be observed through the tubes . Scale bar = 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 01410 . 7554/eLife . 00868 . 015Figure 4—figure supplement 2 . Fluorescence Recovery After Photobleaching ( FRAP ) experiments targeting indicated fluorescent probes . ( A ) Representative fluorescence recovery after FRAP on the cell body of a DiO-stained cell . ( B ) Comparative recovery kinetics of DiO and OMsfGFP after FRAP . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 01510 . 7554/eLife . 00868 . 016Video 5 . Live observations of DiO transfer and formation of DiO+ tubes between two cells . For details see Figure 3A . Corresponding phase contrast and green fluorescence are shown . Pictures were taken every 30 s . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 01610 . 7554/eLife . 00868 . 017Video 6 . Live observations of DiO transfer and formation of DiO+ tubes between two cells . For details see Figure 3C . Corresponding phase contrast and green fluorescence are shown . Pictures were taken every 30 s . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 017 While DiO can be exchanged through the tubes , we did not detect any significant exchange of OMsfGFP or OMmCherry through the tubes ( Figure 4—figure supplement 1 ) . This is probably not surprising because the tubes are narrow extensions and have a relatively short lifespan ( 4 . 2 ± 3 min ) . Thus , large molecules such as OMsfGFP or OMmCherry with lower diffusion rates than DiO ( ∼fourfold , Figure 4—figure supplement 2A , B ) may traffic slowly through the tubes . OM tubes may allow the transfer of small OM molecules , which may be relevant physiologically but they are likely the manifestation of the intimate contact established between transferring cells . The connection of cells by continuous tubes strongly argue that Myxococcus OM-protein transfer involves the formation of a single OM synapse between two connected cells . Where does transfer occur in the Myxococcus biofilm and why is it highly dependent on motility ? Cell alignment in densely packed Myxococcus swarms promotes cell-cell transfer , likely because it favors tight interactions between cells ( Nudleman et al . , 2005; Wei et al . , 2011; Pathak et al . , 2012 ) . However , Cryo-EM studies on the Myxococcus biofilm and our TEM and live observations of the lipid tubes also suggests that large amounts of OM materials may be released in the biofilm matrix , which may constitute a significant transfer reservoir ( Palsdottir et al . , 2009 ) . Interestingly , when we observed gliding cells on cellulose pre-coated EM grids ( ‘Materials and methods’ ) , we found that cells deposit vesicular/tubular material in their wake ( Figure 5A and Figure 5—figure supplement 1 ) . This material was also observed by fluorescence microscopy and must be derived from the OM because dual labeled OMsfGFP/IMmCherry cells deposited trails that were labeled with OMsfGFP but not with IMmCherry ( Figure 5B ) . Gliding Myxococcus cells are known to deposit slime , a self-deposited sugar polymer of unknown composition that facilitates cell adhesion to the underlying substratum ( Ducret et al . , 2012 ) . The slime polymer can be detected selectively by addition of fluorescent Concanavalin A ( ConA-FITC ) in a microfluidic gliding assay ( Ducret et al . , 2012 ) . To test whether the OM materials are specifically associated with the deposited slime , we observed slime trails deposited by an OMmCherry-expressing strain in the presence of ConA-FITC . Figure 5B shows that such cells deposited numerous mCherry+ dots and tubular structures that co-localized with ConA+ trails . EM analysis using gold-labeled ConA confirmed that the deposited OM material is embedded in a sheath of slime polymer ( ‘Materials and methods’ and Figure 5C ) . All together , these results suggest that gliding Myxococcus cells shed a significant amount of their OM during motility and that this material remains attached to the underlying slime polymer . Since gliding Myxococcus cells have long been known to follow trails left by other cells ( Burchard , 1982 ) , a tantalizing possibility is that the transfer of OM materials could also occur when cells follow slime trails , harvesting vesicles and tubes embedded in the slime . Unfortunately , we could not test this possibility directly because the amount of OMsfGFP/OMmCherry labeled material remains too weak to detect a significant transfer to gliding cells by fluorescence microscopy . 10 . 7554/eLife . 00868 . 018Figure 5 . Lipid tubes and vesicles are deposited in slime trails . ( A ) TEM images of lipid tubes deposited in the wake of a moving cell ( left panel ) . A higher magnification view of lipid tubes/vesicles is shown in the right panel . Scale bars = 250 nm . ( B ) Deposition of lipid tubes/vesicles observed by an OMsfGFP+/IMmCherry+ cell . The deposited material is only stained with green fluorescence implying that it is derived from the OM . Scale bar = 1 µm . ( C ) Co-localization of deposited OM materials detected using OMmCherry probe and slime detected using ConA-FITC . Corresponding phase contrast , red fluorescence , green fluorescence and overlay images are shown . Scale bar = 1 µm . ( D ) Lipid tubes/vesicles are embedded in the slime polymer ( Black Arrow ) . Electron dense trails are clearly visible after ConA treatment . White arrows highlight gold particles specifically associated with biotinylated ConA and thus slime . Scale bar = 250 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 01810 . 7554/eLife . 00868 . 019Figure 5—figure supplement 1 . Lipid tubes and vesicles are deposited in slime trails . ( A ) OM materials are deposited in the wake of motile cells and specifically associated with slime . A higher magnification view of lipid tubes/vesicles is shown in the panel ( B ) . Scale bars = 500 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 019
Direct imaging of OM protein transfer between Myxococcus cells uncovers critical aspects of the cell biology and kinetics of transfer . Specifically , we found that the physical contact between two adjacent cells is sufficient to promote transfer of OM proteins and lipids at high efficiency . This explains the results from bulk transfer experiments ( from previous works and reported herein ) suggesting that transfer is a remarkably efficient process . The formation of transient OM tubes between cells is a major indication that transfer indeed occurs by OM fusion: the tubes are continuous extension of cell OM , they form between transferring cells and allow the rapid exchange of lipids . Importantly , the Myxococcus OM transfer system is distinct from reported bacterial nanotubes , which seem to connect the cytosolic contents of connected cells and involve a yet uncharacterized machinery ( Dubey and Ben-Yehuda , 2011 ) . In Myxococcus , the transfer process is restricted to OM proteins and lipids . Transfer only occurs in a subset of cell contact events , suggesting that it is provoked by specific contacts , for example if TraA interactions brought OMs in close apposition locally . OMsfGFP/OMmCherry are fused to type II signal sequences and thus insert in the OM as OM lipoproteins . Since , OM lipoproteins are inserted in the inner leaflet of the OM ( Nakayama et al . , 2012 ) , transfer must involve the fusion of both leaflets of the OM membrane , suggesting that the entire OM is exchanged locally between cells . The formation of OM synapses must therefore create continuity between the periplasmic content of transferring cells . The size of the OM synapse may be estimated from the size of the tubes ( ∼50 nm ) , suggesting that the diameter of the periplasmic lumen may reach up to 20 nm ( for an OM of 10–15 nm thickness [Bayer , 1991; Palsdottir et al . , 2009] ) , providing ample space for periplasmic exchange . However , the PERImCherry probe was poorly if at all exchanged and there is currently no evidence for the physiological transfer of periplasmic proteins , suggesting that OM synapses are not very permeable to periplasmic proteins . Our results also suggest that gliding motility may facilitate transfer by promoting cell–cell alignment but also when cells follow slime trails by incorporating membrane materials embedded in the slime polymer . The shedding of large amounts of membrane materials on the underlying substrate is a common byproduct of surface motility both in eukaryotic and prokaryotic cells . For example , crawling keratinocytes also deposit their plasma membrane due to the activity of acto-myosin motors in focal adhesions ( Kirfel et al . , 2003 ) . In Myxococcus , gliding ( A− ) motility is thought to involve OM dynamics in the form of energized deformations and/or protein movements ( Nan et al . , 2010; Luciano et al . , 2011; Nan et al . , 2011; Sun et al . , 2011 ) . Thus , OM fragments may detach to the substrate due to the interaction between the motility machinery and slime . It is possible that acquisition of the traAB genes allowed Myxococcus cells to recycle this ‘waste’ and co-opt it for cell–cell signaling . A tantalizing possibility would be that slime embedded vesicles contain signals that promote specific recognition , facilitate trail following and promote colony expansion in response to environmental changes . The Myxococcus Tra-dependent cell–cell transfer of OM proteins is a novel mode of bacterial communication that adds to the growing repertoire of bacterial contact-dependent signaling mechanisms . Contrary to known contact dependent protein transfer systems , the type VI secretion ( Silverman et al . , 2012 ) or intercellular nanotubes ( Dubey and Ben-Yehuda , 2011 ) , the distribution of TraA suggests that Tra-dependent OM fusion is restricted to the deltaproteobacteria ( Figure 6A and Figure 6—figure supplement 1 ) . Interestingly , even in Myxococcus xanthus strains , the predicted extracellular N-terminal PA14 domain of TraA shows variability , while the C-terminal region , presumably involved in anchoring to the cell surface is highly conserved ( Figure 6B—figure supplement 2 ) . Importantly , TraA acts as both key and lock for transfer to occur ( Pathak et al . , 2012 ) . Therefore , as already suggested , OM-transfer may have evolved to regulate interactions between cells of the same kin . tra mutants do not show motility or developmental defects in pure culture and thus the contribution of OM transfer to Myxococcus social behaviors is unclear ( Pathak et al . , 2012 ) . Interestingly however , mixing tra mutants with WT cells perturbs motility and development profoundly , consistent with a role in the control population dynamics ( Pathak et al . , 2012 ) . OM exchange by transient fusion may be more widespread than suspected , especially because it is not easily unmasked and likely does not employ a conserved molecular system . Indeed , membrane vesicles and tubes have been observed in other biofilm-forming proteobacteria ( Schooling and Beveridge , 2006; Schooling et al . , 2009 ) and could well be involved in OM exchange behaviors . 10 . 7554/eLife . 00868 . 020Figure 6 . Distribution of TraA is restricted to the deltaproteobacteria . ( A ) TraA homologues in Myxococcus xanthus DK1622 ( gi|108763680 ) , Myxococcus stipitatus ( gi|442324418 ) , Corallococcus coralloides ( gi|383459429 ) , Myxococcus fulvus ( gi|338532052 ) , Stigmatella aurantiaca ( gi|310818240 ) , Cystobacter fuscus ( gi|444910311 ) , Haliangium Ochraceum ( gi|262197466 ) , Sorangium Cellulosum ( gi|162451690 ) . ( B ) The PA-14 domain is variable in Myxococcus xanthus strains . The conservation of the Ct domain is shown for comparison . Sequence database access numbers are shown to the left . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 02010 . 7554/eLife . 00868 . 021Figure 6—figure supplement 1 . TraA homologues in deltaproteobacteria . ClustalW alignment of TraA homolog: Myxococcus xanthus DK1622 ( gi|108763680 ) , Myxococcus stipitatus ( gi|442324418 ) , Corallococcus coralloides ( gi|383459429 ) , Myxococcus fulvus ( gi|338532052 ) , Stigmatella aurantiaca ( gi|310818240 ) , Cystobacter fuscus ( gi|444910311 ) , Haliangium Ochraceum ( gi|262197466 ) , Sorangium Cellulosum ( gi|162451690 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 02110 . 7554/eLife . 00868 . 022Figure 6—figure supplement 2 . ClustalW alignment of TraA in Myxococcus xanthus strains . Note that most amino acids variations are localized in the first 300 N-terminal residues region encompassing the so-called PA14 domain . DOI: http://dx . doi . org/10 . 7554/eLife . 00868 . 022
Primers and plasmids used in this study are listed in Supplementary file 1A , B . See also Supplementary file 1C for strains and their mode of construction . M . xanthus strains were grown at 32°C in CYE rich media as previously described ( Bustamante et al . , 2004 ) . When necessary antibiotics were added: kanamycin ( Km ) at 50 mg/ml , tetracycline ( Tc ) at 12 mg/ml , for M . xanthus , and Km at 50 mg/ml , or Tc at 100 mg/ml for E . coli . Constructs were confirmed by phenotypes , restriction analysis and DNA sequencing . Plasmids were introduced in M . xanthus by electroporation . For colony assays , cells were first grown in CYE , harvested and resuspended to a final concentration of 4 × 109 cfu/ml . Fluorescent donors ( DZ2 PpilA–OMss–mCherry , DZ2 PpilA–IMss–mCherry , DZ2 PpilA–PERIss–mCherry or DZ2 PpilA–OMss–sfGFP ) were mixed 1:1 with fluorescent recipients ( DZ2 aglZ-YFP or DZ2 PpilA–IMss–mCherry ) . Strain mixtures were then spotted on CYE plates ( 1 . 5% agar ) . At various times , cells were scraped from agar plates and resuspended in TPM ( 10 mM Tris [pH 7 . 6] , 8 mM MgSO4 , 10 mM KH2PO4 ) and spotted on agar pads to be counted directly under the micrscope . For each condition and time point , at least 3000 cells were analyzed in triplicate . For single-cell level assays , cells were first grown in CYE , harvested , resuspended to a final concentration of 1 × 107 cfu/ml . To clearly differentiate fluorescent donor to recipient , OMss–sfGFP expressing donors was mixed 1:1 with IMss–mCherry expressing recipient . Cells were then imaged under on agar pads for up to 1 hr . Cells were first grown in CYE , harvested , resuspended to a final concentration of 1 × 107 cfu/ml . To stain cells , 1 µl of Vybrant DiO Cell-Labeling Solution ( Invitrogen , Saint Aubin , France ) was added to 1 ml of cells and incubated for 30 min in the dark at 32°C under agitation . Cells were then pelleted by centrifugation , and washed four times with 1 ml TPM . Cells were then imaged on agar pads for up to 1 hr . Time lapse experiments were performed as previously described ( Ducret et al . , 2009 ) . Microscopic analysis was performed using an automated and inverted epifluorescence microscope TE2000-E-PFS ( Nikon , Champigny sur Marne , France ) . The microscope is equipped with ‘The Perfect Focus System’ ( PFS ) that automatically maintains focus so that the point of interest within a specimen is always kept in sharp focus at all times , in spite of any mechanical or thermal perturbations . Photobleaching was performed with a 488 nm laser . The bleach region of interest ( ROI ) was a circular region ∼1 µm diameter . The ROI was uniformly bleached with a 200 ms laser exposition at 100% intensity . Images were recorded with a CoolSNAP HQ 2 ( Roper Scientific , Roper Scientific SARL , France ) and a 100x/1 . 4 DLL objective . All fluorescence images were acquired with appropriate filters with a minimal exposure time to minimize bleaching and phototoxicity effects . Cell tracking was performed automatically using a previously described macro under the METAMORPH software ( Molecular devices , Evry , France ) ( Ducret et al . , 2009 ) . Typically , the images were equalized , straightened and overlaid under both ImageJ 1 . 40g ( National Institute of Health , United States ) and METAMORPH . Kymographs display the maximum intensity values of green or red signal along the long axis of the cell for each frame , using a 0 . 2 µm wide region . The Y-axis of each kymograph represents the relative position along the cell body , where 0 represents the mid-cell and 1 or −1 the cell poles . The −1 pole is always the pole closer to the bottom of the frames shown in panel . To know the respective diffusions rates of the probes used in this study , we measured the diffusion constant of the DiO and the outer membrane probe OMsfGFP using fluorescence recovery after photobleaching ( FRAP ) . As observed in Figure 4—figure supplement 2A , B , FRAP analysis provided a diffusion coefficient for DiO ( DDiO = 8 . 1 ± 1 . 3 µm2/s ) at least four times faster than the OMsfGFP probe ( DsfGFP = 2 . 3 ± 0 . 5 µm2/s ) . These values are similar to diffusion constants measured for respectively outer membrane probes and outer membrane proteins in E . coli ( Tocanne et al . , 1994; Chow et al . , 2012 ) . For TEM experiments carbon-coated copper grids were first coated with carboxymethylcellulose . Briefly , carbon-coated copper grids were covered with 30 µl of carboxymethylcellulose sodium salt ( Medium viscosity , Sigma-Aldrich , Inc . , St Louis , MO ) diluted in ultrapure water . After 15 min of incubation at room temperature , the coating solution was removed by performing two successive washes with ultrapure water . Carbon-coated copper grids were then covered with the cell suspension previously washed and resuspended in TPM containing 100 mM of CaCl2 ( TPM-Ca2+ ) . After 1 or 15 min incubation , unattached cells were removed by performing two successive washes with TPM . For Lectin-Gold staining procedure , carbon-coated copper grids were first covered for 30 min with 30 µl of ConcanavalinA ( ConA ) -Biotin conjugated ( ConA-Biotin; Sigma-Aldrich , Inc . ) diluted in TPM-Ca2+ to a final concentration of 100 µg/ml and then washed four times with TPM-Ca2+ . The grids were then incubated for 15 min with 10 nm gold-conjugated streptavidin ( Invitrogen , Saint Aubin , France ) diluted in TPM-Ca2+ ( 1/500 ) and then washed four times with TPM-Ca2+ . Grids were postfixed with 1% glutaraldehyde , washed once with TPM , washed four times with water , stained with 1% ( wt/vol ) uranyl acetate , dried , and imaged with a JEM-1011 transmission electron microscope operated at 100 kV . Cells were first observed on standard TEM grids . As observed in Figure 3C , tubes appear as continuous and flexible structures emerging from the cell surface . Unattached tubes and vesicles were also observed in the vicinity of cells suggesting that this material was also released by cells in the media . On standard uncoated TEM grids Myxococcus cells do not glide ( data not shown ) precluding any observation of these structures when cells were moving . To deal with this limitation we pre-coated the TEM grids with cellulose , a linear polysaccharide composed of β ( 1→4 ) linked D-glucose units and previously known to support gliding motility of M . xanthus ( Ducret et al . , 2013 ) . As observed on Figure 5A , D and Figure 5—figure supplement 1A , B , linear depositions of tubes and vesicles were observed in the wake of motile cells when cells were deposited on pre-coated TEM grids and incubated for 20 min before fixation . Linear depositions were not observed ( i ) when motile cells were fixed directly after deposition , ( ii ) with non-motile cells , and ( iii ) with A− cells ( A−S+ strain ) , strongly suggesting that these depositions are specifically associated with the A-motility . Since gliding Myxococcus cells are known to deposit slime , a self-deposited sugar polymer that facilitates cell adhesion to the underlying substratum , we then tested if the deposited material is associated with slime . The slime polymer can be detected selectively by addition of Concanavalin A ( ConA ) . When ConA was added to the TEM grids , electron-dense trails appeared in the wake of motile cells . The tubes and vesicles were clearly embedded in these trails . To verify that the trails are indeed labeled by ConA , we use Biotinylated ConA and colloidal gold-streptavidin . As observed on Figure 5D , gold particles were exclusively associated with the trail proving that vesicles and tubes are associated with the slime polymer . To verify the proper localization of each probe , cells expressing OMmCherry , IMmCherry , PERImCherry or OMsfGFP/IMmCherry were subjected to plasmolysis ( Lewenza et al . , 2008; Wei et al . , 2011 ) . As predicted , OMmCherry , OMsfGFP and PERImCherry retained their envelope localization when IMmCherry probe formed fluorescent cytoplasmic aggregates ( Figure 1—figure supplement 1A , B ) , indicating that following plasmolysis only the IM fusions collapse with the inner membrane . Plasmolysis was performed as previously described ( Lewenza et al . , 2008 ) . Briefly , log phase cells were washed , resuspended in TPM buffer and then immobilized in a hybrid flow chamber ( Ducret et al . , 2009 ) . Cells were imaged before ( control ) and after injection of the plasmolysis solution ( 0 . 5 M NaCl ) . Lectin staining was performed as previously described ( Ducret et al . , 2012 ) . Briefly , cells were injected in a flow chamber pre-coated with Chitosan . Immediately prior to the experiments , the Concanavalin-A stock solution were diluted to a final concentration of 20 µg/ml in TPM containing 100 mM of CaCl2 and 100 µg/ml of bovine serum albumin ( BSA ) . The mixture was then injected into the flow chamber . After 20 min of incubation , the lectins were washed out with TPM . | Bacteria studied in the laboratory are , in general , readily amenable to culture , and they easily form colonies when grown on agar plates . In the wild , however , many bacteria exhibit a range of more complex behaviors , including the growth of super-organisms that contain many cells . The bacterium Myxococcus xanthus can exist either as single cells or as a super-organism . Each cell has an inner and outer plasma membrane separated by a periplasmic space . Previous work has found that individual cells communicate with each other by exchanging the contents of their outer membranes , and that these swaps can govern multicellular behavior . Membrane exchange is known to depend on both donor and recipient cells having the proteins TraA and TraB . TraA proteins are similar to the adhesion factors that hold cells together , and they are found in many species: this suggests that TraA therefore might help the outer membranes of cells to fuse so that they can swap materials . The role of TraB is not known at present . To investigate membrane exchange more closely , Ducret et al . measured the transfer of fluorescent proteins from the periplasm and the inner and outer membranes of the donor cell to the recipient cell , as well as the transfer of fluorescent lipids from the donor’s outer membrane . Both proteins and lipids from the outer membrane were transferred rapidly ( within minutes ) ; although a small amount of protein transfer from the periplasmic space was observed after 36 hr , there was no transfer from the inner membrane . As in previous studies , exchange depended on the presence of TraA . Ducret et al . observed that contact between two cells was sufficient to stimulate transfer of proteins and lipids from the outer membrane . But not all contacts led to a transfer . Importantly , when cells that had swapped fluorescent membrane components moved apart , they appeared to remain connected by tubular structures , suggesting that an inter-membrane junction must form to allow proteins and lipids to be transferred between the cells . This junction is referred to as an outer-membrane synapse . Ducret et al . also noted another phenomenon: cells shed pieces of membrane as they moved across surfaces or separated after outer membrane exchange . This suggests that both synapse formation after direct cell-to-cell contact and the shedding of membrane components can help to propagate bacterial signals , enabling population-wide behavioral changes , including the formation or collapse of super-organisms . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2013 | Direct live imaging of cell–cell protein transfer by transient outer membrane fusion in Myxococcus xanthus |
Regulation of size and growth is a fundamental problem in biology . A prominent example is the formation of the mitotic spindle , where protein concentration gradients around chromosomes are thought to regulate spindle growth by controlling microtubule nucleation . Previous evidence suggests that microtubules nucleate throughout the spindle structure . However , the mechanisms underlying microtubule nucleation and its spatial regulation are still unclear . Here , we developed an assay based on laser ablation to directly probe microtubule nucleation events in Xenopus laevis egg extracts . Combining this method with theory and quantitative microscopy , we show that the size of a spindle is controlled by autocatalytic growth of microtubules , driven by microtubule-stimulated microtubule nucleation . The autocatalytic activity of this nucleation system is spatially regulated by the limiting amounts of active microtubule nucleators , which decrease with distance from the chromosomes . This mechanism provides an upper limit to spindle size even when resources are not limiting .
A general class of problems in biology is related to the emergence of size and shape in cells and tissues . Reaction diffusion mechanisms have been broadly successful in explaining spatial patterns in developmental biology as well as some instances of intracellular structures ( Turing , 1952; Howard et al . , 2011 ) . The mitotic spindle , a macromolecular machine responsible for segregating chromosomes during cell division , is thought to be a classic example of such reaction diffusion processes . A diffusible gradient of the small GTPase Ran emanating from chromosomes has been shown to trigger a cascade of events that result in the nucleation of microtubules , the main building blocks of the spindle ( Kaláb et al . , 2006; Caudron et al . , 2005 ) . The spatial distribution of microtubule nucleation is key for understanding size and architecture of large spindles . This is because microtubules in these spindles are short and turnover rapidly in comparison to the entire structure ( Redemann et al . , 2017; Brugués et al . , 2012; Needleman et al . , 2010 ) . The mechanisms underlying the spatial regulation of microtubule nucleation , however , are still unclear ( Prosser and Pelletier , 2017; Petry , 2016 ) . One possibility is that the interplay between Ran-mediated nucleation and microtubule turnover governs spindle assembly ( Kaláb et al . , 2006; Caudron et al . , 2005 ) . However , the role of the Ran gradient in determining spindle size is still controversial . For instance , in cell culture systems , the length scale of the Ran gradient does not correlate with spindle size ( Oh et al . , 2016 ) . A second possibility is that autocatalytic growth accounts for spindle assembly via microtubule-stimulated microtubule nucleation ( Petry et al . , 2013; Goshima et al . , 2008; Loughlin et al . , 2010; Ishihara et al . , 2016 ) . However , autocatalytic mechanisms suffer from the fact that their growth is hard to control . Although autocatalytic growth can be regulated by limiting the catalyst , such mechanisms are unlikely to function in the large cells of developing eggs such as Xenopus , where resources are not limiting ( Crowder et al . , 2015 ) . Understanding the role of microtubule nucleation in setting the size of spindles is limited by the fact that little is known about the rate , distribution , and regulation of microtubule nucleation in spindles ( Prosser and Pelletier , 2017; Petry , 2016 ) . This is partly because of the lack of methods to measure microtubule nucleation in spindles . Here , we measured microtubule nucleation in spindles assembled in Xenopus laevis egg extract using laser ablation . We show that microtubule nucleation is spatially dependent and requires physical proximity to pre-existing microtubules . Our findings are consistent with a theoretical model in which autocatalytic microtubule nucleation is regulated by the amount of the active form of spindle assembly factors . This mechanism provides a finite size for spindles even when resources are not limiting .
Microtubules grow from the plus ends while minus ends remain stable ( Howard , 2001 ) . Thus , the location of minus ends functions as a marker for microtubule nucleation . However , in spindles microtubules constantly flux towards the poles ( Mitchison , 1989 ) , and measuring the location of a microtubule minus end at a particular time does not correspond to its original site of nucleation ( Brugués et al . , 2012 ) . To decouple microtubule transport from microtubule nucleation , we inhibited kinesin-5 ( Eg5 ) in spindles assembled in Xenopus laevis egg extracts . This inhibition stops microtubule transport and leads to the formation of radially symmetric monopolar spindles ( monopoles ) that have a similar size as regular spindles ( Miyamoto et al . , 2004; Skoufias et al . , 2006 ) ( Figure 1A , Figure 1—figure supplement 1 and Video 1 ) . The location of minus ends in these monopoles corresponds to the location of microtubule nucleation . Three independent measurements show that inhibiting microtubule transport does not affect dynamic parameters of microtubules . First , microtubules in these structures polymerize at 20 . 9 ± 5 . 1 µm/min ( N = 7 monopoles , Figure 1A and Video 1 ) , which is indistinguishable from the polymerization velocity in spindles , 22 . 7 ± 8 . 4 µm/min ( N = 4 spindles ) . Second , microtubules from monopolar and control spindles depolymerize at the same velocity ( 33 . 5 ± 6 . 4 µm/min and 35 . 9 ± 7 . 3 µm/min respectively , see Figure 1—figure supplement 2 ) . Third , microtubule lifetime distributions of monopolar spindles , measured by single-molecule microscopy of tubulin dimers , give an average lifetime of 19 . 8 ± 2 . 2 s , consistent with similar measurements in regular spindles ( Needleman et al . , 2010 ) ( Methods and materials and Video 2 ) . To localize microtubule nucleation events , we measured the density of minus ends in monopolar spindles by analyzing synchronous waves of microtubule depolymerization from laser cuts similar to Ref . ( Brugués et al . , 2012 ) . Briefly , cut microtubules rapidly depolymerize from the newly generated plus ends , while the new minus ends remain stable . The minus end density at the location of the cut can then be obtained from the decrease of the microtubule depolymerization wave , but as opposed to Ref . ( Brugués et al . , 2012 ) , our method resolves the minus end locations with a single laser cut ( see Figure 1B–C , Figure 1—figure supplement 3 , Figure 2—figure supplement 1 , Video 3 , Video 4; a detailed explanation of the method can be found in the Methods and materials and Appendix 1 ) . We define the microtubule nucleation profile at a distance r from the center of the monopole as the number of minus ends per unit length at r divided by 2πr . We measured the microtubule nucleation profile across the entire structure by performing laser cuts at different distances from the center of the monopoles . These measurements revealed that microtubule nucleation extends throughout monopoles , with the highest nucleation near the center and monotonically decreasing far from the center ( see Figure 1D ) , indicating that the strength of microtubule nucleation is spatially regulated . Several mechanisms have been proposed to regulate microtubule nucleation . From a biophysical perspective , these mechanisms can be categorized into two scenarios: ( i ) microtubule-dependent nucleation , in which a pre-existing microtubule stimulates the nucleation of a new microtubule , or ( ii ) microtubule-independent nucleation , in which factors other than pre-existing microtubules ( e . g . diffusible cues in the cytoplasm ) stimulate nucleation ( Prosser and Pelletier , 2017; Petry , 2016; Petry et al . , 2013; Goshima et al . , 2008; Clausen and Ribbeck , 2007; Ishihara et al . , 2014a; Carazo-Salas et al . , 2001 ) . If microtubule nucleation depends on pre-existing microtubules , altering microtubule stability should change the nucleation profile – a microtubule that exists for a longer time would have a higher probability to stimulate the creation of more microtubules . To test this scenario , we increased microtubule stability by inhibiting the depolymerizing kinesin MCAK ( Walczak et al . , 1996 ) using antibodies . MCAK inhibition led to a dramatic increase in monopole size ( see Figure 2A ) . Both the average length and stability of microtubules increased threefold after inhibition ( Figure 2B–C and Figure 2—figure supplement 1 ) as assessed by laser ablation ( 8 . 0 ± 0 . 3 µm versus 23 . 6 ± 3 . 6 µm , see Methods and materials and [Brugués et al . , 2012] ) and single microtubule lifetime imaging ( 19 . 8 ± 2 . 2 s versus 60 . 4 ± 4 . 4 s ) , Video 2 , Video 5 and Supplementary Figure 2 . These measurements are consistent with MCAK modifying the catastrophe rate ( Walczak et al . , 1996; Tournebize et al . , 2000 ) . We measured microtubule nucleation in this perturbed condition and found that the nucleation profile extends further from the center of the monopole , has a larger amplitude , and decays over a larger distance with respect to control monopoles ( Figure 2D ) . Therefore , the number and spatial distribution of nucleated microtubules does indeed scale with microtubule stability in monopolar spindles , which is inconsistent with microtubule-independent nucleation . One possibility is that MCAK-inhibition could by itself increase nucleation independently of microtubules . However , this would only lead to an overall increase of the amplitude of microtubule nucleation , which alone would not be sufficient to account for the dramatic change in the spatial dependence of the nucleation profile we observe in Figure 2D . Thus , microtubule nucleation in these structures depends on the presence and dynamics of microtubules . The presence and dynamics of microtubules could alter microtubule nucleation in two ways: microtubules could nucleate indiscriminately in the cytoplasm without requiring microtubules , but their presence concentrates active nucleators through transient interactions with microtubules ( Oh et al . , 2016 ) , or alternatively , microtubules could directly nucleate new microtubules , requiring active nucleators to bind to microtubules to initiate nucleation . In the latter case , the presence of a microtubule is essential for the nucleation process , whereas in the former , microtubules can still nucleate in the absence of microtubules . To test whether microtubule nucleation requires physical proximity to pre-existing microtubules ( e . g . , a branching process [Petry et al . , 2013] ) , we locally blocked microtubule polymerization by adding inert obstacles near the center of monopoles , at locations where nucleation should be expected according to our measurements ( Figures 2D and 3A , and Video 7 ) . These localized obstacles cannot prevent the diffusion of nucleators , but would prevent microtubules that polymerize towards them to extend further . Consistent with microtubule-stimulated nucleation , the presence of these obstacles inhibited nucleation of new microtubules behind the obstacles , as in a shadow cast by light , whereas microtubules nucleated further around the obstacles , creating a sharp boundary , see Figure 3A . These results suggest that monopolar spindles grow to a size larger than an individual microtubule by microtubule-stimulated microtubule nucleation in physical proximity to pre-existing microtubules , which creates an autocatalytic wave of microtubule growth . For a microtubule structure to have a finite size through an autocatalytic process , each microtubule at the periphery must create on average less than one microtubule at steady state , otherwise the number of microtubules would increase exponentially and the structure would grow unbounded ( Ishihara et al . , 2016 ) . However , measurements of the temporal evolution of microtubule mass in spindles show indeed an initial phase of exponential growth ( Figure 3—figure supplement 1 and ( Clausen and Ribbeck , 2007; Dinarina et al . , 2009 ) . This is also consistent with the observation of microtubules creating more than one microtubule on average when inducing bulk microtubule branching by adding TPX2 and constitutively active Ran ( RanQ69L ) in extracts ( Petry et al . , 2013 ) . These observations raise the question of how spindles reach a finite size through autocatalytic growth ( as in the control and MCAK-inhibited monopoles ) . One possibility is that microtubule dynamics change as a result of limiting amounts of tubulin or microtubule-associated proteins ( Good et al . , 2013; Hazel et al . , 2013 ) . However , since our cell-free system is not confined , availability of tubulin and microtubule-associated proteins is not limiting . Furthermore , inhibiting MCAK leads to larger monopoles with a microtubule polymerization velocity that is indistinguishable from smaller control monopoles ( 20 . 9 ± 5 . 1 µm/min and 18 . 8 ± 5 . 4 µm/min respectively , Video 6 , Video 1 , Figure 3—figure supplement 2 and Table 1 ) , suggesting that the availability of tubulin appears not to be diffusion-limited . Finally , microtubule dynamics do not change spatially throughout MCAK-inhibited monopoles ( Figure 3—figure supplement 2 ) , indicating that spatial variations of tubulin amount or microtubule dynamics cannot explain the finite size of these structures . Another possibility is that microtubule nucleation is limiting . It has been shown that RanGTP is required for spindle assembly . RanGTP is created only in the vicinity of chromosomes ( through the ran nucleotide exchange factor RCC1 ) , which in turn releases spindle assembly factors ( SAFs ) responsible for nucleating microtubules ( Kaláb et al . , 2006; Caudron et al . , 2005 ) . Since the active SAFs are naturally limited by their spatially restricted generation , a limiting amount of an active microtubule nucleation factor would therefore be a good candidate as the limiting component for both autocatalytic growth and size regulation . To test this idea , we added constitutively active Ran ( RanQ69L ) , to pre-existing monopolar spindles . A limiting pool of active nucleators implies that ( i ) activating nucleators everywhere in the cytoplasm would lead to unbounded microtubule growth in the monopole ( similar to large interphase asters in embryos [Wühr et al . , 2010] ) , and ( ii ) new microtubules should nucleate from the pre-existing microtubules of the structure . Adding RanQ69L to pre-existing monopoles immediately started nucleation of new microtubules preferentially at the edge of the pre-existing structures in a wave-like fashion , consistent with microtubule-stimulated growth ( Figure 3B and Videos 8 and 9 ) . This result further suggests that other limiting components that regulate microtubule dynamics alone cannot account for this growth . Taken together , these measurements show that the amount of active nucleators , which is limited by the availability of RanGTP , limits the size of monopolar spindles and is responsible for the bounded growth of these structures . To test whether a limited pool of active nucleators can quantitatively account for the size and microtubule nucleation in these structures , we developed a biophysical model of autocatalytic microtubule nucleation ( see Figure 4A and Appendix 1 ) . In our model , inactive nucleators are present throughout the cytoplasm and can be activated at the surface of chromosomes , which is a simplification of the activation of SAFs by RanGTP . The total amount of active nucleators depends on the balance between the rate of activation at the chromosomes and the rate of inactivation ( accounting for sequestration , hydrolysis , or other processes ) . Once activated at the chromosomes , nucleators can diffuse in the cytoplasm , bind , and unbind from microtubules . When bound to microtubules , active nucleators can nucleate new microtubules at a certain rate , and the newly nucleated microtubules maintain the same polarity as the mother microtubule ( Petry et al . , 2013 ) . This process leads to an autocatalytic wave as a consequence of the self-replicating activity of an extended object . In contrast to a reaction diffusion process , the front propagation is independent of microtubule diffusion and only depends on microtubule dynamics . In our model , the amount and dynamics of active nucleators are the same for both control and MCAK-inhibited monopoles , leading to the prediction that the two microtubule density profiles would only differ in a parameter controlling the microtubule length or lifetime ( see Appendix 1 ) . In particular , both profiles should scale to each other without any fitting parameters by changing the microtubule length as measured independently by laser ablation . To test this prediction , we measured the radial profile of microtubule density of control and MCAK-inhibited monopoles ( Figure 4B ) : These microtubule density profiles are qualitatively different –the density of MCAK-inhibited monopoles increases initially and decreases after reaching a maximum , whereas the control monopole decreases monotonically from the origin . Remarkably , both profiles collapse into each other after the parameter-free rescaling of the MCAK-inhibited monopole predicted by the model ( see Appendix 1 and Figure 4C ) . To test the model beyond scaling , we fit the MCAK-inhibited profile with two independent parameters and an arbitrary amplitude of the density profile , which agrees quantitatively with the data ( see Appendix 1 and Figure 4B ) . By fixing all parameters to the values obtained by this fit ( which are the same for the control monopole , see Appendix 1 , Table 2 ) and using the measured average microtubule length for the control monopole ( Methods and materials , Table 1 ) , the model predicts the control monopole microtubule profile . Finally , we can also predict the MCAK-inhibited and control microtubule nucleation profiles from the fitted parameters up to an arbitrary amplitude ( common for both profiles ) ( Figure 4D ) . Remarkably , this prediction is also consistent with flux-corrected microtubule nucleation in regular spindles obtained by laser ablation ( see Methods and materials , Figure 4D green circles , Videos 10 and 11 ) , showing that the same nucleation mechanism holds for regular spindles . Thus , our model for autocatalytic microtubule nucleation accounts for both the microtubule density and nucleation profiles .
Our data and model are consistent with an autocatalytic mechanism in which microtubule-stimulated microtubule nucleation controls growth in Xenopus laevis egg extract spindles . This process is spatially regulated by a gradient of active nucleators that is established by the interplay between the Ran gradient and microtubule dynamics . Microtubules regulate the nucleator activity because they act as the substrate where active nucleators need to bind to nucleate microtubules . Chromatin acts as a trigger for an autocatalytic wave of microtubule nucleation , and at the same time limits spindle size by controlling the amount of active nucleators through RanGTP . This suggests that the amount of active Ran can tune spindle length , and resolves its controversial relation to spindle length regulation: while a diffusion and inactivation process has a characteristic length scale independent of the amplitude of the gradient – set by the ratio of the squared root of the diffusion and inactivation rate – here we show that both the length scale and amplitude of the gradient of nucleators are involved in regulating the size and mass of spindles . Since the length scale of the gradient is amplified by microtubule-stimulated nucleation , the relevant length scale for setting the size is the distance at which a microtubule generates one or fewer microtubules . Our proposed mechanism therefore allows regulation of spindle size and mass by two means , although microtubule nucleation is the principal control parameter , microtubule dynamics can still fine tune the spindle length ( Reber et al . , 2013 ) . Although our results are restricted to Xenopus laevis spindles , we hypothesize that a similar mechanism may also apply to other spindles with a large number of microtubules . This would be consistent with the fact that components involved in microtubule branching have been identified in many eukaryotic systems ( Dasso , 2002; Hsia et al . , 2014; Sánchez-Huertas and Lüders , 2015 ) . However , further experiments are needed to test this hypothesis . An autocatalytic nucleation process implies that microtubule structures are capable of richer dynamical behaviors than those arising from the classic view of random nucleation in the cytoplasm via a diffusible gradient . Beyond producing finite-sized structures like spindles and ensuring that new microtubules keep the same polarity as the pre-existing ones , it also allows for a rapid switch into unbounded wave-like growth if nucleators become active throughout the cytoplasm . Indeed , the growth of large interphase asters has been hypothesized as a chemical wave upon Cdk1 activation ( Chang and Ferrell , 2013; Ishihara et al . , 2014b ) . These properties , characteristic of excitable media , provide a unified view for the formation of spindles and large interphase asters in embryos ( Ishihara et al . , 2014a ) within a common nucleation mechanism . However , microtubule nucleation differs from regular autocatalytic processes in reaction-diffusion systems such as Fisher-waves and Turing mechanisms ( Turing , 1952; Fisher , 1937 ) in that its growth does not rely on diffusion or advection . Instead , the process of branching displaces the center of mass of the structure . Thus , it emerges as consequence of the finite extension and dynamics of the reactant ( microtubules ) . The interplay between autocatalytic growth and fluxes driven by motors could lead to general principles of pattern formation and cytoskeletal organization in cells .
Cytostatic factor ( CSF ) -arrested Xenopus laevis egg extract was prepared as described previously ( Hannak and Heald , 2006; Murray , 1991 ) . In brief , unfertilized oocytes were dejellied and crushed by centrifugation . After adding protease inhibitors ( LPC: leupeptin , pepstatin , chymostatin ) and cytochalasin D ( CyD ) to a final concentration of 10 µg/ml each to fresh extract , we cycled single reactions to interphase by adding frog sperm ( to 300–1000 sperm/µl final concentration ) and 0 . 4 mM Ca2+ solution , with a subsequent incubation of 1 . 5 hr . While fresh CSF extract containing LPC and CyD was kept on ice , all incubation steps were performed at 18–20°C . The reactions were driven back into metaphase by adding 1 . 3 volumes of fresh CSF extract ( containing LPC and CyD ) . Spindles formed within 1 hr of incubation . To inhibit kinesin-5 ( Eg5 ) in spindles , S-Trityl-L-Cysteine ( STLC ) was added to the reactions to a final concentration of 200 µM . Transitions to monopolar spindles were observed within 30–60 min of incubation . To inhibit the depolymerizing kinesin MCAK in monopolar spindles , we added anti-MCAK antibodies to a final concentration of ∼30 µg/ml ( kind gift from R . Ohi ) . MCAK-inhibited structures reached their steady-state after ∼20 min . Alternatively , we added RanQ69L ( kind gift from K . Ishihara ) to pre-formed monopoles to a final concentration of 30 or 10 µM and imaged immediately . In the control reactions , the same concentrations were added to extract reactions in the absence of pre-formed structures and imaged after ∼20 min incubation . The lower the RanQ69L concentration the later Ran asters formed . Conversely , if a pre-existing structure was present , microtubule nucleation immediately started at the periphery with subsequent growth of the structure . The growth of pre-existing monopolar spindles stopped with the appearance of Ran asters in bulk ( after ∼20 min depending on the concentration of RanQ69L ) , consistent with the sequestering of the additional nucleators activated by RanQ69L . Prior to imaging , Atto565 labeled purified porcine tubulin ( purified according to Ref . [Castoldi and Popov , 2003] ) and Höchst 33342 were added to the reactions to a final concentration of 150 nM and ∼16 µg/ml , respectively , to visualize microtubules and DNA . Control and MCAK-inhibited monopolar spindles were imaged using a Nikon spinning disk microscope ( Ti Eclipse ) , an EMCCD camera ( Andor iXon DU-888 or DU-897 ) , a 60 × 1 . 2 NA water immersion objective , and the software AndorIQ for image acquisition . The room was kept at constant 20∘C . Monopolar spindles after the addition of RanQ69L were imaged using a Nikon wide-field epifluorescence microscope ( Ti Eclipse ) , an sCMOS camera ( Hamamatsu Orca Flash 4 . 0 ) , and a 20 × 0 . 75 NA objective . In this case , image acquisition was performed using µManager ( Edelstein et al . , 2014 ) . The growth of microtubule structures in the presence of obstacles was imaged using a Nikon total internal reflection fluorescence ( TIRF ) microscope ( Ti Eclipse ) , equipped with an Andor iXon3 DU-897 BV back-illuminated EMCCD camera , a 100 × 1 . 49 NA oil immersion objective , and the Nikon software NIS elements . The femtosecond laser ablation setup was composed of a mode-locked Ti:Sapphire laser ( Coherent Chameleon Vision II ) oscillator coupled into the back port of the Nikon spinning disk microscope and delivering 140 fs pulses at a repetition rate of 80 MHz . Cutting was performed using a wavelength of 800 nm and typically a power of 150 mW before the objective . The sample was mounted on a piezo stage that positioned the sample in 3D with sub-micrometer precision . The laser cutting process was automatically executed by a custom-written software that controlled the mechanical shutter in the beam path and moved the piezo stage to create the desired shape of the cut . Lines and circular cuts were performed in several planes to cover a total depth of ∼1–2 µm around the focal plane . We adapted the size and geometry of the cut shapes to each spindle or monopolar structure . Cutting was finished within 2 s . Images were acquired at least every 0 . 5 s during the cutting procedure as well as for ∼1 min after the cut . The depolymerization wave typically disappeared within 30 s . Each microtubule structure was cut only once . We analyzed the depolymerization waves using a custom-written Python code . Briefly , for a given cut at position r , we subtracted the intensities of images ( raw data ) with a time difference δt of 2–3 s to get the differential intensities I ( x , ϕ , t;r ) , where x is the radial coordinate , ϕ is the angle , and t is the time after the cut ( see Figure 1B and Figure 1—figure supplement 3 ) . I corresponds to the quantity of microtubules that depolymerized during the time interval δt . Next , we integrated the differential intensities over ϕ and plotted these integrals with respect to the radial coordinate x . The depolymerization wave appears as a peak that is traveling towards the center of the monopole and broadening over time ( see Figure 1C ) . We fitted Gaussians to these peaks and plotted the area under these Gaussians over time A~ ( t , r ) ( see Laser ablation method and Figure 1—figure supplement 3C ) . We fitted an exponential to the area decays over distance from the cut , and normalized the decays by the amplitude at the location of the cut . The slope at the position of the cut is proportional to the number of minus ends at this location ( see Laser ablation method ) . To take the local microtubule density into account , we multiplied the normalized slopes at the position of the cut by the averaged angular integral of the microtubule fluorescence intensity at this position . This gives the amount of minus ends per unit length nc ( y , r ) at y given a cut performed at r . To obtain the two dimensional minus end density ( number per unit length squared ) , we divided by 2πr , which corresponds to the nucleation profile ( notice that the nucleation profile has arbitrary units ) . Averaged microtubule density profiles were obtained from 82 and 12 fluorescence profiles of monopoles and MCAK-inhibited monopoles , respectively . Additionally , we used angular fluorescence profiles of control and MCAK-inhibited structures from the same extract reaction to determine the ratio between these two nucleation profiles and enable a reliable comparison . Finally , in order to obtain the microtubule length distribution , we fitted an exponential function to nc ( y , r ) as a function of the cut distance r . The slope of nc ( y , r ) at r is proportional to the number of microtubules with minus ends at y and plus ends at r ( see Appendix 1 ) , which after normalization gives the microtubule length distribution . Laser cuts in bipolar spindles were similarly analyzed . Instead of circular cuts , we performed linear cuts perpendicular to the long axis of the spindle , which induced two depolymerization waves traveling towards the poles ( due to the mixed polarity of microtubules ) . The waves were analyzed by integrating the differential intensities along the direction of the cut and plotting these integrals as a function of spindle length . This again lead to the depolymerization waves appearing as peaks that are traveling towards the poles and broadening over time . The subsequent analysis is exactly the same as for monopolar spindles continuing by fitting Gaussians to these peaks as described above . The microtubule polymerization velocity was measured by adding EB1-GFP to extract reactions to a final concentration of ∼0 . 2 µg/ml and analyzing kymographs drawn along the growth direction of a microtubule ( 40 kymographs from seven control monopoles obtained from different reactions on two different days , 68 kymographs from five MCAK inhibited monopoles obtained from different reactions on the same day ) . Microtubule depolymerization velocities were obtained by analyzing the velocity of the fronts after the laser cuts . The maxima of the fitted Gaussians ( see Laser cutting procedure and image analysis ) were used to determine the position of the depolymerization front as a function of time , which was fitted to a linear function . The slope corresponded to the depolymerization velocity of the cut microtubules , which was found to be constant for each laser cut . We measured microtubule lifetimes by adding Atto565 purified frog tubulin ( purified according to Ref . [Groen and Mitchison , 2016] ) to a final concentration of ∼1 nM and subsequent tracking of the speckles using the MOSAIC suite , ( Sbalzarini and Koumoutsakos , 2005 ) ( 5331 speckles from five monopoles from different reactions of 3 different days , 7289 speckles from 3 MCAK-inhibited monopoles from different reactions of the same day ) . We included only those speckles that appeared and disappeared during the length of the movie ( ∼10 min ) . To calculate the average lifetime of microtubules , we used the lifetime distribution 𝒫 ( t ) of a diffusion and drift process to fit it to our data according to 𝒫 ( t ) ∼t-3/2e-t/τ , where τ/4 is the expected lifetime of a microtubule of average length , Ref . ( Bicout , 1997; Needleman et al . , 2010 ) . A summary of the different measured values is given in Table 1 . Coverslips were cleaned by sonication in 2% Hellmanex and used to assemble parafilm channels of ∼3 mm width . Every step of the assay was completed by an incubation at room temperature ( 10 min up to several hours ) and washing of the channel with BRB80 ( 80 mM PIPES , 1 mM MgCl2 , 1 mM EGTA ) . Channels were subsequently filled with anti-biotin antibodies , Puronic F-127 to block the remaining surface , biotinylated Xenopus laevis sperm , biotinylated fluorocarbon oil microdroplets ( produced as described in Ref . [Lucio et al . , 2015] ) or biotinylated polystyrene beads acting as inert obstacles , and freshly prepared extract including Atto565 labeled purified porcine tubulin ( 150 nM final ) , EB1-GFP ( ∼0 . 2 µg/ml final ) , and sodium orthovanadate ( 0 . 5 mM final concentration ) . Image acquisition was performed on a TIRF microscope . To measure the microtubule mass over time , we added frog sperm to extract and immediately started to acquire z-stacks around the DNA over time . After subtracting the background , we integrated the fluorescence intensity of the labeled microtubules over all z-planes and plotted it as a function of time . Passivation of coverslips with Poly-L-lysine-g-polyethylene glycol ( PLL-g-PEG ) was performed according to Ref . ( Field et al . , 2017 ) . In brief , coverslips were placed in a drop of 0 . 1 mg/ml PLL-g-PEG in 10 M HEPES pH 7 . 4 on Parafilm for 20 min at room temperature . They were then washed three times in distilled water and dried with a nitrogen jet . | When cells divide , they first need to create a copy of their genetic material , which they then evenly distribute between their daughter cells . This is done by a complex of proteins known as the mitotic spindle , which divides the chromosomes that carry the genetic material in the form of genes . The mitotic spindle is mainly made of tubulin proteins that are arranged to form hollow cable-like filaments , called the microtubules . Microtubules are dynamic structures that can grow or shrink by adding or removing tubulin proteins . Unlike the spindle , which can ‘live’ up to hours , the microtubules only live for about 20 seconds and need to be constantly renewed to maintain the structure . To successfully distribute the genetic material , spindles need to have the right length . Previous research has shown that the length of a spindle adapts to the size of a cell – the larger the cells , the larger the spindles . However , in very large cells , such as the cells of an embryo when they first divide , spindles have an upper size limit . It is thought that specific proteins produced by the chromosomes help to regulate the formation of new microtubules and thereby also influence the size of the spindle . However , until now it was not clear how exactly they do so and if this also sets the upper size limit . To further investigate microtubule renewal and its relation to spindle size , Decker et al . used spindles assembled in cell extracts from the eggs of the African clawed frog . The results showed that the new microtubules grow off the existing ones , like new branches of a tree . The branching happens when the established microtubules interact with specific molecules emitted by the chromosomes , and the concentration of these molecules decreases with distance from the chromosomes . This concentration gradient regulates how many microtubules grow at different distances from the chromosomes and so sets the size of spindles . These findings help us to understand how biological structures are built out of dynamic and short-lived components . Moreover , a better understanding of how mitotic spindles grow might eventually help to develop new treatments for cancer and other diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods",
"and",
"materials"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2018 | Autocatalytic microtubule nucleation determines the size and mass of Xenopus laevis egg extract spindles |
It is generally assumed that human intelligence relies on efficient processing by neurons in our brain . Although grey matter thickness and activity of temporal and frontal cortical areas correlate with IQ scores , no direct evidence exists that links structural and physiological properties of neurons to human intelligence . Here , we find that high IQ scores and large temporal cortical thickness associate with larger , more complex dendrites of human pyramidal neurons . We show in silico that larger dendritic trees enable pyramidal neurons to track activity of synaptic inputs with higher temporal precision , due to fast action potential kinetics . Indeed , we find that human pyramidal neurons of individuals with higher IQ scores sustain fast action potential kinetics during repeated firing . These findings provide the first evidence that human intelligence is associated with neuronal complexity , action potential kinetics and efficient information transfer from inputs to output within cortical neurons .
A fundamental question in neuroscience is what properties of neurons lie at the heart of human intelligence and underlie individual differences in mental ability . Thus far , experimental research on the neurobiological basis of intelligence has largely ignored the neuronal level and has not directly tested what role human neurons play in cognitive ability , mainly due to the inaccessibility of human neurons . Instead , research has either been focused on finding genetic loci that can explain part of the variance in intelligence ( Spearman’s g ) in large cohorts ( Lam et al . , 2017; Sniekers , 2017; Trampush et al . , 2017; Coleman et al . , 2018 ) or on identifying brain regions in whole brain imaging studies of which structure or function correlate with IQ scores ( Karama et al . , 2009; Hulshoff Pol et al . , 2006; Narr et al . , 2007; McDaniel , 2005; Deary et al . , 2010 ) . Some studies have highlighted that variability in brain volume and intelligence may share a common genetic origin ( Hulshoff Pol et al . , 2006; Posthuma et al . , 2002; Sniekers , 2017 ) , and individual genes that were identified as associated with IQ scores might aid intelligence by facilitating neuron growth ( Sniekers , 2017; Coleman et al . , 2018 ) and directly influencing neuronal firing ( Lam et al . , 2017 ) . Intelligence is a distributed function that depends on activity of multiple brain regions ( Deary et al . , 2010 ) . Structural and functional magnetic resonance imaging studies in hundreds of healthy subjects revealed that cortical volume and function of specific areas correlate with g ( Karama et al . , 2009; Choi et al . , 2008; Narr et al . , 2007 ) . In particular , areas located in the frontal and temporal cortices show multiple correlations of grey matter thickness and functional activation with IQ scores: individuals with high IQ show larger grey matter volume of , for instance , Brodmann areas 21 and 38 ( Choi et al . , 2008; Deary et al . , 2010; Karama et al . , 2009; Narr et al . , 2007 ) . Cortical grey matter consists for a substantial part of dendrites ( Chklovskii et al . , 2002; Ikari and Hayashi , 1981 ) , which receive and integrate synaptic information and strongly affect functional properties of neurons ( Bekkers and Häusser , 2007; Eyal et al . , 2014; Vetter et al . , 2001 ) . Especially higher order association areas in temporal and frontal lobes in humans harbor pyramidal neurons of extraordinary dendritic size and complexity ( Elston , 2003; Mohan et al . , 2015 ) that may constitute variation in cortical thickness , neuronal function , and ultimately IQ . These neurons and their connections form the principal building blocks for coding , processing , and information storage in the brain and give rise to cognition ( Salinas and Sejnowski , 2001 ) . Given their vast number in the human neocortex , even the slightest change in efficiency of information transfer by neurons may translate into large differences in mental ability . However , whether and how the activity and dendritic structure of single human neurons support human intelligence has not been tested . To investigate whether structural and functional properties of neurons of the human temporal cortex associate with general intelligence , we collected a unique multimodal data set from 46 human subjects containing single cell physiology ( 31 subjects , 129 neurons ) , neuronal morphology ( 25 subjects , 72 neurons ) , pre-surgical MRI scans and IQ test scores ( 35 subjects , Figure 1 , data available at the Dryad Digital Repository: https://doi . org/10 . 5061/dryad . 83dv5j7 ) . Human cortical brain tissue was removed as a part of surgical treatment for epilepsy or tumor ( Table 1 ) . The tissue almost exclusively originated from middle temporal gyrus , approximately 4 cm posterior to the temporal pole ( Figure 2b ) as a block of ~1–1 . 5 cm in diameter and was removed to gain access to the disease focus in deeper lying structures such as hippocampus or amygdala . In all patients , the resected neocortical tissue was not part of the epileptic focus or tumor and displayed no structural/functional abnormalities in preoperative MRI investigation , electrophysiological whole-cell recordings or microscopic investigation of histochemically stained tissue ( Mohan et al . , 2015; Testa-Silva et al . , 2014; Testa-Silva et al . , 2010; Verhoog et al . , 2016; Verhoog et al . , 2013 ) . In line with the non-pathological status of tissue , we observed no correlations of cellular parameters or IQ scores with the subject’s disease history and age ( Figure 1—figure supplements 1–2 ) . After resection the tissue was immediately placed in ice-cold artificial cerebro-spinal fluid ( aCSF ) and within 15 min transported to the lab , sliced and maintained to enable single cell physiological recordings and biocytin filling . We recorded action potentials ( APs ) from human pyramidal neurons in superficial layers of temporal cortex and digitally reconstructed their complete dendritic structures . We tested the hypothesis that variation in neuronal morphology can lead to functional differences in AP speed and information transfer and explain variation in IQ scores . In addition to our experimental results , we used computational modelling to understand underlying principles of efficient information transfer in human cortical neurons .
Cortical thickness of the temporal lobe has been associated with IQ scores in hundreds of healthy subjects ( Choi et al . , 2008; Deary et al . , 2010; Hulshoff Pol et al . , 2006;Karama et al . , 2009; Narr et al . , 2007 ) , and we first asked whether this applies to the subjects in our study as well . From T1-weighted MRI scans obtained prior to surgery , we determined temporal cortical thickness in 35 subjects using voxel-based morphometry of temporal cortical areas . These areas included the surgically resected cortical tissue ( Figure 2b ) used for cellular recordings and neuronal reconstructions , which typically came from locations at 4 cm from temporal pole and was 1–1 . 5 cm in diameter ( black circle in Figure 2b ) . The total resected cortical area varied for each patient , but consisted of a larger part of the temporal lobe ( Figure 2b; average resected area in red , maximum in orange ) . The mean distance of resection boundaries from temporal pole was 4 . 2 ± 1 . 7 cm on superior temporal gyrus , 4 . 8 ± 1 . 5 cm on middle temporal gyrus , and 4 . 9 ± 1 . 5 cm on inferior temporal gyrus for the 46 subjects in this study . In MRI images , cortical thickness was measured in temporal lobe that included the resection areas and corresponded to the areas identified to associate with IQ in healthy subjects ( Choi et al . , 2008; Deary et al . , 2010; Hulshoff Pol et al . , 2006; Karama et al . , 2009; Narr et al . , 2007 ) ( Figure 2c; in red ) . The superior temporal gyrus was excluded from this analysis as it contains areas for auditory , gustatory and language processing that are spared during resection . Cortical thickness measurements were collapsed to one mean value for cortical thickness for each subject . In line with findings in healthy subjects ( Choi et al . , 2008; Deary et al . , 2010; Hulshoff Pol et al . , 2006; Narr et al . , 2007;Karama et al . , 2009 ) mean cortical thickness in temporal lobes positively correlated with IQ scores of the subjects ( Figure 2d ) . Cortical association areas in temporal lobes play a key role in high-level integrative neuronal processes and its superficial layers harbor neurons of increased neuronal complexity ( DeFelipe et al . , 2002; Elston , 2003; Scholtens et al . , 2014; van den Heuvel et al . , 2015 ) . In rodents , the neuropil of cortical association areas consists for over 30% of dendritic structures ( Ikari and Hayashi , 1981 ) . To test the hypothesis that human temporal cortical thickness is associated with dendrite size , we used 72 full reconstructions of biocytin-labelled temporal cortical pyramidal neurons from layers 2 , 3 and 4 ( median number of neurons per subject = 2; average 2 . 8; ranging from 1 to 10 ) part of which was previously reported ( Mohan et al . , 2015 ) . We calculated total dendritic length ( TDL ) that included all basal and apical dendrites without apparent slice artifacts for each neuron . We computed TDL from multiple neurons for each subject and correlated these mean TDL values to mean temporal cortical thickness from the same subject . We found that dendritic length positively correlated with mean temporal lobe cortical thickness ( Pearson correlation coefficient r = 0 . 5 , explained variance R2 = 0 . 25 ) , indicating that dendritic structure of individual neurons contributes to the overall cytoarchitecture of temporal cortex ( Figure 3a ) . TDL is in part determined by the soma location within cortical layers: cell bodies of pyramidal neurons with larger dendrites typically lie deeper , at larger distance from pia ( Mohan et al . , 2015 ) . To exclude a systematic bias in sampling , we determined the cortical depth of each neuron relative to the subject’s temporal cortical thickness in the same hemisphere . There was no correlation between IQ score and relative cortical depth of pyramidal neurons indicating that we sampled neurons at similar depths across subjects ( Figure 3b ) . Finally , we tested whether mean TDL and complexity of pyramidal neurons relates to subjects’ IQ scores . We found a strong positive correlation between individual’s pyramidal neuron TDL and IQ scores ( Pearson correlation coefficient r = 0 . 51 , explained variance R2 = 0 . 26; Figure 3c ) as well as between number of dendritic branch points and IQ scores ( r = 0 . 46 , R2 = 0 . 22; Figure 3d ) . Thus , larger and more complex pyramidal neurons in temporal association area may partly contribute to thicker cortex and link to higher intelligence . Dendrites not only receive most synapses in neurons , but dendritic morphology and ionic conductances act in concert to regulate neuronal excitability ( Bekkers and Häusser , 2007; Eyal et al . , 2014; Vetter et al . , 2001 ) . In model simulations where neurons are reduced to balls and sticks , increasing the dendritic membrane surface area , that is the dendritic impedance load , speeds up the onset phase of APs . This is a consequence of the decrease in the effective time constants of the neuron with increasing dendritic size and dendritic impedance load ( Eyal et al . , 2014 ) . Larger dendrites act as a larger sink for currents generated in the axon initial segment during AP onset and result in faster membrane potential changes . Furthermore , we found previously that human neocortical pyramidal neurons , which are three times larger than rodent pyramidal neurons ( Mohan et al . , 2015 ) , have faster AP onsets compared to rodent neurons and are able to track and encode fast synaptic inputs and sub-threshold changes in membrane potential with high temporal precision ( Testa-Silva et al . , 2014 ) . We asked whether the observed differences in TDL between human pyramidal neurons affected their encoding properties and ability to transfer information . To this end , we incorporated the 3-dimensional dendritic reconstructions of the 72 human pyramidal neurons into in silico models , equipped them with excitable properties ( see Materials and methods ) and tested whether their APs have faster onset . We found that TDL of model neurons with realistic dendritic trees positively correlated with the steepness of AP onsets ( r = 0 . 4 , R2 = 0 . 16; Figure 4a , b ) and larger dendrites enabled neurons to generate faster APs . The exact timing of action potential firing allows cortical neurons to pass on temporal information provided by synaptic inputs ( Köndgen et al . , 2008; Ilin et al . , 2013; Testa-Silva et al . , 2014; Linaro et al . , 2018 ) . Single pyramidal neurons do not sustain high frequency firing and generally do not encode high frequency synaptic input content in rate coding . Instead , the precision in timing of AP initiation does allow these neurons to encode incoming high frequency information in their output . In contrast to rodent neurons , human neurons can encode sub-threshold membrane potential changes on a sub-millisecond timescale by timing of APs ( Testa-Silva et al . , 2014 ) . This synaptic input tracking capacity strongly relies on the rapidity of AP onset ( Ilin et al . , 2013 ) . Faster APs allow neurons to respond to fast synaptic inputs , which will be missed if AP generation is too slow . Thereby , neurons with faster APs can translate higher frequencies of synaptic membrane potential fluctuations into AP timing and ultimately encode more information . The aforementioned theoretical work ( Eyal et al . , 2014 ) using ‘ball-and-stick’ neuron models showed that neurons with larger dendritic compartments not only have faster AP onset rapidity , but could also time AP generation to faster changes in membrane potentials , increasing the frequency tracking capabilities of input modulations , and augmenting the input frequency bandwidth of information encoding about three times . However , it is not known whether the same effect holds true for the human cortical pyramidal neurons we recorded from , and whether the range of dendritic compartment sizes we examined might lead to significant quantitative biophysical differences . We tested this by simulating sinusoidal current inputs of increasing frequencies into in silico representations of the neurons we recorded and reconstructed , and studied how the timing of AP firing of these neurons followed sub-threshold membrane potential changes . We find that human neurons with larger TDL can reliably time their APs to faster membrane potential changes , with cut-off frequencies up to 400–500 Hz , while smaller neurons had their cut-off frequencies already at 200 Hz ( Figure 4c , d ) . Furthermore , there was a significant positive correlation between the dendritic length and the cut-off frequency ( Figure 4d ) . Finally , given the same input - composed of the sum of three sinusoids of increasing frequencies - larger neurons were able to better encode rapidly changing temporal information into timing of AP firing , compared to smaller neurons ( Figure 4e ) . Thus , we find that differences in dendritic length of human neurons lead to faster APs and thereby to wider frequency bandwidths of encoding synaptic inputs into timing of AP output . Since cortical pyramidal neurons with large dendrites have faster APs and can encode more information in AP output , and since large dendrites also associate with higher IQ scores , we next asked whether human cortical pyramidal neurons from individuals with higher IQ scores generate faster APs . To test this , we made whole-cell recordings from pyramidal cells in acute slices of temporal cortex ( 31 subjects , 129 neurons , median number of neurons per subject = 3; ranging from 1 to 11 Figure 5 ) and recorded APs at different firing frequencies in response to depolarizing current steps . We determined AP maximum rise speed , which is highly correlated with AP onset rapidity ( r = 0 . 79 p=4 . 29e-14 , n = 60 , data not shown ) , and can more reliably be determined from recordings with sampling frequencies between 10 and 50 kHz . Maximum rise speed of APs depended on the firing history of the cell , with the first AP in the train having the highest AP rise speed and slowing down with increasing instantaneous firing frequency , the time interval between subsequent APs ( Figure 5b–d ) . To test whether AP rise speed differed between IQ groups , we split all AP rise speed data into two groups based on IQ score – above and below 100 . Although the AP rise speed of the first AP was not different between high and low IQ groups ( Figure 5c ) , the AP slowed down stronger in individuals with lower IQ scores compared to APs of individuals with higher IQ scores ( Figure 5d ) . At higher instantaneous firing frequencies ( 20–40 Hz ) , the AP rise speed was higher in individuals with IQ scores above 100 ( Figure 5c right; AP rise speed high IQ = 338 . 4 ± 26 . 03 mV/ms; AP rise speed low IQ = 268 . 1 ± 12 . 20 mV/ms , t-test p=0 . 0113 ) . We next calculated the slowing of APs with increasing instantaneous frequency by normalizing rise speed of APs to the rise speed of the first AP in the train . Relative to first AP , rise speed at 20–40 Hz showed significant slowing in subjects with lower IQ scores and decreased to 74% of the initial AP rise speed . In contrast , in neurons from individuals with higher IQ scores , AP rise speed remained on average at 84% ( Figure 5d right , high IQ = 0 . 84 ± 0 . 014; low IQ = 0 . 74 ± 0 . 024 , t-test p value=0 . 037 ) . We further investigated whether these differences at the group level reflected correlations between individual IQ scores and AP rise speeds . We correlated mean AP rise speeds both of the first AP and AP at 20–40 Hz from all neurons of the same subject to the subject's IQ score . The AP rise speed of the first AP in the train positively correlated with IQ scores ( r = 0 . 41 , R2 = 0 . 17; Figure 5e ) , and this correlation was even stronger for AP rise speeds at instantaneous frequencies of 20–40 Hz ( r = 0 . 46 , R2 = 0 . 21; Figure 5f ) . Importantly , also relative AP values showed significant positive correlations with IQ , indicating that it is the relative slowing of APs at high frequencies that associates with intelligence ( r = 0 . 37 , R2 = 0 . 14; Figure 5g ) . Finally , we asked whether the slowing of APs relates to the dendritic size of the same neurons , as our model results suggest . We find that larger neurons show less slowing of AP rise speed ( higher relative AP speeds ) at 20–40 Hz ( r = 0 . 55 , R2 = 0 . 30; Figure 5h ) . These findings reveal that higher IQ scores are accompanied by faster APs during repeated AP firing , while lower IQ scores associate with increased AP fatigue during elevated neuronal activity . Thus , neurons from individuals with higher IQ scores are better equipped to process synaptic signals at high rates and at faster time scales , which is necessary to encode large amounts of information accurately and efficiently .
Our findings provide a first insight into the possible cellular nature of human intelligence and explain individual variation in IQ scores based on neuronal properties: faster AP rise speed during neuronal activity and more complex , extended dendrites associate with higher intelligence . AP kinetics have profound consequences for information processing . In vivo , neurons are constantly bombarded by high frequency synaptic inputs and the capacity of neurons to keep track and phase-lock to these inputs determines how much of this synaptic information can be passed on to other neurons ( Testa-Silva et al . , 2014 ) . The brain operates at a millisecond time-scale and even sub-millisecond details of spike trains contain behaviorally relevant information that can steer behavioral responses ( Nemenman et al . , 2008 ) . Indeed , one of the most robust and replicable findings in behavioral psychology is the association of intelligence scores with measures of cognitive information-processing speed ( Barrett et al . , 1986 ) . Specifically , reaction times ( RT ) in simple RT tasks provide a better prediction of IQ than other speed-of-processing tests , with a regression coefficient of 0 . 447 ( Vernon , 1983 ) . In addition , high positive correlations between RT and other speed-of-processing tests suggest the existence of a common mental speed factor ( Vernon , 1983 ) . Recently , these classic findings were confirmed in a large longitudinal population-based study counting more than 2000 participants ( Der and Deary , 2017 ) . Especially strong correlations between RT and general intelligence were reported for a slightly more complex 4-choice ( Der and Deary , 2017 ) . Our results provide a biological cellular explanation for such mental speed factors: in conditions of increased mental activity or more demanding cognitive task , neurons of individuals with higher IQ are able to sustain fast action potentials and can transfer more information content from synaptic input to AP output . Pyramidal cells are integrators and accumulators of synaptic information . Larger dendrites can physically contain more synaptic contacts and integrate more information . Indeed , human pyramidal neuron dendrites receive twice as many synapses as in rodents ( DeFelipe et al . , 2002 ) and cortico-cortical whole-brain connectivity positively correlates with the size of dendrites in these cells ( Scholtens et al . , 2014; van den Heuvel et al . , 2015 ) . In this and a previous study ( Mohan et al . , 2015 ) , we find almost 2-fold larger dendritic arbor size ( mean TDL = 14 . 67 ± 4 mm ) and number of dendritic branches ( 64 . 03 ± 17 . 7 ) compared to reports that use post-mortem tissue ( Jacobs et al . , 2001; Bianchi et al . , 2013; Elston et al . , 2001 ) . The differences could be explained by a number of advantages of biocytin filled neurons in surgical resections compared to traditionally used Golgi stainings in human post-mortem tissue . The cortical slices in our study are thicker ( 350 μm compared to 120–250 μm ) and contain neurons with almost completely intact apical and basal dendrites , while other studies use only basal dendrites for quantification ( Jacobs et al . , 2001 ) . Furthermore , only a small number of neurons are filled in a slice , which allows to unambiguously quantify all dendrites from individual cells . Importantly , the tissue comes from a living donor compared to post-mortem tissue collection , and thus does not suffer from post-mortem delays ( de Ruiter , 1983 ) and only still living functional cells are filled . At the same time , post-mortem studies make it possible to make comparative analysis of several cortical areas . A gradient in complexity of pyramidal cells in cortical superficial layers accompanies the increasing integration capacity of cortical areas , indicating that larger dendrites are required for higher-order cortical processing ( Elston et al . , 2001; Jacobs et al . , 2001; van den Heuvel et al . , 2015 ) . Our results align well with these findings , suggesting that the neuronal complexity gradient also exists from individual to individual and could explain differences in mental ability . Within human cortex , association areas contain neurons with larger and more complex dendrites than primary sensory areas , while neuronal cell body density is lower in cortical association areas compared to primary sensory areas ( Elston , 2003; DeFelipe et al . , 2002 ) . Larger neurons are not as tightly packed together within cortical space as smaller cells . A recent study by Genç et al . , 2018 used multi-shell diffusion tensor imaging to estimate parieto-frontal cortical dendritic density and found that higher IQ scores correlated with lower values of dendritic density ( Genç et al . , 2018 ) . This may indicate that parieto-frontal cortical areas in individuals with higher IQ scores have less densely packed neurons , and may suggest that these neurons are larger . In our study , we carefully determined the amount and complexity of dendrite for each neuron , a computational unit within the cortex with well-defined input-output signals . Taking the results of Genç et al . , 2018 and our study together may suggest that the neuronal circuitry associated with higher intelligence is organized in a sparse and efficient manner , where larger and more complex pyramidal cells occupy larger cortical volume . Larger dendrites have an impact on excitability of cells ( Bekkers and Häusser , 2007; Vetter et al . , 2001 ) and determine the shape and rapidity of APs ( Eyal et al . , 2014 ) . Increasing the size of dendritic compartments in silico lead to acceleration of AP onset and increased encoding capability of neurons ( Eyal et al . , 2014 ) . Both in models and in slice recordings , changes of AP initiation dynamics were shown to fundamentally modify encoding of fast changing signals and the speed of communication between ensembles of cortical neurons ( Eyal et al . , 2014; Ilin et al . , 2013 ) . Neurons with fast AP onsets can encode high frequencies and respond quickly to subtle input changes . This ability can be impaired and response speed is decreased when AP onsets are slowed down by experimental manipulations ( Ilin et al . , 2013 ) . Our results not only demonstrate that AP speed depends on dendritic length and influences information transfer , but also show that both dendritic length and AP speed in human neurons correlate with intelligence . Thus , individuals with larger dendrites are better equipped to transfer synaptic information at higher frequencies . Remarkably , dendritic morphology and different parameters of AP waveform are also parameters that we have previously identified as showing pronounced differences between humans and other species ( Mohan et al . , 2015; Testa-Silva et al . , 2014 ) . Human pyramidal cells in layers 2 and 3 have 3-fold larger and more complex dendrites than in macaque or mouse ( Mohan et al . , 2015 ) . Moreover , human APs have lower firing threshold and faster AP onset kinetics both in single APs and during repeated firing ( Testa-Silva et al . , 2014 ) . These differences across species may suggest evolutionary pressure on both dendritic structure and AP waveform and emphasize specific adaptations of human pyramidal cells in association areas for cognitive functions . Our results were obtained from patients undergoing neurosurgical procedure and , thus , may potentially raise questions on how representative our findings are for normal healthy human subjects . Although no healthy controls can be used for single cell measurements , we addressed this issue in the following way . Firstly , in all patients , the resected neocortical tissue was not part of epileptic focus or tumor and displayed no structural or functional abnormalities in preoperative MRI , electrophysiological recordings or microscopic investigation of stained tissue . Secondly , none of the parameters correlated with age at epilepsy onset , seizure frequency , age or disease duration ( Figure 1—figure supplement 1 ) . Thirdly , IQ , dendritic length or AP rise speed were not different across different patient groups ( Figure 1—figure supplement 2 ) . Finally , the cortical thickness correlation with general intelligence we observe in our study was also reported in hundreds of healthy subjects . Taken together , these results indicate that our findings are not likely to be influenced by disease background of the subjects . In this study , intelligence was measured using WAIS IQ score , that combines results of 11 individual subtests of cognitive functioning into a single full-scale IQ score ( Wechsler , 2008; Taylor and Heaton , 2001 ) . This inevitably simplifies and reduces a multi-dimensional human trait to a single number . Although none of the intelligence tests can capture all aspects of human intelligence , IQ tests have proven their validity and relevance . The results of different cognitive subtests are highly correlated and generate a strong general factor – general intelligence or Spearman’s g ( Spearman , 1904 ) . Spearman’s g , calculated based on subtests of WAIS and expressed in total full-scale IQ score , strongly correlates with highly relevant life outcomes , including education , occupation , and income ( Strenze , 2007; Foverskov et al . , 2017 ) . Moreover , intelligence is a stable trait over time in the same individual: the results of intelligence tests at the age of 11 predict the scores at the age of 90 ( Gow et al . , 2011; Deary et al . , 2013 ) . Thus , despite its shortcomings , full scale IQ score provides a relevant and meaningful estimation of general intelligence that lies at the core of cognitive differences between individuals . In conclusion , our results provide first evidence that already at the level of individual neurons , such parameters as dendritic size and ability to maintain fast responses link to general mental ability . Multiplied by an astronomical number of cortical neurons in our brain , very small changes in these parameters may lead to large differences in encoding capabilities and information transfer in cortical networks and result in a speed advantage in mental processing and , finally , in faster reaction times and higher cognitive ability .
All procedures were performed with the approval of the Medical Ethical Committee of the VU University Medical Centre , and in accordance with Dutch license procedures and the Declaration of Helsinki . Written informed consent was provided by all subjects for data and tissue use for scientific research . All data were anonymized . Human cortical brain tissue was removed as a part of surgical treatment of the subject in order to get access to a disease focus in deeper brain structures ( hippocampus or amygdala ) and typically originated from gyrus temporalis medium ( Brodmann area 21 ) . Speech areas were avoided during resection surgery through functional mapping . We obtained neocortical tissue from 46 patients ( 24 females , 22 males; age range 18–66 years , Table 1 ) treated for mesial temporal sclerosis , removal of a hippocampal tumor , low grade hippocampal lesion , cavernoma or another unspecified temporal lobe pathology . From 35 of these patients , we also obtained pre-surgical MRI scans , from 31 patients we recorded Action Potentials from 129 neurons and from 25 patients we had fully reconstructed dendritic morphologies from 72 neurons . In all patients , the resected neocortical tissue was not part of epileptic focus or tumor and displayed no structural/functional abnormalities in preoperative MRI investigation , electrophysiological whole-cell recordings or microscopic investigation of stained tissue . The physiological recordings , subsequent morphological reconstructions , morphological and action potential analysis were performed blind to the IQ of the patients . Total IQ scores were obtained from all 46 subjects using the Dutch version of Wechsler Adult Intelligence Scale-III ( WAIS-III ) ( Taylor and Heaton , 2001 ) and in some cases WAIS-IV ( Wechsler , 2008 ) and consisted of following subtests: information , similarities , vocabulary , comprehension , block design , matrix reasoning , visual puzzles , picture comprehension , figure weights , digit span , arithmetic , symbol search and coding . The tests were performed as a part of neuropsychological examination shortly before surgery , typically within one week . T1-weighted brain images ( 1 mm thickness ) were acquired with a 3T MR system ( Signa HDXt , General Electric , Milwaukee , Wisconsin ) as a part of pre-surgical assessment ( number of slices = 170–180 ) . Cortical reconstruction and volumetric segmentation was performed with the Freesurfer image analysis suite ( http://freesurfer . net ) ( Fischl and Dale , 2000 ) . The processing included motion correction and transformation to the Talairach frame . Calculation of the cortical thickness was done as the closest distance from the grey/white boundary to the grey/CSF boundary at each vertex and was based both on intensity and continuity information from the entire three-dimensional MR volume ( Fischl and Dale , 2000 ) . Neuroanatomical labels were automatically assigned to brain areas based on Destrieux cortical atlas parcellation as described in ( Fischl et al . , 2004 ) . For averaging , the regions in temporal lobes were selected based on Destrieux cortical atlas parcellation in each subject . Upon surgical resection , the cortical tissue block was immediately transferred to ice-cold artificial cerebral spinal fluid ( aCSF ) containing in ( mM ) : 110 choline chloride , 26 NaHCO3 , 10 D-glucose , 11 . 6 sodium ascorbate , 7 MgCl2 , 3 . 1 sodium pyruvate , 2 . 5 KCl , 1 . 25 NaH2PO4 , and 0 . 5 CaCl2 ( 300 mOsm ) and transported to the neurophysiology laboratory ( within 500 m from the operating room ) . The transition time between resection of the tissue and the start of preparing slices was less than 15 min . After removing the pia and identifying the pia-white matter axis , neocortical slices ( 350 μm thickness ) were prepared in ice-cold slicing solution ( same composition as described above ) . Slices were then transferred to holding chambers in which they were stored for 30 min at 34 °C and for 30 min at room temperature before recording in aCSF , which contained ( in mM ) : 125 NaCl; 3 KCl; 1 . 2 NaH2PO4; 1 MgSO4; 2 CaCl2; 26 NaHCO3; 10 D-glucose ( 300 mOsm ) , bubbled with carbogen gas ( 95% O2/5% CO2 ) , as described previously ( Mohan et al . , 2015; Testa-Silva et al . , 2014; Testa-Silva et al . , 2010; Verhoog et al . , 2013; Verhoog et al . , 2016 ) . Cortical slices were visualized using infrared differential interference contrast ( IR-DIC ) microscopy . After the whole cell configuration was established , membrane potential responses to steps of current injection ( step size 30–50 pA ) were recorded . None of the neurons showed spontaneous epileptiform spiking activity . Recordings were made using Multiclamp 700A/B amplifiers ( Axon Instruments ) sampling at frequencies of 10 to 50 kHz , and lowpass filtered at 10 to 30 kHz . Recordings were digitized by pClamp software ( Axon ) and later analyzed off-line using custom-written Matlab scripts ( MathWorks ) . Patch pipettes ( 3–5 MOhms ) were pulled from standard-wall borosilicate capillaries and filled with intracellular solution containing ( in mM ) : 110 K-gluconate; 10 KCl; 10 HEPES; 10 K-phosphocreatine; 4 ATP-Mg; 0 . 4 GTP , pH adjusted to 7 . 3 with KOH; 285–290 mOsm , 0 . 5 mg/ml biocytin . All experiments were performed at 32–35 °C . Only cells with bridge balance of <20 MOhm were used for further analysis . During electrophysiological recordings , cells were loaded with biocytin through the recording pipette . After the recordings the slices were fixed in 4% paraformaldehyde and the recorded cells were revealed with the chromogen 3 , 3-diaminobenzidine ( DAB ) tetrahydrochloride using the avidin–biotin–peroxidase method ( Horikawa and Armstrong , 1988 ) . Slices ( 350 μm thick ) were mounted on slides and embedded in mowiol ( Clariant GmbH , Frankfurt am Main , Germany ) . Neurons without apparent slicing artifacts and uniform biocytin signal were digitally reconstructed using Neurolucida software ( Microbrightfield , Williston , VT , USA ) , using a × 100 oil objective . After reconstruction , morphologies were checked for accurate reconstruction in x/y/z planes , dendritic diameter , and continuity of dendrites . Finally , reconstructions were checked using an overlay in Adobe Illustrator between the Neurolucida reconstruction and Z-stack projection image from Surveyor Software ( Chromaphor , Oberhausen , Germany ) . Only neurons with virtually complete dendritic structures were included; cells with major truncations due to slicing procedure were excluded . Superficial layers pyramidal neurons were identified based on morphological and electrophysiological criteria at cortical depth within 400–1400 µm from cortical surface , that we previously found to correspond to cortical layers 2 , 3 and 4 in humans ( Mohan et al . , 2015 ) . For each neuron , we extracted total dendritic length ( TDL ) of all basal and apical dendrites and number of branch points and computed average TDL and average number of branch points for each subject by pooling data from all cells within one subject ( 1 to 10 neurons per subject ) . Only neurons without major truncations of apical dendrites by tissue sectioning were included for morphological analysis ( Mohan et al . , 2015; Deitcher et al . , 2017 ) . Following previous work ( Eyal et al . , 2014; Eyal et al . , 2016 ) conductance-based multicompartmental ‘Hodgkin and Huxley models’ ( Hodgkin and Huxley , 1952 ) of each of the reconstructed human pyramidal cells were built . To each model , a cylindrical axon ( 1 µm in diameter ) was connected to the soma , consisting of a 50 µm long Axon Initial Segment ( AIS ) and a 1 mm long myelinated part . The AIS consisted of 25 electrical compartments , the rest of the axon of 21 compartments . Simulations were run with the open-source software simulator NEURON v . 7 . 5 ( Carnevale and Hines , 2006 ) ( https://neuron . yale . edu/neuron ) , with dt = 10 µs integration time step at 37 °C . All compartments incorporated passive membrane properties , with specific capacitance Cm = 0 . 75 µF/cm2 , axial resistance Ra = 0 . 1 MOhm/cm , specific resistance Rm = 30 . 3 MOhm/cm2 , and leak-currents with reversal potential E = −70 mV . In the myelinated part of the axon Cm was decreased 37 . 5 times and Rm was increased 5 times . Across all dendritic compartments , Cm was increased by 84% and Rm was decreased by the same amount to account for dendritic spines ( Sarid et al . , 2007; Benavides-Piccione et al . , 2002 ) . Active membrane properties consisted of voltage-dependent fast-inactivating sodium ( Na+ ) and delayed-rectifier potassium ( K+ ) ionic conductances , taken from the SenseLab ModelDB database ( McDougal et al . , 2017 ) ( https://senselab . med . yale . edu/modeldb ) and set to: gNa = 0 pS/µm2 , gK = 0 pS/µm2 in the myelinated axon , gNa = 8000 pS/µm2 and gK = 1500 pS/µm2 in AIS , gNa = 800 pS/µm2 , gK = 320 pS/µm2 in the soma , and gNa = 20 pS/µm2 and gK = 10 pS/µm2 for dendrites . Reversal potentials for Na+ and K+ currents were +50 mV and −85 mV , respectively . Resulting input resistances were 61 . 5±4 . 73 MOhm and resting potentials were −70 . 5±0 . 02 mV . Onset rapidity of simulated action potentials ( APs ) was calculated as the slope of membrane potentials V ( t ) in the phase plane ( i . e . V ( t ) vs dV/dt ) at 10 mV/ms and averaged across APs in simulated trains . The dynamical input-output ‘transfer gain’ ( Linaro et al . , 2018; Köndgen et al . , 2008; Testa-Silva et al . , 2014 ) was determined by injecting sinusoidally oscillating input currents for 120 s at the soma , with amplitude I1 , frequency F ( 1–1’000 cycle/s ) , a DC baseline I0 amplitude , and randomly fluctuating component Inoise: ( 1 ) I ( t ) =I0+I1sin ( 2πFt ) +Inoise ( t ) Inoiset was an exponentially filtered stochastic Gaussian white-noise ( Arsiero et al . 2007 ) , with zero-mean , variance s2 and correlation length τI = 5 ms , by iterating at each simulation time step: ( 2 ) Inoiset+dt=1-dt/τIInoiset+s2dt/τIξtwhere {ξt} is a sequence of independent pseudo-random Gaussian numbers . s2 was set such that membrane potential hyperpolarization resulted in subthreshold potential fluctuations of ~3mV at -75 mV . DC baseline I0 was set to induce mean firing rates of ~10 spike/s . I1 was set to 1/6 of I0 . AP firing times {tk} were detected at the soma and collected across all values of F . The output ‘transfer gain’ r1F at a given frequency F was defined as the amplitude of complex numbers in polar form: ( 3 ) r1F=amplitude∑j=1Nexpj2πFtk/Nwhere N is the number of spikes and j is the imaginary unit . r1F was further normalized tor1F0 , with F0=3 cycle/s . The profile of r1F resembled a low-pass electrical filter , with cut-off frequency Fc defined as the highest frequency at whichr1Fc=r1F0/2 . Input waveforms in Figure 4 , inset , consisted of three rapidly varying components for 240 s: ( 4 ) It=I0+I1sin2πF1t+sin2πF2t+sin2πF3t/3+Inoisetwith F1 = 200 , F2 = 300 , F3 = 450 cycle/s . Action Potential ( AP ) waveforms were extracted from voltage traces recorded in response to intracellular current injections and sorted according to their instantaneous firing frequency . Instantaneous frequency was determined as 1/time to previous AP . Subsequently all APs were binned in 10 Hz bins , while the first APs in each trace were isolated in a separate bin . AP rise speed was defined as the peak of AP derivative ( dV/dt ) . For each analyzed cell , representative APs with all parameters were plotted for visual check to avoid errors in the analysis . For each neuron , the mean values of AP rise speed in a given frequency bin were obtained by averaging all APs within that frequency bin . Relative AP rise speeds were calculated by dividing the mean AP rise speed in each frequency bin ( 1–10 Hz , 11–20 Hz , 21–30 Hz and 31 to 40 Hz ) by the mean first AP rise speed ( first APs in the train of APs ) . To obtain AP values for each subject , AP parameters within each frequency bin were averaged for all neurons from one subject . All AP analysis was performed using customized Matlab scripts ( source code available at https://github . com/INF-Rene/Morphys ( Verhoog et al . , 2018; copy archived at https://github . com/elifesciences-publications/Morphys ) . Statistical significance of all correlations between parameters was determined using Pearson correlation and linear regression ( using Matlab , version R2017a , Mathworks ) . As multiple cells were measured per subject , correlations were calculated on mean parameter values per subject . All Pearson correlation coefficients and p values for correlations are shown in figure insets , R2 coefficients and sample sizes are shown in figure legends and main text . For statistical analysis of AP data , we divided all subjects according to their IQ into two groups: group with IQ > 100 and a group with IQ < 100 . Differences between 2 IQ groups in AP rise times were statistically tested using Student t-test . For analysis of different patient groups ( Figure 1—figure supplements 2 ) an ANOVA test was applied for each parameter separately . | Our brains are made up of almost 100 billion brain cells . Each of them acts like a small chip: they collect , process and pass on information in the form of electrical signals . In brain areas that integrate different types of information , such as frontal and temporal lobes , brain cells have larger dendrites – long projections specialized to collect signals . Theoretical studies predict that larger dendrites help cells to initiate electrical signals faster . Because of difficulty in accessing human neurons , it has been unknown whether any of these features also relate to human intelligence . Previous studies have revealed that people with a higher IQ have a thicker outer layer ( the cortex ) in areas such as the frontal and temporal lobes . But does a thicker cortex also contain cells with larger dendrites and is their role different ? To test whether smarter brains are equipped with faster and larger cells , Goriounova et al . studied 46 people who needed surgery for brain tumors or epilepsy . Each took an IQ test before the operation . To access the diseased tissue deep in the brain , the surgeon also removed small , undamaged samples of temporal lobe . These samples still contained living cells and their electrical signals were measured in the lab . The experiments showed that cells from people with a higher IQ had larger dendrites that transported information more quickly , especially when they are very active . Computer models were then used to understand how these findings can lead to more efficient information transfer in human neurons . Traditionally , research on human intelligence has focused on three main strategies: to study brain structure and function , to find genes associated with intelligence and to study the connection between our mind and behavior . Goriounova et al . are the first to take the single-cell perspective and link cell properties to human intelligence . The findings could help connect these separate approaches , and explain how genes for intelligence lead to thicker cortices and faster reaction times in people with higher IQ . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Large and fast human pyramidal neurons associate with intelligence |
The most common cause of early onset primary dystonia , a neuromuscular disease , is a glutamate deletion ( ΔE ) at position 302/303 of TorsinA , a AAA+ ATPase that resides in the endoplasmic reticulum . While the function of TorsinA remains elusive , the ΔE mutation is known to diminish binding of two TorsinA ATPase activators: lamina-associated protein 1 ( LAP1 ) and its paralog , luminal domain like LAP1 ( LULL1 ) . Using a nanobody as a crystallization chaperone , we obtained a 1 . 4 Å crystal structure of human TorsinA in complex with LULL1 . This nanobody likewise stabilized the weakened TorsinAΔE-LULL1 interaction , which enabled us to solve its structure at 1 . 4 Å also . A comparison of these structures shows , in atomic detail , the subtle differences in activator interactions that separate the healthy from the diseased state . This information may provide a structural platform for drug development , as a small molecule that rescues TorsinAΔE could serve as a cure for primary dystonia .
Torsins belong to the AAA+ ( ATPases associated with a variety of cellular activities ) ATPase family , a functionally diverse group of enzymes , which are fueled by ATP hydrolysis . AAA+ ATPases organize in structurally distinct fashions and interact with various accessory elements to remodel their protein or nucleic acid substrates ( Erzberger and Berger , 2006; Wendler et al . , 2012; White and Lauring , 2007 ) . Torsins are poorly understood AAA+ proteins with yet elusive functions and unknown substrates ( Laudermilch and Schlieker , 2016; Rose et al . , 2015 ) . Among the five human torsins ( TorsinA , TorsinB , Torsin2A , Torsin3A and Torsin4A ) , neuronally expressed TorsinA carries the most clinical significance since it is at the root of primary dystonia . Primary dystonia is a devastating neuromuscular disease that is predominantly caused by the deletion of glutamate 302 or 303 ( ΔE ) in TorsinA ( Goodchild et al . , 2005; Ozelius et al . , 1997 ) . The etiology of primary dystonia is poorly understood ( Breakefield et al . , 2008; Granata and Warner , 2010 ) , and there is currently no known cure for it . TorsinA is an unusual AAA+ ATPase , because , unlike any other family member ( Erzberger and Berger , 2006; Laudermilch and Schlieker , 2016; Rose et al . , 2015; White and Lauring , 2007 ) , it is localized to the endoplasmic reticulum ( ER ) and the contiguous perinuclear space ( PNS ) , and because it is not self-activated , but instead needs the AAA+-like proteins Lamina-associated protein 1 ( LAP1 ) or Luminal domain like LAP1 ( LULL1 ) to catalyze ATP hydrolysis ( Brown et al . , 2014; McCullough and Sundquist , 2014; Sosa et al . , 2014 ) . LAP1 is a type-II transmembrane protein , which resides at the inner nuclear membrane ( INM ) through its association with the nuclear lamina ( Goodchild and Dauer , 2005 ) . LULL1 is a LAP1 paralog , which localizes to the outer nuclear membrane ( ONM ) and the continuous ER , with its N-terminal portion protruding into the cytoplasm ( Goodchild and Dauer , 2005 ) . The structurally similar luminal domains of LAP1/LULL1 interact with TorsinA , and they provide an arginine finger to the TorsinA active site to facilitate torsin’s ATP hydrolysis ( Brown et al . , 2014; Sosa et al . , 2014 ) . Arginine fingers are key structural motifs of AAA+ ATPases because they neutralize the transition state during ATP hydrolysis ( Wendler et al . , 2012 ) . Since torsins lack arginine fingers themselves , this activation mechanism through LAP1/LULL1 is likely critical for their function . As reported by several labs , the disease mutant TorsinA ΔE is compromised in binding to LAP1/LULL1 ( Naismith et al . , 2009; Zhao et al . , 2013; Zhu et al . , 2010 ) . Clearly , this suggests that a probable cause of primary dystonia is the lack of activation of TorsinA . In line with this suggestion , LAP1 deletion shows a similar phenotype to Torsin ΔE , and contributes to disease pathology ( Kim et al . , 2010 ) . To investigate the molecular basis for primary dystonia as a result of the glutamate 302/303 deletion in TorsinA , we took a structural approach . We obtained high-resolution crystal structures of TorsinA as well as TorsinAΔE , each in complex with LULL1 , using a nanobody as crystallization chaperone . These structures likely open a pathway toward rational , structure-based drug design against primary dystonia .
TorsinA is a catalytically inactive AAA+ ATPase ( Brown et al . , 2014; Zhao et al . , 2013 ) , notoriously ill-behaved in vitro , primarily due to its limited solubility and stability . We partially overcame these problems by stabilizing an ATP-trapped E171Q mutant of human TorsinA ( residues 51–332 ) by co-expressing it with the luminal activation domain of human LULL1 ( residues 233–470 ) . This resulted in a better behaved heterodimeric complex ( Figure 1A ) , which , however , was still recalcitrant to our crystallization efforts . To facilitate crystallization , we isolated a nanobody ( VHH-BS2 ) from an alpaca immunized with the TorsinAEQ-LULL1 complex . A stable , heterotrimeric complex of TorsinAEQ-LULL1-VHH-BS2 was readily crystallized in the presence of ATP . We collected a 1 . 4 Å dataset and solved the structure by molecular replacement , using the LULL1-homolog LAP1 and a VHH template as search models ( Sosa et al . , 2014 ) ( Materials and methods , Table 1 ) . TorsinAEQ adopts a typical AAA+ ATPase fold ( Figure 1B , Figure 1—figure supplement 1 ) . The N-terminal nucleotide-binding or large domain ( residues 55–271 ) is composed of a central five-stranded , parallel β-sheet surrounded by 8 α-helices . A small three-helix bundle at its C-terminus ( residues 272–332 ) , forms critical contacts with LULL1 . An ATP molecule is bound in the manner characteristic of P-loop NTPases ( Wendler et al . , 2012 ) . The Walker A and B motifs are positioned to mediate the requisite nucleotide interactions , with sensor 1 and sensor 2 regions sensing the γ-phosphate and thus the nucleotide state ( Figure 1C ) . The luminal LULL1 activation domain ( residues 236–470 ) adopts an AAA+-like conformation , very similar to its paralog LAP1 ( rmsd 1 . 04 Å over 211 Cα positions , Figure 1—figure supplement 1 ) . The AAA+-like domain comprises a central β-sheet embedded within six α-helices ( Figure 1B ) . A C-terminal small domain is not present . Similar to LAP1 , an intramolecular disulfide bond forms at the C terminus of LULL1 , between conserved residues C310 and C468 ( Figure 1—figure supplements 1 , 3 ) . Characteristically , LULL1 lacks nucleotide binding due to the absence of Walker A and B motifs ( Sosa et al . , 2014 ) . LULL1 forms a composite nucleotide-binding site with TorsinA by providing arginine residue 449 ( ‘arginine finger’ ) at the base of helix α5 ( Figure 1C ) . The arginine finger activates ATP hydrolysis by TorsinA ( Brown et al . , 2014; Sosa et al . , 2014 ) . The small domain of TorsinA , including helix α7 featuring glutamates 302 and 303 , is intimately involved in LULL1 binding . Nanobody VHH-BS2 binds both TorsinA and LULL1 at a shallow groove ( Figure 1B , Figure 1—figure supplement 4 ) . Nanobodies contain three complementarity determining regions ( CDRs ) , with CDR3 most often making critical contacts with the antigen ( Muyldermans , 2013 ) . Indeed , the long CDR3 of VHH-BS2 ( residues 97–112 ) is the main binding element in the complex . 10 . 7554/eLife . 17983 . 003Figure 1 . Architecture of the TorsinA-LULL1 complex . ( A ) Schematic diagrams of TorsinA and LULL1 . Important residues and sequence motifs are indicated . The colored areas mark the crystallized segments . Large and small domains of TorsinA are colored in purple and pink , respectively . SS , signal sequence; H , hydrophobic region; TM , transmembrane helix . ( B ) Cartoon representation of the TorsinA-LULL1 complex in two orientations . Color-coding as in ( A ) . A nanobody ( VHH-BS2 , grey; complementarity determining regions , red ) was used as a crystallization chaperone . Numbers refer to secondary structure elements . ( C ) Close-up of the ATP binding site . Key residues are labeled . 2Fo−Fc electron density contoured at 2σ displayed as grey mesh . ( D ) Close-up of the proximal cysteines 280 and 319 next to the adenine base of the bound ATP . 2Fo−Fc electron density is contoured at 1σ . The cysteine pair adopts three alternate conformations , but remains reduced in all of them . DOI: http://dx . doi . org/10 . 7554/eLife . 17983 . 00310 . 7554/eLife . 17983 . 004Figure 1—figure supplement 1 . Structural comparisons . ( A ) Human TorsinA-ATP ( left ) displayed as a cartoon , compared to the D2 domain of the double-ringed AAA+ ATPase ClpB-AMPPCP ( right ) from Thermus thermophilus ( Zeymer et al . , 2014 ) ( PDB code 4LJ9 ) in the same orientation . Important structure motifs are labeled . ( B ) Human LULL1 ( orange ) superposed on human LAP1 ( grey , PDB code 4TVS ) , shown in two orientations . The one region of major structural difference is labeled ( left ) . The disulfide bridge within LAP1/LULL1 is in yellow ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17983 . 00410 . 7554/eLife . 17983 . 005Figure 1—figure supplement 2 . Phylogenetic analysis of Torsins . Maximally diverged torsins are aligned . Secondary structure elements of human TorsinA are displayed above the alignment . Important sequence motifs are boxed . LULL1 contacts , red circles , conserved cysteines , yellow circles . Proximal cysteines 280 and 319 connected with a dashed yellow line . Asterisk denotes putative torsin homologs based on sequence analysis . hs , Homo sapiens; oa , Ornithorhynchus anatinus; gg , Gallus gallus; tr , Takifugu rubripes; dr , Danio rerio; nv , Nematostella vectensis; bf , Branchiostoma floridae; stp , Strongylocentrotus purpuratus; ci , Ciona intestinalis; ce , Caenorhabditis elegans; dm , Drosophila melanogaster; ta , Trichoplax adherens . DOI: http://dx . doi . org/10 . 7554/eLife . 17983 . 00510 . 7554/eLife . 17983 . 006Figure 1—figure supplement 3 . Phylogenetic analysis of LAP1/LULL1 . Maximally diverged LAP1 and LULL1 sequences are aligned . If not experimentally confirmed , sequences were assigned as LAP1 or LULL1 based on the presence of an N-terminal , extraluminal domain with basic signature , characteristic of LAP1 . Secondary structure elements of human LULL1 are displayed above the alignment . The strictly conserved Arg-finger is boxed . TorsinA contacts , red circles , conserved cysteines , yellow circles . Disulfide bridge depicted as a yellow line . hs , Homo sapiens; oa , Ornithorhynchus anatinus; gg , Gallus gallus; tr , Takifugu rubripes; dr , Danio rerio; nv , Nematostella vectensis; bf , Branchiostoma floridae; stp , Strongylocentrotus purpuratus; ci , Ciona intestinalis; ce , Caenorhabditis elegans; dm , Drosophila melanogaster; ta , Trichoplax adherens . DOI: http://dx . doi . org/10 . 7554/eLife . 17983 . 00610 . 7554/eLife . 17983 . 007Figure 1—figure supplement 4 . Nanobody interaction . The heterotrimeric TorsinA ( ATP ) -LULL1-VHH-BS2 complex is shown in two orientations . Nanobody and interacting secondary structure elements of TorsinA and LULL1 are shown in full color , non-interacting elements in faded colors . Complementarity determining region ( CDR ) loops in red . Insets show close-ups with important interacting residues labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 17983 . 00710 . 7554/eLife . 17983 . 008Figure 1—figure supplement 5 . Comparison of sequence motifs of AAA+ ATPases . Torsins and LAP1/LULL1 sequences are compared to the HCLR clade , the most similar branch within the AAA+ ATPase family ( Erzberger and Berger , 2006; Iyer et al . , 2004 ) . Sequential elements characteristic for each of the 3 groups are displayed as WebLogos ( Crooks et al . , 2004 ) . Numbering refers to ClpB-D2 from Thermus thermophilus for the HCLR class , human TorsinA for Torsins , and human LULL1 for LAP1/LULL1 . Grey bars indicate the characteristic motif or residue , surrounded by a few adjacent residues to emphasize the distinct conservation . All three groups have elements that can be used to distinguish them among each other . Since Torsins and LAP1/LULL1 lack a pore loop consensus sequence φφG ( where φ denotes a bulky hydrophobic residue ) , putative pore loop areas have been determined structurally . Dashed grey bars indicate residues which can be structurally aligned to the pore loop motif of the closest HCLR AAA+ clade members . DOI: http://dx . doi . org/10 . 7554/eLife . 17983 . 00810 . 7554/eLife . 17983 . 009Table 1 . X-ray data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 17983 . 009TorsinA-LULL1233-470TorsinAΔE-LULL1233-470PDB Code5J1S5J1TData collectionSpace groupP212121P212121Cell dimensionsa , b , c ( Å ) 75 . 7 , 90 . 7 , 105 . 175 . 4 , 88 . 4 , 105 . 3α , β , γ ( ° ) 90 . 0 , 90 . 0 , 90 . 090 . 0 , 90 . 0 , 90 . 0Resolution ( Å ) 61–1 . 40 ( 1 . 45–1 . 40 ) *68–1 . 40 ( 1 . 45–1 . 40 ) Rsym0 . 06 ( 0 . 88 ) 0 . 10 ( 1 . 98 ) Rpim0 . 03 ( 0 . 43 ) 0 . 03 ( 0 . 60 ) I / σ33 . 0 ( 1 . 5 ) 30 . 8 ( 1 . 3 ) Completeness ( % ) 94 . 7 ( 67 . 5 ) 97 . 9 ( 96 . 5 ) Redundancy5 . 7 ( 4 . 4 ) 12 . 4 ( 11 . 3 ) CC ( 1/2 ) 1 . 00 ( 0 . 65 ) 1 . 00 ( 0 . 58 ) RefinementResolution ( Å ) 61 . 4–1 . 4067 . 7–1 . 40No . reflections132956134333Rwork / Rfree0 . 143/0 . 1880 . 148/0 . 177No . atoms58985927Protein52415244Ligand/ion3547Water622636B factors ( Å2 ) Protein31 . 324 . 0Ligand/ion23 . 217 . 2Water43 . 133 . 6r . m . s . deviationsBond lengths ( Å ) 0 . 0140 . 017Bond angles ( ° ) 1 . 251 . 71RamachandranFavored/allowed/outliers ( % ) 98 . 0/1 . 7/0 . 098 . 6/1 . 4/0 . 0*Values in parentheses are for highest-resolution shell . One crystal was used for each dataset . AAA+ ATPases are organized into a number of structurally defined clades ( Erzberger and Berger , 2006; Iyer et al . , 2004 ) , distinguished by shared structural elements . Comparison with other AAA+ ATPase structures shows that TorsinA fits best into a clade that also contains the bacterial proteins HslU , ClpA/B , ClpX , and Lon ( HCLR clade ) , all of which are involved in protein degradation or remodeling ( Erzberger and Berger , 2006 ) . These AAA+ family members share a β-hairpin insertion that precedes the sensor-I region ( Figure 1—figure supplement 1 ) . TorsinA also contains this structural element , but it adopts a distinctly different orientation compared to other members of the clade; however , the pre-sensor I region may be affected by crystal packing in our structure . Two other distinct regions are present . The protein degrading or remodeling AAA+ ATPases all form hexameric rings with a central pore ( Hanson and Whiteheart , 2005; Olivares et al . , 2016; White and Lauring , 2007 ) . ‘Pore loops’ in each subunit , conserved elements positioned between strand β2 and helix α2 , are critical for threading the protein substrates through the ring ( Sauer and Baker , 2011 ) . Torsins are devoid of a pore loop consensus motif ( Figure 1—figure supplements 2 , 5 ) . TorsinA has two cysteines ( Cys280 , and Cys 319 , which is part of the sensor-II motif ) , positioned near the adenine base of the ATP molecule ( Figure 1D ) . These cysteines do not form a disulfide bridge in our structure . However , the conservation of Cys280 and the Gly-Cys-Lys sensor-II motif at position 318–320 ( Figure 1—figure supplements 2 , 5 ) indicates an important functional role . A redox activity as part of the ATPase cycle therefore seems highly likely , as has been previously speculated ( Zhu et al . , 2008 , 2010 ) . The interaction of TorsinA with its ATPase activators LULL1 and LAP1 is of particular importance , as a prominent mutation causing primary dystonia--the deletion of glutamate 302 or 303--weakens these interactions ( Naismith et al . , 2009; Zhao et al . , 2013; Zhu et al . , 2010 ) . But why and how ? The TorsinA-LULL1 interface extends over an area of 1439 Å2 . The main structural elements involved in this interaction are the nucleotide-binding region as well as the small domain of TorsinA , and helices α0 , α2 , α4 and α5 of LULL1 ( Figure 1 , Figure 1—figure supplements 2 , 3 , Figure 2A ) . The exact position of the small domain of TorsinA relative to the large domain is likely dictated by the sensor II motif , preceding α8 , which directly contacts the γ-phosphate of ATP through Lys 320 , thus serving as an anchor point . A switch to ADP presumably weakens this connection , such that the small domain would become more loosely attached to the large domain . This could explain the observed ATP-dependency of LAP1/LULL1 binding ( Goodchild and Dauer , 2005; Naismith et al . , 2009; Zhao et al . , 2013; Zhu et al . , 2010 ) . Within the small domain , helix α7 , the following loop , and the terminal helix α8 contain all the critical residues . Glutamate 302 and 303 are positioned at the very end of helix α7 , and both are involved in TorsinA contacts . Specifically , Glu 303 forms a prominent charge interaction with Arg 276 of LULL1 . TorsinA Lys113 – LULL1 Glu385 , TorsinA Asp316 - LULL1 Arg419 , TorsinA Lys317 - LULL1 Glu415 , TorsinA Asp327 - LULL1 Lys283 are additional charge interactions . 10 . 7554/eLife . 17983 . 010Figure 2 . Analysis of the TorsinA-LULL1 interface . ( A ) Side-by-side comparison of TorsinA-ATP-LULL1 ( left ) and TorsinAΔE-ATP-LULL1 ( right ) . Zoomed insets show the atomic details of the interactions between TorsinA/TorsinAΔE and LULL1 , with a focus on the ΔE303 area . ( B and C ) Mutational analysis of the TorsinA-LULL1 interface . Substitution or deletion of residues involved in TorsinA-LULL1 binding were probed using a Ni-affinity co-purification assay with recombinant , bacterial-expressed protein . Only TorsinA is His-tagged . SDS-PAGE analysis is shown . Lack of binding is observed by the absence of complex ( uncomplexed His-tagged TorsinA is insoluble ) . t , total lysate , e , Ni eluate . Asterisk denotes an unrelated contaminant . DOI: http://dx . doi . org/10 . 7554/eLife . 17983 . 01010 . 7554/eLife . 17983 . 011Figure 2—figure supplement 1 . Structural mapping of mutations causing dystonia . All known point mutations and deletions that lead to dystonia are marked as green dots and shown in light green color , respectively , on the TorsinA-ATP-LULL1 structure . A modifier TorsinA mutation , D216H , is marked as a blue dot . The structural equivalent of the LAP1 missense mutation ( E482A ) is LULL1 E368A , marked as a green dot . See Table 2 for an explanation of the likely structural consequence . DOI: http://dx . doi . org/10 . 7554/eLife . 17983 . 01110 . 7554/eLife . 17983 . 012Table 2 . Dystonia mutations . DOI: http://dx . doi . org/10 . 7554/eLife . 17983 . 012ProteinMutationStructural consequenceReferenceTorsinA∆E302/303Weakened LAP1/LULL1 binding ( Ozelius et al . , 1997 ) TorsinA∆F323-Y328Weakened LAP1/LULL1 binding ( Leung et al . , 2001 ) TorsinAR288QWeakened LAP1/LULL1 binding ( Zirn et al . , 2008 ) TorsinAF205IFolding problem ( Calakos et al . , 2010 ) TorsinAD194VChange to the conserved , noncatalytic interface ( Cheng et al . , 2014 ) TorsinA∆A14-P15Improper cellular targeting ( Vulinovic et al . , 2014 ) TorsinAE121KCharge inversion at the membrane proximal interface ( Vulinovic et al . , 2014 ) TorsinAV129IFolding problem ( Dobričić et al . , 2015 ) TorsinAD216H ( modifier ) Surface change; consequence unclear ( Kamm et al . , 2008; Kock et al . , 2006 ) LAP1c . 186deiG ( p . E62fsTer25 ) Lack of the luminal activation domain of LAP1 ( Kayman-Kurekci et al . , 2014 ) LAP1E482A*Improper folding ( Dorboz et al . , 2014 ) *Assesment based on the equivalent residue in LULL1 ( E368 ) . To investigate the atomic details of the weakened binding of TorsinAΔE to LAP1/LULL1 , and thus the molecular basis of primary dystonia , we made use of the observation that VHH-BS2 also stabilizes the TorsinAEQΔE ( ATP ) -LULL1 interaction . We were able to crystallize TorsinAEQΔE ( ATP ) -LULL1-VHH-BS2 and determine its structure at a resolution of 1 . 4 Å . Not surprisingly , the overall structure is almost identical to the wild-type protein ( 0 . 34 Å rmsd over 276 Cα atoms for TorsinA , 0 . 27 Å rmsd over 226 Cα atoms for LULL1 ) , except for critical differences in the TorsinA-LULL1 interface ( Figure 2A ) . The principal difference is that helix α7 is shortened due to the missing Glu 303 , with a slight--but significant--restructuring of the loop that follows to establish the connection with helix α8 . For future reference , we suggest renaming the ΔE mutation ΔE303 , rather than ΔE302/303 , since the position of Glu 302 is effectively unchanged . In the dystonia mutant , the TorsinA Glu 303 – LULL1 Arg 276 charge interaction is lost , and the hydrogen-bonding network involving TorsinA Glu 302 , Phe 306 and Arg312 , as well as LULL1 Arg412 and Glu416 is disrupted ( Figure 2A ) . To determine the importance of different TorsinA residues for LULL1 binding , we performed a co-purification assay ( Figure 2B , C ) . His-tagged , ATP-trapped TorsinAEQ ( residues 51–332 ) and mutants thereof were recombinantly co-expressed with LULL1 ( residues 233–470 ) , but without VHH-BS2 , in bacteria . Binding was tested in a co-purification assay using Ni-affinity . The TorsinEQΔE303 mutation abolishes binding in this assay , as expected ( Figure 2B ) . Since unbound TorsinAEQ is largely insoluble , absence of binding is not registered as an appearance of TorsinAEQ alone , but rather as a lack of eluted protein complex altogether . Eliminating the salt bridge between TorsinA Glu303 and LULL1 Arg276 does not disrupt the TorsinA-LULL1 interaction ( Figure 2B ) . However , ΔMet304 and ΔThr305 both phenocopy ΔE303 in abolishing LULL1 binding ( Figure 2C ) . This is in full agreement with published in vivo data using similar mutants ( Goodchild and Dauer , 2004 ) . The intricate network of interactions of the α7-α8 loop of TorsinA is crucial for LULL1 binding . Since the ΔE mutation causes a local change within the small domain of TorsinA rather than protein misfolding , it may be possible to rescue binding by developing a small molecule that resurrects the weakened TorsinAΔE-LAP1/LULL1 interaction . Although TorsinAΔE303 is the most prevalent mutation that causes primary dystonia , it is not the only one ( Laudermilch and Schlieker , 2016; Rose et al . , 2015 ) . We examined the structural consequence of all known mutations ( Figure 2—figure supplement 1 , Table 2 ) . Based on our structural data , we strongly predict that most mutations likely cause protein misfolding or they weaken or abolish LAP1/LULL1 binding . Conversely , the two dystonia-mutations found in LAP1 presumably affect torsin interaction . Our structural data , therefore , clearly support the hypothesis that improper torsin activation is the likely cause of primary dystonia ( Kim et al . , 2010 ) .
The biological function of TorsinA remains enigmatic ( Granata et al . , 2011; Jokhi et al . , 2013; Liang et al . , 2014; Nery et al . , 2008 , 2011 ) . Because TorsinA belongs to the AAA+ ATPase superfamily , with specific homology to the bacterial proteins HslU , ClpX , ClpA/B and Lon , it is generally assumed that TorsinA is involved in protein remodeling or protein degradation ( Laudermilch and Schlieker , 2016; Rose et al . , 2015 ) . However , a substrate of TorsinA has yet to be identified . The TorsinA structure enables a more thorough comparison to other AAA+ ATPases , particularly with regard to the functionally relevant oligomerization state . After the discovery that LAP1/LULL1 are Arg-finger containing TorsinA activators with a AAA+-like structure , it seemed reasonable to suggest that TorsinA and LAP1/LULL1 likely form heterohexameric rings ( ( TorsinA-ATP-LAP1/LULL1 ) 3 ) in order to function ( Brown et al . , 2014; Sosa et al . , 2014 ) . However , the predominant oligomeric form of recombinant TorsinA-ATP-LAP1/LULL1 complex in vitro and in solution is the heterodimer ( Brown et al . , 2014; Sosa et al . , 2014 ) . In addition , torsin variants have been reported to occur in various oligomeric forms as detected by Blue Native PAGE ( BN-PAGE ) ( Goodchild et al . , 2015; Jungwirth et al . , 2010; Vander Heyden et al . , 2009 ) . Our structure now raises doubts about the physiological relevance of a heterohexameric ring ( Figure 3 ) . First , we note that the small domain of TorsinA is essential for LAP1/LULL1 binding ( Figure 2C ) . This is reminiscent of the related HCLR AAA+ clade members where the small domain is known to be critical for hexamerization ( Bochtler et al . , 2000; Mogk et al . , 2003 ) . The importance of the small domain for oligomerization in the context of torsins has also been discussed recently ( Rose et al . , 2015 ) . Neither LAP1 nor LULL1 harbor a small domain , arguing against formation of a stable heteromeric ring , or , alternatively , suggesting a ring of substantially different architecture . Second , ring formation is important for AAA+ ATPases that thread their protein substrate through a central pore for refolding or for degradation . This central pore is lined with conserved ‘pore loops’ that are essential for function ( White and Lauring , 2007 ) . Neither TorsinA and its homologs , nor LAP1/LULL1 have ‘pore loop’ equivalents ( Figure 1—figure supplement 5 ) . TorsinA is therefore unlikely to actually employ a peptide threading mechanism that involves a central pore . Third , the surface conservation of LAP1/LULL1 also argues against a heteromeric ring assembly . Although the catalytic , ATP-containing interface with TorsinA is well-conserved , the presumptive non-catalytic , nucleotide-free interface is not ( Figure 3B ) . Importantly and in contrast to LAP1/LULL1 , the same analysis for TorsinA shows that its ‘backside’ is conserved . TorsinA may therefore interact in homotypic fashion with TorsinA , with other torsin homologs , or even with an additional , yet unidentified protein . This could mean that the previously observed hexameric assemblies ( Goodchild et al . , 2015; Jungwirth et al . , 2010; Sosa et al . , 2014 ) may only contain one LAP1/LULL1 unit , and multiple torsin units , a property that the employed assays would not have differentiated . It is also possible , that the reported hexameric assemblies reflect a vestigial , yet physiologically irrelevant property , perhaps just of the evolutionary origin of the Torsin-LAP1/LULL1 system . Taking all the existing data into account , it is suggestive that TorsinA may be an exceptional AAA+ ATPase in that it simply acts as a heterodimer , together with LAP1 or LULL1 functioning as an activator . As long as the biological function and the substrate for TorsinA are unclear , however , the physiologically relevant oligomeric state of TorsinA ultimately remains a matter of speculation . Given the unique properties of TorsinA , keeping an open mind about TorsinA assembly into its functional state is called for , as it may well differ more than anticipated from well-studied AAA+ ATPase systems . 10 . 7554/eLife . 17983 . 013Figure 3 . Oligomerization of TorsinA-LULL1 . ( A ) Left , Schematic representation of a hypothetical heterohexameric ( TorsinA-LULL1 ) 3 ring model , in analogy to canonical AAA+ ATPases . White star represents ATP . Since LULL1 cannot bind a nucleotide , there would be three catalytic ( nucleotide-bound ) and three non-catalytic interfaces per ring . Open-book representation of the catalytic interface between TorsinA and LULL1 , as seen in this study . Black line marks the outline of the interface . Color gradient marks conservation across diverse eukaryotes . ( B ) The same analysis as in ( A ) , but for the hypothetical ‘non-catalytic’ interface . The interface model on the right is based on swapping the TorsinA and LULL1 positions in the TorsinA-LULL1 complex . DOI: http://dx . doi . org/10 . 7554/eLife . 17983 . 013 The observation that the nanobody VHH-BS2 stabilizes the TorsinAΔE303-LULL1 suggests that it could possibly be used directly as a therapeutic . After all , it could directly rescue TorsinA activity . There are , however , at least two major problems . First , VHH-BS2 only recognizes the TorsinA- ( or TorsinAΔE303- ) LULL1 complex , but not the homologous TorsinA-LAP1 complex . The function of LULL1 is still poorly understood , but a knockdown does not generate an NE blebbing phenotype ( Goodchild et al . , 2015; Turner et al . , 2015; Vander Heyden et al . , 2009 ) , which is symptomatic for a TorsinA knockout ( Goodchild et al . , 2005 ) or a LAP1 knockdown ( Kim et al . , 2010 ) . Therefore , resurrecting activation of TorsinAΔE303 via LULL1 is unlikely to ameliorate the dystonia phenotype . Furthermore , the nanobody interaction site on the TorsinA-LULL1 interface is very likely oriented toward the ER membrane , which can be inferred from the relative positions of the membrane anchor of LULL1 and the hydrophobic , likely membrane-proximal N-terminal region of TorsinA . These topological restraints suggest that the nanobody will not bind in vivo , but that it is of significant use for in vitro studies .
DNA sequences encoding human TorsinA ( residues 51–332 ) and the luminal domain of human LULL1 ( residues 233–470 ) were cloned into a modified ampicillin resistant pETDuet-1 vector ( EMD Millipore ) . TorsinA , N-terminally fused with a human rhinovirus 3C protease cleavable 10xHis-7xArg tag , was inserted into the first multiple cloning site ( MCS ) , whereas the untagged LULL1 was inserted into the second MCS . Mutations on TorsinA and LULL1 were introduced by site-directed mutagenesis . The untagged VHH-BS2 nanobody was cloned into a separate , modified kanamycin resistant pET-30b ( + ) vector ( EMD Biosciences ) . To co-express TorsinA ( EQ or EQ/ΔE ) , LULL1 and VHH-BS2 for crystallization , the E . coli strain LOBSTR ( DE3 ) RIL ( Kerafast , Boston MA ) ( Andersen et al . , 2013 ) was co-transformed with the two constructs described above . Cells were grown at 37°C in lysogeny broth ( LB ) medium supplemented with 100 µg ml−1 ampicillin , 25µg ml−1 kanamaycin and 34 µg ml−1 chloramphenicol until an optical density ( OD600 ) of 0 . 6–0 . 8 was reached , shifted to 18°C for 20 min , and induced overnight at 18°C with 0 . 2 mM isopropyl β-D-1-thiogalactopyranoside ( IPTG ) . The bacterial cultures were harvested by centrifugation , suspended in lysis buffer ( 50 mM HEPES/NaOH pH 8 . 0 , 400 mM NaCl , 40 mM imidazole , 10 mM MgCl2 , and 1 mM ATP ) and lysed with a cell disruptor ( Constant Systems ) . The lysate was immediately mixed with 0 . 1 M phenylmethanesulfonyl fluoride ( PMSF ) ( 50 μl per 10 ml lysate ) and 250 units of TurboNuclease ( Eton Bioscience ) , and cleared by centrifugation . The soluble fraction was gently mixed with Ni-Sepharose 6 Fast Flow ( GE Healthcare ) resin for 30 min at 4°C . After washing with the lysis buffer , bound protein was eluted in elution buffer ( 10 mM HEPES/NaOH pH 8 . 0 , 150 mM NaCl , 300 mM imidazole , 10 mM MgCl2 , and 1 mM ATP ) . The eluted protein complex was immediately purified by size exclusion chromatography on a Superdex S200 column ( GE Healthcare ) equilibrated in running buffer ( 10 mM HEPES/NaOH pH 8 . 0 , 150 mM NaCl , 10 mM MgCl2 , and 0 . 5 mM ATP ) . Following the tag removal by 10xHis-7xArg-3C protease , the fusion tags and the protease were separated from the complex by cation-exchange chromatography on a HiTrapS column ( GE Healthcare ) using a linear NaCl gradient . The flow-through from the cation-exchange chromatography , containing the protein complex , was purified again by size exclusion chromatography on a Superdex S200 column as at the previous step . For the non-structural analysis of TorsinA and LULL1 variants , the pETDuet-1-based expression plasmid was transformed into LOBSTR ( DE3 ) RIL cells without co-expressing nanobody VHH-BS2 . Ni2+-affinity purification was performed as described above and bound protein was eluted . Aliquots from the Ni2+-eluate and the total lysate were collected and analyzed by SDS-PAGE gel electrophoresis . Purified TorsinAEQ-LULL1-VHH-BS2 and TorsinAEQΔE-LULL1-VHH-BS2 complexes were concentrated up to 4–4 . 5 mg/ml and supplemented with 2 mM ATP prior to crystallization . The TorsinAEQ containing complex crystallized in 13% ( w/v ) polyethylene glycol ( PEG ) 6000 , 5% ( v/v ) 2-Methyl-2 , 4-pentanediol , and 0 . 1 M MES pH 6 . 5 . The TorsinAEQΔE containing complex crystallized in 19% ( w/v ) PEG 3350 , 0 . 2 M AmSO4 , and 0 . 1 M Bis-Tris-HCl pH 6 . 5 . Crystals of both complexes grew at 18°C in hanging drops containing 1 μl of protein and 1 μl of mother liquor . Clusters of diffraction quality , rod-shaped crystals formed within 3–5 days . Single crystals were briefly soaked in mother liquor supplemented with 20% ( v/v ) glycerol for cryoprotection and flash-frozen in liquid nitrogen . X-ray data were collected at NE-CAT beamline 24-ID-C at Argonne National Laboratory . Data reduction was performed with the HKL2000 package ( Otwinowski and Minor , 1997 ) , and all subsequent data-processing steps were carried out using programs provided through SBGrid ( Morin et al . , 2013 ) . The structure of the TorsinAEQ-LULL1-VHH-BS2 complex was solved by molecular replacement ( MR ) using the Phaser-MR tool from the PHENIX suite ( Adams et al . , 2010 ) . A three-part MR solution was easily obtained using a sequential search for models of LULL1 , VHH-BS2 , and TorsinA . The LULL1 model was generated based on the published human LAP1 structure ( PDB 4TVS , chain A ) , using the Sculptor utility of the PHENIX suite ( LULL1241–470 and LAP1356–583 share 64% sequence identity ) . The VHH-BS2 model was based on VHH-BS1 ( PDB 4TVS , chain a ) after removing the complementarity determining regions ( CDRs ) . The poly-Ala model of TorsinA was generated based on E . coli ClpA ( PDB 1R6B ) using the MODELLER tool of the HHpred server ( Söding et al . , 2005 ) . The asymmetric unit contains one TorsinAEQ-LULL1-VHH-BS2 complex . Iterative model building and refinement steps gradually improved the electron density maps and the model statistics . The stereochemical quality of the final model was validated by Molprobity ( Chen et al . , 2010 ) . TorsinAEQΔE-LULL1-VHH-BS2 crystallized in the same unit cell . Model building was carried starting from a truncated TorsinAEQ-LULL1-VHH-BS2 structure . All manual model building steps were carried out with Coot ( Emsley et al . , 2010 ) , and phenix . refine was used for iterative refinement . Two alternate conformations of a loop in LULL1 ( residues 428–438 ) were detected in the Fo−Fc difference electron density maps of both structures , and they were partially built . For comparison , the cysteine residues of TorsinA at the catalytic site ( residues 280 and 319 in the TorsinAEQ structure ) were built in the reduced and the oxidized states , respectively . Building them as oxidized , disulfide-bridged residues consistently produced substantial residual Fo−Fc difference density , which disappeared assuming a reduced state . Statistical parameters of data collection and refinement are all given in Table 1 . Structure figures were created in PyMOL ( Schrödinger LLC ) . Torsin and LAP1/LULL1 sequences were obtained via PSI-BLAST ( Altschul et al . , 1997 ) and Backphyre searches ( Kelley and Sternberg , 2009 ) . Transmembrane domains were predicted using the HMMTOP tool ( Tusnády and Simon , 2001 ) . LAP1/LULL1 proteins were distinguished based on the calculated isoelectric point ( pI ) of their extra-luminal portions . The intranuclear domain of LAP1 has a characteristically high pI of ~8 . 5–10 due to a clustering of basic residues , while the cytoplasmic domain of LULL1 is distinctively more acidic . Multiple sequence alignments were performed using MUSCLE ( Edgar , 2004 ) , and visualized by Jalview ( Waterhouse et al . , 2009 ) . To illustrate evolutionary conservation on TorsinA and LULL1 surfaces , conservation scores for each residue were calculated using the ConSurf server with default parameters ( Glaser et al . , 2003 ) . The sequences , which were used to generate the multiple sequence alignments , were also used for preparing the sequence logos of Torsins and LAP1/LULL1 in Figure 1—figure supplement 5 . To obtain the sequence logo of the HCLR clade AAA+ ATPases , Escherichia coli ClpA-D2 ( residues 458–758 ) , Escherichia coli ClpB-D2 ( residues 568–857 ) , Bacillus subtilis ClpE-D2 ( residues 409–699 ) , Saccharomyces cerevisiae Hsp104-D2 ( residues 578–868 ) , Escherichia coli HslU ( residues 13–443 ) , Bacillus subtilis HslU ( residues 15–455 ) , Streptomyces coelicolor ClpX ( residues 71–409 ) , Drosophila melanogaster ClpX ( residues 199–634 ) , Escherichia coli Lon ( residues 320–580 ) , Caenorhabditis elegans Lon ( residues 476–771 ) , Thermus thermophilus ClpB-D2 ( residues 536–845 ) , Escherichia coli ClpX ( residues 64–403 ) , Helicobacter pylori ClpX ( residues 77–430 ) , Haemophilus influenza HslU ( 1–444 ) , Bacillus subtilis Lon ( residues 300–590 ) , Bacillus subtilis ClpC-D2 ( residues 486–802 ) , Saccharomyces cerevisiae Hsp78-D2 ( residues 482–794 ) and Arabidopsis thaliana Hsp101-D2 ( residues 547–849 ) sequences were used . All sequence logos were generated using WebLogo ( Crooks et al . , 2004 ) . The purified human TorsinAEQ-LULL1 complex was injected into a male alpaca ( Lama pacos ) for immunization . Generation and screening of nanobodies was carried out as previously described ( Sosa et al . , 2014 ) . Each of the selected nanobodies was subcloned into a pET-30b ( + ) vector with a C-terminal His6-tag . Each nanobody was bacterially expressed and Ni2+-affinity purified essentially as described ( see above ) . Different from the TorsinA-containing preparations , MgCl2 and ATP were eliminated from all buffer solutions . The Ni2+-eluate was purified via size exclusion chromatography on a Superdex S75 column ( GE Healthcare ) in running buffer ( 10 mM HEPES/NaOH pH 8 . 0 , 150 mM NaCl ) . Nanobody binding was validated by size exclusion chromatography on a 10/300 Superdex S200 column in 10 mM HEPES/NaOH pH 8 . 0 , 150 mM NaCl , 10 mM MgCl2 and 0 . 5 mM ATP . Equimolar amounts of TorsinAEQ-LULL1 and TorsinAEQ-LULL1-VHH were loaded and nanobody binding was monitored by a shift in the elution profile and via SDS-PAGE analysis . After validating VHH-BS2 interaction with TorsinAEQ-LULL1 , the C-terminal His6-tag of VHH-BS2 was removed from the pET-30b ( + ) vector for co-purification experiments . | A group of enzymes known as the AAA+ ATPase family have a wide variety of roles in the cell . They are able to break down a molecule called ATP and use the energy released to change the structure of other ‘target’ molecules . TorsinA is one such AAA+ ATPase and is found primarily in nerve cells inside two cell compartments called the endoplasmic reticulum and the perinuclear space . There , TorsinA interacts with one of two proteins – called LULL1 and LAP1 – that activate TorsinA . In this respect , TorsinA differs from other members of the AAA+ ATPase family , which can activate themselves without the need for additional proteins . TorsinA and other enzymes are made up of building blocks called amino acids . Mutant forms of TorsinA that have lost a particular amino acid cause primary dystonia , an incurable neuromuscular disease . This amino acid is needed for TorsinA to interact with LULL1 and LAP1 . Previous studies have revealed the 3D structure of LAP1 on its own , but the structure of TorsinA remained unknown . One way to study the structure of enzymes is to use a technique called X-ray crystallography . The first step in this technique is to make crystals of the protein of interest . However , it has proved difficult to make crystals of TorsinA . Demircioglu et al . have addressed this problem by using X-ray crystallography to investigate the structure of TorsinA when it is bound to LULL1 . The experiments used a small molecule known as a nanobody that can specifically recognize the human TorsinA enzyme . The nanobody helped TorsinA to stay attached to LULL1 and form the crystals needed for X-ray crystallography . The 3D structures reveal how TorsinA and LULL1 interact in a high level of detail , helping to explain how TorsinA differs from other AAA+ ATPases . In addition , by comparing how normal TorsinA and the mutant form interact with LULL1 , Demircioglu et al . provide more evidence that primary dystonia is likely to be caused by the improper activation of TorsinA . The subtle differences revealed by these structures could be exploited to develop new drugs to fight this disease in the future . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2016 | Structures of TorsinA and its disease-mutant complexed with an activator reveal the molecular basis for primary dystonia |
The balance between Th17 and T regulatory ( Treg ) cells critically modulates immune homeostasis , with an inadequate Treg response contributing to inflammatory disease . Using an unbiased chemical biology approach , we identified a novel role for the dual specificity tyrosine-phosphorylation-regulated kinase DYRK1A in regulating this balance . Inhibition of DYRK1A enhances Treg differentiation and impairs Th17 differentiation without affecting known pathways of Treg/Th17 differentiation . Thus , DYRK1A represents a novel mechanistic node at the branch point between commitment to either Treg or Th17 lineages . Importantly , both Treg cells generated using the DYRK1A inhibitor harmine and direct administration of harmine itself potently attenuate inflammation in multiple experimental models of systemic autoimmunity and mucosal inflammation . Our results identify DYRK1A as a physiologically relevant regulator of Treg cell differentiation and suggest a broader role for other DYRK family members in immune homeostasis . These results are discussed in the context of human diseases associated with dysregulated DYRK activity .
The appropriate and balanced differentiation of naïve CD4+ T helper ( Th ) cells into either pro-inflammatory effector subsets , such as Th1 and Th17 cells , or anti-inflammatory subsets , largely represented by regulatory T ( Treg ) cells , is an important determinant of immune homeostasis , dysregulation of which underlies the pathology of inflammatory diseases and cancer ( Josefowicz et al . , 2012; Cretney et al . , 2013 ) . Genome-wide association studies of inflammatory diseases such as type 1 diabetes ( T1D ) and inflammatory bowel disease ( IBD ) support this notion , implicating genes important for the differentiation and function of Treg cells ( Khor et al . , 2011 ) . The active translational interest in manipulating this process is exemplified by recent attempts using low-dose IL-2 to enhance Treg cells and attenuate the inflammation associated with graft-versus-host disease and HCV vasculitis ( Koreth et al . , 2011; Saadoun et al . , 2011; von Boehmer and Daniel , 2012 ) . While these results are encouraging , IL-2 has numerous effects and reflects the larger issue that more targeted therapies to specifically manipulate individual Th lineages remain lacking , due at least in part to an incomplete understanding of the pathways that contribute to Th differentiation . Our interest has focused on discovering novel pathways that regulate the differentiation of Treg cells , which represent the major anti-inflammatory Th component . The canonical pathways underlying Treg cell differentiation have been well described . In this regard , commitment to the Treg cell lineage is exemplified by expression of the hallmark transcription factor FOXP3 and occurs in either the thymus or the periphery ( Josefowicz et al . , 2012 ) . The canonical cytokine that drives Treg cell differentiation is TGF-β1 and the differentiation of peripheral Treg ( pTreg ) cells requires TGF-β1 signaling through SMAD2 and SMAD3 ( Gu et al . , 2012; Josefowicz et al . , 2012 ) . TGF-β1 also plays a role in thymic Treg ( tTreg ) cell differentiation , as exemplified by the significant but transient decrease in early Treg cell generation upon T cell-specific deletion of the TGF-β1 RI subunit ( Liu et al . , 2008 ) . However , TGF-β1 may be more important for maintaining the pool of Treg cell precursors than instructing Foxp3 expression in this compartment ( Josefowicz et al . , 2012 ) . Similarly to IL-2 , TGF-β1 exerts multiple effects on different cell types and has not proven to be a clinically useful target to manipulate Treg cell differentiation , again pointing to the need to better understand the breadth of pathways involved . In this regard , studies in SMAD2/3 doubly deficient mice point to TGF-β1-dependent , SMAD2/3-independent signals in tTreg cell differentiation and function , demonstrating the relevance of non-canonical pathways , even in the context of well-described cytokines ( Gu et al . , 2012 ) . Elucidating such ancillary pathways in Th differentiation has been approached in several ways . Notable amongst these have been gene expression profiling experiments to identify differentially expressed genes . This approach has been more successful for Th17 cell differentiation , pointing to the transcription factors Batf , Ahr and Ikzf3 and the sodium chloride sensor Sgk1 ( Veldhoen et al . , 2008; Schraml et al . , 2009; Wu et al . , 2013 ) , than for Treg cell differentiation . Such findings have implications for diagnostic efforts and advancing our understanding of disease pathophysiology . For example , the finding that mutations in STAT3 ( which transduces signals from IL-6 , a canonical Th17 cytokine ) can lead to hyper-IgE syndrome ( HIES ) led to the discovery that this subset of HIES patients fail to generate Th17 cells , potentially accounting for their susceptibility to fungal infection ( Ma et al . , 2008 ) . There are also therapeutic implications; for instance , the discovery that SGK1 regulates Th17 cell differentiation led to the hypothesis that increased dietary salt intake may contribute to increased risk of autoimmune disease ( Kleinewietfeld et al . , 2013 ) . Thus , discovering additional pathways that regulate Treg cell differentiation is an important effort that may benefit from other approaches . Integrative computational analyses represent one promising adjunctive approach . Analyses of over 100 gene expression profiles of various CD4+ subsets led to the discovery of novel transcription factors , including Lef1 and Gata1 , that regulate Treg cell differentiation and a model of how they cooperate to establish the Treg cell transcription profile ( Fu et al . , 2012 ) . In another example , the compilation of 557 publicly available microarrays covering over 100 tissues and primary cells facilitated the discovery of Zbtb25 as a transcription factor predominantly expressed in T cells that represses NFAT signaling in response to T cell receptor engagement ( Benita et al . , 2010 ) . Another emerging key approach uses chemical methods to decipher novel nodes that control signal transduction pathways within T cells; this provides an important and complementary view into disease architecture by highlighting druggable connections between disease pathways less easily uncovered genetically . In this regard , defects in autophagy have been associated with IBD . Efforts to find compounds that enhance autophagy led to the observation that some autophagy-enhancing compounds specifically inhibit Th17 cell differentiation while another subset specifically enhances Treg cell differentiation , suggesting that these compounds highlight targets which modulate distinct sets of disease-relevant pathways ( Shaw et al . , 2013 ) . Finally , chemoinformatic methods can help generate high-yield mechanistic hypotheses based on relevant compounds identified by chemical biology approaches . For instance , the use of chemoinformatics to predict novel binding targets for clinically used drugs based on structural similarity to other compounds that bind said targets has helped predict mechanistic explanations for clinically observed side effects ( Keiser et al . , 2007; Lounkine et al . , 2012 ) . Of note , these approaches are not mutually exclusive , but rather are expected to be synergistic . Supporting the value of a chemical biology approach , compounds previously identified to modulate Treg cell differentiation have provided important insights into relevant signaling modules . For example , mechanistic studies of all-trans retinoic acid ( ATRA ) and rapamycin , two well-studied Treg cell enhancers , pointed to roles for RAR-α and mTOR signaling in Treg cell differentiation respectively ( Coombes et al . , 2007; Mucida et al . , 2007; Sun et al . , 2007; Haxhinasto et al . , 2008; Hill et al . , 2008; Sauer et al . , 2008; Hall et al . , 2011 ) . More recently , the discovery of the microbial metabolites proprionate and butyrate as enhancers of Treg cell differentiation , amongst other effects , have highlighted roles for the short-chain fatty acid receptor GPR43 and histone deacetylases in Treg cell differentiation ( Arpaia et al . , 2013; Furusawa et al . , 2013; Smith et al . , 2013 ) . These studies highlight several SMAD-distinct signals in Treg cell differentiation and illustrate how the discovery of novel molecules can facilitate a deeper understanding of the underlying mechanisms and pathways affecting Treg cell differentiation . Th differentiation is a complex cellular process for which there exists no good simplified substitute assay . Chemical biology approaches to study this process have typically either maintained the complexity of the system ( i . e . , used primary CD4+ T cells ) to study one or two selected compounds , for example ATRA , or used larger chemical libraries to interrogate a highly simplified system . Here , we take the novel approach of applying unbiased chemical biology to primary CD4+ T cells in order to discover novel regulators of Treg cell differentiation . We report 14 novel compounds that specifically enhance the differentiation of Treg cells , but of neither Th1 nor Th17 cells . In particular , the β-carboline alkaloid harmine enhances the differentiation of Treg cells and potently inhibits Th17 cell differentiation , at least in part by inhibiting the activity of the kinase DYRK1A . Importantly , we demonstrate that harmine-enhanced Treg cells retain normal suppressive function in vitro and attenuate disease in experimental models of systemic autoimmunity and mucosal inflammation in two distinct compartments . Notably , direct administration of harmine attenuates airway inflammation in an experimental model of asthma . Our approach exemplifies how chemical biology can be applied to a physiologically relevant experimental system with a functional readout to identify DYRKs as a novel and druggable pathway that impacts Treg cell differentiation .
We hypothesized that identifying novel compounds that enhance Treg cell differentiation would enable the discovery of novel pathways that regulate this process . Accordingly , we designed an experimental workflow that feeds into an integrative computational analysis pipeline to identify small molecules that specifically enhance differentiation of Treg cells , but not pro-inflammatory lineages , highlight putative mechanistic classes and demonstrate functional relevance of prioritized small molecule ( s ) ( Figure 1A ) . Primary murine CD4+ T cells were reproducibly differentiated into Treg , Th1 or Th17 lineages in a manner dependent upon the concentration of TGF-β1 , IL-12 or IL-6 and/or IL-1β respectively ( Figure 1—figure supplement 1 ) . Lineage commitment was determined using the gold standard of flow cytometric detection of FOXP3 , IFNγ and/or IL-17 . The role of these canonical cytokines has been well described; the positive control for each lineage was high concentrations of lineage-driving cytokines consistent with published literature while negative controls included cells cultured under Th0 conditions without any lineage-promoting cytokines , as well as cells driven to other lineages ( e . g . , for Treg cells , negative controls were Th0 , Th1 and Th17 conditions ) . Conditions driving the differentiation of sub- or near-maximal levels ( typically about 30% and 95% of maximal levels , respectively ) of Treg , Th1 and Th17 cells ( hereafter Treglow , Th1low , Th17low , Treghi , Th1hi and Th17hi conditions ) were identified ( Figure 1—figure supplement 1 , blue and red arrowheads respectively ) . Addition of a compound to sub-maximal conditions allows quantitation of its ability to enhance lineage-specific differentiation , while addition to near-maximal conditions allows quantitation of its inhibitory effect . To facilitate comparisons between experiments , we used the fractional enhancement ( Fr enhance ) metric , where the compound-driven difference in Th differentiation is normalized against the difference between the positive and negative controls within the experiment ( i . e . , Thhi − Thlow ) . 10 . 7554/eLife . 05920 . 003Figure 1 . Chemical biology approach to identify novel Treg enhancers . All data representative of at least 2 independent experiments . ( A ) Overview of our approach , including key methods applied . ( B ) Dose-response curves showing fractional enhancement ( Fr enhance ) of compounds ( LD50/EC50 > 2 ) for Treg ( blue ) , Th1 ( orange ) and Th17 ( red ) lineages . ( C ) Plot of LD50/EC50 vs maximal fractional Treg cell enhancement showing all 21 Treg-specific enhancers ( 9RA , 9-cis retinoic acid; 13RA , 13-cis retinoic acid; ADQ , amodiaquine; AMIO , amiodarone; AMR , amrinone; ATRA , all-trans retinoic acid; ART , artemisinin; CIS , cisapride; CLO , clotrimazole; HAR , harmine; LOV , lovastatin; MBCQ , 4- ( ( 3 , 4-methylenedioxybenzyl ) amino ) -6-chloroquinazoline;4-quinazolinamine ) ; PEN , pentamidine; PG , proguanil; PYR , pyrvinium pamoate; RAPA , rapamycin; RIB , ribavirin; ROT , rotenone; SER , sertaconazole; SIM , simvastatin; WM , wortmannin; Supplementary file 2 ) . Retinoic acids are in green; compounds with LD50/EC50 < 2 are in gray and LD50/EC50 > 2 are in blue . The orange cluster is described in the text . ( D ) Ability of selected compounds ( LD50/EC50 > 2 and controls ) to inhibit mTOR activity , as measured by S6 phosphorylation ( ***p < 0 . 001 , 1-way ANOVA with Dunnett correction ) . ( E ) Fractional ( Fr ) inhibitory activity of all 21 Treg-specific enhancers on Th1 and Th17 cell differentiation . ( F ) SEA-predicted relationships between all 21 Treg enhancers . Black lines predict binding of compounds ( red circles ) to proteins ( blue diamonds ) with likelihood proportional to line width . Green lines denote connection via curated KEGG pathways . See also Figure 1—figure supplements 1–8 . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 00310 . 7554/eLife . 05920 . 004Figure 1—figure supplement 1 . Titrating Th differentiation conditions . Titrating cytokines for Treg , Th1 , and Th17 conditions identifies sub-maximal ( blue arrowheads ) and near-maximal ( red arrowheads ) lineage-promoting conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 00410 . 7554/eLife . 05920 . 005Figure 1—figure supplement 2 . Culture cellularity affects Treg differentiation . Treglow ( blue ) stimulation with varying initial cell numbers recapitulates the inverse relationship between final culture cellularity and percentage of FOXP3-expressing cells . Treghi ( red ) conditions generate uniformly high levels of FOXP3-expressing cells independent of culture cellularity . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 00510 . 7554/eLife . 05920 . 006Figure 1—figure supplement 3 . Schematic of analytic and hit-calling pipeline . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 00610 . 7554/eLife . 05920 . 007Figure 1—figure supplement 4 . Effect of compounds ( LD50/EC50 < 2 ) on Th differentiation . Dose-response curves showing compound effect on fractional enhancement of Treg ( blue ) , Th1 ( orange ) , and Th17 ( red ) lineages . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 00710 . 7554/eLife . 05920 . 008Figure 1—figure supplement 5 . Modeling analyses to calculate EC50 values , indicated in parentheses , for all 21 Treg-specific enhancers . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 00810 . 7554/eLife . 05920 . 009Figure 1—figure supplement 6 . Modeling analyses to calculate LD50 values , indicated in parentheses , for all 21 Treg-specific enhancers . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 00910 . 7554/eLife . 05920 . 010Figure 1—figure supplement 7 . Similarity clustering analysis of combined Treg , Th1 and Th17 phenotypic data . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 01010 . 7554/eLife . 05920 . 011Figure 1—figure supplement 8 . Euclidean distance clustering analysis of gene expression data from cell lines treated with Treg enhancers first analyzed by principal component analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 011 To find small molecules that enhance Treg cell differentiation , Treglow conditions were used to screen 3281 compounds comprising FDA-approved drugs and tool compounds with known mechanisms ( Shaw et al . , 2013 ) . Our studies revealed a previously unreported negative correlation between the number of live cells in culture and the percentage of FOXP3+ cells in Treglow conditions , not observed in Treghi , Th1low/hi or Th17low/hi conditions ( Figure 1—figure supplement 2 ) . Because the compounds tested exhibit variable effects on cellularity , we accounted for corresponding effects on Treg cell differentiation using a linear regression model ( Figure 1—figure supplement 3 ) . Numerous compounds previously reported to enhance Treg cell differentiation were recovered , including the hypolipidemic statins ( lovastatin and simvastatin ) , artemisinin and ATRA as well as related retinoic acids ( 9-cis retinoic acid and 13-cis retinoic acid ) , validating our experimental approach ( Coombes et al . , 2007; Mucida et al . , 2007; Sun et al . , 2007; Kagami et al . , 2009; Kim et al . , 2010; Zhao et al . , 2012 ) . The fractional enhancement of the weakest of these known enhancers ( artemisinin , 0 . 3 ) was used as a minimum threshold to find all compounds that enhance Treg cell differentiation at least as strongly . By this criterion , 70 compounds were selected for retesting . The compounds prioritized by our efforts described above were retested ( using Treglow conditions ) at 8 doses that typically covered over a 1000-fold difference in concentration , allowing us to capture more optimal concentrations at which a compound might work ( Figure 1B ) . In order to filter for compounds that only enhance Treg cell differentiation , these compounds were also tested under Th1low and Th17low conditions to quantitate their ability to enhance differentiation of pro-inflammatory lineages . Our results also consolidate and validate that the previously described Treg enhancers , including the statins , retinoic acids and artemisinin , specifically enhance Treg differentiation . Simultaneously testing multiple Th lineages and drug concentrations allows a more complete characterization of the effects of a compound on Th differentiation and furthers mechanistic conclusions . We identified 14 compounds hitherto unreported to specifically enhance differentiation of Treg , but neither Th1 nor Th17 , cells ( Figure 1B , C and Figure 1—figure supplement 4 ) . To demonstrate that these novel Treg cell enhancers work distinctly from known canonical pathways , we assessed their activity on mTOR activity . We selected this pathway for three primary reasons . Firstly , the role of mTOR inhibition on enhancing Treg cell differentiation has been well described . Secondly , rapamycin is a well-known mTOR inhibitor and enhancer of Treg cell differentiation , and thus provides a good positive control . Finally , mTOR activity is relatively easily assessed , for example by assessing the phosphorylation state of S6 , which is phosphorylated in the course of mTOR activation . Thus , mTOR inhibition leads to fewer phospho-S6+ cells by flow cytometry . We characterized the effect of all 21 Treg cell enhancers on S6 phosphorylation in primary CD4+ T cells cultured under Treglow conditions ( Figure 1D ) . Only rapamycin , the positive control , significantly inhibited S6 phosphorylation ( 1-way ANOVA with Dunnett correction , threshold p < 0 . 05 ) . Thus , all 14 novel Treg cell enhancers appear to work independently of mTOR and potentially point to undiscovered mechanisms . We sought to identify compounds with minimal impact on cellularity , given its relationship with Treg cell differentiation . To this end , the LD50 and EC50 were determined as the doses at which 50% cytotoxicity and 50% Treg cell enhancement are observed , respectively ( Figure 1—figure supplements 5 , 6 ) . Compounds were classified according to both the LD50/EC50 ratio ( analogous to the therapeutic index ) and the maximal enhancement of Treg cell differentiation , with ideal compounds performing maximally for both parameters ( Figure 1C , blue circles ) . Many compounds , including rapamycin , exhibited significant cytotoxicity with an LD50/EC50 ratio near 1 ( Figure 1C , gray triangles ) . All 21 Treg cell-specific enhancers were additionally tested under Th1hi and Th17hi conditions to accurately quantitate their capacity to inhibit differentiation into these pro-inflammatory lineages ( Figure 1E ) . Th17 cell differentiation was typically more inhibited ( ≥40% ) than Th1 , likely related to Treg cells and Th17 cells arising from a common progenitor ( Figure 1E ) ( Zhou et al . , 2008 ) . Unsupervised analysis of this phenotypic data revealed high similarity between artemisinin , cisapride and sertaconazole , including moderate enhancement of Treg cell differentiation and potent inhibition of both Th1 and Th17 cell differentiation , which may reflect effects on common pathways and direct future studies ( Figure 1C and Figure 1—figure supplement 7 ) . Importantly , these results prioritized MBCQ , harmine , and amrinone as novel enhancers of Treg cell differentiation with favorable phenotypic profiles ( Figure 1—figure supplement 7 ) . In order to further explore potential mechanistic relationships between these Treg cell enhancers , we applied previously described chemoinformatic approaches . Similarity Ensemble Approach ( SEA ) is one such method that utilizes similarities in chemical structure between compounds to predict the likelihood that they could bind common protein targets that have been defined in previous efforts ( Keiser et al . , 2007 ) . Applying SEA to our list of Treg cell enhancers generated several clusters of compounds predicted to bind the same protein ( Figure 1F , black lines ) . Additionally , we recognized a need to account for relationships between these clusters , for example with compounds acting on separate proteins that act within the same pathway . These connections were identified using a manually curated list of KEGG pathways that excludes overly generic and largely populated pathways ( e . g . , Pathways in Cancer ) that would report spurious relationships ( Figure 1F , green lines ) ( Goel et al . , 2014 ) . These results suggested inter-relationships between most of our compounds with two outlier pairs , one comprising the statins and the other comprising harmine and MBCQ ( Figure 1F ) . The L1000 method had previously been used to generate gene expression data from three different cell lines treated with numerous compounds , including most of the Treg cell enhancers identified here ( expression data in GSE5258 ) ( Lamb et al . , 2006 ) . Principal component analysis was used to analyze changes in gene expression after treatment with Treg cell enhancers . These results indicated significant commonalities between most of the Treg cell enhancers identified here , with harmine and the retinoic acids generating the most distinct profiles ( Figure 1—figure supplement 8 ) . Together , our results prioritize harmine for its favorable phenotypic profile and likelihood of mechanistic novelty , given its distinct properties in our chemoinformatic analyses . To further validate our interest in the physiologic relevance of harmine's ability to enhance Treg cell differentiation , we tested the functionality of Treg cells generated under Treglow + harmine ( henceforth abbreviated as TregHAR ) conditions extensively . First , we used an in vitro suppression assay , where Treg cells are co-cultured at increasing dilutions with CFSE-labeled responder CD4+ T cells and their ability to suppress responder T cell proliferation upon anti-CD3/CD28 stimulation is assessed . Naïve CD4+ T cells from Foxp3IRES-GFP mice were cultured under either TregHAR or Treghi conditions and the resulting GFP+ Treg cells were sorted by FACS . Sorted Treg cells generated using either Treghi or TregHAR conditions equivalently suppressed the proliferation of co-cultured responder CD4+ T cells in vitro at each dilution , indicating equal efficacy between both populations of Treg cells ( Figure 2A , red and blue lines ) . Both populations worked better than sorted tTreg cells , likely because they are pre-activated ( Figure 2A , orange line ) . 10 . 7554/eLife . 05920 . 012Figure 2 . Harmine-enhanced Treg cells and harmine attenuate inflammation . ( A–E ) Experiments comparing the suppressive activity of Treg cells generated under either TregHAR ( blue ) or Treghi ( red ) conditions; conditions without Treg cells are shown in black . All data representative of at least 2 independent experiments; in vivo experiments used at least 3 mice per cohort . ( A ) In vitro suppression assay , with unstimulated tTreg cells in orange . ( B ) Treghi-Treg cells and TregHAR-Treg cells similarly delay onset of diabetes in the NOD-BDC2 . 5 model of type 1 diabetes . ( C ) Comparable inhibition of inflammation by Treghi-Treg cells and TregHAR-Treg cells in the CD45RBhi transfer model of colitis . Representative images are shown in ( D ) . Bars represent 100 μm . ( E ) Similar inhibition of airway inflammation by Treghi-Treg cells and TregHAR-Treg cells , as measured by number of eosinophils ( Eos ) in bronchoalveolar lavage ( BAL ) fluid , in a model of asthma . ( F ) Comparison of Treg cells ( as a percentage of total CD4+ T cells ) in the thoracic lymph nodes of mice treated with intranasal harmine , vs mock treatment . ( G ) Intranasal administration of harmine prior to immunization inhibits recall airway inflammation in the asthma model . Right , representative images of inflammation around airway vessels . **p < 0 . 01 , *p < 0 . 05 , ns , not significant , Mantel-Cox test ( B , C ) and Student's t-test ( E , F ) . See also Figure 2—figure supplements 1–2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 01210 . 7554/eLife . 05920 . 013Figure 2—figure supplement 1 . Effects of harmine treatment in vivo on T cell populations . Mice were administered either intranasal harmine HCl ( blue ) or water ( gray ) . T cell populations in thoracic lymph nodes were quantitated as shown , demonstrating relative percentages ( above ) and absolute numbers ( below ) . All statistically significant differences are indicated , **p < 0 . 01 , ***p < 0 . 001 , Student's t-test with Holm-Sidak correction . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 01310 . 7554/eLife . 05920 . 014Figure 2—figure supplement 2 . Effects of harmine treatment in vivo on antigen-presenting cell populations . Mice were administered either intranasal harmine HCl ( blue ) or water ( gray ) . Migratory and classical dendritic cells ( mDC and cDC , respectively ) in thoracic lymph nodes were quantitated ( top right ) , and their respective expression of costimulatory molecules ( CD40 , CD80 , and CD86 ) compared . ns , not significant , Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 014 We also compared the ability of Treghi- and TregHAR-Treg cells to inhibit inflammation in vivo . For this purpose , we selected three experimental models in which inflammation is mediated by T cells and can be suppressed by Treg cells . These models were also selected to represent different genetic backgrounds , inflammation in different sites and different Treg cell antigen specificities . In a model of T1D induced by transfer of NOD-BDC2 . 5+ CD4+ T cells into NOD-scid recipients , diabetes developed rapidly approximately 10 days later without any intervention ( Figure 2B , black line ) . When antigen-specific Treg cells generated from NOD-BDC2 . 5 . Foxp3IRES-GFP mice under either TregHAR or Treghi conditions were co-transferred , a consistent and indistinguishable delay of onset of diabetes was observed , with median time of diabetes onset being delayed by at least 7 days ( Figure 2B , blue and red lines ) ( Herman et al . , 2004; Tarbell et al . , 2004 ) . Using a well-described T cell-dependent model of colitis , transfer of C57Bl/6 CD4+CD45RBhi T cells into C57Bl/6-Rag1−/− hosts led to intestinal inflammation 8 weeks later ( Figure 2C , D , no Treg ) ( Powrie et al . , 1993 ) . This mucosal inflammation was significantly attenuated when antigen-naïve Treg cells generated from C57Bl/6-Foxp3IRES-GFP mice under Treghi conditions were transferred , as determined by histological scoring of intestinal sections by blinded observers ( Figure 2C , D ) ( Smith et al . , 2013 ) . Transfer of Treg cells generated under TregHAR conditions attenuated intestinal inflammation to a level indistinguishable from that achieved by transfer of Treghi-Treg cells ( Figure 2C , D ) . Finally , the functionality of TregHAR-Treg cells was tested in a model of airway inflammation . Here , C57Bl/6 mice are sensitized against ovalbumin and subsequent challenge with intratracheally administered ovalbumin leads to airway inflammation ( Figure 2E ) ( Grainger et al . , 2010 ) . This inflammation was attenuated when antigen-naïve Treg cells generated from C57Bl/6-Foxp3IRES-GFP mice under Treghi conditions were transferred prior to the intratracheal challenge ( Figure 2E ) . Importantly , transfer of Treg cells generated under TregHAR conditions led to a comparable suppression of inflammation ( Figure 2E ) . The observation that harmine promotes the differentiation of Treg cells , at least in vitro , that appear fully functional raises the interesting hypothesis that treatment with harmine itself could attenuate inflammation in vivo . Rapid first pass metabolism ( <2 hr ) consistent with prior reports confounded the interpretation of systemic delivery experiments ( Callaway et al . , 1999 ) . Reasoning that application of harmine to mucosal surfaces might allow for relatively prolonged local presence , we treated mice with harmine intranasally for 5 days and examined the effect on Treg cells in the draining lymph nodes . Compared to mice treated only with vehicle ( water ) , mice treated with harmine exhibited a statistically significant increase ( ∼20% ) in the frequency of Treg cells in the draining thoracic lymph nodes; increases in absolute numbers of effector T cell subsets did not reach statistical significance ( Figure 2F and Figure 2—figure supplement 1 ) . Analyses of dendritic cell populations did not show any effect of treatment with harmine on expression of Treg-relevant costimulatory molecules , consistent with the notion that harmine predominantly acts directly on CD4+ T cells to affect Treg/Th17 differentiation in this model ( Figure 2—figure supplement 2 ) . To determine if this pro-Treg effect might impact inflammation , we adapted the model of airway inflammation described above , where sensitivity to ovalbumin is induced by immunization . Intranasal administration of ovalbumin 5–7 days prior to immunization attenuated the airway inflammation induced by subsequent intratracheal challenge ( Figure 2G ) . This finding supports the notion that exogenous signals can modulate the inflammatory response mounted at the time of immunization . Strikingly , intranasal administration of only harmine during this window inhibited airway inflammation at least as potently as tolerization with ovalbumin ( Figure 2G ) . Our results demonstrate that harmine is a novel , potent , and specific enhancer of Treg cell differentiation with physiologically relevant effects ( Figures 1B , 3A ) . In addition to its pro-Treg effect , harmine strongly inhibits Th17 cell differentiation ( Figure 3A ) . Notably , even in pro-inflammatory Th17low or Th17hi conditions , harmine modestly promotes the paradoxical differentiation of Treg cells approximately twofold ( Figure 3A ) . At the doses used , harmine does not significantly affect culture cellularity , unlike ATRA and rapamycin ( Figure 3B ) . This observation is further substantiated by CFSE studies of cellular proliferation , which show that harmine causes a modest 24% reduction in proliferating cells at day 3 that falls to undetectable levels by day 4 , less than half the reduction caused by rapamycin ( Figure 3—figure supplement 1 ) . Accordingly , harmine enhances absolute numbers of Treg cells to levels approaching Treghi conditions and decreases absolute numbers of Th17 cells ( Figure 3B and Figure 3—figure supplement 2 ) . Importantly , similar effects are observed using human CD4+ T cells , with addition of harmine potently enhancing both percentage and absolute numbers of Treg cells beyond even Treghi conditions ( Figure 3C ) . These findings underscore the physiologic relevance of harmine-related pathways to human Th differentiation . 10 . 7554/eLife . 05920 . 015Figure 3 . Harmine's effects on canonical Treg/Th17 pathways . All data representative of at least 2 independent experiments . ( A ) Effect of harmine on murine Treg , Th1 and Th17 differentiation . ( B ) Effect of harmine on absolute numbers of total live and Treg cells . ( C ) Effect of harmine on human Treg differentiation . ( D ) Effect of harmine on Th differentiation under Th0 conditions as shown by percentage of cells with indicated markers . ( E ) Harmine's pro-Treg effect when added or removed at different times after Treglow stimulation as shown by percentage of maximal Treg enhancement ( % max Treg enh ) . ( F and G ) Volcano plots comparing p value vs fold change in gene expression in naïve CD4+ T cells , Treghi-Treg cells and TregHAR-Treg cells as indicated . Previously reported signature genes for Treg cells ( F ) and activated T cells ( G ) are highlighted in red ( upregulated ) and green ( downregulated ) . Numbers on the right and left reflect genes that are up- and down-regulated in the indicated comparison respectively with χ2 test p values in the middle . ( H ) Median fluorescence intensity ( MFI ) of FOXP3 in Treg cells generated under indicated conditions . ( I ) Time-course analysis of FOXP3 expression in cells cultured under indicated conditions . ( J and K ) Western blot analyses showing effect of Treg enhancers on S6 kinase , SMAD2 and SMAD3 phosphorylation when added under Treglow conditions . Numbers denote fractional phosphorylation relative to Treglow conditions . ( L ) Effect of harmine ( blue ) on RORγT expression and STAT3 phosphorylation in Th17hi conditions ( gray ) . ( M ) qPCR analyses showing effects of harmine on key Th17 genes at 4 days ( left ) and 2 hours ( right ) after stimulation . Gray and blue bars represent Th17hi and Th17hi + harmine conditions , respectively . *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 , ****p < 0 . 0001 , ns , not significant , Student's t-test with Holm-Sidak correction ( B , C , D , M ) . See also Figure 3—figure supplements 1–5 . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 01510 . 7554/eLife . 05920 . 016Figure 3—figure supplement 1 . Effect of harmine on cellular proliferation . CFSE-labelled CD4+ T cells were stimulated under Treglow conditions and proliferation assessed daily . The effect of adding harmine , rapamycin , or high TGFβ ( Treghi ) was compared as shown . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 01610 . 7554/eLife . 05920 . 017Figure 3—figure supplement 2 . Effect of harmine on absolute numbers of Th17 cells . ***p < 0 . 001 , Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 01710 . 7554/eLife . 05920 . 018Figure 3—figure supplement 3 . Effect of addition or removal of harmine at different times on Th17 differentiation . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 01810 . 7554/eLife . 05920 . 019Figure 3—figure supplement 4 . Correlation between gene expression in Treghi-Treg cells and TregHAR-Treg cells , relative to naïve CD4+ T cells . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 01910 . 7554/eLife . 05920 . 020Figure 3—figure supplement 5 . Qualitative analyses of genomewide expression in TregHAR-Treg cells . Previously described methods were used to identify similarities between TregHAR-Treg cells and other specialized Treg subsets ( Joller et al . , 2014 ) . Volcano plots compare relative expression of genes in Treghi-Treg cells and TregHAR-Treg cells . Overlaid are genes of previously identified signatures; red and green reflect genes up- and down-regulated in these signatures respectively . Numbers on the right and left reflect genes that are up- and down-regulated in the indicated comparison respectively with χ2 test p values in the middle . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 020 In the context of neutral Th0 conditions , addition of harmine does not skew Th differentiation towards any lineage , demonstrating that harmine's pro-Treg effect requires exogenous TGF-β1 ( Figure 3D ) . Thus , harmine does not substitute for TGF-β1 but rather acts in conjunction with TGF-β1 . In order to better characterize how harmine acts to enhance Treg cell differentiation , CD4+ T cells were cultured in Treglow conditions and harmine was added 1 , 2 or 3 days later . If the molecular targets of harmine were only expressed early in Th differentiation , then addition of harmine at later time points would have no effect on Treg cell differentiation . Our results show that adding harmine as late as day 3 ( out of 4 ) of culture still significantly enhances Treg cell differentiation ( Figure 3E , top panel ) . The converse experiments , where culture is initiated in TregHAR conditions and harmine removed 4 hr , 1 day or 2 days later , showed complementary results . The earlier harmine was removed , the less Treg cell differentiation was enhanced , although enhancement could be detected with as little as 4 hr of exposure to harmine ( Figure 3E , bottom panel ) . These results not only reinforce the conclusion that the targets of harmine that impact Th differentiation are present throughout the process of differentiation , but also demonstrate that harmine does not impart long-lasting epigenetic signals . If that were the case , maximal Treg cell enhancement would be observed even if harmine were removed after a short time . Corresponding reciprocal results were obtained when harmine was either added or removed at different times in the context of Th17 cell differentiation ( Figure 3—figure supplement 3 ) . To determine the effect of harmine on gene expression in Treg cells , RNA was isolated from FACS-sorted Treg cells , generated under either TregHAR or Treghi conditions , and analyzed by Illumina microarray ( data in GSE67961 ) . These results revealed significant similarity between the expression profiles of Treghi- and TregHAR-Treg cells ( Pearson correlation coefficient = 0 . 95 , Figure 3—figure supplement 4 ) . As might be expected from this result , TregHAR-Treg cells showed concordant regulation of previously described canonical Treg cell signature genes ( Figure 3F ) ( Feuerer et al . , 2010 ) . We found no evidence of significant similarity to Treg cells specialized to suppress Th1 , Th2 or Th17 cells ( CXCR3+ , IRF4+ and GFP-FOXP3-fusion Treg cells respectively , Figure 3—figure supplement 5 ) ( Joller et al . , 2014 ) . However , compared to Treghi-Treg cells , TregHAR-Treg cells showed a bias suggesting increased activation ( Figure 3G ) ( Joller et al . , 2014 ) . In addition , flow cytometry studies showed that , after gating on FOXP3+ Treg cells , TregHAR-Treg cells express FOXP3 at levels at least as high as , if not higher than , those of Treghi-Treg cells ( Figure 3H ) . Together , these results suggest that harmine promotes the differentiation of Treg cells that are of similar to superior function as compared to those driven by high levels of TGF-β1 alone . We next examined harmine's effects on 3 canonical pathways of Treg cell differentiation , namely FOXP3 expression , mTOR activity and TGF-β1/SMAD signaling . To determine if harmine promotes earlier expression of FOXP3 , we analyzed FOXP3 expression by flow cytometry at daily intervals during Treg cell differentiation . The kinetics of FOXP3 expression between Treglow , Treghi and TregHAR conditions were indistinguishable between days 0–2 ( Figure 3I ) . At day 3 , the percentage of FOXP3+ cells consistently decreased in Treglow conditions , while increasing identically in Treghi and TregHAR conditions ( Figure 3I ) . These results argue against the notion that harmine enhances Treg cell differentiation by driving earlier expression of FOXP3 . They also suggest that high levels of TGF-β1 do not accelerate the early kinetics of FOXP3 expression , and the enhanced Treg cell differentiation seen in Treghi conditions may reflect either stabilization of the FOXP3+ state and/or higher TGF-β1 levels available at later timepoints to continue driving Treg cell differentiation . To complement our flow cytometric studies of mTOR activity , we measured phosphorylation of another protein , S6-kinase , by Western blot . Again , only rapamycin inhibited S6-kinase phosphorylation , confirming that harmine does not inhibit mTOR activity ( Figure 3J ) . Finally , we measured phosphorylation of SMAD2 and SMAD3 to determine if harmine enhances TGF-β1 signaling through these canonical molecules . Phosphorylation of both SMAD2 and SMAD3 was increased upon stimulation under Treglow conditions compared to naïve CD4+ T cells , consistent with engagement of TGF-β1 signals ( Figure 3K ) . As previously reported , addition of ATRA leads to increased phosphorylation of SMAD3 , but not SMAD2 ( Figure 3K ) ( Xiao et al . , 2008 ) . Importantly , there was no further increase in SMAD2/3 phosphorylation upon further addition of harmine , rapamycin , or TGF-β1 ( Figure 3K ) . Taken together , these data indicate that harmine does not enhance Treg cell differentiation by amplifying signaling through these canonical pathways . Notably , our finding that Treghi conditions do not enhance SMAD2/3 phosphorylation beyond Treglow conditions further demonstrates that TGF-β1 itself engages pertinent and quantitative SMAD2/3-independent signals outside of tTreg cells ( Figure 3K ) . Th17 cell differentiation centrally involves IL-6 signaling through STAT3 , which in turn leads to expression of RORγT , the hallmark Th17 transcription factor . Kinetic analyses showed increased STAT3 phosphorylation and RORγT expression upon activation in Th17hi conditions ( Figure 3L ) . No difference was observed when harmine was added , indicating that neither signaling event is affected by harmine ( Figure 3L ) . We verified by qPCR that harmine inhibits expression not only of the effector molecules Il17a and Il22 , but also of several key regulators of the Th17 pathway , including Batf , Pou2af1 and Mina ( Figure 3M ) ( Schraml et al . , 2009; Yosef et al . , 2013 ) . These results suggest that harmine works on novel target ( s ) to modulate Treg and Th17 cell differentiation and highlight a druggable point between STAT3/RORγT signaling and other Th17 transcription factors . To gain insight into harmine-regulated genes and pathways , we compared the expression profiles of TregHAR-Treg cells and Treghi-Treg cells . Since harmine also inhibits Th17 cell differentiation , we focused on a 111-gene signature that was concordantly up/down-regulated in TregHAR-Treg cells vs Treghi-Treg cells , as well as in human Treg cells vs Th17 cells ( Figure 4A ) ( Zhang et al . , 2013 ) . To assess if these effects might be relevant to human disease , we evaluated the overlap between these 111 genes ( and the 16 transcription factors whose binding sites were overrepresented therein ) with the 1437 genes that lie within IBD-associated loci as previously reported ( Jostins et al . , 2012; Okada et al . , 2014 ) . The overlap of 16 genes represented a significant ( p < 0 . 01 ) enrichment of 1 . 76-fold , suggesting that harmine impacts Treg cell-relevant genes implicated in IBD ( Figure 4B ) . Of these , we were particularly interested in BACH2 , a transcription factor linked to Treg cell development and inflammation , as well as to pediatric IBD ( Christodoulou et al . , 2013; Roychoudhuri et al . , 2013 ) . BACH2-deficient mice exhibit a progressive wasting disease with autoantibody formation and inflammation in the lung and gut leading to decreased survival due , at least in part , to defective Treg cell development and function ( Roychoudhuri et al . , 2013 ) . Independent qPCR experiments confirmed differential expression of 5 of the 6 TregHAR signature genes with predicted BACH2 binding sites , supporting the notion that harmine enhances Treg cell differentiation at least in part by modulating BACH2 signaling ( Figure 4C ) . During polarization of naïve CD4+ T cells under pro-Treg conditions , BACH2 stabilizes Treg differentiation by suppressing transcriptional programmes associated with other Th lineages; a corresponding BACH2-dependent signature has been identified ( Roychoudhuri et al . , 2013 ) . Intriguingly , a significant number of these BACH2-regulated genes are inversely regulated by harmine , suggesting that harmine may help reverse BACH2-axis defects , for example in IBD ( Figure 3—figure supplement 5 ) . 10 . 7554/eLife . 05920 . 021Figure 4 . Mechanistic dissection of harmine . ( A ) Comparison of expression profiles suggesting a harmine-relevant Treg signature . The black bar separates 2 independently row-normalized experiments . ( B ) Harmine signature genes are enriched for genes in IBD loci . ( C ) Validation of top signature genes by qPCR ( all p < 0 . 05 , Student's t-test ) , including genes with BACH2 binding sites ( B ) . ( D ) Effect on harmine on Treg/Th17 differentiation in BACH2-deficient ( orange ) vs -sufficient ( black ) cells . Conditions in the presence and absence of neutralizing antibodies are indicated by solid and dotted lines , respectively . ( E ) Effect of compounds that inhibit different harmine targets ( indicated in parentheses ) on Treg cell differentiation . ( F ) DYRK inhibitors suppress Th17 cell differentiation . ( G ) Increased levels of DYRK1A in Th17 vs Treg cells . Upper left , Western blot analyses of sorted Treg/Th17 cells with relative expression enumerated below . Histograms show FACS analyses of DYRK1a in Treg cells ( blue ) compared to either Th17 ( red ) or non-Th17 ( black ) cells . ( H ) Knock-down of Dyrk1a ( D1a ) enhances Treg ( left ) and inhibits Th17 ( right ) cell differentiation . Cells treated with non-targeting shRNA ( Ctrl ) and no shRNA ( None ) are shown for comparison . ( I ) Amnis analyses showing nuclear-overlapping NFAT1 signal before ( black ) and after stimulation in Treglow conditions with ( blue ) or without ( red ) harmine . Representative images ( right ) illustrate cytoplasmic and nuclear NFAT1 localization in cells outside and within the gate , respectively . ( J ) Western blot analyses quantitating nuclear fraction of NFAT1 in cells stimulated in Treglow ± harmine conditions . All data in D–J representative of at least 2 independent experiments . See also Figure 4—figure supplements 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 02110 . 7554/eLife . 05920 . 022Figure 4—figure supplement 1 . Effect of harmine on NFAT1 nuclear localization with time . Time-course Amnis analyses showing effect of harmine ( blue ) relative to control Treglow conditions ( black ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 02210 . 7554/eLife . 05920 . 023Figure 4—figure supplement 2 . Comparing effects of DYRK1A deficiency to harmine treatment . RNAseq experiments in primary CD4+ T cells comparing the effects of transduction with Dyrk1a shRNA ( shDyrk1a ) to harmine treatment in control shRNA-transduced cells ( left ) . Further statistical comparison uses volcano plots comparing p value vs fold change in gene expression in harmine-treated and -untreated ( control ) cells ( middle and right panels ) . Genes differentially regulated by Dyrk1a knockdown are highlighted in red ( upregulated ) and green ( downregulated ) . Numbers on the right and left reflect genes that are up- and down-regulated , respectively , in harmine vs control conditions , with χ2 test p values in the middle . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 02310 . 7554/eLife . 05920 . 024Figure 4—figure supplement 3 . Secondary analyses of effects of DYRK1A deficiency compared to harmine treatment . Volcano plots comparing p value vs fold change in gene expression in harmine-treated ( left panels ) and Dyrk1a shRNA-treated ( right panels ) cells . Previously reported signature genes for Treg cells ( top panels ) and TregHAR-Treg cells ( bottom panels ) are highlighted in red ( upregulated ) and green ( downregulated ) . Numbers on the right and left reflect genes that are up- and down-regulated , respectively , in harmine vs control conditions ( left ) or in shDyrk1a vs control conditions ( right ) , with χ2 test p values in the middle . DOI: http://dx . doi . org/10 . 7554/eLife . 05920 . 024 To further dissect this link , we examined the effect of harmine on Treg/Th17 differentiation in BACH2-deficient T cells . Mixed bone marrow chimeras allowed us to simultaneously examine the effect of harmine on wildtype and BACH2-deficient T cells . Our studies reproduced the previously reported cell-extrinsic defect in Treg ( and an even more impressive defect in Th17 ) differentiation which is due in part to the dysregulated production of cytokines like IFNγ by BACH2-deficient cells; this can be attenuated by adding neutralizing antibodies against IFNγ and IL-4 ( Figure 4D , dotted vs solid lines ) ( Roychoudhuri et al . , 2013 ) . Importantly , harmine enhances Treg differentiation and inhibits Th17 differentiation in both wildtype and BACH2-deficient cells , indicating that harmine engages BACH2-independent programs to regulate Treg/Th17 differentiation ( Figure 4D ) . Interestingly , harmine does not regulate the differentiation of BACH2-deficient cells to the same level as wildtype cells , consistent with the notion that harmine also works , at least in part , through BACH2-dependent mechanisms ( Figure 4D ) . Harmine inhibits the activity of several targets including monoamine oxidase A ( MAOA ) , CDC-like kinases ( CLKs ) and dual-specificity tyrosine-phosphorylation regulated kinases ( DYRKs ) ( Bain et al . , 2007; Aranda et al . , 2011 ) . In order to identify those relevant to Treg cell differentiation , we tested compounds that differentially inhibit each of these targets . Again , each compound was tested at multiple doses to capture any effect . Inhibition of either MAOA or CLKs using moclobemide and KH-CH19 , respectively , did not enhance Treg cell differentiation at any dose ( Figure 4E ) ( Fedorov et al . , 2011 ) . However , two other DYRK inhibitors , GSK-626616 and pro-INDY , similarly enhanced Treg and inhibited Th17 cell differentiation ( Figure 4E , F ) ( Ogawa et al . , 2010; Wippich et al . , 2013 ) . The physiological relevance of this finding is supported by FACS and Western blot studies that reproducibly showed higher levels of DYRK1A in Th17 than Treg cells ( Figure 4G ) . This difference is specific; non-Th17 cells generated in the context of pro-Th17 conditions do not exhibit elevated levels of DYRK1A ( Figure 4G ) . Furthermore , knock-down of Dyrk1a in primary CD4+ T cells resulted in increased differentiation of Treg and decreased differentiation of Th17 cells ( Figure 4H ) . Together , these data point to a central role at least for DYRK1A in regulating Treg and Th17 cell differentiation . DYRKs phosphorylate several proteins ( Aranda et al . , 2011 ) . Notable amongst these in the context of T cell biology are NFAT proteins , whose phosphorylation by DYRKs leads to their nuclear exclusion ( Gwack et al . , 2006 ) . Thus , inhibition of DYRKs would be predicted to lead to increased levels of NFAT in the nucleus . To assess this hypothesis , we performed studies using Amnis technology , which combines flow cytometry and high-resolution microscopy to allow precise quantitation of intracellular localization of individual proteins . Naïve CD4+ T cells largely retain NFAT1 in the cytoplasm ( Figure 4I , black line ) . Upon stimulation in Treglow conditions , nuclear translocation of NFAT1 is observed with an accompanying right shift of the nuclear/cytoplasmic ratio ( Figure 4I , red line ) . This nuclear translocation of NFAT1 is reproducibly enhanced approximately 40% with the addition of harmine ( Figure 4I , blue line ) . In support of these results , we independently fractionated nuclei from cells treated with each of these conditions and performed Western blot analyses . These also showed increased NFAT1 in the nuclear compartment upon stimulation in Treglow conditions , with a similar ( 40% ) additional increase when harmine was added ( Figure 4J ) . Thus , harmine enhances nuclear accumulation of NFAT1 at early time points up to 2 hr after stimulation ( Figure 4—figure supplement 1 ) . To gain further insight into the pathways engaged by harmine and DYRK1A , we activated primary CD4+ T cells under Th0 conditions followed either by transduction with Dyrk1a-specific shRNA or control shRNA followed by harmine treatment . RNAseq studies found significant similarities in the expression profiles subsequent to either Dyrk1a knockdown or harmine treatment ( Pearson correlation coefficient = 0 . 65 ) and genes that were up- or down-regulated as a result of Dyrk1a knockdown were predominantly concordantly regulated by harmine ( Figure 4—figure supplement 2 , data in GSE67961 ) . Neither Dyrk1a knockdown nor harmine treatment significantly regulated genes associated with either the canonical Treg signature or our 111-gene TregHAR-Treg signature ( Figure 4—figure supplement 3 ) . This is likely related to our observation that harmine does not promote Treg differentiation in the absence of TGFβ; the differences observed here may be upstream of more Treg-associated expression changes . A relatively small number of genes ( 150 ) were differentially regulated between Dyrk1a knockdown and harmine treatment , which may in part reflect ancillary mechanisms engaged by harmine to regulate Treg/Th17 differentiation . Overall , our data are consistent with the notion that DYRK1A inhibition represents a major mechanism by which harmine regulates Treg/Th17 differentiation .
Treg cells are an important regulator of immune homeostasis and an attractive therapeutic target because of their role in human inflammatory diseases such as IBD and T1D . Nevertheless , there remains a lack of drugs as well as druggable genes and pathways that specifically modulate Th differentiation . Although there is significant interest in manipulating Treg cells to treat IBD and T1D , the most mature efforts are found in the setting of organ transplantation where there is still significant room for improvement , ideal Treg cell subpopulation properties are still unclear and rapamycin is the most cutting-edge compound being used ( Desreumaux et al . , 2012; Long et al . , 2012; Edozie et al . , 2014 ) . To address this issue , we report a systematic , high-throughput pipeline to investigate the effects of small molecules on Th cell differentiation . These efforts enabled us to build a comprehensive profile of how compounds affect T cell viability and differentiation into both pro- and anti-inflammatory Th subsets . Drug selection could be guided by such information; for example , an ideal anti-inflammatory drug would not enhance , and would preferably inhibit , Th differentiation into pro-inflammatory lineages . Indeed , the inclusion of many FDA-approved drugs in our studies illustrates the potential of this approach to be applied to drug repurposing efforts . In this regard , our results reinforce interest in clinically used hypolipidemic statins , including lovastatin and simvastatin , as pro-Treg cell and anti-Th17 compounds and suggest that their targets , whose geranylgeranylation are inhibited , are of fundamental and clinical interest ( Kagami et al . , 2009; Kim et al . , 2010; Zhang et al . , 2008 ) . Clinical studies suggest that statins may be useful in rheumatoid arthritis , with somewhat more mixed results in systemic lupus erythematosus and multiple sclerosis ( Ulivieri and Baldari , 2014 ) . Our cytotoxicity data identify significant toxicity with many known Treg cell enhancers , including rapamycin , supporting the value of a continued search . Furthermore , our computational approaches suggest a framework to bin compounds into mechanistic classes , which holds particular relevance to future efforts to use polypharmacy to modulate the immune response by suggesting combinations that might target the maximal breadth of inflammatory pathways . These studies demonstrate the novel and simultaneous application of three key principles , namely unbiased chemical biology , maximally physiologic experimental system and selection of a phenotypic readout . The advantages of the first two have already been alluded to—studying more compounds intuitively increases the likelihood of discovering novel biology , assuming an accompanying increase in complexity of chemical structures tested , and our use of primary CD4+ T cells , as opposed to a cell line , maximizes the likelihood of our findings being physiologically relevant . Importantly , using a phenotypic primary endpoint significantly extends the scope of previous chemical biology efforts which had largely centered around finding compounds that bind known key regulators of Th differentiation , such as RORγT ( Huh et al . , 2011; Xiao et al . , 2014 ) . Such studies hold therapeutic promise and have highlighted the utility of using larger chemical libraries . However , the nature of the question fundamentally limits the potential mechanistic insight to targets of the pre-identified key regulator . In contrast , our use of a phenotypic endpoint is designed to capture any compound that affects Treg cell differentiation regardless of mechanism . In proof of this concept , we now identify 14 compounds as novel and specific enhancers of Treg cell differentiation , the largest single addition to the Treg cell biologist's chemical toolkit . We fully anticipate that subsequent studies will elucidate these compounds' mechanisms of action , leading us to a fuller understanding of the pathways that regulate Treg cell differentiation . Already , some interesting themes can be observed in this set of Treg cell enhancers . Aside from the retinoic acids and statins , a significant number of them are antimicrobial agents , in particular antifungal agents ( including sertaconazole , clotrimazole and pentamidine ) and antimalarials ( artemisinin , amodiaquine and proguanil ) . This observation suggests how such drugs might simultaneously act on both pathogen and host . It would be interesting to determine if the effect on Treg cells correlates with clinical features of such drugs . In order to develop improved diagnostic and therapeutic options , we will need a fuller understanding of the plethora of genes and pathways that regulate Treg cell differentiation and function . The majority of genetic polymorphisms that affect Treg cell function are unlikely to involve the few canonical genes that have been described . This issue will become increasingly pressing as genome sequencing technologies become more accessible and as our ability to manipulate immune modulation improves , requiring more precise selection of the right therapy for the right patient . Moreover , the identification of additional pathways will highlight new candidate therapeutic targets . It is important to note that these pathways , while ancillary to our current understanding , can be and likely are crucially important . This is underscored by our demonstration that Treghi conditions enhance Treg cell differentiation without increasing SMAD2 or SMAD3 phosphorylation above levels induced by Treglow conditions . Thus , even the best understood Treg-relevant cytokine , TGF-β1 , engages signals that remain to be fully understood . This notion is echoed by earlier discoveries that while ATRA enhances both SMAD3 signaling and Treg cell differentiation , the two are not linked as ATRA can enhance Treg cell differentiation in SMAD3-deficient mice ( Nolting et al . , 2009 ) . Although primary T cells have typically been less amenable to more traditional forward genetic approaches , we show here how chemical biology can rapidly advance our understanding of Th biology . Using a library enriched in tool compounds with known molecular activities enhanced our ability to rapidly make mechanistic insights . In this way , our discovery of harmine as a key compound of interest led us to uncover the novel activity of DYRK1A as a reciprocal regulator of Treg and Th17 cell differentiation . The mechanistic details of how DYRK1A regulates Th differentiation remain to be clearly elucidated . Moreover , DYRK1A is a member of a family of five related proteins ( Aranda et al . , 2011 ) . Whether other DYRK family members regulate Treg and Th17 cell differentiation will require additional experiments to elucidate . While the chemical inhibitors used are significantly more specific for DYRKs as compared to other families of kinases , their ability to distinguish between individual DYRKs is more limited ( Bain et al . , 2007; Ogawa et al . , 2010; Wippich et al . , 2013 ) . The enhanced NFAT1 nuclear translocation we find associated with DYRK inhibition by harmine treatment in primary CD4+ T cells is consistent with previous studies in cell lines showing that DYRKs inhibit NFAT signaling ( Gwack et al . , 2006 ) . On one hand , studies showing progressively severe defects in pTreg cell generation corresponding with increasing loss of NFAT family members raise the possibility that harmine-enhanced NFAT signaling may act in the opposite manner to promote Treg cell differentiation ( Vaeth et al . , 2012 ) . Increased NFAT nuclear translocation might enhance binding to described FOXP3 enhancer elements , thus promoting FOXP3 expression , or even to FOXP3 itself , boosting transcription of FOXP3 targets ( Ruan et al . , 2009; Wu et al . , 2006 ) . However , in counterpoint to this simple association of increased NFAT with increased Treg cell differentiation , decreased NFAT signaling has also been reported to impair Th1 and Th17 cell differentiation ( Ghosh et al . , 2010; Hermann-Kleiter and Baier , 2010 ) . Extrapolating these latter results , one might expect the harmine-enhanced NFAT signaling to concomitantly promote Th1 and Th17 cell differentiation , which we clearly do not observe . One way by which these conflicting predictions might be resolved could involve harmine modulating NFAT activity in a more complex manner , for example involving dynamics of nuclear retention , with a Treg cell-specific net effect . Alternatively , these results in conjunction with our finding that harmine's effect on nuclear localization of NFAT diminishes at later time points raise the possibility that some other target of DYRKs is more relevant in the context of Treg and Th17 cell differentiation . Interestingly , human diseases secondary to perturbed DYRK function , on closer inspection , also exhibit immunological aspects , suggesting that the relationship between DYRKs and Th differentiation is physiologically germane . In this regard , the observation that DYRK inhibition promotes Treg cell differentiation leads to the converse prediction that increased DYRK activity would inhibit Treg cells . Down syndrome is characterized by trisomy of chromosome 21; specifically , the resulting increase in DYRK1A copy number and activity is thought to be a key driver of pathology ( Lepagnol-Bestel et al . , 2009 ) . Notably , patients with Down syndrome have hypofunctional Treg cells and are at increased risk for autoimmune disease ( Pellegrini et al . , 2012 ) . Similarly , gain-of-function mutations in DYRK1B were recently implicated in metabolic syndrome ( Keramati et al . , 2014 ) . Decreased adipose tissue-associated Treg cells contribute to the inflammation that is a central player in obesity-induced metabolic syndrome ( Odegaard and Chawla , 2013 ) . It is tempting to speculate that our findings provide a unifying hypothesis that helps account for these disparate observations , with increased DYRK activity in these patients leading to decreased Treg cell differentiation via effects opposite to harmine's . In addition to extending our understanding of the biology of Treg cell differentiation , our demonstration that harmine-enhanced Treg cells exhibit full functionality in multiple animal models of inflammation differing in genetic background , target organ system and antigen specificity raise interest in the possibility of manipulating this axis therapeutically . This notion is reinforced by our finding that harmine itself , directly administered , can attenuate inflammation . Furthermore , harmine similarly enhances human Treg differentiation , supporting the likely physiologic relevance of the pathways it engages . Taken together , we propose that DYRKs represent a novel , druggable target of particular relevance to tolerance and inflammation . In summary , these results illustrate how unbiased chemical biology approaches can identify novel chemical modulators of Treg cell differentiation , point to interesting mechanistic hypotheses and spark new translational efforts .
Balb/c , C57Bl/6 , Cd45 . 1+/+ , Rag1−/− , Foxp3IRES-GFP , Il17IRES-GFP , NOD-scid and NOD-BDC2 . 5 mice were obtained from Jackson Labs . NOD-BDC2 . 5 . Foxp3IRES-GFP mice were obtained from the JDRF Transgenic Core ( Harvard Medical School , Boston , MA ) . Bach2-knockout mice have been previously described ( Muto et al . , 2004 ) . Mixed chimeras were generated by injecting CD90 . 1+CD45 . 1−Bach2−/− and CD90 . 1−CD45 . 1+Bach2+/+ bone marrow into C57Bl/6 hosts . Antibodies and cytokines used are described in Supplementary file 1A . Chemical compounds were sourced as in Supplementary file 1B . Pro-INDY and GSK-626616 were synthesized as previously described ( Corona et al . , 2010; Ogawa et al . , 2010 ) . CD4+CD62L+ naïve T cells were isolated using CD4 negative enrichment kits ( Stemcell Technologies , Vancouver , Canada ) and CD62L microbeads ( Miltenyi Biotec , San Diego , CA ) and confirmed to be >95% pure by flow cytometry . These were cultured on 96-well plates pre-coated with anti-CD3 and anti-CD28 under conditions outlined in Supplementary file 2 . In particular , the addition of harmine to Treglow conditions is abbreviated as TregHAR . Compounds were pinned using a CyBIO CyBi Well Vario ( 96-well pintool ) ( Cybio , Jena , Germany ) . Treg and Th1 cultures were fed with equal volume of IL-2-supplemented media ( 10 ng/ml ) and retreated with compound at day 2 , split 1:2 into IL-2-supplemented media at day 3 and analyzed at day 4 . Th17 and Th0 cultures were treated similarly except no IL-2 was supplemented . Cell proliferation was monitored using CFSE ( Life Technologies , Carlsbad , CA ) per manufacturer's instructions . Human peripheral mononuclear cells were separated using Ficoll–Paque ( GE Healthcare , Little Chalfont , United Kingdom ) and CD4+CD45RA+ naïve T cells isolated using negative enrichment kits ( Stemcell Technologies , Vancouver , Canada ) per manufacturer's instructions and confirmed to be >90% pure by flow cytometry . These were cultured on 96-well plates pre-coated with anti-CD3 and anti-CD28 under conditions outlined in Supplementary file 2 . Cultures were fed with equal volume of IL-2-supplemented media ( 10 ng/ml ) at day 4 , split 1:2 into IL-2-supplemented media at day 6 and analyzed at day 8 . 5 hr prior to analysis , Th1 and Th17 cultures were restimulated with PMA and ionomycin ( 50 and 500 ng/ml respectively , Sigma Aldrich , St . Louis , MO ) in the presence of Golgistop ( BD Biosciences , San Jose , CA ) . Cells were typically stained with LIVE/DEAD and anti-CD4-FITC before being fixed and permeabilized using either Foxp3 fixation/permeabilization buffers ( eBioscience , San Diego , CA ) or Phosflow Fix/Perm buffers ( BD Biosciences , San Jose , CA ) as indicated . Intracellular staining was performed per manufacturer's instructions . Counting beads ( 10 μm , Spherotech , Lake Forest , IL ) were added at 5000 per well . Acquisition was performed on a FACSVerse ( BD Biosciences , San Jose , CA ) and analyzed using FlowJo software ( Treestar , Ashland , OR ) . Fractional enhancement was determined by increase in percentage lineage-committed cells , relative to maximal cytokine-driven enhancement . Fractional inhibition was calculated relative to percentage lineage-committed cells treated with DMSO . STAT3 phosphorylation was quantitated as previously described ( Chaudhry et al . , 2011 ) . Cell sorting was performed on a FACSVantage ( BD Biosciences , San Jose , CA ) . Each experimental 96-well plate included at least eight wells each of positive and negative controls . Each experimental batch included an additional plate of 48 positive and 48 negative controls and was processed separately . For quality control purposes , data from each experimental plate were first median-centered using data from all wells except positive controls . Median-centered data from all plates were pooled with batch-level negative controls to estimate batch-wide standard deviation . This step was repeated with the positive controls . Each plate was individually assessed if its internal controls met specific standards . Plates where ≤ 75% of controls scored within the expected range or exhibited suboptimal dynamic range were excluded and retested subsequently . The remaining plates were subjected to a similar strategy of pooling median-centered data to estimate robust standard deviation . This measure was first used to select a negative control reference from the pool of in-plate negative controls and compound-treated wells . Next , data from each screening plate were transformed into Z-scores using the mean of select negative control wells and robust standard deviation . Z-normalized data from all screening plates were pooled per experimental batch . Generalized linear regression models were fitted to positive and negative controls using glmfit function in Matlab ( Mathworks , Natick , MA ) . Compounds that performed at ≥30% of the observed levels for positive control ( based on artemisinin's enhancement ) were selected for secondary screening . Dose response curves for fractional enhancement of Treg cell differentiation and culture cellularity were analyzed in Matlab to identify EC50 and LD50 doses , at which 50% Treg enhancement and cytotoxicity are observed , respectively . Each compound was profiled across eight doses selected to sufficiently cover response dynamics . Dose response curves were fitted with either a single sigmoid or a double sigmoid function , depending on whether the response was asymptotic or impulse-like . An impulse function has the form:f=1r1s ( d:d1 , rl , rp , α1 ) ×s ( d:d2 , rh , rp , α2 ) , wheres ( d:dm , ri , rf , α ) = ri+ ( rf−ri ) 11+e−4α ( d−dm ) , is a sigmoid function with a response that ranges from ri to rf with mid-point at dose dm and a slope of α*sign ( rf − ri ) at dose dm . The parameters of this model describe the dose of response onset ( d1 ) , dose of response offset ( d2 ) , initial response at lowest dose ( rl ) , peak response ( rp ) , final response at highest dose ( rh ) , and two slope parameters to model the rate of response onset ( α1 ) and offset ( α2 ) . A single-sigmoid function uses only four parameters ( d , r1 , r2 , α ) . All models were fitted to data using fmincon function in Matlab . Fitted models were reverse-queried to estimate the dose at which 50% of the peak response parameter was observed . Phenotypic data ( LD50/EC50 ratio , maximal Treg enhancement and Th1 and Th17 inhibition ) for all 21 Treg enhancers were combined to form a feature matrix . The data were standardized and pairwise similarity between compounds was computed using Pearson correlation with complete linkage in GENE-E ( http://www . broadinstitute . org/cancer/software/GENE-E/ ) . Transcriptomic profiles examining effects of compounds in three cell lines ( MCF7 , PC3 and HL60 ) , available for 19 of 21 Treg enhancers , were downloaded from the Connectivity Map ( CMAP ) database and analyzed in Matlab ( Lamb et al . , 2006 ) . Expression data from replicate experiments were averaged for each cell line; data from separate doses were not merged . A gene expression amplitude table of 22 , 280 genes and 62 CMAP instances ( reduced from 151 ) was subject to principal components analysis for dimensionality reduction . 43 principal components explained up to 90% variance in expression data , using genes as features . Normalized PC scores for the first 43 components and 62 compound instances were analyzed for pairwise similarity using Euclidean distance with complete linkage in GENE-E . RNA was isolated using RNeasy kits ( Qiagen , Valencia , CA ) and cDNA prepared using iScript cDNA synthesis kit ( Bio-Rad , Hercules , CA ) per manufacturer's instructions . Real-time PCR was performed using iTaq SYBR Green ( Bio-Rad , Hercules , CA ) on a C1000 thermal cycler ( Bio-Rad , Hercules , CA ) equipped with a CFX384 Real Time System ( Bio-Rad , Hercules , CA ) . Cycling conditions were 95°C for 3 min , followed by 40 cycles of 94°C for 15 s , 59°C for 45 s , and 72°C for 6 s . Primers used were Il17a: TTTAACTCCCTTGGCGCAAAA and CTTTCCCTCCGCATTGACAC; Il22: CATGCAGGAGGTGGTACCTT and CAGACGCAAGCATTTCTCAG; Batf: GACACAGAAAGCCGACACC and AGCACAGGGGCTCGTG; Pou2af1: CACCAAGGCCATACCAGGG and GAAGCAGAAACCTCCATGTCA; Mina: TTTGGGTCCTTAGTAGGCTCG and CCGATCCGGTCCTCAGATT; Slc14a1: GGCTCTGGGGTTTCAACA and GCCATCAGGTGTGCCATAC; Trib1: CAGATTGTTTCCGCCGTCG and ACCCTTAATGATGTGAGTATCTTCC; Tfrc: CCGCTCGTGGAGACTACTT and CCCAGAAGATATGTCGGAAAGG; Mybl2: CAAGAATGCCCTGGAGAAGTAC and GCTTTCTCTTCTGCTTCTCGG; Mcm2: GCCCATCATTTCCCGCTTTGA and CCCTTCATCCTTCTTGTTACTGG; Dab2ip: CCATCCTCAGTGCCAAGAC and GGTCCACCTCTGACATCATCA; Snx6: GTTCTACAGGCTGAAACTTCCC and TAAAACCGCAAGGCAGTTCTG; Exoc2: GACAGCGTCACTGAAGAGG and GAGTTTCCAGAAGTTAGGCAGC; Slc9a8: CTGGCAGAAGGAATCTCACTC and CAGTTCGGAGAGTCTGCTG; Rora: ATGCCACCTACTCCTGTCC and GACATCCGACCAAACTTGACAG; Akt3: GGCACACCAGAGTACCTG and GCATCTGAAGAGAGTGTTCGG; Lmna: TGTGGCGGTAGAGGAAGT and GGAAGCGATAGGTCATCAAAGG; St3gal3: GACTGCCATCTTCCCCAG and CAAAAGGTGGCACAAACTCC; Ccl20: TACTGCTGGCTCACCTCTG and CCATCTGTCTTGTGAAACCCAC; Aqp3: TTGGCATCTTGGTGGCTG and GCTCATTGTTGGCAAAGGC; Ypel5: CCAATCGCTCAGAACTCATCTC and ATAACGCTGGCTGTCTTCAG . RNA concentration and purity were measured using a NanoDrop spectrophotometer ( Thermo Scientific , Waltham , MA ) . For microarray studies , RNA was amplified and labeled using the Illumina TotalPrep RNA Amplification kit ( Ambion , Grand Island , NY ) per manufacturer's instructions . Labeled cRNA was then hybridized to the Illumina Mouse WG-6 v2 chip . The Illumina microarray was performed by Partners HealthCare Center for Personalized Genetic Medicine ( PCPGM ) Translational Genomics Core ( Boston , MA ) . BeadChips were scanned per protocol ( Illumina Whole Genome Gene Expression for BeadStation Manual v3 . 2 , Revision A ) using scanning software BeadScan 3 . 5 . 31 . The GenomeStudio Data Analysis Software ( Illumina , San Diego , CA ) was used for data collection . The final report comprising the full dataset was initially processed using the Bioconductor package Lumi by employing a background correction estimate . Subsequently , signal intensities were VST-transformed ( variance-stabilizing transformation ) and RSN-normalized ( robust spline normalization ) using the Lumi package in R . Post-processing and statistical analysis of microarray data was carried out in Matlab . Normalized log2 data was first checked for correlation between replicates ( >0 . 98 on average ) and probes without gene assignments removed . For RNAseq studies , polyA mRNA was isolated using the Dynabeads mRNA DIRECT kit ( Life Technologies , Grand Island , NY ) and cDNA generated using poly-dT priming and Maxima reverse transcriptase ( Life Technologies , Grand Island , NY ) per manufacturer's instructions . After RNase A ( Life Technologies , Grand Island , NY ) treatment and SPRI bead ( Beckman Coulter , Pasadena , CA ) cleanup , second strand cDNA was synthesized using NEBNext mRNA Second Strand Synthesis Module ( New England Biolabs , Ipswich , MA ) followed by SPRI cleanup per manufacturer's instructions . Samples were tagmented using Nextera DNA sample preparation kits ( Illumina , San Diego , CA ) and read on a MiSeq ( Illumina , San Diego , CA ) per manufacturer's instructions . RNAseq reads were aligned using Tophat ( mm10 ) and RSEM-based quantification using known transcripts ( mm10 ) performed , followed by further processing using the Bioconductor package DESeq in R . The data was normalized by library size followed by variance estimation and evaluation of pairwise differential expression based on a negative binomial distribution model . Post-processing and statistical analysis was carried out in Matlab . The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE67961 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE67961 ) ( Edgar et al . , 2002 ) . Gene signatures overlaid were previously reported and χ2 analyses were performed as previously reported ( Hill et al . , 2007; Zheng et al . , 2009; Darce et al . , 2012; Burzyn et al . , 2013; Roychoudhuri et al . , 2013; Joller et al . , 2014 ) . To identify the TregHAR signature , we compared the 2269 genes ( Student's t-test , nominal p ≤ 0 . 05 ) differentially regulated between sorted Treghi and TregHAR cells with the 3492 differentially expressed genes ( Student's t-test , nominal p ≤ 0 . 05 ) between human Treg and Th17 cells ( Zhang et al . , 2013 ) . Gene expression from each experiment was independently scaled on a per-gene basis using min–max normalization , analyzed in GENE-E and marker selection used to identify the 111-gene signature with signal-to-noise ratio > 1 . Transcription factor binding site prediction and enrichment was performed using the Molecular Signatures Database ( MSigDB ) ( Subramanian et al . , 2005 ) . To assess enrichment in disease-associated genes , we downloaded the 1437 genes previously reported to lie within IBD-associated loci ( Jostins et al . , 2012 ) . We used permutation testing as previously reported to assess fold-enrichment and significance of overlap between the TregHAR signature set and the IBD-associated gene set ( Okada et al . , 2014 ) . Briefly , for each of 10 , 000 iterations , we randomly selected a set of 1437 genes from the entire genome , assessing for overlap with the TregHAR signature gene set . This distribution was used as the background to assess fold enrichment . Significance of the enrichment was evaluated using one-sided permutation tests . Lentiviral shRNA constructs were designed based on established TRC guidelines and cloned at the Genetic Perturbation Platform ( Broad Institute , Cambridge , MA ) . Sequences used to knock-down Dyrk1a expression were TATGAAATCGACTCCTTAATA , TTTGAAATGCTGTCCTATAAT , GAGGTCGATCAGATGAATAAA , GAACCCGTAAACTTCATAATA and ACTCGGATTCAACCTTATTAT . Non-targeting sequences used were AGCAGCTGTTCGAGGATAATA , TTTGCACAAGAACAGAATAAT and ACAGATGCCAATGGGAATATT . Naïve CD4+ T cells were isolated and stimulated as described above in Th0 conditions . Cells were infected with 15 μl concentrated lentiviral supernatant at day 1 , fed with media + polarizing cytokines at day 2 ( Treglow or Th17hi conditions as described above ) , split 1:2 at day 4 and analyzed at day 5 . For RNAseq studies , cells were fed with media ± harmine at day 2 without polarizing cytokines and harvested at day 4 . Sorted CD4+ cells were stimulated as indicated , fixed in 3% paraformaldehyde ( Santa Cruz , Dallas , TX ) , permeabilized in PBS + 2% fetal calf serum + 0 . 1% Triton X-100 ( Sigma Aldrich , St . Louis , MO ) and stained with anti-CD4-FITC , rabbit anti-NFAT1 , APC-donkey anti-rabbit and DAPI . Acquisition was performed on an Amnis ImageStream MkII imaging flow cytometer ( EMD Millipore , Billerica , MA ) , combining high-resolution microscopy and flow cytometry to quantitate nuclear/cytoplasmic signal distribution . 10 , 000 event data files were acquired per sample using a 60× objective running a 6-μm core diameter at 44 mm/s . Brightfield , side scatter and fluorescent images were collected using three excitation lasers ( 405 nm , 1 mW Ch07; 642 nm , 100 mW Ch11; 785 nm , 1 . 09 mW for SSC Ch06 ) . Single color controls were acquired and spectral compensation performed post-acquisition . From single , nucleated cells in focus , the DAPI+CD4+NFAT+ population was gated upon . To determine NFAT1 signal distribution , a nuclear morphology mask , which includes all pixels within the outermost image contour , was created from the DAPI+ image ( Channel 07 ) . Nuclear/whole cell NFAT1 ratios were determined by dividing the NFAT1 intensity within the defined nuclear morphology mask by the NFAT1 intensity over the entire cell ( default MC Ch11 mask ) . Nuclear/cytoplasmic histogram values were compared using a similarity score ( a log-transformed Pearson's correlation coefficient that measures the degree to which two images are linearly correlated in the nuclear masked region ) . Cell extracts for Western analyses were prepared using TNN lysis buffer , pH 8 ( 100 mM TRIS-HCl , 100 mM NaCl , 1% NP-40 , 1 mM DTT , 10 mM NaF ) supplemented with protease inhibitor ( Roche , Indianapolis , IN ) and phosphatase inhibitors ( Thermo Scientific , Waltham , MA ) , separated by SDS-PAGE ( Bio-Rad , 456-9035 ) and transferred onto PVDF membrane ( Immobilon-P , Millipore , IPVH20200 , Billerica , MA ) Approximately 106 cells were processed per lane . Antibodies used are described in Supplementary file 1A . Bands were visualized using Western Lightning Plus-ECL ( Perkin Elmer , Waltham , MA ) and/or SuperSignal West Femto substrate ( Thermo Scientific , Waltham , MA ) per manufacturer's instructions . Nuclear isolation was performed using a Nuclei Isolation Kit ( Sigma Aldrich , St . Louis , MO ) per manufacturer's protocol after cells were lysed in RIPA buffer ( 150 mM NaCl , 1% Triton X-100 , 0 . 5% sodium deoxycholate , 0 . 1% SDS , 50 mM TRIS-HCl pH7 . 8 ) supplemented with DTT , protease and phosphatase inhibitors . Band intensity was quantitated using ImageJ ( Schneider et al . , 2012 ) . This was performed as previously described ( Collison and Vignali , 2011 ) . Briefly , sorted CD45 . 1+CD4+CD62L+ Tresponder cells were labeled with CFSE ( Invitrogen , Grand Island , NY ) per manufacturer's protocol , plated at 5 x 104 cells per well in 96-well U bottom plates and co-cultured with sorted CD45 . 2+Foxp3IRES-GFP Treg cells at ratios indicated in the presence of anti-CD3/28 beads ( Dynabead , Grand Island , NY ) and analyzed by flow cytometry 3 days later . As previously described , 5 × 104 CD4+CD62L+ T cells isolated from NOD-BDC2 . 5 mice were administered intravenously to NOD-scid mice with or without 1 × 105 sorted Treg cells generated from NOD-BDC2 . 5 . Foxp3 IRES-GFP mice ( Herman et al . , 2004; Tarbell et al . , 2004 ) . Blood glucose levels were measured with a handheld Contour glucometer ( Bayer , Leverkusen , Germany ) at days 3 , 6 , 8 and every day thereafter . Diabetes was diagnosed when blood sugar was over 250 mg/dl for 2 consecutive days . As previously described , 5 × 105 CD4+CD62L+ T cells were injected into the intraperitoneal cavity of Rag1−/− mice . 10 days later , mice were injected with either PBS or 1 . 5 × 105 sorted Treg cells generated from Foxp3 IRES-GFP mice ( Smith et al . , 2013 ) . Mice were monitored weekly for weight loss and morbidity for 6–8 weeks per protocol . At 8 weeks , mice were euthanized and proximal , middle and distal colon analyzed histologically by blinded observers as previously described ( De Jong et al . , 2000 ) . Allergic airway inflammation was induced in mice as previously described ( Grainger et al . , 2010 ) . In brief , C57Bl/6 mice were injected intraperitoneally with 10 μg of ovalbumin ( Sigma–Aldrich , St . Louis , MO ) and 1 mg of aluminum hydroxide ( Sigma–Aldrich , St . Louis , MO ) suspended in 0 . 5 ml of PBS on days 0 and 10 . Sorted Treg cells generated from Foxp3IRES-GFP mice using Treghi/TregHAR conditions were transferred by retroorbital injection on days 16 and 19 . Mice were challenged intratracheally with 10 μg OVA in PBS on days 17 and 20 and sacrificed 20–24 hr after the last challenge . The trachea was exposed and cannulated with polyethylene tubing followed by bronchoalveolar lavage ( BAL ) with PBS + 0 . 6 mM EDTA . Lavage fluid was centrifuged , and pelleted cells counted and analyzed . The differential cell count was performed as previously described; cells were stained with Diff-Quick ( Dade Behring , Newark , DE ) and macrophages , neutrophils , eosinophils , and lymphocytes on cytocentrifuge preparations enumerated ( Grainger et al . , 2010 ) . At least 200 cells were counted on each slide . C57Bl/6 mice were injected intraperitoneally with 100 μg ovalbumin and 1 mg aluminum hydroxide on day 0 , challenged with 10 μg ovalbumin intratracheally on days 14 , 17 and 21 , and tissues were harvested for analysis 24 hr after the last challenge as previously described ( Curotto de Lafaille et al . , 2008 ) . Tolerance was induced by intranasal sensitization with 100 μg ovalbumin prior to immunization on days −7 , −6 , and −5 ( Curotto de Lafaille et al . , 2008 ) . Mice were treated daily with 1 mg harmine HCl ( Santa Cruz Biotechnology , Dallas , TX ) dissolved in water intranasally from days −8 through −3 as indicated . Tissues were preserved in 10% formalin . Paraffin embedding , sectioning and staining with either hematoxylin and eosin or Periodic acid-Schiff/Alcian Blue were performed by the Histopathology Research Core ( Massachusetts General Hospital , Boston , MA ) Mice were treated with 1 mg harmine HCl intranasally for 5 days , and thoracic lymph nodes harvested on day 6 ( for T cell studies ) or day 2 ( for dendritic cell studies ) . To liberate dendritic cells , lymph nodes were mechanically disrupted and incubated in HBSS ( GE Healthcare , Little Chalfont , United Kingdom ) containing 2 . 5 mg/ml collagenase D ( Roche Diagnostics , Indianapolis , IN ) at 37°C for 30 min . Digestion was neutralized by adding EDTA to 20 mM . Statistical measures , including mean values , standard deviations , Student's t-tests , Mantel–Cox tests , Mann–Whitney tests and one-way ANOVA tests , were performed using Graphpad Prism software and Matlab . Where appropriate , unless otherwise stated , graphs display mean ± standard deviation . All experiments were performed with the approval of the IACUC of Massachusetts General Hospital ( Boston , MA ) . | Inflammation is used by the immune system to protect and repair tissues after an injury or infection . However , if inflammation is too strong , or goes on for too long , it can damage tissues . This is seen in autoimmune diseases such as inflammatory bowel disease and type 1 diabetes . Therefore , precise regulation of the inflammatory response is essential for maintaining human health . White blood cells known as T cells are central regulators of tissue inflammation . To achieve this goal , they develop into subtypes with specialized roles . For example , some T helper cells release chemical signals that trigger inflammation and other immune responses . Regulatory T ( Treg ) cells then shut down these immune responses once they are no longer needed . Many autoimmune and other inflammatory diseases are thought to arise—at least partially—because Treg cells fail to stop the inflammatory response . Boosting the number or the activity of Treg cells could therefore help to treat these diseases . However , technical difficulties have made it difficult to investigate the genes and molecular pathways that control how this subtype of white blood cells develops . Khor et al . thought that discovering new chemicals that increase the number of Treg cells without harming them could help to identify the pathways that control their development . Khor et al . screened over 3000 chemicals , many of which are drugs currently approved for use in humans , for their effect on immature T cells that were taken from mice and grown in the laboratory . This ‘unbiased chemical biology’ approach identified several chemicals that both encouraged the T cells to develop into Treg cells and reduced the numbers that became inflammation-promoting T helper cells . Khor et al . then focused on one of these chemicals , called harmine . Tests in mice showed that harmine reduces the extent of experimentally induced inflammatory reactions . Treg cells generated by treating immature T cells with harmine had the same effect . Further experiments showed that harmine exerts these effects , at least in part , by inhibiting the activity of a protein called DYRK1A . When DYRK1A was removed from maturing mouse T cells grown in the laboratory , the T cells tended to develop into anti-inflammatory Treg cells . These findings therefore identify DYRK1A as part of a pathway that suppresses the development of Treg cells . It remains to be discovered how it does this , and whether other DYRK protein family members have similar roles . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"immunology",
"and",
"inflammation"
] | 2015 | The kinase DYRK1A reciprocally regulates the differentiation of Th17 and regulatory T cells |
Non-clustered δ-protocadherins are homophilic cell adhesion molecules essential for the development of the vertebrate nervous system , as several are closely linked to neurodevelopmental disorders . Mutations in protocadherin-19 ( PCDH19 ) result in a female-limited , infant-onset form of epilepsy ( PCDH19-FE ) . Over 100 mutations in PCDH19 have been identified in patients with PCDH19-FE , about half of which are missense mutations in the adhesive extracellular domain . Neither the mechanism of homophilic adhesion by PCDH19 , nor the biochemical effects of missense mutations are understood . Here we present a crystallographic structure of the minimal adhesive fragment of the zebrafish Pcdh19 extracellular domain . This structure reveals the adhesive interface for Pcdh19 , which is broadly relevant to both non-clustered δ and clustered protocadherin subfamilies . In addition , we show that several PCDH19-FE missense mutations localize to the adhesive interface and abolish Pcdh19 adhesion in in vitro assays , thus revealing the biochemical basis of their pathogenic effects during brain development .
Nervous system function is critically dependent on the underlying neural architecture , including patterns of neuronal connectivity . Cell-cell recognition by cell surface receptors is central to establishing these functional neural circuits during development ( Kiecker and Lumsden , 2005; Steinberg , 2007; Zipursky and Sanes , 2010 ) . The cadherin superfamily is a large and diverse family of cell adhesion molecules that are strongly expressed in the developing nervous system ( Hirano and Takeichi , 2012; Suzuki , 1996; Frank and Kemler , 2002; Shapiro et al . , 2007; Gumbiner , 2005;Chen and Maniatis , 2013 ) . The differential expression of classical cadherins and protocadherins , the largest groups within the cadherin superfamily , suggests that they play important roles in the development of neural circuitry ( Weiner and Jontes , 2013; Hirano and Takeichi , 2012 ) , an idea supported by their involvement in a range of neurodevelopmental disorders ( Redies et al . , 2012; Hirabayashi and Yagi , 2014 ) . In particular , the non-clustered δ-protocadherins have been linked to autism spectrum disorders , intellectual disability , congenital microcephaly and epilepsy . Protocadherin-19 ( PCDH19 ) is a member of the non-clustered δ2-protocadherin subfamily ( Wolverton and Lalande , 2001; Vanhalst et al . , 2005; Gaitan and Bouchard , 2006; Emond et al . , 2009; Liu et al . , 2010 ) that is located on the X-chromosome . Mutations in PCDH19 cause an X-linked , female-limited form of infant-onset epilepsy ( PCDH19 female epilepsy , PCDH19-FE; OMIM 300088 ) that is associated with intellectual disability , as well as compulsive or aggressive behavior and autistic features ( Dibbens et al . , 2008; Scheffer et al . , 2008; Depienne and LeGuern , 2012; van Harssel et al . , 2013; Leonardi et al . , 2014; Thiffault et al . , 2016; Terracciano et al . , 2016; Walters et al . , 2014 ) . To date , well over 100 distinct mutations in PCDH19 have been identified in epilepsy patients , making it the second most clinically relevant gene in epilepsy . Approximately half of these mutations are missense mutations distributed throughout the extracellular domain of the PCDH19 protein . Despite the clear importance of PCDH19 and other non-clustered δ-protocadherins to neural development , their specific roles are only beginning to be revealed . For example , Pcdh7 , Pcdh17 and Pcdh18b are involved in axon outgrowth or arborization ( Piper et al . , 2008; Hayashi et al . , 2014; Biswas et al . , 2014 ) , while several δ-protocadherins , including Pcdh19 , regulate cell motility during early development ( Yamamoto et al . , 1998; Aamar and Dawid , 2008; Biswas et al . , 2010; Emond et al . , 2009 ) . In zebrafish , pcdh19 , regulates the formation of neuronal columns in the optic tectum , and loss of pcdh19 degrades visually-guided behaviors ( Cooper et al . , 2015 ) . However , it is not known how mutations in PCDH19 lead to PCDH19-FE . Cadherins typically mediate adhesion using their extracellular domains , which are made of two or more consecutive extracellular cadherin ( EC ) repeats ( Takeichi , 1990; Brasch et al . , 2012 ) . The adhesion mechanism used by classical cadherins is well known and involves a tip-to-tip interaction that is stabilized by the reciprocal exchange of tryptophan residues at the N-terminal EC1 repeat most distant from the membrane ( Overduin et al . , 1995; Shapiro et al . , 1995; Nagar et al . , 1996; Boggon et al . , 2002; Patel et al . , 2006; Zhang et al . , 2009; Sivasankar et al . , 2009; Harrison et al . , 2010; Ciatto et al . , 2010; Leckband and Sivasankar , 2012 ) . However , PCDH19 along with the rest of the non-classical cadherins lack these critical tryptophan residues and must mediate adhesion by an alternative mechanism ( Emond et al . , 2011; Sotomayor et al . , 2014; Biswas et al . , 2010 ) . In the case of the non-classical protocadherin-15 and cadherin-23 proteins , an adhesive interface is formed by overlapping , antiparallel interactions of their EC1 and EC2 tips ( Elledge et al . , 2010; Sotomayor et al . , 2010; 2012; Geng et al . , 2013 ) . For clustered protocadherins , recent binding assays and structures suggest that adhesion is mediated by an antiparallel interaction of fully overlapping EC1 to EC4 domains ( Rubinstein et al . , 2015; Nicoludis et al . , 2015; Goodman et al . , 2016 ) . Yet how non-clustered δ-protocadherins and PCDH19 form adhesive bonds and how these bonds are altered by disease-causing mutations is unknown . Here we present crystals structures of the highly homologous zebrafish Protocadherin-19 ( Pcdh19 ) encompassing repeats EC1-4 and EC3-4 . The structures allow us to map >70% of the disease-causing missense mutations and provide a structural framework to interpret their functional impact . In addition , the structures suggest two possible homophilic adhesive interfaces , and complementary binding assays validate one of them , which is affected by multiple PCDH19-FE mutations . This interface involves fully overlapping EC1 to EC4 domains and likely represents a general interaction mechanism for the non-clustered δ-protocadherins .
There are 51 PCDH19-FE missense mutations ( out of 70 ) that can be mapped to 43 locations in the Pcdh19 EC1-4 structure ( Figure 1B and Figure 1—figure supplement 2 ) . These mutations can be classified in three groups . The first group ( 18 mutations at 14 locations ) corresponds to residues whose side chains are pointing toward the hydrophobic core of an EC repeat ( Figure 1B , C ) . The second group involves residues whose side chains are at the surface of the protein ( 10 mutations at 10 sites; Figure 1B , D ) . The last group includes residues at calcium-binding motifs , with 19 locations affected by 23 different mutations ( Figure 1B , E–G ) . Mutations in each group are predicted to have different effects on the protein’s structure ( Figure 1—figure supplement 3 ) . PCDH19-FE mutations altering residues in the first group may often cause protein misfolding or structural instability . For instance , mutations L81R and I115K ( corresponding to L58 and I92 in the crystal structure ) would result in impossible conformations in which a positively charged residue side chain is pointing toward the hydrophobic core of EC1 ( Figure 1C ) . Thus , these mutants are unlikely to fold properly . Mutation L25P ( L4 ) will interfere with hydrogen bonding and secondary structure formation , while V72G ( V50 ) is subtler , as it replaces a rather large hydrophobic residue with a different and smaller side chain that may only affect the packing of the EC1 hydrophobic core . The mutation A153T ( A130 ) in EC2 , in which a small hydrophobic residue is replaced by a larger hydrophilic threonine , may result in structural instability as well . A similar analysis can be done for all 18 mutations in this group ( Figure 1—figure supplement 3 ) . Protein misfolding and structural instability caused by these mutations are likely to inhibit PCDH19 adhesive function , either directly , allosterically , or by altering the strength of cell-cell adhesion due to a reduced number of functional molecules on the cell surface . The effect of ten PCDH19-FE mutations on residues with side chains at the protein surface ( second group ) is less clear . Two of them ( S276P and L433P ) may affect packing and folding , as these mutations to proline are predicted to prevent formation of hydrogen bonds important for β strand formation and loop structure . Six of them are involved in putative homophilic interfaces , and their effect on binding is discussed below . The V191L mutation site is not directly involved in homophilic binding , but it is near residues that are , and may allosterically alter binding . Alternatively , this mutation may alter interactions with N-cadherin ( Emond et al . , 2011 ) or other PCDH19 molecular partners yet to be determined . The last mutation , D417H , is not involved in any known interface , but this epilepsy patient has a pair of mutations in PCDH19 ( D417H and D596Y ) . It is unclear whether both mutations contribute to the epileptic syndrome ( Figure 1—source data 1 ) ( Higurashi et al . , 2015; Hoshina et al . , 2015 ) . The third group of mutations involves residues that are at one of the canonical calcium-binding motifs between EC repeats ( XEXBASE and DRE from the first EC repeat , DXNDN from the linker , and DXD and XDXTOP from the second repeat ) . Two of these PCDH19-FE mutations involve charge reversal for a calcium-coordinating residue ( E31K at XEX and E307K at DYE ) , and may result in impaired folding and impaired calcium binding . Twelve PCDH19-FE mutations in this group replace a charged , calcium-coordinating residue by a neutral residue ( D90V at DRE , D121N at DXNDN , D157N at DXD , E199Q at DRE , D230N at DXNDN , E249G at XEX , D264H at DXD , D341G at DXNDN , D375Y at DXD , D377N and D377H at DXD , and E414Q at DRE ) . Some of these mutations only affect charge , but not the size of the side chain ( D to N and E to Q ) , and may decrease the affinity for calcium . Others involve more drastic side-chain size changes ( D to Y or G ) and will not only impair calcium binding , but might also induce protein instability . In addition , three mutations alter the size , but not the charge of a coordinating residue ( E249D at XEX , D341E at DXNDN , and D377E DXD ) , indicating that even subtle perturbations at the calcium-binding linkers might result in impaired function . Three more PCDH19-FE mutations involve substituting a coordinating asparagine residue by a serine ( N232S and N340S at DXNDN , and N234S at DXNDN ) , with one of these mutations present in over twenty unrelated individuals ( N340S ) . Similarly , the mutation NP342-343KT at DXNDN involves a coordinating asparagine residue , but it is mutated to lysine and accompanied by a proline to threonine mutation . In addition , one mutation involves the non-calcium binding residue of the DRE motif ( R198L ) , which may disrupt calcium binding . The last PCDH19-FE mutation in this group involves duplication of three residues ( SEA139-141dup at XEX ) , one of which is directly coordinating calcium . This duplication might change the architecture of the loop and alter calcium binding as well . Overall , mutations at PCDH19 calcium-binding motifs are varied , with some predicted to have drastic effects on protein folding and calcium binding , and others predicted to have a minor effect yet still causing protein malfunction . There are 19 PCDH19-FE missense mutations not found within EC1-4 ( Figure 1—figure supplement 2 ) , 14 of which are at conserved calcium-binding motifs ( N557K , D594H , D596G , H , V , Y ) or at other structurally conserved sites for cadherin repeats ( P451L , G486R , G513R , L543P , P561R , G601D , V642M , L652P ) . Two mutations involve insertion or deletion of residues ( N449_H450insN and S489del ) , and will likely disrupt β strand folding . However , the effect of the remaining three is unclear ( R550P in β strand G of EC5 , P567L in β strand A of EC6 , and D618N likely at the end of β strand D ) ; perhaps they are involved in cis interactions with PCDH19 or other cadherins . To gain insights into the molecular mechanism of the most common PCDH19-FE mutation , N340S ( N317S , Figure 1—source data 1 ) , we introduced this mutation into the Pcdh19 EC3-4 construct and compared its thermal stability with the wild-type ( WT ) Pcdh19 fragment ( Figure 1H ) . The Pcdh19 EC3-4 N317S fragment refolded well as assessed by size exclusion chromatography ( SEC ) , but its melting temperature is considerably lower ( 40 . 7 ± 0 . 6°C vs . 52 . 4 ± 0 . 3°C ) , even in the presence of 2 mM CaCl2 . Another PCDH19-FE mutation of a surface residue ( E313K , equivalent to E290K ) did not show a dramatic shift in melting temperature ( 50 . 4 ± 0 . 1°C ) . These SEC and thermal stability results indicate that the EC3-4 fragment carrying the N317S mutation is folded , and may bind calcium , yet it is not as stable as the wild-type fragment . Crystal structures have previously revealed the adhesive interfaces for classical cadherins , clustered protocadherins , and the protocadherin-15 and cadherin-23 complex ( Nagar et al . , 1996; Boggon et al . , 2002; Patel et al . , 2006; Ciatto et al . , 2010; Sotomayor et al . , 2012; Nicoludis et al . , 2015; Goodman et al . , 2016 ) . Although the Pcdh19 EC3-4 structure does not reveal any relevant interface , the Pcdh19 EC1-4 structure does . The purified Pcdh19 EC1-4 fragment elutes in two well-defined peaks in size exclusion chromatography experiments ( SEC ) , with these peaks most likely representing monomeric and dimeric states in solution ( Figure 2—figure supplement 1 ) . Pcdh19 EC1-4 crystals were grown from the putative dimer SEC peak elution , and two plausible adhesive interfaces are observed in our Pcdh19 EC1-4 structure . The first one , which we will refer to as Pcdh19-I1 , arises from contacts between the two Pcdh19 EC1-4 molecules in the asymmetric unit , and involves a fully-overlapped antiparallel dimer in which EC1 from one molecule interacts with EC4 from the other ( EC1:EC4 ) , EC2 with EC3 ( EC2:EC3 ) , EC3 with EC2 ( EC3:EC2 ) , and EC4 with EC1 ( EC4:EC1; Figure 2A , B ) . Within the same protein crystal structure , the second antiparallel interface ( Pcdh19-I2 ) involves the opposite side of Pcdh19 with observed EC2:EC4 , EC3:EC3 , and EC4:EC2 interactions , as well as potential ( not observed ) EC1:EC5 and EC5:EC1 contacts ( Figure 2—figure supplement 2A ) . Several lines of evidence favor the first interface Pcdh19-I1 as the most likely to mediate biological function . 10 . 7554/eLife . 18529 . 009Figure 2 . A crystallographic Pcdh19 antiparallel interface involves fully overlapped EC1-4 repeats . ( A ) Molecular surface representation of two Pcdh19 EC1-4 molecules interacting in the crystallographic asymmetric unit . In this dimeric arrangement , an interaction interface is formed by fully overlapped and antiparallel EC1-4 protomers ( Pcdh19-I1 ) . Red , dashed boxes indicate three interaction sites highlighted in panels ( C–E ) . ( B ) Side views of the Pcdh19 dimer and the interaction surface exposed with interfacing residues listed and shown in cyan . Sites mutated in PCDH19-FE located at the interface are shown in dark red . Sites with residue side chains pointing to the protein core are labeled in gray text . Three inter-molecular salt bridges are indicated ( *: R40-E328; **: E81-R349; ^: R158-E290 ) . ( C–E ) Detail of antiparallel interface ( red dashed boxes in A ) . Interfacing residues are in cyan and yellow ( PCDH-FE ) . Left panel is in the same orientation as A , middle and right panels are rotated around the dimer’s longest axis . Labels for one of the protomers are in italics . See also Figure 2—figure supplement 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 00910 . 7554/eLife . 18529 . 010Figure 2—figure supplement 1 . Two states for Pcdh19 EC1-4 in solution . Elution profile from a size exclusion chromatography experiment showing two clear , separate peaks of distinct hydrodynamic size . Crystallization of the Pcdh19 EC1-4 fragment was carried out with fractions collected from the peak representing the largest species ( 68 . 4 ml ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 01010 . 7554/eLife . 18529 . 011Figure 2—figure supplement 2 . Alternate crystallographic antiparallel interface involves EC1 to EC5 repeats . ( A ) Molecular surface representation of two Pcdh19 EC1-4 molecules forming an antiparallel dimer that would involve fully overlapped EC1-5 repeats ( Pcdh19-I2 interface ) . Side views and interaction surface exposed with interfacing residues listed and shown in cyan . Sites mutated in PCDH19-FE located at the interface are shown in dark red . Sites with residue side chains pointing to the protein core are labeled in gray text . Labels for one of the protomers are in italics . Two residues involved in a Pcdh19-I2 interface salt-bridge are indicated with a plus sign ( + ) . Buried surface area in this interface is 310 Å2 per interacting EC , compared to 413 Å2 per interacting EC for the Pcdh19-I1 interface ( Figure 2 ) . ( B–C ) Protein G beads coated with full-length extracellular wild-type ( WT ) Pcdh19ECFc ( B ) or an engineered mutation ( C ) imaged after incubation for 1 hr followed by rocking for 2 min in the presence of calcium . Bar – 100 µm . ( D ) Mean aggregate sizes for WT and R364E in the presence of calcium after 1 hr of incubation followed by rocking for 1 min ( R1 ) and 2 min ( R2 ) . Error bars are standard error of the mean ( n = 3 for each construct , Figure 3—source data 1 ) . Aggregation of Pcdh19-I2 interface mutant R364E is within the variation of WT samples ( see also Figure 3H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 01110 . 7554/eLife . 18529 . 012Figure 2—figure supplement 3 . Pcdh19 dimer interfaces and predicted glycosylation and glycation sites . ( A ) Molecular surface representation of the Pcdh19-I1 interface ( left ) with interaction surface exposed ( right ) and with all predicted glycosylation ( green ) and glycation ( light cyan ) sites listed . Interfacing residues are shown in cyan . Sites mutated in PCDH19-FE and at the interface are shown in dark red . None of the predicted glycosylation sites involve interfacial residues , but two glycation sites do ( K156 and K308 ) . To the best of our knowledge , glycation has never been reported for cadherins . ( B ) Molecular surface of the alternate Pcdh19-I2 interface shown as in ( A ) . Glycosylation ( T232 ) and glycation ( K204 ) sites are predicted for interfacial residues . O-linked glycosylation is also predicted for human PCDH19 S204 , equivalent to N202 in Pcdh19 . Non-conserved sites are in gray text . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 012 Analysis of the Pcdh19-I1 antiparallel interface with the Protein Interfaces , Surfaces and Assemblies ( PISA ) server ( Krissinel and Henrick , 2007 ) and with the NOXclass classifier ( Zhu et al . , 2006 ) revealed a large interface ( ~1650 Å2 ) , that is unlikely to be a crystal packing artifact ( 89 . 21% biological , 81% obligate ) . In contrast , the possible antiparallel Pcdh19-I2 interface is predicted by NOXclass to be non physiological , as its smaller interface area ( ~930 Å2 ) and the nature of its contacts matches those of crystal packing interactions ( 42 . 97% biological , 20 . 21% obligate ) . Yet , both interface areas are larger than 856 Å2 , an empirical cut-off that can distinguish biological interfaces from crystal contacts with 85% accuracy ( Ponstingl et al . , 2000 ) , and our analysis of the Pcdh19-I2 interface lacks contributions from possible EC1-EC5 contacts , which might be significant . Moreover , shape correlation ( Lawrence and Colman , 1993 ) for Pcdh19-I1 is lower than for the Pcdh19-I2 interface ( Sc-I1 = 0 . 44 vs . Sc-I2 = 0 . 61 ) , as there is a large gap between the main EC2-EC3:EC3-EC2 contacts and the EC1-EC4:EC4-EC1 interactions zones ( Figure 2A ) . To further differentiate between the possible Pcdh19-I1 and Pcdh19-I2 interfaces , we evaluated whether any of the six PCDH19-FE mutations altering surface residues at crystal contacts , but not necessarily protein structure , could interfere with binding . Five of these mutations ( S139L , T146R , P149S , E313K , and T404I; all at conserved sites ) change residues involved in the Pcdh19-I1 interface ( S116 , T123 , P126 , E290 , T381 respectively in Figure 2B–E ) , where we define residues at a given interface as those with a buried surface area that is at least 20% of their accessible surface area according to PISA . In all cases we predict altered EC1-4 homophilic binding , as the size and nature of the residue is changed by each mutation ( hydrophobic vs . hydrophilic; charged vs . non-charged ) . The remaining mutation ( H203P ) involves a non-conserved residue at the Pcdh19-I2 interface , which could impair its formation ( R180 in Figure 2—figure supplement 2A ) . However , the patient with the H203P mutation also carries another PCDH19-FE mutation ( F206C ) at a location mutated in other epilepsy patients ( Marini et al . , 2012; Depienne et al . , 2011 ) ; thus it is unclear if H203P is contributing to epilepsy . In contrast , all five PCDH19-FE mutations at the Pcdh19-I1 interface are likely causal , which suggests that Pcdh19-I1 is relevant in vivo . We also analyzed predicted glycosylation sites that might interfere with binding and thereby reveal non-physiological interfaces , as observed for VE-cadherin ( Brasch et al . , 2011 ) . There are 14 glycosylation sites within EC1-4 , and none of them involve residues at the Pcdh19-I1 interface ( Figure 2—figure supplement 3A ) . An O-linked glycosylation site is predicted to be at the Pcdh19-I2 interface ( T232 ) , and an additional O-linked glycosylation site is predicted for the human PCDH19 protein at S204 ( the equivalent N202 in Pcdh19 is predicted to be non-glycosylated ) , also at the Pcdh19-I2 interface ( Figure 2—figure supplement 3B ) . Glycation sites , for which sugar molecules might be added randomly and to long-lived proteins , are predicted at both interfaces ( K156 and K308 in Pcdh19-I1 and K204 in Pcdh19-I2 ) , but may not interfere directly with either , since glycation depends on environmental conditions and it has never been reported for cadherins ( Salahuddin et al . , 2014; Simm et al . , 2015 ) . Thus the lack of glycosylation sites at the Pcdh19-I1 interface renders it as the most likely to be functional . While not conclusive , all the analyses presented above favor the Pcdh19-I1 antiparallel dimer over the Pcdh19-I2 interface in terms of physiological relevance . The larger surface area of the Pcdh19-I1 dimer , the nature of the residues involved , the number of PCDH19-FE mutations at this interface , and the lack of predicted glycosylation sites , all suggest that the Pcdh19-I1 interface may occur and be functional in vivo . To conclusively test which binding interface mediates Pcdh19 adhesion , and whether PCDH19-FE mutations at the protein surface can interfere with one of the two possible Pcdh19 interfaces described above , we used modified bead aggregations assays , mutagenesis , and size exclusion chromatography experiments . Previous cell-based assays showed weak homophilic adhesion for the chicken Pcdh19 ( Tai et al . , 2010 ) . In addition , previous assays in which the full-length Pcdh19 extracellular cadherin domain fused to Fc ( Pcdh19ECFc ) was incubated with protein A beads showed calcium-dependent aggregation only when it was co-purified with N-cadherin ( Biswas et al . , 2010; Emond et al . , 2011 ) . To study Pcdh19 homophilic interactions , we modified the previous protocol ( Emond and Jontes , 2014 ) and added a final step in which beads were rocked ( Sano et al . , 1993 ) in a controlled fashion for up to two minutes ( see Materials and methods ) . The modified protocol allowed us to identify clear bead aggregates mediated by Pcdh19ECFc alone ( Figure 3—figure supplement 1 ) . To identify the minimal adhesive unit of Pcdh19 we used our modified protocol with truncated versions of Pcdh19 containing different numbers of EC repeats: Pcdh19ECFc ( EC1-6 ) , Pcdh19EC1-5Fc , Pcdh19EC1-4Fc , Pcdh19EC1-3Fc , Pcdh19EC1-2Fc , and Pcdh19EC2-6Fc ( Figure 3 ) . Bead aggregation was observed only when using Pcdh19ECFc , Pcdh19EC1-5Fc , and Pcdh19EC1-4Fc , thus suggesting that Pcdh19EC1-4 is the minimal adhesive unit and highlighting the biological relevance of the antiparallel Pcdh19-I1 interface , which involves EC1-4 only . 10 . 7554/eLife . 18529 . 013Figure 3 . Minimal adhesive Pcdh19 fragment includes repeats EC1-4 . ( A–F ) Protein G beads coated with full-length ( A ) and truncated versions ( B–F ) of the Pcdh19 extracellular domain imaged after incubation for 1 hr followed by rocking for 2 min in the presence of calcium . Bar – 100 µm . ( G ) Western blot shows efficient production and secretion of full-length and truncated Pcdh19 extracellular domains . ( H ) Aggregate size for full-length and truncated versions of the Pcdh19 extracellular domain after 1 hr of incubation followed by rocking for 1 min ( R1 ) and for 2 min ( R2 ) . Error bars are standard error of the mean ( n = 3 for all aggregation assays and constructs ) . Inset: zoom-in showing pixel size from 15 to 85 ( y axis ) . Bead aggregation was observed for constructs including EC1-6Fc , EC1-5Fc , and EC1-4Fc . Data for EC1-6 is also plotted in Figure 2—figure supplement 2D ( WT ) , Figure 3—figure supplement 1C ( Pcdh19ECFc ( Ca2+ ) ) , and Figure 4H ( WT ( Ca2+ ) ) for comparison to additional constructs . See also Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 01310 . 7554/eLife . 18529 . 014Figure 3—source data 1 . Quantification of aggregation assays . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 01410 . 7554/eLife . 18529 . 015Figure 3—figure supplement 1 . Modified bead aggregation assays can detect calcium-dependent homophilic Pcdh19 interactions . ( A ) Protein G beads coated with full-length extracellular Pcdh19ECFc or NcadECFc imaged after incubation for 1 hr followed by rocking for 2 min in buffer with calcium . Representative images for parallel experiments in the absence of calcium are shown in panels labeled EDTA . Bar – 100 µm . ( B ) Western blot shows efficient production and secretion of Pcdh19ECFc ( also shown in Figure 4G WT lane ) and NcadECFc , the latter present with and without cleavage of its prodomain . ( C ) Aggregate size for Pcdh19ECFc and NcadECFc in the presence and absence of calcium after 1 hr of incubation followed by rocking for 1 min ( R1 ) and for 2 min ( R2; see also Figure 3H ) . ( D ) Detail of plot in ( C ) . Aggregate size for Pcdh19ECFc in the presence of calcium ( Ca2+ , green ) is larger than in the absence of it ( black ) after rocking . See also Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 015 Next , we introduced two PCDH19-FE mutations ( T146R and E313K located at the Pcdh19-I1 interface; T123R and E290K in Figure 2B ) in the full-length Pcdh19 extracellular domain and tested bead aggregation with these protein constructs ( Figure 4 ) . Bead aggregates were not detected when the Pcdh19ECFc carried these mutations under the conditions tested ( Figure 4B–C , E–H ) . In contrast , the mutation R364E , predicted to impair the Pcdh19-I2 interface , did not eliminate bead aggregation ( Figure 2—figure supplement 2B–D ) . Moreover , the presence of a N-cadherin ( Ncad ) fragment known to enhance Pcdh19-mediated adhesion ( Emond et al . , 2011 ) did not qualitatively change the effect of the T146R and E313K mutations . Bead aggregates were greatly diminished for T146R and abolished for E313K in the presence of NcadEC W2A/R14E His , a non-adhesive Ncad mutant previously used to study Pcdh19-mediated homophilic adhesion ( Harrison et al . , 2010; Emond et al . , 2011 ) ( Figure 4—figure supplement 1B–C , E–F ) . In addition , these mutations did not abolish the interaction between Pcdh19ECFc and NcadEC W2A/R14E His ( Figure 4—figure supplement 2 ) . It is possible that the T146R and 313K mutations affect interactions with N-cadherin in a subtle way ( directly or allosterically ) , yet our experimental results suggest that these mutations directly impair Pcdh19 homophilic adhesion . 10 . 7554/eLife . 18529 . 016Figure 4 . PCDH19-FE mutations at Pcdh19 I1 antiparallel interface impair Pcdh19-mediated bead aggregation . ( A–F ) Protein G beads coated with full-length extracellular wild-type ( WT ) Pcdh19ECFc ( A ) and two PCDH19-FE mutants ( B , C ) imaged after incubation for 1 hr followed by rocking for 2 min in the presence of calcium . Representative images for parallel experiments in the absence of calcium are shown in panels D to F ( EDTA ) . All full-length extracellular domains were produced in HEK293 cells . Bar – 100 µm . ( G ) Western blot shows efficient production and secretion of both WT and mutant proteins used for bead aggregation assays . Parallel black lines indicate a two-lane gap between samples . ( H ) Aggregate size for WT and PCDH-FE mutants in the presence ( Ca2+ ) and absence ( EDTA ) of calcium after 1 hr of incubation followed by rocking for 1 min ( R1 ) and for 2 min ( R2 ) . Error bars are standard error of the mean ( n = 3 for all aggregation assays and constructs , Figure 3—source data 1 ) . Aggregation is only observed for Pcdh19 WT in the presence of calcium and after rocking ( see also Figure 3H ) . ( I ) Analytical size exclusion chromatogram traces for WT ( green ) and mutant ( orange and red ) Pcdh19 EC1-4 fragments produced in E . coli . A shift in peak elution volumes indicates impaired homophilic interaction for mutants . ( J ) Schematics of proposed homophilic 'forearm handshake' for the Pcdh19 adhesion complex validated through binding assays with protein carrying PCDH19-FE mutations . See also Figure 4—figure supplement 1–2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 01610 . 7554/eLife . 18529 . 017Figure 4—figure supplement 1 . PCDH19-FE mutations at Pcdh19-I1 impair bead aggregation even in the presence of N-cadherin . ( A–D ) Protein G beads allowed to bind protein from cellular media containing NcadEC W2A/R14E His by itself ( D ) , with Pcdh19ECFc ( A ) , or with a PCDH19-FE mutants ( B–C ) imaged after incubation for 1 hr followed by rocking for 2 min . To test trans interactions mediated by Pcdh19 , we used the NcadEC W2A/R14E mutant , which abolishes Ncad-based homophilic adhesion ( Harrison et al . , 2010; Emond et al . , 2011 ) . ( E ) Western blots show the presence of both Pcdh19ECFc ( wild-type and mutant forms ) and NcadEC W2A/R14E His in the cellular media used for aggregation assays . ( F ) Aggregate sizes for NcadEC W2A/R14E His by itself , with Pcdh19ECFc WT , or Pcdh19ECFc mutants after 1 hr incubation followed by rocking for 1 min ( R1 ) and for 2 min ( R2 ) . Error bars are standard error of the mean ( n = 3 for all aggregation assays ) . Construct labels with a star ( * ) are shortened in graph legend . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 01710 . 7554/eLife . 18529 . 018Figure 4—figure supplement 2 . PCDH19-FE mutations at Pcdh19-I1 do not abolish the interaction between the extracellular domains of Pcdh19 and N-cadherin . Co-immunoprecipitation of Pcdh19ECFc ( WT or with PCDH19-FE mutations at Pcdh19-I1 ) pulls down NcadEC W2A/R14E His . Therefore , the PCDH19-FE mutations at the Pcdh19-I1 interface do not abolish the interaction between Pcdh19 and Ncad extracellular domains . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 018 We also introduced the T146R and E313K mutations at the Pcdh19-I1 interface into the bacterially expressed Pcdh19 EC1-4 protein fragment , and used analytical size exclusion chromatography to determine whether the mutant fragments were eluting as putative dimers or monomers in solution . Both mutations resulted in a shift of the elution peak that indicated a smaller , monomeric state ( Figure 4I ) . Taken together , our crystallographic structural analyses and binding assays including PCDH19-FE mutations strongly support a model in which fully overlapped EC1-4 domains ( Pcdh19-I1 interface ) form the functional adhesive unit of Pcdh19 ( Figure 4J ) . The antiparallel Pcdh19-I1 dimer interface validated above reveals a homophilic 'forearm handshake' binding mechanism for PCDH19 , involving overlap of 4 ECs from each protomer wrapping around each other . This is different from the mechanism used by classical cadherins , only involving EC1 ( Brasch et al . , 2012 ) or the heterophilic 'extended handshake' used by protocadherin-15 and cadherin-23 , involving overlap of only EC1-2 of each protein ( Sotomayor et al . , 2012 ) . The forearm handshake is similar to the binding mechanism recently reported for clustered protocadherins ( Goodman et al . , 2016; Rubinstein et al . , 2015; Nicoludis et al . , 2015 ) and might be used by other non-clustered protocadherins . The Pcdh19-I1 interface involves extended and mostly symmetric , in-register contacts between repeats EC2:EC3 that account for ~58% of the interfacial area , as well as smaller , separate EC1:EC4 contacts ( ~350 Å2 × 2 ) that are slightly off-register . The EC1:EC4 contacts arise as both repeats bend to meet after the C-terminal end of EC3 and the N-terminal end of EC2 separate from each other . In this arrangement , the EC2-3 linkers from each protomer are right next to each other , while the EC3-EC4 linker in one protomer is separated from the EC1-2 linker of the binding partner by a large opening . The interface is generally amphiphilic , with ~49% of the interfacial area involving hydrophobic residues , ~28% hydrophilic , and ~23% charged residues ( Figure 5—figure supplement 1 ) . Interestingly , the contact formed by EC1:EC4 is more hydrophobic ( 58%; 22%; 20% ) than the one formed by EC2:EC3 ( 41%; 33%; 26% ) , yet salt-bridge pairs across protomers are present in both: R40-E328 and E81-R39 enhance the EC1 to EC4 contacts ( Figure 2C ) and R158-E290 links EC2 to EC3 ( Figure 2D ) . While the R40-E328 pair seems to be zebrafish specific , the other two salt-bridges are highly conserved across sequenced species , along with most of the residues involved in the Pcdh19 EC1-4 interface ( Figure 5A and Figure 5—figure supplement 2 ) . The same set of residues is highly variable across different members of the δ1 , δ2 , and the clustered protocadherins ( Figure 5B and Figure 5—figure supplement 3 ) , suggesting that binding mechanisms might differ across subfamilies or that residue variability might confer specificity within a common binding mechanism . 10 . 7554/eLife . 18529 . 019Figure 5 . A common binding mechanism with sequence-diverse interfaces for δ and clustered protocadherins . ( A ) Molecular surface representation of the closed ( left ) and exposed ( right ) Pcdh19-I1 antiparallel dimer . Interfacing residues are colored according to sequence conservation among 102 species ( Figure 5—figure supplement 2 and Figure 5—source data 1 ) . Most of them are highly conserved . Labels as in Figure 2B . ( B ) Antiparallel Pcdh19 EC1-4 dimer shown as in ( A ) , with interfacing residues colored by sequence conservation among selected members of the non-clustered δ1- and δ2-protocadherins , as well as selected α , β , and γ clustered protocadherins ( Figure 5—figure supplement 3 and Figure 5—source data 2 ) . ( C ) Location of interfacing residues for Pcdh19 , Mm pcdhγC3 , Mm pcdhα4 and α7 , Mm pcdhγA1 , and Mm pcdhβ6 and β8 , mapped onto the Pcdh19 topology diagram . Shared structural motifs involved in binding include: The F-G loop along with the beginning of β strands A , G and C in EC1; the A-B loop , most of β strand B , the D-E loop , and the beginning of β strand E in EC2; the EC2-3 linker; the C-D loop , parts of β strands F and G and the F-G loop in EC3; the loop within β strand A , β strand B , and the D-E loop in EC4 . Red/orange circles indicate sites mutated in PCDH19-FE . Common contact zones in EC1 and EC3 , as well as EC2 and EC4 , are highlighted with a brown background . See also Figure 5—figure supplement 1–5 . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 01910 . 7554/eLife . 18529 . 020Figure 5—source data 1 . Protocadherin-19 sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 02010 . 7554/eLife . 18529 . 021Figure 5—source data 2 . Sequences for selected clustered and δ-protocadherins . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 02110 . 7554/eLife . 18529 . 022Figure 5—figure supplement 1 . Pcdh19-I1 antiparallel EC1-4 dimer interface involves charged , hydrophilic , and hydrophobic residues . Molecular surface representation of the Pcdh19-I1 antiparallel dimer with interfacial residues exposed and labeled . Surface is colored according to residue type ( apolar: white; polar: green; negatively charged: red; positively charged and histidines: blue ) . Interfacing residues are labeled as in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 02210 . 7554/eLife . 18529 . 023Figure 5—figure supplement 2 . Sequence alignment of Pcdh19 EC1-4 . Alignment of Human , Zebrafish , Mouse , Rat , Chicken , and Monkey sequences for Pcdh19 EC1-4 . Residues are colored according to conservation based on ConSurf ( Ashkenazy et al . , 2016 ) and an alignment of sequences from 102 species ( Figure 5—source data 1; gray indicates insufficient data due to inadequate diversity ) . Conserved calcium-binding motifs are shown on top of the alignment and labeled . Residues at Pcdh19-I1 are labeled with an orange dot on top of the alignment . Human residues mutated in PCDH19-FE are in bold white or dark red font . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 02310 . 7554/eLife . 18529 . 024Figure 5—figure supplement 3 . Sequence alignment of selected protocadherins . Alignment of selected sequences for δ1 and δ2 protocadherins , as well as selected clustered α , β , and γ protocadherins of known structure ( Figure 5—source data 2 ) . Residues are colored according to conservation based on ConSurf as in Figure 5—figure supplement 2 Conserved Ca2+-binding motifs are shown on top of the alignment . Residues at Pcdh19-I1 are labeled with an orange dot . Predicted glycosylation and glycation sites are labeled with green and light cyan dots , respectively . Pairs of salt-bridges observed at Pcdh19-I1 are indicated in bold red above the alignment . Human residues mutated in PCDH19-FE are in bold white or dark red font . Residues involved in any of the clustered protocadherin interfaces are labeled with a light blue dot below the alignment , and shown in italic bold . Secondary structure of Pcdh19 EC1-4 is shown in gray below the alignment . Long residue insertions were omitted for clarity in the sequences of pcdh7 EC2 ( QEP157 ~209RSS ) , pcdh8 EC4 ( AAP334 ~361GTP ) , pcdh10 EC2 ( GGG192 ~210QRT ) , and pcdh17 EC4 ( VLG377 ~391SVP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 02410 . 7554/eLife . 18529 . 025Figure 5—figure supplement 4 . Structural comparison of Pcdh19-I1 EC1-4 dimer to clustered-protocadherin dimers . ( A ) Two views of structurally aligned monomers ( gray , EC1-EC4 ) and their partners for Pcdh19 ( cyan ) and pcdhα4 ( blue , 5DZW [Goodman et al . , 2016] ) . Arrows point to regions of significant structural differences . ( B ) Two views of Pcdh19 and pcdhα7 ( 5DZV [Goodman et al . , 2016] ) shown as in ( A ) . The right panel shows schematics highlighting differences in dimer arrangement . ( C–F ) Structural alignments as in ( A–B ) for pcdhβ6 ( C , 5DZY [Goodman et al . , 2016] ) , pcdhβ8 ( D , 5DZX [Goodman et al . , 2016] ) , pcdhγA1 ( E , 4ZI9 [Nicoludis et al . , 2015] ) , and pcdhγC3 ( F , 4ZI8 [Nicoludis et al . , 2015] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 02510 . 7554/eLife . 18529 . 026Figure 5—figure supplement 5 . Structural comparison of protocadherin δ1 and δ2 EC3 repeats . ( A ) Ribbon representation of EC3 repeats from Pcdh19 ( δ2 , cyan ) , pcdh7 ( δ1 , dark cyan , PDB 2YST ) , and pcdh9 ( δ1 , ice blue , PDB 2EE0 ) structurally aligned to each other . ( B ) Molecular representation of the Pcdh19 EC3 ( left ) and EC2 ( right ) repeats within the Pcdh19-I1 EC1-4 dimer . Interfacing residues are exposed and shown in cyan , with E290 shown in dark red to indicate its involvement in PCDH19-FE . Location of R158 , which interacts with E290 , is shown in EC2 . The surface is also shown colored according to the residue type ( apolar: white; polar: green; negatively charged: red; positively charged and histidines: blue ) . N and C-termini are indicated . ( C and D ) Molecular surface representations for Pcdh7 EC3 and Pcdh9 EC3 , as in ( B ) , with predicted interfacing residues shown in dark cyan and ice blue , respectively . Charges of the E290-R158 pair are predicted to be swapped in Pcdh7 ( salt-bridge R-D ) and Pcdh9 ( salt-bridge K-E; Figure 5—figure supplement 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18529 . 026 A comparison of our Pcdh19-I1 interface to recently reported models and structures of clustered protocadherin interfaces ( Nicoludis et al . , 2015; Goodman et al . , 2016 ) reveals multiple similarities among them . The most complete models of α and β-protocadherins show similar , fully overlapped antiparallel EC1-4 dimers ( Figure 5—figure supplement 4A–D ) , with the same extended EC2:EC3 antiparallel connection accompanied with smaller EC1:EC4 contacts and salt-bridges across protomers . Structural alignments show that the relative arrangements of protomers within the antiparallel dimers for Pcdh19 , Mm Pcdhα4 ( 5DZW ) , and Mm Pcdhα7 ( 5DZV ) are the most similar to each other with slight shifting in some EC repeats ( Figure 5—figure supplement 4A , B ) . The Mm Pcdhβ6 ( 5DZX ) and Mm Pcdhβ8 ( 5DZY ) structures show similar dimeric interfaces , but the relative arrangement of protomers within the dimer is slightly shifted for all EC repeats ( Figure 5—figure supplement 4C , D ) . Similarly , the Mm PcdhγA1 EC1-3 interface ( 4ZI9 ) matches and aligns well with the Pcdh19 EC1-4 dimer ( Figure 5—figure supplement 4E ) . Mapping of all interaction sites to the Pcdh19 EC1-4 topology diagram ( Figure 5C ) reveals a pattern for common interacting domains in odd and even EC repeats across these structures , which include the F-G β hairpin and β strand A for repeats EC1 and EC3 , as well as the A-B and D-E β hairpins for EC2 and EC4 . While there are differences in some of the interacting domains , dimeric arrangements , and contact details , including diversity of interfacial residues , clustered protocadherins seem to use the same binding mechanism that Pcdh19 uses to mediate adhesion . A conserved RGD sequence motif within Pcdh19 EC2 ( residues 158 to 160 ) at its D-E loop is similar to an integrin-binding RGD site within EC1 ( C-D loop ) in the α-protocadherins ( Ruoslahti , 1996; Mutoh et al . , 2004 ) . The EC1 RGD motif is exposed in the Mm Pcdhα4 and Mm Pcdhα7 homo-dimers while the EC2 RGD motif of Pcdh19 ( also present in Pcdh17 [Kim et al . , 2011] ) is buried at the EC2:EC3 contacts in the Pcdh19-I1 interface . This suggests that homophilic binding could regulate the availability of this potential , untested , integrin-binding site . Pcdh19 belongs to the δ2-protocadherin subfamily , and given the sequence similarity among subfamily members , it is likely that all use the same dimer interface to mediate adhesion . This is less obvious for the δ1 subfamily , with members that have seven EC repeats and that display some critical differences at interaction sites , such as the presence of a positively charged residue ( R or K ) at position 290 , where most δ2 members have a negatively charged glutamate that interacts with an arginine at position 158 ( Figure 2D , Figure 5—figure supplements 3 and 5 ) . The PCDH19-FE E313K mutation at this site ( E290 ) prevents binding ( Figure 4C , H–I and Figure 4—figure supplement 1 ) , suggesting that δ1-protocadherins , which effectively carry the same mutation , should use a different interface to mediate adhesion . Yet , residues at position 157 and 158 in δ1-protocadherins are also charge swapped , with aspartates and glutamates that would restore this critical salt-bridge interaction at the EC2:EC3 interface , and at the same time prevent heterophilic interactions with δ2-protocadherins ( Figure 5—figure supplements 3 and 5 ) . Thus , it is likely that all non-clustered δ-protocadherins use fully overlapped EC1-4 antiparallel interfaces , like the one observed for Pcdh19 , to mediate adhesion .
The non-clustered δ-protocadherins are increasingly linked to human neurodevelopmental disorders , emphasizing both their importance to brain development and their relevance to human health ( Redies et al . , 2005; Redies et al . , 2012; Hirabayashi and Yagi , 2014 ) . In particular , mutations in PCDH19 cause a female-limited form of infant-onset epilepsy ( Dibbens et al . , 2008; Scheffer et al . , 2008; Depienne and LeGuern , 2012; van Harssel et al . , 2013; Leonardi et al . , 2014; Thiffault et al . , 2016; Terracciano et al . , 2016 ) . Therefore , it is imperative to understand the developmental roles of PCDH19 and other non-clustered δ-protocadherins , the structural basis of homophilic adhesion by these molecules , and the functional impact of pathogenic missense mutations . The structural and biochemical data presented here provide a first view on the molecular mechanism of Pcdh19 adhesion , which is likely used by all non-clustered δ and clustered protocadherins . Moreover , our Pcdh19 EC1-4 structural model shows > 70% of the missense mutations identified in PCDH19-FE patients , and reveals the biochemical basis for the deleterious effects for many of these mutations . The Pcdh19 EC1-4 structure reveals an antiparallel dimer that is consistent with a trans adhesive interface , a conclusion supported by multiple lines of evidence . Notably , several missense mutations identified in PCDH19-FE patients localize to this interface . Two of these missense mutations ( T146R and E313K ) impair dimerization , as assessed by analytical gel filtration , and adhesion as assessed in bead aggregation assays , with and without N-cadherin . Sequence analysis suggests that the antiparallel adhesive mechanism presented here is broadly relevant to other , related δ-protocadherins . Recent work with clustered protocadherins , implicated in self-avoidance and self/non-self recognition ( Lefebvre et al . , 2012; Kostadinov and Sanes , 2015; Yagi , 2012 ) , have revealed a similar antiparallel adhesive interface for these clustered protocadherins ( Rubinstein et al . , 2015; Nicoludis et al . , 2015; Goodman et al . , 2016 ) . Thus , the Pcdh19-I1 adhesive interface observed in our Pcdh19 EC1-4 structure likely represents the mechanism used by both non-clustered δ-protocadherins and clustered protocadherins , which , together , represent the largest group within the cadherin superfamily . Our structural data for Pcdh19 , as well as recent work with the clustered protocadherins raises an interesting conundrum . The adhesive interface for protocadherins is extensive and involves interactions extending throughout EC1-4 . This contrasts sharply with the adhesive interface of classical cadherins , which is restricted to EC1 and involves the reciprocal swap of Aβ-strands that is stabilized by burying Trp2 in a hydrophobic pocket ( Brasch et al . , 2012 ) . However , the KD for dimerization of α- and β-protocadherins is in the micromolar range ( similar to classical cadherins ) , bead aggregation and cell-based assays have consistently shown weak adhesion by both non-clustered and clustered protocadherins , and protocadherins are widely recognized as being only weakly adhesive ( Schreiner and Weiner , 2010; Thu et al . , 2014; Sano et al . , 1993; Rubinstein et al . , 2015 ) . This disparity suggests that other mechanisms could modulate protocadherin adhesion in vivo . For instance , cis-oligomerization could compete with trans adhesive interactions , or interactions with other proteins , including N-cadherin , could sequester protocadherins or mask their adhesive interface . Further studies will be required to better understand protocadherin adhesion , how it may be altered in the presence of N-cadherin , and how it is regulated in vivo . In addition to mutations that disrupt adhesion , our data reveal the potential effects of two other classes of mutations . In the first class , many mutations are predicted to directly impair folding and stability , which could lead to reduced levels of protein on the surface , due to impaired trafficking or enhanced protein degradation . In the second , PCDH19-FE mutations affecting calcium-binding sites are likely to cause shifts in calcium affinity as well as protein instability . Similar mutations in cadherin-23 and protocadherin-15 have been shown to decrease protein affinity for calcium , with KD shifts that are relevant in the context of the low calcium concentration to which these proteins are exposed ( Sotomayor et al . , 2010 ) . Yet PCDH19 is expected to be in interstitial space with high calcium concentration , so it is more likely that the relevant effect of PCDH19-FE mutations at calcium-binding sites is compromised stability ( even at saturating calcium concentrations ) , as shown here for the N340S mutation . Finally , analysis of one PCDH19-FE mutation within EC1-4 , and three within EC5-6 , reveal no obvious predicted consequences at the structural level , as they are exposed residues that should not affect calcium-binding , protein stability or adhesion . These mutations may impact a variety of protein-protein interactions . Although the physiological relevance is unclear , both non-clustered and clustered protocadherins can form cis-homo- or cis-hetero-oligomers ( Chen et al . , 2007; Schreiner and Weiner , 2010 ) , and mutations affecting the formation of cis-oligomers could adversely impact protocadherin function . Similarly , protocadherins participate in a variety of protein complexes beyond homophilic trans adhesion: Pcdh19 has been shown to associate in cis with N-cadherin ( Emond et al . , 2011 ) ; protocadherins associate with the Wnt co-receptor , RYK ( Berndt et al . , 2011 ) ; PAPC interacts with Frizzled-7 and FLRT3 ( Chen et al . , 2009; Kraft et al . , 2012 ) ; and Pcdh17 and Pcdh19 have highly conserved RGD sequences , suggesting that they may interact with integrins ( Ruoslahti , 1996; Mutoh et al . , 2004; Kim et al . , 2011 ) . Thus , further experimental characterization of key mutants in vitro and in vivo will continue to reveal correlations between structural defects , cellular-level defects , and different aspects of PCDH19-FE . The non-clustered protocadherins are increasingly recognized as a family of molecules that play important roles during neural development . In addition to the role of PCDH19 in epilepsy , mutation of PCDH12 was found to underlie a syndrome of microcephaly that is associated with epilepsy and developmental disability ( Aran et al . , 2016 ) . Moreover , both PCDH9 and PCDH10 have been associated with autism spectrum disorders ( Marshall et al . , 2008; Morrow et al . , 2008 ) . Ongoing work will likely uncover further links between members of this family and neurodevelopmental disorders . Our Pcdh19 EC1-4 model is the first to show the structural basis of adhesion by the non-clustered δ-protocadherins , and reveals that some of the missense mutations identified in PCDH19-FE occur at the adhesive interface and act by abolishing adhesion . This represents an initial stage in understanding the mechanisms of non-clustered δ-protocadherin homophilic adhesion and provides insight into the biochemical basis of protocadherin-based neurodevelopmental disease .
Zebrafish Pcdh19 repeats EC1-4 and EC3-4 were subcloned into NdeI and XhoI sites of the pET21a vector for bacterial expression . Constructs for mammalian expression were created from previously reported constructs ( Pcdh19 , Pcdh19EC , Ncad , and NcadECW2A/R14E ) and cloned into CMV:N1-Fc and CMV:N1-His backbones , respectively ( Biswas et al . , 2010; Emond et al . , 2011 ) . Truncated versions of Pcdh19 ( Pcdh19EC1-5 , Pcdh19EC1-4 , Pcdh19EC1-3 , Pcdh19EC1-2 , Pcdh19EC2-6 ) were created by PCR subcloning of a Kozak sequence ( GCCACC ) , the signal peptide , and appropriate EC domains into CMV:N1-Fc . Mutations were created in both the bacterial and mammalian expression constructs by site-directed mutagenesis . All constructs were sequence verified . Each construct was expressed in BL21CodonPlus ( DE3 ) -RIPL cells ( Stratagene ) , cultured in TB ( EC1-4 ) or LB ( EC3-4 ) , induced at OD600 = 0 . 6 with 100 µM ( EC1-4 ) or 200 µM ( EC3-4 ) IPTG and grown at 30°C ( EC1-4 ) or 25°C ( EC3-4 ) for ~16 hr . Cells were lysed by sonication in denaturing buffer ( 20 mM TrisHCl [pH7 . 5] , 6 M guanidine hydrochloride , 10 mM CaCl2 and 20 mM imidazole ) . The cleared lysates were loaded onto Ni-Sepharose ( GE Healthcare , Sweden ) , and eluted with denaturing buffer supplemented with 500 mM imidazole . Pcdh19 EC3-4 was refolded by overnight dialysis against 20 mM TrisHCl [pH 7 . 5] , 150 mM NaCl , 400 mM arginine , 2 mM CaCl2 , 2 mM DTT using MWCO 2000 membranes . Pcdh19 EC1-4 was refolded by iterative dilution of the denaturing buffer with refolding buffer ( 100 mM TrisHCl [pH 8 . 5] , 10 mM CaCl2 ) ( Dechavanne et al . , 2011 ) . Refolded protein was further purified on a Superdex200 column ( GE Healthcare ) in 20 mM TrisHCl [pH 8 . 0] , 150 mM NaCl , 2 mM CaCl2 and 1 mM DTT . Crystals were grown by vapor diffusion at 4°C by mixing equal volumes of protein ( Pcdh19 EC3-4 = 14 . 4 mg/ml and Pcdh19 EC1-4 = 7 . 7 mg/ml ) and reservoir solution ( Pcdh19 EC3-4 contained 100 mM calcium acetate , 100 mM sodium cacodylate [pH 6 . 1] , 25% MPD; Pcdh19 EC1-4 contained 200 mM sodium chloride , 100 mM TrisHCl [pH 8 . 1] , 8% PEG 20 , 000 ) . Crystals were cryoprotected in reservoir solution ( Pcdh19 EC3-4 ) or with 25% glycerol added ( Pcdh19 EC1-4 ) , and then cryo-cooled in liquid N2 . X-ray diffraction data were collected as indicated in Table 1 and processed with HKL2000 or HKL3000 ( Minor et al . , 2006 ) . The Pcdh19 EC3-4 structure was determined by molecular replacement using separate homology models for each repeat ( 4AQE_A for EC3 and 1L3W for EC4 ) as an initial search model using MrBUMP ( Keegan and Winn , 2007 ) and PHASER ( McCoy et al . , 2007 ) . Model building was done with COOT ( Emsley et al . , 2010 ) and restrained TLS refinement was performed with REFMAC5 ( Murshudov et al . , 2011 ) . Likewise , the Pcdh19 EC1-4 structure was determined through molecular replacement using Pcdh19 EC3-4 as the initial search model in PHASER . Data collection and refinement statistics are provided in Table 1 . The final model for Pcdh19 EC3-4 is missing residues 243–246 in chain A , and residues 244–248 in chain B ( chains C and D are complete ) . The Pcdh19 EC1-4 model is missing residues 32–36 in chain A , residue V1 in chain B , and side chains for residues K17 , K75 , K419 in chain A and for residues K5 , R71 , and E95 in chain B . All molecular images were generated with VMD ( Humphrey et al . , 1996 ) . The wild-type ( WT ) and mutant Pcdh19 EC3-4 fragments were purified as described above and used for differential scanning fluorimetry ( DSF ) ( Niesen et al . , 2007; Lavinder et al . , 2009 ) . The experiments were repeated three to nine times using protein at 0 . 3 mg/ml for WT ( n = 9 ) , N317S ( n = 9 ) , and E290K ( n = 3 ) in buffer ( 20 mM TrisHCl [pH 8 . 0] , 150 mM NaCl , 2 mM CaCl2 and 1 mM DTT ) mixed with SYPRO Orange dye ( final concentration 5x; Invitrogen ) . Fluorescent measurements were performed in a BioRad CFX96 RT-PCR instrument while samples were heated from 10°C to 95°C in 0 . 2°C steps . Melting temperatures were estimated when the normalized fluorescence reached 0 . 5 . Refolded proteins ( Pcdh19 EC1-4 WT , E313K , and T146R ) were separated from unfolded aggregate protein on a Superdex200 16/60 column ( GE Healthcare ) with 20 mM TrisHCl [pH 8 . 0] , 150 mM NaCl , 2 mM CaCl2 and 1 mM DTT at 4°C . The fraction corresponding to greatest absorbance was run subsequently on a Superdex200 PC3 . 2/3 . 0 column with the same buffer at 4°C . An AKTAmicro system provided a controlled flow rate of 50 µl/min with the sample being injected from a 100 µl loop . Bead aggregation assays were modified from those described previously ( Emond and Jontes , 2014; Emond et al . , 2011; Sivasankar et al . , 2009 ) to detect the weak homophilic adhesion of Pcdh19EC . The Pcdh19ECFc fusion constructs were transfected alone or with NcadEC W2A/R14E His into HEK293 cells using calcium-phosphate transfection ( Kwon and Firestein , 2013; Barry et al . , 2014; Jiang and Chen , 2006 ) . Briefly , solution A ( 10 µg of plasmid DNA and 250 mM CaCl2 ) was added drop-wise to solution B ( 2x HBS ) while mildly vortexing , and the final transfection solution was added drop-wise to two 100 mm dishes of cultured HEK293 cells . The next day , cells were rinsed twice with 1xPBS and serum-free media . Cells were allowed to grow in the serum-free media for 2–3 days before collecting the media containing the secreted Fc fusions . The media was concentrated using ultracel ( Millipore ) and incubated with 1 . 5 µl of protein G Dynabeads ( Invitrogen ) while rotating at 4°C for 1–3 hr . The beads were washed in binding buffer ( 50 mM TrisHCl [pH 7 . 5] , 100 mM NaCl , 10 mM KCl , and 0 . 2% BSA ) and split into two tubes with either 2 mM EDTA or 2 mM CaCl2 . Beads were allowed to aggregate in a glass depression slide in a humidified chamber for 60 min without motion , followed by two 1 min intervals of rocking ( five oscillations/min , ±7° from horizontal ) . Images were collected upon adding EDTA or CaCl2 , after 60 min incubation , and after each rocking interval using a microscope ( AxioStar; Carl Zeiss ) with a 10x objective . Bead aggregates were quantified using ImageJ software as described previously ( Emond et al . , 2011; Emond and Jontes , 2014 ) . Briefly , the images were thresholded , the area of the detected aggregate particles was measured in units of pixels , and the average size was calculated . Assays were repeated three times from separate protein preps and their mean aggregate size ( ± SEM ) at each time point was plotted . Assays were excluded from analysis only if western blots failed to show protein expression . Western blots were performed on a portion of media containing the Fc fusion proteins before incubation with the beads to confirm expression and secretion of the protein . The media was mixed with sample loading dye , boiled for 5 min and loaded onto 10% Bis-Tris NuPAGE gels ( Invitrogen ) for electrophoresis . Proteins were transferred to PVDF membrane ( GE healthcare ) and blocked with 5% nonfat milk in TBS with 0 . 1% tween before incubating overnight with anti-human IgG or anti-His ( 1:200 Jackson ImmunoResearch Laboratories , Inc . ; 1:1000 NeuroMab ) . After several washes , the blot was incubated with anti-goat or anti-mouse HRP-conjugated secondary ( 1:5000 , Santa Cruz Biotechnology; 1:5000 Jackson ImmunoResearch Laboratories Inc . ) for chemiluminescent detection with Western Lightning substrate ( Perkin Elmer ) . HEK293 cells were transfected with Pcdh19ECFc ( wild-type or mutant ) and NcadEC W2A/R14E His constructs using calcium-phosphate transfection as described above . Briefly , solution A ( 8 µg of plasmid DNA , and 250 mM CaCl2 ) was added drop-wise to solution B ( 2x HBS ) while mildly vortexing , and the final transfection solution was added drop-wise to 60 mm dishes of cultured HEK293 cells . 24 hr after transfection , cells were washed twice with 1x PBS and once with serum free media , then cells were allowed to grow in the serum free media for 2–3 days . Media containing the secreted protein was collected and incubated overnight with 10 µl of protein G dynabeads ( Invitrogen ) at 4°C . Beads were washed once in wash buffer ( 20 mM TrisHcl [pH7 . 5] , 150 mM NaCl , 0 . 5% triton X-100 ) , then re-suspended in loading buffer . In addition , loading buffer was added to a small amount of reserved input media for each sample . The samples were loaded onto 10% Bis-Tris NuPAGE gels ( Invitrogen ) for electrophoreses . Proteins were transferred to PVDF membrane ( GE healthcare ) and blocked with 5% nonfat milk in TBS with 0 . 1% tween before incubating overnight with anti-human IgG or anti-his ( 1:200 Jackson ImmunoResearch Laboratories , Inc . ; 1:1000 NeuroMab ) . After several washes in TBS with 0 . 1% tween , the blot was incubated with anti-goat or anti-mouse HRP-conjugated secondary ( 1:5000 , Santa Cruz Biotechnology; 1:5000 Jackson ImmunoResearch Laboratories Inc . ) , washed , and developed with chemiluminescent detection with Western Lightning substrate ( Perkin Elmer ) . For analysis of Pcdh19 residue conservation across species , 102 sequences were obtained from the NCBI protein database and processed manually to include only the extracellular domain through the end of EC4 , using the canonical calcium-binding motifs and SignalP4 . 1 ( Petersen et al . , 2011 ) as guides . These Pcdh19 sequences ( Figure 5—source data 1 ) were then aligned using Clustal Omega ( Sievers and Higgins , 2014 ) and the alignment file was put into ConSurf ( Ashkenazy et al . , 2016 ) to calculate relative conservation of each residue and categorize the degree of conservation into nine color bins . Similarly , conservation between selected δ and clustered protocadherins was calculated in ConSurf . All human δ-protocadherin sequences and sequences for deposited structures of clustered protocadherins were selected and aligned to the sequences from our structure ( 5IU9 ) for input into Consurf ( Figure 5—source data 2 ) . Residue numbering throughout the text and in the structure corresponds to the processed protein , except when referencing human disease mutations for which the number follows standard numbering for the human protein , including the signal peptide ( see also Figure 1—source data 1 ) . The PCDH19 Female Epilepsy ( PCDH19-FE ) disease has been cataloged in the Online Mendelian Inheritance in Man ( OMIM 300088 ) and has previously been referred to by several different names including: Juberg-Hellman syndrome , epilepsy and mental retardation limited to females ( EFMR ) , and Early Infantile Epileptic Encephalopathy-9 ( EIEE9 ) . A thorough list of the currently known PCDH19-FE mutations is presented in Figure 1—source data 1 . Potential Pcdh19 glycosylation sites were predicted for both the human ( NP_001171809 . 1 ) and zebrafish ( ACQ72596 . 1 ) sequences using the following servers: NetNGlyc 1 . 0 ( N-glycosylation , GlcNAc-β-Asn ) , NetOGlyc 4 . 0 ( O-glycosylation , GalNAC-α-Ser/Thr ) ( Hansen et al . , 1998; Steentoft et al . , 2013 ) , and NetCGylc 1 . 0 ( C-glycosylation , Man-α-Trp ) ( Julenius , 2007 ) . In addition , we mapped conserved O-mannosylation sites found in the related δ-protocadherins ( Vester-Christensen et al . , 2013 ) , and mapped the glycosylation sites found in published clustered protocadherin structures from mammalian cells ( Rubinstein et al . , 2015 ) . Potential Protocadherin-19 glycation sites were predicted using the NetGlycate 1 . 0 server for both the human and zebrafish sequences ( Johansen et al . , 2006 ) . Coordinates for Pcdh19 EC1-4 and EC3-4 have been deposited in the Protein Data Bank with entry codes 5IU9 and 5CO1 , respectively . | As the brain develops , its basic building blocks – cells called neurons – need to form the correct connections with one another in order to give rise to neural circuits . A mistake that leads to the formation of incorrect connections can result in a number of disorders or brain abnormalities . Proteins called cadherins that are present on the surface of neurons enable them to stick to their correct partners like Velcro . One of these proteins is called Protocadherin-19 . However , it was not fully understood how this protein forms an adhesive bond with other Protocadherin-19 molecules , or how some of the proteins within the cadherin family are able to distinguish between one another . Cooper et al . used X-ray crystallography to visualize the molecular structure of Protocadherin-19 taken from zebrafish in order to better understand the adhesive bond that these proteins form with each other . In addition , the new structure showed the sites of the mutations that cause a form of epilepsy in infant females . From this , Cooper et al . could predict how the mutations would disrupt Protocadherin-19’s shape and function . The structures revealed that Protocadherin-19 molecules from adjacent cells engage in a “forearm handshake” to form the bond that connects neurons . Some of the mutations that cause epilepsy occur in the region responsible for this Protocadherin-19 forearm handshake . Laboratory experiments confirmed that these mutations impair the formation of the adhesive bond , revealing the molecular basis for some of the mutations that underlie Protocadherin-19-female-limited epilepsy . Other cadherin molecules may interact via a similar forearm handshake; this could be investigated in future experiments . It also remains to be discovered how brain wiring depends on Protocadherin-19 adhesion in animal development , and how altering these proteins can rewire developing brain circuits . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"and",
"conclusions",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics",
"neuroscience"
] | 2016 | Structural determinants of adhesion by Protocadherin-19 and implications for its role in epilepsy |
Due to their economic relevance , the study of plant pathogen interactions is of importance . However , elucidating these interactions and their underlying molecular mechanisms remains challenging since both host and pathogen need to be fully genetically accessible organisms . Here we present milestones in the establishment of a new biotrophic model pathosystem: Ustilago bromivora and Brachypodium sp . We provide a complete toolset , including an annotated fungal genome and methods for genetic manipulation of the fungus and its host plant . This toolset will enable researchers to easily study biotrophic interactions at the molecular level on both the pathogen and the host side . Moreover , our research on the fungal life cycle revealed a mating type bias phenomenon . U . bromivora harbors a haplo-lethal allele that is linked to one mating type region . As a result , the identified mating type bias strongly promotes inbreeding , which we consider to be a potential speciation driver .
Knowledge of basic molecular principles in biology has mainly been gained through intensive studies of relatively few but technically well accessible model systems ranging from prokaryotes to eukaryotes , including the bacterium Escherichia coli , the fungus Saccharomyces cerevisiae , the insect Drosophila melanogaster , the mammal Mus musculus , as well as the dicot plant Arabidopsis thaliana . However , findings in these model organisms cannot always be generalized to more distantly related species ( Mohammadi et al . , 2015; Rine , 2014 ) . This has led to the development of new model systems , closer to species of economic relevance for humans . For temperate poaceous crops like wheat and barley , the small , fast growing grass Brachypodium distachyon has become a promising model organism ( Mur et al . , 2011; Brkljacic et al . , 2011; Vogel and Hill , 2008; Huo et al . , 2008; Yordem et al . , 2011; Garvin , 2008; Draper et al . , 2001 ) . The intrinsic properties of B . distachyon , including self-fertility , its short life cycle , a small sequenced diploid genome , and its genetic accessibility , make it highly suitable as a laboratory model plant . Symbiotic and biotrophic interactions of plants with fungi or other microbes can have major impacts on plant development and crop yield ( Oerke , 2006 ) . Therefore , the study of these interactions is important for research-driven pest control . Biotrophic plant pathogens rely on a living host to proliferate and complete their life cycles . To successfully colonize their host plants , these pathogens employ small secreted molecules , termed effectors . By means of these molecules , biotrophic pathogens are able to avoid host recognition , suppress plant defense responses , and redirect the host metabolism for their needs ( Djamei et al . , 2011; Doehlemann and Hemetsberger , 2013; Djamei and Kahmann , 2012; Deslandes and Rivas , 2012; Bozkurt et al . , 2012 ) . Since most effectors target processes on the host side , the understanding of biotrophy and the molecular study of effectors profits strongly from both , a fully genetically accessible host plant as well as a pathogen that is amenable to genetic manipulation . Among the facultative biotrophic pathogens , the smut fungus Ustilago maydis is a valuable model to study biotrophic interactions ( Djamei et al . , 2011; Brefort et al . , 2014; Kämper et al . , 2006; Horst et al . , 2010; Brefort et al . , 2009; Doehlemann et al . , 2009 ) . Although U . maydis is genetically fully accessible with a small , completely sequenced genome of 20 . 5 Mb and many molecular tools available , its host plant Zea mays is not as suitable as laboratory model organism due to its large size , demanding growth requirements , cross-pollinating nature , elaborate transformation requirements , and complex genome ( Kämper et al . , 2006; Que et al . , 2014; Schnable et al . , 2009 ) . Accordingly , a more suitable model system where both , plant and pathogen are more accessible is highly desirable . The recent rediscovery that the U . maydis-relative U . bromivora can infect Brachypodium sp . provided the impetus to explore the suitability of the U . bromivora-Brachypodium system as a novel plant-pathogen model to study biotrophic interactions ( Barbieri et al . , 2011 ) . Here we describe the characterization of this valuable model pathosystem . This encompasses ( i ) the detailed understanding of the life cycle of U . bromivora , ( ii ) the identification of a sequenced compatible host , ( iii ) the establishment of transformation systems for both the pathogen and the plant , and ( iv ) the sequencing and analysis of the fungal genome . Moreover , the observation of a mating type bias present in U . bromivora provided first insights into the biology of this pathogen .
Descriptive studies on the life cycle and mating behavior of smut fungi have been performed for more than 100 years ( Bauch , 1922; Brefeld , 1883 ) . In all grass smuts ( family Ustilaginaceae ) studied so far , sexual and pathogenic development are coupled ( Begerow et al . , 2014 ) . Infection mostly occurs at the early seedling stage of the respective host plant . After penetrating the plant , the dikaryotic pathogen grows biotrophically inter- and intracellularly in its host and macroscopic symptoms are usually exclusively limited to the inflorescences ( Brefort et al . , 2009 ) . In these plant organs , fungal proliferation finally occurs after systemic growth and eponymous black teliospores are formed . U . maydis is a prominent exception as it induces spore filled galls on all aerial parts of its host ( Brefort et al . , 2009 ) . As described in the early 20th century by Robert Bauch ( Bauch , 1925 ) , U . bromivora grows in its host plant after infection at the seedling stage , and only exhibits macroscopic symptoms during inflorescence development by replacing flowers with black teliospore-filled sori ( Figure 1 ) . Teliospores are the diploid resting stage of these fungi and are able to survive harsh environmental conditions . In a humid , favorable environment , spores germinate , undergo meiosis and form a promycelium from which haploid cells are released . The haploid yeast-like cells are non-pathogenic and grow saprophytically ( Brefort et al . , 2009 ) . To elucidate this part of the life cycle of U . bromivora , we followed germination of fungal spores by widefield and confocal laser scanning microscopy . Under nutrient-rich conditions , U . bromivora spores germinated , formed a promycelium and yeast-like cells were released ( Figure 1A–C , Video 1 ) . The egg-shaped cells of U . bromivora , which are released after germination , show high morphological similarity to the sporidia of the barley-infecting covered smut Ustilago hordei , and their morphology differ from the cigar-like shaped haploid cells of U . maydis ( Figure 1—figure supplement 1 ) . The high morphological similarity between U . bromivora and U . hordei is in line with phylogenetic analysis based on internal transcribed spacer ( ITS ) and large subunit ( LSU ) rDNA sequence comparison , which suggested that U . bromivora is more closely related to U . hordei than to U . maydis ( Stoll et al . , 2005 ) . 10 . 7554/eLife . 20522 . 003Figure 1 . The life cycle of Ustilago bromivora . U . bromivora spores germinate ( A ) and form a promycelium ( B ) . Under high nutrient conditions , haploid yeast-like progeny ( sporidia ) are released and proliferate via budding ( C ) . Under low nutrient conditions intratetrad mating occurs between two adjacent cells of a promycelium by formation of a loop-like mating structure that connects both cells ( D ) . After plant penetration , fungal filaments grow mainly along the stem without triggering macroscopic symptoms ( E ) until flower development occurs . Upon flowering , macroscopic symptoms are detectable as black , smutted spikelets filled with fungal spores ( F ) . Fungal cell walls and nuclei were stained with WGA-Alexa Fluor 488 and DAPI , respectively ( A–D ) . Plant membranes were stained with FM4-64 , fungal hyphae with WGA-Alexa Fluor 488 ( E ) . Scale bars: 5 µm ( A–D ) , 10 µm ( E , right panel ) , or 1 cm ( E , left panel ) and ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 00310 . 7554/eLife . 20522 . 004Figure 1—figure supplement 1 . The morphology of U . bromivora sporidia is similar to U . hordei , but not to U . maydis . U . bromivora UB1 , U . hordei Uh4875-4 , and U . maydis FB1 were grown in axenic culture to an exponential phase and pictures were acquired by widefield microscopy . Scale bars: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 00410 . 7554/eLife . 20522 . 005Video 1 . Spore germination under high nutrient conditions ( PD agar ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 005 During saprophytic growth , daughter cells bud off from the mother cell at the side of the oval tip . For a better visualization of the fungal cells , we employed the chitin stain Wheat Germ Agglutinin ( WGA ) that is conjugated to Alexa Fluor 488 ( Figure 1C ) . The WGA-Alexa Fluor 488 conjugate distinctively stained the tip of the sporidia indicating cell wall composition differences at the sporidial poles . This could be due to either an uneven chitin distribution or an uneven accessibility of chitin for the stain along the sporidial cell wall ( Figure 1C ) . By performing growth analyses we determined the doubling time of U . bromivora sporidia . In comparison to U . maydis sporidia that exhibit a doubling time of about 2 hr ( Steinberg , 2007 ) , U . bromivora sporidia grow much slower ( doubling time in exponential phase at 21°C: 5 ¾ hr; Figure 2 ) . Interestingly , we observed that the lag-phase of the U . bromivora culture could be significantly shortened when U . bromivora was kept for 24 hr at 4°C prior to inoculation ( Figure 2 ) . The underlying mechanism for this phenomenon is unclear and awaits further research . 10 . 7554/eLife . 20522 . 006Figure 2 . Axenic growth of U . bromivora . Growth of U . bromivora UB1 was assessed by monitoring cell density spectrophotometrically at λ = 600 nm in liquid PD medium at 21°C in a time course of 96 hr . Growth was compared between axenic cultures inoculated from plate with cold-treated fungal cell material ( 24 hr at 4°C ) or cell material that was kept at 21°C . Experiments were performed in 3 biological replicates . Significance between cell densities of cold-treated and non-treated cells at each time point was calculated by unpaired t test . **p<0 . 01 , ***p<0 . 001 , ****p<0 . 0001 . The doubling time was calculated by taking the slope of a linear regression during exponential phase . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 006 Prior to pathogenic growth , in case of the well-studied smut fungus U . maydis , two haploid cells of compatible mating type recognize each other via a pheromone-receptor system on the plant surface , grow towards each other , fuse , and form a dikaryotic filament . This dikaryotic filament represents the pathogenic form of the fungus . It penetrates the leaf cuticle and grows intra- and intercellularly inside the maize plant . After proliferation in the aerial parts of the host , the filaments fragment and , after karyogamy , form diploid spores ( Brefort et al . , 2009; Martinez-Espinoza et al . , 2002 ) . The molecular regulation of the dimorphic switch is based on the bi-allelic a locus encoding the pheromone receptor system and the multiallelic b locus coding for a heterodimeric homeodomain transcription factor that controls pathogenic development ( Schulz et al . , 1990; Bölker et al . , 1992 ) . While U . maydis has a tetrapolar mating type system with a and b loci that segregate during meiosis , these two loci are genetically and physically linked in U . hordei , leading to a bipolar mating system with only two mating types , mating type 1 ( MAT-1 ) and mating type 2 ( MAT-2 ) ( Bakkeren and Kronstad , 1994; Lee et al . , 1999 ) . To test for the presence of the mating system described for related smuts we assessed whether U . bromivora spores , that are considered to contain the genetic information of both mating partners , harbor known pheromone-receptor alleles . To this end , we employed a diagnostic PCR approach described by Kellner et al . ( 2011 ) using primers that target conserved regions of either the pheromone receptor allele pra1 or pra2 - . The PCR analysis revealed amplicons for pra1 and pra2 indicating the presence of both mating type alleles ( Figure 3A ) . To subsequently identify haploid U . bromivora cells of compatible mating type after spore germination , 225 randomly chosen progeny of 21 spores were tested . Surprisingly , by testing these progeny , we identified only cells of mating type 1 ( MAT-1; Figure 3A , B ) . These findings indicate a mating type bias after spore germination ( Figure 3C ) . This phenomenon of biased strains was previously described in other related fungi where recessive alleles linked to the mating type-locus were found to be causative for the observed biases ( Nielsen , 1968; Hood and Antonovics , 2000 ) . 10 . 7554/eLife . 20522 . 007Figure 3 . Mating type bias of U . bromivora . ( A ) Diagnostic PCR on genomic DNA derived from spores and spore progeny to test for mating type 1 ( MAT-1 ) or mating type 2 ( MAT-2 ) . To this end , primers targeting a conserved region of pheromone receptor alleles 1 ( pra1 ) and 2 ( pra2 ) , adapted from Kellner et al . ( 2011 ) , were used . Sizes of PCR products are indicated with arrow heads . Representative PCR results are shown . ( B ) Quantification of mating type alleles of 225 progeny derived from 21 spores by PCR as described in ( A ) . ( C ) Schematic model illustrating the observed mating type bias phenomenon . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 007 Due to the putative haplo-lethal allele , we assumed that haploid MAT-2 cells have only a very short period of time to be rescued by finding and mating with a compatible partner , allowing the subsequent formation of a dikaryotic filament . To observe mating events , spores were germinated on water agar and monitored by light microscopy ( Video 2 , Figure 1D ) . After spore germination , the cytoplasmic connection to the spore content becomes separated and , in the majority of germination events , only two cells are visible outside the spores . These cells frequently form a cytoplasmic bridge-like mating structure and fuse ( Video 2 , Figure 1D ) . The mating of progeny derived from the same spore is also known as an intratetrad mating event ( Antonovics and Abrams , 2004 ) . Using DAPI staining at this stage , we most commonly observed a diffuse distribution of signal across the two mating cells and , more rarely , a distinct nuclei-like area ( Figure 1D ) . In contrast to these mating cells , distinct nuclei-like structures could be visualized with this stain in saprophytically growing cells ( Figure 1C ) . The destiny of the other two meiotic products , which , in the majority of observed cases , seemed to stay inside the spore shell , is unclear and awaits further research . The formation of conjugation hyphae that loop directly to neighboring , conjoined cells of the same tetrad could be a necessary result of the mating type bias and ensures an efficient re-entry into the pathogenic stage of the life cycle . At the same time , the haplo-lethal allele that is likely linked to the MAT-2 locus promotes an inbreeding mode of U . bromivora with direct consequences on speciation , genome size , and selection on recessive alleles as it has been observed in other organisms ( Wright et al . , 2008; Ellegren and Galtier , 2016; Joly , 2011 ) . 10 . 7554/eLife . 20522 . 008Video 2 . Spore germination under low nutrient conditions ( water agar ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 008 Although the observed mating type bias is an interesting biological phenomenon , it is an undermining factor for the use of U . bromivora as genetically accessible model system to study biotrophic interactions . It creates a situation where one mating partner can only be maintained in the pathogenic , dikaryotic stage together with its compatible mating type and , as a consequence , it can neither be cultured axenically nor transformed under laboratory conditions . Depending on the physical distance of the MAT-2 locus to the haplo-lethal allele that causes the mating type bias , homologous recombination can lead to an uncoupling and the formation of a haplo-viable MAT-2 strain . Therefore , we screened for such viable haploid MAT-2 recombinants by using a pooled infection assay where a pool of spore-derived progeny was grown saprophytically and used for re-infection of Brachypodium sp . ( for details see Material and Methods ) . By pursuing this approach , we identified one MAT-2 strain that we named UB2 . This strain , which is not derived from the same spore as UB1 , retained the capability to mate with the U . bromivora MAT-1 strain UB1 , which we isolated in a classical spore germination assay . Moreover , upon inoculation of Brachypodium sp . , UB2 along with UB1 formed viable spores demonstrating a successful completion of their life cycle . Infection symptoms of UB1xUB2-inoculated plants were indistinguishable from infected spikelets of spore-inoculated plants ( data not shown ) . Phylogenetic analysis recently led to a taxonomy split within the Brachypodium lineages , separating B . distachyon with cytotype 2n = 10 from Brachypodium stacei ( 2n = 20 ) and the allotetraploid Brachypodium hybridum ( 2n = 30 ) ( Catalan et al . , 2012 ) . The spontaneous infection event by U . bromivora reported by Barbieri et al . ( 2011 ) occurred in the B . hybridum accession Bd28 . The B . hybridum accession Bd28 undergoes self-fertilization , grows fast , and is easy to handle ( Barbieri et al . , 2011 ) . One slightly complicating fact is its allotetraploid genetic background but its genome is currently being sequenced and will be available to the community in near future ( J . Vogel , unpublished ) . In order to identify additional Brachypodium sp . accessions susceptible to U . bromivora , in total , 39 accessions were infected with spores and analyzed for macroscopic infection symptoms in the spikelets . Whereas B . distachyon Bd21 , along with 27 other tested accessions , did not show infection symptoms , we found eleven accessions to be fully susceptible to U . bromivora ( Table 1 ) . The finding that susceptible host plant accessions originate from Europe , Asia , Africa , Australia as well as South America ( Table 1 ) underlines that U . bromivora is considered as a cosmopolite ( Bauch , 1925 ) . The natural host range of U . bromivora comprises various species of the genera Agropyron , Austrofestuca , Brachypodium including B . distachyon , Bromus , Critesion , Elymus , Festuca , Hordeum , Lolium , Sitanion and Trachynia ( Bauch , 1925; Fisher and Holton , 1957; Vanky , 2011 ) and our experiments can confirm at least three host species , B . distachyon , B . hybridum and B . stacei . Among the susceptible accessions are the B . distachyon accessions ABR4 , originally collected from Southern Spain , and Bd1-1 , an inbred line , as well as the B . stacei accession ABR114 ( Figure 4 ) . The genomes of all three susceptible diploid accessions have been sequenced ( J . Vogel , unpublished ) , providing the foundation for valuable tools and studies of our novel plant pathosystem . While , to our knowledge , ABR4 has not been tested for susceptibility to other important fungal pathogens , Bd1-1 shows resistance to the wheat pathogen Zymoseptoria tritici as well as to Puccinia graminis ( Figueroa et al . , 2013; O’Driscoll et al . , 2015 ) . Therefore , Bd1-1 represents a suitable model accession to study compatible interactions with U . bromivora as well as incompatible interactions with P . graminis and Z . tritici . Moreover , the comparison of susceptible B . distachyon accessions such as Bd1-1 and ABR4 , with the resistant accession Bd21 allows the study of compatible and incompatible interactions with U . bromivora without switching pathosystems . The discovery of several Brachypodium accessions resistant to U . bromivora , comprising accessions of B . distachyon , B . stacei , and B . hybridum , suggests that resistance likely evolved in a common ancestor of B . distachyon and B . stacei . As there is evidence that B . hybridum is the product of an interspecies cross between these two diploid taxa ( Petersen et al . , 2011 ) , this scenario would suggest that smut resistance was independently lost several times in each taxon . Alternatively , resistance based on single or even multiple factors could have evolved independently several times in the different closely related species . Within the limits of a relatively small sample size we observe a trend of B . hybridum accessions to be susceptible to U . bromivora . This could be due to suppression of resistance in polyploid genomes , a previously described phenomenon found in tetra- and hexaploid wheat ( Knott , 2000; Kerber , 1991 ) . 10 . 7554/eLife . 20522 . 009Figure 4 . ABR4 , ABR114 , and Bd1-1 are susceptible to U . bromivora . Representative pictures of infected ABR4 , ABR114 , and Bd1–1 spikelets . Sequenced B . distachyon accessions ABR4 and Bd1-1 as well as the sequenced B . stacei accession ABR114 were inoculated with U . bromivora spore material and screened for infection symptoms . Upon flowering macroscopic infection symptoms could be observed as spore-filled spikelets . This figure relates to Table 1 . Scale bars: 1 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 00910 . 7554/eLife . 20522 . 010Table 1 . Tested Brachypodium accessions for U . bromivora susceptibility/resistance . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 010NameSpeciesSequencedSusceptible to U . bromivoraAccession numberSourceCountry of originABR3B . distachyonYN ( 2x ) *ABY-Bs 5088Brachyomics collections ( C . Stace and P . Catalán ) , Aberystwyth , UKSpainABR4B . distachyonYY ( many ) ABY-Bs 5089Brachyomics collections ( C . Stace and P . Catalán ) , Aberystwyth , UKSpainABR6B . distachyonYN ( 2x ) *ABY-Bs 5091Brachyomics collections ( C . Stace and P . Catalán ) , Aberystwyth , UKSpainABR7B . distachyonYN ( 2x ) *ABY-Bs 5092Brachyomics collections ( C . Stace and P . Catalán ) , Aberystwyth , UKSpainABR9B . distachyonYN-unknownCroatiaAdi-10B . distachyonYNW6 39243USDA-ARS-WRPIS; Vogel et al . ( 2009 ) TurkeyAdi-12B . distachyonYNW6 39245USDA-ARS-WRPIS; Vogel et al . ( 2009 ) TurkeyAdi-2B . distachyonYNW6 39235USDA-ARS-WRPIS; Vogel et al . ( 2009 ) TurkeyBd1-1B . distachyonYYPI 170218 / W6 46201GRIN Germplasm; Vogel et al . ( 2006 ) TurkeyBd18-1B . distachyonYNPI 245730 / W6 46204USDA-ARS-WRPIS; Vogel et al . ( 2006 ) TurkeyBd21B . distachyonYNPI 254867 / W6 36678GRIN Germplasm; Vogel et al . ( 2006 ) IraqBd21-3B . distachyonYNW6 39233GRIN Germplasm; Vogel and Hill ( 2008 ) IraqBd2-3B . distachyonYNPI 185133 / W6 46202Vogel et al . ( 2006 ) IraqBd3-1B . distachyonYN ( 2x ) *PI 185134 / W6 46203USDA-ARS-WRPIS; Vogel et al . ( 2006 ) IraqBdTR10CB . distachyonYNW6 39406USDA-ARS-WRPISTurkeyBdTR11IB . distachyonYNW6 39426USDA-ARS-WRPIS; Filiz et al . ( 2009 ) TurkeyBdTR13aB . distachyonYNW6 39430USDA-ARS-WRPIS; Filiz et al . ( 2009 ) TurkeyBdTR13CB . distachyonYNW6 39432USDA-ARS-WRPIS; Filiz et al . ( 2009 ) TurkeyBdTR1IB . distachyonYNW6 39308USDA-ARS-WRPIS; Filiz et al . ( 2009 ) TurkeyBdTR2BB . distachyonYNW6 39314USDA-ARS-WRPIS; Filiz et al . ( 2009 ) TurkeyBdTR2GB . distachyonYNW6 39319USDA-ARS-WRPIS; Filiz et al . ( 2009 ) TurkeyBdTR3CB . distachyonYNW6 39332USDA-ARS-WRPIS; Filiz et al . ( 2009 ) TurkeyBdTR5IB . distachyonYNW6 39366USDA-ARS-WRPIS; Filiz et al . ( 2009 ) TurkeyFoz1B . distachyonYN-Mur et al . ( 2011 ) SpainGaz8B . distachyonYNW6 39269USDA-ARS-WRPIS; Vogel et al . ( 2009 ) TurkeyKah1B . distachyonYNW6 39278USDA-ARS-WRPIS; Vogel et al . ( 2009 ) TurkeyKoz1B . distachyonYNW6 39284USDA-ARS-WRPIS; Vogel et al . ( 2009 ) TurkeyMur1B . distachyonYN-Mur et al . ( 2011 ) SpainS8iiCB . distachyonYN-Ana Caicedo Lab , University of MassachusettsSpainABR114B . staceiYY ( 2x ) *-unknownSpainABR113B . hybridumYN-unknownPortugalABR117 , Bd117 , Bd6B . hybridumNY ( 2x ) *PI - 219965GRIN GermplasmAfghanistanBal-P7B . hybridumNY ( 2x ) *W6 - 39259GRIN GermplasmTurkeyBd23B . hybridumNY ( 2x ) *PI - 287783GRIN GermplasmSpainBd26 B . hybridumNY ( 2x ) *PI - 372187GRIN GermplasmUruguayBd28B . hybridumNY ( many ) *PI - 533015GRIN GermplasmAustraliaBd4B . hybridumNY ( 2x ) *PI - 208216GRIN GermplasmSouth AfricaBd8B . hybridumNY ( 2x ) *PI - 219971GRIN GermplasmAfghanistanIsk-P4B . hybridumNY ( 2x ) *W6 - 39273GRIN GermplasmTurkeyY = yes , N = no; * = times tested Since Bd28 is an excellent host for U . bromivora and needs only very short vernalization to induce flowering and to show infection symptoms , we used this accession for the establishment of a model host . To this end , we developed an efficient growth and infection method providing high germination rate , reliable floral induction , and high infection rate by U . bromivora ( for details see Material and Methods ) . These protocols could be also applied for the diploid and sequenced accession ABR4 . However , in contrast to Bd28 , ABR4 requires a minimum of four weeks vernalization to induce flowering and needs six weeks of vernalization to induce fast bulk flowering . 95 days after the initiation of vernalization , flowering of ABR4 is completed and we observed infection rates of almost 100% . Due to its shorter vernalization requirement , Bd28 can complete its life cycle in approximately 70 days . Interestingly , different time points of spore inoculation as well as the spore load seem to lead to varying infection efficiency . We regularly observed infected plants showing a few healthy spikelets beside spore-filled ones ( data not shown ) . The presence of healthy spikelets might be the evolutionary result of a sustainable virulence strategy of this host-specific , biotrophic pathogen to avoid complete sterilization of its host and therefore its local extinction . The establishment of transformation protocols for both the pathogen and its host plant is an important requirement to employ U . bromivora and Brachypodium sp . as a genetically accessible biotrophic model system . For initial transformation tests of U . bromivora , we used the self-replicating pNEBuC-GFP and pNEBuC-mCherry-HA vectors , which were developed for U . maydis ( Brachmann , 2001 ) . These plasmids contain a gene conferring Carboxin resistance and a gene encoding either the green fluorescent protein ( GFP-HA ) or mCherry-HA protein . GFP-HA or mCherry-HA are under control of the artificial otef promoter , which is constitutively active in axenic culture ( Spellig et al . , 1996 ) . Expression of these genes allows a fast readout of the transformation efficiency . Although different transformation protocols were tested , PEG-mediated protoplast transformation turned out to be most efficient ( average transformation efficiency: 314 colonies per µg plasmid DNA; Figure 5A ) . Since appropriate selection markers are essential for high-efficiency transformations , we concomitantly tested three different antibiotics and selection markers . While wild type cells could not be propagated , transformants harboring self-replicating plasmids conferring Carboxin , Geneticin G418 , or Hygromycin resistance were able to grow on respective selection media ( Figure 5—figure supplement 1 ) . We also tested integrative constructs for various loci such as the mating type region or the predicted pep1 gene locus , encoding an effector ortholog that has been shown to contribute to virulence in U . maydis ( Doehlemann et al . , 2009 ) . However , our results suggest , that in contrast to U . maydis , DNA uptake or other steps during transformation seem to be less efficient in U . bromivora preventing high transformation rates and strongly reduced the number of stable integration events . Therefore , we conducted a restriction enzyme mediated integration ( REMI ) approach which should promote genomic integrations ( Bölker et al . , 1995 ) . By employing this technique , we obtained a UB1 derivative with a stable integration of GFP driven by the otef promoter ( UB1-GFP ) . To ensure that after REMI mutagenesis , UB1-GFP has retained its capability to infect , we inoculated germinating Bd28 caryopses with a mixture of UB1-GFP and UB2 . Spore-filled spikelets and GFP producing progeny derived from these spores demonstrated a successful infection and completion of the fungal life cycle ( Figure 5—figure supplement 2 ) . To establish transformation for the B . hybridum accession Bd28 , we adapted a Bd21 transformation protocol ( Vain et al . , 2008 ) for its specific needs ( Figure 5B ) . As a proof of principle , we generated transgenic lines harboring a fluorescently tagged peroxisome-targeting construct ( eCFP-SKL ) . On the one hand , this construct provides a fast visual read out of a successful transformation; on the other hand , respective transgenic lines might be a valuable tool for studying cell biological questions in Brachypodium sp . ( Figure 5—figure supplement 3 ) . 10 . 7554/eLife . 20522 . 011Figure 5 . Establishment of transformation for the fungal pathogen U . bromivora and its host plant B . hybridum Bd28 . ( A ) Schematic representation and timeline of protoplastation and transformation of U . bromivora UB1 with the autonomously replicating pNEBuC-GFP plasmid conferring Carboxin resistance and encoding GFP . ( B ) Schematic representation and timeline of A . tumefaciens-mediated transformation of B . hybridum Bd28 . Illustrations are not to scale . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 01110 . 7554/eLife . 20522 . 012Figure 5—figure supplement 1 . Several resistance markers can be employed for selection of U . bromivora transformants . UB1 and transformants derived from this strain harboring autonomously replicating plasmids with genes known to confer resistance to Carboxin ( Cbx ) , Hygromycin B ( Hyg ) or Geneticin G418 ( G418 ) were spotted in serial dilutions on PD without and with the respective antibiotic . Antibiotic concentrations are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 01210 . 7554/eLife . 20522 . 013Figure 5—figure supplement 2 . UB2 can mate with UB1-GFP , form filaments , and produce viable spores . ( A ) UB2 , UB1 , and UB1-GFP were grown in PD to an exponential phase ( OD600 nm = 0 . 8 ) and resuspended in H2Odd to an OD600 nm = 2 . Strains were spotted individually and as mixture on charcoal-containing PD medium . White , fuzzy colonies on charcoal-containing medium indicate mating and the formation of filaments . Scale bar: 0 . 5 mm . ( B ) Viable spores are produced by UB1-GFP and UB2 . Germinating Bd28 caryopses were inoculated with an 1:1 mixture of UB1-GFP and UB2 . Upon flower development , plants showed macroscopic symptoms typical for U . bromivora infections . Retrieved spores germinated and produced viable GFP-fluorescent and non-fluorescent progeny . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 01310 . 7554/eLife . 20522 . 014Figure 5—figure supplement 3 . Microscopy of transgenic Bd28 eCFP-SKL marker line . Confocal laser scanning microscopy of leaf epidermis from a transgenic Bd28 line that stably produces eCFP-SKL as peroxisome marker protein . Upper panel shows wild type Bd28 displaying cell wall autofluorescence at 405 nm excitation and 455 nm emission . Lower panel shows transgenic Bd28-eCFP-SKL line . By employing identical settings , specific eCFP marked peroxisomes could be detected in the cytoplasm of epidermal cells in addition to cell wall autofluorescence . Scale bars: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 014 By comparative rDNA analysis , U . hordei has been shown to be the most closely related smut to U . bromivora that has been sequenced to date ( Stoll et al . , 2005 ) . The presence of extensive repetitive elements and transposable elements ( TE ) in the U . hordei genome complicated its genome assembly ( Laurie et al . , 2012 ) . To circumvent possible assembly problems with the related U . bromivora genome , we performed Single Molecule Real-Time ( Pacific Biosciences , Menlo Park , CA ) sequencing of UB1 , the isolated MAT-1 strain . PacBio sequencing has been shown to deliver long reads , facilitating fast and accurate genome assembly . After subread filtering , 376 , 645 reads ( 3 . 1 Gb total ) with 84 . 5% accuracy and a mean length of 8 , 186 bp ( approximately 154x coverage of the ~20 . 7 Mb genome ) were obtained . Assembly and polishing resulted in 25 contigs of which 23 showed good synteny with the optically mapped synthetic chromosomes of U . hordei ( Laurie et al . , 2012 ) . Based on this , U . bromivora has 23 chromosomes ( 20 . 5 Mb ) , the mitochondrial genome and an unassigned smaller contig . Gene model prediction led to the identification of 7 , 233 protein coding genes ( Table 2 ) . All gene models were manually curated by extensive comparative analysis to the existing Ustilaginaceae annotations and 82 . 3% of all gene models were confirmed by assembled transcripts obtained from Illumina-based RNA-seq of axenically grown UB1 . With 7 , 233 protein coding genes , U . bromivora harbors 120 more genes than the close relative U . hordei that has an assembled genome 0 . 65 Mb larger than U . bromivora ( Laurie et al . , 2012 ) . The compactness of the U . bromivora genome is underscored by the fact that 67 . 9% of the genes are intron-less , 19 . 3% have one intron , whereas only 12 . 8% are predicted to contain multiple introns . 10 . 7554/eLife . 20522 . 015Table 2 . Genome comparison of sequenced smut fungi . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 015U . bromivoraU . maydis1S . reilianum2S . scitamineum3U . hordei4M . pennsylvanicum5Assembly statisticsTotal contig length ( Mb ) 19 . 718 . 219 . 520 . 619 . 2Total scaffold length ( Mb ) 20 . 519 . 818 . 419 . 621 . 1519 . 2Average base coverage154x10x20x30x25x339xN50 contig ( kb ) 127 . 450 . 337 . 648 . 743 . 4N50 scaffold ( kb ) 877817 . 8738 . 5759 . 2307 . 7121 . 7Chromosomes23232323GC-content ( % ) 52 . 45459 . 754 . 45250 . 9 coding ( % ) 54 . 456 . 362 . 657 . 854 . 354 non-coding ( % ) 49 . 450 . 554 . 351 . 143 . 446 . 9Coding sequencePercent coding ( % ) 59 . 861 . 165 . 96257 . 556 . 6Average gene size ( bp ) 169918361858181917051734Average gene density ( gene/kb ) 0 . 350 . 340 . 360 . 340 . 330 . 33Protein-coding genes723367866648669371136279Exons111549783977610214109079278Average exon size11011230122111911107Exons/gene1 . 51 . 441 . 471 . 51 . 531 . 48tRNA genes13311196116110126SecretomePredicted secreted proteins409485461466405300Non-coding sequenceIntrons392129973103352131612999Introns/gene0 . 540 . 440 . 460 . 530 . 440 . 48Average intron length ( base ) 163142144130 . 1141191 . 4Average intergenic distance ( bp ) 10541127929111411861328Repeat sequencesDNA Transposon1 . 89%0 . 29%0 . 13%0 . 25%0 . 89%0 . 29%LINE4 . 38%0 . 35%0 . 04%0 . 27%4 . 62%0 . 40%SINE0 . 18%0 . 05%0 . 03%0 . 05%0 . 27%0 . 10%LTR Retrotransposon5 . 83%1 . 15%0 . 13%0 . 69%4 . 82%1 . 17%Unclassified non LTR-Retrotransposon0 . 06%0 . 02%0 . 01%0 . 01%0 . 10%0 . 032Unclassified Retrotransposon2 . 03%0 . 21%0 . 12%0 . 29%1 . 47%0 . 39%Unclassfied0 . 06%0 . 08%0 . 02%0 . 08%0 . 38%0 . 04%Total TE class14 . 33%2 . 11%0 . 45%1 . 60%11 . 84%2 . 32%Simple sequence repeats1 . 31%1 . 75%2 . 00%1 . 59%1 . 59%1 . 54%Total excl . Tandem repeats15 . 72%3 . 90%2 . 49%3 . 23%13 . 56%3 . 95%Tandem repeats5 . 14%4 . 22%6 . 97%4 . 54%5 . 20%5 . 16%Total repeat coverage18 . 51%6 . 70%8 . 26%6 . 68%16 . 45%6 . 72%1Kämper et al . , 2006; 2Schirawski et al . , 2010; 3Dutheil et al . , 2016; 4Laurie et al . , 2012; 5Sharma et al . , 2014 To assess the completeness of the U . bromivora genome , a BLAST search was performed with highly conserved core genes present in higher eukaryotes ( Aguileta et al . , 2008; Parra et al . , 2009 ) . From the expected 248 single-copy orthologs extracted from 21 genomes ( Parra et al . , 2009 ) , 247 are present in the U . bromivora genome ( missing KOG1468 , translation initiation factor eIF-2B ) , indicating that >99% of the gene space is covered by the assembly . To assess the difference of UB1 and its compatible mating partner UB2 on a genomic level , we performed Illumina 125 paired-end sequencing of UB2 and single nucleotide polymorphism ( SNP ) calling . By using stringent parameters , we identified 1 , 323 SNPs between the two strains ( Figure 7—source data 1 ) . 783 of them are found in coding sequences and 429 lead to non-synonymous mutations . Since UB1 and UB2 were independently isolated from different spores , SNPs might be a result of the prevalent inbreeding due to intratetrad mating which over time leads to significant differences between progenies of different spore tetrads . We constructed a phylogenetic tree using one-to-one orthologous genes as identified by OrthoMCL ( Figure 6 ) . The resulting phylogeny shows Sporisorium reilianum and Sporisorium scitamineum forming one cluster and U . bromivora and U . hordei forming a second cluster . U . maydis shows a closer relationship to the tested species of the Sporisorium genus than to the ones of the Ustilago genus that are more closely related to Melanopsichium pennsylvanicum . This is in agreement with the relationships found by Stoll et al . ( 2005 ) and Sharma et al . ( 2014 ) . The phylogeny also illustrates the close relationship between U . hordei and U . bromivora , which share orthologs for a large number of genes and between which protein sequence conservation is much higher than between other fungi ( data not shown ) . 10 . 7554/eLife . 20522 . 016Figure 6 . Phylogeny of U . bromivora and related smuts . Unrooted phylogeny created from 4 , 947 one-to-one orthologs . Branch lengths represent the mean number of substitutions per DNA site . Terminal branch lengths for U . bromivora and U . hordei are not to scale . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 016 The mating type locus of U . bromivora UB1 is located on chromosome 1 . We identified genes encoding a putative pheromone receptor ( UBRO_03901 ) and pheromone ( UBRO_03899 ) as well as bEast ( UBRO_00885 ) and bWest ( UBRO_00887 ) orthologs encoding a putative homeodomain-transcription factor ( Figure 7A ) . The pheromone/receptor genes ( a locus ) are separated from the bEast and bWest genes ( b locus ) by a 183 kb long region highly enriched in transposable elements ( TE; 39 . 85% compared to an average of 17 . 18% over the length of the chromosome and to an overall genome average of 14 . 33%; Figure 7B , Figure 7—figure supplement 1 ) . This bipolar mating system resembles the structural organization found in the close relative U . hordei although the region between the a and b locus in U . hordei is , at ~500 kb , larger ( Figure 7—figure supplement 2 ) ( Bakkeren and Kronstad , 1994 ) . 10 . 7554/eLife . 20522 . 017Figure 7 . The mating type chromosome ( chromosome 1 ) of U . bromivora UB1 ( MAT-1 ) and UB2 ( MAT-2 ) . ( A ) Schematic representation of the a locus ( encoding the predicted pheromone receptor system ) and the b locus ( encoding the putative heterodimeric transcription factor for pathogenic development ) of UB1 ( MAT-1 strain ) and UB2 ( MAT-2 strain ) according to de novo assembly of both strains . Potential rga2 and lga2 orthologs , encoding proteins for uniparental mitochondrial inheritance ( Fedler et al . , 2009 ) , are located in the a2 region between mfa2 and pra2 . *Due to rearrangements in the mating type region of UB2 , the orientation of a2 and b2 locus could not be exactly determined . However , data suggest an inversion of the a2 locus . ( B ) The mating type region of UB1 is enriched for transposable elements . Graph depicts percentage of transposon coverage in 25 kb windows occurring every 12 . 5 kb along the chromosome . ( C ) Mapping of UB2 to the UB1 reference genome shows large non-mapped stretches in the MAT-1 locus indicating sequence differences in this region between MAT-1 and MAT-2 . ( D ) Enrichment of single nucleotide polymorphisms ( SNPs ) in and around the mating type region between the genomes of UB1 and UB2 . Number of SNPs is shown in 5 kb windows . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 01710 . 7554/eLife . 20522 . 018Figure 7—source data 1 . List of Single Nucleotide Polymorphisms ( SNPs ) identified in UB2 . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 01810 . 7554/eLife . 20522 . 019Figure 7—figure supplement 1 . Transposon content along U . bromivora chromosomes . The percentage length of transposons in 25 kb windows occurring every 12 . 5 kb along the chromosome is shown . The horizontal line shows the mean percentage of the length made up by transposons across the entire chromosome . The green line shows a loess regression with the shaded area representing the confidence interval at the 0 . 95 level . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 01910 . 7554/eLife . 20522 . 020Figure 7—figure supplement 2 . Genes of the mating type regions are up to 98% identical between U . bromivora and U . hordei . Genes of the a ( pheromone/receptor ) locus and the b ( heterodimeric transcription factor ) locus were bioinformatically identified in U . bromivora UB1 ( MAT-1 ) and UB2 ( MAT-2 ) and compared to respective orthologs of U . hordei . DNA-Sequence identity of two orthologs is shown between each orthologous pair in percent . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 020 Mapping of Illumina reads from UB2 to the UB1 reference genome and subsequent SNP calling showed that the mating type region flanked by the a and b loci is highly diverse between the two compatible strains , as evidenced by segments depleted of reads , rearrangements and a high number of SNPs in the regions covered ( Figure 7C , D; Figure 7—source data 1 ) . We calculated for the mating type region a SNP frequency of 1 . 8E-03 SNPs bp-1 compared to a frequency of 6 . 5E-05 SNPs bp−1 for the entire genome . This sequence divergence is likely a result of recombination suppression in the mating type region leading to a bipolar mating system at the cost of losing a recombination-based repair in this region . One class of proteins which is of special interest for biotrophic interactions is secreted proteins . In addition to functions such as cell wall remodeling and hydrolytic enzymes for substrate degradation , this class includes effector proteins , which might directly impact the outcome of the biotrophic interaction with the host and often target the host defense system or its metabolism ( Asai and Shirasu , 2015 ) . Several criteria had to be fulfilled for a predicted U . bromivora protein to be considered a secreted protein: it had to contain a signal peptide , fewer than two transmembrane helices , no endoplasmatic reticulum ( ER ) retention signal , and it had to be predicted not to target mitochondria . These strict criteria led to the prediction of 409 secreted proteins encoded in the U . bromivora genome . While for 46 . 3% of all U . bromivora proteins we could not assign a potential function based on BLAST sequence similarity ( 'unclassified proteins' ) , this fraction increases to 69 . 7% for the proteins that are predicted to be secreted . This makes proteins of unknown function the most significantly enriched functional category of U . bromivora secreted proteins ( Figure 8A , Figure 8—figure supplement 1 ) . 10 . 7554/eLife . 20522 . 021Figure 8 . Enrichment of classes of interest among predicted secreted proteins and in planta differentially expressed transcripts . ( A ) Proportions of genes encoding unclassified proteins based on FunCat classification within all annotated U . bromivora genes , within genes encoding secreted proteins , within all genes up- and downregulated in planta twelve days after planting ( dap ) as well as within the genes that are differentially regulated and encode secreted proteins . ( B ) Proportions of genes encoding predicted secreted proteins within all annotated U . bromivora genes , within all genes found to be expressed in axenic culture and in planta , and within genes significantly up- or downregulated in planta . Fisher exact test was used to test whether the proportion of selected genes within a given class differs significantly from the proportion within all annotated genes; ****p-value < 0 . 0001 , ***p-value < 0 . 001 . The total number of genes is shown in brackets below each chart . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 02110 . 7554/eLife . 20522 . 022Figure 8—figure supplement 1 . Functional categorization of putatively secreted proteins and all proteins encoded in the genome . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 022 To provide insights into expression patterns of the genes encoding putatively secreted proteins in U . bromivora during saprophytic growth and in planta , we conducted RNA-seq analyses . To this end , we isolated RNA from axenic UB1 culture and from stems of twelve day old B . hybridum Bd28 plants that were spore-inoculated with U . bromivora . Among the 6 , 756 transcripts found to be expressed in our dataset , 493 were significantly upregulated in planta compared to axenic culture ( logFC > 2 , adjusted p-value < 0 . 1 ) , while 1 , 138 transcripts were significantly downregulated . Notably , transcripts predicted to encode secreted proteins are significantly enriched among the upregulated transcripts compared to all annotated genes ( 30 . 8% compared to 5 . 7%; Fisher exact test , p-value ≤ 2 . 2E-16; Figure 8B ) . Moreover , we could show by functional protein classification that among the putatively secreted proteins that are induced in planta , unclassified proteins are significantly overrepresented ( 84 . 1% compared to 46 . 4% within all annotated genes; p-value = 2 . 2E-16; Figure 8A ) . In contrast , in the subset of down-regulated genes , a functional overrepresentation of genes encoding secreted , unclassified proteins could not be observed . Our findings are in line with the observation that the vast majority of effector proteins are so far functionally not characterized . At the sequence level these proteins are only poorly conserved between distant fungal pathogens making the prediction of protein function difficult . In the case of U . maydis , many of the small secreted protein encoding genes are organized in pathogenicity clusters and a large number of them were shown to play a role in virulence ( Kämper et al . , 2006 ) . To assess the presence of potential pathogenicity clusters in the U . bromivora genome , we defined clusters as containing at least three adjacent genes encoding predicted secreted proteins or three genes encoding secreted proteins interrupted by maximal one single non-secreted protein-coding gene . These in comparison to the analysis of Kämper et al . ( 2006 ) relatively relaxed criteria led to the identification of only eleven secretion clusters comprising a total of 10 . 51% of all putative secreted proteins of U . bromivora . Although some of the putative effector clusters identified in U . maydis also show a syntenic organization in the U . bromivora genome , many others do not and the vast majority of the predicted secreted proteins of U . bromivora are not clustered . Among all clusters , the largest is cluster 18 , related to U . maydis cluster 19 . Whereas cluster 19 ( as located on chromosome 19 ) in U . maydis comprises 24 putative effector genes ( Brefort et al . , 2014 ) , the syntenic cluster in U . bromivora comprises less than half the number of putative secreted protein encoding genes and is located on chromosome 18 ( Figure 9 ) . In the close relative U . hordei the clustering of genes encoding predicted secreted proteins in this region is even less compact and disrupted to a greater extent than in U . bromivora ( Figure 9 ) ( Laurie et al . , 2012 ) . Our analysis of the U . bromivora clusters further showed that the syntenic genes missing from cluster 18 have not moved to other parts of the genome but are simply absent from U . bromivora and occur only in U . maydis or are shared between U . maydis , S . reilianum , and S . scitamineum ( data not shown ) . 10 . 7554/eLife . 20522 . 023Figure 9 . Gene by gene comparison between the largest secreted virulence cluster of U . maydis ( cluster 19 ) with the corresponding region in the U . bromivora and U . hordei genome on chromosome 18 . The scheme depicts U . maydis cluster 19 on chromosome 19 , the predicted U . bromivora cluster 18 on chromosome 18 and the corresponding region on chromosome 18 ( scaffold 5 . 0017 ) of U . hordei . Syntenic orthologs between U . maydis and U . bromivora as well as between U . bromivora and U . hordei are connected with a green bar . Genes encoding predicted non-secreted proteins are displayed in grey , genes encoding putative secreted proteins in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 023 One indication of how suitable a specific model system is for generalizing knowledge to other related species is the presence of orthologous genes . Excluding mitochondrial genes , among the 7 , 193 genes present in the U . bromivora genome , 5 , 121 one-to-one orthologs were predicted with U . maydis , U . hordei , S . scitamineum , S . reilianum , and M . pennsylvanicum . These one-to-one orthologs are genes which have one ortholog in each of the other species and no paralogs . M . pennsylvanicum and U . bromivora share 5 , 470 orthologs . In contrast , U . bromivora shares 5 , 841 orthologs with U . maydis and 6 , 180 orthologs with U . hordei . The lower number of orthologs in M . pennsylvanicum is in line with the observation that this smut genome has lost genes that might be associated with a switch from a monocot to a dicot host plant ( Sharma et al . , 2014 ) . Among the 409 U . bromivora predicted secreted proteins , we found 216 of them among the 5 , 121 one-to-one orthologs . All of the U . maydis effectors functionally characterized or described in the literature , including Tin2 , See1 , Cmu1 , Pep1 , Pit2 and Mig1 ( Djamei et al . , 2011; Redkar et al . , 2015a , 2015b; Hemetsbergeret al . , 2012; Doehlemann et al . , 2009; Mueller et al . , 2013;Basse et al . , 2000; Tanaka et al . , 2014 ) , share orthologs with U . bromivora ( Figure 10A ) . The amino acid similarity between each U . maydis effector and its corresponding U . bromivora ortholog ranges from 41% for the organ specific effector See1 to 70% for the widely conserved peroxidase inhibitor effector Pep1 ( Figure 10A ) . To experimentally test if U . bromivora effectors can functionally complement their respective orthologs in the U . maydis / Zea mays pathosystem , we chose two conserved effectors for that assay , Stp1 and Pep1 ( Doehlemann et al . , 2009; Schipper , 2009 ) . These effectors were recently shown to play an essential role for virulence of U . maydis . The avirulent phenotype of both U . maydis deletion mutants could be complemented by introducing the corresponding U . bromivora ortholog ( Figure 10B ) . This demonstrates that the U . bromivora / Brachypodium system could be indeed suitable to study the functional role and host targets of conserved effectors from related species . 10 . 7554/eLife . 20522 . 024Figure 10 . Testing orthologs of core effectors for functional interchangeability between U . bromivora and U . maydis . ( A ) List of known effector orthologs in U . bromivora and U . maydis and their amino acid identity and similarity . Identity shows the percentage of identical positions in the alignment , taking gaps into account . Percentage identity = 100 ( identical positions / length of alignment ) . Similarity gives a measure of how similar two protein sequences are to one another based on the physical and chemical properties of their amino acids . Sequences were aligned using T-Coffee and identity and similarity scores were given by SIAS ( Sequence Identity and Similarity; http://imed . med . ucm . es/Tools/sias . html ) . ( B ) Core-effector mutants of U . maydis ( SG200∆stp1 and SG200∆pep1 ) can be complemented with the respective U . bromivora ortholog . Disease symptoms of infected plants were scored at twelve days post inoculation ( dpi ) according to Kämper et al . , 2006 . The darker the color , the more severe the symptoms . Numbers of infected plants are indicated above each column . p-values are calculated by Fisher exact test , MTC by Benjamini-Hochberg algorithm , ****p<0 . 0001 . Leaves of representative plants twelve days after inoculation with indicated strains are shown next to the stacked bar plot . Scale bars: 1 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 024 Besides the described orthologs , we found 427 orphan U . bromivora genes that lack orthologs in the five other fungi they were compared with . Of these , 21 are predicted to encode secreted proteins . None of the orphan genes had a predicted function according to either FunCat or Blast2GO . Comparing the currently available smut genomes both , U . hordei and , to an even greater extent , U . bromivora contain a high number of transposable elements and other repetitive sequences , encompassing up to 14 . 33% of the genome ( Table 2 ) . Similar to the other smuts sequenced to date , the U . bromivora genome has a low frequency of small interspersed nuclear elements ( SINEs ) ( 0 . 18% of the genome ) , a class of retrotransposons that lost the coding region for their own reverse transcriptase . However , it contains a high percentage of long terminal repeat ( LTR ) retrotransposons ( 5 . 83% ) , which are independent from other mobile genomic elements and encode all proteins necessary for transposition . Similar ratios of long interspersed elements ( LINE ) have spread in the U . hordei and the U . bromivora genome ( 4 . 62% and 4 . 38% , respectively; Table 2 ) . With few exceptions , such as the mating type chromosome ( Chr1 ) , transposable elements are evenly distributed over the U . bromivora chromosomes ( Figure 7—figure supplement 1 ) . These results are in line with those made in U . hordei , where except for the mating type region , repetitive sequences are also evenly distributed across the genome ( Laurie et al . , 2012; Bakkeren et al . , 2006 ) . While transposable elements have clearly shaped the genome of U . bromivora , their action might be counter-balanced by the presence of a functional core machinery of the RNA interference pathway . Homology searches identified UBRO_08937 that likely encodes the PAZ-domain ( Piwi/Argonaute/Zwille-domain ) containing endoribonuclease DICER . UBRO_20628 , UBRO_01631 , and UBRO_08874 encode three RNA-dependent RNA Polymerases ( RdRP ) which are necessary for the formation of the complementary strands of target RNA , and therefore represent a prerequisite for RNA-directed silencing . UBRO_06256 is predicted to encode the argonaute protein , the catalytic subunit of the RISC-complex . We also identified the small RNA methyltransferase Hen1 ( UBRO_08578 ) . Moreover , the chromodomain ( CD ) protein CHP1 and CHP2 important for heterochromatic gene silencing are with UBRO_05116 and UBRO_07750 as well present . Analysis of dinucleotide frequencies in repetitive regions of the genome shows a lack of CpG dinucleotides similar to that observed in U . hordei ( Laurie et al . , 2012 ) ( Figure 11 ) . This may indicate the presence of repeat induced point-mutations ( RIP ) which can serve as an additional genomic defense mechanism against transposable elements ( Selker , 2002 ) . 10 . 7554/eLife . 20522 . 025Figure 11 . U . bromivora shows a decrease in the occurrence of CpG dinucleotides . Dinucleotide frequencies in repeat regions were determined using RIPCAL and were compared to those of control regions . A noticeable decrease in the occurrence of CpG dinucleotides was detectable for both U . hordei and U . bromivora as well as , to a lesser extent , M . pennsylvanicum . DOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 025 Genes under positive selection are indicative of an ongoing adaptation process often found either upon a host jump , neofunctionalization of a protein , or during the arms race between the host and the pathogen . We analysed the proteins of our six fungal genomes for signs of positive selection . Our analysis of the one-to-one orthologs between the six smut fungi we compared ( Table 3 ) confirmed the previously reported finding of high levels of positive selection in M . pennsylvanicum ( Sharma et al . , 2014 ) . It also showed that U . bromivora has the lowest levels of selection when measured at a false discovery rate ( FDR ) of 0 . 05 but is between U . maydis and S . reillianum when using an FDR of 0 . 01 . Among the genes being under positive selection , only twelve encode predicted secreted proteins . In contrast to the twelve genes found in U . bromivora , M . pennsylvanicum , which adapts to its dicot host , harbors 103 genes encoding predicted secreted proteins that were shown to be under positive selection . The evolutionary pressures driving the observed incidents of selection in U . bromivora remain unclear . 10 . 7554/eLife . 20522 . 026Table 3 . Positive selection among sequenced smut fungiDOI: http://dx . doi . org/10 . 7554/eLife . 20522 . 026U . bromivoraU . maydisU . hordeiS . scitamineumS . reilianumM . pennsylvanicumGenes analysedTotal494749474947494749474947Under selection ( q = 0 . 05 ) 1403451881702562390Under selection ( q = 0 . 01 ) 9611613590911434Non-PSEPsTotal473847354733474847434753Under selection ( q = 0 . 05 ) 1283181681532402287Under selection ( q = 0 . 01 ) 8610512281841377PSEPsTotal209212214199204194Under selection ( q = 0 . 05 ) 1227201716103Under selection ( q = 0 . 01 ) 1011139757*Number of genes under positive selection in each of the six fungi used in this study . The number of genes predicted to be under positive selection is given for both a FDR of 0 . 05 and 0 . 01 and are grouped in three categories: total genes analyzed , the subset of genes that are not predicted to be secreted and the subset of genes that are predicted to be secreted . This analysis is limited to the 4 , 947 one-to-one orthologs that were used for the construction of the phylogeny . Smut fungi , especially U . maydis , have become models for studying recombination ( Holliday , 1964 ) , cell biology ( Steinberg et al . , 2008; Haag et al . , 2015 ) , and , due to their nature , biotrophic interactions ( Brefort et al . , 2009 ) . This has resulted in U . maydis being considered as an important model pathogen in the scientific community ( Dean et al . , 2012 ) . In recent years , the importance of small secreted molecules termed effectors , which shape the biotrophic interaction between the pathogen and its host , has become increasingly evident . One major challenge for effector research is that most effector proteins have no sequence similarity to any known protein and are therefore difficult to functionally characterize . Although these important molecules are produced by the pathogen , in many cases they target host processes and therefore require a completely accessible host system for complementary functional studies in both the host and pathogen . In our search to identify a biotrophic model pair for a smut and a temperate host grass that due to their genetic accessibility will enable these complementary functional studies , we chose the smut fungus U . bromivora and its compatible host grass Brachypodium sp . Although U . bromivora is closely related to other sequenced model smuts , it displays interesting peculiarities in its lifestyle , especially in connection with its mating system . Most strikingly is one major feature: the mating type bias . Upon spore germination and meiosis , the a and b loci located on chromosome 1 co-segregate , and progeny with two different mating types arise from one spore , MAT-1 and MAT-2 . Interestingly , the locus that causes the mating type bias co-segregates with the MAT-2 mating type region leading to the inability to survive under saprotrophic conditions . Independent of its cause , it has important implications for the biology of U . bromivora . The bipolar mating system and the intratetrad mating entail a strong tendency for inbreeding which could be an advantageous driving speciation of this highly specialized plant pathogen ( Hoekstra , 2005 ) . This holds especially true as the pheromone receptor based cell-cell recognition system is rather promiscuous between different smut species and might not be sufficient as a speciation barrier ( Kellner et al . , 2011 ) . Besides the removal of detrimental DNA , like transposable elements and deleterious mutations , sexual reproduction is a way to efficiently reshuffle alleles over generations to provide an opportunity for natural selection to produce efficient allele combinations ( Goddard et al . , 2005; Lee et al . , 2010 ) . The relative abundance of transposable elements in U . bromivora could potentially serve as a source of variation to counterbalance the assumed loss of heterozygosity due to inbreeding between progeny from the same spore . This could enable the population to adapt to evolving challenges such as host defense mechanisms . As a consequence , it might be essential to keep the functional machinery for RNA silencing intact to limit the uncontrolled spreading of transposable elements with potentially deleterious effects . In contrast to U . bromivora , U . maydis has evolved a very efficient recombination system enabling the removal of most of its invasive transposable elements . This was likely a prerequisite to allow the loss of the silencing machinery in U . maydis , which subsequently led to the gain of a competitive advantage through symbiosis with double-stranded RNA totiviruses , which encode for killer toxins that target non-killer containing competitive microbes ( Drinnenberg et al . , 2011; Koltin and Day , 1976 ) . The deleterious allele which leads to the mating type bias in U . bromivora is still unidentified . In the related smut Microbotryum violaceum ( formerly described as Ustilago violacea ) , intratetrad mating was observed as a result of a recessive haplo-lethal allele ( Hood and Antonovics , 2000 ) . In Ustilago nuda , a recessive proline biosynthesis allele linked to the mating type locus led to a similar result in the haploid stage of the dimorphic life cycle ( Nielsen , 1968 ) . Alternatively , a dominant mechanism via meiotic drive elements as described for Neurospora crassa ( Turner and Perkins , 1979 ) and other ascomycetes such as Gibberella fujikuroi ( Fusarium verticillioides ) ( Kathariou and Spieth , 1982 ) and Podospora anserina ( Bernet , 1967 ) could be causative for the observed mating type bias in U . bromivora . While , in this scenario , the ‘killer’-allele and ‘resistance’-allele would form a locus and would be linked to the recombination-suppressed MAT-1 region , the corresponding ‘Non-killer/susceptibility’ alleles would co-segregate with the MAT-2 region . The alleles of the MAT-2 strain would therefore lead to its haplo-lethality . Future research will clarify the cause of the mating type bias in U . bromivora . In summary , the newly established model system U . bromivora and Brachypodium sp . has tremendous potential for the study of biotrophy related questions on both the pathogen and host side as well as evolutionary questions such as sex and speciation . The high quality , manually curated fungal genome , available RNA-seq data , and growth as well as transformation protocols for both the host and the pathogen , provide a solid basis for scientists to gain new insights into biotrophic plant pathogen interactions .
DNA manipulation and plasmid generation were performed according to standard molecular cloning procedures ( Chong , 2001; Ausubel et al . , 1987 ) . All DNA manipulations were performed with E . coli MACH1 ( Thermo Fisher Scientific , Waltham , MA ) . Primers and plasmids are compiled in Supplementary file 1 . All sequences of plasmids created in this study are provided as gb/gbk files in Supplementary file 2 . Strains used in this study are listed in Supplementary file 1 . The Ustilago bromivora spore material used in this study was obtained from Thierry Marcel and originated from spontaneous repetitive infections which occurred in a greenhouse at INRA UMR BIOGER , Avenue Lucien Brétignières BP01 , 78850 Thiverval-Grignon , France ( Barbieri et al . , 2012 ) . U . bromivora UB1 ( formerly named UB2112 ) and UB2 were cultivated in Potato dextrose ( PD ) liquid medium ( 2 . 4% PD dissolved in H2O; Becton , Dickinson and Company , Franklin Lakes , New Jersey ) at 21°C , shaking at 200 rpm in baffled flasks . U . maydis and U . hordei were cultivated according to Kämper et al . ( 2006 ) and Laurie et al . ( 2012 ) . U . maydis strains were generated by gene replacement via homologous recombination as described by Kämper ( 2004 ) or by insertion of p123 derivatives into the ip locus ( Loubradou et al . , 2001 ) . U . maydis virulence assays were performed as described by Kämper et al . ( 2006 ) . In brief , the solopathogenic strain SG200 and its derivatives were cultivated in YepsLight ( 0 . 4% yeast extract , 0 . 4% peptone , 2% sucrose ) at 28°C , under continuous shaking ( 200 rpm ) , until they reached an OD600 nm of 0 . 8 . After centrifugation at 2400 g for 5 min , cultures were adjusted in H2Odd to an OD600 nm = 1 . The suspensions were subsequently used for syringe-inoculation of seven day old maize seedlings ( variety Early Golden Bantam ) . Infection symptoms were scored twelve days post infection employing the scoring system described by Kämper et al . ( 2006 ) . Genomic DNA ( gDNA ) extraction was performed as previously described for U . maydis ( Kämper et al . , 2006 ) . U . bromivora cultures were grown to an exponential phase and subjected to Phenol-Chloroform extraction . For Single Molecule Real-Time ( PacBio ) sequencing of UB1 , Phenol-Chloroform extraction was followed by an additional purification step via the Power Clean DNA Kit ( MO BIO Laboratories , Carlsbad , CA ) . For Illumina sequencing of UB2 , Phenol-Chloroform extracted gDNA was purified using the MasterPure Complete DNA and RNA Purification Kit ( Epicentre , Madison , WI ) . The SMRT bell was produced using the DNA Template Prep Kit 1 . 0 ( Pacific Biosciences , Menlo Park , CA ) . The input genomic DNA concentration was measured using a Qubit Fluorometer dsDNA Broad Range assay ( Thermo Fisher Scientific , Waltham , MA ) . 10 μg of gDNA was mechanically sheared to an average size distribution of 15 kb , using a Covaris gTube ( Kbiosciences , Hoddesdon , UK ) . A Bioanalyzer 2100 12K DNA Chip assay ( Agilent , Santa Clara , CA ) was used to assess the fragment size distribution . 5 μg of sheared gDNA was DNA damage repaired and end-repaired using polishing enzymes . A blunt end ligation reaction followed by exonuclease treatment was performed to create the SMRT bell template . A Blue Pippin device ( Sage Science , Beverly , MA ) was used to size select the SMRT bell template and enrich for large fragments > 10 kb . The size selected library was quality inspected and quantified on an Agilent Bioanalyzer 12 kb DNA Chip and on a Qubit Fluorimeter ( Thermo Fisher Scientific , Waltham , MA ) , respectively . A ready to sequence SMRT bell-Polymerase Complex was created using the P6 DNA/Polymerase binding kit 2 . 0 ( Pacific Biosciences , Menlo Park , CA ) according to the manufacturer’s instructions . The Pacific Biosciences RS2 instrument was programmed to load and sequence the sample on 5 SMRT cells ( v3 . 0; Pacific Biosciences , Menlo Park , CA ) , taking 1 movie of 240 min each per SMRT cell . A MagBead loading ( Pacific Bioscience , Menlo Park , CA ) method was chosen in order to improve the enrichment of longer fragments . After the run , a sequencing report was generated for every cell via the SMRT portal , in order to assess the adapter dimer contamination , the sample loading efficiency , the obtained average read-length , and the number of filtered sub-reads . The genome was assembled with Pacific Bioscience’s SMRTanalysis software version 2 . 2 . 0 and the hierarchical genome-assembly process ( HGAP ) v3 protocol ( Chin et al . , 2013 ) including polishing with Quiver . Default settings were used , except for selecting only reads with a minimum length of 10 , 000 bp and read quality above 0 . 8 . Assembly and polishing resulted in 25 contigs with an overall length of 20 . 7 Mb . U . bromivora UB1 was grown in axenic culture ( 21°C , 200 rpm , PD medium ) in three independent biological replicates to an exponential phase ( OD600 nm = 0 . 8 ) . RNA was extracted using the TRIzol method ( Chomczynski and Sacchi , 2006 ) according to the manufacturer’s protocol ( Thermo Fisher Scientific , Waltham , MA ) . Residual DNA was removed with the DNA-free Kit ( Thermo Fisher Scientific , Waltham , MA ) . The extracted and purified RNA was used for library generation with the NEB Next Ultra RNA Library Prep Kit according to the manufacturer’s protocol ( cutout size 200–800 bp; New England Biolabs , Ipswich , MA ) and was sequenced with an Illumina HiSeq2000 instrument in paired-end 100 mode . The resulting data was pooled to yield 68M read pairs . The overall quality metrics were verified with fastqc ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc ) . The pooled reads were then assembled using Trinity ( vr20140413p1 ) using the built-in trimmomatic quality trimming , in-silico read normalization and jaccard clipping procedures ( Grabherr et al . , 2011 ) . The resulting transcriptome assembly consisted of 11 , 918 contigs with an N50 of 3 , 027 bp . Primary structural annotation was achieved by mapping the protein sequences of U . hordei on the scaffolds using exonerate ( v2 . 2 . 0 ) unless the protein sequences could not be mapped ( Slater and Birney , 2005 ) . As a de novo gene predictor , GeneMark-ES version 2 was applied ( Ter-Hovhannisyan et al . , 2008 ) . In addition , the orthologous protein sequences of U . maydis , S . reilianum and U . hordei were inspected by multi T-Coffee ( v8 . 69 ) alignments to further validate the gene structure in U . bromivora ( Notredame et al . , 2000 ) . As transcriptional evidences , RNA-seq reads were mapped on the genome using tophat2 ( v2 . 0 . 8 ) . The interval for allowed intron lengths was set to a minimum of 20 nt and a maximum of 1 kb ( Langmead et al . , 2009; Trapnell et al . , 2012 ) . The Trinity assembled RNA reads were mapped as transcripts ( Grabherr et al . , 2011 ) . The different gene structures and supporting evidence were displayed in GBrowse ( Donlin , 2009 ) , allowing manual validation and correction of all coding sequences . The final call set comprises 7 , 233 protein coding genes . In addition , 133 tRNA-encoding genes are predicted using tRNAscan-SE ( Lowe and Eddy , 1997 ) . The protein coding genes were analyzed and functionally annotated using the PEDANT system ( Walter et al . , 2009 ) , accessible at http://pedant . helmholtz-muenchen . de/genomes . jsp ? category=fungal . The genome and annotation were submitted to the European Nucleotide Archive ( http://www . ebi . ac . uk/ena ) under the study number PRJEB7751 . The predicted protein set was searched for highly conserved single ( low ) copy genes to assess the completeness of the sequence dataset . Orthologous genes to all 246 single copy genes were identified by BLASTP comparisons ( eValue: 10−3 ) against the single-copy families from all 21 species available from the FunyBASE ( Aguileta et al . , 2008 ) . Additionally , 247 of 248 core-genes commonly present in higher eukaryotes ( CEGs ) could be identified by BLASTP comparisons ( eValue: 10−3 ) ( Parra et al . , 2009 ) . For differential expression analysis , RNA from axenic culture of UB1 ( see 'Transcriptome assembly' ) and from seedlings , 12 days after planting of germinating caryopses that were incubated with spore material for 1 week at 4°C , was isolated using the TRIzol method and DNA was removed with the DNA-free Kit ( Thermo Fisher Scientific , Waltham , MA ) . After library preparation ( cutout size: 200–800 bp; NEB Next Ultra RNA Library Prep Kit; New England Biolabs , Ipswich , MA ) three independent biological replicates were sequenced with an Illumina HiSeq2000 instrument , paired-end 100 bp . RNA-Seq was quantified against the combined transcriptome extracted from Brachypodium distachyon Bd21 ( Bdistachyon_283_v2 . 1 ) and Ustilago bromivora UB1 annotations with kallisto using default parameters ( Bray et al . , 2016 ) . In our dataset we identified 20 cases , where more than one splicing variant encoded by the same gene was present . In all subsequent analyses , splicing variants were treated and counted as individual genes . Differential expression statistics between axenic and in planta samples were computed using the DeSeq2 R package under the assumption that the overall expression level is similar between the two samples ( Love et al . , 2014 ) . Transcripts were considered significantly up- or downregulated in planta , if the log2fold-change compared to axenic culture was ≥2/≤−2 and the Benjamini-Hochberg ( Hochberg and Benjamini , 1990 ) corrected p-value was ≤0 . 1 . Over-/underrepresentation of individual functional classes of interest ( e . g . predicted secreted proteins ) among the in planta up- and downregulated transcripts was tested by Fisher exact test in the R statistical environment ( Core Team , R , 2011 ) . Systematic over-/underrepresentation analysis for all functional classes present in the FunCat annotation of the given dataset was conducted with the FunCat workflow ( Ruepp et al . , 2004 ) . Expression data were submitted to GeneExpressionOmnibus ( http://www . ncbi . nlm . nih . gov/geo/ ) under the accession number GSE87751 . A more detailed description of the methodology can be found on Bio-Protocol ( Czedik-Eysenberg et al . , 2017 ) . To predict putative secreted proteins , protein sequences were retrieved from the PEDANT 3 Genome Database and analysed for specific features that are associated with secreted proteins . We used SignalP ( v4 . 0 ) ( Petersen et al . , 2011 ) to predict the existence of a signal peptide , TMHMM ( v2 . 0c ) ( Sonnhammer et al . , 1998; Krogh et al . , 2001 ) to predict the existence of transmembrane domains , Phobius ( v1 . 01 ) ( Kall et al . , 2004 ) to detect both signal peptides and transmembrane domains , TargetP ( v1 . 1b ) ( Emanuelsson et al . , 2000 ) to predict the final location of a protein , and ScanProsite ( Gattiker et al . , 2002 ) to detect the presence of the ER retention motif [KRHQSA]-[DENQ]-E-L . A list of putative secreted proteins was generated that met the criteria of ( 1 ) signal peptide predicted by both SignalP and Phobius , ( 2 ) fewer than two transmembrane domains predicted by both TMHMM and Phobius , ( 3 ) no ER retention motif and ( 4 ) not predicted to target the mitochondrion . Orthologs were detected using OrthoMCL ( v2 . 0 . 9 ) using default settings ( Fischer , 2011 ) . OrthoMCL takes a list of proteins as an input and was provided with the full proteomes of U . maydis , U . bromivora , U . hordei , S . scitamineum , S . reilianum , and M . pennsylvanicum . Orthologous proteins from the different genomes are then sorted into groups . OrthoMCL makes use of the MCL algorithm ( Enright et al . , 2002 ) . Only nuclear genes were used for the prediction of orthologous relationships . Protein alignments were performed in T-Coffee ( v11 . 00 . 8cbe486 ) using the default settings . BLAST searches were performed with the stand-alone BLAST+ suite ( Camacho et al . , 2009 ) where possible . However , some programs required the older C legacy toolkit . We used RepeatScout ( Price et al . , 2005 ) for the de novo identification of repeat families in combination with the RepBase database ( Jurka et al . , 2005 ) to detect previously published transposable elements , pseudogenes , and retroviruses . The combined library of de novo and RepBase repeats were used to identify individual repeat elements on the genome using RepeatMasker ( Smit et al . , 2010 ) . All determined repeat elements were classified using TEclass ( Abrusan et al . , 2009 ) and analyzed for fingerprints of repeat-induced point mutations ( RIP ) regarding overrepresented dinucleotide frequencies using RIPCAL ( Hane and Oliver , 2008 ) . 5 , 121 genes were predicted as one-to-one orthologs , that means that there was an ortholog detected in each of the six fungal species and there was only a single copy in each genome . After removal of genes that were unsuitable for the positive selection analysis like genes without one-to-one orthologs or which had more than one predicted transcript in one or more species , 4 , 947 sequences remained . The ORF sequences were converted to amino acid sequences with T-Coffee ( v11 . 00 . 8cbe486 ) , aligned and then converted back into nucleotide sequences . These nucleotide alignments were concatenated to form a single alignment which was used as the input for RAxML ( v8 . 1 . 16 ) ( Stamatakis , 2014 ) . The tree was generated using the rapid bootstrapping and best-scoring maximum likelihood tree algorithm , GTRGAMMA nucleotide model and 1000 bootstraps . The individual alignments and the unrooted species tree were used as inputs for the codeml program from PAML 4 . 8A . ( Yang , 1997 ) . Two control files were used , both , allowing two or more dN/dS ratios for branches , checking for positive selection and having an initial omega value of 1 . One allowed the omega value to vary while the second kept it constant . The likelihood ratios under the two models were compared for each gene with the formula ΔLRT = 2 × ( lnL1 – lnL0 ) . The resulting value could be assessed using the chi-squared distribution with 1 degree of freedom to determine if the gene was under positive selection or not . Raw p-values were adjusted to take into account the false discovery rate using the qvalue R package ( http://qvalue . princeton . edu/ , http://github . com/jdstorey/qvalue ) ( Storey , 2015 ) . Genomic DNA of UB2 was used for library generation ( insert size 600–900 bp; NEBNext Ultra DNA Library Prep Kit for Illumina; New England Biolabs , Ipswich , MA ) and sequenced with an Illumina HiSeq2500 instrument in 125 bp paired-end mode . Reads were mapped to the genome of UB1 using CLC Genomics Workbench ( v . 7 . 0 . 3; Qiagen , Hilden , Germany ) with the following parameters: mismatch cost = 2 , insertion cost = 3 , deletion cost = 3 , length fraction = 0 . 5 , similarity fraction = 0 . 8 , no global alignment . SNP calling was performed via the quality-based variant detection mode with the following parameters: neighbourhood radius = 4 , maximum gap and mismatch count = 2 , minimum central quality = 10 , non-specific matches , broken pairs and variants in non-specific regions were ignored , minimum coverage = 90 , minimum variant frequency = 95% , maximum expected alleles = 2 , maximum coverage = 203 , sufficient variant count = 85 , required variant count = 85 . For calling SNPs the presence in both forward and reverse reads was a requirement . Heterozygous SNPs as well as small insertions and deletions were not considered . Moreover , the mitochondrial genome was not included in the analysis . De novo assembly of the MAT-2 strain UB2 was performed with SOAPdenovo2 ( Luo et al . , 2012 ) ( 127mer version 1 . 4 . 10 ) with kmer lengths ranging from 43 to 115 in steps of 6 with the following parameters: max_rd_len = 120 , avg_ins = 470 , asm_flags = 3 , rd_len_cutoff = 120 , rank = 1 , pair_num_cutoff = 3 , map_len = 32 . While assemblies with kmer lengths 73–91 performed good at diverse metrics , kmer 91 , at 4 , 712 bp and 8 , 160 bp , had the best mean size for both contigs and scaffolds , respectively , and the highest number of contigs larger than 100 kb . The resulting assembly ( kmer 91 ) had a contig and scaffold N50 of 23 , 058 bp and 113 , 827 bp . Scaffolds obtained after de novo assembly as well as the raw sequencing reads were submitted to the European Nucleotide Archive ( http://www . ebi . ac . uk/ena ) under the study number PRJEB7751 . Infected spikelets were dehusked and ground with a pistil in a 1 . 5 ml microcentrifuge tube . After the addition of 250 µl water , grinding continued until a black spore suspension was visible . The spore suspension was subjected to CuSO4 treatment to kill vegetative fungal cells and bacteria . CuSO4 was added to the suspension at a final concentration of 1 . 5% , solution and spores were incubated for 15 min , and washed 3 times with water to remove all traces of CuSO4 . CuSO4-treated spore material was resuspended either in 500 µl of water supplemented with 100 µg ml−1 ampicillin , 50 µg ml−1 tetracycline , and 25 µg ml−1 chloramphenicol and plated on PD agarose plates ( 2 . 4% PD supplemented with 2% agarose ) in serial dilutions for the isolation of haploid progeny or resuspendend in 1 ml of the above-mentioned antibiotic solution and spotted on objective slides for microscopy . A more detailed description of this method can be found on Bio-Protocol ( Bosch and Djamei , 2017 ) . Cell walls of fungal hyphae were stained with the chitin specific Wheat Germ Agglutinin - Alexa Fluor 488 conjugate dye ( WGA-AF 488; Thermo Fisher Scientific , Waltham , MA ) . Plant membranes were stained with FM4-64 ( Thermo Fisher Scientific , Waltham , MA ) . Staining and confocal laser scanning microscopy was performed as previously described by Doehlemann et al . ( 2008 ) . For DAPI and WGA staining of germinating spores and sporidia , germinating spores/sporidia were pelleted and fixed by incubating them with acetone for 15 min . Fixation was followed by 10 min incubation with WGA-AF 488 to visualize fungal cell walls and 15 min incubation with 1 µg ml−1 DAPI solution ( Sigma-Aldrich , Taufkirchen , Germany ) to stain nucleic acid . Cell/spore pellets were resuspended in PBS and subjected to confocal laser scanning microscopy . Images were acquired with a LSM780 Axio Observer confocal laser scanning microscope ( Zeiss , Jena , Germany ) . WGA-AF excitation was at 488 nm and detection at 517–552 nm , DAPI excitation at 405 nm and detection at 421–508 nm . CuSO4-treated spore suspensions were spotted on objective slides ( µm dish 35 mm; IBIDI , Planegg , Germany ) . After freezing them for 20 min on a metal block to avoid spreading of the liquid , they were covered with a thin ( <2 mm ) layer of PD agar or 1 . 5% water agar , respectively . After a 15–17 hr incubation at 21°C , the picture acquisition was performed with an inverted microscope ( Zeiss Axiovert 200 M with 40×/1 . 3 Plan-Neofluar Oil objective ) . Picture acquisition occurred every 20 min . Image processing was done with Fiji Imaging software . The transformation protocol was adapted from Schulz et al . ( 1990 ) and Gillissen et al . ( 1992 ) . U . bromivora cells were grown to an OD600 nm of 0 . 3–0 . 6 . Cells were harvested at 2400 g ( 10 min , RT ) , and washed once with Mg2+-MES buffer ( 20 mM MES buffer , pH 5 . 8 , 1 M MgSO4 ) . The Pellet was resuspended in 1 ml Mg2+-MES buffer containing 10 mg ml−1 Glucanex ( Sigma-Aldrich , Taufkirchen , Germany ) and 5 mg ml−1 Yatalase ( Takara Bio , Saint-Germain-en-Laye , France ) and kept on ice . Protoplastation was monitored and stopped by addition of 10 ml ice-cold Mg2+-MES buffer when 30–40% of the cells appeared round due to the loss of the cell wall . Cells were washed 3 times ( 2400 g , 10 min , 4°C ) in Mg2+-MES and once in STC buffer ( STC: 100 mM CaCl2 , 10 mM Tris-HCL pH 7 . 5 , 0 . 9 M sorbitol ) . Protoplastation was followed by PEG-mediated transformation . To this end , 5 µg plasmid DNA and 1 µl 100 mg ml−1 Heparin were added to 100 µl protoplasts and incubated for 30 min on ice . 500 µl STC-PEG ( 40% PEG4000 in STC buffer ) were added and mixed by pipetting gently up and down . After 15 min incubation on ice , the mixture was plated on PD regeneration agar plates composed of an lower layer of PD regeneration agar with twice the concentration of antibiotic ( 2 . 4% PD , 0 . 9 M sorbitol , 1 . 5% agar supplemented with either 4 µg ml−1 Carboxin , 200 µg ml−1 Geneticin G418 , or 200 µg ml−1 Hygromycin B ) overlaid by an antibiotic-free layer of PD regeneration agar . Colonies were visible after 10 to 14 days . For stable genomic integrations , restriction enzyme mediated transformation was performed according to Bölker et al . ( 1995 ) . In brief , 25 U of the restriction enzyme XbaI and XbaI-linearized p123UB-GFP were added to protoplasts and the aforementioned transformation protocol was followed immediately after addition of enzyme and plasmid . Caryopses of B . distachyon accession Bd21 were kindly provided by Phillipe Vain ( Vain et al . , 2008 ) , caryopses of accession ABR4 by John Vogel . For production of donor material , caryopses of each accession were gas-sterilized and transferred to Ø = 10 cm pots with a 4:1 mixture of standard potting soil ( Einheitserde Werkverband e . V . , Sinntal-Altengronau , Germany ) and perlite ( Granuflor , Vechta , Germany ) . Germination and early plant growth took place in a phyto chamber ( Johnson Controls , Milwaukee , WI ) under the following conditions: 20 hr light ( 150 µE ) , 24°C; 4 hr dark , 18°C; 60% humidity . After 10 days , pots were transferred to a cold room to vernalize plants for 6 weeks at 4°C , 40 µE , 13 hr light period . Pots were then moved back to the phyto chamber to allow plants to flower . Four weeks after anthesis , plants were transferred to the greenhouse for caryopses maturation and drying . Harvested caryopses were stored at 4°C in the dark . Caryopses within husks were gas sterilized ( Clough and Bent , 1998 ) . After sterilization , caryopses were moistened with a few ml of sterile water without submerging the caryopses and incubated in the dark at 4°C for 1 week to germinate . The seedlings were then moistened with a spore/water suspension for infection for one additional week at 4°C in the dark . Subsequently the seedlings were potted in a mixture of 3:1:1:1 standard potting soil ( Einheitserde Werkverband e . V . , Sinntal-Altengronau , Germany ) : perlite ( Granuflor , Vechta , Germany ) : silica sand ( min2C GmbH , Melk , Austria ) : germination soil ( Neuhaus ‘Huminsubstrat’ , Klasmann-Deilmann GmbH , Geeste , Germany ) , at a depth of ~1 cm , so that the emerged shoot remained exposed to the air and light . After 10 days in the growth chamber ( 20 hr light ( 150 µE ) , 24°C; 4 hr dark , 18°C; 60% humidity ) the pots were transferred to 4°C , 40 µE , 13 hr light period for 3 weeks ( Bd28 ) or 6 weeks ( ABR4 ) , for vernalization and subsequently returned to the growth chamber . 4–5 weeks later , the infected , spore filled spikelets were visible . U . bromivora spore material was surface sterilized and germinated on PD plates ( see spore recovery and sterilization ) . After spore germination , all derived colonies were pooled and used as an inoculum for further proliferation in liquid PD medium for 24 hr to ensure that only strains , which are viable in axenic culture , would grow . These mixtures were used as an infection inoculum of 300 vernalized B . hybridum caryopses ( Bd28 ) . After six weeks the infected spikelets were harvested , derived spores were surface sterilized and plated on solid PD medium . Colonies derived from single spores were singled out and tested by diagnostic PCR for the presence of pra2 . To infect B . hybridum Bd28 with a pair of compatible mating partners , both strains were grown until they reach the exponential phase ( OD600 nm = 0 . 8 ) and set to an OD600 nm = 2 with H2Odd . Both strains were mixed in equal amounts . One or two days before inoculation , vernalized B . hybridum caryopses were placed to RT to promote germination . Upon inoculation with the fungal mixture , the onset of coleoptiles should be visible . Seedlings were moistened with the fungal mixture and incubated in an appropriate tube , e . g . 2 ml microcentrifuge tube , at 21°C in the dark . After 24 hr incubation , seedlings were potted . Mature embryos from dormant caryopses were used for callus culture , plant regeneration , and transformation . In brief , dry dormant caryopses were surface-sterilized for 45 min with 6% NaClO including 0 . 03% Tween20 . After surface-sterilization , caryopses were rinsed five times with sterile tap water . Embryo preparation was carried out in two steps . First , the embryo was cut away from the endosperm . Second , the embryo was cut in a longitudinal direction into two pieces . Forty bisected embryos were transferred to a Ø = 10 cm petri dish with longitudinal wound in direct contact with the callus induction medium ( CIM ) . Callus induction of B . hybridum Bd28 took place on CIM ( 4 . 3 g l−1 MS No . 4 ( Murashige and Skoog medium modification No . 4; Duchefa Biochemie , Haarlem , The Netherlands ) , 30 g l−1 maltose , 11 . 1 µM 2 , 4-D , 2 mM NH4NO3 , 1 . 9 mM MES-monohydrate , 1x B5 vitamin mixture ( Duchefa Biochemie , Haarlem , The Netherlands ) , 3 . 5 g l−1phytagel ( Sigma-Aldrich , Taufkirchen , Germany ) , pH 5 . 8 . Before transformation , bisected embryos were pre-cultured for 6–8 weeks at 24°C without light . Every 14 days , calli were transferred to fresh CIM . Developing roots and shoots were cut away . Agrobacterium tumefaciens strain AGL1 was used for transformation ( Lazo et al . , 1991 ) . The plant transformation vector p6U contains a hygromycin phosphotransferase gene driven by the Zea mays ubiquitin promoter to confer hygromycin resistance to transformed plant cells . For B . hybridum Bd28 transformation , AGL1 harboring the respective p6U derivative was cultivated at 28°C and 210 rpm in Erlenmeyer flasks containing 10 ml MG/L medium ( Jones et al . , 2005 ) supplemented with 100 µg ml−1 Carbenicillin , 50 µg ml−1 Rifampicin , and 100 µg ml−1 Spectinomycin . After 22 hr , 500 µM Acetosyringone ( Sigma-Aldrich , Taufkirchen , Germany ) was added and the p6U-containing AGL1 strain was cultivated for additional 2 hr . For inoculation of pre-cultured B . hybridum calli , the A . tumefaciens culture was diluted with infection solution ( 4 g l−1 Chu ( N6 ) minerals ( Duchefa Biochemie , Haarlem , The Netherlands ) , 6 . 75 µM 2 , 4-D , 36 g l−1 glucose , 68 . 4 g l−1 sucrose , 0 . 7 g l−1 , 6 mM L-proline , 1x Chu ( N6 ) vitamin mixture ( Duchefa Biochemie , Haarlem , The Netherlands ) , 500 µM Acetosyringone ( Sigma-Aldrich , Taufkirchen , Germany ) , pH 5 . 2 ) to OD550 nm = 0 . 8 . 6–8 weeks after callus induction , forty calli were transferred into 15 ml tubes and inoculated with A . tumefaciens suspension . After 30 min , the A . tumefaciens suspension was discarded , calli were transferred to sterile filter paper and dried for 30 min . Calli were then transferred to Co-culture-medium ( 2 g l−1 Chu ( N6 ) minerals , 34 . 2 g l−1 sucrose , 2 mM CaCl2 , 20 µM Dicamba , 25 mM L-proline , 2 . 3 mM MES-monohydrate , 3 . 3 mM L-cysteine , 1x B5 vitamin mixture ( Duchefa Biochemie , Haarlem , The Netherlands ) , 500 µM Acetosyringone ( Sigma-Aldrich , Taufkirchen , Germany ) , 3 . 5 g l−1 phytagel , pH 5 . 8 ) and co-cultivated for 3 days , at 21°C without light . After co-cultivation , calli were transferred to CIM supplemented with 300 mg l−1 Timentin ( Duchefa Biochemie , Haarlem , The Netherlands ) for a 5 days resting phase ( counterselection of agrobacteria ) at 24°C without light . After the resting phase , calli were transferred to CIM supplemented with 300 mg l−1 Timentin and 50 mg l−1 Hygromycin B ( Sigma-Aldrich , Taufkirchen , Germany ) for an 8–12 week selection phase at 24°C without light . During that time , calli were transferred to fresh selection medium every two weeks . After these 8–12 weeks of selection , surviving calli were transferred to regeneration medium ( K4N ) ( Kumlehn et al . , 2006 ) including 150 mg l−1 Timentin and 25 mg l−1 Hygromycin B and cultivated for 8–12 weeks at 25°C with a 16/8 hr ( light/dark ) photoperiod . Regenerating plantlets from independent calli were transferred to pots with a substrate mixture of 3:1:1:1 Einheitserde:perlite:sand and grown as described previously ( see 'Plant growth conditions and sexual propagation ) . Cyan fluorescent protein peroxisome ( eCFP-SKL ) Bd28-marker lines were tested by PCR and confocal laser scanning microscopy . | Fungi cause many diseases in plants , and reduce the yield of important crops like wheat , corn and rice – all of which belong to the family of grasses . Much research into how disease-causing fungi infect plants will look at a given fungus that infects a specific plant in order to understand plant diseases in general . Over the years , scientists have generated suites of research tools to study these pairs of fungi and plants . However , many of these organism pairs ( often called “model pathosystems” ) have drawbacks when it comes to research in the laboratory , either on the side of the fungus or the side of plant . Brachypodium is a small grass that grows quickly and , unlike crop plants , it grows well in the laboratory . These characteristics make Brachypodium a promising model organism for studying many aspects of plant biology . Recently , a fungus called Ustilago bromivora – which is related to a fungus that infects corn – was reported to infect Brachypodium . This raised the question: could this fungus and this small grass become a new model pathosystem ? Rabe , Bosch et al . set out to answer this question and now provide a toolkit that will help to establish U . bromivora and Brachypodium as a new model pathosystem . In all of U . bromivora’s close relatives , two compatible strains must meet and mate before the fungus can infect the plant; first Rabe , Bosch et al . confirmed that this is also the case for U . bromivora . Studying the life cycle of the U . bromivora fungus also unexpectedly revealed that while both mating partners are needed to infect the plant , only one of the strains survives outside of the plant after the infection . This phenomenon , referred to as a “mating type bias” , has been described for a few other related fungi . Next , Rabe , Bosch et al . conducted a genetic screen and identified two compatible strains that can grow without the plant as yeast-like cells . This means that these cells can be manipulated genetically , and indeed protocols to grow and genetically engineer the fungus and plant to address different research questions are included in the toolkit as well . Other new tools include the complete genetic sequence of the fungus with all its genes annotated , and a dataset of which genes are active in U . bromivora growing yeast-like in liquid culture versus those active when the fungus grows as a pathogen inside the plant . Together these new tools and datasets will provide a foundation to study different aspects of the interactions between grasses and disease-causing fungi . This in turn may lead to new methods to reduce fungal growth and reduce yield losses caused by fungal diseases in crop plants . Finally , the discovery that U . bromivora shows a mating type bias could provide a starting point for future studies into sexual reproduction in fungi and how new species arise . | [
"Abstract",
"Introduction",
"Results",
"and",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | A complete toolset for the study of Ustilago bromivora and Brachypodium sp. as a fungal-temperate grass pathosystem |
Although millions of distinct virus species likely exist , only approximately 9000 are catalogued in GenBank's RefSeq database . We selectively enriched for the genomes of circular DNA viruses in over 70 animal samples , ranging from nematodes to human tissue specimens . A bioinformatics pipeline , Cenote-Taker , was developed to automatically annotate over 2500 complete genomes in a GenBank-compliant format . The new genomes belong to dozens of established and emerging viral families . Some appear to be the result of previously undescribed recombination events between ssDNA and ssRNA viruses . In addition , hundreds of circular DNA elements that do not encode any discernable similarities to previously characterized sequences were identified . To characterize these ‘dark matter’ sequences , we used an artificial neural network to identify candidate viral capsid proteins , several of which formed virus-like particles when expressed in culture . These data further the understanding of viral sequence diversity and allow for high throughput documentation of the virosphere .
There has been a rush to utilize massive parallel sequencing approaches to better understand the complex microbial communities associated with humans and other animals . Although the bacterial populations in these surveys have become increasingly recognizable ( Lloyd-Price et al . , 2017 ) , a substantial fraction of the reads and de novo assembled contigs in many metagenomics efforts are binned as genetic ‘dark matter , ’ with no recognizable similarity to characterized sequences ( Krishnamurthy and Wang , 2017; Oh et al . , 2014 ) . Some of this dark matter undoubtedly consists of viral sequences , which have remained poorly characterized due to their enormous diversity ( Simmonds et al . , 2017; Paez-Espino et al . , 2016; Emerson et al . , 2018 ) . Recent efforts have shown that our understanding of viral diversity , even of viruses known to directly infect humans , has been incomplete ( Pastrana et al . , 2018; Turnbaugh et al . , 2007; Gilbert et al . , 2010 ) . To increase the power of future studies seeking to more comprehensively catalog the virome and find additional associations between viruses and disease , reference genomes for all clades of the virosphere need be identified , annotated , and made publicly accessible . Virus discovery has typically proven to be more difficult than discovery of cellular organisms . Whereas all known cellular organisms encode conserved sequences ( such as ribosomal RNA genes ) that can readily be identified through sequence analysis , viruses , as a whole , do not have any universally conserved sequence components ( O'Leary et al . , 2016; Brister et al . , 2015; Sullivan , 2015; Rohwer and Edwards , 2002 ) . Nevertheless , some success has been achieved in RNA virus discovery by probing for the conserved sequences of their distinctive RNA-dependent RNA polymerase or reverse transcriptase genes in metatranscriptomic data ( Shi et al . , 2016 ) . Also , many bacteriophages of the order Caudovirales , such as the families Siphoviridae , Podoviridae , and Myoviridae , have been reported in high numbers due to their and their hosts' culturability and their detectability using viral plaque assays ( Pope et al . , 2015; Grose and Casjens , 2014; Grose et al . , 2014 ) . The relatively abundant representation of these families in databases has allowed new variants to be recognized by high-throughput virus classification tools like VirSorter ( Roux et al . , 2015; Gregory et al . , 2019; Roux et al . , 2019b ) . In contrast , many small DNA viruses are not easily cultured ( Bedell et al . , 1991 ) , use diverse genome replication strategies , and typically lack DNA polymerase genes such as those in large DNA viruses ( Koonin et al . , 2015 ) . An additional challenge is that small DNA viruses with segmented genomes may have segments that do not encode recognizable homologs of known viral genes . Therefore , small DNA viruses are more sparsely represented in reference databases . However , some groups have been successful in discovery of small DNA genomes in a wide range of viromes ( Blinkova et al . , 2010; Pastrana et al . , 2018; Dayaram et al . , 2015; Dayaram et al . , 2016; Labonté and Suttle , 2013; Rosario et al . , 2018; Victoria et al . , 2009 ) . Despite the apparent challenges in detecting small DNA viruses , many have physical properties that can be leveraged to facilitate their discovery . In contrast to the nuclear genomes of animals , many DNA virus genomes have circular topology , which allows selective enrichment through rolling circle amplification ( RCA ) methods ( Kim et al . , 2008 ) . Further , the unique ability of viral capsids to protect nucleic acids from nuclease digestion and to mediate the migration of the viral genome through ultracentrifugation gradients or size exclusion columns allows physical isolation of viral genomes . The current study grew out of an effort to find papillomaviruses ( small circular DNA viruses ) in humans and economically important or evolutionarily informative animals ( Pastrana et al . , 2018; Peretti et al . , 2015 ) . The sampling included several types of animals that might serve as laboratory models ( e . g . , mice , fruit flies , soil nematodes ) . A number of papillomaviruses were detected among a vastly larger set of circular DNA sequences that were not easily identifiable in standard BLASTN searches . The goal of the present study is to catalog and annotate the circular DNA virome from these animal tissues to understand the diversity and evolution of viral sequences . We developed a comprehensive bioinformatics pipeline , Cenote-Taker , to classify and annotate over 2500 candidate viral genomes and generate GenBank-compliant output files . Cenote-Taker is available for free public use with a graphical user interface at http://www . cyverse . org/discovery-environment .
We have previously developed methods for discovery of new polyomavirus and papillomavirus species in skin swabs and complex tissue specimens ( Peretti et al . , 2015 ) . Nuclease-resistant DNA from purified virions was amplified by random-primed rolling circle amplification ( RCA ) and subjected to deep-sequencing . Reads were de novo assembled into contigs and analyzed with a bioinformatics pipeline , Cenote-Taker ( a portmanteau of cenote , a naturally occurring circular water pool , and note-taker ) , to identify and annotate de novo-assembled contigs with terminal direct repeats consistent with circular DNA molecules . In this pipeline , putative-closed circular sequences of greater than 1000 nucleotides ( nt ) were queried against GenBank’s nucleotide database using BLASTN to remove circles with extensive nucleotide identity ( >90% across any 500 nt window ) to known sequences . Sequences with >90% identity to previously reported viral sequences represented less than 1 . 5% of circular contigs and are not included in further analysis . Approximate taxonomy was determined by BLASTX to a protein database derived from RefSeq virus proteins and GenBank plasmid proteins ( only hits better than 1 × 10−5 were considered ) . Open reading frames ( ORFs ) from remaining unidentified circular DNA sequences > 240 nucleotides ( nt ) in length were translated and used for RPS-BLAST queries of GenBank’s Conserved Domain Database ( CDD ) . ORFs that did not yield E values better than 1 × 10−4 in RPS-BLAST were subjected to BLASTP searches of viral sequences in GenBank’s nr database ( Altschul et al . , 1990; Marchler-Bauer and Bryant , 2004; Marchler-Bauer et al . , 2015 ) . For ORFs that were not confidently identified in BLAST searches , HHBlits ( Remmert et al . , 2012 ) was used to search the CDD , Pfam ( El-Gebali et al . , 2019 ) , Uniprot ( UniProt Consortium , 2019 ) , Scop ( Chandonia et al . , 2019 ) , and PDB ( Burley et al . , 2017 ) databases . The results were used to annotate and name each sequence in a human-readable genome map as well as a format suitable for submission to GenBank . After checking the Cenote-Taker output of each genome , minor revisions were made , as needed , and files were submitted to GenBank ( BioProject Accessions PRJNA393166 and PRJNA396064 ) . All annotations meet or exceed recently proposed standards for uncultivated virus genomes ( Roux et al . , 2019a ) . Plasmid sequences were frequently detected and were discarded . Circular sequences were considered to be plasmid-like if they: 1 ) had a best BLASTX hit to a plasmid and 2 ) had no detectable virion structural genes . Viral enrichment of the analyzed samples ( based on ViromeQC [Zolfo et al . , 2019] , with alignment to prokaryotic single-copy housekeeping genes ) was typically high ( Supplementary file 1 ) . However , even in the samples where enrichment was low , quality viral genomes could still be identified based on the bioinformatic analyses . Of the novel circular sequences detected in the survey , 1844 encode genes with similarity to proteins of ssDNA viruses and 55 encode genes with similarity to dsDNA viral proteins ( Figure 1A ) . The large majority of genomes from this study are highly divergent from RefSeq entries ( Figure 1—figure supplement 1 ) . We discovered 868 genomes that had similarity to unclassified eukaryotic viruses known as circular replication-associated protein ( Rep ) -encoding single-stranded DNA ( CRESS ) viruses . The group is defined by the presence of a characteristic rolling circle endonuclease/superfamily three helicase gene ( Rep ) ( Zhao et al . , 2019; Kazlauskas et al . , 2019 ) , but has not been assigned to families by the ICTV or RefSeq . We estimate that 199 non-redundant unclassified CRESS virus genomes had been previously deposited in GenBank , and 85 are curated in RefSeq ( Figure 1B ) . Also abundant was the viral family Microviridae , a class of small bacteriophages , with 670 complete genomes . This represents a substantial expansion beyond the 459 non-redundant microvirus genomes previously listed in GenBank ( of which 44 were curated in the RefSeq database ) . Other genomes that were uncovered represent Anelloviridae ( n = 170 ) , Inoviridae ( n = 70 ) , Genomoviridae ( n = 58 ) , Siphoviridae ( n = 18 ) , unclassified phage ( n = 14 ) , Podoviridae ( n = 10 ) , Myoviridae ( n = 7 ) unclassified virus ( n = 6 ) , Papillomaviridae ( n = 4 ) , Circoviridae ( n = 3 ) , unclassified Caudovirales ( n = 3 ) , Bacilladnaviridae ( n = 2 ) , Smacoviridae ( n = 2 ) , and CrAssphage-like ( n = 2 ) ( Figure 1B , Supplementary file 2 ) . Viral families were found in association with 23 different animal species ( Figure 1C ) . It was not surprising to find bacterial viruses , as all animals are presumed to have microbial communities and our sampling included tissues where these communities reside . It is difficult to assign a host to most of the viruses from this study due to their divergence from known viral sequences . However , we searched the CRISPR database at ( https://crispr . i2bc . paris-saclay . fr/crispr/BLAST/CRISPRsBlast . php ) , and three viruses had exact matches to CRISPR spacers in bacterial genomes ( Siphoviridae sp . ctcj11:Shewanella sp . W3-18-1 , Inoviridae sp . ctce6:Shewanella baltica OS195 , Microviridae sp . ctbe523:Paludibacter propionicigenes WB4 ) and one virus had an exact match to the CRISPR spacer of an archaeon ( Caudovirales sp . cthg227:Methanobrevibacter sp . AbM4 ) , implying that these organisms are infected by these viruses . Further , the 142 anelloviruses found in human blood samples ( Supplementary file 2 ) are almost certain to be bona fide human viruses based on their relatedness to known human anelloviruses . In addition to circular genomes with recognizable similarity to known viruses , 609 circular contigs appeared to represent elements that lacked discernable similarity to known viruses ( Figure 1A , C ) . The vast majority of the de novo assembled circular genomes were <10 kb in length ( Figure 1—figure supplement 2 ) . This is largely due to the fact that large genomes are typically more difficult to de novo assemble from short reads . Despite these technical obstacles , our detection of a new tailed bacteriophage with a 419 kb genome ( Myoviridae sp . isolate ctbc_4 , GenBank Accession: MH622943 ) , along with 45 other >10 kb circular sequences ( Figure 1—figure supplement 2 ) , indicates that the methods used for the current work can detect large viral genomes . There has been a recent renewal of interest in the hypothesis that viruses may be etiologically associated with degenerative brain diseases , such as Alzheimer's disease ( Itzhaki et al . , 2016; Eimer et al . , 2018 ) . Conflicting literature suggests the possible presence of papillomaviruses in human brain tissue ( Coras et al . , 2015; Chen et al . , 2012 ) . Samples of brain tissue from individuals who died of Alzheimer’s disease ( n = 6 ) and other forms of dementia ( n = 6 ) were subjected to virion enrichment and deep sequencing . Although complete or partial genomes of known papillomaviruses , Merkel cell polyomavirus , and/or anelloviruses were observed in some samples ( Supplementary file 3 ) , no novel complete viral genomes were recovered ( Supplementary file 2 ) . No viral sequences were detected in a follow-up RNA deep sequencing analysis of the brain samples . It is difficult to know how to interpret these negative data . It is conceivable that the known viral DNA sequences observed in the Optiprep-RCA samples represent virions from blood vessels or environmental sources . It has recently become apparent that certain nucleic acid extraction reagents are contaminated with viral nucleic acids ( Asplund et al . , 2019 ) . To ensure we were not merely reporting the sequences of the ‘reagent virome , ’ we performed our wet bench and bioinformatic pipeline on three independent replicates of reagent-only samples . We found no evidence of sequences of any viruses reported here or elsewhere . Further , cross-sample comparison of contigs showed that almost no sequences were found in different animal samples , aside from technical replicates . In total , six viral genomes were observed in multiple unrelated samples from at least two sequencing runs ( Supplementary file 4 ) . It is unclear whether this small minority of genomes ( 0 . 24% of the genomes reported in the current study ) represent reagent contamination , lab contamination , or actual presence of the sequences in different types of samples . Given the stringent requirements for sequences to be considered as belonging to a complete viral genome , as well as the largely unexplored nucleotide space of the virome , it is unsurprising that , in most samples , most reads did not align to the genomes reported in this study or virus genomes from RefSeq ( Figure 1—figure supplement 3 ) ( Supplementary file 5 ) . Single stranded DNA viruses , in general , have vital genes encoding proteins that mediate genome replication , provide virion structure , and , in some cases , facilitate packaging of viral nucleic acid into the virion . Being structurally conserved , these genes also tend to be important for evolutionary comparisons and can serve as important ‘hallmark genes’ for virus discovery and characterization . However , even structurally conserved proteins sometimes do not have enough sequence conservation as to be amenable to high confidence BLASTP searches . We therefore set out to catalog hallmark ssDNA virus genes based using protein structural prediction . Structures of hallmark genes of exemplar isolates from most established ssDNA virus families have been solved and deposited in publicly available databases such as PDB ( Protein Data Bank ) ( Burley et al . , 2017 ) . Using bioinformatic tools , such as HHpred , one can assign structural matches for a given gene based on the predicted potential folds of a given amino acid sequence . HHpred has been extensively tested and validated for computational structural modeling by the structural biology community ( Meier and Söding , 2015; Huang et al . , 2014 ) . The method proves especially useful for protein sequences from highly divergent viral genomes that have little similarity to annotated sequences in current databases . We extracted protein sequences from our dataset and compiled nonredundant proteins from circular ssDNA viruses in GenBank and used them as queries in HHpred searches against the PDB , PFam , and CDD databases . We then grouped structurally identifiable sequences into hallmark gene categories and aligned them pairwise ( each sequence was compared to all other sequences ) using EFI-EST ( Gerlt et al . , 2015 ) . The resulting sequence similarity networks ( SSNs ) were visualized with Cytoscape ( Su et al . , 2014 ) , with each node representing an predicted protein sequence ( Figures 2–3 , Figure 2—figure supplement 1 ) . Nodes ( sequences ) with significant amino acid similarity are connected with lines representing BLAST similarity scores better than a threshold E value . Sequence similarity network analyses , it has been proposed ( Iranzo et al . , 2017 ) , represent relationships between viral sequences better than phylogenetic trees . Further , SSNs have previously been used for viral protein and genome cluster comparison ( Bolduc et al . , 2017; Lima-Mendez et al . , 2008; Lefeuvre et al . , 2019; Kazlauskas et al . , 2019 ) and can be used to display related groups of viral genes in two dimensions ( Bin Jang et al . , 2019 ) . These clusters were also used to guide the construction of meaningful phylogenetic trees ( Figure 2A–B , Figure 2—figure supplement 2 ) . In Figure 2 , sequences that showed a structural match to a known eukaryotic circular ssDNA virus capsid protein are displayed as a network . This general capsid type features a single beta-jellyroll fold and assembles into T = 1 virions of 20–30 nm in diameter . The network shows that sequences from this study expand and link smaller disconnected clusters of sequences found in GenBank entries ( Figure 2A–C ) . Perhaps more importantly a number of previously unknown clusters were identified , providing insight into highly divergent hallmark sequences and making this capsid sequence space amenable to BLAST searches in GenBank ( Figure 2C ) . Although the satellite tobacco necrosis virus ( STNV ) capsid protein encapsidates an RNA molecule , it has previously been noted that its structure is highly similar to the capsid proteins of geminiviruses and other ssDNA viruses ( Koonin et al . , 2015; Kraberger et al . , 2015; Krupovic et al . , 2009; Hipp et al . , 2017; Bottcher et al . , 2004; Zhang et al . , 2001 ) and was included as a model for populating this network . A similar pattern can be seen in sequence similarity networks for the Rep genes of CRESS viruses ( Figure 3 ) . Rep genes have been the primary sequences used for taxonomy of CRESS viruses ( Zhao et al . , 2019 ) . In this case , it was determined that a network with alignment cutoffs with E values of 1 × 10−60 could split the data neatly into ‘family-level’ clusters ( Fontenele et al . , 2019; Kraberger et al . , 2019 ) , precisely mirroring ICTV taxonomy of CRESS viruses . Many additional family-level clusters can be discerned from unclassified CRESS viruses . Other eukaryotic and prokaryotic ssDNA virus hallmark gene networks are shown in Figure 2—figure supplement 1 . Phylogenetic trees of networks from Figures 2 and 3 and Figure 2—figure supplement 1 are displayed in Figure 2—figure supplement 2 . Cytoscape files of sequence similarity networks and phylogenetic trees can be found at https://ccrod . cancer . gov/confluence/display/LCOTF/DarkMatter . Although no single family of viruses accounts for the majority of genomes in this study , these results expand the knowledge of the vast diversity of CRESS viruses , which appear to be ubiquitous among eukaryotes ( Krupovic et al . , 2016; Zerbini et al . , 2017; Rosario et al . , 2017; Varsani and Krupovic , 2018 ) and are likely to also infect archaea ( Díez-Villaseñor and Rodriguez-Valera , 2019; Kazlauskas et al . , 2019 ) . Characterized CRESS viruses have small icosahedral virions ( 20–30 nm in diameter ) with a simple T = 1 geometry ( Khayat et al . , 2011 ) . This capsid architecture likely limits genome size , as nearly all previously reported CRESS virus genomes and genome segments are under 3 . 5 kb . Exceptions to this size rule are bacilladnaviruses , which have 4 . 5–6 kb genomes ( Tomaru et al . , 2011 ) and cruciviruses , which have 3 . 5–5 . 5 kb genomes ( Quaiser et al . , 2016 ) . Interestingly , the genomes of these larger CRESS viruses encode capsid genes that appear to have been acquired horizontally from RNA viruses ( Kazlauskas et al . , 2017 ) . In our dataset , eight CRESS-like circular genomes exceed 6 kb in length ( Figure 4—figure supplement 1 ) . Further , this study's large CRESS genomes are apparently attributable to several independent acquisitions of capsid genes from other taxa and/or capsid gene duplication events . Notably , a large CRESS genome ( CRESS virus isolate ctdh33 , associated with rhabditid nematodes that were serially cultured from a soil sample ) encoded three separate genes with structural homology ( HHpred probability scores 97–99% ) to STNV capsid ( Figure 4—figure supplement 1G ) . The three predicted STNV capsid homologs in the nematode virus are highly divergent from one another , with only 28–30% amino acid similarity , but also highly divergent from other amino acid sequences in GenBank . A possible explanation for this observation is that the capsid gene array is the result of gene duplication events . CRESS genomes ctba10 , ctcc19 , ctbj26 , ctcd34 , and ctbd1037 ( ranging from 3 . 5 to 6 . 2 kb in length ) also each encode two divergent capsid gene homologs ( Figure 4—figure supplement 1A , B , C , E , H ) . Single genomes encoding multiple capsid genes with related but distinct amino acid sequences have been observed in RNA viruses ( Agranovsky et al . , 1995 ) and giant dsDNA viruses ( Schulz et al . , 2017 ) , but we believe that this is the first time it has been reported in ssDNA viruses . Two related large CRESS viruses ( ctdb796 and ctce741 ) encode capsid proteins similar to those of bacilladnaviruses ( Figure 4—figure supplement 1K , M ) . Interestingly , the Rep genes of the two viruses do not show close similarity to known bacilladnavirus Reps and are instead similar to the Reps of certain unclassified CRESS viruses , suggesting that CRESS ctdb796 and CRESS ctce741 are representatives of a new hybrid CRESS virus family . Two other CRESS virus genomes ( isolates ctca5 and ctgh4 ) encode capsid genes that show amino acid similarity to distinct groups of icosahedral T = 3 ssRNA virus capsids ( Makino et al . , 2013 ) ( tombus- and tombus-like viruses ) , but not to cruciviruses or bacilladnaviruses ( Figure 4 , Figure 4—figure supplement 1D , J , Figure 4—figure supplement 2A ) . Further , a 6 . 6 kb CRESS virus ( isolate ctbd466 ) ( Figure 4—figure supplement 1L ) was found to encode a gene with some similarity to the capsid region of the polyprotein of two newly described ssRNA viruses ( ciliovirus and brinovirus ( Figure 4—figure supplement 2B ) ( Makino et al . , 2013; Greninger and DeRisi , 2015 ) . Protein fold predictor Phyre2 ( Kelley et al . , 2015 ) showed a top hit ( 58% confidence ) for the capsid protein of a norovirus ( ssRNA virus with T = 3 icosahedral capsid ) for isolate ctbd466 ( see GenBank: AXH73946 ) . Two CRESS genomes ( ctbe30 and ctbc27 ) from separate Rhesus macaque stool samples combine Rep genes specific to CRESS viruses with several genes specific to inoviruses , including inovirus-like capsid genes , which encode proteins that form a filamentous virion ( Figure 4—figure supplement 1F , N ) . The bacteriophage families Inoviridae and Microviridae are ssDNA viruses that replicate via the rolling circle mechanism , but they are not considered conventional CRESS viruses because they exclusively infect prokaryotes and do not encode Rep genes with CRESS-like sequences . Other inovirus-like genes encoded in the ctbe30 and ctbc27 genomes include homologs of zonular occludens toxin ( ZOT , a packaging ATPase ) and RstB ( a DNA-binding protein required for host genome integration ) ( Falero et al . , 2009 ) ( Figure 4—figure supplement 1F , N ) . TBLASTX searches using ctbe30 and ctbc27 sequences yielded large segments of similarity to various bacterial chromosomes ( e . g . , GenBank accession numbers AP012044 and AP018536 ) , presumably representing integrated prophages . This suggests that ctbb30 and ctbc27 represent a previously undescribed bacteria-tropic branch of the CRESS virus supergroup . Viral genomes discussed in this section were validated by aligning individual reads back to the contigs followed by visual inspection . No disjunctions were detected , indicating that illegitimate recombinations are not evident ( see Figure 4—figure supplement 2C for an example ) . We defined potential viral ‘dark matter’ in the survey as circular contigs with no hits with E values < 1 × 10−5 in BLASTX searches of a database of viral and plasmid proteins . We posited that leveraging sequence similarity networks would be useful both for analyzing groups of gene homologs and for discerning which gene combinations tended to be present on related circular genomes . To categorize the 609 dark matter elements based on their predicted proteins , we used pairwise comparison with EFI-EST . A majority of translated gene sequences could be categorized into dark matter protein clusters ( DMPCs ) containing four or more members ( Figure 5A ) . Further , groups of related dark matter elements ( i . e . dark matter genome groups ( DMGGs ) ) , much like viral families , could be delineated by the presence of a conserved , group-specific marker gene . For example , DMPC1 can be thought of as the marker gene for DMGG1 . Certain DMPCs tend to co-occur on the same DMGG . For instance , DMPC7 and DMPC17 ORFs are always observed in genomes with a DMPC1 ORF ( i . e . , DMGG1 ) ( Figure 5B ) . This pro tempore categorization method is useful for visualizing the data , but we stress that is not necessarily taxonomically definitive . HHpred , was again employed to make structural predictions for these data ( Zimmermann et al . , 2018 ) . Instead of querying individual sequences , alignments were prepared using MAFFT ( Katoh and Standley , 2013 ) for each major DMPC to identify conserved residues and increase sensitivity . Then , each alignment was used for an HHpred query . The results indicate that ten DMPCs are likely viral capsid proteins and 11 are rolling circle replicases ( Figure 5A ) . While most of the circular dark matter in the survey could be characterized using these methods , dark matter contigs represent a small remaining fraction in some samples ( Figure 5—figure supplement 1 ) . In contrast to viral genes such as Rep , with conserved enzymatic functions , sequences of the capsid genes are often poorly conserved , even within a given viral family ( Buck et al . , 2016 ) . Moreover , it appears that capsid proteins have arisen repeatedly through capture and modification of different host cell proteins ( Krupovic and Koonin , 2017 ) . This makes it challenging to detect highly divergent capsid proteins using alignment-based approaches or even structural modeling . We therefore turned to an alignment-independent approach known as iVireons , an artificial neural network trained by comparing alignment-independent variables between a large set of known viral structural proteins and known non-structural proteins ( Seguritan et al . , 2012 ) ( https://vdm . sdsu . edu/ivireons/ ) . As an example of the approach , iVireons scores for DMPCs associated with DMGG1 are shown in Figure 5C . Other sets of iVireons scores can be seen in Figure 5—figure supplement 2 . Of the 17 DMGGs for which HHPRED did not identify capsid genes , iVireons predicted that ten contain at least one DMPC predicted to encode some type of virion structural protein ( median score of cluster >0 . 70 ) . This allowed us to generate the testable hypothesis that some of these predicted structural proteins would form virus-like particles ( VLPs ) if expressed in cell culture . A subset of predicted capsid proteins were expressed in human-derived 293TT cells and/or in E . coli and subjected to size exclusion chromatography . Electron microscopic analysis showed that several of the predicted capsid proteins formed roughly spherical particles , whereas a negative control protein did not form particles ( Figure 6 ) . Although the particles were highly irregular , the DMGC11 isolate ctgh70 preparation was found to contain nuclease-resistant nucleic acids , consistent with nonspecific encapsidation . The results suggest that , in multiple cases , we were able to experimentally confirm that iVireons correctly predicted the identity of viral capsid proteins .
Massive parallel DNA sequencing surveys characterizing microbial communities typically yield a significant fraction of reads that cannot be mapped to known genes . The present study sought to provide the research community with an expanded catalog of viruses with circular DNA genomes associated with humans and animals , as well as a means to characterize future datasets . We hope that the availability of this expanded viral sequence catalog will facilitate future investigation into associations between viral communities and disease states . Our annotation pipeline , Cenote-Taker , can be accessed via http://www . cyverse . org/discovery-environment . The CyVerse version of Cenote-Taker can readily annotate circular or linear DNA viruses . RNA viruses with polyproteins or frameshifts will require post hoc manual editing . Efforts could be made , for example , to apply the pipeline to previously published viromes to uncover additional viral genomes missed by other methods . At the present time , GenBank’s RefSeq database includes complete sequences for approximately 9000 viral genomes , most of which fit into 131 families recognized by the International Committee on Taxonomy of Viruses ( ICTV ) ( King et al . , 2018 ) . Similarly , the IMG/VR database contains over 14 , 000 circular virus genomes from hundreds of studies , though some of these appear to be redundant with each other and are not comprehensively annotated ( Paez-Espino et al . , 2019 ) . The current study , which focused on circular DNA viruses with detergent-resistant capsids , found 2514 new complete circular genomes . The availability of these comprehensively annotated genomes in GenBank contributes new information and understanding to a broad range of established , emerging , and previously unknown taxa . Figure 3 shows dozens of potential family-level groupings within the unclassified CRESS virus supergroup . Sequences from this study contribute to 40 of such groupings and constitute the only members of seven groups . There are also 192 singleton CRESS sequences that could establish many additional family-level groups . Although small ssDNA viruses are ubiquitous , they are often overlooked in studies that only characterize sequences that are closely related to reference genomes . In addition , ssDNA is not detected by some current DNA sequencing technologies unless second-strand synthesis ( such as the RCA approach used in the current study ) is conducted . While many of the viruses discovered in this study appear to be derived from prokaryotic commensals , it is important to note that bacteriophages can contribute to human and animal diseases by transducing toxins , antimicrobial resistance proteins , or genes that alter the physiology of their bacterial hosts ( Waldor and Mekalanos , 1996 ) . Furthermore , interaction between animal immune systems and bacteriophages appears to be extensive ( Hodyra-Stefaniak et al . , 2015 ) . Over 100 distinct human anellovirus sequences were found in human blood . Anelloviruses have yet to be causally associated with any human disease , but this study indicates that we are likely still just scratching the surface of the sequence diversity of human anelloviruses . It will be important to fully catalog this family of viruses to address the field’s general assumption that they are harmless . Several of the CRESS viruses detected in this study are larger than any other CRESS virus genomes that have been described previously . In some cases , the larger size of these genomes may have been enabled by a process involving capsid gene duplication events . Further , CRESS virus acquisition of T = 3 capsids from ssRNA Nodaviridae and Tombusviridae families has been previously suggested as the origin of bacilladnaviruses ( Kazlauskas et al . , 2017 ) and cruciviruses ( Steel et al . , 2016; Dayaram et al . , 2016; Roux et al . , 2013; Krupovic et al . , 2015 ) , respectively . We present evidence of additional independent recombination events between CRESS viruses and ssRNA viruses and ssDNA bacteriophages . In light of these findings , it should be reiterated that only DNA ( not RNA ) was sequenced in our approach , so DNA/RNA in silico false recombination does not seem plausible . These data suggest that CRESS viruses are at the center of a tangled evolutionary history of viruses in which genomes change not just via gradual point mutations but also through larger scale recombination and hybridization events . It is likely that some dark matter sequences detected in this study share a common ancestor with known viruses but are too divergent to retain discernable sequence similarity . In some cases , the dark matter circles may represent a more divergent segment of a virus with a multipartite genome . Alternatively , some of these sequences likely represent entirely new viral lineages that have not previously been recognized .
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact , Chris Buck ( buckc@mail . nih . gov ) . All reads and annotated genomes associated with this manuscript can be found on NCBI BioProject Accessions PRJNA393166 and PRJNA396064 . Cenote-Taker , the viral genome annotation pipeline , can be used by interested parties on the Cyverse infrastructure: http://www . cyverse . org/discovery-environment . Relevant protocols on lab website: https://ccrod . cancer . gov/confluence/display/LCOTF/Virome . | When scientists hunt for new DNA sequences , sometimes they get a lot more than they bargained for . Such is the case in metagenomic surveys , which analyze not just DNA of a particular organism , but all the DNA in an environment at large . A vexing problem with these surveys is the overwhelming number of DNA sequences detected that are so different from any known microbe that they cannot be classified using traditional approaches . However , some of these “known unknowns” are undoubtedly viral sequences , because only a fraction of the enormous diversity of viruses has been characterized . This “viral dark matter” is a major obstacle for those studying viruses . This led Tisza et al . to attempt to classify some of the unknown viral sequences in their metagenomic surveys . The search , which specifically focused on viruses with circular DNA genomes , detected over 2 , 500 circular viral genomes . Intensive analysis revealed that many of these genomes had similar makeup to previously discovered viruses , but hundreds of them were totally different from any known virus , based on typical methods of comparison . Computational analysis of genes that were conserved among some of these brand-new circular sequences often revealed virus-like features . Experiments on a few of these genes showed that they encoded proteins capable of forming particles reminiscent of characteristic viral shells , implying that these new sequences are indeed viruses . Tisza et al . have added the 2 , 500 newly characterized viral sequences to the publicly accessible GenBank database , and the sequences are being considered for the more authoritative RefSeq database , which currently contains around 9 , 000 complete viral genomes . The expanded databases will hopefully now better equip scientists to explore the enormous diversity of viruses and help medics and veterinarians to detect disease-causing viruses in humans and other animals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2020 | Discovery of several thousand highly diverse circular DNA viruses |
Dorsal/ventral ( DV ) patterning of the sea urchin embryo relies on a ventrally-localized organizer expressing Nodal , a pivotal regulator of the DV gene regulatory network . However , the inceptive mechanisms imposing the symmetry-breaking are incompletely understood . In Paracentrotus lividus , the Hbox12 homeodomain-containing repressor is expressed by prospective dorsal cells , spatially facing and preceding the onset of nodal transcription . We report that Hbox12 misexpression provokes DV abnormalities , attenuating nodal and nodal-dependent transcription . Reciprocally , impairing hbox12 function disrupts DV polarity by allowing ectopic expression of nodal . Clonal loss-of-function , inflicted by blastomere transplantation or gene-transfer assays , highlights that DV polarization requires Hbox12 action in dorsal cells . Remarkably , the localized knock-down of nodal restores DV polarity of embryos lacking hbox12 function . Finally , we show that hbox12 is a dorsal-specific negative modulator of the p38-MAPK activity , which is required for nodal expression . Altogether , our results suggest that Hbox12 function is essential for proper positioning of the DV organizer .
Patterning of the embryonic ectoderm along the dorsal/ventral ( DV ) axis , also known as oral/aboral axis , has been extensively studied in various species of sea urchins . DV polarity is not firmly established in the unfertilized egg , but rather relies on a combination of inherited maternal information and inductive interactions among early blastomeres , becoming morphologically recognizable from the gastrula stage onward ( Brandhorst and Klein , 2002; Angerer and Angerer , 2003; Molina et al . , 2013 ) . The ectoderm of the pluteus larva is noticeably partitioned into four main domains: ( 1 ) the oral/ventral ectoderm , a thickened epithelium surrounding the mouth , ( 2 ) the aboral/dorsal ectoderm , a squamous epithelium that covers most of the rest of the larval body , ( 3 ) the ciliary band , a belt of ciliated cells positioned at the border between oral and aboral ectoderm , and ( 4 ) the apical neurogenic domain . The genetic landmark of polarization along the secondary axis is the zygotic expression of the TGF-β superfamily member Nodal on the future oral side , which behaves as an organizing centre imposing DV polarity in all three germ layers of the embryo ( Duboc et al . , 2004; Flowers et al . , 2004; Duboc et al . , 2010; Materna et al . , 2013 ) . Targets of Nodal signaling within the oral ectoderm include genes encoding the TGF-β pathway extracellular components Lefty , BMP2/4 and Chordin ( Angerer et al . , 2000; Duboc et al . , 2004 , 2008; Bradham et al . , 2009; Lapraz et al . , 2009; Yaguchi et al . , 2010 ) . Although they are expressed by the same cells , Lefty is thought to diffuse more rapidly than Nodal , thus acting as a long-range Nodal inhibitor ( Bolouri and Davidson , 2010; Duboc et al . , 2004 , 2008 ) . The BMP2/4 ligand acts instead as a relay to specify the aboral ectoderm , to which its signaling activity is confined , due to inhibition of BMP2/4 reception by Chordin within the oral ectoderm ( Angerer et al . , 2000; Duboc et al . , 2004; Lapraz et al . , 2009; Chen et al . , 2011 ) . The amount of details available on molecular circuits that govern DV patterning downstream of nodal expression is growing rapidly ( Su , 2009; Saudemont et al . , 2010; Yaguchi et al . , 2010; Chen et al . , 2011; Li et al . , 2012 , 2013 ) . As opposite , only fuzzy clues are known as to the early steps that trigger the DV gene regulatory network . According to the current models , at the early blastula stage a maternally related anisotropy in redox gradient would transiently inactivate the p38 kinase in the future dorsal ectoderm ( Bradham and McClay , 2006; Coffman , 2009 ) , somehow leading to activation of the maternal bZIP and Oct1/2 factors on the ventral side ( Nam et al . , 2007; Range et al . , 2007; Range and Lepage , 2011 ) . The cis-regulatory apparatus of nodal responds to these factors , as well as to the maternal positive inputs of SoxB1 and Univin ( Range et al . , 2007 ) , directing the expression of the gene within a discrete sector of the ectoderm that corresponds to the presumptive oral ectoderm ( Duboc et al . , 2004; Flowers et al . , 2004; Saudemont et al . , 2010 ) . Such a spatial expression profile is then consolidated by a positive feedback mechanism related to the Nodal signal transduction system , and by the concurrent Nodal-dependent production of the Nodal antagonist Lefty . It should be noted that all of the known positive inputs converging on the nodal cis-regulatory apparatus are broadly distributed in the embryo as early as nodal transcription occurs , raising the question of whether additional negative regulators are involved in the initial repression of nodal in all but the oral territories . In strict accordance with this possibility , it has been shown that the transcription repressor FoxQ2 contributes , together with Lefty , to suppress nodal expression in the apical neurogenic ectoderm ( Yaguchi et al . , 2008 ) . On the other hand , a similar negative function acting on nodal within the presumptive dorsal ectoderm cells has not yet been uncovered . An extremely interesting candidate for this role is the zygotically-expressed hbox12 homeobox-containing gene . In Paracentrotus lividus , hbox12 expression precedes the onset of nodal transcription , then declines after the 60-cell stage and is no longer detectable by hatching ( Di Bernardo et al . , 1995; this paper ) . Intriguingly , hbox12 transcripts are asymmetrically distributed along the DV axis , being confined in cells that become aboral ectoderm , as revealed by cis-regulatory analysis ( Cavalieri et al . , 2008 ) . The observed pattern of expression is so far unique in sea urchin development , and suggests that the Hbox12 transcription factor could act as a precocious input within the gene regulatory network that directs DV patterning . In agreement with this hypothesis , we have previously shown that disrupting the function of the Otx activator , a driver of hbox12 , downregulates hbox12 transcription and dramatically affects embryo polarization along the DV axis ( Cavalieri et al . , 2008 ) . Here we extend these findings providing more direct evidence indicating that hbox12 is a key upstream gene in the symmetry-breaking sequence of events , functioning to prevent the ectopic activation of nodal transcription within the prospective dorsal side of the early sea urchin embryo .
The timing and spatial expression profile of hbox12 described earlier ( Di Bernardo et al . , 1995 ) led us to initially hypothesize an involvement of such a gene in the negative control of nodal expression . This is further supported by directly comparing the dynamic temporal and spatial patterns of hbox12 and nodal transcription during very early development . As expected , hbox12 expression begins at least two cell divisions earlier with respect to that of nodal ( Figure 1A ) . Double-labeling whole mount in situ hybridization ( WMISH ) at the early blastula stage showed that the spatial domains of expression of the two genes occupy opposite sectors of the embryo ( Figure 1B ) . These results , coupled with those of the cis-regulatory analysis reported previously ( Cavalieri et al . , 2008 ) , lead us to definitively conclude that hbox12 is expressed in the prospective dorsal ectoderm of the early embryo . Thus , hbox12 is expressed at the right time and in the right place to regulate nodal expression . 10 . 7554/eLife . 04664 . 003Figure 1 . Expression of hbox12 and nodal genes during early embryogenesis of P . lividus . ( A ) Temporal expression profiles examined by qPCR . Values at the different stages are shown as a percentage of the maximum signal intensity . Absolute numbers of transcripts per embryo given at the 60-cell stage are averages of the results of two independent experiments using distinct batches of cDNA . Abbreviations of the examined developmental stages: 2 , 2-cell; 4 , 4-cell; 8 , 8-cell; 16 , 16-cell; 32 , 32-cell; 60 , 60-cell; Mor , morula; eB , early blastula; HB , hatching blastula . ( B ) Spatial restriction of the hbox12 and nodal transcripts observed following WMISH at the indicated stage . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 003 As a first approach to reveal the role of hbox12 during embryogenesis , we microinjected into zygotes the synthetic full-length mRNA at dosage ranging from 0 . 01 to 0 . 4 pg . When eggs from the same batches were injected with equal amounts of the control out-of-frame strim1 transcript ( Cavalieri et al . , 2011 ) , or low amounts of the hbox12 mRNA ( <0 . 05 pg ) , the embryos developed normally ( Figure 2A–C ) . By contrast , 50% ( n > 1500 ) of the specimens ubiquitously expressing 0 . 1 pg of the functional hbox12 mRNA exhibited a highly reproducible strong perturbation . Development of these embryos was apparently normal until the mesenchyme blastula stage ( Figure 2D ) . At gastrula stage , when control embryos displayed a clear DV polarity as shown by the thickening of the ventral side and the symmetric ventral-lateral arrangement of the two primary mesenchyme cell ( PMC ) clusters ( Figure 2B ) , embryos translating exogenous hbox12 mRNA appeared quite rounded and their PMCs were irregularly dispersed into the blastocoel ( Figure 2E ) . 10 . 7554/eLife . 04664 . 004Figure 2 . Disruption of embryonic DV polarity by ectopic expression of hbox12 . ( A–G ) 0 . 1 pg of the full-length hbox12 mRNA ( D–F ) or the hd-En mRNA ( G ) , as well as a control out-of-frame strim1 transcript ( A–C ) , were injected into zygotes and embryos were observed at the indicated stages . Overexpression of either hbox12 or hd-En severely perturbed DV axis formation , inflicting morphological defects that appeared from the gastrula stage onward . ( H ) qPCR measurements of nodal transcript abundance in embryos injected with increasing amounts of the hbox12 mRNA . Values are shown as a percentage of the nodal mRNA level in control uninjected embryos . Further detail for the qPCR procedure is given in ‘Materials and methods’ . ( I ) Changes in gene expression level of nodal and other territorial marker genes assessed by qPCR in hbox12-injected embryos . Data are indicated as normalized ΔCt ( ΔΔCt , left ordinate ) , and as the corresponding fold difference in transcript abundance ( right ordinate ) , with respect to control embryos , at the same stage of development , derived from zygotes injected with the strim1 out-of-frame transcript . The gray region represents ΔΔCt values corresponding to less than threefold difference . Although this is commonly considered the limit of significance for qPCR assays , the relevance of our measurements is reinforced by WMISH results ( see text for details ) . Error bars are standard errors for the qPCR replicates . Oligonucleotide primer pairs used for qPCR reactions and amplicon lengths are indicated in Supplementary file 1 . Abbreviations: lG , late gastrula; VEB , very early blastula . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 00410 . 7554/eLife . 04664 . 005Figure 2—figure supplement 1 . Overexpression of the HD-En obligate repressor and effect on nodal and gsc gene expression . Changes in gene expression levels was assessed by qPCR in cDNA samples resulting from hd-En-injected embryos . Data are normalized and indicated as in Figure 2I . The gray region represents ΔΔCt values corresponding to less than threefold difference . See also Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 00510 . 7554/eLife . 04664 . 006Figure 2—figure supplement 2 . Spatial distribution of ectoderm- and PMC-specific markers in control and hbox12 overexpressing embryos . Control strim1 out-of-frame transcript ( A–D ) and hbox12 mRNA injected ( E–K ) embryos were fixed at the mesenchyme blastula ( A–C , E–G and I–K ) or prism ( D and H ) stages and analysed by chromogenic WMISH with the indicated probes . The embryos shown in ( D ) and ( H ) are oriented in a oral/ventral and lateral view , respectively , while all the other embryos are in a vegetal view . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 006 A striking phenotype was even more apparent at the pluteus stage . Control-injected embryos were normal angular-shaped larvae exhibiting the characteristic bilateral symmetry ( Figure 2C ) . By contrast , hbox12-injected embryos appeared almost spherical ( Figure 2F ) . In these specimens , PMCs retained a certain biomineralizing activity , producing two calcareous elements which did not elongate as much as those of the control embryos at the same stage . These embryos displayed a straight archenteron and did not form a stomodeum . As judged by morphological observation , their ectoderm was composed of only a thick and a squamous epithelium , respectively coating the animal and the vegetal side of the larva . Exactly the same phenotype was obtained following injection of similar amounts of a synthetic mRNA encoding for the chimeric repressor HD-En ( Figure 2G ) , in which the homeodomain of Hbox12 was joined to the repressor domain of Drosophila Engrailed . It follows that Hbox12 normally function as a transcriptional repressor in the early embryo . Altogether , the phenotypes showed in Figure 2 are broadly similar to those obtained by the knock-down of nodal function ( Duboc et al . , 2004 ) , again suggesting a potential negative effect of the Hbox12 transcription factor on nodal expression . Indeed , microinjection of equal amounts of either hbox12 or hd-En mRNA caused a dose-dependent attenuation in the level of nodal transcript at a very similar extent , as revealed by qPCR analysis from hbox12-injected embryos at morula stage ( Figure 2H and Figure 2—figure supplement 1 ) . Expression of the additional oral ectoderm marker goosecoid ( gsc ) as well as the aboral ectoderm marker tbx2/3 was also reduced ( Figure 2I and Figure 2—figure supplement 1 ) , as expected from the perturbed expression of nodal normally necessary for the establishment of the entire DV axis ( Duboc et al . , 2004 ) . The variations in transcript abundance shown in Figure 2H , I could appear somewhat modest , although they are of the same order of magnitude as those reported by other authors ( Agca et al . , 2010; Bergeron et al . , 2011; Coffman et al . , 2004 ) . A possible explanation is that the qPCR data refer to the whole embryo population , affected and not affected . In fact , in our experiments only about half of the hbox12-injected embryos developed as aberrant larvae that , unfortunately , could not be isolated at early stages because they were virtually indistinguishable from the unaffected ones . In addition , the phenotype exhibited by the affected embryos was broadly similar but not perfectly identical to that obtained following knock down of nodal ( Duboc et al . , 2004 ) , leading us to speculate that nodal expression was decreased to some extent in the overtly affected embryos . In close agreement with this hypothesis , WMISH showed that the expression of nodal , gsc , and tbx2/3 genes was apparently reduced ( n = 36/93; Figure 2—figure supplement 2E–G ) or even nullified ( n = 8/93; Figure 2—figure supplement 2I–K ) in roughly 47% ( n = 93 ) of hbox12-injected embryos . Although we were not able to quantify subtle differences , if any , in gene expression between control- and hbox12-injected embryos by a mere analysis of Dig-probe staining , the overall fraction of embryos showing defective expression of nodal , gsc , and tbx2/3 was somewhat consistent with that of embryos that ultimately displayed unambiguous loss of DV polarization at later stages ( Figure 2E , F ) . These findings substantially override the lack of significance in of qPCR expression data , that although showed less than threefold difference , are strengthened by the WMISH results . Altogether , our findings indicate that ectopic expression of hbox12 disrupted the patterning of the ectodermal domains along the DV axis . By WMISH with the PMC-specific marker msp130 we also confirmed that in the hbox12-injected embryos the complement of PMCs was quite congruent with that of control embryos at the same stage but , in most embryos ( 70% , n = 250 ) , the distribution of PMCs into the blastocoel was disorganized ( compare panels J and K of Figure 2—figure supplement 2 ) . This finding indicates that defects in PMC arrangement are due to a failure in the ectoderm to provide adequate patterning information . In close agreement , the transcript level of ectoderm-specific genes that have been previously implicated in skeletogenesis was reduced in hbox12-overexpressing embryos ( Figure 2I ) . As hbox12 transcripts are present in a quadrant of the early embryo including the presumptive aboral ectoderm founder cells of the animal hemisphere and part of the veg1 tier ( Di Bernardo et al . , 1995; Cavalieri et al . , 2008 ) , we assessed hbox12 misexpression in each domain by transplantation experiments . Zygotes were injected with the full-length hbox12 mRNA together with the Texas Red-conjugated dextran ( TRCD ) lineage tracer and allowed to develop up to the 16-cell stage . Then , animal and vegetal halves from these embryos were separated and recombined with their complementary halves derived from control uninjected embryos ( Figure 3A ) , and resulting chimeras were observed at pluteus stage . When hbox12 was overexpressed exclusively in vegetal halves , almost all chimeras ( n = 9/10 ) developed into larvae that were indistinguishable from controls ( Figure 3C ) . In these embryos , descendants of vegetal cells normally formed the endomesoderm territories which , owing to TRCD , appeared red fluorescent . 10 . 7554/eLife . 04664 . 007Figure 3 . Overexpression of hbox12 in chimeric embryos . ( A ) Schematic illustrating that , at the 16-cell stage , animal and vegetal halves of hbox12/TRCD-injected and control uninjected embryos were isolated and recombined . ( B–C ) Side views of representative examples of the resulting reciprocal chimeras examined at 48 hpf . The composition of the chimeras is shown in the diagrams in the left panels ( Ba and Ca ) . Images for each embryo are shown under DIC optics with simultaneous plane polarized light illumination ( Bb and Cb ) , epifluorescence ( Bc and Cc ) , and aggregate merging ( Bd and Cd ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 007 By contrast , 6/10 of the reciprocal chimeric embryos , in which hbox12 was present in the animal hemisphere ( Figure 3B ) , produced the same phenotype as that observed with ubiquitous hbox12 overexpression . Indeed , DV polarization was apparently impaired , stomodaeum did not form , and skeletal elements remained poorly elongated . Taken together , these findings reveal that overexpression of the transcriptional repressor Hbox12 in cells derived from the animal hemisphere impinges on the establishment of the DV axis . To study in more detail the role of hbox12 , we first attempted to block the expression of the protein by injecting morpholino-substituted antisense oligonucleotides directed against the translation initiation site . Unfortunately , injected embryos always developed to the pluteus stage with no overt abnormalities . As multiple copies of the hbox12 gene exist in P . lividus ( Cavalieri et al . , 2008 ) , this failure may be due to gene copies with different translation initiation sites . The Hbox12 protein includes the homeodomain close to the N-terminus and two serine-rich octapeptide repeats in the C-terminal region , which probably account for transcription repressor activity ( Cavalieri et al . , 2008 ) . Therefore , to disrupt the hbox12 function we expressed a truncated form of the protein , referred to as HD , which sequence ends just after the homeodomain . In principle , HD should efficiently compete with the endogenous Hbox12 for binding to DNA , quenching the repressor activity on target genes . Again , almost all embryos injected with the control strim1 transcript developed without deleterious effects ( Figure 4Aa–b ) . By contrast , embryos injected with hd mRNA exhibited a failure of the DV axial patterning . At the gastrula stage , when controls exhibited a correctly partitioned ectoderm ( Figure 4Aa ) , the vast majority of the hd-injected embryos ( 80% , n > 1200 ) appeared to be constituted by a uniformly thickened epithelium and no discernable ciliary band was identified ( Figure 4Ac ) . The PMCs were homogeneously distributed around the straight archenteron , without distinguishable clusters , forming six to eight triradiate spicule rudiments ( Figure 4Ac–c′ ) . At the pluteus stage , HD-expressing embryos exhibited a range of phenotypes that could be ranked in order of severity . The most dramatically affected embryos ( 62% , n > 1200 ) have developed an archenteron . However , it never bent to fuse with the oral ectoderm but did grow from the centre of a spherical embryo ( Figure 4Ad ) . In these specimens , PMCs adopted a full radial distribution , as confirmed by WMISH with the msp130 probe ( Figure 4Bh , n = 31/45 ) , and formed a grossly mispatterned skeleton around the circumference of the embryo ( Figure 4Ad–d′ ) . Moreover , pigment cells were not detected at all and embryos appeared uniformly albino ( Figure 4Ad ) . A smaller fraction of hd-injected embryos ( 28% ) showed a slightly less severe phenotype ( Figure 4Ae–e′ ) , indicating that axial specification was partially impaired in these embryos . 10 . 7554/eLife . 04664 . 008Figure 4 . Impairing hbox12 function and effects on DV axis formation . ( A ) Zygotes were injected with the control out-of-frame strim1 RNA ( Aa–Ab ) or the hd mRNA ( Ac–Ae ) , and the resulting embryos were observed from a vegetal view at the indicated stages . ( B ) Control- ( Ba–Bd ) and hd-injected ( Be–Bh ) embryos were fixed at the mesenchyme blastula stage and analysed by WMISH with the indicated probes . The embryo shown in ( Ba ) is oriented in a lateral view , while all the other embryos are in a vegetal view . ( C ) Changes in gene expression level of territorial marker genes assessed by qPCR during development of hd-injected embryos . Data are normalized and indicated as in Figure 2I . The gray region represents ΔΔCt values corresponding to non-significant variation ( less than threefold difference ) . See also Supplementary file 1 . Abbreviations: VEB , very early blastula; EB , early blastula; MB , mesenchyme blastula; lG , late gastrula . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 00810 . 7554/eLife . 04664 . 009Figure 4—figure supplement 1 . Multiple comparison of the homeodomain of Hbox12 and Otp proteins from P . lividus . Identical residues are shaded by differently coloured boxes , while double and single dots indicate decreasing degree of conservation; divergent amino acids are indicated by blank spaces . The Glutamine at position 50 is marked by an asterisk . Accession numbers were: Hbox12 , X83675 . 1 , and Otp , AJ007501 . 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 00910 . 7554/eLife . 04664 . 010Figure 4—figure supplement 2 . Overexpression of isolated homeodomains and effect on nodal and gsc gene transcription . Zygotes were injected with the synthetic mRNAs coding for the isolated homeodomain of either Hbox12 or Otp and observed at the early blastula stage ( upper panels ) . Total RNA was extracted from embryos at the morula stage , and changes in gene expression level of nodal and gsc determined by qPCR measurements . Data are indicated as in Figure 2I . The gray region represents ΔΔCt values corresponding to non-significant variation ( less than threefold difference ) . See also Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 010 To ascertain the specification of different cell types in hd-injected embryos , we used a set of marker genes specific for the major embryonic territories . By WMISH , we first examined the expression of nodal and gsc , which are expressed exclusively in ventral ectoderm cells of control mesenchyme blastulae ( Figure 4Ba , 4Bb ) . In sharp contrast , we found that HD expression in embryos at the same stage caused broadened ectodermic transcription of both genes in approximately 70% ( n > 80 ) of the resulting embryos ( Figure 4Be , 4Bf ) . The mRNA abundance of nodal and gsc was examined by qPCR and , as expected , upregulation was detected for both genes during development ( Figure 4C ) . The precise mechanisms that dictate the physiological function and target range of individual homeodomain proteins are in general either unknown or incompletely delineated ( Svingen and Tonissen , 2006 ) . As the DNA-binding properties of the homeodomains could not be , by themselves , sufficiently discriminating to distinguish between different sets of target genes in vivo ( Hoey and Levine , 1988 ) , to assess the specificity of HD on nodal and gsc gene transcription , we injected an equal amount of otp-hd , a synthetic mRNA coding for the closely related homeodomain of Orthopedia ( Di Bernardo et al . , 1999; Cavalieri et al . , 2003 ) . Otp and Hbox12 homeodomains belong both to the Q50 class and show a very similar helix-III ( Figure 4—figure supplement 1 ) . As shown in Figure 4—figure supplement 2 , all the early developing embryos expressing such an isolated DNA-binding domain did not display perceptible phenotypic aberration with respect to control injected embryos at the same stage . Most important , the transcript abundance of both nodal and gsc did not change significantly in embryos expressing the otp-hd . Altogether , these findings imply that impairing Hbox12 function through expression of HD specifically biased DV patterning toward ventralization . We also noted that the mRNA abundance of the dorsal specific gene tbx2/3 did fluctuate following a peculiar trend during development of hd-injected embryos . Although it was dramatically increased at early blastula stages ( Figure 4C ) , it appeared significantly down-regulated at mesenchyme blastula stage ( Figure 4Bg , n = 34/42 , and Figure 4C ) and almost completely abolished at late gastrula stage ( Figure 4C ) . It should be emphasized that tbx2/3 is expressed at a very low level during early embryogenesis ( Chen et al . , 2011; Ben-Tabou de-Leon et al . , 2013 ) . Hence , the simplest interpretation of the qPCR results is that the initial rise in tbx2/3 transcription level could be due to an early pulse of BMP2/4 before the HD-induced overexpression of Nodal swamps the system . In close agreement , it has been shown in P . lividus that injection of nodal mRNA abrogates the expression of tbx2/3 across the gastrula-stage embryo ( Duboc et al . , 2004 , 2010; Saudemont et al . , 2010 ) . Consistently with the derepression of nodal , the expression of four ventro-lateral ectoderm specific markers , strim1 , fgf , otp and pax2/5/8 , was robustly down-regulated in hd-injected gastrulae ( Figure 4C ) , strongly supporting the hypothesis that ectoderm patterning was impaired in these embryos . As described , an adjunctive defect characterizing most of the hd-injected embryos consisted in the lack of pigment cells . Known to be a cohort of secondary mesenchyme cells which expresses the specific marker gcm , the pigment cell precursors occupy a dorsal sector of the vegetal plate at the late mesenchyme blastula stage ( Ransick et al . , 2002; Röttinger et al . , 2006; Duboc et al . , 2010 ) . At this stage , hd-injected embryos had downregulated gcm expression ( Figure 4C ) . It is known that Nodal signaling antagonizes the specification of pigment cells ( Duboc et al . , 2010 ) , and our result is perfectly congruent with the reported loss of gcm expression , and the albino phenotype , provoked by the ectopic expression of nodal across the embryo ( Duboc et al . , 2010 ) . On this basis , we plausibly assume that the HD-induced overexpression of nodal similarly affected the specification of pigment cells in HD-expressing embryos . Hd transcript was injected into zygotes along with the TRCD-fluorescent tracer and then , at the 16-cell stage , animal and vegetal halves from these embryos were separated and recombined with their complementary halves derived from uninjected embryos . When HD was expressed only in vegetal halves , the resulting chimeras developed into normal plutei ( n = 10/10 ) , which endomesodermal territories appeared red-fluorescent ( Figure 5B ) . Thus , hbox12 function is not required in the vegetal half for DV polarization . By contrast , the reciprocal chimeric embryos , in which hbox12 function was disrupted in the animal hemisphere , phenocopied the morphologies of the non-chimeric hd-injected embryos , and the strongly perturbed phenotype prevailed ( n = 7/10; Figure 5A ) . These results support the contention that inhibition of hbox12 activity in the animal hemisphere , where it is normally expressed , is sufficient to impede the establishment of the DV axis . In addition , these findings indicate that skeletogenic disorders originate not in PMCs , but are due to a failure in the overlying ectoderm to provide adequate patterning information . 10 . 7554/eLife . 04664 . 011Figure 5 . Block of hbox12 function in the animal hemisphere of chimeric embryos , and effects on DV patterning . ( A–B ) Diagrams in ( Aa ) and ( Ba ) show the composition of the reciprocal chimeras resulting from animal and vegetal half recombination of hd/TRCD-injected and control uninjected embryos at the 16-cell stage . ( Ab–Ae and Bb–Be ) Side views of the resulting chimeras examined at 48 hpf . Images for each embryo are shown under DIC optics ( Ab and Bb ) , plane polarized light illumination ( Ac and Bc ) , epifluorescence ( Ad and Bd ) , and aggregate merging ( Ae and Be ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 011 To further strengthen confidence in this evidence , we attempted to disrupt the hbox12 activity exclusively in aboral ectoderm founder cells . This was accomplished by introducing into zygotes the expression construct phbox12-HD-GFP , in which the HD-GFP fusion protein was specifically placed under the control of the wild type cis-regulatory sequences of the hbox12 gene . This construct or phbox12-GFP , the control lacking HD ( Cavalieri et al . , 2008; see also ‘Materials and methods’ ) , was injected into zygotes and developing embryos were scored for GFP expression . In this experiment , the transgenes were delivered along with TRCD , to discriminate among injected embryos and those that escaped microinjection ( not-fluorescent ) . As expected , both constructs were expressed at a similar extent during early embryogenesis , as indicated by roughly comparable green fluorescence of embryos observed at the early blastula stage ( Figure 6—figure supplement 1A–B ) . In agreement with previous observations ( Cavalieri et al . , 2008 ) , at the gastrula stage green fluorescence was specifically detected in the ectoderm of an average of 60% ( n > 500 ) of injected embryos ( Figure 6A–B ) . However , whereas phbox12-GFP expression occurred in large ectoderm patches ( Figure 6Ac–d ) , scattered and less fluorescent cells were observed in phbox12-HD-GFP-injected embryos ( Figure 6Bc–d ) . This evidence and the fact that the number of HD-GFP-stained cells progressively extinguished as development proceeded ( not shown ) most likely indicate a low stability of the chimeric protein . Strikingly , despite this circumstance , most of the phbox12-HD-GFP-injected gastrulae expressing the transgene did show an equally thickened ectoderm and synthesized supernumerary spicule rudiments ( Figure 6Ba–b ) , resembling embryos at the same stage that received exogenous hd mRNA ( Figs 3Ac–c′ and 4Ab–c ) . 10 . 7554/eLife . 04664 . 012Figure 6 . Block of hbox12 function in aboral ectoderm cells of transgenic embryos , and effects on DV patterning . ( A–E ) Zygotes were injected with the indicated transgenes , and the resulting embryos were observed from a vegetal view at the indicated stages . Bright field ( Ca , Da , and Ea ) , DIC ( Aa , Ba , Cb , Db , and Eb ) , Dark field ( Ab , Bb , Cc , Dc , and Ec ) , epifluorescence ( Ac , Bc , Cd , Dd , and Ed ) , and aggregate merging ( Ad , Bd , Ce , De , and Ee ) images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 01210 . 7554/eLife . 04664 . 013Figure 6—figure supplement 1 . Expression of the HD-GFP fusion protein during early embryogenesis . ( A–C ) Zygotes were injected with either the phbox12-GFP ( A ) or phbox12-HD-GFP ( B ) DNA constructs or with the hd-GFP synthetic transcript ( C ) , and embryos were observed at the indicated stages . ( D ) Control embryo injected with TRCD . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 013 To obtain statistically relevant results we scored thousands of injected embryos at the early blastula stage . Specimens expressing HD-GFP at high level were transferred into a distinct plate filled with filtered sea water , and finally scored for the phenotype at 48 hr post-fertilization ( hpf ) . At this stage , all the phbox12-GFP injected embryos were normal pluteus larvae expressing the transgene reporter in their aboral ectoderm ( n > 500; Figure 6C ) . By contrast , about one half ( n > 800 ) of the phbox12-HD-GFP-injected embryos phenocopied the abnormalities observed following hd mRNA overexpression ( Figure 6D , E ) . Despite the expected absence of GFP fluorescence at this stage ( Figure 6Dd , 6Ed ) , most of the affected embryos ( 78% ) displayed profound defects in the establishment of the DV polarity , as shown by the absence of the larval arms and aboral vertex , and the reduced number of pigment cells ( Figure 6E ) . Worth mentioning , more than 60% of these embryos maintained such an abnormal phenotype for up to a week when maintained in culture , excluding that they were stunted embryos that could become normal with more developmental time . Otherwise , a smaller fraction ( 22% ) of the phbox12-HD-GFP-injected embryos showed rather severe radialization coupled to the presence of multiple ectopic spicules ( Figure 6D ) . In this experimental assay , the prevalence of the milder phenotype could be explained by the combination between the mosaic incorporation of the exogenous DNA construct in the embryo ( Flytzanis et al . , 1985; Franks et al . , 1988; Hough-Evans et al . , 1988 ) , and the rapid turnover of the HD-GFP protein . The number of the dark-red pigment cells was quantified in late stage pluteus larvae using DIC optics . The total average number of these cells was about 45 ± 5 in phbox12-GFP-expressing larvae developed from two distinct batches of zygotes ( n = 35 embryos counted in each experiment ) . This number did not differ from that observed in control uninjected plutei . In striking contrast , pigment cell population was greatly decreased in more than one half of phbox12-HD-GFP injected embryos observed at the same stage . In particular , specimens showing the milder phenotype differentiated less than 16 pigment cells ( Figure 6E ) , and this number did not vary culturing the embryos up to a week ( not shown ) . No pigment cells were instead detected in the phbox12-HD-GFP injected embryos exhibiting the fully radialized phenotype ( Figure 6D ) , once again supporting the hypothesis that loss of hbox12 function biased axial specification toward ventralization . The results described in the previous sections strongly suggest that hbox12 acts upstream of nodal , being involved in the asymmetrical establishment of the DV organizing centre . To better demonstrate the specificity of such a functional relationship , we performed a rescue assay in which the spatially restricted knock-down of nodal was superimposed on HD-expressing embryos . The experimental assay is depicted in Figure 7A . The hd-GFP mRNA was first microinjected into zygotes . At the 4-cell stage , the morpholino-oligonucleotide directed against nodal mRNA sequence ( Mo-nodal ) ( Duboc et al . , 2004 ) was successively injected into a single randomly chosen blastomere . To follow the fate of the re-injected cells , Mo-nodal was delivered together with the rhodamine-labelled dextran ( RLDX ) tracer . As expected , embryos injected with the hd-GFP mRNA and observed at the hatching blastula stage showed a diffused green fluorescence , indicating that the synthetic transcript was efficiently translated throughout the early embryo ( Figure 6—figure supplement 1C ) . Since , as mentioned , the abundance of the HD-GFP protein is rather quickly eroded following embryogenesis , early developing specimens simultaneously displaying GFP and RLDX fluorescence were selected for microscopic observation . 10 . 7554/eLife . 04664 . 014Figure 7 . Rescue of DV polarity by clonal knock-down of nodal into hd-injected embryos . ( A ) At the 4-cell stage , one blastomere of hd-GFP mRNA injected embryos was co-injected with a morpholino oligonucleotide directed against nodal ( Mo-nodal ) together with the RLDX red fluorescent tracer . The resulting embryos were scored for simultaneous GFP/RLDX fluorescence at the early blastula stage , and eventually examined by microscopic observation at the pluteus stage . ( B–G ) Representative examples of embryos injected with either the hd-GFP mRNA ( B and E ) or the Mo-nodal alone ( C and F ) , and of double-injected rescued embryos ( D–G ) . Note that in both the rescued pluteus larvae shown in ( D ) and ( G ) , the progeny of the blastomere that received Mo-nodal was embedded into the dorsal structures . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 014 Sister batches of zygotes were injected with either the hd-GFP mRNA or Mo-nodal alone , and observed at 48 hpf . Once again , the former embryos exhibited a typically ventralized phenotype ( Figure 7B , E ) , while embryos resulting from Mo-nodal injection developed into bell-shaped larvae with multiple entangled spicules ( Figure 7C , F ) . In these embryos , most of the ventral and dorsal ectoderm was replaced by a thick ciliated epithelium which , as demonstrated by other authors ( Duboc et al . , 2004 ) , represents the default state of the ectoderm . Thus , both hd-GFP and Mo-nodal injected embryos , albeit with peculiar differences , never acquired any bilateral symmetry . Conversely , almost all double-injected embryos ( n > 250 ) developed into normal pluteus larvae with a harmoniously patterned DV axis ( Figure 7D , G ) . Remarkably , inspection of these embryos under fluorescence illumination clearly revealed that the progeny of the blastomere injected with Mo-nodal was always found on the dorsal face of the rescued pluteus larvae ( Figure 7Db–c , 7Gb–c ) . As it has been extensively shown that injection of Mo-nodal in P . lividus embryos abrogates translation of the Nodal ligand ( Duboc et al . , 2004 , 2010; Range et al . , 2007; Saudemont et al . , 2010; Bessodes et al . , 2012 ) , we reasonably infer that in the double hd-GFP and Mo-nodal injected embryos , clonal knock-down of nodal was able to restore the asymmetrical production of the Nodal ligand , which in turn sufficed to resume the entire DV axis . According to current models , the activation of nodal requires a phosphorylation event catalyzed by the p38 MAPK ( Bradham and McClay , 2006; Nam et al . , 2007; Range et al . , 2007; Coffman et al . , 2009 ) . As mentioned , p38 is globally active , but is transiently inactivated in the prospective dorsal side of the early embryo , immediately before the onset of nodal expression ( Bradham and McClay , 2006 ) . By this time , that in P . lividus corresponds to the 60-cell stage , the peak of hbox12 transcription has just been accomplished ( Figure 1 ) , and therefore , anisotropic inactivation of p38 should correlate with hbox12 expression . To confirm that prediction , the impact of perturbing hbox12 function on p38 activity was determined by expressing a GFP-tagged p38 , to reflect kinase activation as GFP nuclearization in living embryos ( the inactive kinase is instead cytoplasmic ) . Embryos expressing such a fusion protein developed normally whereas , at the 60-cell stage , p38 was cleared for a brief interval from nuclei on one side of the embryo ( Figure 8A ) , which is thought to be the dorsal side . Impairing Hbox12 function , by co-injection of the hd mRNA , efficiently supplanted the p38 asymmetry , as shown by the uniform nuclear GFP staining in roughly 80% ( n > 250 ) of the resulting embryos observed at the 60-cell stage ( Figure 8B ) . Thus , loss of hbox12 function revoked the transitory phase of p38 inactivation in dorsal cells . Reciprocally , uniform ectopic expression of the full-length hbox12 mRNA completely reversed this pattern , restraining p38-GFP to the cytoplasm in all blastomeres of early embryos ( >80% , n > 250; Figure 8C ) . 10 . 7554/eLife . 04664 . 015Figure 8 . Functional correlation between hbox12 and p38 MAPK . ( A–C ) Embryos injected with the p38-GFP mRNA either alone ( A ) or in the presence of the hd mRNA ( B ) or the hbox12 mRNA ( C ) , respectively , and analyzed by live confocal microscopy at the 60-cell stage . Images are internal projections of 10–15 sections from a confocal z-series . ( D ) At the 4-cell stage , one blastomere of p38-GFP mRNA injected embryos was co-injected with the hbox12 mRNA , and the resulting embryos were examined by live confocal microscopy at the early blastula stage . Representative examples of control p38-GFP mRNA injected ( E ) and double-injected ( F ) embryos are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 015 To exclude that the ubiquitous p38-GFP nuclear clearance did reflect a visualization artefact , we analyzed the p38 activation in embryos misexpressing hbox12 in one blastomere at the 4-cell stage , rather than into the zygote . Indeed , in this experimental assay , three-fourths of each embryo that resulted from the uninjected blastomeres provides an internal control . Thus , following injection of the p38-GFP mRNA into zygotes , the functional hbox12 mRNA was injected , along with TRCD , into one blastomere of embryos at the 4-cell stage , and the resulting embryos were scored for GFP localization at the early blastula stage ( Figure 8D ) . At this time , active p38 had recovered its ubiquitous distribution , as indicated by the GFP staining present in all nuclei of control embryos ( 92% , n = 200 ) injected with the p38-GFP mRNA alone ( Figure 8E ) . Remarkably , in almost all the double-injected embryos ( 87% , n > 200 ) , the progeny of the blastomere that received exogenous hbox12 mRNA firmly detained GFP staining within the cytoplasm ( Figure 8F ) , consistent with p38 being a downstream target of Hbox12 . Importantly , nuclear TRCD-fluorescence clearly highlights that the hbox12-misexpressing cells were perfectly intact and divided synchronously with respect to surrounding cells of the double injected embryos . It should be emphasized that at the 60-cell stage p38 is normally activated in a sector encompassing about 250° ( Figure 8A ) . On the basis of this observation , we reasoned that inhibition of p38 activity through clonal misexpression of hbox12 could only be attained , at most , in about one third of such a spatial domain . Therefore , it was not surprising to observe that the double-injected embryos developed as normal-looking pluteus larvae ( not shown ) . To sum up , these results very well correlate with either the ectopic or reduced expression of nodal in hd-injected and hbox12 overexpressing embryos , respectively . Most importantly , our findings also strongly support the hypothesis that Hbox12 defines the future dorsal side of the embryo by transient inactivation of p38 activity at a key time point , thereby allowing the asymmetric expression of nodal on the opposite side .
During early development of bilaterian embryos symmetry breaking is imposed through establishing of distinct polarities , which are precursors of the larval axes . Although polarization is morphologically not apparent in the zygote , fates of different embryonic regions are patterned on the axial coordinate system by a combination of maternally inherited factors and differential zygotic gene expression ( De Robertis , 2009 ) . In sea urchins , the embryonic DV axis formation is intimately linked to a small group of cells that specifically begin to express nodal by the early blastula stage and behaves as a DV organizing centre . Production of the Nodal ligand by these cells is pivotal for both defining of their ventral identity and accomplishing the DV patterning program ( Duboc et al . , 2004; Flowers et al . , 2004; Yaguchi et al . , 2007 ) . Indeed , Nodal signaling locally induces the production of BMP2/4 , a diffusible relay molecule which is translocated on the opposite side of the embryo to specify dorsal cell fates ( Angerer et al . , 2000; Lapraz et al . , 2009; Chen et al . , 2011 ) . The DV organizer is also a source of Lefty , which restricts Nodal signaling to the ventral side , and Chordin , which prevents BMP2/4 signaling in the ventral ectoderm ( Bradham et al . , 2009; Lapraz et al . , 2009 ) . Because polarization requires the concerted action of all these secreted proteins , unveiling the mechanisms that allow competence for spatial positioning of the DV organizer activities is crucial for understanding embryonic patterning . The data presented here establish that hbox12 is a key upstream gene in patterning the DV axis of the sea urchin embryo , where it functions to prevent ectopic activation of nodal transcription within the prospective dorsal ectoderm . To our knowledge , hbox12 represents the earliest known zygotic regulatory gene expressed by non-organizer cells and involved in the restriction of the DV organizer field . As a first piece of evidence to support this assertion , we found that overexpression of hbox12 severely perturbs DV polarity in embryos ubiquitously translating the synthetic full-length hbox12 mRNA as well as in chimeric specimens bearing the hbox12 transcript in the animal hemisphere . An almost perfect equivalence of effects is obtained by the expression of HD-En obligate repressor , clearly indicating that Hbox12 acts as a transcriptional repressor in the embryo . Remarkably , the transcript abundance of nodal suffered a dose-dependent attenuation following misexpression of hbox12 through injection of either hbox12 or hd-En synthetic mRNAs into zygotes . As a consequence , gsc and tbx2/3 genes , which are differentially expressed along the DV axis , downstream to nodal signaling , were also down-regulated in both the experimental assays . The notion of an important role of hbox12 in limiting the DV organizer function is even further strengthened by the loss-of-function experiments following injection of the hd RNA , which revealed that axial specification strayed off the canonical track , eventually culminating in strong ventralization . In particular , the morphology of the hd-injected embryos was steadily spherical during development , never evolving into the characteristic easel-like shape of unperturbed embryos . Their ectoderm did not partition into morphologically distinguishable domains , PMCs were radially distributed around the straight archenteron , forming a mispatterned circular-shaped skeleton , and pigment cells were not produced . As described , all these phenotypic traits have been reported to arise following ectopic expression of nodal ( Bradham and McClay , 2006; Duboc et al . , 2004; Saudemont et al . , 2010; Yaguchi et al . , 2010 ) . Accordingly , molecular analyses revealed a dramatic expansion of the nodal expressing domain . The specificity exerted by HD on both nodal and gsc gene transcription is well supported by the absence of effect on these marker genes following overexpression of the isolated homeodomain of Otp , indicating individual discrimination of the two homeodomains . Furthermore , clonal expression of HD imposed by either blastomere transplantation experiments or gene transfer assays unequivocally highlights that Hbox12 action in presumptive dorsal ectoderm cells is necessary to harmoniously pattern the embryo along its DV axis . This evidence is even better corroborated by the fact that injection of the morpholino oligonucleotide against nodal mRNA into a single blastomere at the 4-cell stage could substantially restore the DV polarity of embryos ventralized by the injection of hd RNA . Lineage tracing clearly showed that the progeny of the injected blastomere was invariably embedded into the dorsal structures of the rescued larvae . Such a result is fully consistent with the regulative nature of the DV axis specification , historically highlighted since the experiment of Driesch demonstrating the totipotency of each blastomere of the 4-cell stage embryo ( Driesch , 1892 ) . In fact , cell fates along the DV axis are progressively determined through early development by means of intercellular signaling ( Hurley et al . , 1989; Duboc et al . , 2004 ) . Accordingly , hbox12 transcription is not activated cell-autonomously , being almost undetectable in dissociated embryos ( Di Bernardo et al . , 1995 ) . Altogether , our findings show that the equilibrium of DV patterning is severely disrupted from the beginning in the absence of the Hbox12 function . The events that drive the transcriptional regulation of nodal are not completely understood . An interesting line of questioning to pursue in the future would be to evaluate whether Hbox12 directly represses nodal transcription in dorsal cells . Intriguingly , several consensus binding sites for homeodomain-containing factors do exist within the promoter sequence of the nodal gene ( Range et al . , 2007 ) . On this basis , we cannot exclude the direct association of Hbox12 to the cis-regulatory apparatus of nodal ( Figure 9 ) . 10 . 7554/eLife . 04664 . 016Figure 9 . Model for establishment of the DV organizing centre in the sea urchin embryo . In the early embryo , hbox12 transcription is initiated by combinatorial positive inputs from Otx and probably Sox in the future dorsal ectoderm ( Cavalieri et al . , 2008 ) . hbox12-dependent suppression of nodal gene expression in these cells is mediated by the transient inactivation of p38 and/or probably by direct repression . On the ventral side of the embryo , hbox12 expression is negatively regulated by unidentified repressors ( Cavalieri et al . , 2008 ) . In these cells , active p38 stimulates nodal expression probably through Oct1/2 or other intermediate transcription factors ( Range and Lepage , 2011 ) , allowing the establishment of the DV organizer and patterning along the secondary axis . DOI: http://dx . doi . org/10 . 7554/eLife . 04664 . 016 On the other hand , several lines of evidence indicate that redox signaling is involved in transcriptional activation of nodal . Old and new intriguing experiments suggest that a respiratory gradient in the early sea urchin embryo could bias DV axis orientation ( Pease , 1941; Child , 1948; Czihak , 1963; Coffman and Davidson , 2001 ) . As an additional clue , the mitochondrial enzyme cytochrome oxidase has higher activity in the presumptive ventral side of the early embryo ( Czihak , 1963 ) , probably due to a local enrichment in mitochondria ( Coffman et al . , 2004 ) . A possible link between the redox gradient and the initiation of nodal transcription is represented by the stress activated kinase p38 , which is responsive to reactive-oxygen-species in the sea urchin embryo ( Coffman et al . , 2009 ) and is required for nodal expression ( Bradham and McClay , 2006 ) . In agreement with this hypothesis , we show that Hbox12 is functionally upstream of p38 . This conclusion is firstly supported by the fact that impairing Hbox12 function , by injection of the hd RNA , efficiently supplanted the anisotropic inactivation of a GFP-tagged p38 in the prospective dorsal side of the 60-cell stage embryo . By contrast , and in accordance with the previous result , either the ectopic or clonal expression of the hbox12 mRNA specifically prevents p38-GFP nuclear translocation . These findings strongly suggest that a disrupted distribution of active/inactive p38 realistically accounts for either the ectopic or reduced expression of nodal in hd-injected and hbox12 overexpressing embryos , respectively , and concomitant abnormalities in DV patterning . Taken together , these results show clearly that the negative regulation of nodal gene transcription in presumptive dorsal cells is under the control of hbox12 by means of specific p38 inactivation during early steps of embryogenesis ( Figure 9 ) . The mechanism by which Hbox12 suppresses p38 function remains an open question . As the functional status of p38 depends upon the balance between specific kinase and phosphatase activities , Hbox12 might switch such a balance in dorsal cells . For instance , this could be accomplished by suppressing locally the expression of the kinase involved in p38 activation . In a distinct scenario , Hbox12 might work in a regulatory tandem together with a downstream repressor , constituting a so-called double-negative gate which in turn would allow the dorsal-restricted expression of the phosphatase acting on p38 . Such an exclusion effect is not uncommon , especially during early embryogenesis , and numerous examples have been described across metazoans ( Oliveri and Davidson , 2007 ) . Another fascinating possibility will require learning whether p38 directly interacts with the Hbox12 regulator as it does with other homeodomain-containing transcription factors ( Houde et al . , 2001; Zhou et al . , 2013 ) . Whatever is the mechanism , our results shed new light on the understanding of symmetry breaking events acting upstream of Nodal signaling for the establishment of the DV organizer activity during early embryogenesis of the sea urchin . This makes hbox12 the earliest regulatory gene integrated in the molecular circuit that initiate DV axis formation in the embryo . The identification and characterization of potential asymmetric input drivers involved in the control of hbox12 transcription might further improve our understanding of the nature of DV polarization . On this basis , an intriguing issue to be elucidated would be to know whether maternal redox information somehow integrates at the level of the cis-regulatory apparatus of the hbox12 gene . Experiments are currently underway to explore this possibility . Among the paired-like class of homeodomain-containing proteins identified in sea urchins , Pmar1 from Strongylocentrotus purpuratus ( Oliveri et al . , 2002 ) , and its ortholog Micro1 from Hemicentrotus pulcherrimus ( Kitamura et al . , 2002 ) , appeared to be the most similar to Hbox12 . We reasoned that if they were orthologous genes they should show high identity , same cis-regulation , as well as spatiotemporal expression pattern , and function . For instance , the homeodomain of the Otx regulator is 100% identical between S . purpuratus , H . pulcherrimus and P . lividus and fulfils the same function . By contrast , Hbox12 and Pmar1/Micro1 proteins display only an average of 74% identity , with differences found even in the homeodomain . The spatial expression pattern , and to some extent the timing of activation , of hbox12 and pmar1 are profoundly different ( Figure 1; Di Bernardo et al . , 1995; Oliveri et al . , 2002 ) . In addition , while pmar1 expression is thought to be cell-autonomous in micromeres ( Oliveri et al . , 2002 ) , hbox12 transcription depends on cell–cell interactions ( Di Bernardo et al . , 1995 ) . Notably , we have demonstrated that the hbox12 cis-regulatory system is transiently active in the prospective dorsal ectoderm , congruent with the spatiotemporal restriction of the endogenous gene ( Cavalieri et al . , 2008 ) . Given this premise , a clear cut discordance between the functional outputs of Pmar1 and Hbox12 was expected . Indeed , Pmar1 act as a anti-repressor inhibiting the transcription of the ubiquitous repressor HesC , which negatively regulates the repertoire of early micromere specification genes ( Oliveri et al . , 2003; Yamazaki et al . , 2005; Revilla-i-Domingo et al . , 2007 ) . Our results rather suggest an involvement of Hbox12 at the top of the regulatory hierarchy implicated in the polarization of the embryo along the DV axis . Noteworthy , multiple copies of both hbox12 and pmar1/micro1 genes are clustered in P . lividus and S . purpuratus/H . pulcherrimus genomes respectively , suggesting that numerous rounds of duplication have occurred at these loci during evolution . From this and aforementioned considerations , we suppose that hbox12 and pmar1 could be paralogs , rather than orthologs . Considering that the time point for the split between the Strongylocentrotid species and the more recent Parechinid P . lividus species is estimated to be about 35–50 Myr ( Smith , 1988 ) , in such a speculative scenario hbox12 may have arisen by a P . lividus lineage-specific duplication from a common ancestor with pmar1 . In this regard , of interest is the sobering clue that the two cited sea urchin species differ in aspects of DV axis determination . For instance , in S . purpuratus the DV axis passes through a plane about 45° clockwise from the first cleavage furrow ( Cameron et al . , 1989 ) , indicating that secondary axis specification is initiated between fertilization and first cleavage , which is consistent with the asymmetric distribution of mitochondria within eggs and early embryos of this species ( Coffman et al . , 2004 ) . Classical studies in P . lividus embryos instead demonstrate that DV axis is randomly oriented with respect to the first cleavage plane , and that it is established between the fifth and eighth cleavage stage ( Horstadius and Wolsky , 1936 ) , which broadly corresponds to the peak of hbox12 expression .
Microinjection was conducted as described ( Cavalieri et al . , 2007 , 2009a ) . Synthetic mRNAs were resuspended in 30% glycerol and , in selected experiments , TRCD ( Molecular Probes , Italy ) was added at 5% . For overexpression experiments , capped mRNAs were synthesized from the linearized pCS2 constructs using the mMessage mMachine kit ( Ambion , Italy ) . Approximately 1–2 pl of the purified RNAs were then injected at the following concentrations: hbox12 and strim1 out-of-frame , 0 . 01–0 . 4 pg/pl; hd , hd-GFP , and otp-hd , 0 . 4 pg/pl . Mo-nodal was instead injected at 0 . 5 mM along with RDLX . For DNA constructs , approximately 5000 molecules of the linearized phbox12-HD-GFP or the control phbox12-GFP transgenes were injected per zygote . The phbox12-GFP transgene corresponds to the construct initially referred to as 1 . 45GFP ( Cavalieri et al . , 2008 ) . Such a construct contains a genomic fragment including 1 . 45 kb of the promoter sequence of hbox12 , abutting at the 3′ end the ATG start codon fused in frame with the GFP coding sequence ( further details are provided in Cavalieri et al . , 2008 ) . For all experiments , several hundreds of injected embryos were observed and each experiment was repeated at least three times with different batches of eggs . Recombination of animal and vegetal halves were carried out as described ( Cavalieri et al . , 2011 ) . Briefly , control or injected P . lividus embryos at the 16-cell stage were transferred into a modified Kiehart chamber in Ca2+-free sea water and manipulated with fine glass needles under a Leica M165FC stereomicroscope equipped with micromanipulators ( Narishige , UK ) . After surgery , the embryos were returned to regular sea water and reared until the desired stage . Transplanted and non-chimeric injected embryos at the desired stage were harvested , mounted on glass slides and examined under a Leica DM-4500B upright fluorescent microscope . Digital images were captured and processed using Adobe Photoshop CS6 . Reverse-transcription and qPCR analysis were performed as described ( Cavalieri et al . , 2009b , 2011 , 2013 ) . Briefly , total RNA from batches of unfertilized eggs and embryos grown at the desired stage was extracted by using the Power SYBR Green Cells-to-CT kit ( Ambion , Italy ) and reverse transcribed following the manufacturer's recommendations . The resulting cDNA sample was further diluted and the equivalent amount corresponding to one embryo was used as template for qPCR analysis , using the oligonucleotide primers indicated in Supplementary file 1 . qPCR experiments were performed from two distinct batches and all reactions were run in triplicate on a 7300 Real-Time PCR system ( Applied Biosystems , Italy ) using SYBR Green detection chemistry . ROX was used as a measure of background fluorescence and , at the end of the amplification reactions , a ‘melting-curve analysis’ was run to confirm the homogeneity of all amplicons . Calculations from qPCR raw data were performed by the RQ Study software version 1 . 2 . 3 ( Applied Biosystems , Italy ) , using the comparative Ct method . Primer efficiencies ( i . e . , the amplification factors for each cycle ) were found to exceed 1 . 9 . In every experiment , a no-template control was included for each primers set . A cytochrome oxidase or the mbf1 mRNA , which are known to be expressed at a constant level during development ( Cavalieri et al . , 2008 , 2009b ) , were used to normalize all data , in order to account for fluctuations among different preparations . For developmental expression analysis of hbox12 and nodal , the number of transcripts per embryo at the 60-cell stage was estimated assuming a reference standard number of 1000 copies/embryo of the z12 mRNA ( Wang et al . , 1995; Materna et al . , 2010 ) . Chromogenic whole mount in situ hybridization procedure was performed as described ( Cavalieri et al . , 2011 ) , with Digoxigenin-labeled antisense RNA probes and staged embryos . For the simultaneous detection of hbox12 and nodal transcripts we followed the double two-color WMISH procedure ( Cavalieri et al . , 2011 ) . In this experiment , embryos were co-hybridized with a Digoxigenin-labeled hbox12 probe along with a Fluorescein-labeled nodal probe . Probes were then sequentially revealed using alkaline phosphatase-conjugated antibody with NBT/BCIP ( for the dig-probe ) or BCIP ( for the fluo-probe ) as chromogenic substrate . | Embryos begin as a collection of identical cells . As the embryo develops further , the cells in different regions must take on different structures and roles in order to form the complex tissues and organs seen in the fully developed organism . Therefore , a key task in early development is to inform cells where they are in a developing embryo . Signaling proteins released by special groups of organizing cells are responsible for providing the information about where a cell is located . Networks of genes controlled by these proteins then inform embryonic cells of where they are and what they should , or should not , become . One such signaling protein is called Nodal , and is needed to perform a number of tasks in the developing embryo , including helping to form the basic tissues of the organism . Many animals depend on Nodal to develop correctly—from mice and humans , to zebrafish and sea urchins . During sea urchin development , Nodal establishes where the mouth of a larva forms , setting up what is called the dorsal/ventral axis of the embryo; this separates the front and back of the embryo . To do so , the Nodal protein is mostly produced at the front of the embryo . Although much is already known about the network of genes that the Nodal protein controls , the genes and proteins that ensure that the initial source of Nodal is present at the right time and place are largely unknown . Another protein called Hbox12 was also thought to be important for setting up the dorsal/ventral axis . Now , Cavalieri and Spinelli reveal that Hbox12 regulates Nodal during the development of a sea urchin embryo . In the early developing sea urchin , the gene that produces Hbox12 is activated in the region of the embryo that will become its back , directly opposite where Nodal is present . This activation normally occurs just before the gene that produces Nodal is turned on . If the hbox12 gene function is impaired , the Nodal protein is produced in both the front and the back sections of the embryo . Conversely , if Hbox12 is introduced into regions where Nodal is present , the amount of Nodal decreases . Furthermore , disrupting Hbox12 prevents any signs of the dorsal/ventral axis forming . Cavalieri and Spinelli propose that Hbox12 inhibits the production of Nodal by briefly inactivating another protein that is required to activate the nodal gene . By doing so , Hbox12 sets up the dorsal/ventral axis by restricting Nodal to the cells that will make up the front half of the embryo . Most complex organisms have asymmetric bodies , and failure to establish these body asymmetries can result in disease and other disorders in humans . Deciphering how the dorsal/ventral asymmetry in the sea urchin embryo is established should improve our understanding of how the mechanisms that form body shapes have evolved . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2014 | Early asymmetric cues triggering the dorsal/ventral gene regulatory network of the sea urchin embryo |
The mammalian circadian clock is driven by a transcriptional–translational feedback loop , which produces robust 24-hr rhythms . Proper oscillation of the clock depends on the complex formation and periodic turnover of the Period and Cryptochrome proteins , which together inhibit their own transcriptional activator complex , CLOCK-BMAL1 . We determined the crystal structure of the CRY-binding domain ( CBD ) of PER2 in complex with CRY2 at 2 . 8 Å resolution . PER2-CBD adopts a highly extended conformation , embracing CRY2 with a sinuous binding mode . Its N-terminal end tucks into CRY adjacent to a large pocket critical for CLOCK-BMAL1 binding , while its C-terminal half flanks the CRY2 C-terminal helix and sterically hinders the recognition of CRY2 by the FBXL3 ubiquitin ligase . Unexpectedly , a strictly conserved intermolecular zinc finger , whose integrity is important for clock rhythmicity , further stabilizes the complex . Our structure-guided analyses show that these interspersed CRY-interacting regions represent multiple functional modules of PERs at the CRY-binding interface .
Life on Earth evolved a self-sustaining molecular timing system that synchronizes cellular activities with the solar day . This endogenous clockwork prepares an organism for periodic environmental fluctuations and coordinates numerous physiological and behavioral processes ( Reppert and Weaver , 2002 ) . At the molecular level , the mammalian circadian clock operates through an auto-regulatory transcription–translation feedback loop composed of four core components—the transcriptional activator proteins , CLOCK and BMAL1 , and the transcriptional repressors , Periods ( PERs ) and Cryptochromes ( CRYs ) . The heterodimeric CLOCK and BMAL1 complex acts as the positive arm of the loop by recognizing E-box elements and promoting the expression of clock-controlled genes , including Per1 , Per2 , Cry1 , and Cry2 . The PER and CRY proteins function as the negative arm of the loop by blocking the activity of CLOCK-BMAL1 and inhibiting the transcription of their own and all other clock-controlled genes . The cyclic accumulation , localization , and degradation of the PER and CRY proteins are necessary to manifest a 24-hr rhythm ( Lowrey and Takahashi , 2011 ) . Earlier studies suggested that CRYs are the predominant inhibitors of CLOCK-BMAL1 ( Griffin et al . , 1999; Kume et al . , 1999 ) . Independent of PERs , overexpressed CRY1 and CRY2 can each potently inhibit the CLOCK-BMAL1-induced transcription of a luciferase reporter gene in cultured cells ( Griffin et al . , 1999; Kume et al . , 1999 ) . This transcriptional repression activity of CRYs likely occurs through their direct interactions with BMAL1 ( Griffin et al . , 1999; Shearman et al . , 2000; Partch et al . , 2014 ) and CLOCK ( Huang et al . , 2012 ) . Despite the important repressor function of CRYs , the PER proteins have been suggested as the rate-limiting factor of the rhythmic negative feedback loop ( Lee et al . , 2001 ) . With its protein abundance tightly regulated during the circadian cycle , PERs mediate the formation of the PER-CRY complexes and their nuclear localization . Once in the nucleus , PERs might physically bridge CRYs and CLOCK-BMAL1 and promote their interactions ( Chen et al . , 2009 ) . The critical role of PERs in driving the molecular clock is underscored by the complete loss of circadian rhythmicity upon constitutive overexpression of PERs , but not CRY1 , in vitro and in vivo ( Chen et al . , 2009; McCarthy et al . , 2009; Ye et al . , 2011 ) . Periodic degradation of PERs and CRYs represents another crucial step in the negative feedback loop . The F-box proteins , β-TrCP and FBXL3 , have been discovered as the key ubiquitin ligases , responsible for promoting the polyubiquitination of PERs and CRYs , respectively ( Shirogane et al . , 2005; Busino et al . , 2007; Godinho et al . , 2007; Reischl et al . , 2007; Siepka et al . , 2007 ) . Phosphorylation of a degron sequence serves as the signal for PER ubiquitination by β-TrCP ( Shirogane et al . , 2005 ) , whereas recognition of CRYs by FBXL3 is made through a large protein-interaction interface without the involvement of a canonical degron motif or any post-translational modification ( Xing et al . , 2013 ) . This CRY-FBXL3 interface is susceptible to disruption by both the CRY cofactor flavin adenine dinucleotide ( FAD ) and the PER proteins , which have been suggested to control the stability of CRYs by directly competing with FBXL3 ( Xing et al . , 2013 ) . Although genetic studies have firmly established a central role of PERs in clock regulation , the molecular mechanisms by which PERs orchestrate the dynamic clock protein network remain elusive . Binding of PERs to CRYs , CLOCK , and BMAL1 have been detected both in vivo and in vitro ( Kiyohara et al . , 2006; Ye et al . , 2011; Partch et al . , 2014 ) . However , the role of PER2 in coordinating the repression complex assembly is controversial . In addition , how PER–CRY interaction might interfere with FBXL3 for CRY binding also remains unclear . Here , we report the crystal structure of a PER2–CRY2 complex , which provides the missing structural framework for understanding the multiple functions of PERs in driving the molecular clock .
Mammalian PER1 and PER2 share ∼50% sequence identity and a common domain architecture comprised of tandem N-terminal PER-ARNT-SIM ( PAS ) domains , a central CK1δ/ε-binding region , and a ∼100 amino acid long C-terminal CRY-binding domain ( CBD ) , which is necessary and sufficient for CRY binding ( Yagita et al . , 2002 ) . The isolated PER2 CBD can stabilize CRY1/2 in vivo and compete with FBXL3 for CRY1/2 binding in vitro ( Chen et al . , 2009; Xing et al . , 2013 ) . In mouse embryonic fibroblasts ( MEFs ) , overexpression of PER2-CBD alone was able to completely disrupt the circadian bioluminescence rhythm of the luciferase activity of a PerLuc reporter gene ( Chen et al . , 2009 ) . To first characterize the PER–CRY interaction , we performed an alanine-scanning mutagenic analysis of PER2-CBD . We initially targeted stretches of residues strictly conserved among vertebrate PER1/2 orthologs ( Figure 1A , Figure 1—figure supplement 1 ) . Surprisingly , none of the 10 single mutants , which were distributed along the length of the CBD , showed any detectable defect in CRY1 binding ( Figure 1—source data 1 ) . The PER2–CRY1 interaction was only abolished when alanine mutations were simultaneously introduced to two adjacent stretches of residues in the C-terminal , but not N-terminal half of PER2-CBD ( Figure 1B ) . These results suggested an unusual binding mode of PER2-CBD onto CRYs and the importance of the C-terminal half of the CBD in complex formation . 10 . 7554/eLife . 03674 . 003Figure 1 . Overall structure of the murine CRY2–PER2 complex at 2 . 8 Å . ( A ) PER2 CBD sequence alignment . 49% of PER2 CBD residues interact with CRY2 ( blue dots ) . The zinc-coordinating residues are conserved throughout vertebrates ( highlighted in yellow ) . Blue and green boxes correspond to the mutA and mutB constructs , respectively , and indicate regions of PER2-CBD that were mutated to alanines . Dashed lines indicate crystallographically disordered regions . Black squares indicate residues mutated under structure guidance . ( B ) Co-immunoprecipitation of mutant PER2-CBD-FLAG constructs , only mutB was able to abolish CRY1-MYC binding . Western blot of an immunoprecipitaion of COS7 cells transfected with PER2-NLS-FLAG and CRY1-MYC . Proteins were precipitated with α-FLAG and then analyzed by Western blots using α-MYC and α-FLAG . ( C ) CRY2 PHR ( gray ) adopts an overall fold identical to its apo and complexed forms ( e . g . , FAD , FBXL3 , and KL001 ) . PER2 CRY-binding domain ( CBD ) ( orange ) shows a highly extended binding mode around CRY2 . PER2 flanks the CRY2 C-terminal helix and coordinates a zinc ion with CRY2 within a CCCH-type intermolecular zinc finger motif . ( D ) Crystallographic data identify the location of alanine scanning mutants . Importantly , the mutB construct is centered around the CRY2 C-terminal helix . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 00310 . 7554/eLife . 03674 . 004Figure 1—source data 1 . Alanine scanning mutants of PER2 CBD . Each individual mutation was unable to disrupt PER-CBD-CRY1 binding . Only the combined mutation of two nearby stretches of residues on the C-terminal half of PER2-CBD was able to abolish binding . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 00410 . 7554/eLife . 03674 . 005Figure 1—figure supplement 1 . Sequence alignment and structural elements of vertebrate CRY . Alignment and secondary structure assignments of CRY2 orthologs from Mus musculus ( Mm ) , Homo sapiens ( Hs ) , Gallus gallus ( Gg ) , Danio rerio ( Dr ) , and Xenopus laevis ( Xl ) . Strictly conserved residues are colored in red . Blue and green dots indicate mPER2-CBD- , and hFBXL3-interacting residues , respectively . Yellow squares indicate residues that interact with both PER2 and FBXL3 . Black dots indicate residues that are involved in zinc coordination . Colored boxes represent the boundaries of structurally dynamic loops . Dashed line represents the regions outside the PHR . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 005 Mammalian CRY1 and CRY2 paralogs contain a highly similar photolyase-homology region ( PHR ) and a more diverse Cryptochrome C-terminal Extension ( CCE ) sequence ( Figure 1—figure supplement 1 ) . Their PER-binding activity has previously been mapped to the PHR , which is made of an α/β photolyase domain and an α-helical domain ( Figure 1C ) . Consistent with their high sequence homology ( 86% ) , the crystal structures of CRY1-PHR and CRY2-PHR can be superimposed with a root-mean-square deviation ( RMSD ) of 0 . 43 Å out of 377 aligned Cα atoms . To gain structural insights into the general interaction between PERs and CRYs , we purified a representative PER2-CBD-CRY2-PHR complex and determined its crystal structure at a resolution of 2 . 8 Å ( Table 1 ) . 10 . 7554/eLife . 03674 . 006Table 1 . Data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 006CRY2-PER2Data collection Space groupP41 Cell dimensions a , b , c ( Å ) 97 . 67 , 97 . 67 , 163 . 21 α , β , γ ( ° ) 90 , 90 , 90 Resolution ( Å ) 2 . 9 ( 2 . 8 ) Rmeas0 . 06 ( 0 . 8 ) I/σI18 . 8 ( 2 . 1 ) Completeness ( % ) 99 . 6 ( 98 . 2 ) Redundancy4 . 2 ( 4 . 2 ) Refinement Resolution ( Å ) 42 . 7–2 . 8 No . reflections37541 ( 3671 ) Rwork/Rfree20 . 5/27 . 7 No . atoms9342 Protein9292 Ligand/ion2 Water48 B-factors97 . 3 Protein97 . 5 Ligand/ion114 . 1 Water66 . 3 R . m . s . deviations Bond lengths ( Å ) 0 . 009 Bond angles ( ° ) 1 . 3 PER2-CBD adopts a highly extended structure , devoid of a hydrophobic core . It folds into five α-helices of variable length , which are dispersed along an otherwise linear polypeptide ( Figure 1C ) . In the crystal , PER2-CBD meanders along one side of CRY2-PHR and sinuously wraps around the region . With nearly half of the PER2 residues involved in binding , the two proteins bury a total 2800 Å2 of solvent accessible surface area at the interface , which stretches over a distance of more than 215 Å . This unusually extensive interface provides a plausible explanation for the high-affinity binding between the two clock proteins and their insensitivity to mutational disruption . In comparison to its FBXL3- , KL001- , and FAD-complexed forms , CRY2 adopts the same global fold when bound to PER2-CBD ( Figure 4—figure supplement 2 ) . The largest structural variations take place in two local regions , the interface loop next to the FAD-binding pocket and a serine-rich loop neighboring a secondary pocket ( see below ) . The majority of PER2-contacting residues on CRY2 ( 85% ) are strictly conserved between mammalian CRY1 and CRY2 , suggesting that the two cryptochrome proteins share a common PER2 binding mode . The two stretches of residues , whose alanine mutations abrogated CRY1 binding , are mapped to a loop flanked by two α-helical regions in the C-terminal half of PER2 ( Figure 1D ) . The PER2-CBD α3 helix preceding this loop packs against the long CRY2 C-terminal helix at an approximately 30° angle , while the region C-terminal to the loop locks onto the same CRY2 helix from the other side ( Figure 1C–D ) . Together , these PER2-CBD structural elements encircle the CRY2 C-terminal helix like an U-shaped clamp . Arg501 and Lys503 in the CRY2 C-terminal helix have previously been documented to be important for PER2 binding ( Ozber et al . , 2010 ) . In the crystal , these two positively charged residues of CRY2 project in opposite directions and latch onto the surrounding PER2 regions by forming salt bridges with Asp1167 and Asp1206 , respectively ( Figure 2A ) . To confirm the critical role of the CRY2 C-terminal helix in binding PERs , we mutated two hydrophobic residues , Ile505 and Tyr506 , at the end of this CRY2 helix , which are involved in fixing the α-helix to the rest of the CRY2 α-helical domain ( Figure 2B ) . As expected , mutating both residues to aspartate completely abolished the PER2-binding activity of CRY2 ( Figure 2C ) . The same effect was also achieved when negative charges were introduced to the side chains of a stretch of four nearby residues ( amino acids 1171–1174 ) in the α3 helix of PER2-CBD ( Figure 2B , Figure 2—figure supplement 1A ) . Based on these results , we conclude that the CRY2 C-terminal helix represents a key anchoring site for PER2 binding . 10 . 7554/eLife . 03674 . 007Figure 2 . CRY2 C-terminal helix is the central locus of both PER2 and FBXL3 interactions . ( A ) PER2 ( orange ) forms three salt-bridges along CRY2 C-terminus helix ( gray ) R501 and K503 have been previously reported as critical binding residues . ( B ) A close-up view of the PER2-CRY2 interface at the end of CRY2 C-terminal helix . While the upper portion of the CRY2 C-terminal helix maintains ionic interactions with PER2 , the lower is predominantly mediated by hydrophobic interactions . CRY2 and PER2 residues chose for subsequent mutational analysis are shown in sticks . ( C ) Concurrent mutations of hydrophobic residues on the CRY C-terminal helix ( I505D and Y506D ) prevent PER-CRY complex formation . Co-immunoprecipitations were performed with transfected full-length PER2-V5 and MYC-CRY2 in HEK293 cells with α-MYC beads and analyzed by Western blotting using α-V5 and α-MYC . See Figure 2—figure supplement 1A for corresponding PER2 mutants . ( D ) Surface mapping of FBXL3- and PER2-binding sites on CRY2 . Residues that share contacts with PER2 and FBXL3 are colored in yellow and are clustered along the C-terminal helix . Other residues involved in binding PER2 and FBXL3 are colored in orange and green , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 00710 . 7554/eLife . 03674 . 008Figure 2—figure supplement 1 . Mutational and structural analysis of the PER2-CRY2 interface . ( A ) Diminished PER2-CRY2 interaction was replicated in a co-immunoprecipitation assay , in which the CXXC motif of the full-length FLAG-tagged PER2 protein was mutated to four alanines . Co-immunoprecipitations were performed with transfected full-length PER2-V5 and MYC-CRY2 in HEK293 cells with α-MYC beads . ( B ) Superimposition analysis demonstrates the direct competition of PER2-CBD ( orange ) and FBXL3 ( green ) binding to CRY ( gray ) , which is centered around the CRY2 C-terminal helix ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 008 The close interaction between the C-terminal half of PER2-CBD and CRY2 C-terminal helix is immediately reminiscent of the docking mode between FBXL3 and CRY2 . In the crystal structure of the FBXL3-CRY2 complex , the leucine-rich repeat ( LRR ) domain of FBXL3 engages CRY2 at the same site as PER2-CBD does in the PER2-CRY2 complex . The interface between FBXL3-LRR and CRY2 is also centered around the long C-terminal helix of the Cryptochrome protein . In fact , the CRY2 surface regions involved in contacting FBXL3-LRR and PER2-CBD share extensive overlapping regions ( Figure 2D ) . Superposition analysis reveals that FBXL3 and PER2 cannot be simultaneously engaged with CRY2 without clashing into each other ( Figure 2—figure supplement 1B ) . PERs , therefore , have the capability of protecting CRYs from FBXL3-mediated ubiquitination and degradation by directly competing with the ubiquitin ligase for binding CRYs . Amino acid sequence alignment of vertebrate PER1/2 orthologs reveals that their sequence conservation ends at a CXXC motif near the C-terminus ( Figure 1A ) . In the complex structure , these two cysteine ( C1210 and C1213 ) residues face a pair of cysteine and histidine residues in CRY2 ( C432 and H491 ) , which are also invariant among vertebrate CRY1/2 proteins ( Figure 1—figure supplement 1 ) . Together , these four residues sequester a strong density at the center , hinting at the coordination of a Zn2+ ion at the end of the PER2–CRY2 interface ( Figure 3A , Figure 3—figure supplement 1A ) . Indeed , we were able to validate the identity of the Zn2+ ion by both anomalous dispersion measurements and inductively coupled plasma mass spectrometry ( Figure 3—figure supplement 1A , B ) . Although a Zn2+ ion has been previously reported to mediate protein–protein interactions ( Somers et al . , 1994 ) , to our knowledge , this is the first CCCH-type intermolecular zinc finger that has been identified in a protein complex . Interestingly , the electron density of the PER2 sequence preceding the CXXC motif is not as strong as other regions of PER2-CBD , suggesting that the intermolecular zinc finger might have evolved to stabilize a flexible region of the PER–CRY interaction by acting as a ‘molecular clasp’ . 10 . 7554/eLife . 03674 . 009Figure 3 . The intermolecular zinc finger is important for PER2–CRY2 complex formation . ( A ) Four conserved , contributing residues from PER2 ( C1210 and C1213 ) and CRY2 ( C432 and H491 ) form a CCCH-type zinc finger . ( B ) GST-pull-down assay with recombinant GST-tagged PER2ΔCXXC CBD and untagged CRY2-PHR protein show compromised CRY binding in the zinc finger mutant compared to WT PER2-CBD . ( C ) Similarly diminished interaction was replicated in a co-immunoprecipitation assay . Alanine mutations were introduced to CRY2 zinc-coordinating residues , C432 and H491 , individually or in combination . Co-immunoprecipitations were performed with transfected full-length PER2 and CRY2 in HEK293 cells with α-MYC beads . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 00910 . 7554/eLife . 03674 . 010Figure 3—figure supplement 1 . Analysis of the intermolecular zinc finger . ( A ) Zinc-coordinating residues of PER2-CBD ( orange ) and CRY2 ( gray ) with zinc anomalous signal ( λ = 1 . 284 Å ) contoured at 7σ ( blue mesh ) . ( B ) Inductively coupled plasma mass spectrometric analysis of metal isotopes . Purified PER2-CBD-CRY2 complex was dehydrated , dissolved in concentrated HNO3 overnight , diluted to 1% vol/vol HNO3 , and titrated . Zn ( purple ) isotopes ( 64Zn , 66Zn , 67Zn , 68Zn , 70Zn ) were the only ones that showed a greater than sixfold increase in mean signal intensity above the blank , dose-dependent increase , and approximated the predicted intensity of the standard ( orange line ) . 24Mg ( green ) and 56Fe ( gray ) are shown as controls . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 010 To assess the role of the intermolecular zinc finger in mediating PER–CRY association , we first tested the CRY2-binding activity of a recombinant mutant PER2-CBD , which lacks the CXXC motif . In comparison to the wild-type polypeptide , the ability of the PER2-CBD mutant to bind CRY2 was substantially compromised ( Figure 3B ) . Similarly weakened interaction was also observed in a co-immunoprecipitation assay , in which the CXXC motif of the full-length PER2 protein or the two zinc-coordinating residues of CRY2 were mutated to alanines ( Figure 2—figure supplement 1A , Figure 3C ) . Together , these results highlight the importance of the intermolecular zinc finger in strengthening the PER–CRY interface . Cryptochromes and DNA photolyases belong to the same superfamily of flavoproteins , whose common PHR fold is characterized by two large surface pockets , one for binding flavin adenine dinucleotide ( FAD ) and the other for binding a photoantenna cofactor , which is used by light-sensitive photolyases to catalyze FAD-dependent DNA repair ( Glas et al . , 2009; Figure 4A ) . Previously , we have identified the FAD-binding pocket as a regulatory ‘hot spot’ , which is targeted by FAD , the extreme carboxyl tail of FBXL3 , and the clock-modulating small molecule , KL001 ( Nangle et al . , 2013; Xing et al . , 2013; Figure 4—figure supplement 1A ) . However , the functional significance of the secondary pocket remained unexplored . 10 . 7554/eLife . 03674 . 011Figure 4 . The secondary pocket is involved in CRY-CLOCK-BMAL1 complex assembly and repression . ( A ) Relative positions of the two large pockets on CRY2 . ( B ) Surface representation of CRY2 with side chains of the serine loop shown in sticks . PER2 α1 helix inserts into a hydrophobic cleft . Compared to other CRY2 complexed forms , the serine loop flips up and engages PER2 . ( C ) The serine loop lies opposite to the CRY α4 helix , which together frame the secondary pocket , the α4 helix contains three residues ( CRY1 G106R and R109Q , CRY2 E121K ) , whose mutations result in a weak repression phenotype . ( D–F ) Co-immunoprecipitation assays show that the CRY1 R109Q mutant is unable to bind CLOCK-BMAL1 , but retains PER1 and PER2 binding . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 01110 . 7554/eLife . 03674 . 012Figure 4—figure supplement 1 . Locations of structurally plastic loops on CRY2 . ( A ) The phosphate-binding loop and interface loop frame opposite sides of the FAD-binding pocket . Superimposition analysis shows the FAD-binding pocket as a regulatory hotspot , which can bind metabolic cofactor , FAD ( yellow ) , clock-modulating small molecule , KL001 ( cyan ) , and FBXL3 C-terminal tail ( green ) . ( B ) Locations of CRY2 loops; interface loop ( yellow ) , phosphate-binding loop ( red ) , serine loop ( purple ) , and protrusion loop ( light blue ) , which in light-sensitive CRYs occludes part of the FAD-binding pocket but is pushed outward and maintains an open FAD-binding pocket in vertebrate CRYs . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 01210 . 7554/eLife . 03674 . 013Figure 4—figure supplement 2 . CRY-PHR superposition: including CRY1 apo ( red ) , CRY2 apo ( light blue ) , KL001-bound ( green ) , FAD-bound ( orange ) , FBXL3-bound ( cyan ) , and PER2-CBD-bound ( gray ) CRY . ( A ) Serine loop undergoes a large conformational change after PER2-CBD binding . ( B and C ) The interface loop and phosphate-binding loop are also sites of high structural plasticity . ( D ) Overall CRY-PHR showing the global structure adopts a common fold . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 01310 . 7554/eLife . 03674 . 014Figure 4—figure supplement 3 . Major differences between CRY1-PER2-CBD and CRY2-PER2-CBD complex structures . Superposition of the two structures reveals major structural dissimilarities between the two paralogs at the CRY secondary pocket and a residual fusion-protein sequence ( yellow ) in CRY1-bound PER2-CBD . The PER2-CBD ( dark blue ) N-terminus together with the artifactual sequence ( AGLEVLFQGPDSM ) forms a β-hairpin and induces an inward conformation of the CRY1 ( light green ) serine loop . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 014 In the PER2-CRY2 crystal , the N-terminal half of PER2-CBD diverges from the FBXL3-binding site of CRY2 and reaches the rim of the secondary pocket after traversing around the α-helical domain ( Figure 4A ) . With a highly conserved sequence , the N-terminal end of PER2-CBD is embedded in a V-shaped cleft formed between the two globular domains of CRY2-PHR , burying a PER2 tryptophan residue ( Trp1139 ) at the junction ( Figure 4B , Figure 2—figure supplement 1A ) . One side of the cleft is constructed by a serine-rich loop in CRY2 , which we name as ‘serine loop’ . Distinct from its surrounding regions , this loop adopts different conformations in several available crystal structures of CRY ( Figure 4—figure supplement 2 ) . Remarkably , PER2 binding induces yet another distinct structural configuration of the loop , thereby , defining a unique structural state of the local area next to the secondary pocket . Although CRYs are known to not engage a second cofactor ( Zoltowski et al . , 2011; Xing et al . , 2013 ) , our previous cell-based random mutagenesis screen has identified three residues within this secondary pocket ( Gly106 and Arg109 in CRY1 , Glu121 in CRY2 ) ( Figure 4C ) , whose missense mutations effectively abolished the repressor activity of CRYs ( McCarthy et al . , 2009 ) . Among these three residues , Arg109 is exposed to the solvent and decorates one side of the pocket . Co-immunoprecipitation analysis of the R109Q mutant showed that alteration of this single amino acid is sufficient to abrogate CLOCK-BMAL1 , but not PER1 or PER2 binding ( Figure 4D–F ) . Thus , the secondary pocket of CRYs represents an important docking site for the heterodimeric transcriptional activators . Anchoring of PER2 at the edge of this CRY pocket not only reinforces its function as a previously unrecognized locus for protein–protein interactions , but also suggests a possible role of PERs in modulating the repressor functions of CRYs . To functionally characterize the multiple interfaces on CRYs mapped by the crystal structures , we systematically assessed several representative CRY mutants for their abilities to rescue rhythmicity in Cry-deficient MEF cells . Consistent with previous studies , wild-type CRY1 was able to repress the expression of the P ( Per2 ) -dLuc reporter gene and produce robust bioluminescence rhythms . By contrast , the ‘IY’ mutant of CRY1 , which confers severe structural disruption in the C-terminal helix , failed to restore any level of circadian rhythm , although it has the ability to repress CLOCK-BMAL1 as seen by the constitutively low luciferase signal ( Figure 5A ) . Because the C-terminal helix of CRYs is a critical region for binding both FBXL3 and PERs , this result underscores the importance of CRY ubiquitination and degradation in establishing clock rhythmicity and suggests the ability of CRY1 to inhibit CLOCK-BMAL1 in a PER-independent manner . In agreement , two CRY1 mutants unable to coordinate zinc , C414A and H473A , were also capable of transcriptional repression , even though their PER-binding activities are largely compromised ( Figure 5B , C ) . 10 . 7554/eLife . 03674 . 015Figure 5 . Real-time circadian rescue assays . Cry1−/−/Cry2−/− MEFs were transfected 24 hr after plating with dLuc reporter plasmid and mCry1 expression or mutant vector . 72 hr after transfection , the cells were synchronized with dexamethasone . Bioluminescence ( raw counts/s ) monitoring was performed continuously for 70 s every 10 min using a photomultiplier tube at 37°C . Traces are shown as mean ± SEM and are representative of triplicate samples . Mutants are shown in blue and WT control in black . Only CRY1 , not CRY2 is able to reconstitute robust circadian rhythmicity . ( A ) CRY1 I487D Y488D ( CRY2 I505 Y506 ) ‘IY’ mutant abolishes rhythmicity but maintains repression compared to WT , suggesting that PER is not required for transcriptional repression . ( B and C ) Zinc-coordinating residues on CRY1 C414 and H473 ( CRY2 C432 and H491 ) show blunted rhythm amplitude . ( D ) A nearby cysteine residue , C412 ( CRY2 430 ) , when mutated to alanine , does not show a significantly different phenotype from the WT control . ( E ) A critical residue on the secondary pocket , CRY1 R109 ( CRY2 R127 ) shows a severely weakened repression phenotype when mutated to a glutamine . Traces are shown as mean ± SEM and are representative of duplicate samples . ( F ) Mutations of two serine residues in the serine loop , CRY1 S44D S45D ( CRY2 S62 S63 ) , show near WT rhythmicity and repression but with a 1-hr shorter period . For all mutants , corresponding CRY2 residues are in parenthesis . DOI: http://dx . doi . org/10 . 7554/eLife . 03674 . 015 We noticed that the two zinc finger CRY1 mutants still sustained circadian rhythms . However , they showed defects in their bioluminescence oscillations ( Figure 5B , C ) . Such a phenotype was not observed for a mutant with a nearby residue , Cys412 , mutated to alanine , which did not perturb PER or FBXL3 binding as previously documented ( Figure 5D; Xing et al . , 2013 ) . The contrast between the two zinc finger-defective mutants and the wild-type-like C412A mutant confirms the functional role of the zinc-coordinating residues in the negative arm of the feedback loop . Consistent with its impaired CLOCK-BMAL1 binding activity , the CRY1 R109Q mutant showed significant derepression in the rescue assay ( Figure 5E; McCarthy et al . , 2009 ) . This single amino acid mutation highlights the key role of the secondary pocket of CRYs for repression . Intriguingly , double serine to aspartate mutations ( S44D S45D ) in the nearby serine loop at the opposite side of the CRY secondary pocket completely rescued the circadian rhythm , although the period of the bioluminescence rhythms rescued by the mutant was reliably shorter than the wild-type CRY1 by about 1 hr ( Figure 5F ) . In our co-immunoprecipitation experiments , this double serine mutation weakened PER2 binding to a lesser degree than the zinc finger mutations , which did not elicit a similar period-shortening effect ( Figure 2C ) . Therefore , the period-shortening effect induced by the double serine mutation is likely specific to the defects of the local PER–CRY interface instead of their overall binding . It is conceivable that PERs might engage with CRYs near the CLOCK-BMAL1 docking site to control a periodicity-related step of negative feedback different from what they do at the predominant PER–CRY interface .
Previous studies have established a critical role of PERs in driving the rhythmic negative feedback loop ( Reppert and Weaver , 2002 ) . To fulfill this role , PERs have been suggested to act through multiple mechanisms , including mediating CRY nuclear entry , coupling CRYs to CLOCK-BMAL1 , and competing with FBXL3 to stabilize CRYs . Our structural and mutagenic analyses of the PER2-CBD-CRY2 complex reveal a surprisingly robust binary assembly , which is resilient to mutational disruption . This stable complex is enabled by an extended binding mode of PER2-CBD , which spreads several distinct functional modules over a mostly linear interface . The hallmark of the PER–CRY interactions is its steric incompatibility with the FBXL3–CRY complex , which provides the structural basis for the competition of PERs and the FBXL3 ubiquitin ligase for controlling CRY stability . Interestingly , distant from the FBXL3–CRY interface , PERs also anchor themselves next to the putative CLOCK-BMAL1-binding pocket of CRYs , possibly regulating a specific step of transcriptional repression . Despite intensive genetic and cell-based studies , the precise spatial and temporal steps undertaken by PERs to coordinate transcriptional repression in the molecular clockwork remain unclear . On the one hand , PERs have been reported to be essential for CRYs to interact with CLOCK-BMAL1 ( Chen et al . , 2009 ) . On the other hand , emerging evidence suggests that PERs binding might interfere with complex formation between CRYs and CLOCK-BMAL1 at certain steps during repression ( Ye et al . , 2011; Akashi et al . , 2014 ) . Conceivably , by interacting with the CRY C-terminal helix , PERs could compete with the C-terminus of the BMAL1 transactivation domain for CRY binding ( Czarna et al . , 2011 ) . While detailed biochemical studies are necessary to resolve this controversy , our results offer the structural framework for in-depth mechanistic investigations . Apart from the PER2-CBD-CRY2 complex , the crystal structures of CRY2 have been determined for four additional functional states , apo , FAD- , FBXL3- , and KL001-bound ( Nangle et al . , 2013; Xing et al . , 2013 ) . Together , these structures outline a rich landscape for the functional surfaces of mammalian CRYs , which distinguishes them from other members of the cryptochrome/photolyase superfamily . In their C-terminal α-helical domain , CRYs feature the conserved FAD-binding pocket , which is also targeted by the FBXL3 C-terminal tail and the clock-modulating drug , KL001 . In their N-terminal α/β photolyase domain , CRYs have evolved the secondary pocket into a critical site for CLOCK-BMAL1 binding . Importantly , both CRY surface pockets are demarcated by structural elements with noticeable structural plasticity ( Figure 4—figure supplements 1 and 2 ) . The FAD-binding pocket is framed by the phosphate-binding loop and the interface loop on opposite edges , whereas the secondary pocket is guarded by the serine loop on one side . With the exception of the phosphate-binding loop , both the interface and serine loop have been shown to directly mediate protein–protein interactions . Lastly , the extreme C-terminal α-helix of the mammalian CRYs presents yet another important surface area , which is responsible for the mutually exclusive binding of FBXL3 and PERs . Remarkably , all these molecular interacting sites likely represent an incomplete functional map of CRYs . Numerous mutants identified in our random mutagenesis screen of functionally deficient CRY1 and CRY2 bear mutations of amino acids located outside these sites ( McCarthy et al . , 2009 ) . Future structural studies are needed to paint a complete picture of CRY functional surfaces . Our crystal structure of the PER2-CBD-CRY2 complex unveils a structurally important intermolecular zinc finger , which might function as a stabilizing ‘molecular clasp’ . Although the evolutionary significance of the zinc-coordinating residues is apparent , as evidenced by their strict conservation across vertebrates , the functional significance of this unusual binding interface requires further investigation . On the one hand , the intermolecular zinc finger might be an intermediate product of the still evolving PER–CRY interface . On the other hand , it is plausible that this special protein interaction interface confers sensitivity to the fluctuating abundance of intracellular zinc ( Wang et al . , 2012 ) , which might serve as a tissue-specific clock-modulating ion . During the preparation of this manuscript , the complex structure of mammalian CRY1-PHR and PER2-CBD was reported ( Schmalen et al . , 2014 ) . With high sequence conservation between CRY1 and CRY2 , PER2-CBD adopts a similar CRY-binding mode with a tetrahedral coordination of a zinc ion by an intermolecular CCCH zinc-binding motif . The major structural difference lies at the interface of the N-terminal region of PER2-CBD and the CRY secondary pocket . The CRY1-bound PER2-CBD fragment contains a residual fusion-protein sequence , which forms an artifactual ß-hairpin with the first five amino acids of the PER2-CBD ( Figure 4—figure supplement 3 ) . In contrast to the PER2-bound CRY2 serine loop , but reminiscent of the Drosophila CRY antenna loop ( Zoltowski et al . , 2011 ) , the otherwise disordered ( Czarna et al . , 2013 ) CRY1 serine loop adopts an inward conformation and occludes the secondary pocket . This conformational difference reveals a substantial degree of structural plasticity , which might be necessary for differential binding and regulation at this site . Interestingly , Schmalen et al . ( 2014 ) identified a potential redox sensor involving a disulfide bond near the zinc finger between Cys412 and Cys363 , which modulates CRY1-PER2 binding . However , in our circadian reporter assay , we did not detect any difference between the CRY1 wild type and C412A mutant ( Figure 5D ) . More in-depth analyses can now exploit the specific structural differences between the two complexes to explain the non-redundant roles of the two Cryptochrome proteins . True to their name , Period proteins act as the master timekeepers in the circadian clock pathway , and likely use their multiple functional modules to simultaneously mediate the negative and positive phases of the clock through CRY stability and CRY-CLOCK-BMAL1 repression complex assembly .
The mouse CRY2 ( amino acids 1–512 ) was expressed as a glutathione S-transferase ( GST ) fusion protein in High Five ( Invitrogen , Carlsbad , CA ) suspension insect cells and isolated by glutathione affinity chromatography using buffer containing 20 mM Tris–HCl pH 8 , 200 mM NaCl , 10% glycerol , 5 mM DTT ( dithiothreitol ) . The protein was cleaved on-column by tobacco etch virus ( TEV ) protease then purified further by cation-exchange chromatography . Proteolytically stable murine PER2 ( amino acids 1095–1215 ) was expressed as a GST-fusion protein in Escherichia coli expression system and isolated through glutathione affinity chromatography using buffer containing 20 mM Tris–HCl pH 8 , 300 mM NaCl , 5 mM DTT . The protein was cleaved on-column by TEV protease then purified further by anion-exchange and size-exclusion chromatography . Both proteins were combined , concentrated , and further purified by size-exclusion chromatography using buffer containing 20 mM Tris–HCl pH 8 , 300 mM NaCl , 5 mM DTT , 10% glycerol to establish stoichiometric binding . The crystals of the CRY2-PER2 complex were grown at 4°C by the hanging-drop vapor diffusion method , using 2 μl protein complex sample mixed 2:1 with reservoir solution containing 100 mM HEPES pH 7 . 5 , 200 mM NaCl , 15% PEG 3350 . Diffraction-quality crystals were subjected to a cryo-protectant procedure by gradually increasing the concentration of ethylene glycol to 25% ( vol/vol ) and then frozen in liquid nitrogen . The native and zinc anomalous data sets were collected at the BL8 . 2 . 1 beamline at the Advanced Light Source of the Lawrence Berkeley National Laboratory . Reflection data were indexed , integrated , and scaled with the HKL2000 ( Otwinowski and Minor , 1997 ) . The CRY2-PER2 complex was determined by molecular replacement using CRY2 from the murine CRY2-KL001 complex structure ( PDB:4MLP ) as the search model . The structural models were manually built , refined , and rebuilt with the programs COOT ( Emsley et al . , 2010 ) , PHENIX ( Adams et al . , 2010 ) , and CCP4 ( Winn et al . , 2011 ) . PER2 was built in following density modification . All figures were made using PyMOL ( Schrödinger , LLC ) . Buried surface area was calculated using CNS ( Brunger et al . , 1998 ) . GST-tagged mCRY2 ( amino acids 1–512 ) was over-expressed in High Five insect cells suspension culture . GST-tagged mPER2 WT ( amino acids 1095–1215 ) and GST-tagged mPER2ΔCXXC ( amino acids 1095–1209 ) were over-expressed in E . coli and purified as previously described . Equal volumes CRY2-PHR was incubated with immobilized PER2 at 4°C for 1 hr . Glutathione beads were rigorously washed , and GST-PER2-CRY2 was released from the beads with SDS sample buffer , analyzed by SDS-PAGE and detected by Coomasssie stain . N-terminal Myc-tagged Cry2 ( 0 . 25 μg ) and a C-terminal V5-tagged Per2 ( 0 . 5 μg ) were transfected ( Fugene 6 , Madison , WI ) into HEK293 cells . After 48 hr , cells were harvested and lysed by centrifugation . α-MYC-conjugated beads were used to immobilize MYC-CRY2 . Beads were washed with buffer containing 50 mM Tris–HCl pH 7 . 5 , 100 mM NaCl , 5% glycerol , 0 . 5 mM DTT , 0 . 5% Triton X-100 , protease inhibitor ( 1:50 ) . Protein was released from beads with SDS sample buffer and analyzed by Western blot using α-MYC and α-V5 for CRY2 and PER2 , respectively . Real-time circadian rescue assays performed as described in Ukai-Tadenuma et al . ( 2011 ) . Cry1−/−/Cry2−/− MEFs were plated in 35-mm dishes at a density of 5 × 105 cells per dish . 24 hr later , cells were transfected with FuGene6 with 4 μg of pGL3-P ( Per2 ) -dLuc reporter plasmid and 150 ng of the pMU2-mCry1 expression vector ( Ukai-Tadenuma et al . , 2011 ) or mutant forms of this vector . 72 hr after transfection , the cells were synchronized by a 2-hr incubation in medium ( DMEM/10% FBS/antibiotics ) with dexamethasone ( 1 μM ) . The medium was then replaced with medium prepared from powdered DMEM without phenol red ( Corning 90-013-PB ) containing 4 . 5 g/l glucose and supplemented with 10 mM HEPES pH 7 . 2 , 100 μM luciferin , 1 mM sodium pyruvate , 0 . 035% sodium bicarbonate , 10% FBS , antibiotics , and 2 mM L-glutamine . Bioluminescence monitoring was performed using a LumiCycle ( Actimetrics , Inc . Wilmette , IL ) to record from each dish continuously for ∼70 s every 10 min using a photomultiplier tube at 37°C . | Since the very simplest organisms emerged on earth , the rhythms of life have been synchronized with the rising and setting of the sun . Even the most basic life forms have internal clocks that help them to maintain daily routines and adapt to shifting seasons . In animals , these internal clocks regulate processes such as the release of hormones that wake an animal up and the expression of genes necessary to carry out the activities of daily life . Later on , the clocks then trigger the release of hormones that cause drowsiness and the expression of the genes that are active during rest . In mammals , these internal circadian rhythms are maintained by a feedback loop governed by four key proteins . Two of these proteins—CLOCK and BMAL1—work together to begin a process called transcription , whereby sections of DNA are used as a template to copy the information needed to make a protein . The two activating proteins CLOCK and BMAL1 recognize the sections of DNA where the genes that are controlled by the circadian clock are located and selectively turn on the expression of those genes . Expression of the two other key circadian proteins—Period and Cryptochrome—is switched on by CLOCK and BMAL1 . As Period and Cryptochrome proteins accumulate , they begin to inhibit the activity of CLOCK and BMAL1 , helping to reduce the rate at which the circadian genes are transcribed as the day progresses . Nangle et al . provide new insights into how the Period and Cryptochrome proteins interact with each other , using X-ray crystallography to reveal the molecular level details of the bond between the two proteins . Period stretches out as it ‘embraces’ Cryptochrome . One end of the Period protein then tucks into part of the Cryptochrome structure that is next to a large pocket . This pocket is where the Cryptochrome protein binds to CLOCK and BMAL1 , suggesting that Period can influence whether this binding occurs . The other end of the Period protein covers one end of the Cryptochrome protein . By doing so , enzymes cannot bind there , and so cannot break down Cryptochrome . Nangle et al . also discovered that a finger-like projection that includes a zinc ion acts as a clasp , strengthening the bond between Period and Cryptochrome . These findings help to demonstrate how Period proteins act as a timekeeper that regulates how long Cryptochrome can turn down the activity of CLOCK and BMAL1 . A deeper understanding of the molecular choreography among the four clock proteins holds promise for developing medications to treat the sleep disorders and circadian clock disruptions associated with a modern lifestyle . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Molecular assembly of the period-cryptochrome circadian transcriptional repressor complex |
Nuclear pore complexes ( NPCs ) conduct massive transport mediated by shuttling nuclear transport receptors ( NTRs ) , while keeping nuclear and cytoplasmic contents separated . The NPC barrier in Xenopus relies primarily on the intrinsically disordered FG domain of Nup98 . We now observed that Nup98 FG domains of mammals , lancelets , insects , nematodes , fungi , plants , amoebas , ciliates , and excavates spontaneously and rapidly phase-separate from dilute ( submicromolar ) aqueous solutions into characteristic ‘FG particles’ . This required neither sophisticated experimental conditions nor auxiliary eukaryotic factors . Instead , it occurred already during FG domain expression in bacteria . All Nup98 FG phases rejected inert macromolecules and yet allowed far larger NTR cargo complexes to rapidly enter . They even recapitulated the observations that large cargo-domains counteract NPC passage of NTR⋅cargo complexes , while cargo shielding and increased NTR⋅cargo surface-ratios override this inhibition . Their exquisite NPC-typical sorting selectivity and strong intrinsic assembly propensity suggest that Nup98 FG phases can form in authentic NPCs and indeed account for the permeability properties of the pore .
Cell nuclei lack protein synthesis and therefore import required proteins from the cytoplasm . Nuclei , on the other hand , supply the cytoplasm with ribosomes , mRNAs , tRNAs , and other ‘nuclear products’ . Nuclear pore complexes ( NPCs ) conduct this nucleocytoplasmic transport ( Reviewed by Brohawn et al . , 2009; Hetzer and Wente , 2009; Rothballer and Kutay , 2013 ) . They are embedded into the nuclear envelope ( NE ) and equipped with a remarkable permeability barrier that suppresses an intermixing of nuclear and cytoplasmic contents . This barrier is freely permeable for small molecules but becomes increasingly restrictive as the size of the diffusing species approaches a limit of ≈30 kDa in mass or ≈5 nm in diameter ( Mohr et al . , 2009 ) . Larger objects are essentially excluded from passage but may traverse the pore when bound to an appropriate nuclear transport receptor ( NTR ) . NTRs are in continuous circulation between nucleus and cytoplasm ( reviewed in Görlich and Kutay , 1999; Weis , 2003; Cook et al . , 2007 ) . They select cargo molecules on one side of the NE , traverse the NPC barrier in a facilitated manner , release cargo into the destination compartment , and return to the initial compartment for the next round of transport . There are several NTR categories . Importins and exportins use coupling to the RanGTPase system for active transport against concentration gradients . With molecular masses between 90 and 140 kDa , they are relatively large in size . NTF2 , the import receptor for RanGDP ( Ribbeck et al . , 1998 ) , is smaller ( ≈30 kDa for the homodimer ) . It is structurally unrelated to importins/exportins ( Bullock et al . , 1996 ) but homologous to the Mex67p/Mtr2p heterodimer , which functions in RNA export ( See e . g . , Strasser et al . , 2000 ) . Hikeshi represents yet another category . It mediates nuclear import of Hsp70 ( Kose et al . , 2012 ) . The NPC barrier has a remarkably high capacity , supporting up to 1000 facilitated translocation events or a mass flow of 100 MDa per pore per second ( Ribbeck and Görlich , 2001 ) . Facilitated NPC passage can be completed within a few milliseconds , at least when smaller cargoes are transported ( Yang et al . , 2004; Kubitscheck et al . , 2005; Tu and Musser , 2011 ) . Facilitated translocation per se does not consume metabolic energy ( Weis et al . , 1996; Schwoebel et al . , 1998; Englmeier et al . , 1999; Ribbeck et al . , 1999 ) . Instead , it is based on a higher permeability of NPCs for NTRs as compared to inert molecules ( Ribbeck and Görlich , 2001 ) . NPCs become active and highly efficient cargo pumps , when cargo-release from NTRs is enforced at destination , which is typically achieved by the RanGTPase system ( Görlich et al . , 2003 ) . The NPC barrier is , however , not just an obstacle to overcome . It is also essential for any active directional transport because it prevents a dissipation of the transport-driving RanGTP-gradient and retains already transported cargoes at destination . Facilitated translocation is evident for objects of a wide range of sizes . NTF2 ( diameter: ≈5 nm ) exemplifies a small translocation species . It traverses NPCs 100 times faster than the equally sized GFP ( Ribbeck and Görlich , 2001; Mohr et al . , 2009 ) . 60S ribosomal subunits ( diameter: ≈25 nm ) are on the other end of the allowed size spectrum . Such large cargoes require multiple NTR molecules for efficient NPC passage . At least three shuttling NTRs ( Xpo1/CRM1 , Xpo5 , and Gle2/ Rae1 ) contribute to nuclear export of human 60S ribosomal subunits ( Wild et al . , 2010 ) , while Xpo1/CRM1 , the Mex67 Mtr2 dimer , Arx1 , and Gle2 synergise in exporting S . cerevisiae 60S particles ( Ho et al . , 2000; Bradatsch et al . , 2007; Yao et al . , 2007; Occhipinti et al . , 2013 ) . This NTR-cooperation effect is also evident for smaller cargoes . Nuclear import of an IBB-2xMBP-M3 fusion ( 88 kDa ) , for example , occurs ≈10 times faster if it is simultaneously bound by two importin molecules rather than just a single one ( Ribbeck and Görlich , 2002 ) . Likewise , single molecule tracking revealed that multiple importin molecules are required for transporting a ≈500 kDa M9-β-Gal tetramer through NPCs ( Tu et al . , 2013 ) . FG nucleoporins ( Nups ) are critical for NPC function ( Strawn et al . , 2004 ) . They comprise FG domains with often hundreds of residues and up to 50 FG ( phenylalanine–glycine ) dipeptide motifs . Vertebrate NPCs accommodate ≈11 different FG Nups , which together should contribute ≈13 MDa of FG domain mass and >5000 FG motifs per pore ( Ori et al . , 2013 ) . Nup98 ( Radu et al . , 1995; Powers et al . , 1997 ) is an FG Nup of special interest for this study . In its canonical form , it is initially produced as a fusion protein with the NPC scaffold component Nup96 . An auto-proteolytic domain in Nup98 subsequently cleaves the fusion ( Rosenblum and Blobel , 1999 ) and thereafter anchors Nup98 to the NPC scaffold ( Griffis et al . , 2003 ) . Mammalian Nup98 comprises an FG domain of about 500 residues , including a binding site for the mRNA export mediator Gle2/ Rae1 ( GLEBS domain; Bailer et al . , 1998; Pritchard et al . , 1999 ) . Nup98 has been estimated to occur in ≈48 copies per NPC ( Ori et al . , 2013 ) . The yeast Saccharomyces cerevisiae possesses three Nup98 paralogs , Nup145 , Nup100 , and Nup116 ( Wente et al . , 1992; Wente and Blobel , 1994 ) , whereby only Nup145 is produced as a fusion protein and only Nup116 contains a GLEBS domain . The FG domains of Nup100 ( 570 residues ) and Nup116 ( 740 residues ) are considerably longer than that of Nup145N ( 200 residues ) . FG domains are intrinsically disordered ( Dosztanyi et al . , 2005; Denning and Rexach , 2007 ) and have therefore been appearing ‘featureless’ on high-resolution cryo-electron tomography averages ( Beck et al . , 2004 ) . Nevertheless , they engage in critical interactions . First , FG motifs are binding sites for NTRs ( Iovine et al . , 1995; Rexach and Blobel , 1995 ) and this interaction is a prerequisite for any facilitated translocation ( Bayliss et al . , 1999 ) . Second , certain FG domains also bind constituents of the NPC scaffold ( Patel et al . , 2007; Schrader et al . , 2008; Andersen et al . , 2013; Xu and Powers , 2013 ) . ‘Cohesive’ interactions between FG repeats themselves represent a third category . They are multivalent and primarily driven by the hydrophobicity of the FG motifs ( Ribbeck and Görlich , 2001; Frey et al . , 2006; Patel et al . , 2007; Frey and Görlich , 2009; Ader et al . , 2010; Hülsmann et al . , 2012; Labokha et al . , 2013; Xu and Powers , 2013 ) . Several lines of evidence suggest that FG domains actually constitute the NPC permeability barrier . First , certain genetic deletions of FG domains relax the barrier of S . cerevisiae nuclear pores ( Patel et al . , 2007 ) . Second , the collapse of the permeability barrier during poliovirus infection coincides with a cleavage of FG domains ( Park et al . , 2008 ) . Third , replacing Nup98 in Xenopus nuclei by a variant lacking the FG domain renders NPCs non-selectively permeable ( Hülsmann et al . , 2012 ) . Additional striking evidence came from the observation that jellified FG domains show permeability properties very similar to those of authentic NPCs ( See e . g . , Frey and Görlich , 2007; Labokha et al . , 2013 , and below ) . The perhaps greatest challenge in the field is to explain why interactions between an NTR and the stationary FG domains do no delay but instead greatly accelerate NPC passage in comparison to inert ( i . e . , non-FG-binding ) reference molecules . Several solutions to this paradox have been proposed . The ‘virtual gate’ model ( Rout et al . , 2003 ) regards FG domains as entropic brushes that repel inert material while NTRs overcome this entropic barrier by binding to them . The selective phase model assumes that cohesive interactions between FG repeats cross-link them into a sieve-like FG hydrogel , whose mesh size sets an upper size limit for passive NPC passage ( Ribbeck and Görlich , 2001; Frey and Görlich , 2007 ) . It further assumes that repeat–repeat contacts disengage locally and transiently when a transport receptor binds the corresponding FG motifs . This should allow the receptor to ‘melt’ with its cargo through the gel . Several lines of evidence support this model: first , replacing the highly cohesive FG domain of Xenopus Nup98 by non-cohesive ( or by partially cohesive ) ones resulted in non-selectively permeable NPCs ( Hülsmann et al . , 2012 ) . This held true even if the replacing non-cohesive FG domain was fully proficient in NTR-binding . Second , numerous FG domains can indeed form hydrogels that behave like the NPC's permeability barrier , i . e . they exclude inert macromolecules but allow rapid influx of NTRs and NTR⋅cargo complexes ( Frey and Görlich , 2007 , 2009; Milles and Lemke , 2011 ) . A systematic analysis of Xenopus FG domains revealed that the Nup98 FG domain , which appears to be the most critical one for NPC function , also yielded the most selective FG hydrogel ( Hülsmann et al . , 2012; Labokha et al . , 2013 ) . A critical gap in the argument relates , however , to the facts that ( i ) gels with a true NPC-like permselectivity were obtained only if the FG domain concentration exceeded a certain ‘saturation limit’ , typically around 200 mg/ml ( Frey and Görlich , 2007 ) and ( ii ) that it was so far unclear if and how the relevant FG domains could possibly become sufficiently concentrated to form highly selective gels . We now observed that Nup98 FG domains from animals , fungi , plants , amoebas , ciliates , and excavates ( chosen to represent all major eukaryotic clades ) indeed phase-separate spontaneously from aqueous solutions into FG hydrogels . Phase-separation was a matter of no more than seconds and occurred at critical concentrations ranging from 20 to 700 nM or 1 to 50 µg/ml FG domains , which corresponds to 1–35 µM FG motifs . These critical concentrations are exceeded by at least 2–3 orders of magnitude when the NPC scaffold initially recruits Nup98 molecules during NPC assembly . Notably , NTRs did not prevent the phase-separation of FG domains , even when present in larger excess than NPCs could possibly accommodate . The self-assembly process yielded micrometer-sized near-spherical FG particles with a preferred protein concentration of ≈250 mg/ml . Strikingly , these particles behaved like the predicted selective phase: they excluded inert macromolecules ( such as the ≈25 kDa-sized mCherry ) but at the same time allowed efficient NTR entry at rates that appeared mostly limited by mass transport within the surrounding buffer . NTRs diffused through these particles with a rate constant in the order of 0 . 1 µm2/s , i . e . fast enough to traverse a 40-nm thick NPC barrier made of the same material within ≈10 milliseconds . With remarkable clarity , the FG particles also recapitulated the effect that a larger cargo must recruit several NTR molecules in order to efficiently pass an NPC . The simplest explanation for these parallels is that the actual NPC barrier is formed by such phase-separated FG hydrogel material . Our data further suggest that a mere binding of a translocating species to FG domains is insufficient for facilitated passage . Instead , it is consistent with the notion that facilitated translocation involves an NTR-mediated melting of obstructing meshes and that this ‘melting’ is restricted to the immediate vicinity of the NTRs .
Clearly , the recruitment of FG Nups to the NPC scaffold already represents a first concentration step . A lower limit for the resulting FG concentration can thus be estimated based on three assumptions: ( i ) only Nup98-type FG domains participate in the barrier , ( ii ) these domains occur in 48 copies per NPC ( Ori et al . , 2013 ) , and ( iii ) not only dwell inside the central channel ( length: 80 nm; width: 40 nm ) but reach 70 nm into the nuclear and cytoplasmic space . The latter value represents the farthest distance from the NPC plane at which epitopes of an FG Nup ( Nup358 ) had previously been detected by immuno-EM ( Walther et al . , 2002 ) . These numbers give an accessible volume of 1 . 5 × 10−18 litres and concentrations of 50 µM ( 2 . 5 mg/ml ) for the FG domain and 2 mM for the FG motifs . When considering non-Nup98 FG domains as well , the lower bound increases to 6 mM FG motifs . However , this is still far lower than the above-mentioned threshold for hydrogel selectivity . An effective hydrogel-based barrier should therefore assemble within NPCs only if some mechanism increases the local FG domain concentration further by a factor of 10–100 . In the simplest case , local concentrations can increase because the barrier-forming FG domains have an intrinsic propensity to self-concentrate and hence to phase-separate from aqueous solutions . However , even then the question still remains if the resulting protein-rich phase has an appropriate FG repeat concentration and structure to function as a barrier with NPC-like permselectivity . Given that these issues are central for our understanding of NPC function , we decided to address them systematically . As an initial example , we chose the FG domain of S . cerevisiae Nup100 , because it is a well-studied Nup98 homolog that can restore a functional barrier in FG Nup-depleted Xenopus NPCs ( Hülsmann et al . , 2012 ) , but it has not yet been tested if this domain is also sufficient for assembling a selective hydrogel in vitro . In the first series of experiments , we prepared a 1000 µM Nup100 FG domain stock solution , supplemented by 2 M guanidinium hydrochloride in order to initially keep the domain molecules in a non-interacting state . The solution was then rapidly diluted in 100 volumes of a neutral Tris/NaCl buffer , which lowered the protein concentration to 10 µM and the guanidinium ion concentration to negligible levels ( Figure 1A ) . Remarkably , the Nup100 FG domain solution turned instantaneously turbid , pointing to a very rapid phase-separation and formation of small particles or liquid droplets , which can be easily recovered by centrifugation ( Figure 1B ) . We tried to record a time course of this reaction ( by static light scattering ) but had to realise that the reaction had already reached its endpoint before we could place samples into the instrument and start the measurement ( after 10–30 s ) . 10 . 7554/eLife . 04251 . 003Figure 1 . Dilute Nup100 FG domain solutions spontaneously undergo phase-separation . ( A ) Illustration of the experimental design . ( B ) Cohesive FG domains self-assemble into FG phases that can be collected by centrifugation . Two stock protein solutions were prepared in 2 M guanidinium hydrochloride ( GuHCl ) , 100 mM Tris/HCl pH 8 . 0 . They contained 300 µM of a Z-domain tandem fusion ( labelled 1:1 with Atto565 maleimide ) and 300 µM FG domain from either or Nsp1 ( residues 274–601 ) or Nup100 . 5% of the FG domain molecules carried an Atto488 maleimide label . 16 . 7 µl of each solution was diluted with 500 µl 50 mM Tris/HCl pH 7 . 5 , 150 mM NaCl ( TBS ) . Photographs show test tubes after ultracentrifugation ( 100 , 000×g , 30 min ) , illuminated at 366 nm . Note that the Nup100 FG domain pelleted , while the non-cohesive Nsp1 FG repeats and the globular ZZ-domain remained soluble . ( C ) FG particle formation at different concentrations . Label-free ScNsp1276-601 and ScNup100 FG domains were diluted from 400 µM stocks ( in 2 M GuHCl ) to the indicated concentrations with TBS . Formed particles were analysed by Dynamic light scattering ( DLS ) using a DynaPro NanoStar instrument ( Wyatt Technologies ) . Two data sets , comprising each 100 acquisitions à 5 s , were averaged for each dilution . The Dynamics 7 . 1 . 5 software was used for autocorrelation analysis and computation of size distributions . ( D ) Confocal laser-scanning microscopy ( CLSM ) images showing an overview and zoom-in of ScNup100 FG particles . ( E ) FG particles exclude inert molecules . Particles were formed with 10 µM ScNup100 FG domain and 0 . 5 µM Atto390-tracer . Particles were mixed with Alexa488-labeled maltose binding protein ( MBP ) , which remained excluded from the particle and thus qualified as an internal standard for Alexa488 fluorescence . ( F ) Estimation of FG domain concentration within FG particles . Particles were formed with 10 µM unlabelled , 0 . 5 µM Atto390- and 14 nM Alexa488-labelled ScNup100 FG-domain . CLSM images were taken after adding different dilutions of MBP-Alexa488 , which served as an internal fluorescence standard . Correlating extra-particle Alexa488 signals ( originating from MBP ) with the known supplied MBP concentrations and matching them with the intra-particle Alexa488 signals ( originating from 1/715th of the FG-domain molecules ) suggests that an average particle contains ≈4 . 5 mM ( ≈275 mg/ml ) FG-domain . This corresponds to ≈200 mM FG motifs . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 003 When applying dynamic light scattering ( DLS ) to the 10 µM Nup100 FG domain sample , we observed a very prominent particle population with diameters ranging mostly between 2 and 8 µm ( Figure 1C ) . At a lower concentration ( 0 . 625 µM ) , we observed a broader main peak with 0 . 4–4 µm particles ( accounting for ≈60% by mass ) , another peak with 25 nm assembly intermediates ( 10% ) and residual monomers ( 30% ) . The remaining monomer concentration can be taken as an estimate for the critical concentration for this phase-separation , which hence should be in the range of ≈200 nM FG domain or ≈10 µM FG repeat units . Yet , already the anchorage of FG domains to the NPC scaffold should result in a ≥200 times higher local concentration ( 2 mM repeat units; see above ) . This ≥200-fold oversaturation should thus drive FG phase-separation to completion also in real NPCs . The rapid onset of the in vitro FG particle formation further supports this view , as it indicates that self-association of Nup100 FG domains is not impeded by a kinetic hurdle . As a control , we analysed the regular and highly charged part of the S . cerevisiae Nsp1p FG domain ( residues 274–601 ) and observed that it failed to form particles at 10 µM domain concentration ( Figure 1B , C ) . This confirms the earlier observations that this FG subdomain is of low cohesiveness ( Ader et al . , 2010; Yamada et al . , 2010 ) and fails to assemble a functional barrier in Xenopus NPCs that lack the Nup98 FG domain ( Hülsmann et al . , 2012 ) . Confocal laser-scanning microscopy ( CLSM ) of Nup100 FG particles , formed in the presence of 5% ( vol/vol ) Atto390-labeled FG domain tracers , indicated that the particle size increased with higher initial concentration ( not shown ) . Most particles were of nearly spherical shape , whereby the evident deviations from perfect spheres indicated that the phase-separated objects represent solids rather than liquids ( Figure 1D; further evidence for this assumption is provided below ) . As a next step , we wanted to determine the Nup100 FG domain concentration in the formed particles . For that we recorded the local fluorescent intensity of the added Alexa488 FG domain tracer , determined absolute fluorophore concentrations by calibration with a series of internal Alexa488-MBP standards ( Figure 1E , F ) , and used this number to derive an average intra-particle concentration of the Nup100 FG domain of 275 mg/ml ( corresponding to 4 . 7 mM FG domain or 200 mM FG motifs ) . Further assuming a specific partial volume for the protein part of 0 . 73 ml/g , one can estimate that the polypeptide accounts for 20% and water ( respectively buffer ) for the remaining 80% of the particle volume . These numbers are well in line with the assumption that these particles indeed represent hydrogels . We subsequently used CLSM to assess how fluorescent permeation probes would partition into the self-assembled Nup100 FG particles . As shown in Figure 2A , the particles allowed for a strong accumulation of probes with NTR-like properties . This applied to NTF2 , Importin β complexed with an Atto488-labelled IBB-domain or an IBB-GFP fusion , or even a 520 kDa octameric complex comprising four copies of Importin β and the IBB-ZsGreen homotetramer . The particles were imaged after just ≈3 min of NTR influx . Nevertheless , particle:buffer partition coefficients of ≈200 were observed , and the NTRs showed already a rather even distribution between the centres and peripheries of the 5–10 µm-sized particles . A comparison with computer simulations ( Figure 3 ) suggests that the influx was so rapid that diffusion within the surrounding buffer , but not the actual particle-entry , must have become rate-limiting . The intra-particle NTR distributions are consistent with intra-particle diffusion coefficients of ≥0 . 1 µm2/s , implying that it would take these transported species no more than 10 milliseconds to traverse a 40-nm thick NPC barrier made of the same material . This agrees well with previously observed NPC dwell times of NTF2 or importin β during a successful pore passage ( Yang et al . , 2004; Kubitscheck et al . , 2005; Yang and Musser , 2006 ) . 10 . 7554/eLife . 04251 . 004Figure 2 . Permeability properties of ScNup100 FG particles . ( A ) FG particles were formed at 10 µM ScNup100 FG domain concentration ( including 5% Atto390-tracer ) , as described above . 3 µM of the passive permeation probe MBP-mCherry or 1 µM of the indicated active permeation species were added ( concentration referring to substrate monomers ) . CLSM images were taken ≈2–3 min later , using the 405 nm , 488 nm , or 561 nm laser lines for exciting the FG tracer , active or passive permeation probes , respectively . ( B ) Intra-FG particle dynamics of FG domains and NTR⋅cargo complexes . ScNup100 FG particles were formed as in Figure 1E and challenged with 1 µM of a yeast Impβ•IBB-MBP-GFP complex . CLSM images show two particles . One of them was photobleached at 405 and 488 nm in one hemisphere . Fluorescence recovery of the FG domain tracer as well as of the NTR⋅cargo complex was detected over time . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 00410 . 7554/eLife . 04251 . 005Figure 3 . Estimate for kinetic parameters for Importinβ•IBB-GFP influx into ScNup100 FG particles . Influx of the NTR⋅cargo complex into the FG particles had to be performed in suspension , which implied that we had to wait until particles had settled to the bottom of the slide ( ≈3 min ) before concentration profiles across a particle and surrounding buffer could be recorded . Then , however , the endpoint of accumulation was essentially reached already . In order to nevertheless estimate kinetic parameters for particle-entry , we simulated the process and asked which parameter set would be consistent with the observed cargo distribution at the 3 min time point . These parameters included the partition coefficient ( 220 ) and 7 µm particle diameter ( both measured directly ) , the diffusion coefficient in buffer ( Dbuffer = 50 µm2/s , derived by the Stokes–Einstein equation from the radius of the Impβ•IBB-GFP complex and the viscosity of the buffer ) , as well as the intra-particle diffusion coefficient ( DParticle = 0 . 1 µm2/s ) , which was the smallest that allowed an even intra-particle distribution of the cargo at the 3 min timepoint . ‘Capture efficiency’ describes the probability that a colliding NTR⋅cargo complex gets captured by the particle . Simulations were performed in Mathematica 9 . 0 and exploited the spherical symmetry of the particle to simplify the system of differential equations ( see Supplementary file 1 for the Mathematica code and more detailed explanations ) . ( A ) ScNup100 FG particles were formed at 10 µM and 30 min later challenged with 1 µM Impβ•IBB-GFP complex . CLSM image was taken after another 3 min . ( B ) Impβ•IBB-GFP concentration profile across the area indicated in panel A . Signal was normalized to the concentration in buffer . ( C ) Simulation of influx for indicated parameters and time points . ( D ) Sensitivity analysis , varying the parameters used in C . It revealed that diffusion in buffer , the partition coefficient , and diffusion inside the particle , but not the capture efficiency , are limiting for the influx process . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 005 At the same time , we observed that the inert MBP-mCherry fusion ( 65 kDa ) remained well excluded from the particles' interior ( see Figure 2A ) . The respective partition coefficients of <0 . 1 were far lower than the estimated fraction of the particles' volume that is occupied by the solvent , which in turn is consistent with the assumption of a sieve structure causing a size-exclusion effect . The phase-separation of cohesive FG domains from dilute aqueous solutions draws an interesting parallel to RNA granules , which are membrane-free compartments formed by multivalent interactions between RNAs and RNA-binding proteins . RNA granules are also of near-spherical shape and this shape has been attributed to a liquid state and surface tension effects ( Brangwynne et al . , 2009 ) . Given these similarities , we wanted to probe the physical conditions of the obtained FG phases and used FRAP for this purpose ( Figure 2B ) . After bleaching the Atto390 fluorophore of labelled FG domains , we observed that bleached patterns within the Nup100 FG particles remained stable for some minutes at least . This indicates a solid state of the particles and stable interactions between FG domains . The simultaneously bleached NTR-signal was , however , far more dynamic . The particles' peripheries showed a clear signal recovery already within two seconds , indicating a very rapid exchange with the surrounding medium . This implies that this NTR-species can rapidly exit the particle—despite its very high partition coefficient . The fluorescent signal in the centre of a particle with a radius of ≈4 µm recovered with a half time of ≈40 s . This clearly shows that NTRs are actually very mobile within a rather static FG phase . We had previously prepared FG hydrogels by quickly dissolving the TFA ( trifluoroacetic acid ) salt of a lyophilised FG domain to ≈200 mg/ml , and one could argue that gels formed only because the domains were forced artificially to such a high local concentration . We now demonstrated that initially very dilute aqueous solutions of Nup98 FG domains phase-separate into an FG-rich phase ( see Figure 1 and below ) . However , in order to suppress intermolecular contacts prior to this dilution step , we had to keep the concentrated protein solution in ≥2 M guanidinium hydrochloride . One could therefore argue that such transient denaturation is still non-physiological and might have caused an artificial FG hydrogel formation that otherwise would not have occurred . In order to address this issue , we avoided any denaturing treatment during the next steps . Specifically , we tested whether Nup98-derived FG domains also self-assemble into selective phases , when they are simply expressed in the bacterium Escherichia coli and thus are not exposed to any potentially structure-changing manipulation . Following recombinant expression ( after modest induction ) , we resuspended the bacteria in a physiological buffer , gently disrupted the cells by lysozyme treatment , and subjected the lysate to a 10 , 000×g centrifugation step . We observed that the FG domains from Nup100 and Nup116 ( the second S . c . Nup98 paralog ) pelleted under these conditions ( Figure 4A , B ) . Given that the centrifugation was performed with only a k-factor of ≈1000 S , it thus appeared that these FG domains had formed rather large structures . 10 . 7554/eLife . 04251 . 008Figure 4 . In vivo assembly of selective FG phases . For the undistorted in vivo formation of FG bodies , the ScNup100 and ScNup116 FG domains were expressed in Escherichia coli ( NEB Express ) following induction with 0 . 5 mM IPTG at 30°C . Cells were pelleted , resuspended in TBS , and lysed with 1 mg/ml lysozyme , 20 µg/ml DNAse I and 0 . 5% Tween 20 . Insoluble ‘FG bodies’ were recovered by centrifugation at 10 , 000×g . SDS-PAGE analysis of supernatant ( ‘Sup’ ) and pellet fractions and the permeability properties of the recovered ScNup100 ( A ) and ScNup116 ( B ) FG bodies are shown . Crude FG bodies were washed twice in TBS and their permselectivity analysed as described in Figure 2 . ( C ) Mobility of NTR⋅cargo complexes in FG bodies . ScNup100 FG bodies were challenged with 1 µM Impβ•IBB-GFP and photobleached , and fluorescence recovery was detected over time . Note that the NTR⋅cargo complex was mobile also within the in vivo formed FG bodies . The white dashed line in the 50 s frame indicates the region analysed for the line plots shown below the images . ( D ) Fluorescence recovery over time in the area indicated by the blue box in the zoom-in of the particle outlined by the dashed lines in C . The analysed region lies approximately 1 µm inside of the particle . Fluorescence recovery occurred with a time constant of ≈100 s . Note that this involved not only intra-particle diffusion , but also the uptake from the buffer against a ≈200-fold concentration gradient . ( E ) In contrast to the FG bodies , actin inclusion bodies did not enrich the Impβ•IBB-GFP species , but in fact excluded it like MBP-mCherry . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 008 At first glance , the insoluble material resembled ordinary inclusion bodies that result when ‘difficult-to-fold-proteins’ ( for example recombinant actin; Figure 4E ) form irreversible aggregates during overexpression in a prokaryotic host . One critical difference is , however , that the initially insoluble ‘FG bodies’ slowly dissolved when soluble FG domain molecules were removed from the equilibrium , for example during repeated steps of pelleting and resuspending in fresh aliquots of buffer ( not shown ) . Microscopic analysis revealed that the Nup100 and Nup116 ‘FG bodies’ were 10 µm-sized irregularly shaped particles that probably had been ‘glued’ together from smaller species during the centrifugation steps . They nearly perfectly excluded our passive permeation marker MBP-Cherry and yet allowed a very efficient intra-particle accumulation of the NTR⋅cargo complex Impβ⋅IBB-GFP , reaching again a particle:buffer partition coefficient of around 200 ( Figure 4A , B ) . In Figure 4C , we bleached the Impβ⋅IBB-GFP signal in an area that comprised an entire Nup100 FG body . Fluorescence recovery occurred with a time constant of ≈100 s in the interior of the particles ( Figure 4D ) . This is remarkably fast , considering that ( i ) recovery could occur only through the influx from the buffer , ( ii ) that the concentration in the buffer was just 0 . 5% of that inside the particles , and ( iii ) that the dimension of the particle was almost 200-fold larger than the presumed NPC barrier , i . e . when scaled down to NPC dimensions , recovery would have occurred with a time constant of a few milliseconds . Earlier studies already reported that overexpressing , for example , a YFP-Nup100 FG repeat fusion in S . cerevisiae ( Patel et al . , 2007 ) or GFP-fused to the human Nup98 FG domain in HeLa cells ( Xu and Powers , 2013 ) , can result in characteristic intra-cellular foci . Yet , the composition , dependence on host factors and barrier properties of such assemblies had not been evaluated . Figure 4 now demonstrates that in vivo formed minimalistic Nup98 FG phases feature a striking NPC-like permselectivity . It also shows that phase-separation requires neither accessory eukaryotic factors nor sophisticated experimental manipulations . Instead , the assembly of FG bodies relies exclusively on the strong intrinsic propensity of these FG domains to interact . If self-assembly of barrier-critical FG domains into selective FG phases is fundamental for NPC function , then one can expect this phenomenon to be conserved throughout the eukaryotic tree of life . We decided to test this assumption by comparing ten different Nup98 FG domains from nine divergent species , namely: Nup98 from human ( Radu et al . , 1995 ) , Branchiostoma floridae ( representing lancelets ) , Drosophila melanogaster ( representing insects; Presgraves et al . , 2003 ) , NPP-10/Nup98 from Caenorhabditis elegans ( representing nematodes; Voronina and Seydoux , 2010 ) , the two already mentioned paralogs Nup100 and 116 from Saccharomyces cerevisiae ( representing fungi; Wente et al . , 1992 ) , Dictyostelium discoideum Nup220 ( representing amoebas ) , Arabidopsis thaliana Nup98B ( representing plants; Tamura et al . , 2010 ) , the macronuclear MacNup98A from Tetrahymena thermophila ( representing ciliates; Iwamoto et al . , 2009 ) as well as Nup158 from Trypanosoma brucei ( representing euglenozoans/excavates; DeGrasse et al . , 2009 ) . Criteria for this selection had been a wide sampling of species and sequence diversity , a preference for well-studied model organisms , and exclusion of FG domains shorter than 400 residues . The resulting selection covered all major eukaryotic clades ( See Figure 5A ) , with the exception of Rhizaria , where genome analysis has lagged far behind and only Nup98s with apparently short FG domains were listed in databases at the time of our analysis . 10 . 7554/eLife . 04251 . 009Figure 5 . Nup98 FG domains of diverse evolutionary origin bind human Importin β . ( A ) This study analyses ten Nup98 FG domains from nine species . Cartoon illustrates their positions within the eukaryotic tree of life ( adapted from Keeling et al . , 2005 ) . See Supplementary file 2 for complete sequences . LCEA denotes the position of the last common ancestor to all eukaryotes . ( B ) Indicated His-tagged FG domains ( 30 µg each ) were immobilized on 30 µl PEG-passivated Ni ( ii ) chelate beads and rotated for 2 hr at 4°C with 1 µM untagged human Importin β ( 400 µl ) . Bound prey and immobilized baits were co-eluted with SDS/Imidazol and analysed by SDS-PAGE/Coomassie-staining . Note the incomplete elution of the Saccharomyces Nup116 and Dictyostelium Nup220 FG domains . Phenyl-Sepharose served as a positive control ( Ribbeck and Görlich , 2002 ) . Binding was in 25 mM Tris/HCl pH 7 . 5 , 100 mM NaCl , 1 mM MgCl2 , 0 . 5% PEG4000 , 5 mM DTT . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 009 As a first step , we constructed bacterial expression vectors , and recombinantly expressed , purified , and immobilized all ten Nup98 domains . We observed that all of them bound human Importin β specifically and to a similar extend ( Figure 5B ) . This suggests that the mode of FG domain⋅NTR-interaction has not changed dramatically during eukaryotic evolution . We next extended our analysis of spontaneous FG phase formation to the entire set of Nup98 FG domains . We observed that all of them showed the same readiness to phase-separate and form FG particles as the ScNup100 FG domain ( Figures 6 and 7 ) . The size distribution of particles differed between the various FG domains , which perhaps reflects different particle seeding and growth rates . But otherwise , the particles were mostly spherical and showed a rather even intra-particle FG domain distribution . In this species comparison , they were also remarkably similar in their intra-particle FG domain concentration ( average 250 mg/ml , range 175–350 mg/ml; see Table 1 ) . Furthermore , not in a single case did the critical FG domain concentration for the phase-separation exceed 1 µM or 50 µg/ml , even when FG domains were O-glycosylated ( Table 1; see also Labokha et al . , 2013 ) . Remember that the local Nup98 FG domain concentration at NPCs should be at least 50 times higher , even if one assumes that the domains initially do not interact . Thus , a phase-separation of Nup98 FG domains into selective FG hydrogels should be thermodynamically highly favoured for the stoichiometry settings of authentic NPCs . Moreover , the striking conservation of the underlying biophysical properties throughout the eukaryotic kingdom can be taken as a very strong argument for a fundamental functional relevance . 10 . 7554/eLife . 04251 . 006Figure 6 . Different modes of inter-FG repeat interactions within FG particles . The indicated Nup98 FG domains were studied . Representative repeat sequences are shown in single letter code ( see Supplementary file 2 for complete sequences ) . Particles were formed at 5 µM ( HsNup98 , TtMacNup98 , TbNup158 ) or 10 µM FG domain concentration ( all other FG domains ) . The suspensions were afterwards supplemented with 3 µM MBP-mCherry and 1 µM ThioflavinT , a diagnostic dye for the presence of amyloid-like cross-β-structures . Particles were detected based on exclusion of MBP-mCherry . ThioflavinT was excited at 405 nm and detected in a 460–500 nm window . Graphs show quantitations for the measured signals ( gray value scales ) . ScNup100 FG particles gave the strongest signal . For Table 1 , all Thioflavin signals were normalized to the Nup100 signal . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 00610 . 7554/eLife . 04251 . 007Figure 7 . Nup98 FG domains from distant eukaryotic clades self-assemble into highly selective FG particles . Particles were formed as described in Figure 6 , followed by the addition of either 1 µM NTF2-Atto488 or Impβ•IBB-GFP and 3 µM mCherry as active and passive permeation probes , respectively . The experimental setup was as in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 00710 . 7554/eLife . 04251 . 010Table 1 . Key descriptors of FG particle constitutionDOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 010Molecular weight*Number of residues*Number of FG motifs*Estimated critical concentration†Estimated intra-particle concentration‡NQ-content*Relative Thioflavin-T signal§FG domainFG motifsHomo sapiens HsNup98≈49 kDa50039≈25 nM≈1 µg/ml≈1 µM≈175 mg/ml11%11%Branchiostoma floridae BfNup98≈46 kDa47940≈20 nM≈1 µg/ml≈1 µM≈200 mg/ml8%8%Branchiostoma floridae BfNup98 + GlcNAc≈55 kDa47940≈150 nM≈10 µg/ml≈6 µMND8%ND#Drosophila melongaster DmNup98≈56 kDa58146≈200 nM≈10 µg/ml≈9 µM≈300 mg/ml10%3%Caenorhabditis elegans CeNup98≈48 kDa49436≈700 nM≈40 µg/ml≈25 µM≈300 mg/ml18%52%Saccharomyces cerevisiae ScNup100≈58 kDa57843≈175 nM≈10 µg/ml≈7 . 5 µM≈275 mg/ml28%100%Saccharomyces cerevisiae ScNup116≈65 kDa73747≈700 nM≈50 µg/ml≈33 µM≈350 mg/ml26%78%Dictyostelium discoideum DdNup220≈68 kDa71956≈125 nM≈10 µg/ml≈7 µM≈300 mg/ml12%4%Arabidopsis thaliana AtNup98B≈66 kDa66852≈25 nM≈1 µg/ml≈1 . 5 µM≈200 mg/ml13%1%Tetrahymena thermophila TtMacNup98A≈61 kDa66642≈25 nM≈1 µg/ml≈1 µM≈175 mg/ml18%2%Trypanosoma brucei TbNup158≈50 kDa56558≈300 nM≈15 µg/ml≈17 . 5 µM≈250 mg/ml12%2%*All values are given for full-length FG domains ( including GLEBS domains ) . The molecular weight of glycosylated BfNup98 was estimated by SDS-PAGE . †The critical concentrations for phase separation were estimated as described in Figure 1C and the methods section . ‡Intra-particle FG domain concentrations were estimated as described in Figure 1F and the methods section . §Thioflavin-T signals are normalised to the ScNup100 Thioflavin-T signal; also see Figure 6 . #Experiments with macroscopic hydrogels suggest that the Thioflavin-T signal of glycosylated BfNup98 FG particles is even lower than the observed signal for the non-glycosylated BfNup98 FG particles . As expected from the extreme evolutionary distances separating the host species , the sequences of the ten analysed Nup98 FG domains are actually rather diverse . They differ , for example , quite widely in their NQ-content ( Table 1; Supplementary file 1 ) , i . e . , in a parameter that has been linked to the formation of amyloid-like structures , also in NQ-rich hydrogels ( Alberti et al . , 2009; Ader et al . , 2010 ) . To test if these sequence differences also translate into different contents of amyloid-like cross-β sheets , we stained the particles with ThioflavinT—a compound whose 480 nm fluorescence emission is enhanced upon binding to NQ-rich and other amyloid-like cross-β structures ( Vassar and Culling , 1959; Khurana et al . , 2005 ) . Particles derived from the most NQ-rich FG domains , Nup100 ( 28% ) and Nup116 ( 26% ) showed indeed by far the strongest ThioflavinT signal , supporting that cross-β structures can occur not only in amyloid fibres , but also in self-assembled FG particles ( Figure 6 and Table 1 ) . The NQ content alone is , however , not necessarily a reliable predictor for the presence of cross-β sheets . The C . elegans Nup98 and the Tetrahymena MacNup98A FG domains , for example , have a very similar NQ-fraction ( 18% ) , yet we observed a clear ThioflavinT signal only for C . elegans FG particles . The Tetrahymena MacNup98A FG particles were essentially ThioflavinT-negative , perhaps because their high glycine content counteracts β-sheet formation and/or stability . FG particles from the other species also showed at most a weak ThioflavinT signature . Taken together , this suggests that particles from the selected FG domains sample different modes of inter-FG repeat interactions , and we were curious to find out if this would translate into different permselectivities . Strikingly , we observed that Nup98 FG particles from all ten Nup98 FG domains efficiently excluded not only the still rather large MBP-mCherry fusion ( ≈75 kDa ) ( Figure 6 ) but also rejected the far smaller mCherry ( ≈25 kDa ) ( Figure 7 ) . At the same time , these FG phases allowed a high or very high accumulation of NTF2 ( Figure 7 ) . In each case , the partition coefficient of NTF2 was at least 1000 times higher than that of mCherry . A FRAP analysis revealed that the solid state of the Nup100 FG particles and low mobility of the phase-separated Nup100 FG domain were no exception but also applied to other FG particles , for example , formed by the evolutionary very distant Tetrahymena MacNup98A FG domain ( Figure 8 ) . Our analysis also revealed an extremely rapid exchange of NTF2 between the TtMacNup98A particles and the surrounding buffer , as well as a very high mobility of NTF2 inside these particles . 10 . 7554/eLife . 04251 . 011Figure 8 . Intra-FG particle dynamics of FG domains and NTRs . TtMacNup98A FG particles were formed with 5 µM unlabeled and 25 nM Atto390-labeled TtMacNup98A FG domain and challenged with 1 µM NTF2-Atto488 . 5 min after NTR addition , particles were photobleached at 405 and 488 nm in one hemisphere , and fluorescence recovery of the tracer and NTR followed over time . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 011 We observed that aqueous Nup98 FG domain solutions spontaneously separate into FG phases with just the ‘right’ FG domain concentration and structure for creating a barrier with NPC-typical permselectivity . This held true for Nup98 FG domains from nine evolutionary very distant species—a clear indication for a strong evolutionary pressure to maintain the underlying physicochemical properties . As an additional indication , we noticed extreme sequence conservation amongst vertebrates , covering ≈400 million years of evolution . The Nup98 FG domains from human and the fish Lepisosteus oculatus , for example , share ≈70% identical residues , which is close to the ≈75% identities between the corresponding globular autoproteolytic Nucleoporin 2 domains ( Figure 9 ) . Moreover , most of the observed exchanges were just conservative permutations between T , S , A , and N within the spacers , while spacer lengths and the type of FG motif at a given position remained extremely conserved . The Nup98 FG domain is thus an exception from the rule that intrinsically disordered domains change rapidly during evolution ( Denning and Rexach , 2007 ) . In fact , this indicates that the Nup98 FG domain engages in critical interactions along its entire sequence and that deviations from this optimal sequence are not well tolerated . 10 . 7554/eLife . 04251 . 012Figure 9 . High sequence conservation of the Nup98 FG domain amongst vertebrates . Nup98 from human ( GI: 530395413 ) , the frog Xenopus tropicalis ( GI: 523580018 ) and the fish Lepisosteus oculatus ( GI: 573879996 ) were aligned . FG motifs are in bold; deviations from the human sequence are marked in yellow . The domain structure is annotated . Unlike typical intrinsically disordered domains , the Nup98 FG domain shows a similar conservation ( ≈70% identity ) as the globular , folded nucleoporin2 domain ( ≈75% identity ) . The majority of exchanges between the FG domains are very conservative , that is mostly permutations between T , S , N , A , and to a lesser extent , with G . Please also note that many exchanges from the human sequence are identical in frog and fish , further supporting the notion of slow evolution . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 012 With greater evolutionary distance , differences in the preferred FG sequence context and inter-FG spacer compositions became evident . Ciliates , for example , prefer GLFG motifs , fungi SLFG or GLFG , plants PFG , PAFG , or xFG and Trypanosomes GGFGQ motifs ( Table 2 ) . Likewise , fungi prefer very NQ-rich inter-FG spacers , lancelets very T-rich , and trypanosomes very GA-rich spacers . 10 . 7554/eLife . 04251 . 013Table 2 . Sequence features of the studied FG domainsDOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 013FG motif density ( occurrence per 100 aa ) . Hs 98Bf 98Dm 98Ce 98Sc 100Sc 116Dd 220At 98BTt Mac98ATb 158All FG dipeptides7 . 88 . 47 . 97 . 37 . 46 . 47 . 87 . 86 . 310 . 3 ( G/A ) FG2 . 41 . 62 . 10 . 60 . 71 . 20 . 81 . 90 . 11 . 6 ( S/T ) FG1 . 21 . 90 . 71 . 01 . 20 . 20 . 31 . 20 . 40 . 2GLFG1 . 62 . 71 . 02 . 21 . 92 . 73 . 20 . 04 . 00 . 0SLFG0 . 40 . 40 . 92 . 22 . 10 . 31 . 10 . 10 . 10 . 0PFG0 . 80 . 60 . 50 . 20 . 50 . 51 . 51 . 60 . 10 . 7PAFG0 . 00 . 00 . 70 . 00 . 00 . 10 . 01 . 20 . 00 . 0GFGQ0 . 00 . 00 . 00 . 00 . 00 . 00 . 00 . 00 . 07 . 8Other FG1 . 41 . 22 . 01 . 11 . 01 . 40 . 91 . 81 . 60 . 0Bold data represents dominant FG motifsFraction of hydrophobic and charged residues ( w/o GLEBS domains; in % ) . Hs 98Bf 98Dm 98Ce 98Sc 100Sc 116Dd 220At 98BTt Mac98ATb 158FILVM17161614171514161613FILVMP21192218201821262019DE0 . 40 . 00 . 50 . 00 . 20 . 00 . 00 . 30 . 00 . 2RK1 . 61 . 71 . 71 . 62 . 31 . 80 . 10 . 70 . 61 . 8Amino acid composition of the spacer regions ( i . e . FG domains w/o FG motifs and GLEBS domains; in % ) . Hs 98Bf 98Dm 98Ce 98Sc 100Sc 116Dd 220At 98BTt Mac98ATb 158T1625141111918131210S1157131611111913G1013999101362224A7716947671019N63410191326101Q5568913107811P4374347956Bold data represents dominant amino acids At this point , we wondered if Nup98 FG domain sequences have anything in common that could possibly explain their unique biophysical properties . To achieve the best possible sampling , we performed exhaustive database searches and identified 666 sequences that matched several stringent criteria for representing Nup98 FG domains ( Figure 10 ) . Analysis of the sequences indeed revealed several features that appear conserved across all eukaryotic clades , namely: a rather constant number of FG dipeptide motifs per domain ( median 43 ± 6 ) , a similar FG motif density ( one FG motif per 12 . 5 ± 0 . 7 residues ) and domain-length ( median 549 ± 87 residues per FG domain ) , as well as a very strong bias for G , T , S , A , N , Q , and P in the inter-FG spacers ( Figure 10 ) . 10 . 7554/eLife . 04251 . 014Figure 10 . Compilation and analysis of a Nup98 ortholog-derived FG domain database . ( A ) The Nucleoporin2 domains of human Nup98/96 ( GI: 56549643 ) , Arabidopis thaliana Nup98A ( GI: 22329468 ) , and Tetrahymena thermophila MicNup98B ( GI: 289623519 ) were used as BLAST templates to identify further Nup98 homologues in the non-redundant NCBI protein database . The 913 sequences identified in all three searches were then analysed further as subsequently described . ( B ) Taxonomic classification of identified full-length Nup98 candidates . ( C ) Cartoon illustrates domain structure of a canonical Nup98-Nup96 fusion protein that includes an FG repeat domain , an embedded Gle2p-binding site ( GLEBS domain ) , an intervening domain , the Nucleoporin2 domain , as well as the Nup96 part . The number of Nup98 candidates that comprise a given module is written underneath . ( D ) FG domains included residues from translation start till the last FG dipeptide but excluded the GLEBS domain ( as defined by alignment with the ScNup116 GLEBS domain ) . For subsequent analyses , only the 666 FG domains with 400–1000 residues were considered . The histogram illustrates the FG dipeptide density distribution with a median of 9 FG dipeptides per 100 residues ( or one FG dipeptide per 11 residues ) . Outliers to higher densities ( >10 FG/100 residues ) mainly represent domains dominated by less hydrophobic FG motifs ( e . g . GFGQ motifs ) than the often dominating LFG motifs . ( E ) Average amino acid composition of Nup98 FG domains and corresponding standard deviations . Note that the inter FG spacers are dominated by G , T , S , A , N , Q , and P , while F and L dominate the hydrophobic residues . F shows the smallest coefficient of variation . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 014 A comparison of typical intrinsically disordered ( IDP ) regions ( as represented in the DisProt database; Sickmeier et al . , 2007 ) and globular proteins ( as represented by the PDB database ) revealed that Nup98 FG domains are very distinct from these two domain categories ( Figure 11 ) . They have far fewer charged residues ( median 2 . 5 ± 0 . 4% ) than either average IDP regions ( 28 ± 7% ) or globular proteins ( 24 ± 3% ) . They are clearly more hydrophobic than typical IDP regions and , in fact , comparably hydrophobic as globular proteins . Thus , the Nup98 FG domains should have a similar potential to bury hydrophobic side chains from water as globular proteins . In contrast to the latter however , this does not result in a globular fold , but in strong inter-FG repeat cohesion . Taken together , the strong selection against charges and maintenance of hydrophobicity is well in line with our observation that Nup98 FG domains experience water as a “poor solvent” and consequently phase-separate from even rather dilute solutions . The evolutionary conservation of the underlying sequence features suggests again a fundamental functional relevance . 10 . 7554/eLife . 04251 . 015Figure 11 . Comparison of Nup98 FG domains to intrinsically disordered and globular proteins in terms of charged residue contents and hydrophobicity . Analysis was similar to Uversky et al . ( 2000 ) , the differences being ( i ) that we considered not the net charge , but the total fraction of charged residues and ( ii ) we used a more strictly defined hydrophobicity scale that is not biased by globular protein structures . For each protein sequence , the mean fraction of charged residues was determined by counting D , E , K , and R and dividing this sum by the sequence length . Mean hydrophobicity was calculated according to a scale based on partitioning of Nα-acetyl-amino acid amides between 1-octanol and water at neutral pH ( given in Table 2 in Fauchere and Pliska , 1983 ) . For clarity , we re-scaled their numbers linearly to range between 0 and 1 , and thus used the following parameters: ( R , 0 ) ; ( K , 0 . 006 ) ; ( D , 0 . 012 ) ; ( E , 0 . 113 ) ; ( N , 0 . 132 ) ; ( Q , 0 . 242 ) ; ( S , 0 . 298 ) ; ( G , 0 . 31 ) ; ( H , 0 . 35 ) ; ( T , 0 . 39 ) ; ( A , 0 . 405 ) ; ( P , 0 . 46 ) ; ( Y , 0 . 604 ) ; ( V , 0 . 684 ) ; ( M , 0 . 687 ) ; ( C , 0 . 782 ) ; ( L , 0 . 831 ) ; ( F , 0 . 859 ) ; ( I , 0 . 862 ) ; ( W , 1 ) . The brightness in the heat maps reflects the number of proteins in a given regime of the plot . ( A ) 667 intrinsically disordered protein ( IDP ) regions were extracted from the DisProt database ( Sickmeier et al . , 2007 ) and analysed as described above . Note their wide distribution in the plot , their high content of charges residues and low hydrophobicity . ( B ) Analysis of 34 , 551 non-redundant protein sequence entries from the PDB ( Bernstein et al . , 1978 ) , representing mostly globular , folded proteins . Note that these sequences are on average less charged and considerably more hydrophobic than the bulk of IDPs . ( C ) Analysis of the 666 identified Nup98 FG domains ( excluding the GLEBS domain ) . Despite also being intrinsically disordered , they strongly cluster in a very narrow region with extremely very low charge density and a hydrophobicity very close to globular proteins . The few outliers with slightly less hydrophobicity represent NQ-rich sequences and reflect the facts ( i ) that N and Q are more hydrophilic than other typical inter-FG spacer residues ( A , T , S , G , P ) and ( ii ) probably that NQ-rich stretches contribute to cohesiveness by conferring very hydrophilic ( cross-β ) contacts ( Ader et al . , 2010 ) . ( D ) For direct comparison , plots A ( in red ) , B ( in blue ) , and C ( in green ) were overlaid in a single plot . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 015 Early evidence for the importance of hydrophobic contacts in maintaining the permeability barrier of authentic NPCs came from the use of 7% ( wt/vol ) trans-cyclohexane 1 , 2 diol to interfere with such interactions in semi-permeabilised HeLa cells . This treatment collapsed the NPC barrier and allowed rapid nucleocytoplasmic equilibration of an MBP reporter , which could be reversed by removal of the reagent ( Ribbeck and Görlich , 2002 ) . Later , similar effects of n-hexane 1 , 6 diol on the S . cerevisiae NPC barrier were published , whereas ethanol appeared to have little direct consequence on passive nuclear influx of cytoplasmic reporters ( Shulga and Goldfarb , 2003 ) . The effect of the hexanediols was attributed to a reversible disruption of inter-FG repeat cohesion ( Ribbeck and Görlich , 2002; Patel et al . , 2007 ) . Hence , one would expect that hexanediols should also interfere with the formation of FG phases . Figure 12 shows that trans-1 , 2-cyclohexanediol or 1 , 6 hexanediol indeed effectively suppressed the assembly of S . cerevisiae Nup116 and Tetrahymena Mac98A FG particles , while ethanol had no disruptive effect . Taken together , these findings further affirm the very close relationship between authentic NPCs and the Nup98 FG phases described in this work . 10 . 7554/eLife . 04251 . 016Figure 12 . Influence of hexanediols on FG particle formation . FG particles were assembled in a volume of 200 µl with 10 µM ScNup116 ( A ) or TtMacNup98A ( B ) FG domain in the presence of increasing amounts of trans-1 , 2-cyclohexanediol , 1 , 6-hexanediol or ethanol . After 60 min of incubation , formed particles were collected as pellets in a 10-min 20 , 000×g centrifugation step . They were analysed together with the soluble supernatants by SDS-PAGE . Both hexanediols clearly disrupted the FG particles , though TtMacNup98A FG particles appear slightly more resistant than ScNup116 FG particles ( consistent with the lower saturation concentration of the Mac98A FG domain ) . In contrast , ethanol had no disruptive effect , but rather precipitated the FG domains . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 016 When analysing Tetrahymena MacNup98a FG particles in more detail , we noticed that they accumulated NTF2 very efficiently in their interiors , but arrested the Importin β⋅IBB-GFP complex at their surfaces ( Figure 13A ) . As a similar arrest at real NPCs would be very problematic , we decided to dig deeper into this problem . At first glance , this was reminiscent of macroscopic Xenopus Nup98 FG hydrogels , which were permeable for Importin β-type NTRs only when O-glycosylated with GlcNAc ( Labokha et al . , 2013 ) . However , since there is no indication for a similar modification in ciliates and as the Tetrahymena Nup98A domain contains far fewer serines and threonines as potential modification sites than its vertebrate counterpart , we also had to consider other possibilities . 10 . 7554/eLife . 04251 . 017Figure 13 . Effect of cargo domains and NTR stoichiometry on entry into TtMacNup98A FG particles . FG particles ( with 5% Atto390-labeled tracer ) were formed with 5 µM TtMacNup98A FG domain . CLSM images show how NTR⋅cargo complexes of different sizes and NTR-to-cargo ratios partition between FG phase and bulk solvent . IBB ( recognized by Importin β ) and M9 ( recognized by Transportin , Trn ) represent two orthogonal nuclear import signals . See Figure 13—figure supplement 1 and main text for additional information . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 01710 . 7554/eLife . 04251 . 018Figure 13—Figure supplement 1 . Partitioning of various NTR cargo complexes into TtMacNup98 FG particles . The experiment is identical to the corresponding panels of Figure 13 , the only difference being that a larger field and more particles were imaged at a lower resolution . This was done in order to document that the described differences apply to entire particle populations . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 018 Apart from size , there is actually another difference between NTF2 and the Importin β⋅IBB-GFP complex: NTF2 is a species with a ‘pure’ NTR-surface , whereas Importin β also had to ‘squeeze’ the bound IBB-GFP molecule through the meshes of the gel . It is therefore possible that their spontaneous opening is just too slow or energetically too costly to allow an efficient entry . If this were true , then one would expect that an NTR⋅cargo complex with a smaller cargo should not experience this problem . To test this , we replaced the ≈28 kDa GFP moiety by a single Atto488 molecule . This shrunk the cargo to the ≈6 kDa IBB domain , which gets entirely enwrapped by the Importin β molecule leaving no cargo surface exposed ( Görlich et al . , 1996; Cingolani et al . , 1999 ) . As expected , the Importin β⋅IBB-Atto488 complex did not remain stuck at the particles' surface but accumulated in their interiors ( Figure 13A ) . In parallel controls with the same batch of particles , we observed that the Importin β⋅IBB-MBP-GFP complex ( ≈170 kDa ) with a further enlarged cargo domain got again firmly arrested at the surface . Much to our surprise , however , we observed that a far larger ≈530 kDa complex , namely a complex comprising four Importin β molecules and an IBB-ZsGreen tetramer , crossed the buffer-particle boundary efficiently and accumulated strongly inside ( Figure 13A and 13—figure supplement 1 ) . How can this be explained ? ZsGreen/zFP506 is a tetrameric homolog of GFP ( Matz et al . , 1999; Pletneva et al . , 2007 ) . Consequently , the Importin β⋅IBB-GFP and Importin β⋅IBB⋅ZsGreen complexes have the same NTR:cargo mass ratios , but the tetramerisation buries much of the otherwise exposed cargo surface that counteracts entry into the FG phase . The tetramerisation should thus allow the Importin to dominate the surface properties of the entire complex , and hence to melt together with the cargo through the meshes of the gel . At this point , one could still argue that ZsGreen and GFP are different proteins and that differences other than the fraction of exposed cargo surface account for the difference in Importin-mediated particle-entry . In a next step , we therefore used one and the same cargo ( an IBB-MBP-GFP-M9 fusion ) , and just varied the number of bound NTRs . This cargo includes two orthogonal nuclear import signals , the IBB-domain recruiting Importin β ( Görlich et al . , 1996 ) and the M9-domain conferring nuclear import by Transportin ( Pollard et al . , 1996 ) . Figure 13B shows that neither Importin β nor Transportin alone were sufficient to ferry the cargo across the buffer-particle boundary . Yet , when present together on the same cargo molecule , the two importins synergised and mediated efficient influx and strong accumulation of the cargo in the interior of the particles ( Figure 13B ) . Given that the Tetrahymena MacNup98A FG particles are particularly tight towards inert macromolecules , it is possible that they exaggerate the inhibitory effect of an exposed cargo domain on barrier-passage . This effect is , however , also evident ( with somewhat shifted size limits ) for particles from other Nup98 FG domains . S . cerevisiae Nup116 FG particles , for example , brightly accumulate NTF2 and the Importin β⋅IBB-Atto488 complex ( Figure 14A ) . The accumulation was already weaker for the Importin β⋅IBB-GFP complex , while an Importin β⋅IBB-MBP-GFP complex got stuck at the particles' surface . The larger Importin β⋅IBB-ZsGreen complex ( with its minimized exposed cargo surface ) could again overcome the boundary and accumulated inside . Likewise , the IBB-MBP-GFP-M9 fusion accumulated inside the particles only when Importin β and Transportin were simultaneously bound to this cargo molecule ( Figure 14B ) . 10 . 7554/eLife . 04251 . 019Figure 14 . Effect of cargo domains and NTR stoichiometry on entry into ScNup116 FG particles . FG particles ( with 5% Atto390-labeled tracer ) were formed with 10 µM S . cerevisiae Nup116 FG domain . See Figure 13 and main text for additional information . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 019 Hence , the self-assembled FG phases very nicely recapitulate a key sorting criterion of authentic NPCs , namely that large cargoes require multiple NTRs and thus a good surface coverage by NTRs for an effective pore passage . In fact , the phenomenon of NTR-cooperation was originally observed with a very similar cargo , namely with an IBB-2xMBP-M9 fusion . This cargo bound well to NPCs in the presence of importin β alone , but rapidly traversed NPC only when both , Importin β as well as Transportin , had been recruited ( Ribbeck and Görlich , 2002 ) . Likewise , Tu et al . , ( 2013 ) recently showed that a single Transportin molecule could confer efficient NPC targeting of a large M9-β-Gal fusion complex , but successful NPC barrier-passage required several Transportin molecules . These remarkable parallels indeed strongly suggest that the permselectivities of in vitro assembled Nup98 FG phases and authentic NPCs are governed by the same mechanistic principles and by functionally equivalent structures . We found that Nup98 FG domains assemble from ( even dilute ) aqueous solutions into very protein-rich phases that reject inert molecules ≥25 kDa in size , but at the same time allow influx of far larger NTR-complexes . They even reproduce the multi-NTR requirement for larger cargoes and recapitulate the kinetics of NPC passage . The most straightforward interpretation of these data is that such phases also assemble inside authentic NPCs and account for the permselectivity of nuclear pores , as previously predicted by the selective phase model . Yet , the concept of cohesive FG repeat interactions as the basis of NPC transport selectivity has been facing a remarkable resistance from the nuclear transport field over the past 13 years . In light of the available data however , it needs to be emphasised that all suggested alternatives to the cohesion concept appear far more complicated and tied to two rather unlikely assumptions . First , some still elusive process would have to suppress phase-separation and counteract cohesive inter-FG repeat interactions in authentic NPCs . Second , an alternative selectivity mechanism would then have to generate the same permselectivity as observed for self-assembled Nup98 FG phases . The assembly of selective Nup98 FG phases within NPCs should a priori be a thermodynamically highly favoured process because the local FG motif concentration exceeds the critical concentration at least 100-fold and the system should thus be highly over-saturated with respect to an FG phase-separation . Any block of such an assembly would therefore require a masking of the sticky parts , i . e . of thousands of cohesive FG repeat units per NPC . If effective ‘cohesion suppressors’ existed , then they would have to be exceedingly abundant NPC ligands , which only leaves NTRs as possible candidates . We tested this and observed that NTRs do not block Nup98 FG phase-separation even if present in 10-fold molar excess over FG domain ( Figure 15 ) . This is not surprising because ( i ) NTR⋅FG domain interactions are governed by an extreme multivalency ( with ≈40 NTR-binding sites on the FG domain and ≈10 FG-binding sites on importin β; Isgro and Schulten , 2005 ) , and ( ii ) such multivalency is prone to drive phase-separation ( Li et al . , 2012 ) —at least for molar ratios that can be accommodated in the context of NPCs . The tested 10-fold molar excess is probably already far greater than what nuclear pores could possibly hold , because an NPC with 100 anchored FG domains would then be liganded with 1000 NTR molecules ( which would at least double the observed mass of an NPC ) . We consider it , however , possible that NTRs can suppress ectopic FG phase-separation outside NPCs , because there , NTRs are present in an ≈100-fold molar excess over Nup98 FG domains ( Hahn and Schlenstedt , 2011; Hülsmann et al . , 2012 ) . 10 . 7554/eLife . 04251 . 020Figure 15 . Formation of FG particles in the presence of NTRs . ( A ) NTRs do not suppress FG particle formation at molar ratios expected within NPCs . FG particles were formed by dilution of TtMacNup98A to 5 µM ( including 5% Atto390-labeled tracer ) with Tris-buffered saline containing the indicated concentrations of Importin β . Approximately , 60 s after particle formation , 3 µM mCherry was added . The CLSM images show that FG particles can still form in the presence of a 10-fold molar excess of NTRs . ( B ) An excess of NTRs does not compromise barrier function . Zoom-ins on the particles indicated in A show that mCherry is excluded in all cases . ( C ) Importin β increases the critical concentration for FG particle formation only when added in very high excess . FG particles were formed by dilution of TtMacNup98A to 1 µM ( including 5% Atto488-labeled tracer ) with Tris-buffered saline containing 5 µM Atto565-labeled ZZ-domain and the indicated concentrations of Importin β . After 10 min of incubation , particles were collected by ultra-centrifugation at ≈125 , 000×g for 1 hr , and the amount of FG domains in the pellet and supernatant analysed by SDS-PAGE . For detection of the labelled FG domains , a Fujifilm FLA-9000 fluorescence imager was used . ( D ) Unlike Importin β , NTF2 does not influence the critical concentration of FG particle formation . Experimental setup as in ( C ) , with the exception that TtMacNup98A FG particles were formed by dilution with buffer containing the indicated concentrations of NTF2 . DOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 020 Alternative models for NPC function , avoiding cohesive interactions as a selectivity principle , have been repeatedly proposed . As we discuss in the following , they lack , however , direct experimental evidence or even collide with experimental observations . The ‘virtual gate’ model ( Rout et al . , 2003 ) , for example , regards FG domains as entropic brushes that repel inert material while NTRs overcome this entropic barrier by binding to them . This model implies that non-cohesive FG domains are sufficient to create a barrier . This is incompatible with the observation that Xenopus NPCs loose their selectivity if the highly cohesive Nup98 FG domain is replaced by non-cohesive ones ( Hülsmann et al . , 2012 ) . The reversible collapse model extended the virtual gate model by the assumptions that an initially highly extended FG brush collapses or contracts when an NTR binds and that this contraction moves the NTR⋅cargo complex towards the anchor point and eventually across the pore ( Lim et al . , 2007 ) . This model relied on an experiment interpreted such that Importin β collapses the Nup153 FG domain , with a half maximum effect occurring at 30 femtomolar Importin β . Considering that the cellular NTR concentration is 300 million times higher ( ≥10 µM; Hahn and Schlenstedt , 2011 ) , this interpretation , however , implies that those domains are collapsed at all times . More generally , this model is inconsistent with the observation that NPCs still exhibit superb permselectivity when saturated with NTRs ( Ribbeck and Görlich , 2001; Yang and Musser , 2006 ) . Moreover , neither of the aforementioned alternative models would have predicted that NPC-like permselectivity can be reconstituted with FG domains alone , i . e . independently of any grafting to a channel's inner face . None of them can explain why the propensity of Nup98 FG domains for spontaneous phase-separation is so perfectly conserved in evolution . And neither can they explain why larger cargoes require multiple NTRs for efficient NPC passage ( Ribbeck and Görlich , 2002; Tu et al . , 2013 ) . The NTR-cooperation effect indeed emphasizes that a simple binding of a translocating species to FG domains is insufficient for facilitated passage , which in fact had been an explicit prediction from the selective phase model ( Ribbeck and Görlich , 2002 ) . As discussed in the following , the hydrogel model provides two complementary perspectives on this effect . In the first one , we consider the barrier as a ‘solvent’ , into which NTRs , but not inert cargo domains , easily partition . An NTR would improve the cargo's solubility through its own FG affinity as well as by shielding the cargo domain . A complete NTR-assisted partitioning of the cargo into the FG phase would then be an essential stage during a successful NPC passage . Is the cargo , however , too large ( in proportion to the NTR ) , then one would expect ‘arrested’ intermediates , in which the NTR has already partitioned into the FG phase , while the cargo domain is still excluded , because the energetic penalty is too high for rapidly immersing into the FG phase . Additional NTR molecules should minimise this penalty by further reducing the cargo's accessible surface area and by their own propensity to partition into the FG gel . Indeed , the experiments of Figure 13 and Figure 14 visualise such behaviour with remarkable clarity . The second perspective explicitly considers the inter-FG repeat contacts , which obstruct the passage for inert material , but transiently and locally melt when an NTR binds the corresponding FG motifs . This melting should be restricted to the immediate vicinity of the NTR and thus become inefficient at distant parts of a large cargo . A second NTR molecule acting at this location should then solve this problem . Notably , the selective phase model per se does not make a prediction as to which cohesive FG domains conduce to barrier formation . Yet , the following observations suggest a prominent contribution of Nup98 FG domains: ( i ) their extreme propensity to form selective barriers on their own ( this study ) , ( ii ) their IDP-untypical evolutionary conservation ( this study ) , ( iii ) their high copy numbers ( Ori et al . , 2013 ) , and ( iv ) their capacity to restore a selective permeability barrier in FG Nup-depleted NPCs ( Hülsmann et al . , 2012 ) . Other FG domains will , however , also contribute by supplying additional ( cohesive or adhesive ) FG mass , by forming distinct FG gel layers ( of similar or distinct selectivity ) and perhaps by forming composite gels with the Nup98 FG domains . Such interplay between distinct FG domains will probably improve the robustness of the system , and it might indeed be an efficient way to fine-tune permselectivity . In the case of Tetrahymena we expect , for example , that such blending will relax the extremely tight TtMacNup98A FG hydrogel to a physiological optimum . Moreover , a perfect control of transport through the NPC requires not only the presence of barrier-forming material , but also that these barrier-elements span the entire cross-section of the central channel and make a tight seal to the scaffold of the pore . This is apparently achieved not only by the direct anchorage of the FG domains , but also by additional interactions between FG repeats and scaffold Nucleoporins , such as yeast Nic96p ( Patel et al . , 2007; Schrader et al . , 2008 ) , Nup188 , Nup192 ( Andersen et al . , 2013 ) , or vertebrate Nup93 ( Xu and Powers , 2013 ) . The resulting scaffold-FG contacts can thus be seen as a functionally important extension of the cohesive inter-FG meshwork .
Every protein used in this study was expressed in the E . coli strains BLR or NEB Express . Constructs and purification strategies for the following proteins have been described before: Transportin and NTF2 ( Ribbeck and Görlich , 2001 ) , IBB-MBP-mEGFP , IBB-ZsGreen , mCherry , MBP-mCherry ( Frey and Görlich , 2009 ) . For all other proteins , new expression vectors have been generated ( See Table 3 ) . If required , coding sequences were adapted and optimized for expression in E . coli . This was particularly important for the FG domain from Tetrahymena because this species uses a non-universal genetic code . Sequences and plasmid maps are available on request . 10 . 7554/eLife . 04251 . 021Table 3 . Proteins and corresponding bacterial expression constructs used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 04251 . 021Protein namePlasmidEncoding forUsed in figuresHsNup98 FGpHBS491His18-HsNup981-499-Cys5B , 6 , 7BfNup98 FGpHBS505His18-BfNup981-478-Cys5B , 6 , 7DmNup98 FGpHBS503His18-DmNup981-580-Cys5B , 6 , 7CeNup98 FGpHBS504His18-CeNup981-493-Cys5B , 6 , 7ScNup100 FGpHBS512His18-ScNup1002-580-Cys1B-F , 2 , 3A , 5B , 6 , 7ScNup100 FGpHBS697His18-ScNup1001-580-Cys4A , 4CScNup116ΔGLEBS FGpHBS514His18-ScNup1162-109 , 167-715-Cys5B , 6 , 7 , 12 , 13ScNup116 FGpHBS698His18-ScNup1161-736-Cys4BDdNup220 FGpHBS241His14-TEV-DdNup2201-718-Cys5B , 6 , 7AtNup98B FGpHBS383His14-TEV-AtNup98B1-668-Cys5B , 6 , 7TtMacNup98A FGpHBS418His18-TtMacNup98A1-666-Cys5B , 6 , 7 , 8 , 12 , 13 , 14 , 15TbNup158 FGpHBS249His14-TEV-TbNup1581-565-Cys5B , 6 , 7ScNsp1274-601 FGpSF654His10-TEV-ScNsp1274-601-Cys1B-CScImpβpMR676His14-brSUMO-ScKap95p2 , 3A , 4 , 13 , 14 , 15A-CHsImpβpICH005His14-scSUMO-HsImpβ5HsTrnpKK006His10-mEGFP-TEV-HsTransportin11B , 13B , 14BHsNTF2pAL239His14-brSUMO-HsNTF22A , 7 , 8 , 13A , 14A , 15DScIBBpHBS695His14-ZZ-brNEDD8-Srp1p2-63-Cys2A , 13A , 14AScIBB-GFPpSF807His14-TEV-ScSrp1p2-63-mEGFP2A , 3A , 4 , 13A , 14AScIBB-MBP-GFPpHBS45His14-TEV-ScSrp1p2-63-MBP-mEGFP2B , 13A , 14AScIBB-MBP-GFP-HsM9pHBS704His14-TEV-ScSrp1p2-63-MBP-mEGFP-hnRNP A1268-30613B , 14BScIBB-ZsGreenpSF881His14-TEV-ScSrp1p2-63-ZsGreen2A , 13A , 14AZZpHBS237His10-ZZ-TEV-Cys1B , 15C-DmCherrypSF846His14-TEV-mCherry7 , 15BGFPpHBS349His18-mEGFP5BMBP-mCherrypSF844His14-TEV-MBP-mCherry2A , 4A-B , 4D , 6MBPpSF1911His14-brSUMO-MBPGly260Cys–His61EHsActinpKG017His14-brSUMO-HsβActin4D All NTRs , transport substrates and inert molecules were purified by virtue of N-terminal His-tags and native Ni ( II ) chelate chromatography . Elution was performed with either imidazole or by on-column protease cleavage ( Frey and Görlich , 2014a , 2014b ) . The tags of all imidazole-eluted proteins were cleaved off in solution with TEV protease , proteins were further purified by gel filtration on a Superdex200 column equilibrated with 44 mM Tris pH 7 . 5 , 290 mM NaCl , 4 . 4 mM MgCl2 , 5 mM DTT , and eventually snap-frozen in liquid nitrogen after addition of 250 mM sucrose . FG domains were purified using Ni ( II ) chelate chromatography under denaturing conditions ( 100 mM Tris pH 8 , 8 M GuHCl , 10 mM DTT ) . Elution was with imidazole in 100 mM Tris pH 8 , 20% formamide . If necessary , they were further purified by covalent chromatography , whereby an engineered C-terminal cysteine was allowed to form a disulfide bond with a 2-thiopyridine-activated SH-silica matrix ( described below ) and elution was achieved by reducing the disulfides with DTT . FG domains were finally re-buffered to 20% acetonitrile , 0 . 08% TFA and lyophilised . O-GlcNAc modification of FG domains was performed as previously described ( Labokha et al . , 2013 ) . The 2-thiopyridine-activated SH-silica matrix was produced by the following steps: ( i ) modifying macroporous silica ( Davisil XWP 1000 Å 35–70 micron [Alltech Grom GmbH , Worms , Germany] ) with 2% ( vol/vol ) 3-glycidoxypropyl-trimethoxysilane ( CAS#2530-83-8 ) in xylene o/n @ 60°C , ( ii ) washing in xylene , ethanol , and finally water , ( iii ) reaction with aqueous 0 . 5 M dithiothreitol buffered with 0 . 1 M Tris/HCl pH 7 . 5 , o/n , 40°C , under argon , ( iv ) thorough washing in oxygen-free water , transfer in degassed isopropanol:water ( 50:50 ) and reaction with 0 . 1 M 2 , 2' dithiodipyridine ( CAS#2127-03-9 ) , 0 . 05 M Tris/HCl pH 7 . 5 in degassed isopropanol:water ( 50:50 ) at 20°C under argon . The matrix was stored after washing in 100% degassed isopropanol at 4°C under argon and transferred to binding buffer immediately prior to use . Covalent chromatography was performed at room temperature . For regeneration , the procedure was repeated at step ( iv ) . For estimation of the critical concentrations for phase-separation , FG particles were formed by diluting soluble , unfolded FG domain stocks in 2M GuHCl with TBS ( 50 mM Tris/HCl pH 7 . 5 , 150 mM NaCl ) to a range of FG domain concentrations ( 10 µM , 5 µM , 2 . 5 µM , 1 . 25 µM , 0 . 6 µM , 0 . 3 µM , 0 . 1 µM , 0 . 06 µM ) . As a first estimate for the saturation concentration , we used the lowest FG domain concentration that yielded detectable particles in the DLS setup ( see Figure 1C ) . This number was then refined by measuring the concentration of apparent monomers that co-existed with already formed particles . Additional method details are given in the legends . | Cells of eukaryotic species—which include plants , animals , and fungi—have a nucleus that harbours the organism's genome . Two membrane layers surround the nucleus and separate its contents from the cytoplasm , where proteins are made . This separation is essential for a correct interpretation of the genetic information . Yet , various molecules , such as proteins , need to move into or out of the nucleus for the cell to work properly . This transit has to occur without an uncontrolled mixing of the contents of the nucleus and the cytoplasm happening . Structures called nuclear pore complexes span the double membrane and allow material to be exchanged between the nucleus and the cytoplasm . Small molecules can freely pass through these complexes , while larger molecules can only be transported when bound as “cargo” to so-called nuclear transport receptors . Nuclear pore complexes are large assemblies of proteins called nucleoporins . FG nucleoporins are special in that they contain regions with a repeating pattern of two amino acids , phenylalanine ( ‘F’ ) and glycine ( ‘G’ ) . These regions are called FG domains . They bind to nuclear transport receptors and have been suspected to form a barrier that decides which molecules may pass through the nuclear pore complex . Exactly how this control is exerted has been a matter of debate . Versions of a particular FG nucleoporin called Nup98 are found in all branches of eukaryotic life , i . e . in animals , fungi , plants , amoebas , and even in the evolutionarily most distant protozoans . When Schmidt and Görlich dispersed small amounts of Nup98 FG domains in an aqueous solution , the domains rapidly attracted each other to form ‘FG particles’ , regardless of which species the proteins came from . These FG particles were so dense that they repelled ‘normal’ macromolecules , yet they allowed nuclear transport receptors , along with their bound cargoes , to rapidly enter . Taken together , the work of Schmidt and Görlich suggests that such FG particles form the transport barrier in nuclear pore complexes . Based on these findings , Schmidt and Görlich refine a model where the FG domains form a mesh in the nuclear pore complexes that acts like a ‘smart sieve’ . Smaller molecules can move through gaps in the meshwork , but larger molecules are hindered . Schmidt and Görlich suggest that nuclear transport receptors help large molecules to move through nuclear pore complexes by ‘melting’ the FG meshwork locally , creating a path for the molecule to move through . The reconstitution of these smart barriers in the laboratory will now allow researchers to analyse the process of receptor-mediated nuclear pore passage in unprecedented ( mechanistic ) detail . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2015 | Nup98 FG domains from diverse species spontaneously phase-separate into particles with nuclear pore-like permselectivity |
Progression through the mitotic cell cycle requires periodic regulation of gene function at the levels of transcription , translation , protein-protein interactions , post-translational modification and degradation . However , the role of alternative splicing ( AS ) in the temporal control of cell cycle is not well understood . By sequencing the human transcriptome through two continuous cell cycles , we identify ~1300 genes with cell cycle-dependent AS changes . These genes are significantly enriched in functions linked to cell cycle control , yet they do not significantly overlap genes subject to periodic changes in steady-state transcript levels . Many of the periodically spliced genes are controlled by the SR protein kinase CLK1 , whose level undergoes cell cycle-dependent fluctuations via an auto-inhibitory circuit . Disruption of CLK1 causes pleiotropic cell cycle defects and loss of proliferation , whereas CLK1 over-expression is associated with various cancers . These results thus reveal a large program of CLK1-regulated periodic AS intimately associated with cell cycle control .
Alternative splicing ( AS ) is a critical step of gene regulation that greatly expands proteomic diversity . Nearly all ( >90% ) human genes undergo AS and a substantial fraction of the resulting isoforms are thought to have distinct functions ( Pan et al . , 2008; Wang et al . , 2008 ) . AS is tightly controlled , and its mis-regulation is a common cause of human diseases ( Wang and Cooper , 2007 ) . Generally , AS is regulated by cis-acting splicing regulatory elements that recruit trans-acting splicing factors to promote or inhibit splicing ( Matera and Wang , 2014; Wang and Burge , 2008 ) . Alterations in splicing factor expression have been observed in many cancers and are thought to activate cancer-specific splicing programs that control cell cycle progression , cellular proliferation and migration ( David and Manley , 2010; Oltean and Bates , 2014 ) . Consistent with these findings , several splicing factors function as oncogenes or tumor suppressors ( Karni et al . , 2007; Wang et al . , 2014 ) , and cancer-specific splicing alterations often affect genes that function in cell cycle control ( Tsai et al . , 2015 ) . Progression through the mitotic cell cycle requires periodic regulation of gene function that is primarily achieved through coordination of protein levels with specific cell cycle stages ( Harashima et al . , 2013; Vermeulen et al . , 2003 ) . This temporal coordination enables timely control of molecular events that ensure accurate chromatin duplication and daughter cell segregation . Periodic gene function is conventionally thought to be achieved through stage-dependent gene transcription ( Bertoli et al . , 2013 ) , translation ( Grabek et al . , 2015 ) , protein-protein interactions ( Satyanarayana and Kaldis , 2009 ) , post-translational protein modifications , and ubiquitin-dependent protein degradation ( Mocciaro and Rape , 2012 ) . Although AS is one of the most widespread mechanisms involved in gene regulation , the relationship between the global coordination of AS and the cell cycle has not been investigated . Major families of splicing factors include the Serine-Arginine rich proteins ( SR ) proteins and the heterogeneous nuclear ribonucleoproteins ( hnRNPs ) , whose levels and activities vary across cell types . SR proteins generally contain one or two RNA recognition motifs ( RRMs ) and a domain rich in alternating Arg and Ser residues ( RS domain ) . Generally , RRM domains confer RNA binding specificity while the RS domain mediates protein-protein and protein-RNA interactions to affect splicing ( Long and Caceres , 2009; Zhou and Fu , 2013 ) . Post-translational modifications of SR proteins , most notably phosphorylation , modulate their splicing regulatory capacity by altering protein localization , stability or activity ( Gui et al . , 1994; Lai et al . , 2003; Prasad et al . , 1999; Shin and Manley , 2002 ) . Dynamic changes in SR protein phosphorylation have been detected after DNA damage ( Edmond et al . , 2011; Leva et al . , 2012 ) and during the cell cycle ( Gui et al . , 1994; Shin and Manley , 2002 ) , suggesting that regulation of AS may have important roles in cell cycle control . However , the functional consequences of SR protein ( de ) phosphorylation during the cell cycle are largely unclear . Through a global-scale analysis of the human transcriptome at single-nucleotide resolution through two continuous cell cycles , we have identified widespread periodic changes in AS that are coordinated with specific stages of the cell cycle . These periodic AS events belong to a set of genes that is largely separate from the set of genes periodically regulated during the cell cycle at the transcript level , yet the AS regulated set is significantly enriched in cell cycle- associated functions . We further demonstrate that a significant fraction of the periodic AS events is regulated by the SR protein kinase , CLK1 , and that CLK itself is also subject to cell cycle-dependent regulation . Moreover , inhibition or depletion of CLK1 causes pleiotropic defects in mitosis that lead to cell death or G1/S arrest , suggesting that the temporal regulation of splicing by CLK1 is critical for cell cycle progression . The discovery of periodic AS thus reveals a widespread yet previously underappreciated mechanism for the regulation of gene function during the cell cycle .
To systematically investigate the regulation of AS during the cell cycle , we performed an RNA-Seq analysis of synchronously dividing cells using a total of 2 . 3 billion reads generated across all stages ( G1 , S , G2 and M ) of two complete rounds of the cell cycle ( Figure 1—figure supplement 1A ) . To maximize the detection of regulated AS events , we used the complementary analysis pipelines , MISO and VAST-TOOLS ( Katz et al . , 2010; Irimia et al . , 2014 ) . These pipelines have different detection specificities and employ partially overlapping reference sets of annotated AS events , and therefore afford a more comprehensive analysis when employed together . Both pipelines were used to determine PSI ( the percent of transcript with an exon spliced in ) and PIR ( the percent of transcripts with an intron retained ) . Alternative exons detected by both pipelines had highly correlated PSI values ( Figure 1—figure supplement 1G; see below ) . Consistent with previous results ( Bar-Joseph et al . , 2008; Whitfield et al . , 2002 ) , transcripts from approximately 14 . 2% ( 1182 ) of expressed genes displayed periodic differences in steady-state levels between two or more cell cycle stages ( see below ) . Remarkably , 15 . 6% ( 1293 ) of expressed genes also contained 1747 periodically-regulated AS events , among a total of ~40 , 000 detected splicing events ( FDR < 2 . 5%; Figure 1A and Figure 1—figure supplement 1B , D ) . 10 . 7554/eLife . 10288 . 003Figure 1 . Global detection of periodic cell cycle-dependent alternative splicing . ( A ) Heat map representation of periodically spliced events . Row-normalized relative PSI values are shown . Diagram below indicates cell cycle phase . ( B ) Overlap between periodically spliced genes and periodically expressed genes detected by RNA-Seq . ( C ) Heat map representation of enriched Gene Ontology terms shown as log ( p-value ) . Three gene sets were analyzed separately: all genes with periodic AS , genes with periodic AS only , and genes with both periodic AS and periodic expression . ( D ) Real-time quantitative PCR analysis of periodic retained introns and total mRNAs for three selected genes . Cells were synchronized by double thymidine block and samples were collected 0 , 3 , 6 , 9 , 12 and 15 hr post release . Errors bars represent standard deviation of the mean . Diagram below indicates cell cycle stage . ( E ) Schematic representation of AURKB AS pattern . Line graph showing the relationship between intron retention and mRNA levels for the AURKB gene across the cell cycle . Percent intron retention ( solid red line ) across cell cycle was used to determine the fraction of total mRNAs ( solid blue line ) not containing an intron , i . e . ‘corrected’ mRNA levels ( dashed blue line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 00310 . 7554/eLife . 10288 . 004Figure 1—figure supplement 1 . Identification of periodic AS by multiple analysis pipelines . ( A ) Number of sequencing reads per sample ( top ) . RNA-Seq reads and periodic seeds used for the identification of all periodically expressed and spliced genes ( bottom , see methods ) . ( B , C ) Dot plot of periodic score and false discovery rate ( FDR ) for each exon analyzed by the MISO and VAST-TOOLS analysis pipelines . Dashed lines show FDR and periodic score cutoff ( see methods ) . ( D ) Heat map representation of periodically-spliced events identified by the VAST-TOOLS pipeline . Data are row-normalized . Diagram below indicates cell cycle stage . ( E ) Bar graphs showing the number of periodic AS events identified separated by event type and shown as a fraction of total events identified ( SE: skipped exon , RI: retained intron , A3: alternative 3’splice site , A5: alternative 5’splice site ) . MISO analysis ( left panel in blue ) and VAST-TOOLS analysis ( right panel in red ) . ( F ) Venn diagram representation of the overlap between periodic AS identified by VAST-TOOLS and periodically expressed mRNAs ( top ) . Venn diagram representation of the overlap between periodic AS events as identified by both VAST-TOOLS and MISO ( bottom , see methods ) . ( G ) Spearman’s rank correlation analysis of each cell cycle time point according to commonly detected alternative exons by MISO and VAST-TOOLS . Spearman’s rho values are shown in heat map . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 004 Importantly , as has been observed previously for AS regulatory networks ( Pan et al . , 2004 ) , the majority of genes with periodic AS events did not overlap those with periodic steady-state changes in mRNA expression . This indicates that genes with periodic changes in AS and transcript levels are largely independently regulated during the cell cycle ( Figure 1B ) . Further supporting this conclusion , we did not observe a significant correlation ( positive or negative ) between exon PSI values and mRNA expression levels for genes with both periodic expression and periodic exon skipping ( data not shown ) . A gene ontology ( GO ) analysis reveals that genes with periodic AS , like those with periodic transcript level changes ( Bar-Joseph et al . , 2008; Whitfield et al . , 2002 ) , are significantly enriched in cell cycle-related functional categories , including M-phase , nuclear division and DNA metabolic process ( Figure 1C; adjusted p<0 . 05 for all listed categories , FDR<10% ) ( Supplementary file 1 ) . Similar GO enrichment results were observed when removing the relatively small fraction ( 10% ) of periodically spliced genes that also display significant mRNA expression changes across the cell cycle ( Figure 1C ) . These results thus reveal that numerous genes not previously linked to the cell cycle , as well as previously defined cell cycle-associated genes thought to be constantly expressed across the cell cycle , are in fact subject to periodic regulation at the level of AS ( Supplementary file 1 for a full list ) . Among the different classes of AS analyzed ( cassette exons , alternative 5'/3' splice sites and intron retention [IR] ) , periodically regulated IR events were over-represented ( relative to the background frequency of annotated IR events ) by ~2 . 2 fold whereas periodically regulated cassette exons , represent the next most frequent periodic class of AS ( p=2 . 2×10-16 , Fisher’s exact test , Figure 1—figure supplement 1E ) . Quantitative RT-PCR assays across different cell cycle stages validated periodic IR events detected by RNA-Seq ( Figure 1D ) . Interestingly , one of these IR events is in transcripts encoding aurora kinase B ( AURKB ) , a critical mitotic factor regulated at the levels of transcription , protein localization , phosphorylation and ubiquitination ( Carmena et al . , 2012; Lens et al . , 2010 ) . The AURKB retained intron is predicted to introduce a premature termination codon that elicits mRNA degradation through nonsense mediated decay , and is thus expected to result in reduced levels of AURKB protein . The splicing of the retained intron lags behind changes in the total AURKB mRNA expression ( Figure 1E ) . We computationally corrected levels of fully spliced , protein coding AURKB mRNA by taking into account the fraction of intron-retaining ( i . e . non-productive ) transcripts across the cell cycle stages ( Figure 1E ) . The expression curve for corrected AURKB mRNA levels is substantially different from total AURKB transcript levels , with a shifted peak coinciding with mitosis . Periodically-regulated IR events detected in other genes , including those with known cell cycle functions such as HMG20B and RAD52 , are similarly expected to affect the cell cycle timing of mRNA expression ( Figure 1A , D ) . Collectively , these results provide evidence that the temporal control of retained intron AS provides an important mechanism for establishing the timing of expression of AURKB mRNA and protein , as well as of the timing of expression of additional genes during the cell cycle . Alternative splicing is generally regulated by the concerted action of multiple cis-elements that recruit cognate splicing factors . Consistently , analysis of our RNA-seq data revealed 96 RNA binding proteins ( RBPs ) with periodic mRNA expression , including RS domain-containing factors like SRSF2 , SRSF8 , TRA2A and SRSF6 ( Figure 2—figure supplement 1A ) . These 96 RBPs were significantly enriched in the GO term 'splicing regulation' ( adjusted p=10-4 , Figure 2—figure supplement 1B ) , indicating that periodic AS is likely controlled by multiple RBPs . Correlations between these RBPs and periodic splicing events were also identified ( Figure 2—figure supplement 1C ) . For example , SRSF2 expression is significantly correlated with the splicing pattern of a retained intron in the SRSF2 transcript . Further supporting a role for these RBPs in controlling periodic splicing was the identification of RNA motifs bound by a subset of periodically expressed RBPs ( Figure 2—figure supplement 1D ) . To further examine periodic RBP regulation during cell cycle , we measured the abundance of known splicing regulatory proteins at different stages of the cell cycle by immunoblotting ( Figure 2A ) . Among the proteins analyzed , CDC-like kinase 1 ( CLK1 ) , an important regulator of the Ser/Arg ( SR ) repeat family of splicing regulators , displayed the strongest cyclic expression peaking at the G2/M phase ( Figure 2A , B ) , consistent with the results of a recent mass-spectrometry-based screen for cycling proteins ( Ly et al . , 2014 ) . CLK1 is one of four human CLK paralogs ( CLK1-4 ) and is known to regulate AS via altering the phosphorylation status of multiple SR proteins ( Duncan et al . , 1997; Jiang et al . , 2009; Ninomiya et al . , 2011; Prasad et al . , 1999 ) . Notably , the levels of other detectable CLK paralogs , as well as members of another SR protein kinase , SRPK1 , did not change significantly at the level of RNA and/or protein during the cell cycle ( Figure 2B and Supplementary file 1 ) . 10 . 7554/eLife . 10288 . 005Figure 2 . Cell cycle-dependent regulation of CLK1 . ( A ) Immunoblot analysis of proteins involved in splicing regulation in synchronized HeLa cells after release from double thymidine block . ( B ) Immunoblot analysis of selected proteins in asynchronous HeLa cells or cells arrested at different cell cycle stages . Stably expressed exogenous CLK1 levels were also assessed during the cell cycle ( bottom panel ) . ( C ) Immunoblot of endogenous CLK1 ( top ) and exogenously-expressed wild type ( CLK1wt ) or kinase catalytically inactive ( CLK1KD ) proteins ( bottom ) upon treatment with 10 µM TG003 . ( D ) Co-expression of CLK1WT and CLK1KD at different ratios . ( E ) Immunoprecipitation of CLK1 proteins co-expressed with myc-ubiquitin . Cells were treated with 10 μM TG003 and 10 μM MG132 prior to sample collection . ( F ) Immunoblot analysis of lysates from cells synchronized upon early S phase ( double thymidine ) release with or without TG003 treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 00510 . 7554/eLife . 10288 . 006Figure 2—figure supplement 1 . Periodic expression of RBPs . ( A ) Heat map representation of RNA-bound proteins ( RBPs ) with periodic expression . Row-normalized FPKM levels are shown . ( B ) GO analyses for the functional enrichment in the periodic RBPs . ( C ) Number of periodic AS events that significantly correlate ( Spearman’s Rho > | . 75| , p< . 05 ) with the expression pattern of each RBP during cell cycle . Expression pattern of known two known splicing factors , SRSF2 and ESRP2 , is shown in inset . ( D ) Average PSI values of periodic that peak at either G1 ( red line ) or M phase ( blue line ) ( top panel ) . k-mer enrichment in periodic exons as judged by Z score and separated by cell cycle phase ( y-axis = G1-S and x-axis = G2-M ) ( bottom panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 00610 . 7554/eLife . 10288 . 007Figure 2—figure supplement 2 . Regulation of CLK1 proteins levels during the cell cycle is degradation-dependent . ( A ) CLK1 mRNAs as measured in synchronized cells by RNA-Seq ( left ) or in cells arrested at each cell cycle phase , followed by quantitative RT-PCR ( right ) . ( B ) Diagram depicting alternative the splicing pattern for CLK1 pre-mRNA ( left ) . The short form represents skipping of exon 4 that introduces a premature stop codon and is targeted by the non-sense mediated decay pathway , while long form represents the full-length active isoform . Right panels show levels of CLK1 variants , as measured by semi-quantitative RT-PCR with primers that simultaneously detect both forms , in cells upon early S phase release or in cells arrested at specific stages . ( C ) Protein stability of CLK1 is affected by its activity . Cyclohexamide chase experiment was used to measure the stability of CLK1 in cells expressing either CLK1wt or the catalytic mutant CLK1KD . MG132 was used to block proteasomal degradation . ( D ) Immunoprecipitation of Flag-CLK1 from cells after 3 hrs of MG132 treatment and subsequent detection of polyubiquitination by immonoblotting . The arrow indicates the expected position of unmodified CLK1 . Bottom panel shows FLAG-CLK1 protein . ( E ) HeLa cells stably expressing Flag-CLK1 were synchronized by double thymidine block in the presence or absence of TG003 . Phosphorylated histone 3B was detected as cell cycle marker . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 007 Given that both CLK1 protein levels and known CLK1 substrates are periodically expressed , we decided to further investigate the role of CLK1 in the context of cell cycle . The levels of total CLK1 mRNA , as well as the levels of specific CLK1 splice variants , did not change significantly during the cell cycle ( Figure 2—figure supplement 2A , B ) indicating that periodic expression of CLK1 is controlled at the level of protein translation and/or turnover . Consistent with this , an exogenously expressed CLK1 protein displayed cell cycle-dependent fluctuations similar to those observed for endogenous CLK1 protein ( Figure 2B ) . Moreover , CLK1 was rapidly degraded upon inhibition of translation by cycloheximide , and this effect was reversed by co-treatment with the proteasome inhibitor MG132 ( Figure 2—figure supplement 2C ) . Additionally , polyubiquitination of Flag-tagged CLK1 was detected following immunoprecipitation with anti-Flag antibody from cells treated with MG132 ( Figure 2—figure supplement 2D ) . These data suggest that the levels of CLK1 protein are controlled by ubiquitin-mediated degradation in a cell cycle-dependent manner . Periodically regulated protein levels are often controlled through negative feedback circuits involving auto-regulatory loops . CLK1 has been reported to auto-phosphorylate on several residues ( Ben-David et al . , 1991 ) . To investigate whether auto-phosphorylation of CLK1 affects its periodic regulation , we tested whether blocking its kinase activity affects its stability . Inhibition of CLK1 kinase activity using a selective inhibitor , TG003 ( Muraki et al . , 2004 ) , markedly stabilizes both endogenous and exogenously expressed CLK1 proteins ( Figure 2C ) . Moreover , activity-dependent destabilization of CLK1 was observed with a wild type ( WT ) protein , but not with a catalytically inactive ( KD ) mutant ( Figure 2C , left panel ) . We further observe that WT CLK1 is rapidly degraded upon cycloheximide treatment , whereas the KD mutant is more stable ( Figure 2—figure supplement 2C ) . We next tested whether CLK1 activity is sufficient to trigger its own degradation by co-expressing KD CLK1 with increasing amounts of WT CLK1 . As expected , increasing amounts of WT CLK1 reduces levels of KD CLK1 ( Figure 2D ) . Consistent with these results , WT CLK1 is more highly polyubiquitinated compared to the KD mutant ( Figure 2E , compare lanes 2 to 4 ) , and treatment with TG003 reduces polyubiquitination levels ( Figure 2E , lane 2 vs . 3 and lane 4 vs . 5 ) . Decreased polyubiquitination of WT CLK1 is more prevalent than is apparent upon TG003 treatment , as CLK1 is stabilized by TG003 inhibition and thus more total Flag-CLK1 is immunoprecipitated ( Figure 2E ) . To further examine whether this auto-feedback loop is required for changes in CLK1 protein levels during the cell cycle , we treated synchronized cells with TG003 ( or DMSO as a control ) and measured CLK1 protein levels . We observed that CLK1 inhibition prevents its turnover after the G2/M phase for both endogenous and exogenously expressed kinases ( Figure 2F and Figure 2—figure supplement 2E ) . Taken together , these results provide strong evidence that CLK1 protein levels are controlled by ubiquitin-mediated proteolysis in a cell cycle stage-specific manner , and that an activity-dependent negative feedback loop is required for this periodic regulation . These results further suggest that changes in the levels of CLK1 could account for many of the periodically regulated AS transitions we have detected during the cell cycle . RNA-Seq analysis of cells treated with TG003 revealed 892 AS events ( in 665 genes ) that significantly change after CLK1 inhibition ( Figure 3A ) , including known CLK1-regulated splicing events ( e . g . exon 4 of CLK1 [Duncan et al . , 1997] ) . It is worth noting that TG003 can also inhibit CLK4 ( although to a lesser extent than inhibition of CLK1 ) . However , RNAi of CLK1 is sufficient to recapitulate the phenotype of CLK1/4 inhibition ( see below and Figure 4 ) ( Fedorov et al . , 2011; Muraki et al . , 2004 ) . Intron retention and cassette exons are the most overrepresented types of AS affected by TG003 ( Figure 3A ) . Most ( 70% ) of the CLK1-regulated exons display increased skipping upon CLK1 inhibition , whereas 87% of CLK1-regulated introns show increased retention ( Figure 3—figure supplement 1B and C ) , consistent with a recently reported role for CLK1 in the regulation of retained introns ( Boutz et al . , 2015 ) . Of nine analyzed TG003-affected AS events detected by RNA-Seq analysis , all were validated by semi-quantitative RT-PCR assays ( Figure 3B ) . These observations indicate that CLK1 inhibition mainly suppresses splicing , consistent with a general requirement for phosphorylation of SR proteins to promote splicing activity ( Irimia et al . , 2014; Prasad et al . , 1999; Tsai et al . , 2015 ) . Importantly , there is a significant overlap between genes with cell cycle periodic AS events ( Figure 1 ) and those with CLK1-regulated AS events , involving 156 genes ( p=8 . 5×10-10 , hyper-geometric test ) . In contrast , consistent with the results in Figure 1 , we do not observe a significant overlap between genes containing CLK1-regulated AS events and periodically expressed genes . These results thus support a widespread and rapidly acting role for CLK1 in controlling cell cycle-regulated AS . Indeed , CLK1 inhibition induces rapid ( within 3–6 hr ) changes in AS among several analyzed cases ( Figure 3—figure supplement 1D ) . 10 . 7554/eLife . 10288 . 008Figure 3 . CLK1 regulates a network of genes that control cell cycle progression . ( A ) Identification of endogenous CLK1 targets by RNA-Seq . Numbers of different AS types affected by treatment with the CLK1 inhibitor TG003 ( left graph ) . SE , skipped exon; RI , retained intron; A3E , alternative 3’ exon; A5E , alternative 5’ exon . Fraction of total analyzed events that were affected by TG003 treatment ( right graph ) . ( B ) Validation of TG003-responsive AS events by semi-quantitative RT-PCR . The bar graph shows the max-delta PSI for each AS event tested in a 24-hr time course of inhibition with 20μM TG003 . ( C ) Representation of cell cycle control genes with CLK1-dependent AS events , organized by cell cycle phase and function . ( D ) Schematic representation of CHEK2 alternative splicing , showing that exon 9 encodes a region overlapping the kinase domain ( upper panel ) . Semi-quantitative RT-PCR assessment of CHEK2 isoforms after treatment with TG003 or over-expression of the indicated factors ( lower left panel ) . RNAi of SRSF1 in cells and subsequent analysis of CHEK2 splicing by semi-quantitative RT-PCR ( lower right panel ) . PSI values are shown below gel . ( E ) Normalized CENPE total mRNA expression during an unperturbed cell cycle ( triangle denotes mitosis , left panel ) and diagram of CENPE splicing ( right ) . TG003-treatment of HeLa cells released from G1/S arrest followed by semi-quantitative RT-PCR analysis of CENPE isoforms ( bar graph ) . ( F ) Schematic of RNA-Seq analysis of CLK1 inhibition during cell cycle ( left ) . AS events that were identified as being differentially regulated between G1 and G2 phase ( top bar of bar graph ) and number of events that were blocked by the indicated conditions ( bottom 4 bars ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 00810 . 7554/eLife . 10288 . 009Figure 3—figure supplement 1 . CLK1 regulates a network of genes that control cell cycle progression . ( A ) Identification of endogenous CLK1 targets by RNA-Seq . Cells were treated with 10 μM TG003 for 18 hr and poly-A+ RNA was sequenced . Genes significantly up-regulated or down-regulated upon CLK1 inhibition ( log ( FPKM ) are plotted , all plotted genes meet p<10-7 ) . ( B ) Scatter plot representation of cassette exon inclusion after TG003 treatment by MISO analysis . Log Bayes factor values are shown on y-axis and delta PSI on x-axis . ( C ) Scatter plot representation of retained intron levels after TG003 treatment . Log Bayes factor values are shown on y-axis and delta PSI on x-axis . ( C ) Time course experiment after TG003 treatment . Cells were collected at the indicated time points and the absolute delta PSI is plotted . ( E ) Immonoblot analysis of reciprocal co-immunoprecipation experiments between CHEK2 isoforms as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 00910 . 7554/eLife . 10288 . 010Figure 4 . CLK1 is required for cell cycle progression and proliferation . ( A ) Immunoblot analysis of CLK1 proteins after stable shRNA knockdown in HeLa cells . Bottom , DNA content as measured by propidium iodide staining following flow cytometry . ( B ) Immunofluorescence microscopy of A549 cells depleted of CLK1 by shRNA ( top row ) , cells treated with 10 µM TG003 for 12 hr ( middle row ) , and a control treatment with DMSO ( bottom row ) ; green: tubulin , red: emerin ( nuclear envelope ) , and blue: DAPI . Scale bar 10 µm . Right bar graph shows the quantification of multinucleated cells . p values determined using Student’s t-test . ( C ) Static frames from a live-cell high-content imaging movie of HeLa cells expressing Histone H2B-GFP and treated with TG003 ( top panel ) . Time after start of the experiment is indicated; EP , end point ( ~960 min ) . TG003 treated cells with apparent cell division defects ( indicated by arrowheads in the bottom field ) are shown in two independent fields . ( D ) Synchronized HeLa cells were treated with 20 µM TG003 at the indicated time points ( 0 , 5 , and 10 hr ) and analyzed by propidium iodide staining and flow cytometry to measure DNA content . Percent of 2N ( lower bar graph ) and 4N ( upper bar graph ) cells were quantified at each time point as indicated in the treatment scheme ( top ) . ( E ) Colony formation assay of HeLa cells depleted of CLK1 by shRNA , or continuously treated with TG003 or KHCB-19 at the indicated concentrations . ( F ) Box plot representation of CLK1 mRNA expression levels in paired normal and tumorous kidney tissue . 72 cases were analyzed . ( G ) Kaplan-Meier plot showing survival differences between patients with kidney tumors with high CLK1 ( red , upper quartile ) or reduced CLK1 ( blue , lower three quartiles ) expression . ( H ) Number of cancer-associated AS events that are also regulated by CLK1 in different tumor types . BRCA , Breast invasive carcinoma; COAD , Colorectal adenocarcinoma; KIRC , Kidney renal clear cell carcinoma; LUAD , Lung adenocarcinoma; LUSC , Lung squamous cell carcinoma; LIHC , liver hepatocellular carcinoma . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 01010 . 7554/eLife . 10288 . 011Figure 4—figure supplement 1 . Loss of CLK1 results in cell cycle defects in multiple cell types . ( A ) Cell cycle composition as measured by propidium iodide staining of DNA and flow cytometry analysis of cells that have been depleted of CLK1 by the indicated shRNAs . ( B ) Representative histograms of propidium iodide stained H157cells to determine cell cycle defect after RNAi of CLK1 . ( C ) HeLa cells treated with TG003 also have defective cell division . Immunofluorescence microscopy was used to detect multi-nucleated cells , and a representative field is shown ( green:tubulin , red:emerin , blue:DAPI ) . ( D ) Representative histograms of DNA content in synchronous HeLa cells treated with TG003 . The time after early S phase release ( when TG003 was added ) is indicated , These data are associated with Figure 4 of the main text . ( E ) Representative image from anchorage-independent growth assays ( soft agar assay ) of HeLa cells after depletion of CLK1 . ( F ) Relative mRNA expression levels of CENPE and HMMR during cell cycle . Data obtained from RNA-Seq analysis . Dashed line represents 6 hr after release from G1/S . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 01110 . 7554/eLife . 10288 . 012Figure 4—figure supplement 2 . CLK1 mis-regulation in human cancer . ( A ) Overlap of AS events that were altered in kidney cancers ( as compared to normal kidney samples ) and AS that was altered in asynchronous cells treated with TG003 ( left panel ) . Pie chart denoting if AS in cancer occurred in the expected direction , that is , normal kidney resembled TG003 treatment while tumor kidney resembled untreated cells ( see methods ) . ( B ) Boxplot representation of CLK1 , CLK2 , CLK3 and CLK4 mRNA levels in 72 paired normal vs . cancer kidney cancer ( cRCC ) samples . Kolmogorov-Smirnov test significance for each factors is as follows CLK1: p=3 × 10-5 , CLK2: p=1 . 2 × 10-7 , CLK3: p=2 . 8 × 10-12 , CLK4: p=2 . 2 × 10-16 . ( C ) Kaplan-Meier plot of kidney ( cRCC ) patients with tumors expressing high CLK4 ( red , upper quartile ) vs . normal CLK4 ( blue , 1–3 quartile ) . ( D ) PARD3 exon ( chr10:34661426–34661464 ) PSI levels in five cancer cases are shown as an example of TG003-sensitive AS which is altered between normal and cancer tissues . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 012 Supporting an important role for CLK1 in cell cycle progression , genes whose AS levels are affected by CLK1 inhibition are significantly enriched in the GO terms cell cycle phase , M-phase , DNA metabolic processes , nuclear division , DNA damage response , and cytokinesis ( adjusted p<0 . 05 and FDR <20% for all listed GO terms; full list in Supplementary file 2 ) . The affected genes function at various stages of cell cycle including the G1/S transition ( Figure 3C ) . Mitotic processes were , however , associated with the largest number of CLK1-target genes with AS changes and included examples that function in centriole duplication ( CEP70 , CEP120 , CEP290 , CEP68 , CDK5RAP2 ) , metaphase and anaphase ( e . g . CENPK , CENPE , CENPN ) , and cytokinesis ( e . g . SEPT2 , SEPT10 , ANLN ) ( additional examples in Figure 3C ) . To further investigate the functional consequence of CLK1-dependent AS , we selected two examples in genes that have important roles in the cell cycle: checkpoint kinase 2 ( CHEK2 ) , a tumor suppressor that controls the cellular response to DNA damage and cell cycle entry ( Paronetto et al . , 2011; Staalesen et al . , 2004 ) , and centromere-associated protein E ( CENPE ) , a kinetochore-associated motor protein that functions in chromosome alignment and segregation during mitosis ( Kim et al . , 2008 ) . We detected a TG003 dose-dependent increase in CHEK2 exon 9 inclusion , whereas overexpression of WT CLK1 induced exon 9 skipping , an event that removes the CHEK2 kinase domain ( Figure 3D ) . Expression of WT CLK1 in the presence of TG003 , or a catalytically inactive CLK1 , had little to no effect on the splicing of this exon ( Figure 3D , bottom panels ) , indicating that the catalytic activity of CLK1 is essential for regulating CHEK2 AS . Over-expression of the SR protein splicing regulator SRSF1 , a known target of CLK1 ( Prasad et al . , 1999 ) , had a similar effect as over-expression of CLK1 , resulting in CHEK2 exon 9 skipping , whereas knockdown of SRSF1 had the opposite effect ( Figure 3D , right panel ) . Furthermore , the activation of CHEK2 requires homodimerization ( Shen et al . , 2004 ) , and we observe that the CHEK2 isoform lacking exon 9 still interacts with full-length protein ( Figure 3—figure supplement 1E ) , suggesting that this CLK1-regulated isoform may function in a dominant-negative manner to attenuate CHEK2 activity . CENPE is known to be tightly controlled at multiple levels ( including transcription , localization , phosphorylation and degradation ) , and disruption of its regulation leads to pronounced mitotic defects . CENPE AS generates long and short isoforms ( Supplementary file 2 ) , with the predominant variant being the short isoform that lacks amino acids 1972–2068 . Inhibition of CLK1 rapidly shifts CENPE splicing to produce predominantly the long isoform ( Figure 3E ) , and this is accompanied by a reduction in CENPE protein levels during G2/M phase , presumably due to the instability of the long isoform ( Figure 3E ) . These data thus show that CLK1 controls the AS of major cell cycle regulators , and therefore suggest that inhibition of CLK1 may alter cell cycle progression . To investigate this , we next performed an RNA-Seq analysis of synchronized cells at G1 and early G2 ( when CLK1 accumulates ) , following inhibition of CLK1 with TG003 . As a specificity control , we performed a parallel RNA-Seq analysis using a structurally distinct CLK1 inhibitor , KHCB-19 ( Fedorov et al . , 2011 ) . Strikingly , of 1498 AS events that change between G1 ( t=0 ) and G2 ( t=6 ) ~94% display reduced changes following treatment with the two drugs ( Figure 3F ) , with 65% commonly affected by both drugs , thus supporting an important role for CLK1 in controlling cell-cycle dependent splicing . Given that CLK1 regulates the AS of many cell cycle factors ( Figure 3C ) , we next examined whether it is necessary for cell cycle progression . Knockdown of CLK1 using shRNAs led to an accumulation of cells with 4N DNA content in multiple cell types , specifically HeLa , H157 , and A549 ( Figure 4A , Figure 4—figure supplement 1A , B ) , and the extent of this accumulation correlated with the degree of knockdown ( Figure 4—figure supplement 1A ) . We also observed a significant increase in multi-nucleation , a common consequence of defective chromosome segregation or cytokinesis , following shRNA-knockdown or TG003 inhibition of CLK1 in the treated cells ( Figure 4B and Figure 4—figure supplement 1C ) . To visualize the effect of CLK1 inhibition on mitosis at a single cell level , we performed time-lapse high-content microscopy on live cells stably expressing a GFP-histone 2B fusion protein , to track changes in chromatin . TG003-treated cells entered mitosis normally , as measured by nuclear envelope breakdown , but displayed delayed or aberrant cytokinesis , typically resulting in multi-polar divisions , increased time in metaphase , failure to undergo chromatin de-condensation and eventual cell death ( Figure 4C and Videos 1–5 ) . To further determine at what cell cycle stage CLK1 activity is required , we inhibited CLK1 using TG003 at different time points after early S phase release . Consistent with the imaging data , both control and TG003-treated cells entered mitosis normally , as measured by 4N DNA content . However , inhibition of CLK1 before late S-phase impaired progression through mitosis , whereas cells treated 5 hr after early S phase release underwent a round of normal mitotic division , although failed to enter the next cell cycle ( Figure 4D and Figure 4—figure supplement 1D ) . These results suggest that the primary defects caused by CLK1 inhibition occur in late S-phase and G2 phase , which is when CLK1 levels normally begin to rise ( Figure 2A ) . This conclusion is further supported by the observation that CLK1-dependent AS targets , such as those detected in HMMR and CENPE , are periodically expressed during cell cycle and peak during G2 and M phase ( Figure 4—figure supplement 1E ) . Taken together with the earlier results , these data support an important and multifaceted role for CLK1 in the control of cell cycle progression through its function in the global regulation of periodic AS . 10 . 7554/eLife . 10288 . 013Video 1 . Live-cell imaging of control HeLa cells stably expressing a GFP-H2B . Cells were synchronized by single-thymidine block and released and imaged at 10X magnification ( every ~15 min ) for 960 min . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 01310 . 7554/eLife . 10288 . 014Video 2 . Live-cell imaging of control HeLa cells stably expressing GFP-H2B . Cells were synchronized by single-thymidine block and released into 1 μM of TG003 and imaged at 10X magnification ( every ~15 min ) for 960 min . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 01410 . 7554/eLife . 10288 . 015Video 3 . Zoom of live-cell imaging of control HeLa cells . Data associated with Figure 4D . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 01510 . 7554/eLife . 10288 . 016Video 4 . Zoom of live-cell imaging of TG003-treated HeLa cells . Data associated with Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 01610 . 7554/eLife . 10288 . 017Video 5 . Zoom of live-cell imaging of TG003-treated HeLa cells . Data associated with Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10288 . 017 The importance of CLK1 for faithful progression through the cell cycle suggests it may play a role the control of cell proliferation in cancer . Supporting this , shRNA knockdown or chemical inhibition of CLK1 with TG003 or KHCB-19 in HeLa cells results in a near complete block in cell proliferation , as measured by anchorage-dependent and -independent colony formation assays ( Figure 4E and Figure 4—figure supplement 1F ) . As mentioned above , CLK1 is likely the primary target of inhibition in these experiments since RNAi of CLK1 recapitulates the phenotype seen with these chemical inhibitors . Using RNA-Seq data from the Cancer Genome Atlas ( TCGA ) ( Cancer Genome Atlas Network , 2013 ) we observe that CLK1 displays significantly higher expression in 72 kidney tumors compared to matched normal tissue samples ( Figure 4F , p=10-5 , Kolmogorov-Smirnov test ) . Consistently , most CLK1-controlled AS events that are altered in tumors have expected splicing changes ( Figure 4—figure supplement 2A ) . Furthermore , patients with tumors that have elevated CLK1 expression ( i . e . the upper quartile of all samples ) have significantly lower survival rates relative to other patients in the comparison group ( Figure 4G , p=0 . 007 ) . While there was also an increase in CLK2 , CLK3 and CLK4 mRNA expression in these tumors , CLK1 displayed the highest relative mRNA expression levels compared to CLK2 and CLK3 ( Figure 4—figure supplement 2B ) . Levels of the CLK4 , which is ~80% identical to CLK1 , did not correlate with survival differences despite its increased levels in tumors ( Figure 4—figure supplement 2C ) . Consistent with this , CLK1-regulated AS events , as defined by the RNA-Seq analysis in Figure 3 , were also altered across multiple tumor types , including breast , colon , lung and liver ( Figure 4H and Figure 4—figure supplement 2D ) . These data are further consistent with a multi-faceted role for CLK1 in regulating cell cycle progression , and also suggest that CLK1 contributes to increased cell proliferation in cancer , at least in part through its role in controlling periodic AS .
Previous studies have shown that splicing and the cell cycle are intimately connected processes . Indeed , cell cycle division ( CDC ) loci originally defined in S . cerevisiae , namely cdc5 and cdc40 , were subsequently shown to encode spliceosomal components ( Ben-Yehuda et al . , 2000; McDonald et al . , 1999 ) . Moreover , genome-wide RNAi screens for new AS regulators of apoptosis genes in human cells revealed that factors involved in cell-cycle control , in addition to RNA processing components , were among the most significantly enriched hits ( Moore et al . , 2010; Tejedor et al . , 2015 ) . An RNAi screen performed in Drosophila cells for genes required for cell-cycle progression identified numerous splicing components ( Björklund et al . , 2006 ) as well as a Drosophila ortholog of CLK kinases , Darkener of apricot Doa ( Bettencourt-Dias et al . , 2004 ) . In other studies , negative control of splicing during M phase was shown to be dependent on dephosphorylation of the SR family protein , SRSF10 ( Shin and Manley , 2002 ) , and the mitotic regulator aurora kinase A ( AURKA ) was shown to control the AS regulatory activity of SRSF1 ( Moore et al . , 2010 ) . Our study shows for the first time that AS patterns are subject to extensive periodic regulation , in part via a global control mechanism involving cell cycle fluctuations of the SR protein kinase CLK1 . At least one likely function of this periodic AS regulation is to control the timing of activation of AURKB ( Figure 1 ) , as well as of numerous other key cell cycle factors shown here to be subject to periodic AS . The definition of an extensive , periodically-regulated AS program in the present study thus opens the door to understanding the functions of an additional layer of regulation associated with cell cycle control and cancer . Since many RBPs are found to be periodically expressed ( Figure 4A ) , it is likely that other aspects of mRNA metabolism are also coordinated with cell cycle stages . For example , differential degradation could potentially contribute to the observed periodic fluctuation of splice isoforms during cell cycle . The degradation of mRNA is closely linked with alternative polyadenylation , which has emerged as a critical mechanism that controls mRNA translation and stability . Generally , shortened 3’ UTRs are found in rapidly dividing cells and more aggressive cancers ( Mayr and Bartel , 2009 ) . This study identified 94 cases periodic alternative poly-A site usage ( data not shown ) in genes known to regulate cell cycle and/or proliferation , including SON , CENPF and EPCAM ( Ahn et al . , 2011; Bomont et al . , 2005; Chaves-Perez et al . , 2013 ) . In addition , many mRNAs have recently been found to be translated in a cell cycle dependent fashion ( Aviner et al . , 2015; Maslon et al . , 2014; Stumpf et al . , 2013 ) . Interestingly , a fraction of periodically translated genes are also periodically spliced , including key regulators of cell cycle ( e . g . , AURKA , AURKB , TTBK1 and DICER1 ) . This observation is consistent with recent findings that the regulation of AS and translation may be coupled ( Sterne-Weiler et al . , 2013 ) . In summary , we have demonstrated that AS is subject to extensive temporal regulation during the cell cycle in a manner that appears to be highly integrated with orthogonal layers of cell cycle control . These results thus provide a new perspective on cell cycle regulation that should be taken into consideration when studying this fundamental biological process , both in the context of normal physiology and diseases including cancers .
HeLa ( a kind gift from J . Trejo ) , HEK 293T ( from ATCC CRL-3216 ) and A549 ( kind gift from W . Kim ) cells were maintained in DMEM ( Gibco ) medium supplemented with 10% FBS ( Gibco ) . All cells were cultured in humidified incubators with 5% CO2 . Cell cycle synchronization was adapted from the protocol of Whitfield et al . ( Whitfield et al . , 2002 ) ; ~750 , 000 log phase HeLa cells were plated in 15 cm dishes in complete media and allowed to attach for 16 hr , reaching <30% confluence . Cells were subsequently treated with 2 mM thymidine ( Sigma-Aldrich , St . Louis , MO ) for a total of 18 hr , washed 2 times with 1xPBS , and supplemented with fresh complete media for 10 hr . 2 mM thymidine was subsequently added for a second block of 18 hr and washed as described previously . Mitotic block was performed by double thymidine arrest ( as above ) and release in fresh media for 3 . 5 hr followed by addition of nocodazole 100 μM ( Sigma ) for 10 hr . G1 block was performed by serum starvation for 72 hr in DMEM containing 0 . 05% FBS . For RNA-Seq , cells ( both adherent and detached ) were harvested every 1 . 5 hr for 30 hr and frozen immediately for purification of total RNA . To block the activity of CLK1 , cells were treated with TG003 ( Sigma ) , KHCB-19 ( Tocris , Bristol , UK ) . To block activity of the proteasome cells were treated with MG132 ( Sigma ) . Drugs were-suspended in DMSO and added to growing cultures at the indicated concentrations and times . Cells were harvested with trypsin treatment , washed 2 times in cold 1xPBS and subsequently fixed in 80% ice cold ethanol for at least 4 hr . Cells were then washed twice with 1xPBS and suspended in propidium iodide/RNase staining buffer ( BD Pharmingen , cat # 550825 ) . Cells were analyzed by flow cytometry to count 10 , 000 cells that satisfied gating criteria . Data collected were analyzed using ModFit software to discern 2N ( G1 ) , S-phase , and 4N ( G2 and M ) composition . RNA-Seq reads were mapped to the human genome ( build hg19 ) using the MapSplice informatics tool with default parameters ( Wang et al . , 2010 ) . The mapped reads were further analyzed with Cufflinks to calculate the level of gene expression with FPKM ( Fragments Per Kilobase per Million mapped reads ) ( Trapnell et al . , 2010 ) . The levels of alternatively spliced isoforms were quantified with MISO ( Mixture-of-Isoforms ) probabilistic framework ( Katz et al . , 2010 ) using the annotated AS events for human hg19version 2 . The levels of alternatively spliced isoforms were also quantified with VAST-TOOLS using the event annotation as previously described ( Irimia et al . , 2014 ) . Each AS event was assigned a PSI or PIR value to represent the percent of transcripts with the exon spliced in , or the intron retained , respectively . For identification of periodic AS raw PSI/PIR values were normalized as:normalized ( Φns ) =Φns− ΦminsΦmaxs where s = 1 to 32 , 109 for all splicing events; n = 1 to 14 for the 14 samples; Φmin is the minimum and Φmax is the maximum PSI value among the 14 samples . To identify periodic AS events , normalized gene expression values ( normalized FPKM values as en ) for the well-known periodic gene , CCNB1 , CCNA2 , CCNB2 , and CENPE , were used as a starting point to subsequently add curves with broader or sharper peaks as well as shifted to left or right , resulting in 7 periodic expression curves that cover all the phases of cell cycle ( Figure 1—figure supplement 1A ) . We term these 'ideal seed curves' , which capture intermittent peak times and phase shifts that were not well represented within the initial known periodic genes . To identify genes with similar splicing patterns across the cell cycle , we computed the EuclideanDistanceED of each AS event s to the model seed curves m as follows:EDm , s=∑n=114|normalized ( enm ) −normalized ( Φns ) | where m = 1 to 7 for all model seed curves , s = 1 to 32 , 109 for all AS events . Based on the ranking of distance , a similar cutoff of ED≤2 . 75 was set as a minimum requirement for periodic AS . Lastly , we calculated a false discovery rate ( FDR ) by shuffling PSI values across the 14 time points 10 , 000 times and calculating how often a random shuffle had a better periodic score than the true periodic score for that event . A maximum FDR of 2 . 5% was required for a splicing event to be periodic . Heat maps , hierarchical clustering , and Pearson correlations were generated using GENE-E ( www . broadinstitute . org/cancer/software/GENE-E/ ) . All heat maps shown are row-normalized for presentation purposes . Spearman’s rank correlation with average linkage was used for clustering . DAVID ( http://david . abcc . ncifcrf . gov/gene2gene . jsp ) was used for all gene ontology enrichment; terms shown are for biological process ( GOTERM_BP_FAT ) . To test for significance in overlap analysis , overlapping genes in two data sets ( i . e . TG003-treatment and periodic AS ) and a background set of only co-detected events was used ( i . e . genes detected in both experiments ) . Significance of overlapping gene sets was assessed using the hyper-geometric test . For overlap and correlation analysis between VAST-TOOLS and MISO we used two MISO AS event annotations ( HG18 and HG19 ) , due to differences in the input annotation files for these two pipelines . To identify the 4 , 343 overlapping exons in the heatmap , we used MISO hg19v1 annotations and MISO hg18 annotations , and the VAST-TOOLS annotations . Student’s t-test was used to measure significant in cell cycle defects ( multi-nucleation and flow-cytometry ) as well as semi-quantitative RT-PCR assessment of splice variants . To identify AS events blocked by TG003 or KHCB19 , a Student’s t-test statistic was used . If a change between G1/S ( t=0 ) and G2 ( t=6 ) was significant , but not significant in the presence of inhibitors we consider that event to be blocked . For over-representation of periodic introns , we performed Fisher’s exact test in a 2x2 contingency table as compared to skipped exons . For differences in expression of CLK1 mRNAs in kidney cancers a Kolmogorov-Smirnov test was performed . For survival differences , the survdiff function in the R survival package was used ( as discussed in methods below ) . The expression constructs were generated by cloning the cDNA of CLK1 into pCDNA3 ( for transient expression ) or pCDH ( for stable transfection ) backbones with different epitope tags ( HA or Flag ) at the N- or C-terminus . The Myc-His-Ubiquitin expression vector is a gift from Dr . Gary Johnson’s lab , and the Histone H2B-GFP expression vector is gift from Dr . Angelique Whitehurst’s lab . Plasmid transfections were performed using Lipofectamine 2000 ( Invitrogen ) according to the manufacturer’s protocol . The lentiviral vectors of shRNAs were obtained from Addgene in pLKO . 1 TRC cloning plasmid through UNC core facility as part of mammalian gene knockdown consortium . Lentiviral infections were performed according to the manufacturer's instruction from System Biosciences ( SBI ) . Cells transfected with shRNA constructs or treated with TG003 were lysed and total RNA was extracted using the Trizol method ( Life Technologies ) . Purified RNA was treated with 1U of RNAse-free DNAase ( Promega ) for 1 hr at 37°C and reverse transcribed using random hexamer cDNA preparation kit ( Applied Biosystem ) . One-tenth of the RT product was used as the template for PCR amplification ( 25 cycles of amplification , with a trace amount of Cy5-dCTP in addition to non-fluorescent dNTPs ) using gene specific primers listed in Supplementary file 4 . The resulting gels were scanned with a Typhoon 8600 Imager ( GE Healthcare ) , and analyzed with ImageQuant 5 . 2 software ( Molecular Dynamics/GE Healthcare ) . Real time PCR was carried out using the SYBR Green kit ( Invitrogen ) and GAPDH as an internal control . Proteins were extracted in lysis buffer ( CHAPS 1% w/v , 150 mM NaCl , 50 mM MgCl2 with protease inhibitor ) , resolved by SDS-PAGE and transferred onto PVDF membrane . For immunoprecipitaion experiments to detect ubiquitination , cells were co-transfected with Flag-CLK1 and myc-ubiquitin constructs as above . 36 hr later , TG003 ( 20 μM ) was added for 18 hr . 4 hr prior to harvesting , 10 μM of MG132 was added to the media . Proteins were extracted in lysis buffer as above with the addition of NEM . Incubation with EZ-View FLAG Beads ( Sigma ) was performed for 2 hr at 4°C . Samples were extensively washed according to the manufacturer’s protocol and subjected to immunoblotting . Antibodies and dilutions are listed in Supplementary file 4 . For immunofluorescence microscopy , cells were plated on glass coverslips coated with poly-L-Lysine . Cells were then washed twice with 1xPBS , fixed with 4% formaldehyde ( Sigma ) , permeabilized with 0 . 05% Triton X-100 ( Promega ) and blocked with 3% BSA ( Fisher ) ; all dilutions were made in 1XPBS . For live cell imaging , HeLa cells transduced with Histone H2B-GFP was stably selected as described previously ( Cappell et al . , 2010 ) . Cells were plated in a 6-well format and treated with 2 mM thymidine for 24 hr , subsequently washed and released in fresh complete medium with or without TG003 ( 20 µM ) . Cells were imaged using the BD Pathway Microscope with a 10X objective . HeLa cells stably producing shRNAs targeting CLK1 or control shRNAs were plated at low density ( 1000 cells/6 cm2 ) in standard culture medium and allowed to proliferate for 9 days . Cells were then fixed and stained with crystal violet at room temperature . The dried plates were used for estimations of colony diameter and number . RNA-Seq data from the The Cancer Genome Atlas ( Ciriello et al . , 2013 ) was processed as previously described ( Tsai et al . , 2015 ) . Briefly , for mRNA expression , RSEM expression values for the indicated genes were analyzed in 79 paired KIRC ( tumor and normal ) samples and the paired ks-test was used to test significance . For alternative splicing analysis data from BRCA: Breast invasive carcinoma , COAD: Colorectal adenocarcinoma , KIRC: Kidney renal clear cell carcinoma , LUAD: Lung adenocarcinoma , LUSC: Lung squamous cell carcinoma , LIHC: liver hepatocellular carcinoma were analyzed through the MISO pipeline as described above ( Tsai et al . , 2015 ) . Relapse-free survival was analyzed using Kaplan Meier plots . All plots and statistical analyses ( survdiff ) were generated using the R package version 3 . 1 . 1 survival function . | Mitosis is a key step in the normal life cycle of a cell , during which one cell divides into two new cells . As a cell progresses through the cell cycle , it must carefully regulate its gene activity to switch particular genes on or off at specific moments . When a gene is activated its sequence is first copied into a temporary molecule called a transcript . These transcripts are then edited to form templates to build proteins . One way that a transcript can be edited is via a process called alternative splicing , in which different pieces of the transcript are cut and pasted together to form different versions of the final template . This allows different instructions to be obtained from a single gene , introducing an added layer of biological complexity . However , the role of alternative splicing in the timing of key events of the cell life cycle is not well understood . Dominguez et al . have now looked for the genes that undergo alternative splicing during the cell cycle . The sequences of gene transcripts produced within human cells were collected while the cells went through two rounds of division . This approach revealed that around 1 , 300 genes are spliced in different ways at different stages of each cell cycle . Many of these genes were known to play roles in controlling the cell’s life cycle , but few of the genes showed large changes in the amount of total transcript that is generated over time . Dominguez et al . also showed that an enzyme called CLK1 influences about half of the 1 , 300 periodically spliced genes during the cell cycle . The production of CLK1 is itself carefully controlled throughout the cell cycle , and the enzyme’s activity prevents its own overproduction . Further experiments showed that blocking CLK1’s activity while a cell is replicating its DNA halts the cell cycle , but blocking this enzyme’s activity after the cell had replicated its DNA did not . Given this pivotal role in the cell cycle , Dominguez et al . also examined the role of CLK1 in cancer cells and found that high levels of CLK1 in tumours were linked to lower survival rates . These findings indicate that CLK1 warrants further investigation , particularly in relation to its role in cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2016 | An extensive program of periodic alternative splicing linked to cell cycle progression |
At the origin of multicellularity , cells may have evolved aggregation in response to predation , for functional specialisation or to allow large-scale integration of environmental cues . These group-level properties emerged from the interactions between cells in a group , and determined the selection pressures experienced by these cells . We investigate the evolution of multicellularity with an evolutionary model where cells search for resources by chemotaxis in a shallow , noisy gradient . Cells can evolve their adhesion to others in a periodically changing environment , where a cell’s fitness solely depends on its distance from the gradient source . We show that multicellular aggregates evolve because they perform chemotaxis more efficiently than single cells . Only when the environment changes too frequently , a unicellular state evolves which relies on cell dispersal . Both strategies prevent the invasion of the other through interference competition , creating evolutionary bi-stability . Therefore , collective behaviour can be an emergent selective driver for undifferentiated multicellularity .
The evolution of multicellularity is a major transition in individuality , from autonomously replicating cells to groups of interdependent cells forming a higher-level of organisation ( Buss , 2014; Smith and Szathmary , 1995 ) . It has evolved independently several times across the tree of life ( Grosberg and Strathmann , 2007; Parfrey and Lahr , 2013 ) . Comparative genomics suggests ( Knoll , 2011 ) , and experimental evolution confirms ( Boraas et al . , 1998; Ratcliff et al . , 2012 ) that the increase of cell–cell adhesion drives the early evolution of ( undifferentiated ) multicellularity . Increased cell adhesion may be temporally limited and/or may be triggered by environmental changes ( e . g . in Dictyostelids and Myxobacteria [Du et al . , 2015; Kaiser et al . , 1979] ) . Moreover , multicellular organisation may come about either by aggregation of genetically distinct cells or by incomplete separation after cell division ( King , 2004; Du et al . , 2015 ) . The genetic toolkit and the cellular components that allow for multicellularity - including adhesion proteins - pre-date multicellular species and are found in their unicellular relatives ( Rokas , 2008; Prochnik et al . , 2010; Du et al . , 2015; Richter et al . , 2018 ) . Aggregates of cells can organise themselves by exploiting these old components in the new multicellular context , allowing them to perform novel functions ( or to perform old functions in novel ways ) that may confer some competitive advantage over single cells . Greater complexity can later evolve by coordinating the division of tasks between different cell lineages of the same organism ( e . g . in the soma-germline division of labour ) , giving rise to embryonic development . Nevertheless , the properties of early multicellular organisms are defined by self-organised aggregate cell dynamics , and the space of possible multicellular outcomes and emergent functions resulting from such self-organisation seems large – even with limited differential adhesion and signalling between cells . However , the evolution of emergent functions as a consequence of adhesion-mediated self-organisation has received little attention to date . Mathematical models can define under which conditions multicellularity evolves , in terms of fitness for individual cells vs . the group , or in terms of the resulting spatial and temporal organisation . The formation of early multicellular groups has been studied in the context of the evolution of cooperation: by incorporating game theoretical interactions and transient compartimentalisation ( Garcia et al . , 2014 ) or the possibility of differential assortment ( Joshi et al . , 2017 ) , it was found that adhering groups of cooperating individuals evolve . Alternatively , reproductive trade-offs can give rise to division of labour ( Solari et al . , 2013 ) and lead to the formation of a higher-level proto-organism capable of self-regeneration in a structured environment ( Duran-Nebreda et al . , 2016 ) . A plethora of multicellular life-cycles can emerge by simple considerations about the ecology of the uni-cellular ancestor and the fitness benefit that cells acquire by being in groups ( Staps et al . , 2019 ) . Once multicellular clusters are established , the spatial organisation of their composing cells can play an important role in determining group-level reproduction - possibly leading to the evolution of cell-death ( Libby et al . , 2014 ) or different cell shapes ( Jacobeen et al . , 2018 ) , and to specific modes of fragmentation of the aggregate ( Pichugin et al . , 2017; Gao et al . , 2019 ) that increase overall population growth . In these models , multicellularity is either presupposed or its selective pressure is predetermined by social dynamics , by directly increasing fitness of cells in aggregates or by adverse environmental conditions that enforce strong trade-offs . Here we investigate the origin of this selective pressure , motivated by the idea that multicellular groups emerge as a byproduct of cell self-organisation and cell-environment interactions , and subsequently alter the evolution of their composing cells . We expect that a selective pressure to aggregate can arise from the emergent functions of the multicellular group , without requiring explicit selective advantages and disadvantages for cells in a group . We therefore present a computational model of an evolving population of cells where fitness is based solely on how adequately a cell responds to a spatially and temporally heterogeneous environment , regardless of whether they belong to an aggregate . In this study , we draw inspiration from collective movement of groups of cells , such as the aggregate phase of the slime mould Dictyostelium discoideum ( Schaap , 2011 ) , other simple multicellular organisms ( Kaiser , 2003; Schaap , 2011; Smith et al . , 2019 ) and many processes within complex multicellular organisms , for example , embryogenesis , tissue repair and cancer ( Weijer , 2009; Friedl and Gilmour , 2009 ) . Previous models have shown how cell collectives are able to integrate noisy information from the environment , for instance when moving up a shallow chemoattractant gradient . ( Marée et al . , 1999; Szabó et al . , 2006; Kabla , 2012; Szabó et al . , 2010; Camley and Rappel , 2017; George et al . , 2017; Camley , 2018; Varennes et al . , 2017 ) . We use the Cellular Potts Model ( Graner and Glazier , 1992 ) ( CPM ) to study collective cell movement as an emergent driver of multicellularity during evolution . The CPM formalism is a spatially extended , mesoscopic description of cells which explicitly accounts for cell shape and size , and allows for a straightforward implementation various cellular processes within complex and potentially self-organised environments . We include four key elements: cells are placed in a seasonally changing environment that periodically introduces new resources at different locations , they can perform chemotaxis by sensing a chemoattractant produced by these resources , they reproduce depending on their proximity to resources and they can evolve their adhesion to other cells . Because the gradient generated by the resources is noisy and shallow , we find that individual cells follow the chemotactic signal very inefficiently . Instead , cells that adhere to each other within groups transfer information about the gradient in a self-organised manner , allowing for efficient chemotaxis in our model . We show that for longer seasons , this emergent property of cell groups is sufficient to select for high levels of adhesion and multicellularity , despite the fact that fitness is only defined at the cell level .
To explore the evolutionary dynamics of a population of cells , we seasonally change the location of the resources , and therewith the direction of the gradient , every τs MCS ( Figure 2 ) . Longer seasons ( larger values of τs ) correspond to more persistent resources . During each season ( i . e . one period of τs MCS ) cells move due to chemotaxis and persistent migration . Depending on the ligands and receptors expressed on the cell surfaces , they may either adhere to one another or disperse from one another ( Figure 2a ) . At the end of the season , cells are given a chance to divide , followed by a culling phase to keep the number of cells constant . To reflect the assumption that more nutrients are present at higher concentrations of the signal , the division probability is inversely proportional to the distance of the cell to the gradient peak and cells very close to the gradient peak may divide multiple times . Cells divide along their short axis to create two daughter cells ( after Hogeweg , 2000 ) , after which we let cells regrow to target size for 5 MCS . The daughter cells inherit mutated copies of the ligand and receptor , so that their adhesive properties can change with respect to the parent . This allows cells to evolve their adhesion strength . Cell size AT , strength of chemotaxis μχ and migration persistence μp do not evolve . After cell division , the population is brought back to N cells by randomly culling cells , at which point the new season begins ( Figure 2b ) . Note that we do not include cell dispersal after replication , therefore related cells remain close at the beginning of the new season . Simulations last 400 seasons ( i . e . 400×τs MCS ) , which is sufficient to reach evolutionary steady state under all conditions . We do not select for multicellularity directly: the fitness function rewards cells for their proximity to resources , and we do not explicitly incorporate a fitness advantage or disadvantage for the multicellular state . Therefore , multicellular clusters ( Figure 2a , c , d ) can arise only because they perform an emergent task that single cells cannot perform . See Table 1 for parameter values , and Materials and methods Section for the details of the model and parametrisation . We first assessed how well groups of cells with different adhesion strengths could reach the source of the chemotactic signal . We placed a connected cluster of cells on one side of the lattice , opposite to the location of the gradient peak . We then recorded their travel distance over a fixed amount of time and compare it to the travel distance of single cells ( i . e . from simulations with only one cell ) , by measuring both the position of the centre of mass of the group ( Figure 3a ) and the position of the cell closest to the peak of the gradient ( Figure 3b ) . Single cells perform chemotaxis inefficiently ( Video 1 ) , whereas a group of adhering cells migrates up the same gradient more accurately ( Figure 3a , γ>0 ) : the centre of mass of this group takes much less time than single cells do to reach the peak of the gradient ( Video 2 ) . Groups of cells can also perform collective chemotaxis when they do not adhere , and when they do not have a preference for medium or cells , although with lower efficiency in both cases ( Figure 3a , respectively γ<0 and γ=0 ) . Chemotaxis is inefficient , because these cells tend to lose contact from one another ( Video 3 ) and once isolated they behave like those from simulations with one cell ( Figure 3a , b ‘one cell’ ) . Single cells also show large variance between different simulations ( Figure 3b ) . While cell clusters perform chemotaxis efficiently only when cells adhere , the speed of the cell closest to the peak of the gradient is roughly the same regardless of adhesion strength ( Figure 3b ) . Thus , in a non-adhering population some cells reach the peak of the gradient almost as quickly as an adhering cluster does . Adhering cells have large chemotactic persistence - as shown by the super-linear shape of the Mean Square Displacement ( MSD ) plot ( Figure 3c , γ=6 ) and by a diffusive exponent consistently larger than 1 ( Figure 3d; the diffusive exponent is obtained as the derivative of the log-log transformed MSD/time curve , see Appendix 1 . 1 ) . Instead , the MSD of a single cell ( Figure 3c , one cell ) is approximately linear and its diffusive exponent tends to 1 , indicating that cells’ movement is much more dominated by diffusion . Interestingly , there is no difference in the instantaneous speed of cells when they are in a cluster or when they are alone ( Figure 3e ) , so the higher rate of displacement of a group of adhering cells is only due to larger persistence in the direction of motion . Figure 4 shows the movement of a cluster of strongly adhering cells ( γ=6 ) compared to the movement of a single cell , over the typical setup of the simulation system . Although the cluster moves straight towards the source of the gradient , individual cells follow noisy trajectories . A possible explanation for collective chemotaxis is that a cluster averages individual cells’ polarisation , leading to a linear relationship between the accuracy of chemotaxis and the number of cells in the cluster ( Varennes et al . , 2017 ) . Instead , we found that cluster speed saturates quickly with the number of cells , at a smaller speed than that of individual cells ( cf . Appendix 1 . 2 with Figure 3e ) . We conclude that individual contributions to cluster chemotaxis are not simply averaged . Therefore , we look at how cells self-organise to understand how collective chemotaxis comes about . Through persistent migration , a cell pushes other cells within an adhering cluster , and is pushed by them . The resulting forces are resolved when cells align and form streams within the cluster ( see Video 4 ) . These streams are persistent over a much longer time scale than a cell’s persistence τp=50 MCS ( since the video frame rate is 50 MCS and streams are visible over multiple frames ) . Through streaming , these small clusters generate extensions , retractions and rotations ( Video 4 ) , so that the entire cluster visually resembles a single amoeboid cell ( Video 2 ) . This behaviour is not influenced by the presence of the chemotactic signal , since the flow field is identical when the chemotactic signal is removed ( Appendix 1 . 3 ) . Thus , the effect of persistent migration is to align the direction of motion of the cells in a cluster . This in turn speeds up collective chemotaxis , as cell streams preferentially align towards the direction of the gradient , although aligning is not strictly required for chemotaxis ( Appendix 1 . 4 ) . Clusters perform chemotaxis faster than individual cells over a large range of values for persistent migration strength μp and chemotactic strength μχ ( Appendix 1 . 5 and Appendix 1 . 6 ) , with larger μp increasing collective chemotaxis speed ( and to a lesser extent individual chemotaxis speed ) more than μχ . Because larger cells perceive a larger area of the chemotactic signal , chemotactic migration improves with cell size ( Appendix 1 . 7 ) . We calculated the deviation of each individual cell’s measurement of the gradient as the angle θ ( X→ , χ→ ) between the true direction of the gradient X→ and the direction of the gradient locally measured by the cells χ→ ( so that θ ( X→ , χ→ ) =0 is a perfect measure ) . We found that the measurements of individual cells deviate significantly from the true direction of the gradient ( Figure 3f ) . Despite this , they are carried in the right direction by the other cells . To assess how cells in a cluster alter each others’ ( short-timescale ) trajectories we extracted the straight segments from the cell tracks and assessed both the length of these segments and their orientation with respect to the gradient source ( Appendix 1 . 8 ) . We find that cells in a cluster tend to migrate for longer in straight lines , and that these straight lines are also more likely to be oriented towards the source of the gradient ( Figure 3g ) . For single cells , there is no such bias . In conclusion , cluster organisation emerges from cells altering each others’ paths by exerting pushing and pulling forces through their persistent migration , which in turn results in efficient collective chemotaxis . The emergence of reliable chemotactic behaviour in adhering cell clusters suggests an evolutionary path to multicellularity: a population of cells may aggregate if collective chemotaxis allows cells to find resources more reliably . While cells could improve their ability to sense the gradient individually by becoming bigger , there are many factors that restrict cell size , such as the complexity of the metabolism and cellular mechanisms such as cell division ( Björklund and Marguerat , 2017; Marshall et al . , 2012 ) . We therefore assume that cell size is fixed , and we let cells evolve adhesion - that is , the receptor and ligands expressed by the cells - in response to a seasonally changing environment , where the gradient is generated by a volatile resource that periodically changes position . Cells closer to the peak of the gradient have a higher chance to reproduce at the end of the season , and related cells remain close to each other at the beginning of the new season ( there is no cell dispersal phase , see also model setup and Materials and methods ) . The receptors and ligands of the initial population are chosen such that cells neither adhere to one another nor disperse from one another ( γ=0 ) . When the season lasts τ=100×103 MCS , the average adhesion between cells readily increases after only few generations ( Figure 5a ) : Jcell , cell decreases and Jcell , medium increases ( see also Video 5 and Figure 2 for snapshots ) . At evolutionary steady state , all cells adhere strongly and with roughly the same energy to one another ( Appendix 2 . 1 ) . Figure 5b shows that two evolutionary steady states are possible , depending on the duration of the season τs . For τs<20×103 MCS , cells evolve to become unicellular , as cell–cell interactions are characterised by strong repulsion ( γ<0 ) . Figure 5c suggests that by selecting for γ<0 cells disperse efficiently throughout the grid . Although non-adhering cells follow the chemotactic signal only weakly , the spreading over the course of multiple seasons ensures that at least some cells end up close to the source of the gradient at the end of the season ( Video 6 ) . In contrast , a cluster of adhering cells is at disadvantage when seasons are short because it does not have enough time to reach the source of the chemotactic signal . Over the course of multiple seasons , an adhering cluster ends up in the centre of the lattice ( Video 7 ) and all its composing cells have the same ( low ) fitness . Furthermore , the connectedness of a cluster of adhering cells is locally disrupted when excess cells are culled between seasons ( Figure 2b ) , which briefly reduces the efficiency of collective migration . Because this phase is short-lived - cells reconnect within 2000 MCS - we expect that culling plays a minor role in the evolutionary outcome of the system . For τs>40×103 MCS , cells evolve to adhere to one another , i . e . γ>0 ( see Figure 5c for a snapshot ) . When seasons are sufficiently long , clusters of adhering cells have enough time to reach the source of the gradient . At this point , the fitness of cells within a cluster outweighs that of non-adhering cells , because clustering increases the chances of reaching the peak of the gradient . Finally , for intermediate season duration , 20×103≤τs≤40×103 MCS , both repulsion and adhesion are evolutionary ( meta ) stable strategies , and the outcome of the simulation depends on the initial value of γ ( for τs=20×103 MCS , the steady state with γ>0 is very weakly stable ) . Because different values for migration parameters affect collective chemotaxis speed , we checked that the evolution of multicellularity is qualitatively robust to changes in the values of persistent migration strength μp and chemotactic strength μχ ( respectively in Appendix 2 . 2 and Appendix 2 . 3 ) . Results are also robust to changes in gradient shape ( assuming that resources are located over an entire side of the lattice , we tested a gradient with straight isoclines in Appendix 2 . 4 ) and to steeper , noiseless gradients ( Appendix 2 . 5 ) . Furthermore , the evolution of multicellularity does not depend on the precise mechanism for collective chemotaxis . To show this , we relax the assumption that individual cells sense the gradient by implementing a recently proposed mechanism of emergent collective chemotaxis that relies only on concentration sensing ( Camley et al . , 2016 ) . Following Camley et al . , 2016 , we assume that cell polarisation is inhibited at the sites of cell–cell contact ( a phenomenon called contact inhibition of locomotion , see Mayor and Carmona-Fontaine , 2010 for a review ) , and that the magnitude of their polarisation is proportional to the concentration - not the gradient - of the chemoattractant . In Appendix 3 . 1 , we show that results are robust to this modification of the chemotaxis mechanism . We next investigated what causes the evolutionary bi-stability in adhesion strategies for season duration 20×103≤τs≤40×103 MCS . We performed competition experiments between two populations of cells , one adhering ( γ=6 ) and one non-adhering ( γ=-4 ) , to determine whether a strategy can invade in a population of cells using the other strategy . We simulated non-adhering mutants invading a resident population of adhering cells by placing a large cluster of adhering cells in front of a small group of non-adhering ones ( Figure 6a ) , and conversely , a small cluster of adhering cells invading a large group of non-adhering cells ( Figure 6b ) . This initial configuration is analogous to the beginning of a season in the evolutionary experiments , as mutants are in small numbers and furthest away from the new peak because they are likely born from cells that replicate most , that is , those closest to the previous location of the peak . In both cases , after 30×103 MCS , the resident population physically excludes the invading one from the path to resources , and thus the distance travelled by the invading population is limited . This shows that the adhesion energy of the resident population ( whether cells adhere or not ) determines the outcome of the invasion ( for the values of τs where we find evolutionary bistability ) . We also considered a scenario where a whole population - rather than few mutants - invades another with the opposite strategy . We studied the spatial competition dynamics of two clusters of equal size ( N=100 cells ) when adhering cells are positioned in front of the non-adhering ones ( Figure 6c ) , and when the position of the two clusters is swapped ( Figure 6d ) . The distance to the peak after 30 × 103 MCS of a cluster of adhering cells is larger ( i . e . their fitness is smaller ) if they are hindered by a population of non-adhering cells in front of them . Taken together , these results show that there is interference competition ( i . e . direct competition due to displacement ) between populations of cells with different strategies . In the evolutionary experiments , mutants with a slightly different strategy are generated during reproduction at the end of each season and interference competition continually prevents their successful invasion for intermediate season duration . This explains why the two strategies are meta-stable . This result may also provide a simple explanation for the fact that many unicellular organisms do not evolve multicellularity despite possessing the necessary adhesion proteins . Moreover , evolutionary bi-stability protects the multicellular strategy from evolutionary reversal to unicellularity over a large range of environmental conditions . So far , we showed that the evolutionary benefit of uni- or multi-cellular strategies is indirect , as it is mediated by the fittest form of self-organisation for a given season duration . For simplicity , we did not incorporate any cost to evolving multicellularity . However , evolving multicellular organisms may incur fitness costs that are not present at the unicellular level ( Rebolleda-Gómez and Travisano , 2018; Rainey and Rainey , 2003; Ratcliff et al . , 2012; Yokota and Sterner , 2011; Kapsetaki and West , 2019 ) . We incorporated costs in our system by assuming that cells spend energy to maintain their bonds with other cells , with a cost cm ( per unit of cell boundary , per MCS ) . This metabolic cost accumulates over time when cells are in contact with one another , and translates into a fitness penalty at the end of the season for cells that spent more time in contact with others . Costs range from cm=0 , the cost-free model presented so far , to cm=1 ( the maximum cost ) which zeroes the fitness of a cell that spent the entire season completely surrounded by other cells . Multicellularity evolved for sufficiently long seasons when costs were not too high ( cm≤0 . 5 ) , with larger costs shifting the transition to multicellularity to longer seasons , while only the uni-cellular strategy evolved when costs were high ( cm=0 . 75 , for the season duration we tested; Appendix 4 . 1 ) .
We demonstrated that undifferentiated multicellularity can evolve in a cell-based model as a byproduct of an emergent collective integration of spatial cues . Previous computational models have shown that multicellularity can be selected by reducing the death rate of cells in a cluster ( Staps et al . , 2019; Pichugin et al . , 2017 ) , through social interaction ( Garcia and De Monte , 2013; Joshi et al . , 2017 ) , by incorporating trade-offs between fitness and functional specialisation ( Ispolatov et al . , 2012 ) , or by allowing cells to exclude non-cooperating cells ( Pfeiffer and Bonhoeffer , 2003 ) . In these studies , direct selection for forming groups is incorporated by conferring higher fitness to the members of a cluster . Earlier work found that multicellular structuring can emerge without direct selection when cells are destabilised by their internal molecular dynamics ( e . g . the cell cycle ) ( Furusawa and Kaneko , 2002 ) , or because of a toxic external environment ( Duran-Nebreda et al . , 2016 ) . In both cases , cell differentiation stabilises cell growth and arises as a consequence of physiological or metabolic trade-offs . With our model setup , we show that division of labour - although important - is not a strict requirement for emergent aggregation . Nevertheless , our work bears some similarity with these models because we do not explicitly incorporate a fitness benefit for being in a group: selection acts on individual cells only on the basis of how close they are to the source of the gradient , regardless of migration strategy . Thus the fitness function does not dictate which evolutionary strategy , that is , uni- or multi-cellularity , should be followed . A limitation of the current model is that cells have a narrow set of possibilities for adapting to the environment , as the only mutable traits are their ligands and receptors . Therefore , their adaptation to the environment is solely mediated by their adhesion to one another and selection for multicellularity can only occur because adhering clusters always perform chemotaxis better than individual cells . Despite the advantage of clusters over individuals , an alternative strategy can evolve that does not rely on collective behaviour . This uni-cellular strategy evolves because non-adhering cells disperse throughout the field over multiple seasons . By chance - and aided by inefficient chemotaxis - some cells will be located near the peak of the gradient at the end of each season . When seasons change rapidly , a multicellular cluster does not have the time to reach the peak of the gradient . It is therefore at disadvantage over cells evolving a unicellular strategy . This further illustrates that the selection pressure to become multicellular emerges from the structure of the environment in our model , rather than being an explicit part of the fitness function . Whichever evolutionary strategy maximises fitness , be it multi- or uni- cellularity , will evolve within the ( limited ) complexity of the model . A second limitation of the model is that resources are modelled only implicitly - through the chemoattractant gradient they generate and through season duration , i . e . how long they persist . The precise seasonality of these resources might be realistic if resources are deposited in the system by periodic phenomena ( e . g . tides , or daily and yearly cycles ) , whereas other types of resources might be more stochastic ( such as preys ) . However , if the stochasticity of resources is not too extreme , we expect that evolution converges to the average resource duration . In many ways , the evolution of multicellularity can be compared to the evolution of collective dynamics . Previous studies on the evolution of herding behaviour showed that aggregating strategies can also evolve in response to highly clumped food even though the pack explores the space slowly and inefficiently before finding food ( Wood and Ackland , 2007 ) . When gradient sensing and social behaviour are both costly , a combination of strategies evolves in response to selection for distance travelled ( Guttal and Couzin , 2010 ) . Some individuals pay the cost for actively sensing the gradient , while others invest in social behaviour to move towards others and align their direction of motion with them , leading to the formation of migrating herds ( Guttal and Couzin , 2010 ) . These models of collective migration represent individuals as active particles , which is similar to the behaviour of our cells . However , group movement requires an explicit rule for alignment , whereas in our model it emerges naturally from interactions between deformable cells . Modelling cells with an explicit shape and size ( including both CPM and , we expect , self-propelled particles ) allows for spatial self-organisation and can generate interesting ecological dynamics , such as interference competition between the unicellular and multicellular search strategies . The ensuing evolutionary bi-stability stabilises unicellularity despite these cells possessing the surface protein toolkit to adhere to each other , and prevents multicellular organisation from evolutionary reversal into single cells ( over a range of environmental conditions ) . The ‘automatic’ outcome of spatial self-organisation provides an initial , non-genetic robustness , which can be further stabilised by later adaptations ( Libby et al . , 2016 ) . In our model , cells retain their spatial distribution between seasons . This reinforces spatial self-organisation , and consequently bistability , because genetically similar cells remain close to one another . However , we expect bistability also if cells were dispersed between seasons: few adhering cells scattered in a cloud of non-adhering ones would not be likely to meet ( and collectively chemotax ) if seasons are short . In contrast , a large number of adhering cells would meet frequently after scattering and thus displace non-adhering cells in their march towards the peak of the gradient . This suggests that the two strategies are not mutually invadable over some intermediate season length , hence bistability . The driver for the evolution of adhesion in our model is collective chemotaxis . This is reminiscent of the aggregate phase of the life cycle of Dictyostelium discoideum ( Schaap , 2011 ) , in that a cluster of cells moves directionally as a unit following light or temperature , while individual cells inefficiently identify the correct direction of motion ( Miura and Siegert , 2000 ) . There are some important differences between our model and D . discoideum , however . Information about the direction of the gradient is transmitted mechanically within cell clusters in our model . In D . discoideum photo- and thermo-taxis are coordinated by waves of cAMP secretion that travel through the slug . The lack of extra chemical cues to organise movement within a cell cluster in our model makes for a simpler scenario without large-scale transmission of information throughout the aggregate . Nevertheless , computational modelling has shown that long-range chemical signalling coupled to cells’ differential adhesion suffices to reproduce D . discoideum’s migration ( Marée et al . , 1999; Marée and Hogeweg , 2001 ) . Another important difference between our model and D . discoideum is the absence of dispersal at the end of the life cycle in our model . In D . discoideum , the slug transforms into a fruiting body at the top of a stalk of terminally differentiated cells . Extending the current model with the evolution of dispersal would enrich our understanding of D . discoideum evolution towards partial multicellularity . Our model of collective movement is an example of the ‘many wrongs’ principle ( Simons , 2004 ) : the direction error of each cell is corrected by the interactions with the other cells in the cluster . However , in our model there is no explicit mechanism for transferring gradient information between cells . Therefore our results differ from previous work on rigid clusters of cells , where cells’ polarisation towards the perceived gradient translates linearly and instantaneously to cluster movements ( Camley et al . , 2016; Varennes et al . , 2017 ) . In models where cells readily exchange neighbours , simple rules for cell adhesion and migration led to self-organisation of cells into highly persistent , migrating tissue with emergent global polarity ( Smeets et al . , 2016 ) ( earlier observed also at large cell density without adhesion rules [Beltman et al . , 2007; Szabó et al . , 2006] ) . Similarly , in our model , cells convey gradient information through such emergent collective streaming , which becomes biased towards the ( weak ) chemotactic signal . However , we expect that the evolutionary results described here are independent of the particular cell model choice , or the mechanism for chemotaxis provided that cells were able to polarise or move also in the absence of other cells . Indeed , we found similar results with an alternative model of collective chemotaxis ( Camley et al . , 2016 ) in which individual cells do not sense the gradient . We opted for a computational cell-based model - the Cellular Potts Model - because it allowed us to explore the spatial interactions of cells , and because it enabled straightforward implementation of the evolvable receptor-ligand system . The visual nature of our results may guide the future development of analytical approaches to generalize the results of this work . For instance , analytical work may provide a more detailed explanation of the ‘many wrongs’ principle for a cell cluster in which cells are highly motile and change their neighbours often , in which case positional information is transmitted by pulling and pushing on each other . Moreover , the simplicity of our model setup makes our results easily testable in vitro . The importance of a bottom-up approach to study the evolution of multicellularity has been repeatedly emphasised ( van Gestel and Tarnita , 2017; De Monte and Rainey , 2014 ) , and a broader understanding of cells self-organisation and evolution may have applications to clinically relevant multiscale evolutionary problems , such as the evolution of collective metastatic migration of cancer cells ( Coffey , 1998; Stuelten et al . , 2018; Disanza et al . , 2019; Lacina et al . , 2019 ) . Our work highlights that the properties of single cells emergently give rise to novel properties of cell clusters . These novel properties - in a downward causative direction - generate the selection pressure to form the first undifferentiated multicellular groups .
The model is a hybrid Cellular Potts Model implemented with the Tissue Simulation Toolkit ( Daub and Merks , 2015 ) . A population of N cells exists on a regular square lattice Λ1⊂ℤ2 . The chemotactic signal is located on a second plane Λ2 , of the same size and spacing as Λ1 . A cell c consists of the set of ( usually connected ) lattice sites x→∈Λ1 to which the same spin s is assigned , that is , c ( s ) ={x→∈Λ1∣σ ( x→ ) =s} . The spin value is a non-negative integer , it is unique and positive for each cell , and it is used as the cell identifier . The medium is assigned spin σ=0 . Cell movement arises from deformation of its boundaries through stochastic fluctuations . These fluctuations minimise a cell’s energy , whose terms correspond to biophysically motivated cell properties ( but see Glazier , 2007 for a discussion on the statistical mechanics of the CPM ) . The energy minimisation occurs through the Metropolis algorithm ( a Monte Carlo method ) , as follows . Fluctuations in cell boundary attempt to copy the spin value σ ( x→ ) of a randomly chosen lattice site x→ to a site x→′ in its Moore neighbourhood . One Monte Carlo Step ( MCS ) consists of L2 attempted copying events , with L2=|Λ1| ( the size of the lattice , and L one of its dimensions on a regular square lattice ) . Throughout this work L=500 . Whether an attempted spin copy is accepted depends on the contribution of several terms to the energy H of the system , as well as other biases Y . A copy is always accepted if energy is dissipated , that is , if ΔH+Y<0 ( with ΔH=Hafter copy−Hbefore copy ) , and may be accepted if ΔH+Y≥0 because of ‘thermal’ fluctuations following a Boltzmann distribution:P ( ΔH , Y ) =e- ( ΔH+Y ) Twith T=16 the Boltzmann temperature , a temperature-like parameter ( in Arbitrary Units of Energy AUE ) that controls the overall probability of energetically unfavourable fluctuations ( allowing escape from local energy minima ) . The Hamiltonian H of the system consists of two terms , corresponding to adhesion and cell size maintenance:H=Hadhesion+Hcell size The copy biases , or ‘work terms’ , Y consist of terms corresponding to cell migration and chemotaxis:Y=Ymigration+Ychemotaxis Adhesion between cells and to medium contribute to the Hamiltonian as:Hadhesion=∑ ( x→ , x→ ′ ) J ( σ ( x→ ) , σ ( x→ ′ ) ) ( 1−δ ( σ ( x→ ) , σ ( x→ ′ ) ) ) where the sum is carried out over all the neighbour pairs ( x→ , x→′ ) , and δ ( σ ( x→ ) , σ ( x→′ ) ) is the Kronecker delta which restricts the energy calculations to neighbouring lattice sites at the interface between two cells , or a cell and medium . J ( σ ( x→ ) , σ ( x→′ ) ) is the contact energy between two adjacent lattice sites x→ and x→′ with different identity ( i . e . J=0 when σ ( x→ ) =σ ( x→ ′ ) ) . In order to calculate the values of J ( σ ( x→ ) , σ ( x→′ ) ) , we assume that cells express ligand and receptor proteins on their surface . Ligands and receptors are modelled as binary strings of fixed length ν ( Figure 1 , inspired by Hogeweg , 2000 ) . Two cells adhere more strongly ( experience lower J values ) when their receptors R and ligands I are more complementary , i . e . when the Hamming distance D ( R , I ) =∑i=1νδ ( Ri , Ii ) between them is larger . Thus , given two cells with spin values σ1 and σ2 and their corresponding pairs of receptors and ligands ( R ( σ1 ) , I ( σ1 ) ) and ( R ( σ2 ) , I ( σ2 ) ) :J ( σ1 , σ2 ) =Jα+2ν-D ( R ( σ1 ) , I ( σ2 ) ) -D ( R ( σ2 ) , I ( σ1 ) ) with Jα=4 chosen so that the final calculation yields values for J ( σ1 , σ2 ) in the interval [4 , 52] . For any particular receptor R there is a single ligand I which is maximally complementary , leading to a J value of 4; and a single I which is maximally similar , leading to a J value of 52 . Adhesion of a cell with medium is assumed to depend only on the cell ( the medium is inert , that is , J ( σmedium , σmedium ) =0 ) , and in particular it depends only on a subset of the ligand proteins of a cell . This subset consists of the substring of I which begins at the initial position of I and has length ν′ . The value of J ( σ1 , σmedium ) is calculated as:J ( σ1 , σmedium ) =Jα′+∑i=1ν′F ( i ) IiF ( i ) ={4if i=13if i=22if i=31if 4≤i≤60if i>6with Jα′=8 and F ( i ) a piece-wise defined function ( a lookup table ) . The J values range in the interval ( Du et al . , 2015; Jacobeen et al . , 2018 ) . Encoding the energy values for cell adhesion in terms of receptor-ligand binding allows for flexibility and redundancy . Two cells that have the same receptors and ligands ( i . e . given R ( σ1 ) , I ( σ1 ) and R ( σ2 ) , I ( σ2 ) with R ( σ1 ) =R ( σ2 ) and I ( σ1 ) =I ( σ2 ) ) can have any J value , by virtue of the particular receptor and ligand combination . The lookup table for the J value with the medium was chosen to allow for a wide variety of possible J values with a small number of bits . Finally , implementing receptors and ligands in terms of binary strings allows for a simple evolutionary scheme , where mutations consist of random bit-flipping ( more on this below ) . The numerical values of the various constants are chosen with four criteria in mind: ( 1 ) the receptor-ligand system has to be long enough that many different combinations are possible , so that its evolution is more open-ended; ( 2 ) two cells with random receptors and ligands do not - on average - adhere preferentially to each other or to the medium; ( 3 ) the range of adhesion energy must allow for strong clustering and strong dispersal while cells maintain their integrity; ( 4 ) although we are not fitting cell behaviour to any specific system , the adhesion energies must be in the typical range used to quantitatively model eukaryotic cells with CPM ( Graner and Glazier , 1992; Marée and Hogeweg , 2001; Ouchi et al . , 2003; Magno et al . , 2015 ) . With these constraints we set receptor and ligand lengths to ν=24 . On average , two cells with random receptors and ligands will neither preferentially adhere to each other nor to the medium if their surface tension γ=J ( σcell , σmedium ) −J ( σcell , σcell ) /2 ( see main text ) is zero . We numerically checked ( by generating a large number of ligands and receptors ) that ⟨γ ( cells with random ligand receptors ) ⟩=⟨J ( σcell , σmedium ) −J ( σcell , σcell ) /2⟩=⟨[8 , 20]⟩−⟨[4 , 52]⟩/2=0 . Moreover γmax=18 and γmin=−18 ( parameter values in Graner and Glazier , 1992; Marée and Hogeweg , 2001; Ouchi et al . , 2003; Magno et al . , 2015 ) . Cell size A ( c ) =|c ( s ) | , the number of lattice sites that compose a cell , is assumed to remain close to a target size AT ( equal for all cells ) . This is achieved by adding an energy constraint in the Hamiltonian that penalises cell sizes that are much larger or smaller than AT:Hcell size=∑c∈Cλ ( A ( c ) -AT ) 2with C the set of cells c present in the lattice configuration , and λ a scaling factor for cell stiffness . This formulation captures the fact that cells are elastic objects that resist deformation from a preferred size ( AT ) . Unless otherwise specified , AT=50 lattice sites , chosen small enough to reduce computational load while large enough to avoid lattice anisotropy effects ( Magno et al . , 2015 ) . The numerical value of λ ( set to 5 throughout the paper ) is large enough to preserve cell size but not too large to freeze cells in place ( see Graner and Glazier , 1992; Ouchi et al . , 2003 for details ) . We model migration ( following Beltman et al . , 2007 ) by biasing cell movement to their previous direction of motion p→ ( c ) : extensions of a cell are energetically more favourable when they are closer to the direction of that cell’s p→:Ymigration=−μpcos ( θp ) Where μp is the maximum energy contribution given by migration , and θp is the angle between p→ and the vector that extends from the centre of mass of the cell to the lattice site into which copying is attempted . Every τp MCS the vector p→ is updated: its new value is the vector corresponding to the actual direction of displacement of the cell over the past τp MCS ( scaled to unit ) ( Figure 1 ) . Persistent migration occurs if τp≫1 , and captures the observation that a cell’s cytoskeleton takes some time to re-polarise ( Ridley et al . , 2003 ) . In line with previous CPM-based models of cell migration ( Vroomans et al . , 2012; Vroomans et al . , 2015 ) we set τp=50 MCS . Note that all cells have the same τp , but their initial moment of updating is randomised so that they do not update all at the same time . Individual cells are able to migrate towards the perceived direction of a chemoattractant gradient . The slope of the gradient is very shallow , making it difficult to perceive the direction over the typical length of a cell . Moreover , several sources of noise are introduced: cell’s sampling error due to small size , noise due to integer approximation , and noise due to random absence of the signal . The chemotactic signal is implemented as a collection of integer values on a second two dimensional lattice ( Λ2⊂ℤ2 , with the same dimensions as the CPM lattice ) . The ( non-negative ) value of a lattice site represents the local amount of chemotactic gradient . This value remains constant for the duration of one season ( τs MCS ) . The amount of chemotactic signal χ is largest at the peak , which is located at the centre of one of the lattice boundaries , and from there decays linearly in all directions , forming a gradient: χ ( d ) =1+ ( kχ/100 ) ( L-d ) , where kχ is a scaling constant , d is the Euclidean distance of a lattice site from the peak of the gradient , and L is the distance between the source of the gradient and the opposite lattice boundary; L=|Λ1| for a square lattice . Non integer values of χ are changed to ⌈χ⌉ ( the smallest integer larger than χ ) with probability equal to ⌈χ⌉-χ , otherwise they are truncated to ⌊χ⌋ ( the largest integer smaller than χ ) . Moreover , the value of χ is set to zero with probability pχ=0 to create ”holes’ in the gradient . Setting kχ=1 and pχ=0 . 1 generates a shallow and noisy gradient . In a subset of simulations we used an alternative gradient , assumed to be generated by resources homogenously distributed on an entire side of the lattice , so that concentration isoclines are straight lines , see Appendix 2 . 4 . A cell has limited knowledge of the gradient , as it only perceives the chemotactic signal on the portion of Λ2 corresponding to the cell’s occupancy on Λ1 . We define the vector χ→ ( c ) as the vector that spans from the cell’s centre of mass to the centre of mass of the perceived gradient . Copies of lattice sites are favoured when they align with the direction of the vector χ→ ( c ) , i . e . when there is a small angle θc between χ→ ( c ) and the vector that spans from the centre of mass of the cell to the lattice site into which copying is attempted ( Figure 1 ) :Ychemotaxis=-μχcos ( θc ) where μχ is the maximal propensity to move along the perceived gradient , and is set to μχ=1 in line with previous studies on cell migration ( Vroomans et al . , 2012 ) ( chemotactic behaviour is robust to changes in μχ however , see Appendix 1 . 6 ) . A uniform random θc∈[0 , 2π] is chosen whenever |χ→ ( c ) |=0 , that is , when , locally , there is no gradient ( which may happen for very shallow gradients ) . In a subset of simulations we implemented an alternative mechanism of collective chemotaxis ( proposed by Camley et al . , 2016 ) that does no rely on individual cells’ gradient sensing . The mechanism works by combining three elements: cell–cell adhesion , contact inhibition of locomotion ( Mayor and Carmona-Fontaine , 2010 ) and larger cell polarisation with higher concentration of the chemoattractant . The implementation of this mechanism in the CPM is straightforward . cell–cell adhesion was kept the same as explained above , and the chemoattractant is distributed to form the same gradient as in the previous paragraph . Every MCS each cell measures the average concentration of chemoattractant over the surface it covers χ ( c ) ( note that this is a scalar ) . Then , in the copy biases Y we substitute a new term YCIL to the term Ychemotaxis , with:YCIL=μCILχ ( c ) andμCIL={-3if cell attempts spin copy into medium0if cell attempts spin copy with another cell3if medium attempts spin copy into cell This definition of μCIL introduces contact inhibition of locomotion by decreasing the probability that cells copy into each other , and increasing the probability that cells copy into medium . A population of N cells undergoes the cell dynamics described above for the duration of a season , i . e . τs MCS . At the end of the season the evolutionary dynamics take place . The evolutionary dynamics are decoupled from the cell dynamics for the sake of simplicity , and consist of fitness evaluation , cell replication with mutation , and cell death to enforce constant population size . The evolutionary experiments last 400 seasons - that is , 400 cycles of mutation-selection-dynamics . This value is larger than the time to reach evolutionary steady state in all simulations . Changes in τs result in qualitatively different evolutionary dynamics , as reported in the main text . Fitness , that is , the probability of replication , is calculated at the end of each season for each cell . We do not include any explicit advantage or disadvantage due to multicellularity , and instead assume that fitness is based only on individual properties of the cells . Therefore , any multicellular behaviour is entirely emergent in this simulation . The fitness F ( c ) of a cell c∈C depends only on the distance d=d ( c ) of the centre of mass of a cell c from the peak of the gradient as a sigmoid function which is maximal when d=0 , and decreases rapidly for larger values of d:F ( c ) =11+ ( hdd ) 2with hd being the distance at which F ( c ) =1/2 . In a subset of simulations ( see Appendix 4 . 1 ) we include a fitness penalty due to the metabolic costs of maintaining adhesion with other cells . We compute the average amount of boundary a cell has in contact with other cells over the course of a season ⟨m⟩ . The fitness of a cell F ( c ) at the end of the season is then multiplied by a decreasing function of ⟨m⟩ . For simplicity we use a linear function: 1-cm⟨m⟩ , with cm the metabolic cost of adhesion , which can vary in [0 , 1] . With small costs ( cm∼0 ) there is little penalty associated with adhering , whereas with large costs ( cm∼1 ) the fitness penalty punishes adhering cells more severely than non-adhering ones . When cm=1 , a cell that spent the entire season completely surrounded by other cell has fitness 0 , that is , it will not reproduce . For each cell i∈C with fitness F ( i ) , the probability of replicating is P ( cell i replicates ) =F ( i ) /∑c∈CF ( c ) . We allow for N replication events , each calculated with the same probabilities , choosing only cells that were already present in the previous season ( so not their offspring ) . Cells with larger fitness may be chosen multiple times for replication . Each replicating cell divides along its short axis ( see e . g . Hogeweg , 2000 ) , ensuring that related cells start close to each other at the beginning of the new season . One of the two daughter cells , chosen randomly , can re-enter the competition for replication . All the lattice sites belonging to the other daughter cell are assigned a new ( unique ) spin value and the cell can mutate its receptor and ligand . The bitstrings of the receptor and ligand may be mutated with a per-position probability μR , I . Mutations flip individual bits ( from 0 to 1 , and vice versa ) . Because repeatedly halving a cell’s area would quickly lead to very small cells , we run a small number η of steps of the cell dynamics ( without cell migration and chemotaxis ) between two replication events that affect the same cell , so that cells can grow back to target size ( η=5 MCS suffices ) . After replication , there are 2N cells on the lattice . In order to restore the initial population size N , half of the cells are removed from the lattice at random . When the initial population size is restored , the season ends . The new season begins by randomly placing the peak of a new gradient at the mid-way point of a randomly chosen boundary ( different from the previous one ) . The remaining cells will then undergo the cell dynamics for the following τs MCS . | All multicellular organisms , from fungi to humans , started out life as single cell organisms . These cells were able to survive on their own for billions of years before aggregating together to form multicellular groups . Although there are trade-offs for being in a group , such as sharing resources , there are also benefits: in a group , single cells can divide tasks amongst themselves to become more efficient , and can develop sophisticated mechanisms to protect each other from harm . But what compelled single cells to make the first move and aggregate into a group ? One way to answer this question is to study the behaviour of slime moulds . These organisms exist as single cells but form colonies when their resources run low . Researchers have observed that slime mould colonies can navigate their environment much better than single cells alone . This property suggests that the benefits of moving together as a collective could be the driving factor propelling single cells to form groups . To test this theory , Colizzi et al . developed a computer model to examine how well groups of cells and lone individuals responded to nearby chemical cues . Unlike previous simulations , the model created by Colizzi et al . did not specify that being in a group was necessarily more favourable than existing as a single cell . Instead , it was left for evolution to decide which was the best option in response to the changing environmental conditions of the simulation . The mathematical model showed that groups of cells were generally better at sensing and moving towards a resource than single cells acting alone . Single cells moved at the same speed as groups , but they often sensed their environment poorly and got disorientated . Only when the environment changed frequently , did cells revert back to the single life . This was because it was no longer beneficial to band together as a group , as the cells were unable to sense the environmental cues fast enough to communicate to each other and coordinate a response . This work provides insights into what drove the early evolution of complex life and explains why , under certain conditions , single cells evolved to form colonies . Additionally , if this model were to be adopted by cancer biologists , it could help researchers better understand how cancer cells form groups to move and spread around the body . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"computational",
"and",
"systems",
"biology"
] | 2020 | Evolution of multicellularity by collective integration of spatial information |
Genetic alterations which impair the function of the TP53 signaling pathway in TP53 wild-type human tumors remain elusive . To identify new components of this pathway , we performed a screen for genes whose loss-of-function debilitated TP53 signaling and enabled oncogenic transformation of human mammary epithelial cells . We identified transglutaminase 2 ( TGM2 ) as a putative tumor suppressor in the TP53 pathway . TGM2 suppressed colony formation in soft agar and tumor formation in a xenograft mouse model . The depletion of growth supplements induced both TGM2 expression and autophagy in a TP53-dependent manner , and TGM2 promoted autophagic flux by enhancing autophagic protein degradation and autolysosome clearance . Reduced expression of both CDKN1A , which regulates the cell cycle downstream of TP53 , and TGM2 synergized to promote oncogenic transformation . Our findings suggest that TGM2-mediated autophagy and CDKN1A-mediated cell cycle arrest are two important barriers in the TP53 pathway that prevent oncogenic transformation .
The TP53 ( known as p53 ) protein is a central component of the tumor suppressive network that monitors oncogenic transformation ( Vogelstein et al . , 2000 ) . Activation of TP53 by oncogenic stress stimulates the transcription of a host of genes , in particular those involved in cell cycle arrest , apoptosis , metabolism , and autophagy ( Bieging and Attardi , 2012 ) . The relative contribution of these genes to tumor suppression by TP53 is likely to be tissue- and context-dependent , but it is clear that they play complementary roles . For example , mutations in CDKN1A ( known as p21 ) , which is one of the best-characterized direct target genes of TP53 and prevents cell cycle progression ( Abbas and Dutta , 2009 ) , are much less frequent than mutations in TP53 . Moreover , Cdkn1A knockout mice have a much lower tumor penetrance than TP53 knockout mice ( Martin-Caballero et al . , 2001 ) , suggesting that additional TP53 targets must contribute to tumor suppression ( Brady et al . , 2011 ) . It has been shown that TP53 activity is required to prevent tumorigenesis in vivo ( Bieging and Attardi , 2012 ) and transformation in vitro ( Hahn et al . , 1999 ) . For example , primary human mammary epithelial cells ( HMECs ) can be fully transformed to form colonies in soft agar and tumors in immunocompromised mice by overexpressing TERT , HRASV12 , and the SV40 oncoproteins large T and small T , which inactivate TP53 and RB1/pRB , and PP2A , respectively ( Elenbaas et al . , 2001; Hahn et al . , 2002 ) . This in vitro transformation model is particularly powerful for identifying and studying putative tumor suppressor genes in the TP53 pathway ( Drost et al . , 2010; Voorhoeve et al . , 2006 ) , especially compared to cancer-derived cell lines or spontaneously immortalized cells such as MCF10A cells in which the tumor suppressive network has been inactivated in a variety of ways ( Kadota et al . , 2010 ) . Given the crucial role of the TP53 pathway in tumor suppression , the significant proportion of tumors that still express wild-type TP53 are likely to harbor alternative lesions that override TP53 activity , most prominently MDM2 overexpression or loss of CDKN2A ( p14ARF ) expression ( Vogelstein et al . , 2000 ) . In addition , a significant number of TP53 wild-type breast cancer tumor lose expression of BRD7 , a transcriptional cofactor of TP53 , compared to TP53 mutant tumors ( Drost et al . , 2010; Miller et al . , 2005 ) . Therefore , to identify genes that modulate the TP53 pathway for tumor suppression , we developed a loss-of-function screen employing HMECs . In HMECs , the TP53 pathway is intact , but the RB1/pRB pathway is disrupted due to silencing of the CDKN2A ( INK4A/p16 ) promoter ( Stampfer and Yaswen , 2000 ) . Utilizing these characteristics , we established a primary HMEC malignant transformation system that is genetically defined to depend on the loss of TP53 activity for full transformation , which can be assessed by analyzing colony formation in a soft agar assay . This screen uncovered tissue transglutaminase 2 ( TGM2 ) as a tumor suppressor that inhibits oncogenic transformation of HMECs . We showed that TGM2 expression is regulated by TP53 to suppress oncogenic transformation of , and tumor formation by , primary HMECs . We provide evidence that reduced TGM2 expression induces colony formation in soft agar possibly due to defects in autophagy , specifically autophagic protein degradation and autolysosome clearance . Importantly , simultaneous knockdown of TGM2 and CDKN1A synergistically promotes transformation , revealing the complementary and essential roles of TP53-induced autophagy and cell cycle arrest in tumor suppression .
To identify new genes within the TP53 tumor suppressor pathway , we established an assay in which the loss of TP53 signaling promotes oncogenic transformation . We employed human mammary epithelial cells ( HMECs ) since the TP53 pathway is intact , but the RB1/pRb pathway is disrupted due to silencing of the CDKN2A ( INK4A/p16 ) promoter ( Stampfer and Yaswen , 2000 ) . HMECs require TERT , oncogenic ER-HRASV12 , SV40 large T and small T antigen for full transformation ( Hahn et al . , 2002 ) . However , we did not use large T antigen since it would perturb the TP53 pathway . ER-HRASV12 is a fusion protein of HRASV12 with the hormone-binding domain of the estrogen receptor , and is activated by the addition of 4-Hydroxy-Tamoxifen ( 4-OHT ) ( Voorhoeve et al . , 2006 ) . This inducible HRAS system allowed us to minimize the emergence of aberrant clones arising from HRAS oncogenic stress . These cells are referred to as HMECTERT/ST/ER-RasV12 cells throughout this paper . To develop a soft agar screen that suppresses colony formation in TP53 wild-type but not TP53 depleted cells , we first plated HMECTERT/ST/ER-RasV12 cells in medium supplemented with 4-OHT ( to activate HRASV12 ) , EGF , insulin , and hydrocortisone ( Drost et al . , 2010; Hahn et al . , 2002 ) . Unexpectedly , many colonies grew in soft agar under these conditions , even though the TP53 pathway was not specifically inhibited ( Figure 1—figure supplement 1 , first column ) . In addition , the number of colonies was not significantly increased by TP53 shRNA ( Voorhoeve and Agami , 2003 ) ( Figure 1—figure supplements 1 and 2 ) , suggesting that TP53 activity does not inhibit oncogenic transformation under these conditions . Therefore , we tested more stringent conditions that would avoid transformation due to potentially oversaturated growth supplements . We found that HMECTERT/ST/ER-RasV12 cells produced significantly fewer colonies when they were grown in medium with only 4-OHT for the first 3 days , followed by medium with 4-OHT , EGF , insulin , and hydrocortisone ( Figure 1A , first column ) . Importantly , knockdown of TP53 substantially increased the number of colonies , suggesting that the loss of TP53 activity is required for transformation under these conditions ( Figure 1A and Figure 1—figure supplement 3 ) . Therefore , we used these conditions to identify genes whose loss compromises the TP53 pathway . 10 . 7554/eLife . 07101 . 003Figure 1 . TGM2 suppresses transformation of primary human mammary epithelial cells in soft agar . ( A ) HMECTERT/ST/ER-RasV12 cells were transduced with retroviral vectors encoding control or TP53 shRNAs and plated in soft agar in medium with 4-OHT ( to activate RasV12 ) . Growth supplements ( EGF , insulin , hydrocortisone ) were withheld for the first 3 days . Results ( left panel ) shown are the average colony number ± SD in biological triplicates . Representative MTT-stained colonies are shown in the right panel . ( **p<0 . 01 compared to control cells , student’s t-test ) ( B ) Flow diagram for the shRNA screen . The candidate gene list of 122 genes is selected by comparing genes with lower expression in a significant number of TP53 wild-type tumor samples versus TP53 mutant tumor samples using expression array ( GSE3494 ) consisting of 251 breast cancer samples with TP53 mutation status , followed by designing and cloning of shRNAs against the genes . These genes could be under selective pressure to lose expression only in TP53 wild-type tumors , thus could be potential members of the TP53 pathway . HMECTERT/ST/ER-RasV12 cells were generated by overexpressing TERT , SV40 small T antigen and ER-HRASV12 . The shRNAs against the candidate genes were introduced into the cells and observed for colony formation in soft agar in the primary and secondary screen . ( C ) Soft agar analysis for HMECTERT/ST/ER-RasV12 cells expressing control , TGM2- ( denoted as TGM2#1 or TGM2#2 ) , TP53- , or CDKN1A- shRNAs using the conditions described in ( B ) . Quantification shows average colony number ± SD in biological triplicates . ( *p<0 . 05; **p<0 . 01 compared to control cells , student’s t-test ) ( D ) Knockdown efficiency of TGM2 with two independent shRNAs . TGM2 protein expression was analyzed by Western blotting . β-ACTIN serves as the loading control . ( E ) TGM2 mRNA expression was quantified by qPCR , normalized to TBP expression and to control vector in biological triplicates , and represented as the average fold change ± SD . ( **p<0 . 01 compared to control cells , student’s t-test ) ( F ) Soft agar assay analysis of HMECTERT/ST/ER-RasV12 cells transduced with a retrovirus expressing mCherry with either control or TGM2#1 shRNAs . The populations were verified to have more than 70% mCherry positive cells , and additionally transduced and selected to express an empty vector ( EV ) or a shRNA-resistant TGM2 cDNA ( TGM2RcDNA ) by retroviruses . Quantification shows average colony number ± SD in biological triplicates . ( **p<0 . 01 , student’s t-test ) ( G ) Western blot analysis of TGM2 protein expression for ( F ) . β-ACTIN serves as the loading control . ( EV , Empty Vector ) ( H ) Soft agar assay analysis of BJTERT/ST/ER-RasV12/shp16 cells transduced with a retrovirus expressing control , TGM2- , or TP53- shRNA . Quantification shows average colony number ± SD in biological triplicates . ( *p<0 . 05; **p<0 . 01 , compared to control cells , student’s t-test ) ( I ) Soft agar assay analysis of NIH 3T3ER-RasV12 cells transduced with a retrovirus expressing control or Tgm2-shRNAs . Quantification shows average colony number ± SD in biological triplicates . ( **p<0 . 01 compared to control cells , student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 00310 . 7554/eLife . 07101 . 004Figure 1—figure supplement 1 . The effect of TP53 on colony formation in HMECTERT/ST/ER-RasV12 cells in the presence of EGF , insulin , and hydrocortisone . Soft agar assay analysis of HMECTERT/ST/ER-RasV12 cells transduced with retroviral vectors encoding control or TP53 shRNAs . Cells were plated in soft agar in medium containing growth supplements ( EGF , insulin , hydrocortisone ) and 4-OHT ( to activate RasV12 ) . Results ( left panel ) shown are the average colony number ± SD in biological triplicates . Representative MTT-stained colonies are shown in the right panel . ( ns: not significant , student’s t-test ) DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 00410 . 7554/eLife . 07101 . 005Figure 1—figure supplement 2 . Knockdown efficiency of TP53 shRNA . Western blot analysis corresponding to Figure 1—figure supplement 1 to assess the knockdown efficiency of TP53 shRNA . β-ACTIN serves as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 00510 . 7554/eLife . 07101 . 006Figure 1—figure supplement 3 . Knockdown efficiency of TP53 shRNA for Figure 1A . Western blot analysis corresponding to Figure 1A to assess the knockdown efficiency of TP53 shRNA . β-ACTIN serves as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 00610 . 7554/eLife . 07101 . 007Figure 1—figure supplement 4 . Generating the candidate gene list for screening . By utilizing a publicly available breast cancer dataset ( GSE3494 ) , a chi-square test was used to identify genes with an asymmetric distribution in breast tumors with ‘low’ expression ( lowest = 0 , highest = 1 ) in 205 TP53 wild-type samples and 46 TP53-mutant samples . The cut-off level was determined at the average minus one standard deviation cut-off ( horizontal axis ) . A total of 122 genes with lower expression in a significant number of TP53 wild-type samples compared to TP53 mutant samples were selected based on their χ2 p values ( p<0 . 01 ) . MINA was selected as one of the genes in the candidate list and its normalized mRNA signal is shown as an example . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 00710 . 7554/eLife . 07101 . 008Figure 1—figure supplement 5 . Generating the candidate gene list for screening . Known TP53 downstream target genes , BAX , and transcriptional cofactor of TP53 , BRD7 , were also in the candidate list and downregulated in a subset of TP53 wild-type tumors relative to TP53 mutant tumors by chi-square analysis ( p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 00810 . 7554/eLife . 07101 . 009Figure 1—figure supplement 6 . Generating the candidate gene list for screening . TGM2 expression analysis is shown to illustrate lower expression in a subset of TP53 wild-type tumors compared to TP53 mutant tumors by chi-square analysis ( p<0 . 01 ) DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 00910 . 7554/eLife . 07101 . 010Figure 1—figure supplement 7 . Pictures of soft agar assay for Figure 1C . Representative MTT-stained colonies corresponding to Figure 1C are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 01010 . 7554/eLife . 07101 . 011Figure 1—figure supplement 8 . Protein expression for Figure 1C . Western blot of TGM2 , CDKN1A and TP53 corresponding to Figure 1C . β-ACTIN serves as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 01110 . 7554/eLife . 07101 . 012Figure 1—figure supplement 9 . Pictures of soft agar assay for Figure 1H . Representative MTT-stained colonies corresponding to Figure 1H are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 01210 . 7554/eLife . 07101 . 013Figure 1—figure supplement 10 . Protein expression for Figure 1H . Western blot of TGM2 and TP53 corresponding to Figure 1H . β-ACTIN serves as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 01310 . 7554/eLife . 07101 . 014Figure 1—figure supplement 11 . Pictures of soft agar assay for Figure 1I . Representative MTT-stained colonies corresponding to Figure 1I are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 01410 . 7554/eLife . 07101 . 015Figure 1—figure supplement 12 . Protein expression for Figure 1I . Western blot of TGM2 corresponding to Figure 1I . β-ACTIN serves as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 015 We searched a publicly available breast cancer expression array dataset for genes with reduced expression in a significant number of TP53 wild-type tumor samples compared to TP53 mutant tumor samples ( GSE3494 ) ( Figure 1B and Supplementary file 1 ) ( Miller et al . , 2005 ) . We reasoned that there is selective pressure to abrogate TP53 signaling during carcinogenesis , and that the loss of expression of TP53 pathway components would be more frequent in a subset of TP53 wild-type tumors compared to TP53 mutant tumors . Thus , genes with reduced expression in a subset of TP53 wild-type ( see red circle in Figure 1—figure supplements 4 , 5 , and 6 ) but not in mutant TP53 tumors are potential members of the TP53 pathway . We selected 122 candidates with significantly lower expression in a subset of TP53 wild-type tumor samples compared to TP53 mutant tumor samples using Chi-square analysis ( Figure 1B and Supplementary file 1 ) . We individually transduced shRNAs for each of these 122 candidate genes into HMECTERT/ST/ER-RasV12 cells , and evaluated the effect of knockdown on colony formation in soft agar assays ( Figure 1B and Supplementary file 1 ) . CDKN1A shRNA ( Voorhoeve et al . , 2006 ) was used as a positive control , since it is a well-known downstream target gene of TP53 and since reduced CDKN1A expression promotes cell transformation ( Schaefer et al . , 2010; Zhang et al . , 2013; Zou et al . , 2002 ) . Among the 122 candidate genes , there were four shRNAs that produced colonies in soft agar from the primary screen ( Supplementary file 1 ) . To exclude off-target effects , we constructed additional shRNAs against the four genes in a secondary screen ( Supplementary file 2 ) , and found that knockdown of only one , TGM2 , produced colonies in soft agar with at least two independent shRNAs ( denoted as TGM2#1 and TGM2#2 ) ( Figure 1C and Figure 1—figure supplements 7 and 8 ) . The number of colonies formed correlated with the efficiency of TGM2 knockdown ( Figure 1D and E ) . To exclude the possibility that these two independent shRNAs against TGM2 share off-target activity , we restored TGM2 expression in HMECTERT/ST/ER-RasV12 cells expressing TGM2#1 shRNAs with an shRNA-resistant cDNA ( TGM2R ) . TGM2 restoration at physiological levels significantly suppressed colony formation ( Figure 1F and G ) . Taken together , these findings uncover TGM2 as a putative tumor suppressor gene that functions within the TP53 pathway to prevent oncogenic transformation of HMECs . We further validated the effect of TGM2 knockdown on colony formation in soft agar using different cell types . Human foreskin fibroblast BJ cells were retrovirally transduced with TERT , ER-HRASV12 , SV40 small T , and p16INK4a shRNA ( to disrupt the Rb pathway ) . Two independent shRNAs against TGM2 were further transduced into these cells and the number of colonies was evaluated . Consistent with the results from HMECs , knockdown of TGM2 enhanced the colony formation in BJTERT/ST/ER-RasV12/shp16 cells ( Figure 1H and Figure 1—figure supplements 9 and 10 ) . A similar result was also obtained with mouse embryonic fibroblast NIH 3T3 cells expressing ER-HRASV12 ( Figure 1I and Figure 1—figure supplements 11 and 12 ) . These results suggest that TGM2 has a tumor suppressive role not only in human mammary epithelial cell ( HMECs ) , but also in BJ human fibroblasts and NIH 3T3 mouse fibroblasts . TGM2 could act downstream , upstream , or as a co-regulator of TP53 to support tumor suppression . To distinguish between these possibilities , we assessed TP53 expression and activity in TGM2 knockdown HMECTERT/ST/ER-RasV12 cells . Depletion of TGM2 expression did not reduce TP53 expression , or its transcription factor activity , as measured by the expression of TP53 target genes such as CDKN1A and MDM2 ( Figure 2A ) . Thus , TGM2 is not an upstream regulator or a co-factor of TP53 . In contrast , we observed a significant reduction in TGM2 mRNA and protein expression in cells expressing TP53 shRNAs ( Figure 2B and C ) , suggesting that TGM2 is induced by TP53 . Conversely , reconstitution of TP53 expression in HMECTERT/ST/ER-RasV12 cells co-expressing a TP53 shRNA with a TP53 shRNA-resistant cDNA ( His-tag TP53R ) restored TGM2 protein and mRNA expression ( Figure 2D and E ) , further validating that TGM2 is regulated by TP53 . Consistent with these findings , the levels of TGM2 mRNA was lower in BJTERT/ST/ER-RasV12/shp16 cells expressing TP53 shRNA compared to control BJTERT/ST/ER-RasV12/shp16 cells ( Figure 2F ) . Furthermore , a reduction in TGM2 mRNA was also observed in Tp53 knockout MEFs compared to wild-type MEFs ( Figure 2G ) . Together , these data strongly suggest that TGM2 is regulated by TP53 . 10 . 7554/eLife . 07101 . 016Figure 2 . TGM2 expression is dependent on TP53 . ( A ) qRT-PCR analysis of the effect of control or TGM2 shRNAs ( denoted respectively as control and TGM2#1 ) on mRNA expression of either TP53 , CDKN1A , MDM2 , or TGM2 in HMECTERT/ST/ER-RasV12 cells . The levels of mRNA were normalized to TBP expression and to control cells . The data indicate the average ± SD of biological triplicates . ( **p<0 . 01 , student’s t-test to control cells ) ( B ) Western blot analysis of HMECTERT/ST/ER-RasV12 cells stably transduced with retroviruses expressing control or TP53 shRNAs . β-ACTIN serves as the loading control . ( C ) qRT-PCR analysis of the cells in ( B ) . The levels of TGM2 mRNA were normalized to TBP expression and to control cells . The data indicate the average ± SD of biological triplicates . ( **p<0 . 01 , student’s t-test to control cells ) ( D ) Western blot analysis of HMECTERT/ST/ER-RasV12 cells stably transduced with a retrovirus expressing mCherry with either control or TP53 shRNAs . The populations were verified to have more than 70% mCherry positive cells , and then retrovirally-transduced and selected to express an empty vector or a 6x His-tag TP53 shRNA-resistant overexpression vector ( His-tag TP53RcDNA ) . β-ACTIN serves as the loading control . ( EV , Empty Vector ) ( E ) qRT-PCR analysis of cells in ( D ) for TGM2 mRNA normalized to TBP expression and to control cells . The data indicate the average ± SD of biological triplicates . ( *p<0 . 05; **p<0 . 01 , student’s t-test ) ( F ) qRT-PCR analysis of BJTERT/ST/ER-RasV12/shp16 cells stably transduced with retroviruses expressing control or TP53 shRNAs . The levels of mRNA were normalized to TBP expression and to control cells . The data indicate the average ± SD of biological triplicates . ( **p<0 . 01 , student’s t-test to control cells ) ( G ) qRT-PCR analysis of wild-type and Tp53-/- MEFs . The levels of mRNA were normalized to Gapdh expression and to control cells . The data indicate the average ± SD of biological triplicates . ( **p<0 . 01 , student’s t-test to control cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 016 To examine if TGM2 is induced by TP53 activation , we treated HMECTERT/ST/ER-RasV12 cells and BJTERT/ST/ER-RasV12/shp16 with Nutlin-3a , a small molecule inhibitor of the TP53-MDM2 interaction ( Tovar et al . , 2006; Vassilev et al . , 2004 ) . MDM2 promotes degradation of TP53 , thus inhibiting this interaction with Nutlin-3a effectively stabilizes TP53 . In both cell types , TGM2 expression was increased by Nutlin-3a treatment in a TP53-dependent manner ( Figure 3A and B ) , suggesting that TGM2 could be a transcriptional target gene of TP53 . 10 . 7554/eLife . 07101 . 017Figure 3 . TGM2 is a potential target gene of TP53 . ( A ) qRT-PCR analysis of HMECTERT/ST/ER-RasV12 cells stably transduced with retroviruses expressing control or TP53 shRNAs . Cells were treated with Nutlin-3a ( 5 µM or 10 µM ) for 2 days . The levels of mRNA were normalized to TBP expression and to control cells . CDKN1A is used as the positive control to see TP53 activation . The data indicate the average ± SD of biological triplicates . ( B ) qRT-PCR analysis of BJTERT/ST/ER-RasV12/shp16 cells stably transduced with retroviruses expressing control or TP53 shRNAs . Cells were treated with Nutlin-3a ( 10 µM ) for 2 days . The levels of mRNA were normalized to TBP expression and to control cells . CDKN1A is used as the positive control to see TP53 activation . The data indicate the average ± SD of biological triplicates . ( C and D ) Luciferase reporter assays using a series of promoter deletion mutants of the TGM2 gene . The number ( bp ) indicates the position relative to the translational start site ( ATG ) . Reporter plasmids containing the indicated deletion constructs were transfected into H1299 cells with control or TP53 plasmid , and luciferase activity was monitored . The average value of the luciferase activity from the cells transfected with CMV-TP53 and the reporter plasmid containing TGM2 ( -1530 to -78 ) promoter fragment is set at 1 , and the relative activity is shown . The data indicate the average ± SD of biological triplicates . ( E ) TP53 binds to the TGM2 promoter . ChIP assay was performed with an antibody detecting endogenous TP53 , or IgG ( negative control ) using HMECTERT/ST/ER-RasV12 cells . The potential TP53 response element in the TGM2 promoter identified in ( D ) was analyzed by PCR . CDKN1A and GAPDH are served as the positive and negative control respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 01710 . 7554/eLife . 07101 . 018Figure 3—figure supplement 1 . Luciferase reporter assays of TGM2 promoter . Luciferase reporter assays using a series of promoter deletion mutants of the TGM2 gene . The number ( bp ) indicates the position relative to the translational start site ( ATG ) . Reporter plasmids containing indicated deletion constructs were transfected into H1299 cells with control or TP53 plasmid , and luciferase activity was monitored . The average value of the luciferase activity from the cells transfected with CMV-TP53 and the reporter plasmid containing TGM2 ( -159 to -78 ) promoter fragment is set at 1 , and the relative activity is shown . The data indicate the average ± SD of biological triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 01810 . 7554/eLife . 07101 . 019Figure 3—figure supplement 2 . Diagram of the TGM2 promoter region . The genomic sequence and location of 82 bp fragment required and sufficient for TP53-mediated transactivation is shown . Chromosome coordinates are given according to GRCh38/hg38 Human Genome Reference . The sequence and location of the primers used for ChIP analysis are also shown ( the position is indicated by the arrow ) . Genbank accession number ( NM_004613 . 2 ) of TGM2 mRNA is shown . The TGM2 gene structure , TP53 responsive element , and primers were drawn proportionally to their sizes . The location of the primers for ChIP analysis were chosen to avoid the difficulty of amplifying the highly GC-rich region on the TGM2 promoter to increase the sensitivity of the ChIP analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 019 It has been reported that the TGM2 promoter contains two predicted binding sites for TP53 , although these sites were not tested for TP53 binding and TP53-mediated transactivation ( Ai et al . , 2012 ) . Several other potential TP53 binding motifs within the TGM2 promoter were also predicted using computer software , p53MH algorithm described previously ( Hoh et al . , 2002 ) . To determine if the TGM2 promoter contains TP53 target sites , we engineered a reporter system which contained a luciferase gene under the control of the partial TGM2 promoter sequences ( Figure 3C ) . We investigated non-overlapping regions from -5980 to -78 base pairs upstream of the TGM2 translational start site ( ATG ) . Co-transfection of the -1530 to -78 region of the TGM2 promoter/luciferase reporter with a TP53 expression vector into TP53-null H1299 cells significantly increased luciferase activity ( Figure 3C ) . We evaluated several deletion constructs corresponding to this region , and unexpectedly found that an element from -159 to -78 of the TGM2 promoter , which does not contain a TP53 binding consensus , was necessary and sufficient for TP53-mediated transactivation ( Figure 3D ) . Further deletion of 5 or 10 nucleotides at the 5’- or 3’-end from this 82 bp element significantly reduced luciferase activity ( Figure 3—figure supplement 1 ) , suggesting that it represents the minimal region for TP53-mediated activation within the TGM2 promoter ( Figure 3—figure supplement 2 ) . Further , we used a TP53 antibody for chromatin immunoprecipitation ( ChIP ) and found that this genomic region of the TGM2 promoter was specifically immunoprecipitated with endogenous TP53 from HMECTERT/ST/ER-RasV12 cells ( Figure 3E and Figure 3—figure supplement 2 ) , suggesting that TP53 directly binds to this element to regulate TGM2 expression . Together , these findings suggest that the TGM2 promoter contains a novel target site for TP53 binding and activation . TP53 is known to induce autophagy , the catabolic breakdown of cellular components by the lysosome ( Budanov and Karin , 2008; Crighton et al . , 2006; Feng et al . , 2005; Kenzelmann Broz et al . , 2013 ) , and autophagy can have a tumor suppressive function ( Karantza-Wadsworth et al . , 2007; Mathew et al . , 2007 ) . Various studies have shown that the absence of growth factor signaling can also induce autophagy ( Cheng et al . , 2010; Eom et al . , 2014; Lum et al . , 2005 ) . We wanted to determine whether the deprivation of growth supplements ( EGF , insulin , and hydrocortisone ) in our soft agar assay induced autophagy in HMECTERT/ST/ER-RasV12 cells and , if so , whether autophagy was TP53-dependent . We used a well-established approach employing a tandem RFP-GFP-LC3 fusion construct to monitor autophagy by fluorescence microscopy ( Mizushima et al . , 2010 ) . MAP1LC3A ( known as LC3 ) is modified to LC3-II , a critical component of autophagosomes , which engulf cellular components and fuse with lysosomes during autophagy ( Kimmelman , 2011 ) . In the absence of autophagy , cells display diffuse colocalization of both red and green signals from RFP-GFP-LC3 . In contrast , the fusion of autophagosomes with lysosomes during autophagy results in rapid quenching of the GFP signal from RFP-GFP-LC3 , since it is more sensitive to the acidic conditions of the autolysosomal lumen than RFP . Therefore , RFP signals without GFP , visualized as red punctae in the cytoplasm , represent acidic compartments such as autolysosomes and signify autophagic flux ( Kimura et al . , 2007; Klionsky et al . , 2012; Mizushima , 2009; Wu et al . , 2010 ) . Since we used a TERT-H2B-GFP construct , which expresses TERT and H2B-GFP , to generate HMECTERT/ST/ER-RasV12 cells ( Kolfschoten et al . , 2005 ) , there is a ubiquitious GFP signal in the nucleoplasm due to the nuclear localization signal of histone H2B . Thus , the green signal from H2B-GFP overlapped with the green signal from RFP-GFP-LC3 in nuclei . Nevertheless , we were still able to monitor autophagic flux by the presence of red punctae in the cytoplasm . About 80% of HMECTERT/ST/ER-RasV12 cells ( denoted as TP53+/+ ) transiently transfected with the tandem RFP-GFP-LC3 construct and cultured in the absence of growth supplements displayed extensive red punctae and no green signal in the cytoplasm , indicative of autophagy ( Figure 4A and B ) ( Boland et al . , 2008 ) . To determine whether the observed autophagy was dependent on TP53 , we generated TP53 knockout HMECTERT/ST/ER-RasV12 cells ( denoted as TP53-/- ) by CRISPR/Cas technology . After transfection of CRISPR plasmids , single clones were isolated and mutation of the TP53 locus was validated by DNA sequencing ( Figure 4—figure supplement 1 ) . In contrast to TP53+/+ cells , only about 40% of TP53-/- cells transfected with RFP-GFP-LC3 exhibited red punctae , and more cells showed complete overlap of green and red fluorescence signals ( Figure 4A and B ) . To exclude potential effects from the clonal selection of TP53-/- cells , we also transfected the RFP-GFP-LC3 construct into cells expressing TP53 shRNA . Consistent with the results from TP53-/- cells , TP53 knockdown cells displayed a lower percentage of cells having red punctae without green signal in the cytoplasm compared to control cells ( Figure 4C and D ) . These observations suggest that the depletion of growth supplements induces TP53-dependent autophagy , which may limit colony formation in our soft agar assay . 10 . 7554/eLife . 07101 . 020Figure 4 . Absence of growth supplements induces TP53-dependent autophagy . ( A ) Formation of red punctae following autophagy induction in the absence of growth supplements . HMECTERT/ST/ER-RasV12 cells ( TP53+/+ ) or TP53 CRISPR knockout HMECTERT/ST/ER-RasV12 cells ( TP53-/- ) cells were transfected with the plasmid RFP-GFP-LC3 . Cells were incubated in medium without EGF , insulin , and hydrocortisone for 24 hr before visualization on a confocal microscope . Scale bar: 5 μm ( B ) Quantification of the fraction of red punctate cells within the total population of transfected cells shown in ( A ) . Red punctate cells were counted as cells containing only RFP signal without visible overlap of GFP signal in the cytoplasm; transfected cells were counted as cells containing either RFP signals or a mix of RFP and GFP signals in the cytoplasm ( >250 cells were counted ) . ( **p<0 . 01 , student’s t-test ) ( C ) Control or TP53 knockdown ( TP53 shRNA ) HMECTERT/ST/ER-RasV12 cells were seeded , treated and visualized as in ( A ) . ( D ) Quantification of the fraction of red punctate cells within the total population of transfected cells shown in ( C ) , treated as in ( B ) . ( **p<0 . 01 , student’s t-test ) ( E ) Western blot analysis of TP53+/+ cells , TP53-/- CRISPR knockout HMECTERT/ST/ER-RasV12 clones , as well as control and TP53 shRNA cells treated with or without chloroquine ( CQ , 50 μM , 2 hr ) after incubation in medium without EGF , insulin , and hydrocortisone for 24 hr . β-ACTIN serves as the loading control . ( F ) Western blot analysis of HMECTERT/ST/ER-RasV12 cells and two independent TP53 CRISPR knockout HMECTERT/ST/ER-RasV12 clones . Cells were incubated in medium in the presence or absence of EGF , insulin , and hydrocortisone ( denoted as EIH ) for 48 hr . β-ACTIN serves as the loading control . ( G ) qRT-PCR analysis of cells in ( F ) . TGM2 mRNA expression was normalized to TBP mRNA expression . The mean value of TGM2 mRNA expression in TP53+/+ cells with presence of EIH is set at 1 , and relative expression is shown . ( **p<0 . 01 , ns: not significant , student’s t-test ) ( H ) Western blot analysis of HMECTERT/ST/ER-RasV12 cells expressing either control or TP53 shRNA . Cells were treated the same as in ( F ) . ( I ) qRT-PCR analysis of cells in ( H ) . Data are shown as in ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 02010 . 7554/eLife . 07101 . 021Figure 4—figure supplement 1 . DNA sequencing of TP53-/- clones . DNA sequencing result of TP53-/- clones showing frameshift mutations in exon 4 of TP53 gene for CRISPR TP53 knockout HMECTERT/ST/ER-RasV12 ( TP53-/- ) clone 1 and clone 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 021 To further assess the role of TP53 in autophagic flux , we treated cells with chloroquine ( CQ ) , an agent that prevents the acidification of lysosomes and inhibits autophagic flux by preventing lysosomal protein degradation ( Shintani and Klionsky , 2004 ) . During active autophagic flux , LC3-I is conjugated to phosphatidylethanolamine ( PE ) to form LC3-II , a widely-used autophagic marker ( Kabeya et al . , 2004 ) . The addition of chloroquine to cells undergoing autophagy prevents LC3-II degradation by lysosomal enzymes , leading to the accumulation of LC3-II protein and an increase in the ratio of LC3-II to LC3-I ( Klionsky et al . , 2012; Mizushima and Yoshimori , 2007 ) . However , if cells do not have active autophagic flux , the levels of both LC3-I and LC3-II will be limited and their ratio will not be affected by chloroquine treatment ( Mizushima and Yoshimori , 2007 ) . The addition of chloroquine to parental HMECTERT/ST/ER-RasV12 cells and to HMECTERT/ST/ER-RasV12 cells expressing control shRNAs triggered an increase in the LC3-II to LC3-I ratio , indicating rapid autophagic flux ( Figure 4E , lane 1 , 2 , 7 , and 8 ) ( Klionsky et al . , 2012 ) . However , TP53 knockdown or knockout cells displayed decreased LC3-I and LC3-II protein levels , and minimal accumulation of LC3-II by chloroquine treatment , indicating that TP53 contributes to active autophagic flux ( Figure 4E ) . To determine if TGM2 is also involved in the autophagy induced by a depletion of growth supplements , we analyzed the expression of TGM2 in the presence or absence of EGF , insulin , and hydrocortisone . Indeed , depletion of these growth supplements increased TGM2 protein and mRNA levels in TP53 wild-type ( TP53+/+ ) HMECTERT/ST/ER-RasV12 cells ( Figure 4F and G ) . In contrast , removal of growth supplements in TP53 knockout HMECTERT/ST/ER-RasV12 cells ( TP53-/- clone 1 and clone 2 ) , as well as in TP53 shRNA cells did not induce TGM2 protein and mRNA expression ( Figure 4F , G , H , and I ) . Thus , TGM2 is induced by the depletion of growth supplements in a TP53-dependent manner , suggesting a potential role for TGM2 in mediating the TP53-induced autophagic program . Previous studies have reported that TGM2 promotes autophagy ( D'Eletto et al . , 2009 ) . Therefore , we hypothesized that a depletion of growth supplements induces TP53-dependent autophagy in part through TGM2 . To test this directly , we transfected the RFP-GFP-LC3 plasmid into HMECTERT/ST/ER-RasV12 cells expressing TGM2 shRNA and cultured them without growth supplements . Contrary to our expectation , knockdown of TGM2 did not affect the fraction of cells having red punctae without green signal , unlike knockdown of TP53 ( Figure 5A and B ) . This finding indicates that reduced TGM2 expression does not prevent fusion of autophagosomes with lysosomes to form autolysosomes . Interestingly , however , we noticed that TGM2 knockdown cells displayed a two-fold enlargement in the size of red punctae compared to control cells ( Figure 5A and C ) . Furthermore , we observed similarly enlarged yellow punctae in both control and TGM2 knockdown cells after chloroquine treatment ( Figure 5A and C ) . Addition of chloroquine blocks the acidification of lysosomes , thereby preventing autophagic protein degradation , enlarging the size of autolysosome and preventing the quenching of GFP fluorescence in the autolysosome . The formation of enlarged red punctae upon knockdown of TGM2 , even without chloroquine treatment , suggests a defect in the later steps of autophagy such as autolysosome clearance , leading to autolysosome enlargement ( Boland et al . , 2008; Nixon et al . , 2005 ) . 10 . 7554/eLife . 07101 . 022Figure 5 . Loss of TGM2 expression inhibits autophagic protein degradation and autolysosome clearance . ( A ) Formation of red punctae following autophagy induction in the absence of growth supplements . HMECTERT/ST/ER-RasV12 cells ( control ) and TGM2 shRNA cells ( TGM2#1 ) were transfected with the plasmid RFP-GFP-LC3 . Cells were incubated in medium without EGF , insulin , and hydrocortisone with or without chloroquine ( 20 µM ) for 24 hr before visualization on confocal microscope . Scale bar: 5 μm ( B ) Quantification of fraction of cells showing only red punctate within the total population of transfected cells in ( A ) . Cells were categorized and counted as in Figure 4B ( >250 cells counted ) . ( ns: not significant , student’s t-test ) ( C ) Quantification of puntate size in control and TGM2 shRNA ( TGM2#1 ) cells with and without chloroquine treatment . Average area of each vesicle per cell in ( A ) was analyzed by Image J software and represented as box plot . ( **p<0 . 01 , ns: not significant , student’s t-test ) ( D ) Western blot analysis of HMECTERT/ST/ER-RasV12 cells expressing control and TGM2 shRNA using independent hairpins ( TGM2#1 and double hairpins TGM2#2/3 ) treated with or without chloroquine ( CQ , 50 μM , 2 hr ) after incubation in medium without EGF , insulin , and hydrocortisone for 24 hr . Note that for TGM2#2/3 , cells were generated by transducing with retrovirus carrying TGM2#2 and TGM2#3 shRNAs to achieve similar knock-down efficiency to TGM2 shRNA#1 . β-ACTIN serves as the loading control . ( E ) Transmission electron microscopy images for control and TGM2 shRNA ( TGM2#1 ) cells . Cells were incubated without EGF , Insulin , and hydrocortisone for 24 hr before fixing and imaging . The higher magnification micrograph ( x40 , 000 ) shows presence of undigested protein aggregates ( arrowheads ) in autophagic vesicles . Lower magnification was set at x10 , 000 . Scale bar: 1 μm at x10 , 000 and 0 . 2 μm at x40 , 000 . ( F ) Flow cytometry analysis of cells stained with lysotracker after incubation in media without EGF , insulin , and hydrocortisone for 24 hr . The mean fluorescence intensity of control and TGM2 shRNA ( TGM2#1 ) cells was quantified by Flowjo software . The data indicate the mean ± SD of biological triplicates . ( ns: not significant , student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 022 To test whether TGM2 promotes autophagic protein degradation and autolysosome clearance , we treated control and TGM2 knockdown cells with chloroquine ( CQ ) and monitored LC3-I and LC3-II protein levels . A block in autophagic protein degradation and autolysosome clearance in cells would be predicted to lead to increased LC3-II protein levels even in the absence of chloroquine , and the addition of chloroquine in these cells would result in only a minimal increase in LC3-II and in the ratio of LC3-II to LC3-I ( Mizushima and Yoshimori , 2007 ) . The ratio of LC3-II/LC3-I in control HMECTERT/ST/ER-RasV12 cells increased from 1 . 3-fold to 5 . 9-fold after addition of chloroquine , indicating rapid autophagic flux in untreated control cells ( Figure 5D , compare lane 1 and 2 ) ( Klionsky et al . , 2012 ) . In contrast , knockdown of TGM2 with independent hairpins ( TGM2#1 and TGM2#2/3 ) increased the LC3-II/LC3-I ratio even in the absence of chloroqiune ( Figure 5D , compare lane 1 , 3 and 5 ) , to a similar extent as that observed in control cells after chloroquine treatment ( Figure 5D , lane 2 ) . Furthermore , the addition of chloroquine did not further elevate the levels of LC3-II or the LC3-II/LC3-I ratio in TGM2 knockdown cells compared to control cells ( Figure 5D , lane 3–6 ) , suggesting that TGM2 knockdown leads to a defect in autophagic protein degradation . The SQSTM1/p62 protein is known to bind ubiquitinated proteins and transport them to the autophagy machinery for their degradation ( Kimmelman , 2011 ) . A block in autophagy leads to the accumulation of SQSTM1 , since the SQSTM1 protein itself is degraded by autophagy ( Mizushima and Yoshimori , 2007 ) . We observed an accumulation of SQSTM1 protein in TGM2 knockdown cells ( Figure 5D ) , providing additional evidence that loss of TGM2 impairs autophagic protein degradation and autolysosome clearance . To further clarify how loss of TGM2 impacts autophagy , we used transmission electron microscopy to visualize autophagic vesicles at the ultrastructural level ( Mizushima et al . , 2010 ) . We observed an increase in the size and number of vesicles , as well as an accumulation of undigested protein , seen as black aggregates within the vesicles in TGM2 knockdown cells compared to control cells , consistent with impairment in autophagic protein degradation and autolysosome clearance ( Figure 5E ) . To determine if loss of TGM2 altered the acidity of autolysosomes , thereby preventing protein degradation , we incubated cells with Lysotracker Red , a fluorescent dye which labels acidic vesicles in living cells . We used flow cytometry to quantify the fluorescence intensity , and thus the acidity of vesicles ( Chikte et al . , 2014 ) . Interestingly , we did not observe a significant difference in the fluorescence levels between control and TGM2 knockdown cells ( Figure 5F ) , consistent with our observation that the percentage of cells having red punctae without green signal was similar between control and TGM2 knockdown cells transfected with the RFP-GFP-LC3 construct ( Figure 5A and B ) . These data suggest that the loss of TGM2 does not alter the acidity of autolysosomes , but rather impairs their content degradation and clearance . Taken together , our results suggest a role of TGM2 in autophagy by promoting autophagic protein degradation and autolysosome clearance . We showed that loss of either TGM2 or CDKN1A could stimulate colony formation in soft agar ( Figure 1C ) . The major function of CDKN1A is to promote cell cycle arrest ( Abbas and Dutta , 2009; Chang et al . , 2000 ) , whereas we show here that one function of TGM2 is to promote autophagy by facilitating autophagic protein degradation and autolysosome clearance ( Figure 5 ) . Next , we examined whether these different functions of TGM2 and CDKN1A cooperate to protect against cell transformation . To this end , we performed combined knockdown of TGM2 and CDKN1A in HMECTERT/ST/ER-RasV12 cells to determine the effect on colony formation ( Figure 6A and Figure 6—figure supplement 1 ) . We used two independent shRNAs against TGM2 ( #1 and #2 ) . TGM2#1 shRNA showed a higher knockdown efficiency compared to TGM2#2 shRNA ( Figure 1D and E , and Figure 6—figure supplement 2 ) . We found that the number of colonies formed with the simultaneous knockdown of TGM2 and CDKN1A was significantly greater than with the knockdown of each gene individually ( Figure 6A , column 2 , 3 , 4 , 8 , and 10 and Figure 6—figure supplement 1 ) . Unexpectedly , double knockdown of CDKN1A/TGM2#1 gave rise to more colonies than single knockdown of TP53 ( Figure 6A , column 5 and 8 ) . This could be due to the residual expression of TGM2 in TP53 knockdown cells compared to CDKN1A/TGM2#1 knockdown cells , arising from the insufficient down regulation of TGM2 by a single TP53 shRNA ( compare lane 1 , 5 , and 8 in Figure 6—figure supplement 2 ) . Indeed , cells expressing both TGM2#2 shRNA , which has a lower knockdown efficiency of TGM2 ( Figure 1D and E , and Figure 6—figure supplement 2 ) , and CDKN1A shRNA generated a similar number of colonies as TP53 knockdown cells ( Figure 6A , column 5 and 10 ) . Knockdown of CDKN1A and TP53 together generated a similar number of colonies as loss of TP53 alone , indicating that these two genes act in the same pathway ( Figure 6A , column 5 and 6 ) . Furthermore , the effect of TP53/TGM2#2 knockdown was also comparable to the potency of TP53 knockdown alone for inducing colony formation in soft agar ( Figure 6A , column 5 and 9 ) . TP53/CDKN1A knockdown and CDKN1A/TGM2#2 knockdown cells also did not generate significantly different colony numbers ( Figure 6A , column 6 and 10 ) . These data suggest that CDKN1A and TGM2 suppress colony formation mainly through the TP53 pathway . The cooperative effect of CDKN1A and TGM2 knockdown indicates that they provide complementary contributions to tumor suppression and that loss of each gene function is critical for oncogenic transformation . 10 . 7554/eLife . 07101 . 023Figure 6 . Loss of TGM2 expression synergizes with loss of CDKN1A to promote the transformation of HMECTERT/ST/ER-RasV12 cells . ( A ) Soft agar assay analysis of HMECTERT/ST/ER-RasV12 cells transduced with a retrovirus expressing mCherry with either control , CDKN1A or TP53 shRNAs . The populations were verified to have more than 70% mCherry positive cells . The cells were then additionally transduced with the indicated shRNA constructs , selected with 4 µg/ml of blasticidin , and evaluated by soft agar assay analysis . The results shown are the average colony number ± SD from biological triplicates . ( *p<0 . 05 , ns: not significant , student’s t-test ) ( B ) Formation of tumors in NOD/SCID mice . 500 , 000 HMECTERT/ST/RasV12 cells transduced with the indicated vectors were injected subcutaneously in mice ( n=6 ) . ( Top ) The number of tumors observed after 4- and 6-weeks from the time of injection . ( Bottom ) Pictures of tumors excised 4 weeks after injection . ( C and D ) HMECTERT/ST/ER-RasV12 cells were transduced with a retrovirus expressing mCherry with either control or CDKN1A shRNAs . The populations were verified to have more than 70% mCherry positive cells . The cells were then additionally transduced with the indicated shRNA constructs , selected with 4 µg/ml of blasticidin , and evaluated by soft agar assay analysis . The results shown are the average colony number ± SD from biological triplicates . ( *p<0 . 05 , **p<0 . 01 , ns: not significant , student’s t-test ) ( E and F ) HMECTERT/ST/ER-RasV12 cells were transduced with a retrovirus expressing mCherry with either control or TGM2 ( TGM2#1 ) shRNAs . The populations were verified to have more than 70% mCherry positive cells . The cells were then additionally transduced with the indicated shRNA constructs , selected with 4 µg/ml of blasticidin , and evaluated by soft agar assay analysis . The results shown are the average colony number ± SD from biological triplicates . ( *p<0 . 05 , ns: not significant , student’s t-test ) . ( G ) Western blot analysis of HMECTERT/ST/ER-RasV12 cells treated with TGM2 inhibitors . The cells were incubated in medium without EGF , insulin , and hydrocortisone in the presence of Z-DON ( 50 µM ) or LDN 27219 ( 10 µM ) for 24 hr . β-ACTIN serves as the loading control . ( H ) Soft agar assay analysis of HMECTERT/ST/ER-RasV12 cells transduced with a retrovirus expressing mCherry with either control or TGM2 ( TGM2#1 ) shRNAs . The populations were verified to have more than 70% mCherry positive cells , and additionally transduced and selected with 4 µg/ml of blasticidin to express an empty vector ( EV ) or a TGM2 cDNA resistant to TGM2 shRNA wild-type ( WT ) , C277S , or R580A mutants ( denoted as TGM2 WTR , TGM2 C277SR , or TGM2 R580AR cDNA ) by retroviruses . Quantification shows average colony number ± SD in biological triplicates . ( **p<0 . 01 , *p<0 . 05 , ns: not significant , student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 02310 . 7554/eLife . 07101 . 024Figure 6—figure supplement 1 . Pictures of soft agar assay for Figure 6A . Representative MTT-stained colonies corresponding to Figure 6A are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 02410 . 7554/eLife . 07101 . 025Figure 6—figure supplement 2 . Protein expression for Figure 6A . Western blot analysis of TGM2 , TP53 , and CDKN1A corresponding to Figure 6A are shown . β-ACTIN serves as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 02510 . 7554/eLife . 07101 . 026Figure 6—figure supplement 3 . Pictures of soft agar assay for Figure 6C . Representative MTT-stained colonies corresponding to Figure 6C are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 02610 . 7554/eLife . 07101 . 027Figure 6—figure supplement 4 . Protein expression for Figure 6C . Western blot analysis corresponding to Figure 6C are shown . β-ACTIN serves as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 02710 . 7554/eLife . 07101 . 028Figure 6—figure supplement 5 . Protein expression of BECN1 , ATG5 , ATG12 , and LC-3 . Western blot analysis of HMECTERT/ST/ER-RasV12 cells expressing either control , ATG5 , BECN1 , or ATG12 shRNA . β-ACTIN serves as the loading control . Knockdown of ATG5 , BECN1 and ATG12 resulted in the accumulation of LC3-I and decrease in the ratio of LC3-II / LC3-I . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 02810 . 7554/eLife . 07101 . 029Figure 6—figure supplement 6 . Pictures of soft agar assay for Figure 6D . Representative MTT-stained colonies corresponding to Figure 6D are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 02910 . 7554/eLife . 07101 . 030Figure 6—figure supplement 7 . Protein expression for Figure 6D . Western blot corresponding to Figure 6D . β-ACTIN serves as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 03010 . 7554/eLife . 07101 . 031Figure 6—figure supplement 8 . Pictures of soft agar assay for Figure 6E . Representative MTT-stained colonies corresponding to Figure 6E are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 03110 . 7554/eLife . 07101 . 032Figure 6—figure supplement 9 . Protein expression for Figure 6E . Western blot analysis corresponding to Figure 6E are shown . β-ACTIN serves as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 03210 . 7554/eLife . 07101 . 033Figure 6—figure supplement 10 . Pictures of soft agar assay for Figure 6F . Representative MTT-stained colonies corresponding to Figure 6F are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 03310 . 7554/eLife . 07101 . 034Figure 6—figure supplement 11 . Protein expression for Figure 6F . Western blot corresponding to Figure 6F . β-ACTIN serves as the loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 03410 . 7554/eLife . 07101 . 035Figure 6—figure supplement 12 . Pictures of soft agar assay for Figure 6H . Representative MTT-stained colonies corresponding to Figure 6H are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 03510 . 7554/eLife . 07101 . 036Figure 6—figure supplement 13 . Protein expression for Figure 6H . Western blot analysis of TGM2 expression corresponding to Figure 6H are shown . β-ACTIN serves as the loading control . Note the comparable expression levels of endogenous TGM2 , and the shRNA-resistant wild-type and mutant TGM2 cDNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 036 Next , we investigated whether combined loss of TGM2 and CDKN1A would enhance tumorigenesis of HMECTERT/ST cells expressing constitutively active HRASV12 ( denoted as HMECTERT/ST/RasV12 cells ) in a xenograft model . Subcutaneous injection of HMECTERT/ST/RasV12 cells expressing control shRNAs into immunocompromised NOD/SCID mice did not lead to tumors after 4 weeks , and formed only a few , small tumors after 6 weeks ( Figure 6B ) , which is consistent with the small number of colonies observed in soft agar ( Figure 6A ) . In contrast , HMECTERT/ST/RasV12 cells in which TP53 expression was reduced , and more importantly in which both TGM2 and CDKN1A expression were simultaneously reduced , formed six tumors out of six subcutaneous injections in NOD/SCID mice after 4 weeks ( Figure 6B ) . HMECTERT/ST/RasV12 cells expressing shRNAs targeting either TGM2 or CDKN1A alone produced tumors with a reduced penetrance and increased latency compared to those expressing TP53 shRNA ( two tumors out of six injections after 4 weeks ) ( Figure 6B ) , consistent with their performance in the soft agar assay ( Figure 6A ) . Taken together , our data show that loss of both TGM2 and CDKN1A expression allows HMECs to overcome TP53-dependent tumor suppression in vitro and in vivo . We hypothesized that reduced TGM2 expression enabled colony formation in soft agar by interfering with TP53-dependent autophagy . Therefore , inhibiting autophagy by other means should phenocopy TGM2 knockdown and also synergize with CDKN1A knockdown to promote colony formation in soft agar as shown in Figure 6A . To test this hypothesis , we evaluated the role of ATG12 , which controls autophagosome formation ( Kimmelman , 2011 ) , in tumor suppression . We expressed ATG12 shRNA ( Lock et al . , 2011 ) in HMECTERT/ST/ER-RasV12 stably expressing control- or CDKN1A- shRNAs and analyzed colony formation in soft agar . Cells with reduced ATG12 expression produced significantly more colonies in soft agar than control cells , similar to CDKN1A knockdown , consistent with a role of autophagy in tumor suppression ( Qu et al . , 2003; Takamura et al . , 2011; Yue et al . , 2003 ) ( Figure 6C and Figure 6—figure supplement 3 ) . We observed that ATG12 knockdown cells have elevated CDKN1A protein expression ( Figure 6—figure supplement 4 , lane 1 and 2 ) , which may trigger the suppression of colony formation and could explain why loss of ATG12 did not generate as many colonies as loss of TGM2 ( Figure 6C ) . However , the simultaneous knockdown of ATG12 and CDKN1A led to substantially more colonies than a reduction in the expression of each gene individually ( Figure 6C , column 3 to 5 , and Figure 6—figure supplement 3 ) , indicating a synergistic effect between ATG12 and CDKN1A in preventing cell transformation , similar to the synergistic effect observed between TGM2 and CDKN1A ( Figure 6A ) . This synergy was comparable to the effect of loss of TP53 expression ( Figure 6C , column 5 and 6 ) . To further confirm that inhibiting autophagy phenocopies TGM2 knockdown in colony formation , we performed similar RNAi experiments to analyze two additional autophagy regulator genes , ATG5 and BECN1 ( known as Beclin 1 ) . ATG5 functions in autophagosome elongation ( Kimmelman , 2011 ) and BECN1 functions autophagosome nucleation at the initiation of autophagy ( Kimmelman , 2011 ) . Knockdown of ATG5 , BECN1 , or ATG12 resulted in the accumulation of LC3-I and a decrease in the ratio of LC3-II/LC3-I ( Figure 6—figure supplement 5 ) , consistent with a defect in the conversion of LC3-I to LC3-II and a block in the early stage of autophagy , as expected ( Liu et al . , 2012; Mizushima and Yoshimori , 2007; Otomo et al . , 2013; Papandreou et al . , 2008; Tang et al . , 2013; Thorburn et al . , 2014 ) . We found that cells with reduced ATG5 or BECN1 expression produced significantly more colonies in soft agar than control cells , recapitulating our observations with ATG12 knockdown cells ( Figure 6D , column 1 to 5 , and Figure 6—figure supplements 6 and 7 ) . Furthermore , the simultaneous knockdown of ATG5 or BECN1 with CDKN1A led to substantially more colonies than reduced expression of each gene individually ( Figure 6D ) . Indeed , the cells with the double knockdown produced the comparable number of colonies to TP53 knockdown cells ( Figure 6D , column 7 to 11 ) . These results suggest that the inhibition of autophagy , together with loss of CDKN1A expression , strongly promotes transformation in HMECs . In order to determine if loss of TGM2 stimulates colony formation primarily through inhibition of autophagy , we expressed TGM2 shRNA and ATG12 shRNA in HMECTERT/ST/ER-RasV12 cells . We observed no additional stimulation of colony formation by simultaneous loss of TGM2/ATG12 compared to single loss of TGM2 ( Figure 6E and Figure 6—figure supplements 8 and 9 ) . We also co-expressed TGM2 shRNA with ATG5 shRNA or BECN1 shRNA in HMECTERT/ST/ER-RasV12 cells . Similar to the double knockdown of TGM2/ATG12 , there was no significant additional stimulation of colony formation by simultaneous loss of TGM2/ATG5 or TGM2/BECN1 compared to single loss of TGM2 ( Figure 6F , column 6 to 10 , Figure 6—figure supplements 10 and 11 ) . These data suggest that knockdown of TGM2 promote colony formation possibly by inhibiting autophagy . Taken together , our data suggest that efficient autophagic flux through autophagic protein degradation and autolysosome clearance by TGM2 , together with cell cycle arrest by CDKN1A , are complementary barriers for tumor suppression in the TP53 pathway , and that simultaneous loss of these barriers is important for oncogenic transformation in HMECs . TGM2 is a multifunctional enzyme with two well-established activities: crosslinking as a transglutaminase and binding and hydrolyzing GTP as a GTPase ( Chhabra et al . , 2009; Lorand and Graham , 2003 ) . To investigate which of these functions are important to promote autophagic protein degradation and autolysosome clearance , we treated cells with the TGM2 inhibitors Z-DON and LDN 27219 . Z-DON is a peptide-based inhibitor that specifically inhibits the crosslinking activity of TGM2 ( McConoughey et al . , 2010 ) , whereas LDN 27219 inhibits the GTPase activity of TGM2 ( Case and Stein , 2007 ) . HMECTERT/ST/ER-RasV12 cells were deprived of growth supplements , with or without TGM2 inhibitors , and the levels of SQSTM1 and LC3-I and -II were assessed . Interestingly , treatment with LDN 27219 , but not Z-DON , led to a significant accumulation of SQSTM1 protein ( Figure 6G ) . Furthermore , LC3-II protein levels and the ratio of LC3-II/LC3-I were increased dramatically in LDN 27219 treated cells , and partially in Z-DON treated cells ( Figure 6G ) , suggesting that the GTPase activity of TGM2 is more important than its crosslinking activity for promoting autophagic protein degradation . We showed that the GTPase activity of TGM2 is important for promoting autophagic protein degradation ( Figure 6G ) . Therefore , we hypothesized that colony formation will be stimulated in soft agar if the GTPase activity of TGM2 is inhibited . Instead of using TGM2 inhibitors , which may not be effective in agar and may have a side effect by prolonged culture during colony formation , we evaluated TGM2 mutants that alter the critical amino acid residues for GTP binding ( R580A ) or for transamidation ( C277S ) ( Gundemir et al . , 2012; Kumar et al . , 2012; Kumar and Mehta , 2012 ) . We expressed wild-type or mutant TGM2 constructs in TGM2 knockdown cells . Consistent with Figure 1F , ectopic expression of wild-type TGM2 suppressed the colony formation in TGM2 knockdown cells ( Figure 6H , column 1 to 3 ) . Interestingly , the GTP binding site mutant ( R580A ) completely ablated the tumor suppressive effect of TGM2 in colony formation , whereas the transamidation site mutant ( C277S ) displayed only a slightly reduced tumor suppressive effect compared to wild-type TGM2 ( Figure 6H and Figure 6—figure supplement 12 ) . The comparable expression levels of endogenous TGM2 , and the shRNA-resistant wild-type and mutant TGM2 are shown in Figure 6—figure supplement 13 . These findings are consistent with the observation that the GTPase inhibitor prevented autophagic protein degradation ( Figure 6G ) . Thus , the GTPase activity of TGM2 is required to promote autophagy and suppress cell transformation of HMECTERT/ST/ER-RasV12 cells .
Our loss-of-function screen identified TGM2 as a putative tumor suppressor gene within the TP53 signaling pathway that prevents oncogenic transformation and tumor formation by primary HMECs expressing TERT , activated HRASV12 and SV40 small T antigen . The role of TGM2 in cancer is quite complex and remains poorly understood . TGM2 has been shown to induce apoptosis ( Fok and Mehta , 2007 ) or differentiation ( Liu et al . , 2007 ) , and inhibit angiogenesis ( Jones et al . , 2006 ) . Additionally , the TGM2 gene locus is epigenetically silenced via methylation in some breast tumors and gliomas ( Ai et al . , 2008; Dyer et al . , 2011 ) . In contrast , TGM2 is overexpressed in other types of tumors ( Iacobuzio-Donahue et al . , 2003; Jin et al . , 2012; Miyoshi et al . , 2010 ) . TGM2 was also reported to be upregulated in cancer cell lines by several important signaling pathways involved in tumor progression or metastasis , such as NFKB1/NF-κB , TGFB1/TGF-beta , RARA/RAR-alpha ( Ai et al . , 2012; Cao et al . , 2012; Jung et al . , 2007; Ranganathan et al . , 2007; Rebe et al . , 2009 ) , and upon genotoxic stress ( Caccamo et al . , 2012; Shin et al . , 2004 ) . Although the role of TGM2 in tumorigenesis is likely context-dependent , our data clearly reveal a tumor suppressive role of TGM2 in a variety of cell lines that represent an early step in transformation and carcinogenesis . We found that the expression of TGM2 is dependent on TP53 , and that TGM2 is a potential direct target gene of TP53 . Although there are several putative consensus TP53 binding motifs in the TGM2 promoter , we found that an 82 bp region which does not contain a predicted TP53 binding consensus motif is necessary and sufficient for TP53-mediated transactivation . Our ChIP analysis also showed that endogenous TP53 binds to this region in HMECs . Recent genome-wide approaches have revealed that around 10% of the validated TP53 responsive elements are novel sequences that are not clearly related to the canonical TP53 binding consensus ( Menendez et al . , 2009 ) , underscoring the complexity of the TP53 network ( Contente et al . , 2002; Jordan et al . , 2008; Menendez et al . , 2013; Tebaldi et al . , 2015 ) . The TP53 binding sequence in the TGM2 promoter could be one of these 10% which do not have a canonical consensus . Our database search did not identify other promising transcription factor candidates that bind to this region . Although TP53-mediated transactivation of a reporter construct containing this region was observed within 24 hr after transfection in TP53-null H1299 cells , induction of TGM2 mRNA by Nutlin-3a in HMECs and BJ cells required 48 hr . The distinct kinetics of TGM2 induction by TP53 in these different contexts may reflect differences in epigenetics , co-factors , repressors , or posttranscriptional modifications of TP53 , which remain to be elucidated . TGM2 is a pleiotropic enzyme with well-known transglutaminase and GTPase activities ( Lorand and Graham , 2003 ) . The transamidation activity of TGM2 has been implicated in apoptosis by interacting with BAX ( Rodolfo et al . , 2004 ) or cross-linking CASP3/Caspase 3 and RB1/pRB to inhibit apoptosis ( Boehm et al . , 2002; Oliverio et al . , 1997; Yamaguchi and Wang , 2006 ) in various cancer cell lines . On the other hand , the GTP/GDP binding but not the transamidation domain of TGM2 has been shown to function in the epithelial-to-mesenchymal transition in immortalized MCF10A cells ( Mann et al . , 2006 ) . Our data suggest that the GTPase function of TGM2 is required for autophagy and suppresses transformation of HMECTERT/ST/ER-RasV12 cells . It is possible that the cell type , TP53 status or cell culture conditions influence the biochemical activities of TGM2; for instance , high Ca2+ concentrations induce TGM2 transamidation activity but inhibit its GTPase activity ( Chhabra et al . , 2009; Lorand and Graham , 2003 ) . The predominant activity of TGM2 in specific contexts may determine whether it functions as a tumor-promoting or -suppressive protein ( Chhabra et al . , 2009 ) . Similarly , the role of autophagy in cancer is quite complex , and may have tumor suppressive or promoting effects depending on the model and stage of tumorigenesis . Therefore , its role in cancer must be determined for each context . Substantial evidence suggests that autophagy supports the survival of established tumors by providing nutrients under metabolic stress . Alternatively , autophagy can act as a tumor suppressor by enhancing the degradation of damaged proteins and organelles to maintain tissue homeostasis and genomic stability in normal cells or in the early stages of cancer development ( Green and Levine , 2014; Lorin et al . , 2013; Mizushima and Komatsu , 2011 ) . Tumor suppressive roles for autophagy were demonstrated in mice with Becn1/Beclin-1 heterozygosity , systemic mosaic Atg5 deletion or liver-specific deletion of Atg7 ( Green and Levine , 2014; Lorin et al . , 2013; Mizushima and Komatsu , 2011 ) . Consistent with a tumor suppressive role , we found that TGM2 promotes autophagy and prevents an early step of HMEC transformation , the acquisition of anchorage-independent growth . Our data suggest that TGM2 enhances autophagic protein degradation and autolysosome clearance , thereby promoting autophagic flux . Previous work described a role for TGM2 in autophagosome maturation ( D'Eletto et al . , 2009 ) . These findings suggest that TGM2 is an important regulator of autophagy . Although our HMEC transformation model suggests that TGM2-mediated autophagy suppresses early events during tumor initiation , the autophagic function of TGM2 may promote tumor progression by facilitating the survival of established tumors under nutrient stress . In fact , TGM2 is overexpressed in subset of tumors ( Iacobuzio-Donahue et al . , 2003; Jin et al . , 2012; Miyoshi et al . , 2010 ) . TGM2 has been also reported to stimulate epithelial-to-mesenchymal transition ( EMT ) and remodel the extracellular matrix ( Kotsakis and Griffin , 2007 ) , supporting a positive role of TGM2 in the later stages of tumorigenesis . In our model , TGM2 contributes to a TP53-induced autophagy program and suppress transformation; however , TP53 has diverse roles in autophagy . Nuclear TP53 promotes autophagy through many of its target genes , such as DRAM1 , C12orf5/TIGAR , DAPK1 , SESN2/SESTRIN2 , ULK1 , ULK2 , BBC3/PUMA , BAX , BAD , and BNIP3 ( Balaburski et al . , 2010; Berkers et al . , 2013; Itahana and Pervaiz , 2014; Levine and Abrams , 2008 ) . On the other hand , cytoplasmic TP53 and mutant TP53 inhibit autophagy ( Berkers et al . , 2013; Itahana and Pervaiz , 2014 ) . Therefore , the role of TP53 in autophagy must be determined in each model system and cell context . We observed that TP53 promotes autophagic flux and autophagosome formation , an early step of autophagy , in HMECs . However , these findings do not exclude a role for TP53 in later steps of autophagy through TGM2 . Interestingly , we found that TGM2 and CDKN1A provide complementary functional contributions to tumor suppression in the TP53 pathway . Furthermore , knockdown of core autophagy genes ( ATG12 , ATG5 , and BECN1 ) synergized with loss of CDKN1A but not with loss of TGM2 to induce colony formation . In addition , inhibition of the GTPase activity of TGM2 prevents autophagic protein degradation as well as colony formation , supporting the conclusion that TGM2 contributes to tumor suppression , at least in part , by promoting autophagy . Although TGM2 may also have non-autophagic functions to suppress the transformation , our study suggests that cell cycle arrest , mediated by CDKN1A , and autophagy , mediated by TGM2 , are two critical TP53-dependent tumor suppressive barriers that prevent oncogenic transformation of HMECs ( Figure 7 ) . 10 . 7554/eLife . 07101 . 037Figure 7 . A model of the tumor suppressive functions of TGM2 in HMECTERT/ST/ER-RasV12 cells . Stress , in this case the depletion of growth supplements , induces autophagy and TP53-dependent expression of TGM2 . TGM2 facilitates autophagic flux by promoting autophagic protein degradation and autolysosome clearance . Loss of TGM2 expression synergizes with loss of CDKN1A expression to promote malignant transformation of HMECs . Therefore , TGM2-mediated autophagy and CDKN1A-mediated cell cycle arrest are potentially two critical barriers in the TP53 pathway that prevent oncogenic transformation of HMECs . DOI: http://dx . doi . org/10 . 7554/eLife . 07101 . 037 Knockdown of multiple genes by shRNAs can potentially lead to a synergistic effect , even if genes work in the same pathway , due to the incomplete loss of transcripts . For example , we observed that the combined knockdown of TP53 and TGM2 , using TGM2#1 shRNA , produced a greater number of colonies compared to TP53 knockdown alone . However , we do not exclude the possibility of TP53-independent functions of TGM2 in suppressing colony formation . Further work will be necessary to elucidate other tumor suppressive functions and regulation of TGM2 . The canonical functions of TP53 are the induction of cell cycle arrest , senescence , and apoptosis upon cellular stress . However , recent evidence challenges this long held view of TP53-mediated tumor suppression and highlight the importance of non-canonical , diverse functions for TP53 such as in autophagy ( Bieging and Attardi , 2012; Brady et al . , 2011; Li et al . , 2012; Valente et al . , 2013 ) . In this manuscript , we provide evidence that TGM2 suppresses an early event in tumorigenesis , anchorage-independent growth , and participates in TP53-induced autophagy which can collaborate with CDKN1A-mediated cell cycle arrest , the canonical tumor suppressive function of TP53 ( Figure 7 ) . We showed that TGM2 is a potential direct target gene of TP53 and revealed a role of TGM2 in suppressing colony formation by promoting autophagic flux through autophagic protein degradation and autolysosome clearance . These findings are consistent with a recent report showing that induction of autophagy is part of the TP53 tumor suppressive response during early tumorigenesis ( Kenzelmann Broz et al . , 2013 ) . Therefore , inhibition of autophagy as a therapeutic strategy for cancer may have unintended , tumorigenic effects in cases where autophagy is a critical part of the early tumor suppressive response . Weakening the TP53-dependent tumor suppressive barrier by inhibiting autophagy may allow early lesions with low level oncogenic signaling to progress to more aggressive lesions ( Junttila et al . , 2010 ) . This caveat needs to be considered before autophagy inhibition is used for cancer therapy in the clinic .
The procedure to select gene candidates for new TP53 pathway components was previously described ( Drost et al . , 2010 ) . Briefly , the candidates were selected from the Miller breast expression array ( GSE3494 ) consisting of 251 breast cancer samples with TP53 mutation status ( Miller et al . , 2005 ) . Normalized mRNA signals from gene probes were arranged as lowest = 0 , highest = 1 . By comparing TP53 wild-type tumors with TP53 mutant samples , we selected 122 gene candidates with significant downregulation ( p<0 . 01 ) in gene signal in a subset of TP53 wild-type tumors using Chi-square ( χ2 ) analysis . The signal level cut-off is indicated at average minus one standard deviation ( horizontal line ) . For each gene , we constructed a shRNA using the RNAi consortium library database ( http://www . broadinstitute . org/rnai/trc/lib ) , retrovirally transduced it into HMECTERT/ST/ER-RasV12 cells , and evaluated the effect of knockdown on colony formation in soft agar . Antibodies were obtained from: GeneTex , Inc ( TGM2 , #GTX111702 ) , Cell Signaling ( CDKN1A , #2947; LC3B , #2775; ATG12 , #D88H11; ATG5 , #D5F5U; BECN1 , #D40C5 ) , Santa Cruz Biotechnology ( TP53 , DO-1 , #sc-261 ) , Millipore ( β-actin , #MAB1501 ) , and Novus Biologicals ( SQSTM1/p62 , 2C11 , #H00008878-M01 ) . The RFP-GFP-LC3 plasmid was a gift from T . Yoshimori ( Osaka University , Osaka , Japan ) ( Kamimoto et al . , 2006 ) . Retroviral vectors expressing oncogenic SV40 small T antigen , HRASV12 , ER-HRASV12 , and TERT , and pRetroSuper vectors targeting TP53 , CDKN1A , and p16INK4a were described previously ( Mullenders et al . , 2009; Voorhoeve and Agami , 2003; Voorhoeve et al . , 2006 ) . The retroviral pRetroSuper vector expressing mCherry with various shRNA was constructed by replacing the blasticidin marker with mCherry . The retroviral vector expressing TERT-H2B-GFP was sub-cloned into pBabe vector to express TERT with H2B-GFP as selection marker ( Kolfschoten et al . , 2005 ) . pRetroSuper vectors targeting TGM2 ( TGM2#1; 5’-ACAGCAACCTTCTCATCGAGT , and TGM2#2; 5’-CCACCCACCATATTGTTTGAT ) , ATG12 ( 5’-TGTTGCAGCTTCCTACTTCAA-3’ ) , ATG5 ( ATG5#1; ATTCCATGAGTTTCCGATTGATGGC , and ATG5#2; CCTTTGGCCTAAGAAGAAA ) , BECN1 ( BECN1#1; GATACCGACTTGTTCCTTA , and BECN1#2; CTAAGGAGCTGCCGTTATA ) and mouse Tgm2 ( Tgm2#1; GCTGGACCAACAGGACAATGT , and Tgm2#2; GCGAGATGATCTGGAACTTCC ) ( Lock et al . , 2011 ) were driven by a U6 promoter . The double knockdown vectors of TGM2 and CDKN1A were created by cloning the CDKN1A shRNA expression cassette into the pRetroSuper shRNA vector targeting TGM2 ( hairpins #1 and #2 ) . The human TGM2 ORFs were cloned into miR-Vec expression vectors ( Voorhoeve et al . , 2006 ) . The TGM2R ORF contains five synonymous changes in nucleotides 931–939 of transcript variant 1 ( CTTCTCATC to TTGTTGATT ) . The human TP53 ORF was cloned into the pMSCV blast vector together with a 6x His-tag at the C-terminus . The TP53 construct designed to resist TP53 shRNA knockdown contains four synonymous changes in nucleotides 983–988 of transcript variant 1 ( AGTGGTAA to TCCGGAAA ) that preserve the amino acid sequence . TGM2 C277S and R580A mutants were generated by site-directed mutagenesis using TGM2 wild-type ORF as the template . All constructs were verified by sequencing . Primary human mammary epithelial cells ( HMECs ) ( Lot no . 7F3286 , #CC-2551 , Lonza ) were cultured in MEGM media supplemented with Bullet kit containing EGF , insulin , hydrocortisone , bovine pituitary extract , and GA-1000 ( gentamicin and amphotericin ) as recommended by the manufacturer , and transduced to express the ecotropic receptor and TERT-H2B-GFP as described ( Voorhoeve and Agami , 2003 ) , selected , and frozen down as early passages . Upon analysis , most of these cells showed a normal karyotype , and 3/13 cells analyzed had trisomy for Chromosome 20 . Fully transformed HMECs recovered as a colony from soft agar were analyzed by spectral karyotyping ( SKY ) and confirmed to have a normal karyotype with no translocations . HMECs that were retrovirally transduced to express TERT-H2B-GFP , SV40 small T , and ER-HRASV12 were referred to as HMECTERT/ST/ER-RasV12 cells . For deprivation of growth supplements , EGF , insulin , and hydrocortisone were removed from the medium for 24 hr after 3 days of treatment with 300 nM of 4-OHT ( #H6278 , Sigma ) . HEK293T cells , human foreskin fibroblast BJ cells and mouse embryonic fibroblast NIH 3T3 cells ( ATCC ) were grown in Dulbecco’s modified Eagle medium supplemented with 10% FCS and antibiotics . Early-passage wild-type and Tp53-/-mouse embryonic fibroblasts ( MEFs ) previously described ( Itahana et al . , 2007; Itahana and Zhang , 2008 ) were kindly provided by Dr . Yanping Zhang ( UNC , Chapel Hill , NC ) , and cultured similarly . All cells were maintained in a 5% CO2 incubator at 37°C . Nutlin-3a was purchased from Sigma ( #SML0580 ) for TP53 activation . For the treatment of TGM2 inhibitor , HMECTERT/ST/ER-RasV12 cells were incubated with 300 nM of 4-OHT together with Z-DON ( #Z006 , Zedira , 50 µM ) or LDN 27219 ( #4602 , Tocris Bioscience , 10 µM ) for 2 days , followed by the deprivation of growth supplements for 24 hr in the presence of the inhibitors . Ecotropic retroviruses were generated as previously described ( Brummelkamp et al . , 2002 ) . Briefly , cells at 60–70% confluence were transduced overnight in the presence of 8 µg/ml polybrene with ecotropic retroviruses ( Brummelkamp et al . , 2002 ) and selected with blasticidin ( 4 µg/ml ) , puromycin ( 2 µg/ml ) , or hygromycin ( 300 µg/mL ) 48 hr after transduction . HMECTERT/ST/ER-RasV12 cells were transfected with plasmids for 4 hr with JetPrime Polyplus ( Bioparc , France ) following the manufacturer’s instructions . TP53 knockout cells were generated by first transfecting HMECTERT/ST/ER-RasV12 cells with plasmids expressing Cas9 and a gRNA ( Addgene #41815 and #41824 ) ( Mali et al . , 2013 ) targeting exon 4 of TP53 ( GGCAGCTACGGTTTCCGTCT ) followed by plating in soft agar and picking of single clones after two weeks of incubation ( Ho et al . , 2013 ) . Single cell clones were expanded , followed by DNA extraction . DNA sequencing was performed to verify frame-shifting mutations in both alleles of TP53 ( Figure 4—figure supplement 1 ) . A bottom layer of 2 ml 1% agar was overlaid with a suspension of 20 , 000 HMECs in 2 ml 0 . 4% low gelling agarose ( #A9045 , Sigma ) in DMEM with 10% FCS . Agar was topped up with 1 ml DMEM containing 500 nM of 4-OHT for the first 3 days , and then replaced with DMEM containing EGF ( 5 ng/ml ) , insulin ( 5 μg/ml ) , hydrocortisone ( 500 ng/ml ) , and 4-OHT ( 500 nM ) . The top media was replaced after one week . After 2 weeks , colonies were stained with 500 μl of 5 mg/ml MTT ( #5655 , Sigma ) at 37°C for 1 hr and photographed with a stereomicroscope ( SZX16 , Olympus ) . Counting of colonies was automated using a MATLAB script based on the intensity and size of stained colonies . Statistical analysis of data was done by unpaired student’s two-sided t-test . For BJ cells , a bottom layer of 2 ml of 1% agar was overlaid with a suspension of 40 , 000 BJ cells in 2 ml of 0 . 4% low gelling agarose . Agar was topped up with 1 ml DMEM containing 500 nM of 4-OHT and 10% FCS for the first 1 week . After 1 week , the top medium was replaced with DMEM containing EGF ( 5 ng/ml ) , insulin ( 5 μg/ml ) , hydrocortisone ( 500 ng/ml ) , 4-OHT ( 500 nM ) , and 10% FCS . For NIH 3T3 cells , a bottom layer of 2 ml of 1% agar was overlaid with a suspension of 20 , 000 NIH 3T3 cells in 2 ml of 0 . 4% low gelling agarose . Agar was topped up with 1 ml DMEM containing 500 nM of 4-OHT and 10% FCS . The top media was refreshed after one week . After 2 weeks , colonies were stained with MTT and analyzed in the same way as HMEC colonies . RNA was extracted using Trizol ( Life Technologies ) and 1 μg of total RNA was used for cDNA synthesis ( iScript , Biorad ) . RT-qPCR was performed using SsoFast EvaGreen supermix ( Biorad ) on a CFX96 machine ( Biorad , CA , USA ) according to manufacturer's instructions . TBP or Gapdh were used as an internal control . Statistical analysis of data was done by unpaired student’s two-sided t-test . Specific qPCR primers ( 5’ to 3’ ) : TGM2 forward: ACTACAACTCGGCCCATGAC TGM2 reverse: TGGTCATCCACGACTCCAC TP53 forward: CAACAACACCAGCTCCTCTC TP53 reverse: CCTCATTCAGCTCTCGGAAC CDKN1A forward: GCAGACCAGCATGACAGATTT CDKNA1 reverse: GGATTAGGGCTTCCTCTTGGA MDM2 forward: GAATCTACAGGGACGCCATC MDM2 reverse: TCCTGATCCAACCAATCACC TBP forward: TCCTGTGCACACCATTTTCC TBP reverse: CGCCGAATATAATCCCAAGC Mouse Tgm2 forward: GGCCACTTCATCCTGCTCTA Mouse Tgm2 reverse: TCCAAGGCACACTCTTGATG Mouse Cdkn1A forward: CCTGGTGATGTCCGACCTG Mouse Cdkn1A reverse: CCATGAGCGCATCGCAATC Mouse Gapdh forward: AGGTCGGTGTGAACGGATTTG Mouse Gapdh reverse: TGTAGACCATGTAGTTGAGGTCA Cells were washed once with Phosphate Buffered Saline ( PBS ) and lysed with 2% SDS lysis buffer ( 2% SDS , 50 mM Tris-HCl [pH 6 . 8] , 10% glycerol ) . Protein concentration was determined with the BCA protein assay kit ( Thermo Scientific , MA , USA ) . Equal amounts of protein were separated by SDS-polyacrylamide gel electrophoresis and transferred to PDVF membranes . Membranes were blocked with 4% non-fat milk ( Biorad , CA , USA ) and incubated with the indicated antibodies . Detection of blots was done with Western Lightning Plus-ECL reagent ( PerkinElmer , MA , USA ) for antibody conjugated with HRP , or done with Odyssey Infrared Imaging system ( LI-COR Biosciences ) for fluorescent-labeled antibody . The TGM2 promoter from -5980 to -78 base pairs upstream of the translational start site ( ATG ) were amplified by PCR using the genomic DNA from HMECTERT/ST/ER-RasV12 cells as the template , and cloned into the pGL4 . 11-Luc reporter plasmid ( Promega ) . Deletions of this genomic fragment were also amplified by PCR and cloned into the same reporter plasmids . pGL4 . 11-Luc reporter plasmid containing TGM2 genomic fragments were transfected into H1299 cells ( TP53-null ) along with CMV-TP53 and pRL-CMV Plasmid . The luciferase and renilla luciferase activities were measured 24 hr after transfection using a dual-luciferase reporter assay system ( Promega ) ( Itahana et al . , 2015 ) . Renilla luciferase activity was used as an internal control to normalize transfection efficiency . HMECTERT/ST/ER-RasV12 cells were seeded onto µ-plates ( ibidi , Martinsried , Germany ) before transfecting with the RFP-GFP-LC3 construct . EGF , insulin , and hydrocortisone were removed 24 hr post-transfection to stimulate autophagy . After 24 hr , the autophagic vesicles labeled with RFP-GFP-LC3 were then acquired using a 561 nm and a 488 nm lasers on a confocal microscope ( Carl Zeiss LSM 710 , Jena , Germany ) equipped with oil-immersion objective lens ( NA 1 . 40 , 63x; Plan Apochromat , Carl Zeiss ) and ZEN 2010 software ( version 6 . 0 . 0 . 485; Carl Zeiss ) . The size of the vesicle was obtained from acquired 8-bit images of cells using Image J software ( version 1 . 45f; National Institutes of Health ) . Briefly , a minimum and a maximum threshold value of 0 and 42 , respectively , were applied to each single cell images . Background noise of 1 pixel was removed before applying the watershed function to obtain the outlines of the vesicles . The size of vesicle , in µm2 , was obtained using the analyze particles function with the following parameters: size range of 0 . 2 to 11 . 0 µm2 and circularity of 0 . 7 to 1 . 0 . Subsequently , the average area of each vesicle per cell was represented as box plot generated using the GraphPad Prism 5 software ( version 5 . 03 ) . Statistical analysis of data was done by unpaired student’s two-sided t-test . HMECTERT/ST/ER-RasV12 cells were seeded onto 4-chambered coverglass ( Lab-Tek Chambered Coverglass System ) ( Nalgene-Nunc , Rochester , NY , USA ) and incubated in medium without EGF , insulin , and hydrocortisone for 24 hr to stimulate autophagy . Samples were fixed in 2 . 5% glutaraldehyde in PBS at 4°C for 1 hr before osmication with 1% osmium tetroxide , pH7 . 4 for 1 hr . Subsequently the samples were dehydrated through an ascending series of ethanol at room temperature before infiltration with acetone and resin , followed by final embedding in resin which was allowed to polymerise at 60°C for 24 hr . Samples were cut by an ultra-microtome ( Leica ) , mounted on formvar-coated copper grids and stained with lead citrate . The grids were viewed in a JEOL JEM 1010 transmission electron microscopy ( TEM ) . Chromatin Immunoprecipitation assay was performed as previously described ( Itahana et al . , 2015 ) . Briefly , HMECTERT/ST/ER-RasV12 cells were fixed and lysed . The protein lysates were then sonicated and immunoprecipitated with mouse anti-p53 antibody ( Santa Cruz , DO-1 ) or mouse IgG as the negative control . The bound DNA in the immunocomplex was eluted and used as the template for PCR . PCR primers used are ( 5’ to 3’ ) ; TGM2 forward: TGGGCTAGTTGTGTGTCCCTGTCC TGM2 reverse: AGGCGGAGAGCGGCGCTAACTTAT CDKN1A forward: GTGGCTCTGATTGGCTTTCTG CDKNA1 reverse: CTGAAAACAGGCAGCCCAAG GAPDH forward: GTATTCCCCCAGGTTTACAT GAPDH reverse: TTCTGTCTTCCACTCACTCC HMECs expressing TERT , SV40 small T antigen , HRASV12 and the respective shRNAs were resuspended in 50% PBS and 50% matrigel ( #354248 , BD Falcon ) and 500 , 000 cells in 100 μl were injected in each flank of immunocompromised NOD/SCID female mice . Tumor growth was monitored every 3–4 days . Mice were sacrificed after 4–6 weeks of injection when tumors were noticeable and less than 2 cm in diameter . All work was done under an approved animal protocol ( IACUC#2013/SHS/815 ) . | Cancers grow from rogue cells that manage to defy the strict rules that normally stop a cell from dividing when it should not . Each cell contains many proteins that are responsible for implementing these rules , and thus help to prevent tumors from forming . One of these proteins – p53 ( which is also called TP53 ) – plays a central role in this process . Information about many processes within and around a cell filters through the p53 protein , before being passed on to a range of different proteins . The proteins that are alerted by p53 are commonly referred to as its 'downstream effectors' , and it is these proteins that stop cells from dividing too much . For example , the protein p21 ( also called CDKN1A ) – which is the best understood of p53’s downstream effectors – hinders the machinery that causes cells to divide . Other p53 effectors can cause cells to kill themselves to prevent cancer growth . However , recent experiments with mice predicted that there may be other p53’s effectors that are important too . Yeo , Itahana et al . have now depleted the proteins that potentially work in p53’s network , one by one , in human cells called mammary epithelial cells , to test if these cells can become cancerous in the laboratory . The experiments showed that another downstream effector protein of p53 – an enzyme called transglutaminase 2 – contributes to prevent these mammary epithelial cells from becoming cancerous . Transglutaminase 2 promotes a process known as autophagy , which recycles damaged and old components of the cell , and therefore normally helps to keep cells healthy . Yeo , Itahana et al . also demonstrated that the effects of both p21 and transglutaminase 2 are critical to stop human mammary epithelial cells grown in the laboratory from dividing too much and from forming tumors when injected into mice . These experiments provide a deeper understanding of how most cells manage to remain healthy rather than becoming cancerous and reveal a potential new target for the early detection of cancer . Further investigations could now explore whether therapies could re-activate this enzyme to prevent or treat cancer . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"cancer",
"biology"
] | 2016 | Transglutaminase 2 contributes to a TP53-induced autophagy program to prevent oncogenic transformation |
Nramp family transporters—expressed in organisms from bacteria to humans—enable uptake of essential divalent transition metals via an alternating-access mechanism that also involves proton transport . We present high-resolution structures of Deinococcus radiodurans ( Dra ) Nramp in multiple conformations to provide a thorough description of the Nramp transport cycle by identifying the key intramolecular rearrangements and changes to the metal coordination sphere . Strikingly , while metal transport requires cycling from outward- to inward-open states , efficient proton transport still occurs in outward-locked ( but not inward-locked ) DraNramp . We propose a model in which metal and proton enter the transporter via the same external pathway to the binding site , but follow separate routes to the cytoplasm , which could facilitate the co-transport of two cationic species . Our results illustrate the flexibility of the LeuT fold to support a broad range of substrate transport and conformational change mechanisms .
The amino acid-polyamine-organocation ( APC ) superfamily of secondary transporters encompasses a broad range of evolutionarily-related proteins that transport diverse substrates including neurotransmitters , metabolites , and transition metals in organisms throughout the tree of life ( Vastermark et al . , 2014; Wong et al . , 2012 ) . In humans alone , the APC superfamily encompasses 11 subfamilies of distinct solute carrier proteins ( Perland and Fredriksson , 2017 ) . These transporters harness the energy stored in preexisting transmembrane ion gradients . The LeuT fold ( Yamashita et al . , 2005 ) is the core structural unit that undergoes conformational rearrangements necessary for alternating access-based transport in the APC superfamily . This fold consists of ten transmembrane ( TM ) segments , divided into two pseudosymmetric , interlocking five-TM repeats , although many members have additional TMs . Primary substrates bind in a pocket formed by non-helical regions of TM1 and TM6 , close to the center of the membrane . Co-transported coupling ions—typically Na+ and/or H+—bind at the interface between two proposed domains ( Perez and Ziegler , 2013; Rudnick , 2013 ) : a ‘bundle’ formed by TMs 1 , 2 , 6 , and 7; and a ‘scaffold’ or ‘hash’ domain comprising most or all of the remaining six TMs ( Forrest and Rudnick , 2009 ) . When all substrates are bound , conformational rearrangement closes an external vestibule between ‘bundle’ and ‘scaffold’ and opens an intracellular vestibule between the two domains to allow substrate release ( Boudker and Verdon , 2010; Forrest et al . , 2011; Shi , 2013 ) . Despite the common fold , many APC members have little-to-no sequence identity , consistent with mechanistic divergences , including variance in the identity and stoichiometry of the coupled ions ( Ma et al . , 2012; Shaffer et al . , 2009 ) and in which helices move the most to open and close the inner and outer gates ( Kazmier et al . , 2017; Kazmier et al . , 2014a; Kazmier et al . , 2014b; Krishnamurthy and Gouaux , 2012; Malinauskaite et al . , 2014; Ressl et al . , 2009; Shimamura et al . , 2010; Weyand et al . , 2008 ) . Natural resistance-associated macrophage proteins ( Nramps ) are APC-superfamily transition metal transporters that enable uptake of rare micronutrients such as Mn2+ in plants and bacteria and Fe2+ in animals ( Cellier , 2012; Courville et al . , 2006; Nevo and Nelson , 2006 ) . Nramps bind and/or transport biologically-essential divalent metals such as Mn2+ , Fe2+ , Co2+ , Ni2+ , Cu2+ , Zn2+—and toxic metals like Cd2+ , Pb2+ , and Hg2+—but not the abundant alkaline earth metals Mg2+ and Ca2+ ( Bozzi et al . , 2016a; Ehrnstorfer et al . , 2014 ) . Metal uptake by Nramps is typically stimulated by acidic pH and accompanied by proton influx ( Chen et al . , 1999; Ehrnstorfer et al . , 2017; Gunshin et al . , 1997 ) . However , many homologs also display considerable proton uniport—proton transport in the absence of added metal that suggests loose , if any , coupling between the two substrates ( Chen et al . , 1999; Gunshin et al . , 1997; Mackenzie et al . , 2006; Nelson et al . , 2002; Xu et al . , 2004 ) . To date no studies have conclusively demonstrated that Nramp is in fact a thermodynamically coupled secondary transporter capable of harnessing a favorable gradient of metal or proton to power electrochemically-uphill transport of the other substrate . Nramps have 11 or 12 TMs , the first ten forming a LeuT fold , as seen in structures of three bacterial Nramp homologs ( Bozzi et al . , 2016b; Ehrnstorfer et al . , 2014; Ehrnstorfer et al . , 2017 ) , including our model system Deinococcus radiodurans ( Dra ) Nramp ( Bozzi et al . , 2016b ) . Conserved aspartate , asparagine , and methionine residues in TM1 and TM6 coordinate transition metal substrates as observed in an inward-open state ( Ehrnstorfer et al . , 2014 ) , while only a metal-free outward-open state has been reported ( Ehrnstorfer et al . , 2017 ) . Here , we provide the first complementary structures of the same Nramp homolog in multiple conformations , including the first metal-bound outward-open Nramp structure , and a novel inward-occluded structure . These allow us to fully illustrate the transport cycle for DraNramp . We also show that metal transport requires the expected alternating access bulk conformational change , whereas proton transport can occur via a more channel-like mechanism in the outward-open state . Using the structures and accompanying biochemical data , we delineate separate conserved transport pathways for metal and proton substrates and provide a mechanistic model encompassing substrate binding , release , and the conformational change process . We demonstrate novel modes of conformational rearrangement and ion shuttling in DraNramp compared to other LeuT-fold transporters , thus expanding the known repertoire of intramolecular dynamics and substrate transport mechanisms possible within this important protein family .
A previously determined structure of a Fab-bound DraNramp in an inward-open conformation revealed the intracellular metal permeation pathway , or vestibule , between TMs 1a , 2 , 5 , 6b , 7 , and 8 ( Bozzi et al . , 2016b ) . This structure was stabilized in an inward-open state by patches of mutations to intracellular loops 4–5 , 6–7 , and 10–11 , and we thus refer to it as the Patch mutant . To observe additional conformational states of a transport cycle in a single Nramp homolog at high resolution , we developed two complementary conformationally-locked constructs for crystallization . Adding steric bulk along TM1a—for example a G45R mutation , which mimics a human anemia-causing mutation of a conserved glycine ( Barrios et al . , 2012 ) —prevented the opening of the extracellular vestibule and eliminated metal transport , emphasizing the importance of the alternating-access mechanism to DraNramp function ( Bozzi et al . , 2016b ) . Based on these findings we pursued the G45R mutant as a new inward-locked crystallization construct . To develop a complementary outward-locked DraNramp construct , we adapted an approach previously described for the lactose transporter LacY ( Kumar et al . , 2014; Smirnova et al . , 2013 ) . By mapping extensive cysteine accessibility data onto the inward-open structure , we identified the external vestibule between TMs 1b , 6a , 3 , 8 , and 10 ( Bozzi et al . , 2016b ) . We created a panel of 11 tryptophan point mutants lining this predicted external vestibule ( Figure 1A ) to destabilize the inward-open state . An outward-locking mutation should severely impair metal transport , and indeed several mutants had impaired in vivo Co2+ uptake when expressed in Escherichia coli ( Figure 1B and Figure 1—figure supplement 1A ) . We chose to pursue G223W—on TM6a one helical turn above the unwound metal-binding region—which like G45R eliminated Co2+ and Fe2+ metal transport ( Figure 1—figure supplement 1B ) . Importantly , a tryptophan modeled in the inward-open DraNramp Patch mutant structure at position 223 clashes with the top of TM10 . In contrast , in the recent outward-open structure of Eremococcus coleocola ( Eco ) Nramp ( 33% identity with DraNramp ) the analogous glycine lines a wide aqueous channel with adequate room for tryptophan’s bulk ( Ehrnstorfer et al . , 2017 ) . To further validate this G223W construct , we measured bulk solvent accessibility of two single-cysteine reporters: A61C on TM1b , which is exposed only in DraNramp’s outward-open state ( Bozzi et al . , 2016b ) ; and A53C on TM1a just below the metal-binding D56 , a putative inward-open reporter based on comparing the Patch mutant and EcoNramp structures ( Figure 1C and Figure 1—figure supplement 1C ) . WT-like DraNramp ( with the indicated reporter cysteine and a C382S mutation to remove the lone endogenous cysteine ) maintains a dynamic conformational equilibrium—even in the absence of added metal substrate—such that either reporter can be fully modified by the thiol-specific N-ethylmaleimide ( NEM ) at high concentrations ( Figure 1D–E ) . G45R slightly increased A53C accessibility but fully protected A61C , indicating an outward-closed state that we will refer to as inward-locked based on these data and the structure described below . In contrast , G223W significantly increased A61C accessibility while fully protecting A53C , consistent with an outward-locked state ( Figure 1D–E and Figure 1—figure supplement 1D ) . We have thus identified two complementary constructs that trap DraNramp in outward-locked ( G223W ) and inward-locked ( G45R ) states ( Figure 2—figure supplement 1A ) . Using lipidic cubic phase ( LCP ) to mimic the hydrophobic membrane environment , we crystallized and determined the structures of G45R and G223W to resolutions of 3 . 0 and 2 . 4 Å , respectively , both significantly improved from our earlier DraNramp structure ( 3 . 94 Å ) ( Table 1 and Figure 2—figure supplements 1–2 ) . The new high-resolution structures also allowed us to re-refine our original structure , including correction of a sequence registry error in TM11 . Unexpectedly , the G45R structure is not in an inward-open conformation as seen previously with our Fab-bound Patch mutant ( Figure 2A ) but instead adopts an inward-occluded , metal-free state ( Figure 2B ) that may represent an intermediate conformation between inward-open and outward-open states in the DraNramp transport cycle ( Figure 2D ) . As in the inward-open apo state , the external vestibule remains sealed , with TM1b and TM6a forming tight hydrophobic packing with the tops of TM3 and TM10 , and most TMs undergo little apparent displacement ( Figure 2F–G ) . The major exception is TM1a , which swings ~45° to partially seal the inward aqueous cavity in the G45R structure , a motion we previously showed to be essential to the transport cycle ( Bozzi et al . , 2016b ) . The intracellular ends of TM4 and TM5 also move slightly compared to their position in the inward-open state , further sealing the metal-binding site from the cytosol . Comparisons of the G45R and G223W structures indicate that , rather than preventing inward motion of TM1a as we had hypothesized ( Bozzi et al . , 2016b ) , the G45R mutation precludes TM4-TM5 from fully closing the inner gate , as any bulkier replacement for that absolutely-conserved glycine in our outward-open G223W structure would clash with E176 on TM5 . Consequently , the intracellular vestibule to the metal-binding site is highly constricted yet there is no aqueous pathway to the binding site from the external side ( Figure 2B ) . Structural alignments with the inward-open Staphylococcus capitis ( Sca ) Nramp ( Ehrnstorfer et al . , 2014 ) and outward-open EcoNramp ( Ehrnstorfer et al . , 2017 ) also indicate an intermediate conformation for the G45R structure , albeit closer to the inward-open state ( Figure 2E and Figure 2—figure supplement 2D ) , confirming our assignment as inward-occluded . The G223W structure ( Figure 2C ) represents an outward-open , metal-bound state that superimposes best with the outward-open EcoNramp structure ( Figure 2E and Figure 2—figure supplement 2E ) . As predicted , the exogenous tryptophan lines a periplasmic-facing aqueous cavity leading to a bound Mn2+ in the center of the transporter , with close helix packing below precluding metal passage to the cytoplasm . We also determined a G223W apo structure ( Table 1 ) , which lacks electron density attributable to metal substrate in the binding site ( Figure 2—figure supplement 2C ) but is otherwise similar to the metal-bound state ( Cα RMSD = 1 . 08 Å ) ; hence we used the metal-bound structure for all further analyses . Compared to the inward-open and inward-occluded structures , in the outward-open state TM1b , TM6a , and the top of TM10 are splayed open , and loop 1–2 is displaced by ~4 Å , to form a wide aqueous pathway to the metal-binding site ( Figure 2H–I ) . On the cytoplasmic side , TM4 and TM5 move significantly ( by ~8 Å ) straddling TM8 and approaching TM1a , while TM1a also approaches TM8 to fully shut the interior aqueous vestibule ( Figure 2H–I ) . Based on overall superpositions of the three DraNramp structures , TMs 1 , 4 , 5 , 6 , and 10 show the largest displacements to switch metal-binding site accessibility ( Figure 2F–I ) . The remaining TMs ( 2 , 3 , 7 , 8 , 9 , and 11 ) would thus form a ‘scaffold , ’ which adjusts to accommodate the more significant movements of the other five TMs ( Video 1 ) . To more objectively compare the intramolecular rearrangements that occur during the transport cycle , we calculated difference distance matrices ( Richards and Kundrot , 1988 ) , averaged by TM , for each pair of structures ( Figure 3 ) . These matrices confirm that TMs 1 , 4 , 5 , 6 , and 10 undergo the most significant displacements relative to the rest of the protein between the different structures . But rather than moving as a rigid body such as proposed in the ‘rocking bundle’ model for LeuT ( Forrest and Rudnick , 2009 ) , these five TMs are also significantly displaced relative to each other . Like other LeuT-family members , DraNramp relies on unwound regions of its TMs to bind substrates . A conserved DPGN sequence is non-helical in TM1—with the helix-breaking proline-glycine pair separating two metal-binding residues—while a conserved MPH sequence that includes the metal-binding methionine ends an unwound region in TM6 ( Figure 4A–C ) . We used these two canonical motifs to generate an alignment of 6878 Nramp sequences ( Figure 2—source data 1 ) and calculate the conservation of other important residues ( Figure 2—figure supplement 1C ) . Interestingly , a third proline , P386 ( 83% conserved ) , enables the top of TM10 to swing to open the metal-binding site from the periplasm , while T228 ( 80% ) on TM6 and N426 ( 99% ) on TM11 stabilize the extended unwound TM6 region in the G223W structure ( Figure 4—figure supplement 1A ) . The inward-open ScaNramp structure revealed a metal-binding site consisting of three conserved sidechains corresponding to D56 , N59 , and M230 in DraNramp , and a backbone carbonyl of A227 ( Figure 4B ) ( Ehrnstorfer et al . , 2014 ) . In outward-open DraNramp , a Mn2+ binds both D56 and M230 ( 2 . 9 and 3 . 0 Å ) , with N59 slightly further away ( 3 . 4 Å ) ( Figure 4A ) . The increased unwinding of TM6a displaces the A227 carbonyl too far ( 6 . 5 Å ) to coordinate the metal ( Figure 4—figure supplement 1A–B ) . Instead , the A53 carbonyl coordinates the Mn2+ ( 2 . 0 Å ) —our structure is thus the first to implicate this residue in the metal transport cycle . Interestingly , A53 and A227 are at analogous positions within the two inverted repeats of the LeuT fold . Two waters ( 2 . 7 and 2 . 8 Å ) complete a Mn2+-coordination sphere . While the resolution remains too low to definitively define the coordination geometry , the electron density is consistent with Mn2+ interacting with both D56 oxygens and thus seven total ligands—rare but not unprecedented for Mn2+ ( Barber-Zucker et al . , 2017; Glasfeld et al . , 2003 ) . An ordered water network expands into the external vestibule as part of the extended metal coordination sphere ( Figure 4—figure supplement 1A ) . A water is also tethered to the conserved H232 directly below the metal-binding M230 , perhaps poised to hydrate the cation upon conformational change . The inward-occluded G45R binding site contains no metal . The A53 carbonyl is farther from the other metal-binding residues than the A227 carbonyl ( Figure 4C ) . This is consistent with a model in which Nramp metal transport involves a switch of ligands , perhaps with the A53 and A227 carbonyls both coordinating the metal substrate in an as-yet-uncaptured intermediate state . In G45R , flexing of TM10 above its P386 pivot shifts Q378 ( 86% conserved , with another 11% of sequences , including HsNramp2 , having an N at this position ) ~5 Å to within hydrogen-bonding distance of metal-binding A227 and D56 , perhaps stabilizing the negative charge on a deprotonated D56 during the empty transporter’s return to outward-open . While Q378 does not bind the metal substrate in either outward DraNramp ( 7 . 9 Å ) or inward ScaNramp ( Ehrnstorfer et al . , 2014 ) ( 4 . 5 Å ) , its position in the G45R occluded intermediate suggests it may transiently bind during the transport process . Indeed , two independent molecular dynamics ( MD ) simulations of the inward-open ScaNramp showed a metal interaction with the Q378 oxygen ( Bozzi et al . , 2016a; Pujol-Giménez et al . , 2017 ) , and mutations at this position impaired metal transport in HsNramp2 ( Pujol-Giménez et al . , 2017 ) . To test the importance of the three conserved sidechains that coordinate Mn2+ ( D56 , N59 , and M230; Figure 4A–B ) and Q378 to metal transport , we purified a panel of mutants and reconstituted them into proteoliposomes ( Figure 4D ) . D56A and D56N eliminated Mn2+ and Cd2+ transport ( Figure 4E–F and Figure 4—figure supplement 1C–D ) , confirming the importance of D56 . N59A severely and N59D moderately reduced transport of both metals ( Figure 4E–F and Figure 4—figure supplement 1C–D ) . Both M230A and M230T transport both metal substrates ( Figure 4—figure supplement 1C–D ) , but with lower apparent affinity than WT ( Figure 4E–F ) . Consistent with our previous findings , removing M230—the lone sulfur-containing metal-binding residue—affects Cd2+ more than Mn2+ transport , reflecting the importance of the semi-covalent interaction Cd2+ can form with sulfur ( Bozzi et al . , 2016a ) . Lastly , Q378S and Q378N both preserved significant transport of both metals ( Figure 4—figure supplement 1C–D ) . However , these mutants increased the KM for Mn2+—but interestingly not Cd2+ ( Figure 4E–F ) . That the native glutamine is essential for efficient transport of the biological substrate Mn2+ but dispensable for Cd2+ uptake ( Figure 4—figure supplement 1E ) suggests that the two metals interact differently with their surrounding ligands during the transport process , corroborating the differential effects of M230 mutations . We reconstituted G45R and G223W into proteoliposomes to assess their metal and proton transport ( Figure 5A ) . Consistent with our in vivo findings , neither G45R nor G223W transported Mn2+ , Cd2+ , or Co2+ ( Figure 5B and Figure 5—figure supplement 1A–B ) . Surprisingly , in the presence of a favorable negative membrane potential established using K+ gradients and the K+-ionophore valinomycin , G223W enabled larger basal H+ influx than WT , while G45R had no H+ flux ( Figure 5C ) . Interestingly , while Mn2+ transport stimulated H+ uptake for WT , adding Mn2+ did not further stimulate either G45R or G223W ( Figure 5C ) . These results suggest that Nramp metal and proton transport can proceed via separate routes , with proton transport requiring only that the protein sample the outward-open state . To test this new hypothesis , we reconstituted the A53C and A61C mutants , with single cysteines located just below or above the metal-binding site respectively ( Figure 5—figure supplement 1D ) . Both retain significant metal transport ( Figure 5E , G and Figure 5—figure supplement 1E–F ) , and can be targeted with cysteine-specific modifiers to post-translationally add bulky and/or charged wedges to impede conformational change ( Figure 5D and Figure 5—figure supplement 1C ) . Charged , and thus membrane-impermeable , MTSET nearly eliminated metal transport by A61C , while uncharged NEM or MTSEA moderately impaired or did not affect transport , respectively ( Figure 5E and Figure 5—figure supplement 1E ) , a result consistent with our previous in vivo findings that adding steric bulk , but not formal charge , is tolerated at this position ( Bozzi et al . , 2016b ) . In addition , MTSET-treated A61C replicated G223W’s H+-transport behavior , with higher H+ uniport compared to unmodified transporter , but little stimulation by Mn2+ ( Figure 5F ) . In contrast MTSET had no effect on Mn2+ and Cd2+ transport by A53C , but membrane-permeable MTSEA and NEM both eliminated transport ( Figure 5G and Figure 5—figure supplement 1F ) . Regarding H+ uptake , MTSEA- and NEM-treated A53C resembled G45R , with no basal uniport or Mn2+ stimulation , while again unmodified and MTSET-treated A53C both behaved similarly to WT ( Figure 5H ) . These findings show that essentially all activity in proteoliposomes comes from outside-out DraNramp . MTSET , which should inhibit any inside-out A53C , did not affect transport ( Figure 5G and Figure 5—figure supplement 1F ) , indicating that inside-out transporters contribute negligibly to the total activity in this assay . Consistently , while MTSET treatment spares inside-out A61C from labeling , it nevertheless nearly eliminated metal transport ( Figure 5E and Figure 5—figure supplement 1E ) , further supporting the assertion that outside-out WT-like transporters provided most of the detected transport activity . To confirm that a mix of inside-out and outside-out transporters were indeed present in proteoliposomes , we assessed the susceptibility of DraNramp to thrombin cleavage at a naturally-occurring site ( Gallwitz et al . , 2012 ) in the protein’s non-conserved , disordered N-terminal region ( Figure 5—figure supplement 2A–C ) . While thrombin fully cleaved DraNramp in detergent , in proteoliposomes the cleaved protein population plateaued at ~50% ( Figure 5—figure supplement 2D–F ) . This cleaved portion likely corresponds to inside-out oriented protein with an exposed N-terminal region , with the remaining ~50% of protein therefore outside-out oriented with the N-terminus inside the liposome and thus protected from thrombin cleavage ( Tsai et al . , 2012; Tsai et al . , 2013 ) . In summary , these experiments with permanently-locked crystallization constructs or chemically-locked cysteine mutants demonstrated that while metal transport requires complete conformational cycling , proton transport does not require large-scale conformational change and can proceed through DraNramp’s outward-open state but not its inward-open state . In addition , metal transport through DraNramp is much more efficient in the outside-to-inside direction than in the inside-to-outside direction under the physiological-like conditions set up in our in vitro assay . Our in vitro results suggested that proton transport occurs via a pathway separate from the intracellular metal-release vestibule , which remains closed to bulk solvent in the proton-transporting G223W mutant . Below the metal-binding site begins a network of highly-conserved hydrophilic residues , including at least seven potentially protonatable sidechains , that leads from the metal-binding D56 through a tight corridor between TMs 3 , 4 , 8 , and 9 to the cytoplasm ( Figure 6A–C ) . In contrast to the external and intracellular vestibules proposed as metal entrance and release pathways , the helices and residues within this polar network undergo little rearrangement between the three DraNramp structures , except the intracellular end of TM4 ( Figures 2F–I and 3 ) . Highly-conserved residues surrounding the metal-binding D56 include H232 on TM6b ( 100% as defined ) and E134 ( TM3 , 98% ) —which have each been proposed as the Nramp proton transfer point ( Ehrnstorfer et al . , 2017; Pujol-Giménez et al . , 2017 ) . Across from E134 lies a conserved salt-bridge pair: D131 ( TM3 , 93% ) and R353 ( TM9 , 78% ) . Approximately 9 Å below , a second conserved salt-bridge , E124-R352 ( 94% and 87% ) , links the same two helices . This network could provide the route for proton uniport in the outward-open conformation . To assess whether these residues could be proton carriers , we calculated predicted pKa values for our outward-open and inward-occluded structures ( Figure 6D ) ( Dolinsky et al . , 2004 ) . Surprisingly , D56 is the only residue with a pKa in the ideal 6–7 range to facilitate proton exchange at a typical external pH . About 4 Å from D56 , E134’s high pKa ( ~11 . 5 ) indicates a near permanently-protonated state , while H232—4 Å below E134 and 7 Å from D56—has too low a pKa ( ~3 . 5 ) to easily protonate , as does H237 ( ~4 . 0 ) further down TM6b . While E134 and H232 have separately been suggested as the Nramp proton-binding site ( Ehrnstorfer et al . , 2017; Pujol-Giménez et al . , 2017 ) , our pKa predictions suggest otherwise , as maintaining a formal change , especially on a histidine , would not be favorable in the protein core . In addition , previous studies showed the analogous E-to-Q mutant in EcoNramp maintained WT-like proton transport ability ( Ehrnstorfer et al . , 2017 ) as did the analogous H-to-A mutant in rat Nramp2 ( Mackenzie et al . , 2006 ) , which argues against those two residues as essential transfer points if the Nramp family shares a common transport mechanism . Within the TM3-TM9 salt-bridge network , R352 and R353 are likely protonated and positively charged , while their respective partners E124 and D131 are likely deprotonated and negatively charged . The predicted pKa values of D131 and E124 indicate their amenability to protonation . Indeed , as D56’s predicted pKa drops to 3 . 2 with Mn2+ bound , D131 becomes the best candidate to receive a proton . We observed three distinct voltage-driven H+ transport phenotypes within a panel of mutants to highly-conserved residues ( Figure 6E and Figure 6—figure supplement 1 ) . First , removing either metal-binding residue N59 or M230 had little effect . Second , neutralizing any member of the D56-E134-H232-D131 network or the H237Q mutation drastically reduced H+ transport . Third , mutating any of E124 , R352 , R353—farthest from D56—increased H+ uniport across multiple voltages . Outward-reporter A61C accessibility ( Figure 1C ) is consistent with each mutant sampling the outward-open state needed for proton transport ( Figure 5 ) , ruling out a conformation-locking explanation for the loss-of-function mutants . While some mutations perturbed the transporter’s conformational preference , A61C remained at least somewhat accessible in all cases except H237Q ( Figure 6F ) . In summary , based on our H+ transport measurements , structure-calculated residue pKa values , and prior studies using mutants of other Nramp homologs , D56 is the likely initial protonation point , with E134 and H232 positioned to chaperone the proton transfer to D131 , while R352 , R353 , and E124 restrain this process ( Figure 6G ) .
We propose a structure-based model for conformation cycling in DraNramp ( Figure 7A–B ) . Starting from the outward-open state seen in our G223W structure ( Figure 7A , left panel ) , metal binding ( and perhaps resulting proton entry into its release pathway ) may trigger bulk conformational rearrangement ( see below for details ) . To close the external vestibule , TM6a , TM10 , and to a lesser extent TM1b move closer to each other above their respective non-helical hinge regions , with the TM6a movement propagated through the TM5-6 linker to reorient TM5 and thus begin to open the inner gate . From this transient occluded conformation similar to our G45R structure ( Figure 7A , middle panel ) , additional movement of TM4-TM5 allows TM1a to bend upward to fully open the inner gate , enabling solvent access to and release of the metal , as the protein achieves a state similar to the Patch mutant DraNramp structure ( Figure 7A , right panel ) ( Bozzi et al . , 2016b ) . Analogously , to return to the outward-open state and complete the transport cycle , TM1a swings in to reach a conformation similar to G45R , then TM4-TM5 fully close on TM1a to seal the cytoplasmic vestibule while TM1b , TM6a , and TM10 separate to open the external vestibule . Our in vitro assays showed that while DraNramp metal transport requires sampling of both outward- and inward-open states , proton uniport occurs in sterically outward-locked constructs ( Figures 1 and 5 ) . This supports a model where protons and metal travel through distinct pathways on the cytoplasmic side of the protein ( Figure 7C–D ) , such that proton uniport is a feature of DraNramp’s outward-open state , whereas metal transport requires bulk rearrangement . In contrast , both protons and metal likely enter through the same aqueous pathway , as inward-locked proteins do not transport either substrate . From structure-based pKa calculations ( Figure 6D ) , H+-transport data for a large panel of mutants to conserved protonatable residues in DraNramp ( Figure 6E ) , and previous mutagenesis studies with other Nramp homologs ( Ehrnstorfer et al . , 2017; Mackenzie et al . , 2006; Pujol-Giménez et al . , 2017 ) , we propose that proton uniport occurs via a network of conserved protonatable residues leading from D56 in the metal-binding site to D131 in a salt-bridge network between TMs 3 , 4 , 8 , and 9 . This proton pathway is accessible in the outward-open state , thus enabling the well-documented proton uniport ( Chen et al . , 1999; Gunshin et al . , 1997; Mackenzie et al . , 2006; Nelson et al . , 2002; Xu et al . , 2004 ) . The proton uniport—common to the general Nramp family—that occurs under physiological conditions wastes electrochemical energy by dissipating the transmembrane proton gradient without contributing to metal uptake . Considering the relatively low abundance and slow kinetics of Nramp transporters , this proton uniport property may be an evolutionarily-tolerated consequence of the transporter’s design , or it could instead confer an as-yet-undetermined functional advantage . The predicted protonation of D56 and subsequent transfer to D131 , mediated through an E134/H232-stabilized transition state , may however serve to restrain H+ entry . For metal-stimulated proton transport , Mn2+ binding likely stimulates proton transfer into the same salt-bridge network , perhaps by directly ejecting a proton from D56 in the metal-binding site . Indeed , in a separate study we show that neutralizing mutations to the same four residues that eliminated H+ uniport ( Figure 6E ) also eliminated ( D56 , D131 , H232 ) or severely reduced ( E134 ) H+ fluxes in the presence of added Mn2+ , with mutants to D131 , E134 , and H232 retaining significant metal transport despite a lack of proton transport ( Bozzi et al . , 2018 ) . However , the precise order of events for proton and metal transport , including whether it is indeed a thermodynamically coupled symport mechanism , remains undetermined , and additional transport mechanisms are possible . In our G223W structure , two water molecules coordinate Mn2+: one lies between the metal and A227’s carbonyl , the other the metal and Q378 ( Figures 4A and 7E ) . We propose that after Mn2+ binds to D56 , M230 , A53 , and N59 as in our G223W structure , the A227 carbonyl and Q378 both displace the two waters as the outer gate closes . DraNramp would thus reach a fully dehydrated metal-bound state not yet visualized but which may resemble our apo G45R inward-occluded structure ( Figure 7E ) . Next , as the inner gate opens , the A53 carbonyl would exchange with a nearby water—such as the one bound to H232 in our G223W structure—as would Q378 , to yield an inward-open metal-bound state similar to the ScaNramp structure ( Figures 4B and 7E ) ( Ehrnstorfer et al . , 2014 ) . In this conformation the Mn2+-coordination sphere would include four residues and two waters—analogous to the G223W structure—thus facilitating eventual metal release . The proposed transition from four to six to four Mn2+-coordinating residues could help preferentially stabilize the occluded transition state ( Shilton , 2015 ) through the free energy ( entropy-driven ) gains of releasing the two water ligands . Furthermore , the rearrangements needed to achieve the hypothetical intermediate six-residue Mn2+ coordination—the helical extension and inward movement of TM6a , and the toppling of TM10’s top half—also close the external vestibule , providing a potential mechanistic link between local metal-coordination changes and bulk conformational change . To return to the outward-open state , the transporter must pass through an apo-occluded state as seen in the G45R structure , in which the N59 and Q378 sidechains reorient to stabilize D56 in the absence of the divalent cation carried during the outward-to-inward transition ( Figures 4C and 7E ) . As the transporter reaches the outward-open state seen in EcoNramp ( Ehrnstorfer et al . , 2017 ) and our apo G223W structure , a protonation event at D56 may prime the binding site to receive another incoming metal ion . Future molecular dynamic simulations and/or experiments will be essential to test these predictions . The mechanism described above for the DraNramp transport cycle—developed from structures of the same Nramp homolog in three distinct conformations and supported by metal and proton transport data—differs significantly from those previously observed for other LeuT-fold transporters . Mhp1 , BetP , and ( to a lesser degree ) LeuT generally obey a ‘rocking bundle’ model in which the rigid-body movement of four TMs that contain the primary substrate binding site ( 1 , 2 , 6 , and 7 ) against the remaining TMs ( 3-5 , 8-10 ) leads to conformational change ( Forrest and Rudnick , 2009; Kazmier et al . , 2017; Shi , 2013 ) . In DraNramp TMs 4 , 5 , and 10 join TMs 1 and 6 to form the substrate-binding ‘mobile domain , ’ while TMs 2 and 7 join the remaining TMs as part of the scaffold . Furthermore , the mobile helices do not move as rigid bodies , as conserved helix-breaking motifs free TMs 1a , 6a , and the top of TM10 to move independent of TMs 1b , 6b and the bottom of TM10 . In contrast , the fully-helical TM5 wholly reorients , and may thus coordinate the opening and closing of the inner and outer gates , connecting TMs 1a , 4 , and 6b with TM6a ( Figures 2 and 7A–B ) . In comparison to other APC superfamily members , the large TM1a displacement in DraNramp most closely resembles its dramatic movement in LeuT ( Krishnamurthy and Gouaux , 2012 ) . Gating roles for TMs 5 and 10 have been ascribed for BetP , Mhp1 , and MhsT ( Malinauskaite et al . , 2014; Ressl et al . , 2009; Shimamura et al . , 2010 ) , although not as extensive as we propose here in DraNramp . Not surprisingly , the DraNramp conformational changes are most similar to those predicted by comparing structures of two other bacterial Nramp homologs in complementary conformations ( Ehrnstorfer et al . , 2014; Ehrnstorfer et al . , 2017 ) , suggesting conservation within the Nramp clade of the LeuT-fold family . Whereas the distinct conformational changes of DraNramp demonstrate the diverse repertoire of dynamics available to the LeuT-fold family , the most striking mechanistic differences between DraNramp and other structurally-studied LeuT-fold transporters concern the substrate transport routes . Most well-characterized members ( including LeuT , BetP , and Mhp1 ) are Na+-driven symporters of small organic molecules which have one or two Na+-binding sites ( Perez and Ziegler , 2013; Rudnick , 2013 ) . Sodium binding at the highly-conserved Na2 site connects the ‘bundle’ ( TM1 ) and ‘scaffold’ ( TM8 ) domains while also shifting the conformational equilibrium to favor the outward-open state ( Claxton et al . , 2010; Tavoulari et al . , 2016; Zhao et al . , 2011 ) . This Na2 site consists of hydroxyls from two consecutive serines/threonines on TM8 and four main-chain carbonyls ( one from TM8 , three from the unwound-region of TM1 ) ( Perez and Ziegler , 2013; Yamashita et al . , 2005 ) . Intriguingly , the analogous location in DraNramp also contains highly-conserved hydroxyl-providing TM8 residues S327 ( 92% ) and S328 ( 20% conserved , with another 74% as T ) , which may be remnants of the ancestral Na2 site conversion into a H+ site in the Nramp clade . This hypothetical evolutionary switch has precedent within the LeuT-fold family , as the proton-coupled amino acid transporter ApcT analogously uses a conserved TM5 lysine ( K158 ) , whose sidechain protrudes into the Na2 location , as its primary proton-binding site ( Shaffer et al . , 2009 ) . The sodium-to-proton switch may have evolved in Nramps to avoid simultaneously coordinating two metal cations ( Na+ coordination , like Mn2+ coordination , requires ~6 oriented ligands , whereas H+ binding requires a single sidechain ) . LeuT and other bacterial homologs also antiport a proton as they return to an outward-open state ( Kantcheva et al . , 2013; Zhao et al . , 2010; Zomot et al . , 2007 ) via a conserved glutamate ( E290 ) on TM7 ( Malinauskaite et al . , 2016 ) , analogous to the highly-conserved N275 ( 100% ) that lines DraNramp’s intracellular vestibule . Available structures and MD simulations suggest that proton symport in ApcT and antiport in LeuT likely occur through the bulk opening and closing of the same permeation pathways used by the primary substrates ( amino acids ) ( Krishnamurthy and Gouaux , 2012; Malinauskaite et al . , 2016; Shaffer et al . , 2009; Shi and Weinstein , 2010 ) . In contrast DraNramp does H+ uniport even when mutationally ( G223W ) or chemically ( A61C-MTSET ) precluded from opening the intracellular vestibule ( Figure 5 ) . We propose a proton route from D56 through D131 and into a conserved salt-bridge network between TMs 3 , 4 , 8 , and 9 ( Figure 6 ) , which remain relatively stationary during the conformational change process ( Figure 2 ) . Indeed , evolutionary analysis reveals that this polar network is unique to the Nramp clade of the LeuT-family ( Cellier , 2016 ) ; this region is mainly hydrophobic in both LeuT and ApcT ( Shaffer et al . , 2009; Yamashita et al . , 2005 ) . A parallel transport pathway for protons could alleviate the electrostatic problem of simultaneously stabilizing three added positive charges ( the proton and divalent metal cation ) in close proximity throughout a conformational change process , although other unrelated transporters are known to accommodate multiple positive charges within their binding sites during the transport cycle ( Vandenberg and Ryan , 2013 ) . The observed proton uniport in Nramp , requiring only subtle conformational rearrangements , is more reminiscent of H+ shuttling in the CLC family of Cl-/H+ antiporters ( Accardi and Miller , 2004; Accardi and Picollo , 2010; Basilio et al . , 2014; Miller , 2006 ) than the canonical Na+ transport seen in LeuT-family symporters . It remains to be demonstrated whether the observed Nramp metal and proton transport truly constitute symport . The DraNramp proton and metal transport mechanism we outline , where primary and driving substrates enter via a common permeation pathway but exit via separate routes to the cytoplasm—with H+ transfer perhaps triggering bulk conformational rearrangement needed for Mn2+ release to occur if the substrates are in fact coupled—is thus far unique to the Nramp clade within the APC superfamily . This new model for Nramp transport therefore illustrates the evolutionary flexibility and adaptability of the shared LeuT fold .
WT and mutant DraNramps were cloned in pET21-N8H as described ( Bozzi et al . , 2016b ) . All constructs were full-length , except the G223W crystallization construct was N-terminally truncated to residue 35; this deletion did not affect metal transport ( Bozzi et al . , 2016b ) . Mutations were made using the Quikchange mutagenesis protocol ( Stratagene ) and confirmed by DNA sequencing . Single-cysteine constructs also included the C382S mutation to remove the lone endogenous cysteine . The C41 ( DE3 ) E . coli strain was used for protein expression and in vivo assays . Six liters of DraNramp C41 ( DE3 ) cells were cultured as described ( Bozzi et al . , 2016b ) , pelleted and flash-frozen in liquid nitrogen . Proteins were purified at 4°C . Thawed cells were lysed by sonication in 40 mL load buffer ( 20 mM sodium phosphate , pH 7 . 5 , 55 mM imidazole , 500 mM NaCl , 10% ( v/v ) glycerol ) plus 1 mM PMSF , 1 mM benzamidine , and 0 . 3 mg/mL each DNAse I and lysozyme . Lysates were cleared for 20 min at 27 , 000 × g . Membranes were pelleted from supernatant at 230 , 000 × g for 70 min , homogenized in 65 mL load buffer and flash-frozen in liquid nitrogen . Thawed membranes were solubilized for 1 hr , adding 1 . 5% ( w/v ) n-dodecyl-β-D-maltopyranoside ( DDM ) , then spun at 140 , 000 × g for 35 min to pellet debris . Pre-equilibrated Ni-sepharose beads ( 3 mL; GE Healthcare ) were incubated with the supernatant for 1 hr , and washed with load buffer containing sequentially 0 . 03% DDM , 0 . 5% lauryl maltose neopentyl glycol ( LMNG ) , and 0 . 1% LMNG . Protein was eluted in 20 mM sodium phosphate , pH 7 . 5 , 450 mM imidazole , 500 mM NaCl , 10% ( v/v ) glycerol , 0 . 01% LMNG , concentrated to <0 . 5 mL in a 50 kDa cutoff centrifugal concentrator , and loaded onto a Superdex S200 10/300 ( GE Healthcare ) pre-equilibrated with SEC buffer ( 10 mM HEPES pH 7 . 5 , 150 mM NaCl , 0 . 003% LMNG ) . Peak fractions were pooled , concentrated to ~20–40 mg/mL , aliquoted , and flash-frozen in liquid nitrogen . Protein ( 10–15 μL ) was loaded into a 100 μL glass syringe attached to an LCP coupling device ( Formulatrix ) . A second 100 μL syringe containing 1 . 5 volumes of liquid monoolein ( T > 37°C ) was attached to the coupling device , and the two solutions were mixed for 100 cycles using an NT8 ( Formulatrix ) in LCP mixing mode at 5 mm/s . LCP boluses ( 50–100 nL ) and precipitant ( 600–1000 μL ) were dispensed into 96-well LCP glass plates and incubated at room temperature ( RT ) . Crystals of G223W ( 25 mg/mL ) with 2 . 5 mM MnCl2 in SEC buffer grown in 28% PEG400 , 5 mM MnCl2 , 100 mM MES pH 6 , 50 mM succinic acid pH 6 , 10 mM spermidine pH 7 , ( 10–30 μm square plates ) were harvested after nine days using mesh loops ( MiTeGen ) to scoop the bolus , then flash-frozen in liquid nitrogen . Similar G223W crystals in the apo state were obtained after seven days with 26% PEG400 , 100 mM MES pH 6 , 50 mM succinic acid pH 6 , 20 mM spermidine pH 7 . Crystals of the analogous NEM-modified G223C/C382S ∆N34 ( G223C retains close to WT-level metal transport before NEM-labeling ( Bozzi et al . , 2016b ) ) were obtained in the same condition . Crystals of G45R ( 22 mg/mL ) grown in 20% PEGMME 550 , 150 mM NaCl , 100 mM HEPES , pH 7 . 0 ( ~100 μm rods ) were harvested after ~10 days . Data were collected at Advanced Photon Source beamline 24ID-C . Crystals were located by grid scanning with a 70 μm beam at 70% transmission followed by focused grid scanning with a 10 μm beam at 100% transmission . Data wedges were typically collected from −30° to +30° in 0 . 2° increments using a 10 μm beam at 100% transmission . Two wedges from two crystals ( G45R ) , 23 wedges from ~15 crystals ( G223W with Mn2+ ) , or 39 wedges from ~20 crystals ( G223W apo ) were independently indexed and integrated then combined during scaling using HKL2000 ( Otwinowski and Minor , 1997 ) to obtain complete datasets . Structures were determined using software provided by SBGrid ( Morin et al . , 2013 ) . Initial phases were obtained by molecular replacement in PHASER ( McCoy et al . , 2007 ) using our first DraNramp structure ( PDB: 5KTE ) as a search model for G45R and an in-progress G45R model for G223W . Model building and refinement were iterated in Coot ( Emsley and Cowtan , 2004 ) and PHENIX ( Adams et al . , 2010 ) , respectively . For all structures , positional and B-factor refinement with TLS restraints were used throughout , with torsion angle and NCS restraints for G45R , and secondary structure restraints for the Patch mutant . G45R contains two protein molecules in the asymmetric unit—chain A with residues 45–167 and 174–436 and chain B with residues 44–168 and 175–436 ( RMSD 0 . 86 Å over 2899 atoms , 0 . 50 Å over all 386 Cαs ) —and six fully or partly modeled monoolein molecules . Chain A was used for figures and analyses . The G223W Mn2+-bound structure includes residues 39–436 , ten full or partial monooleins , one spermidine molecule , and two Mn2+ ions—one in the metal-binding site , one at a crystal-packing interface . The G223W apo structure includes residues 39–436 and six full or partial monooleins . The two G223W structures align with RMSD 1 . 41 Å over 3012 atoms , 1 . 08 Å over all 398 Cαs ) . The electron density for each TM is shown in Figure 2—figure supplement 2A for G45R and Figure 2—figure supplement 2B for G223W with Mn2+ , while Figure 2—figure supplement 2C shows the metal-binding site for both the Mn2+-bound and apo G223W structures . The inward-open Patch mutant structure was updated to correct the position of intracellular loop 10–11 and the registry of TM11 , and extend the N-termini of TMs 5 , 7 and 9 , the C-terminus of TM7 , and extracellular loop 7–8 , and improve the geometry of the Fab . The new model comprises residues 43–165 , 170–236 , 256–341 , 351–436 of DraNramp , 1–129 and 132–213 , and 1–213 of the Fab heavy and light chains , respectively , and three Os3+ ions . Metal uptake assays in E . coli were performed as described previously ( Bozzi et al . , 2016a ) . For each biological replicate reported in figure legends , a separate culture of transformed E . coli was grown and induced to express the exogenous Nramp construct . DraNramp constructs were cloned , expressed , and purified as described above , with the following changes: protein was purified from cell pellets in a single day , and washed/eluted from nickel beads in buffers with 0 . 03% DDM . Protein was concentrated to 2 . 5 mL and buffer-exchanged into 100 mM NaCl , 10 mM HEPES pH 7 . 5 , 0 . 1% n-Decyl-β-D-Maltopyranoside ( DM ) on a PD10 desalting column . Protein concentrations were normalized to 1 . 2 mg/mL and aliquots were flash frozen in liquid nitrogen . Single-cysteine constructs A53C and A61C were purified in the presence of 1 mM DTT . Adjusting the lipid composition ( Ehrnstorfer et al . , 2017 ) of a previous protocol ( Bozzi et al . , 2016a; Tsai et al . , 2014 ) , 75% w/w 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine ( POPE ) was mixed with 25% w/w 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol ( POPG ) in chloroform ( Avanti Polar Lipids ) , and then dried under nitrogen in a warm water bath , re-dissolved in pentane , and dried again . Lipids were resuspended at 20 mg/mL in 5 mM DM in KCl + NaCl/MOPS buffer ( typically ~90 mM KCl , 30 mM NaCl , 0 . 5 mM or 10 mM MOPS pH 7 ) . Protein was added at a 1:400 w/w ratio to lipid , and the mixture dialyzed at 4°C to remove the detergent in 10 kDa molecular weight cutoff dialysis cassettes against KCl + NaCl/MOPS buffer with 0 . 2 mM EDTA for 1 day , then with 0 . 1 mM EDTA for 1–3 days , then overnight at room temperature ( RT ) against KCl + NaCl/MOPS buffer . For A53C and A61C , 1 mM , and 0 . 5 mM DTT was included in the first two dialysis steps . Fluorescent dye ( either 1:49 v/v 5 mM Fura-2 pentapotassium salt or 1:66 v/v 10 mM 2' , 7'-bis ( carboxyethyl ) −5 ( 6 ) -Carboxyfluorescein ( BCECF ) in dimethyl sulfoxide ) was incorporated into proteoliposomes permeabilized by three freeze-thaw cycles in dry ice-ethanol and RT water baths ( and sometimes stored at −80°C after the third freeze ) . Proteoliposomes were extruded through a 400 nM filter to create uniform-sized vesicles , buffer-exchanged 1–2 times on a PD10 desalting column pre-equilibrated with NaCl/ or KCl/10 mM MOPS pH 7 buffer . Peak proteoliposome-containing fractions were pooled to remove unincorporated dye . Proteoliposomes loaded with either 100 μM Fura-2 or 150 μM BCECF were diluted into buffer containing appropriate [KCl] to establish the desired membrane potential ( Fitzgerald et al . , 2017; Uzdavinys et al . , 2017 ) and aliquoted into 96 well black clear-bottom plates . Following baseline fluorescence measurement , 5X metal ( 750 μM final concentration unless otherwise noted ) and valinomycin ( 100 nM final concentration ) were added . Stocks of 100 mM CdCl2 , MnCl2 , and Co ( NO3 ) 2 , as well as appropriate serial dilutions in water for concentration range experiments , were freshly diluted into appropriate NaCl or KCl buffer with pre-added valinomycin . To pre-modify cysteines ( A53C or A61C when applicable ) , liposomes were diluted into buffer ( 120 mM NaCl , 10 mM MOPS pH 7 ) containing 3 mM MTSET , 3 mM MTSEA , or 4 mM NEM , and incubated at least 30 min at RT before beginning transport assays . Metal transport was monitored by measuring Fura-2 fluorescence at λex = 340 and 380 nm , at λem = 510 nm . Proton transport was monitored by measuring BCECF fluorescence at λex = 450 and 490 nm , at λem = 535 nm . To calculate concentrations of imported metal , the Fura-2 340/380 ratio and an experimentally determined KD value ( Hinkle et al . , 1992 ) was used for Cd2+ as described previously ( Bozzi et al . , 2016a ) . For Mn2+ and Co2+ , the fraction of Fura-2 340 and 380 fluorescence quenched , normalized to maximum observed quenching , was used to estimate imported metal . For proton uptake , the BCECF 450/490 ratio was used to calculate internal pH , which along with the known total internal buffer ( 0 . 5 mM ) and dye ( 150 μM ) concentration was used to calculate net proton import via the Henderson-Hasselbalch equation . The effect of divalent cations on BCECF fluorescence under analogous conditions to the liposome assay was tested ( Figure 5—figure supplement 3 ) , which showed that Mn2+ , Co2+ , Zn2+ , Cd2+ , Ca2+ had no effect while Fe2+ had a slight effect but of a much smaller magnitude than adding an equivalent concentration of H+ or OH- . Initial rates were calculated in Excel and Michaelis-Menten parameters were fit using MATLAB . For each technical replicate reported in figure legends , a separate aliquot of dye-loaded proteoliposomes was diluted into the appropriate outside buffer , including cysteine modifiers if applicable , then fluorescence time course data were collected before and after the addition of valinomycin , metal substrate , and/or ionomycin . DraNramp proteoliposomes with a 1:200 w/w ratio of protein to lipid were formed as described above , in 90 mM KCl , 30 mM NaCl , 10 mM MOPS pH 7 , and extruded 19 times . Additional purified DraNramp was diluted to 0 . 1 mg/ml in the same buffer with 0 . 1% DM . A 1/16 vol of 250 mM Tris pH 8 . 25 was added to the proteoliposomes or detergent-solubilized protein to adjust the pH to 8 . 0 for optimal thrombin activity . After removing a 0 min aliquot , thrombin from human plasma ( EMD Biosciences ) was added to final concentrations of 2 . 5 or 10 U/mL , and timed aliquots were removed and quenched by adding excess PMSF ( ~3 mM ) and sample buffer . Samples were run on SDS-PAGE and stained with Coomassie . Band intensities corresponding to the full-length ( 48 . 2 kDa ) and thrombin-cleaved ( 42 . 9 kDa ) proteins were calculated using ImageJ64 , and the fraction of the protein in the lower band ( corrected for the molecular weights ) was determined . Cells grown as for the uptake assay were washed once in labeling buffer ( 100 mM Tris pH 7 . 0 , 60 mM NaCl , 10 mM KCl , 0 . 5 mM MgCl2 , 0 . 75 mM CaCl2 ) , resuspended at OD600 = 2 , and aliquoted 100 μL per well in a 96-well plate . A 1:1 NEM dilution series was prepared in labeling buffer at 8 mM; 100 μL of the appropriate 2X NEM solution was added to each well and incubated 15 min at RT . L-cysteine ( 10 μL of 200 mM ) was added to quench NEM . Cells were washed twice in labeling buffer , pelleted , resuspended in 30 μL lysis and denaturing buffer ( 6 M urea , 0 . 1% SDS , 100 mM Tris pH 7 ) with 0 . 5 mM DTT and incubated 1 hr at 37°C . The lysate ( 10 μL ) was mixed with 3 . 5 μL of 6 mM 5K-PEG maleimide ( Creative PEGWORKS ) in lysis and denaturing buffer , incubated 1 hr at 37°C , and the reaction terminated by adding sample buffer with β-mercaptoethanol . Protein was detected via SDS-PAGE and western blotting using an Alexa 647-conjugated anti-His-tag antibody ( QIAGEN ) and a Typhoon Imager ( GE Healthcare ) . ImageJ64 was used to determine the % modification as described ( Bozzi et al . , 2016b ) . For each biological replicate reported in figure legends , a separate culture of transformed E . coli was grown and induced to express the exogenous Nramp construct . An alignment of 9289 Nramp sequences was obtained from a HMMER ( Finn et al . , 2011 ) search using the DraNramp sequence with an E-value of 1 , filtered for sequences with just one domain , then filtered for sequences 400–600 residues long . Incomplete sequences and sequences lacking the canonical Nramp TM1 ‘DPGN’ and TM6 ‘MPH’ motifs were removed , yielding 6878 sequences . A seed of 92 diverse sequences were aligned using MUSCLE ( Edgar , 2004 ) , then HMMER was used with a gap threshold of 0 . 99 to create the final alignment ( Figure 2—source data 1 ) . Per-residue Cα RMSD values were calculated using the ColorByRMSD PyMOL script , and whole-structure Cα RMSD values using the PyMOL align command with cycles = 0 . To generate the distance difference matrices , the pairwise distances between all Cα atoms were calculated for each structure ( inward , inward-occluded , and outward ) . Then , for each combination of two conformations , a distance difference matrix was calculated by taking the difference between the distance matrices corresponding to each conformation . These distance-difference values were then averaged for each pair of TM helices ( after dividing TMs 1 , 6 , and 10 into ‘a’ and ‘b’ segments ) using an RMSD-like calculation to obtain a 14×14 matrix for each pair of conformations . The python code used to perform these calculations and generate the resulting plots is available at GitHub: https://github . com/GaudetLab/coarse-grained-DDMP ( copy archived at https://github . com/elifesciences-publications/coarse-grained-DDMP ) . pKa values were calculated using PROPKA ( Dolinsky et al . , 2004 ) with CHARMM forcefields . The accession number for the DraNramp crystal structures reported in this paper are G45R inward-occluded , PDB ID: 6C3I; G223W ∆N34 outward-open with Mn2+ , PDB ID: 6BU5; G223W ∆N34 outward-open apo , PDB ID: 6D91; revised inward open , PDB ID: 6D9W . The unprocessed diffraction images were deposited in the SBGrid Data Bank ( https://data . sbgrid . org/ ) with SBGDB ID: 567 ( G45R ) ; 564 and 576 ( G223W ∆N34 Mn2+-bound and apo respectively ) ; and 332 , 333 , and 334 ( inward-open ) . The raw biochemical data that support the findings of this study are available from the corresponding author upon reasonable request . | Cells use transport proteins embedded in their membrane to acquire many of the nutrients they need to survive and grow . Different transport proteins transport different nutrients; for example , the Nramp transporters move transition metal ions across cell membranes . Nramps are found in a wide range of organisms . Bacteria use them to acquire the metals they need during the course of an infection , and humans rely on Nramps to absorb iron from food . Nramps can also transport hydrogen ions ( known as protons ) . Understanding how the structure of an Nramp transporter changes as it transports metal ions and protons can help researchers to understand how it works . These structures can be studied using a technique called X-ray crystallography , which captures snapshots of the proteins at different stages of their task . Bozzi , Zimanyi et al . used X-ray crystallography to study the structures of an Nramp transporter from the bacterium Deinococcus radiodurans . The results reveal four of the shapes that the Nramp transporter takes on at different stages in its transport process , including the first structure to show an Nramp binding to a metal ion from the outside of the cell . Taken together , the structures suggest a new transport mechanism that has not been seen in previously studied transport proteins with similar structures . An unexpected feature of this mechanism is that Nramps transport metal ions and protons along different pathways . Studying the transport mechanisms used by Nramp transporters will help researchers to understand how cells maintain appropriate levels of metal ions , an important component of human health . The mechanisms of relatively few transport proteins are understood at a structural level , yet many share common origins and have shared characteristics . Understanding how Nramps work could therefore help us to understand how wider classes of transporters work as well . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2019 | Structures in multiple conformations reveal distinct transition metal and proton pathways in an Nramp transporter |
The reconstruction of ancient insect ectoparasitism is challenging , mostly because of the extreme scarcity of fossils with obvious ectoparasitic features such as sucking-piercing mouthparts and specialized attachment organs . Here we describe a bizarre fly larva ( Diptera ) , Qiyia jurassica gen . et sp . nov . , from the Jurassic of China , that represents a stem group of the tabanomorph family Athericidae . Q . jurassica exhibits adaptations to an aquatic habitat . More importantly , it preserves an unusual combination of features including a thoracic sucker with six radial ridges , unique in insects , piercing-sucking mouthparts for fluid feeding , and crocheted ventral prolegs with upward directed bristles for anchoring and movement while submerged . We demonstrate that Q . jurassica was an aquatic ectoparasitic insect , probably feeding on the blood of salamanders . The finding reveals an extreme morphological specialization of fly larvae , and broadens our understanding of the diversity of ectoparasitism in Mesozoic insects .
The early evolution of insect ectoparasites and their associations with hosts are poorly known ( Labandeira , 2002; Wappler et al . , 2004; Grimaldi and Engel , 2005 ) . Although several Mesozoic insects were regarded as putative ectoparasites , only giant fleas have been widely accepted as definite terrestrial ectoparasitic insects on dinosaurs , pterosaurs , or mammals ( Gao et al . , 2012 , 2013b; Huang et al . , 2012 , 2013 ) . Here we report on an aquatic ectoparasitic insect based on five well-preserved specimens from the Middle Jurassic Daohugou beds of China . These fossils are extremely rare among the approximately 300 , 000 fossil insects in the collections of the Nanjing Institute of Geology and Palaeontology and Shandong Tianyu Museum of Nature .
Order Diptera Linnaeus , 1758 Family Athericidae Stuckenberg , 1973 Qiyia jurassica gen . et sp . nov . Qiyia is from the Chinese ‘qiyi’ meaning bizarre; jurassica is a reference to the Jurassic age of the fossils . Holotype STMN65-1 . Paratypes STMN65-2 , NIGP156982 , NIGP156983 , NIGP156984 . All specimens are preserved as carbonaceous impressions on the surface of grey tuffaceous siltstone ( Wang et al . , 2013 ) . From the Middle Jurassic Daohugou beds ( approximately 165 million years old ) of Ningcheng County , Inner Mongolia , China ( Liu et al . , 2006 ) . Three thoracic segments fused , with a ventral sucker; two pairs of dorsal spines on abdominal segments 1–7; abdominal segments 1–6 with paired ventral prolegs bearing upward directed bristles and apical crochets; extended seventh proleg; two pairs of anal papillae; sclerotized terminal processes with stiff setae . Body elongate , 18–24 mm long ( Table 1 ) . Head greatly reduced and partly retractile into thorax ( Figure 1A , B ) ; antennae and eyes not visible ( Figure 1C ) ; a pair of sclerotized tentorial rods ( Figure 2B ) . Mandibles approximately 0 . 6 mm long , heavily sclerotized , sickle-shaped , oriented to move parallel to each other in vertical plane , with external groove on adoral surface extending whole length of mandible ( Figure 1E ) . Thoracic segment swollen , slightly narrower than abdomen ( Figure 2A ) . Sucker retractile , diameter about 2 mm , located ventrally on thoracic segment and consisting of a circular suction disc with central opening about one quarter of disc diameter; peripheral area of disc thin and flexible ( Figure 1D ) . Six robust , sclerotized ridges on sucker , radially arranged , covered by soft skin with small spines ( Figure 2D , E ) ; distal part of each ridge thickened , probably with three processes embedded in musculature ( Figure 2E ) . Three pairs of small spines with simple shafts on dorsolateral margins of thorax , two pairs on dorsolateral margins of abdominal segments 1–7 , and one pair on abdominal segment 8 ( Figure 2A ) . Abdomen with eight distinct segments , covered by many short setae . Abdominal segments 1–6 with a pair of cylindrical , ventral prolegs covered by stiff , upward directed bristles; each proleg nearly half width of body with two rows of six crochet hooks apically ( Figure 1F ) ; seventh proleg approximately three times longer than other prolegs with only three or four apical hooks; abdominal segment 8 with two pairs of slender , tapering anal papillae: first pair long , approximately quarter body length; second pair half the length of the first pair ( Figure 1A , B ) ; one pair of unsegmented , sclerotized terminal processes fringed with stiff setae , approximately one-tenth body length; each process with about 10 spiracles ( Figure 1G , Figure 2C ) . 10 . 7554/eLife . 02844 . 003Table 1 . Measurements of specimens of Qiyia jurassicaDOI: http://dx . doi . org/10 . 7554/eLife . 02844 . 003Holotype STMN65-1Paratype STMN65-2Paratype NIGP156982Paratype NIGP156983Paratype NIGP156984OrientationlaterallateraldorsaldorsallateralBody23 . 822 . 122 . 9∼2218 . 1Head∼1∼1∼1–0 . 8Thorax2 . 82 . 52 . 6∼2 . 52 . 3Thoracic sucker diameter2 . 01 . 9––1 . 6Ridge0 . 60 . 6––0 . 5Abdominal segments 1–7 ( average ) ∼2 . 3∼2 . 2∼2 . 3∼2 . 2∼1 . 9Prolegs 1–6 ( average ) ∼1 . 5∼1 . 5∼1 . 5∼1 . 5∼1 . 3Seventh proleg4 . 03 . 8––3 . 0First anal papilla6 . 16 . 0–∼64 . 8Second anal papilla3 . 73 . 2–––Terminal process2 . 92 . 73 . 02 . 72 . 3All measurements are in mm and lengths except where otherwise indicated . ∼: approximately; –: unknown . 10 . 7554/eLife . 02844 . 004Figure 1 . Qiyia jurassica from the Middle Jurassic epoch of Daohugou , China . ( A ) Holotype STMN65-1 . ( B ) Paratype STMN65-2 under alcohol ( horizontal mirror image ) . ( C ) Head capsule of paratype STMN65-2 . ( D ) Head and thorax of holotype STMN65-1 . ( E ) Enlargement and reconstruction of the mandible of holotype STMN65-1; note the longitudinal groove . ( F ) Fifth proleg of holotype STMN65-1; note stiff , upward directed bristles which are distinctly longer than setae on body . ( G ) Last abdominal segment of holotype STMN65-1 . ap , anal papilla; p , proleg; pr , process of ridge; tp , terminal process . ( Scale bars: 5 mm in A , B , 1 mm in D , F , G , and 0 . 5 mm in C , E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02844 . 00410 . 7554/eLife . 02844 . 005Figure 2 . Qiyia jurassica from the Middle Jurassic epoch of Daohugou , China . ( A ) Paratype NIGP156982 under alcohol . ( B ) Head and thorax of paratype NIGP156982; note the underlying thoracic sucker . ( C ) Terminal processes of paratype NIGP156982 . ( D ) Reconstruction of sucker . The sucker consists of a circular suction disc with central opening and thin peripheral area . Six robust , radially arranged ridges are covered by soft skin with small spines . ( E ) Head and thorax of paratype NIGP156984; note the deformed mandible . ap , anal papilla; p , proleg; pr , process of ridge; tp , terminal process; tr , tentorial rod . ( Scale bars: 5 mm in A , 1 mm in B , C , E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02844 . 005
Three specimens are laterally compressed ( STMN65-1 , STMN65-2 , NIGP156984 ) and two are dorsoventrally compressed ( NIGP156982 , NIGP156983 ) , thereby providing side and top views of the detailed morphology of the larva . Q . jurassica is attributed to the Tabanomorpha by the reduced and retractable head and sickle-shaped mandibles shifted into a vertical plane ( Yeates , 2002; Zloty et al . , 2005 ) . It possesses two noticeably plesiomorphic features: mandibles with external grooves ( Zloty et al . , 2005 ) and well-developed anal papillae ( Wichard et al . , 1999 ) , while sharing two potential synapomorphies with extant athericid larvae: paired prolegs with crochet hooks ( Yeates , 2002; Kerr , 2010 ) and long terminal processes fringed with setae ( Dobson , 2013 ) . This combination of primitive and derived features demonstrates that Q . jurassica is a stem lineage representative of the Athericidae ( water snipe flies ) , a family sister to the more familiar horse flies ( Tabanidae ) . The earliest known Athericidae and Tabanidae ( all represented by preserved adults ) are from the Early Cretaceous of southern England ( Mostovski et al . , 2003 ) . Our new fossils are the earliest record of athericid flies and extend the lineage back to the Middle Jurassic , an age which is consistent with predicted divergence times based on molecular studies ( estimated at the Early or Middle Jurassic ) ( Wiegmann et al . , 2011 ) . Q . jurassica displays adaptations to an aquatic habitat , much like extant Athericidae which are today aquatic predators in fast-flowing water ( as adults some athericids feed on mammalian or amphibian blood ) ( Mostovski et al . , 2003; Nagatomi and Stuckenberg , 2004 ) . The paired sclerotized terminal processes are morphologically comparable to the modifications of beetle urogomphi in the aquatic larvae of some families such as Dytiscidae ( Wichard et al . , 1999 ) . About 10 spiracles are present on each process of Q . jurassica ( Figure 1G , Figure 2C ) , confirming that they were used for breathing air , functionally similar to the unsclerotized ones of extant athericid larvae ( Nagatomi and Stuckenberg , 2004 ) . Q . jurassica also possesses two pairs of anal papillae which are useful for extracting dissolved oxygen from water in aquatic flies and also play an important part in salt absorption to maintain ionic concentrations in the body fluids ( Wichard et al . , 1999 ) . These organs are common in nematoceran larvae and in some lower brachyceran larvae , but are reduced in extant tabanomorphan larvae ( Wichard et al . , 1999; Dobson , 2013 ) . In the case of the fossil larva , their development implies a plesiomorphic condition . The most notable structure of these newly discovered fossils is the ridged thoracic sucker which is a unique evolutionary adaptation among holometabolous insects . The round sucker has six radial ridges which are considered to be highly modified thoracic legs ( Figure 2D ) . These six robust , sclerotized ridges could increase both the suction area and surface friction , thus providing more adhesion and increasing lateral stability whilst reducing slippage , like the radial grooves in modern octopus suckers ( Kier and Smith , 2002 ) and supporting ribs in man-made suction cups ( Monkman et al . , 2007 ) . The dense vestiture of small spines may be used for better anchoring on the corrugated skin of a salamander , like the sucker-ring teeth and knobs on squid suckers ( e . g . , Miserez et al . , 2009 ) . To our knowledge , among insect larvae , only extant blepharicerids ( Diptera ) have six well-developed suckers , but these are small and without ridges on the abdominal sternites . As blepharicerid larvae graze on periphyton on rocks , they use the suckers to adhere to the substrate in fast-flowing streams ( Frutiger , 2002 ) . However , the excellent preservation of our new fossils suggests that Q . jurassica did not travel long distances and , unlike crown group Athericidae , most probably lived in still water near to or in the Daohugou palaeolake , a low-energy preservation environment ( Wang et al . , 2013 ) . The thoracic sucker on Q . jurassica is strongly cephalad on the body so , when anchored to the substrate , it probably restricted the movement of the small , short head ( Figure 1D , Figure 2E ) , a condition that is clearly suitable for piercing and sucking ( Figure 3 ) . Suckers are widespread in aquatic ectoparasites such as leeches , fish lice , and lampreys ( Kearn , 2004 ) which require more suction power to avoid becoming dislodged; other aquatic ectoparasites without attachment organs embed themselves in skin or muscle , such as cyclopoid copepods ( anchor worms ) ( Kearn , 2004 ) . In addition to the sucker , the stiff , upward directed bristles and apical hooks on the prolegs ( Figure 1F ) are also specialized attachment structures . These morphological adaptations provide compelling evidence that Q . jurassica adhered to a host as an ectoparasite , providing further specialization for a dense , watery habitat . 10 . 7554/eLife . 02844 . 006Figure 3 . Reconstruction of Qiyia jurassica in lateral view . DOI: http://dx . doi . org/10 . 7554/eLife . 02844 . 00610 . 7554/eLife . 02844 . 007Figure 3—figure supplement 1 . Ecological restoration of Qiyia jurassica from the Middle Jurassic epoch of Daohugou , China . One larva is shown attached to the salamander . Larvae could be located on unexposed body zones , such as on the axil or behind the gill , where there are many blood vessels and the skin is thinner . DOI: http://dx . doi . org/10 . 7554/eLife . 02844 . 007 Bloodsucking is considered to have evolved independently at least 12 times in true flies ( Lukashevich and Mostovski , 2003; Wiegmann et al . , 2011 ) . It started with free-living scavengers or predators which subsequently became opportunistic feeders on vertebrates , such as the notorious Congo floor maggot ( Auchmeromyia ) that consumes the blood of sleeping humans ( Lehane , 2005 ) . Bloodsuckers are present as adults in three families of extant Tabanomorpha ( Nagatomi and Stuckenberg , 2004 ) . Although hitherto known larval Tabanomorpha are mainly predators , some larvae suck the body fluids of vertebrates such as anurans ( Jackman et al . , 1983 ) . Predatory fly larvae commonly have morphological and physiological adaptations ( such as efficient protein-digesting enzymes and salivary glands ) , facilitating the switch to bloodsucking ( Balashov , 1984; Lehane , 2005 ) . Q . jurassica has a pair of sickle-shaped mandibles with external grooves ( Figure 1E ) , which is a groundplan character of Tabanomorpha ( Wichard et al . , 1999; Yeates , 2002 ) , forming a channel when the left and right mandibles are occluded ( Zloty et al . , 2005 ) and used for sucking blood or other body fluids ( Marshall , 1981 ) . In the Daohugou deposits fish are completely absent but salamanders are extremely abundant ( several thousand specimens recovered to date ) ( Liu et al . , 2006 ) . The most common salamanders at Daohugou , Chunerpeton tianyiensis and Jeholotriton paradoxus , have body lengths of 500 mm and 150 mm , respectively ( Wang and Rose , 2005 ) . Both species display neotenic features and are fully aquatic in all stages of their life cycle ( Gao et al . , 2013a ) . Salamander skin is glabrous and thin , and could easily have been penetrated by the mandibles of a larva such as Q . jurassica . The Daohugou salamanders match Q . jurassica well in size as well as co-occurrence , suggesting a possible parasite-host relationship . Some extant fly larvae parasitize anurans by burrowing into the skin , including Calliphoridae , Sarcophagidae , and Chloropidae ( Hoskin and McCallum , 2007 ) , and sometimes cause substantial mortality in their hosts ( Bolek and Coggins , 2002 ) . Q . jurassica , however , could simply have been anchored to the salamander skin using its sucker and prolegs ( Figure 3—figure supplement 1 ) , in a similar manner to leeches and fish lice ( Kearn , 2004 ) . Despite a great taxonomic diversity of extant ectoparasitic insects ( Marshall , 1981 ) , previous definite Mesozoic records were confined to the terrestrial giant fleas from the Middle Jurassic and Early Cretaceous epochs ( Gao et al . , 2012 , 2013b; Huang et al . , 2012 ) . Q . jurassica , which is arguably the earliest known aquatic ectoparasitic insect , reveals an unexpected morphological specialization of fly larvae and highlights the diversity of ectoparasitism in the Mesozoic .
The electronic edition of this article conforms to the requirements of the amended International Code of Zoological Nomenclature , and hence the new names contained herein are available under that Code from the electronic edition of this article . This published work and the nomenclatural acts it contains have been registered in ZooBank , the online registration system for the ICZN . The ZooBank LSIDs ( Life Science Identifiers ) can be resolved and the associated information viewed through any standard web browser by appending the LSID to the prefix ‘http://zoobank . org/’ . The LSID for this publication is: urn:lsid:zoobank . org:pub: 99FE7164-CF29-4EAE-B7B2-40C727CAC4FA . The electronic edition of this work was published in a journal with an ISSN , and has been archived and is available from the following digital repositories: PubMed Central , CLOCKSS , Linyi University , Steinmann Institute at University of Bonn , and Nanjing Institute of Geology and Palaeontology ( CAS ) . Printed copies are deposited in six major publicly accessible libraries including Linyi University , Nanjing Institute of Geology and Palaeontology ( CAS ) , Steinmann Institute at University of Bonn , University of Kansas , Natural History Museum ( London ) , and Muséum National d’Histoire Naturelle in Paris . | Parasites have been exploiting other organisms for millions of years . However , little is known about ancient parasitic insects , as it is rare to find fossils that are preserved well enough for them to be identified as parasites . This is particularly true for ectoparasitic insects , which live on the skin of their hosts . As a result , the only widely accepted ectoparasitic insect from the Mesozoic era is the giant flea , which infested dinosaurs , pterosaurs or mammals . Now , Chen , Wang , Engel et al . have discovered a new genus and species of ancient aquatic fly , which may be the earliest currently known aquatic ectoparasitic insect . Named Qiyia jurassica—after the Chinese word for ‘bizarre’ and the Jurassic period when it lived—its larva has a combination of features that mark it out as a parasitic ancestor of modern water snipe flies . In addition , the well-preserved fossilised larvae used to identify Q . jurassica have some more unusual features . The mouth of Q . jurassica had a structure commonly found in ectoparasites , designed to pierce skin and suck blood . The larva also had several features that were particularly well-adapted for gripping the host animal while underwater . The prolegs—stumpy fleshy structures found on the abdomen—were covered in bristles that pointed upwards , anchoring the larva in place . Q . jurassica also had an unusual sucker on its thorax that would have provided a firm grip that held its head still during feeding . Although many modern aquatic ectoparasites—like leeches—have suckers , the Q . jurassica sucker may be unique amongst insect larvae , as it has six large ridges and is covered in spines . Both features may have provided extra grip . Chen , Wang , Engel et al . suggest that Q . jurassica feasted on the blood of salamanders , as many salamander fossils have been found in the same region . The larvae could have attached to unexposed areas of the salamander—behind the gills , for example—where feeding would also have been easier due to the rich supply of blood vessels , and the thinner , more easily pierced skin . The wide range of features found on Q . jurassica suggests that Mesozoic ectoparasitic insects were more diverse than previously thought . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"short",
"report",
"evolutionary",
"biology"
] | 2014 | Extreme adaptations for aquatic ectoparasitism in a Jurassic fly larva |
A key question in human genetics and evolutionary biology is how mutations in different genes combine to alter phenotypes . Efforts to systematically map genetic interactions have mostly made use of gene deletions . However , most genetic variation consists of point mutations of diverse and difficult to predict effects . Here , by developing a new sequencing-based protein interaction assay – deepPCA – we quantified the effects of >120 , 000 pairs of point mutations on the formation of the AP-1 transcription factor complex between the products of the FOS and JUN proto-oncogenes . Genetic interactions are abundant both in cis ( within one protein ) and trans ( between the two molecules ) and consist of two classes – interactions driven by thermodynamics that can be predicted using a three-parameter global model , and structural interactions between proximally located residues . These results reveal how physical interactions generate quantitatively predictable genetic interactions .
Mutations often have outcomes that change depending upon additional genetic variation carried by an individual , making their effects difficult to predict ( Lehner , 2011 ) . The unexpected outcomes obtained when two or more mutations are combined are referred to as genetic interactions or epistasis ( Phillips , 2008 ) . One approach that has been taken to better understand how mutations interact to alter phenotypes has been to systematically combine together gene deletions or representative hypomorphic alleles ( Baryshnikova et al . , 2013 ) . In budding yeast , this has been undertaken on a genomic scale , with the resulting network of interactions referred to as the ‘genetic landscape’ of a cell ( Costanzo et al . , 2010 , 2016; Tong et al . , 2004 ) . However , gene deletions are rare in nature – most genetic variation consists of point mutations not deletions or null alleles . Point mutations can have very diverse and difficult to predict effects ( Shendure and Akey , 2015 ) . These range from no consequence , through partial loss-of-function , to very strong effects or the creation of new functions . To date , however , there has been no systematic effort to map how point mutations in two genes combine together to alter biological functions . Protein-protein interactions ( PPIs ) represent the backbone of a cell’s functional organization . Mutations affecting PPIs lead to disease , to functional innovations , and hence are subject to selection ( Diss et al . , 2013 ) . It has long been appreciated that pairs of mutations in two physically interacting proteins can have non-additive outcomes ( Horovitz , 1996; Lehner , 2011 ) . However , to date , the effects of mutations on PPIs have only been quantified for deep mutant libraries of one protein in combination with a small number of targeted mutants in a physical interaction partner ( Aakre et al . , 2015; Araya et al . , 2012; Raman et al . , 2016 ) . A thorough understanding of the patterns of mutation outcome between interacting proteins requires a non-biased , systematic mutagenesis of both interacting proteins . Here , we present a high-throughput technique based on the protein fragment complementation assay ( Tarassov et al . , 2008; Diss et al . , 2017 ) ( PCA ) called deepPCA that quantifies how mutations of diverse individual effect combine to alter protein interactions . We used the assay to systematically and comprehensively determine the effects of combinations of mutations in the proto-oncogenes FOS and JUN on the formation of the AP-1 transcription factor complex ( Shaulian and Karin , 2002 ) . Fos and Jun interact through their leucine zipper domains that consist of five heptad repeats ( Figure 1A ) ; this interaction has been previously extensively investigated ( Mason et al . , 2006; Ransone et al . , 1989 ) . We first quantified the consequences of combining thousands of pairs of mutations in trans between the two proteins . We then compared these results to the effects of thousands of pairs of mutations in cis within one of the proteins ( Fos ) . The resulting dataset presents a global view of how hundreds of mutations of diverse individual effects in different genes combine to alter a biological function through two major mechanisms related to the thermodynamics of a PPI and the structural interactions between proximal residues .
To quantify how mutations of diverse individual effect combine to alter protein interactions , we developed deepPCA , a protein-protein interaction assay that uses PCA and deep sequencing to quantify thousands of protein-protein interactions in parallel in a single assay ( Figure 1B , see Materials and methods ) . deepPCA uses deep sequencing to quantify the effects on a PPI of thousands of combinations of point mutations within one or both physically interacting proteins . The method is inspired by deep mutation scanning experiments on individual proteins ( Fowler et al . , 2010; Fowler and Fields , 2014 ) and uses physical linkage on a plasmid to read out the frequency of each pair of mutations after a competitive selection for growth dependent on the physical interaction between two proteins ( Figure 1B , see Materials and methods ) . Briefly , the two proteins of interest are fused to complementary halves of a methotrexate-resistant variant of murine dihydrofolate reductase and expressed in yeast . If the two proteins interact , the two fragments complement each other and reform an active enzyme , allowing growth in the presence of methotrexate . PCA is highly quantitative because the growth rate is correlated to the abundance of the complementation complex ( Freschi et al . , 2013; Levy et al . , 2014; Schlecht et al . , 2012 ) so cells expressing strongly interacting variants of the two proteins will hence grow faster and be enriched in the population while cells expressing weakly interacting variants and variants that don’t interact will be depleted . These changes in frequency between the pre- and post-selection populations ( input and output , respectively ) are then quantified by paired-end deep sequencing . The final PPI score quantifies the strength of interaction relative to the wild-type protein ( Figure 1B ) . We used deepPCA to quantify the effects of systematically mutating the leucine zipper domains of FOS and JUN . We obtained reliable ( input reads > 10 and output reads > 0; Figure 1—figure supplement 1A , B; see Materials and methods ) measurements for 607 and 608 of the 608 ( 32 positions x 19 substitutions ) possible single amino acid ( aa ) changes within the targeted regions of Fos and Jun , respectively . PPI scores measured by deepPCA are highly reproducible between biological replicates ( mean Pearson correlation R = 0 . 95 between the three pairs of replicates , n = 108 , 840 mutation combinations , Figure 1C , Figure 1—figure supplement 1C and Supplementary file 1 ) and also with mutation effects tested individually ( R = 0 . 95 for 14 variants chosen randomly , Figure 1D ) . The PPI scores for single amino acid changes in both proteins show a bimodal distribution ( Figure 2—figure supplement 1A ) , with ~20% and 15% of substitutions severely detrimental for the interaction and significantly different from the wild-type ( PPI score ≤ 0 . 64 , FDR < 0 . 05 , one sample t-test against a mean of 1; Figure 2—figure supplement 1B ) . However , the individual substitutions altered the interaction across the entire dynamic range , with 25 and 10 aa changes in each protein strengthening the interaction ( PPI score > 1 . 04 , FDR < 0 . 05 , average SEM of these 35 variants = 0 . 0054; Figure 2—figure supplement 1C ) . Mutations in the hydrophobic core of the interaction interface ( heptad positions a and d ) are most detrimental , followed by mutations at salt-bridge positions ( positions e and g , Figure 2A–B ) . Mutations in the hydrophilic far side of the zipper ( positions b , c and f ) were generally of small effect ( Figure 2A–C ) . Changes in the physico-chemical properties of the amino acids ( hydrophobicity , charge , α-helical stability etc , see Supplementary file 2 ) provide good prediction of the mutation effects ( percentage of variance explained from 35% to 98% across Fos and Jun positions ) , with properties related to α-helical stability most informative for predicting single mutation outcomes ( Figure 2—figure supplement 1D and Supplementary file 2 ) . Identical substitutions in the same positions in Fos and Jun often had similar effects . For instance , mutations in one protein that disrupted the interaction ( PPI scores < 0 . 64 ) were also very likely to disrupt the interaction when made in the other protein ( odds ratio = 31 . 6 , p<2 . 2×10−16 , Fisher’s exact test; Figure 2D ) . Near neutral or strengthening mutations ( PPI scores > 0 . 96 ) in one protein were also more likely to have a similar effect in the other one ( odds ratio = 7 . 3 , p < 2 . 2×10−16 , Fisher’s exact test ) . However , a substantial number of substitutions had effects that differed between the two proteins ( n = 381 out of 581 , FDR < 0 . 05 , paired t-test between the three replicate measurements in Fos and Jun ) , underlining the importance the structural context in which they occur . These mutations are enriched in intermediate effects in one or both proteins ( odds ratio = 8 . 1 , p < 2 . 2×10−16 , Fisher’s exact test ) . The average PPI score per position was also generally conserved between the two proteins , but revealed positions asymmetrically involved in the interaction such as the salt bridge positions ( Figure 2E ) . Considering pairs of substitutions in the two proteins , we obtained data for 107 , 625 of the 369 , 664 possible double mutants ( input read count above 10 and output read count above 0 , Supplementary file 1 , see Materials and methods ) . The double mutant PPI scores also show a bimodal distribution , but with proportionally more severely detrimental ( ~26% ) and fewer near-neutral outcomes ( ~21% ) than for the single mutants ( Figure 3A ) . The outcome of the double mutations was well predicted by multiplying the PPI scores of the constitutive single mutants ( percentage of variance explained of 85–86% in all three replicates , Figure 3B ) , that is , by assuming no genetic interaction between mutations . We calculated a genetic interaction score for each double mutant as the difference between the observed and predicted PPI scores ( Supplementary file 1 ) . Negative and positive genetic interactions ( 16 , 394 and 11 , 653 cases , respectively , at a 20% FDR , one-sample t-test ) thus represent double mutants with lower or higher interaction strength than expected , respectively . The genetic interaction scores are well correlated between replicates with a distribution centered on zero and long tails of positive and negative scores ( Figure 3—figure supplement 1 ) . Thus , as observed in other systems ( Araya et al . , 2012; Olson et al . , 2014 ) , genetic interactions make an important contribution to the outcome of double mutations . The genetic interaction scores are , however , strongly dependent on the single mutant PPI scores ( Figure 3C ) . Combining two mutants that both moderately reduce PPI strength is likely to result in a negative genetic interaction ( Figure 3C ) . Positive genetic interactions are , however , generally detected between two mutations that greatly weaken the interaction and also often when combining strength-increasing and strength-decreasing mutations ( Figure 3C ) . To account for these trends , we considered the thermodynamics of a PPI , relating the concentration of the bound and total subunits to the free energy of a dimeric complex ( equation 9 in the Materials and methods ) . This model has only three free parameters that need to be fitted from the data , representing the total concentration of each protein and the background growth in the PCA selection ( see Materials and methods ) . In the model the changes in free energy ( ΔΔG , expressed in arbitrary units ) for the mutations are additive but there is a sigmoidal relationship between PPI scores and ΔΔGs ( Figure 3D ) . Fitting the three parameters from the data ( Figure 3—figure supplement 2A–B ) reveals that the model provides very good prediction of how mutations in the two proteins combine together ( percentage of variance explained of 89–90% in all three replicates , n = 107 , 618 mutation combinations , Figure 3E ) . The model also removes the systematic trend in the genetic interaction scores across mutation pairs with different individual effects ( Figure 3F–G ) . Indeed , because of the sigmoidal nature of the model , a single mutant that decreases ΔΔG will increase PPI scores to a lower extent in the wild-type context than when combined with a mutation that destabilized the complex because of the saturation effect caused by the plateau of the sigmoid . To investigate the remaining genetic interactions not accounted for by the thermodynamic model , we calculated residual genetic interaction scores as the difference between the observed double mutant PPI score and the thermodynamic model prediction . These new genetic interaction scores also correlate well amongst the three replicates , with a narrow peak centered on zero interaction and long tails of rare strong positive and negative genetic interactions ( Figure 3—figure supplement 2C ) . We observed more cases of strong negative ( 1711 , 1 . 6% ) than positive ( 883 , 0 . 82% ) genetic interactions ( absolute score > 0 . 1 , FDR = 0 . 2 , Figure 3—figure supplement 2D–G ) . These strong interactions are enriched between particular Fos and Jun residues ( Figure 4—figure supplement 1 ) , with positive genetic interactions concentrated between positions close in the sequence of heptad positions ( along the diagonal of the matrix in Figure 4A ) and negative genetic interactions more spread-out in the structure and less enriched between specific pairs of positions ( Figure 4A–B and Figure 4—figure supplement 2 ) . Both directions of interaction are enriched between residues at the interface of the PPI and between residues close in space ( Figure 4C–D and Figure 4—figure supplement 3 ) , with this stronger for positive than for negative interactions . Positive interactions are therefore particularly enriched between contacting residues , identifying ‘lock and key’ specificity residues ( Horovitz , 1996 ) . We refer to these interactions beyond the interactions predictable from the global thermodynamic model as structural genetic interaction . For instance , in the wild-type PPI , the Glu residue in position 3g of Fos establishes a salt-bridge with the Arg residue in position 4e of Jun ( Figure 4E ) . The individual mutations Glu3gLys and Arg4eGlu both destabilize the PPI by replacing the salt-bridge by repulsive electro-static interactions ( Glu-Arg replaced by Lys-Arg and Glu-Glu with average PPI scores of 0 . 84 and 0 . 71 , respectively ) . However , the two mutations compensate each other by recreating the salt-bridge ( Glu-Arg replaced by Lys-Glu ) and restoring a neutral PPI score of 0 . 98 . Additional examples are shown in Figure 4E . In addition to combining pairs of mutations in the two different proteins , we also quantified the effects of 17 , 688 double amino acid changes within Fos alone ( 99% of cis double mutant combinations reachable through combinations of single nucleotide changes; for all comparisons between the trans and cis libraries below , we only consider mutants reachable by single nucleotide changes in both libraries; Figure 5A , Figure 5—figure supplement 1A and Supplementary file 3 and 4 ) . The PPI scores for single mutants correlate very well between the two libraries ( R = 0 . 96 , Figure 5—figure supplement 1B ) , further validating the reproducibility of the deepPCA method . There is a good correlation between the PPI scores of cis and trans double mutants consisting of exactly the same pairs of substitutions at the same positions ( R = 0 . 77 , n = 5451 identical double mutants quantified in cis and trans , Figure 5B ) . This correlation indicates that the structural determinants of mutation effects in FOS and JUN remain well conserved despite sequence divergence over long evolutionary timescales . However , the distributions of double mutant effects are quite different for the cis and trans combinations ( Figure 5C ) . This could be either due to different levels of genetic interactions or merely to the combination of different distribution of single mutant effects ( FOS x FOS in cis and FOS x JUN in trans ) . To control for differences in the distributions of single mutant effects in the two libraries , we sub-sampled the libraries to match their single mutant effect distributions ( Figure 5—figure supplement 2A , see Materials and methods ) . This revealed that , even when controlling for single mutant effect sizes , two mutations within Fos are more likely to increase the strength of the PPI than one mutation in Fos combined with a second mutation in Jun ( p < 10−3 over 1000 sub-samplings , 5 . 2% vs 3 . 4% , respectively , for PPI scores > 1 . 04 , Figure 5—figure supplement 2B ) . Two mutations in Fos are slightly less likely to destroy the PPI than a trans mutation combination ( 25 . 5% vs 27 . 8% , respectively , p < 10−3 for PPI scores < 0 . 64 , Figure 5—figure supplement 2C ) but have slightly more intermediate negative effects ( p < 10−3 , 39 . 4% vs 35 . 3% , respectively , for PPI scores between 0 . 64 and 0 . 92 , Figure 5—figure supplement 2D–E ) . Whether mutations of the same individual effect sizes combine together in cis or in trans therefore influences the double mutant outcome . Because leucine zippers , including Fos and Jun , fold upon binding ( Patel et al . , 1990; Thompson et al . , 1993 ) , the same thermodynamic model based on a two-state equilibrium between the two unfolded proteins and the complex can describe how mutations combine in cis as well as in trans . We tested how well the thermodynamic model with parameters fitted on the trans double mutants predicted the cis library data and found that it gave very good prediction ( percentage of variance explained of 82–83% for cis vs . 90–91% for trans combinations , Figure 5—figure supplement 3A–B ) . Similarly , fitting the thermodynamic model on the cis double mutants gave very good prediction of the trans library data ( percentage of variance explained of 84% for cis and 90% for trans , Figure 5—figure supplement 3A–B ) . A common thermodynamic model therefore accounts very well for how mutations combine in both cis and trans to change the PPI ( Figure 5D and Figure 5—figure supplement 3C ) . Therefore , just as in trans- , cis-genetic interactions have a component that results from the non-linear relationship between free energy and protein complex concentration . We then tested whether the residual component of cis-genetic interactions are also enriched for structural interactions . The strongest cases of structural cis interactions ( absolute genetic interaction score > 0 . 1 and FDR < 0 . 2 in both libraries , Figure 5—figure supplement 4 ) are indeed also enriched between proximally located residues but are less restricted to pairs of positions that are both at the PPI interface and involve more far side positions compared to trans-genetic interactions ( Figure 5—figure supplement 5 ) . cis-genetic interactions are also less enriched at specific positions and more dispersed throughout the structure ( Figure 5—figure supplement 6 ) . These results are robust to the magnitude threshold used to call strong genetic interactions and to the differences in single mutant effects between the two libraries ( Figure 5—figure supplement 7 and Figure 5—figure supplement 8 ) . Thus , cis-genetic interactions can be subdivided into the same two components as trans-genetic interactions , genetic interactions that results from the non-linearity of the general relationship between protein complex concentration and free energy and specific structural interactions . Interestingly , structural genetic interactions explain more of the variance in double mutant PPI scores when mutations are combined in cis than in trans ( Figure 5D ) . Indeed , both positive and negative structural genetic interactions ( genetic interactions not accounted for by the thermodynamic model; Supplementary files 3 and 4 ) are more abundant in cis than in trans ( 1493 vs . 835 true cases of positive and 1319 vs . 1128 true cases of negative genetic interactions in cis and trans , respectively , at p < 0 . 031 , one sample t-test , FDR < 0 . 15 and 0 . 2; Figure 5E and Figure 5—figure supplement 9 ) . This higher prevalence of genetic interactions in cis can potentially be explained by a higher number of contacts between positions within FOS than across the interface ( 1040 cis vs . 518 trans mutants pairs at positions within 5 Å of each other ) , supporting our previous result that proximity is a major determinant of genetic interactions both in cis and trans in a PPI interface .
Here , we have presented a protein-protein interaction assay – deepPCA – that allowed us to quantify the effects of >120 , 000 combinations of mutations in both cis and trans on the physical interaction between the products of the FOS and JUN proto-oncogenes . This provided a comprehensive and systematic data for how a very large number of different mutations in two genes combine to alter a biological function , allowing us to investigate the causes of genetic interactions between mutated genes and the extent to which genetic interactions can be quantitatively predicted . In its current form , deepPCA is limited to small domains because of the limit in the amplicon size for paired-end sequencing . However , the use of barcodes ( Hiatt et al . , 2010 ) would allow the assay to be applied to longer proteins . Our data reveal that physical interactions in the cell generate two distinct types of genetic interaction: interactions due to the sigmoidal relationship between the concentration of a protein complex and the free energy of an interaction ( Figure 3 ) and specific , structural interactions ( Figure 4 ) . The general genetic interactions that arise in the physical interaction between molecules is one of several non-linear mappings that can occur between changes in genotype and changes in phenotype . Additional non-linearities occur in the folding of individual proteins or RNAs ( ‘threshold robustness’ ) ( Bershtein et al . , 2006; Olson et al . , 2014; Tokuriki and Tawfik , 2009 ) , in saturating enzyme flux ( Kacser and Burns , 1973; Stiffler et al . , 2015 ) , and in regulatory dynamics ( Gjuvsland et al . , 2007; Omholt et al . , 2000 ) . This type of genetic interaction is cumulative and easily predictable , for example a three parameter thermodynamic model accounts for ~90% of the variance in our dataset of >120 , 000 genotypes . The magnitude of this type of genetic interaction can also be predicted when combining three or more mutations together . A better knowledge of all the sources of non-linearities between the genotype and the phenotype is therefore critical to model how genotypic variation translates into phenotypic changes . The second type of genetic interactions generated by molecular interactions is thermodynamically non-additive . These interactions are enriched between physically contacting and proximal residues , but can also involve some long-range indirect interactions ( Halabi et al . , 2009 ) . Structural genetic interactions have a more complex basis and are therefore more difficult to predict . Gathering comprehensive data similar to that described here for additional PPIs will help to further elucidate the structural determinants of genetic interactions and the rules for predicting them . This second type of genetic interactions generated by a protein-protein interaction was more important when combining mutations in cis within the same protein than when combining mutations in trans between the two molecules . This results in a different distribution of double mutant outcomes when combining mutations in cis and trans . Whether a second mutation happens in cis or in trans can therefore impact an evolutionary outcome . A substantial fraction of genetic interactions could however not be explained by structural contacts . Some other mechanisms not accounted for by the model could be at play . For instance , non-linearities between the growth rate and complementation complex in the protein-complementation assay could artificially produce genetic interactions . However , such saturation effects are unlikely in the range of expression and binding affinities in this study because they would lead to diminishing returns when combining two strength-increasing mutations , which is not observed ( Figure 3C ) . Moreover , Levy et al . have shown that growth is correlated to complementation complex concentration over a wide-range of concentrations ( Levy et al . , 2014 ) . A more likely source of actual genetic interactions could come from JUN’s ability to form homodimers , which are however less stable than the Fos-JunN heterodimer ( Chinenov and Kerppola , 2001 ) . Mutations affecting the equilibrium between the Jun-Jun homodimer and the Fos-Jun heterodimer could indeed have effects that would not be predicted by our thermodynamic model . Elucidating the remaining mechanisms of genetic interactions will thus require further studies that take these effects into account . Our approach complements the large-scale efforts to comprehensively map genetic interactions between gene deletions or representative alleles of yeast genes ( Tong et al . , 2004; Costanzo et al . , 2010; Costanzo et al . , 2016 ) . Gene deletions are , however , rare in nature , with most genetic variation consisting of point mutations of diverse and difficult to predict effects . Our data provides a comprehensive view of how point mutations within two genes interact to affect a biological function . It will be interesting to extend this strategy to quantifying the effects of point mutation combinations on additional phenotypes beyond PPIs , including applying it to gene pairs that do not encode directly physically interacting proteins but instead participate in regulatory interactions or the same biological process .
All experiments were performed in BY4742 ( MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0 ) . Two intermediate plasmids were constructed each carrying the CYC promoter , a DHFR F[1 , 2] ( GGGGS ) 4 linker fusion or DHFR F[3] ( GGGGS ) 3- ( GGGAS ) linker fusion , respectively , a CYC terminator and a LEU2 or URA3 selection cassette , respectively . The backbone fragments containing the CYC terminator , selection cassette and propagation elements were amplified from pAG415GPD-ccdB-EGFP and pAG416GPD-ccdB-EGFP ( Addgene plasmid # 14196 ) , respectively , by Polymerase Chain Reaction ( PCR ) using primer pairs OGD075-OGD077 and OGD076-OGD077 , respectively . The CYC promoter fragment ( identical for both constructs ) was amplified by PCR from pYM-N10 ( Janke et al . , 2004 ) using primer pair OGD041-OGD042 . The DHFR F[1 , 2] and DHFR F[3] fragments were amplified by PCR from pAG25-D1 , 2 and pAG32-D3 ( Tarassov et al . , 2008 ) , respectively , using primer pairs OGD051-OGD080 and OGD053-OGD081 , respectively . The linkers were encoded as overhangs on primers used to amplify DHFR fragments ( 3’ side ) and plasmid backbones ( CYC terminator 5’ side ) . All PCR fragments were purified using a QIAquick PCR purification kit ( Qiagen , Netherlands ) . Plasmids were constructed by Gibson assembly using 15 fmol of backbone fragment , 45 fmol of DHFR fragment and 45 fmol of promoter fragment in 20 μL reactions . The reactions were incubated at 50˚C for 60 min and 1 μL was transformed into homemade DH5α chemo-competent cells . Correct assembly was checked by colony PCR and the absence of mutations in the coding region by Sanger sequencing . The resulting sequence-confirmed plasmids were named pGD002 and pGD006 , respectively . The complete empty vector was then assembled by inserting the CYC promoter – DHFR F[1 , 2] – Linker fragment ( amplified by PCR from pGD002 using primer pair OGD089-OGD101 ) into pGD006 ( linearized by PCR using primer pair OGD087-OGD099 ) at the 3’ end of the CYC terminator . All fragments were purified using a QIAquick PCR purification kit ( Qiagen , Netherlands ) . Gibson assembly was performed as described above with 30 fmol of backbone and 90 fmol of fragment and transformed in homemade DH5α chemo-competent cells . Correct assembly was checked by colony PCR and the absence of mutations in the coding region by Sanger sequencing . The resulting sequence-confirmed plasmid was named pGD009 . FOS and JUN basic leucine zipper domains , as defined by SMART ( Letunic et al . , 2015 ) , were amplified from human cDNA provided by Juan Valcarcel using primer pairs OGD098-OGD110 and OGD113-OGD125 , respectively . These fragments correspond to residues 135 to 199 and 250 to 314 , respectively . The CYC terminator was amplified by PCR using primer pair OGD099-OGD135 . The backbone was prepared by digestion with BamHI and NheI . All fragments were purified using a QIAquick PCR purification kit ( Qiagen , Netherlands ) . Gibson assembly was performed as described above with 10 fmol of backbone and 50 fmol of each of the CYC terminator FOS and JUN fragments . Correct assembly was checked by colony PCR and the absence of mutations in the coding region by Sanger sequencing . The resulting sequence-confirmed plasmid was named pGD012 . | Proteins , the molecular workhorses of the cell , are made of small units called amino acids attached together like the links of a chain . Each protein is composed of a unique combination of amino acids , which is determined by a specific sequence of DNA called a gene . A change in a gene – a mutation – can create a variation in the protein it codes for , for instance by swapping a type of amino acid for another . Different mutations in the same gene can alter a protein in different ways . Some of these changes are harmless , but other can hinder how the protein performs its role . For example , a small change in the structure of a protein could affect how it will bind to other molecules . It is possible for people to have identical mutations in the same genes , but experience different consequences . For instance , two persons could carry the same disease-inducing mutation , but one has a severe version of the condition and the other only mild symptoms . One reason is that changes in other genes cancel out or enhance the effect of a mutation . This phenomenon is known as a genetic interaction and it remains poorly understood , especially at the molecular level . Here , Diss and Lehner developed a method , called deepPCA , to study the consequences of mutations in proteins in the laboratory . The experiments focused on two human genes which code for two proteins that normally attach to each other . Two mutations were artificially created , either one in each gene , or two in one of them . Diss and Lehner then examined how strongly the two mutated proteins could still attach to each other . By repeating this process with over 120 , 000 different pairs of mutations , it became possible to study how one mutation can have different effects depending on the presence of other mutations in the same protein or in the binding partner . Overall , Diss and Lehner found that genetic interactions are the result of two mechanisms . In the first one , the two mutations together cause specific structural changes that modify how proteins bind to each other . In the second one , the changes solely depend on the magnitude of the initial , thermodynamic effects of individual mutations , but not on their specific physical and chemical properties . To predict the consequences of this second type of genetic interactions , knowing the identity or the exact effects of the two mutations is not necessary . Understanding and predicting genetic interactions is important to develop personalized medicine , where treatments are tailored based on the genetic make up of an individual . This knowledge will also help to study how genes have evolved together . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology"
] | 2018 | The genetic landscape of a physical interaction |
Genetic variation conferring resistance and susceptibility to carcinogen-induced tumorigenesis is frequently studied in mice . We have now turned this idea to melanoma using the collaborative cross ( CC ) , a resource of mouse strains designed to discover genes for complex diseases . We studied melanoma-prone transgenic progeny across seventy CC genetic backgrounds . We mapped a strong quantitative trait locus for rapid onset spontaneous melanoma onset to Prkdc , a gene involved in detection and repair of DNA damage . In contrast , rapid onset UVR-induced melanoma was linked to the ribosomal subunit gene Rrp15 . Ribosome biogenesis was upregulated in skin shortly after UVR exposure . Mechanistically , variation in the ‘usual suspects’ by which UVR may exacerbate melanoma , defective DNA repair , melanocyte proliferation , or inflammatory cell infiltration , did not explain melanoma susceptibility or resistance across the CC . Instead , events occurring soon after exposure , such as dysregulation of ribosome function , which alters many aspects of cellular metabolism , may be important .
Cutaneous malignant melanoma ( MM ) is well known to be associated with high levels of sun exposure . However , this is only true for intermittent rather than chronic exposures , with indoor workers having a higher risk for MM than outdoor workers ( Gandini et al . , 2005 ) . Examples of intermittent sun exposures include number of waterside vacations and number of severe sunburns . One melanoma subtype , lentigo maligna melanoma ( LMM ) , is invariably linked to chronic sun exposure . Individual risk for the development of MM is in part due to the number of UVR-induced mutations incurred ( i . e . the environmental factor ) , but it is also due to genetic variation that controls skin color ( degree of protection due to pigment ) , DNA repair capability ( DiGiovanna and Kraemer , 2012 ) , propensity to burn ( inflammation ) , failure of programmed death of a damaged cell , and/or other factors that control melanocyte behavior ( Sample and He , 2018 ) . Genome-wide association studies ( GWAS ) point to genes regulating aspects of cell division control ( e . g . CDKN2A/MTAP , PLA2G6 , TERT ) , DNA repair ( e . g . PARP1 , APEX1 , ATM ) , and pigmentation ( e . g . MC1R , ASIP , TYR , SLC45A2 , TYR ) as the main players in conferring MM risk ( Gerstenblith et al . , 2010; Duffy et al . , 2018 ) . However only a small fraction of the variance in MM risk is explained by these genes , suggesting that there are still other genes involved in conferring MM risk in the general population ( Hulur et al . , 2017 ) . Clearly , MM is a heterogeneous disease with respect to both innate and somatic genetics , and also to environmental factors since each individual with MM was exposed to different levels of sun exposure . The most common superficial spreading melanomas ( SSM ) are sometimes found on sun-exposed body sites , but more often on non-sun exposed sites , and there can be great diversity in terms of the number of UVR signature mutations in individual lesions ( Mukhopadhyay et al . , 2017 ) . While the number of UVR signature mutations in MMs strongly suggests a role for UVR in MM , the number of mutations present in a MM is due to not just obvious levels of exposure , and protective and repair differences between individuals , but also other factors , for instance the type of exposure . There is in vivo evidence that less intense exposures may be more important for skin cancer induction than more intense doses which cause more apoptosis ( Lan et al . , 2016 ) , and even for a single exposure there are significant differences in skin responses between the same dose ( i . e . the same levels of DNA damage ) administered with high intensity over a short period versus a low intensity for a longer period ( Iida et al . , 2016 ) . Our skin can protect itself via pigmentation responses ( tanning ) , but also by ‘photo-adaption’ , which is independent of pigmentation levels ( Palmer et al . , 2006 ) . Thus , the skin of different individuals can respond and adapt to various forms of sun exposure in different ways , and there are potentially multiple and interacting mechanisms which might explain how UVR exposure could initiate or accelerate MM development in the general population . For the above reasons , most experimental work on UVR carcinogenesis has used animal models . Natural genetic variation can confer resistance to many cancer types in mice ( Balmain , 2002 ) , and it is of great interest to determine why this is so . Most mouse MM models rapidly develop tumours after neonatal UVR exposure . Such models have provided tractable experimental systems to determine a MM action spectrum ( De Fabo et al . , 2004 ) , to assess which type of UVR-induced DNA adducts are required , and to study the role of UVR-induced DNA damage , inflammation , and immunosuppression ( Walker , 2008 ) . It is not yet clearly known why a single neonatal UVR exposure so efficiently accelerates MM onset . We have shown previously that it is not via the acquisition of unrepaired UVR-induced damage leading to mutations in important cancer genes ( Mukhopadhyay et al . , 2016 ) . A number of factors have been proposed to play a role: 1 ) there is a muted inflammatory response to UVR in neonates associated with immunosuppression ( Wolnicka-Glubisz et al . , 2007 ) ; 15 , 16 ) , Muller et al . , 2008 ) there is a heightened sensitivity of neonatal melanocytes to proliferate following UVR ( Walker et al . , 209 ) that is driven by inflammatory cytokines , especially interferon-γ ( Zaidi et al . , 2011 ) . We used the Cdk4R24C::Tyr-NRASQ61K ( hereafter termed Cdk4::NRAS ) mouse as a UVR-induced MM model . Somatic NRAS mutation is carried by 27% of MMs ( http://www . cbioportal . org ) , and in ~90% of MMs the p16/CDK4/pRb pathway is deregulated via mutations in CDKN2A , CDK4 , or RB1 , and/or CDK4 or CCND1 amplification ( Sheppard and McArthur , 2013 ) . We studied the development of MM in mice of 70 diverse genetic backgrounds carrying these transgenes . To do so , we utilized the Collaborative Cross ( CC ) , a set of recombinant inbred mouse strains generated from eight original founder strains , designed to enable rapid gene mapping ( Churchill et al . , 2004; Morahan et al . , 2008 ) . The CC is ideal for systematic analysis studies to discover modifier genes for complex diseases . Mice from each inbred CC strain may be considered as ‘clones’ of each other . Related to the CC is the diversity outbred ( DO ) population , in which mice descended from the same eight founders are generated as outbred stock ( Churchill et al . , 2012 ) . The DO system allows very high levels of heterozygosity and recombination of CC founder alleles , but each DO mouse is genetically unique and not reproducible for experimentation requiring testing of multiple mice . The CC system has allowed us to study the influence of germline genetic background on MM induction using experimentally controlled UVR exposures . This approach tries to explain UVR-induced MM susceptibility and resistance by integrating the complex interaction of many kinds of genetic and biological information , and as such should provide much more realistic insights into MM than simple disease models focusing on single genes or proteins in isolation ( e . g . Hamilton and Yu , 2012 ) .
Before embarking on the screen for melanoma modifier genes in mice , we assessed whether there may be better murine models to work with . All models tested were on the FVB strain background . Given that BRAFV600E mutation is more common than NRASQ61K in MM overall , we studied the inducible BrafV600E model developed by the MacMahon lab ( Dankort et al . , 2009 ) combined with the knock-in mutant Cdk4R24C mouse . Cdk4R24C/R24C::Tyr-CreER::BrafV600E mice were studied in three groups . In one group , the spontaneous MM group , Braf was induced by topical tamoxifen ( tam ) at P1 , P2 , and P3 . In the next group , we applied Tam at P1 , P2 , and P3 , then exposed the mice to a single neonatal UVB dose at post-natal day 3 ( P3 ) ( Figure 1A ) . For the final group , we first exposed to UVR at P3 , then treated with Tam at P7 , P8 and P9 ( Figure 1B ) . Surprisingly , we saw no significant difference in MM age of onset between any cohort ( Figure 1C ) . Melanoma is not observed in Cdk4R24C/R24C mice without carrying a melanocyte-specific Ras pathway mutation , with or without neonatal UVR ( Hacker et al . , 2006 ) , showing that in our experiments with the Cdk4R24C/R24C::Tyr-CreER::BrafV600E model , BrafV600E must have been induced by the tamox application . In contrast , using the Cdk4R24C/R24C::Tyr-NRASQ61K model ( Ferguson et al . , 2010 ) the single neonatal UVR exposure significantly accelerated MM age of onset ( Figure 1D ) . As another context in which to assess the role of the engineered mutation in mouse models of UVR-induced MM , we studied the Trp53F/F::Tyr-Cre ( ER ) ::Tyr-NRAS model in which the Trp53 deletion is induced by tamox application ( 26 ) ( Figure 1E ) , whereas in these mice the NRASQ61K mutation is not inducible , so is present through development . Tamox treatment ( i . e . Trp53 deletion in melanocytes ) accelerated both spontaneous and UVR-induced MM . But there was no difference in MM onset whether or not Trp53 was deleted before or after neonatal UVR . Thus for both BrafV600E and Trp53-inducible models the ability of neonatal UVR to accelerate MM may not be dependent upon whether the engineered mutation is present in melanocytes at the time of UVR exposure . Instead , it may be due to a more generalized effect via differences in DNA repair , melanocyte number and proliferative response , or inflammatory response , as has been outlined previously ( Mukhopadhyay et al . , 2016; Walker et al . , 2009; Zaidi et al . , 2011 ) . But an oncogenic mutation in melanocytes seems to be a prerequisite . In sum , the Cdk4::NRAS model was best suited for breeding with CC mice to look for QTLs associated with UVR-dependent MM . We tested 38 CC strains to discover QTLs that modify median age of onset of spontaneous MM per strain in CC X Cdk4::Tyr-NRAS progeny ( Ferguson et al . , 2015 ) . The phenotype was encoded as median age of MM onset per strain ( Figure 2B ) , and genetic analyses performed using the Gene Miner platform ( Ram and Morahan , 2017 ) . This software uses a logistic regression matrix model over the reconstructed haplotypes matrix to produce genome-wide distribution of P values ( ANOVA chi-squared ) . We used a false discovery rate of p<0 . 001 to define significant genome wide linkage . We identified a major effect QTL on mouse chromosome ( chr ) 16 . The -Log10 ( P ) −1 interval = 14 . 8–21 . 4 megabases ( Mb ) , a region containing 45 genes ( Figure 2C ) . Examination of the founder haplotype coefficients in the significantly linked interval on chromosome 16 showed that the causal variant for early age of onset of MM was derived from the 129/SvJ founder ( hereafter termed 129S ) . We ascertained which genes within the interval were the best candidates by cataloguing DNA variants that are carried on the causal 129S haplotype , and not any of the other founder haplotypes . Mining the Sanger Mouse Genomes and ENCODE databases for 129S-specific variants revealed eight candidates , two of which ( Prkdc and Arvcf ) carry non-synonymous mutations ( Figure 2D ) , while all 8 ( Prkdc , Pkp2 , Yars2 , Arvcf , Gp1bb , and Abcc5 ) had 129S-specific single nucleotide potentially regulatory polymorphisms ( SNPs ) in their 5’ or 3’ UTR , or introns . It is possible that one of the 129S-specific SNPs near the candidates may regulate another gene elsewhere in the genome , or that there are other 129S-specific regulatory SNPs nearby any of the 45 genes in the region that were not detected by ENCODE . Next , we looked at skin gene expression of the six candidates . They are expressed in neonatal mouse skin and/or hair follicle , including in melanocytes , as published in the Hair Gel database ( Sennett et al . , 2015 ) ( Figure 2E ) . We then performed global gene expression analysis ( RNAseq ) ( Supplementary file 1 ) of skin from adult mice from various laboratory strains ( AJ , NOD , B6 , FVB , and 129S ) ( Figure 2F ) , four of which are CC founders , for which the genome sequences are available in the Sanger database We found no differences between susceptible ( 129S allele-carrying ) and resistant strains ( non 129S allele- carrying; AJ , NOD , FVB , B6 ) that would help demarcate a candidate ( s ) . Only one gene ( Lamp3 , lysosomal-associated membrane protein 3 ) of the potential expressed sequences within the region was significantly differentially expressed between the susceptible and resistant strains assessed ( Figure 2F ) , but the nearest 129S-specific regulatory SNP is located 10 Kb away , near Gp1bb , and the next about 500 Kb away , near Avcrf . Unfortunately , there is no skin expression quantitative trait locus ( eQTL ) data for mouse skin , and it was not possible to consult the GTEX eQTL database for eQTLs in human skin since the LAMP3 , GP1BB , and ACVRF genes in humans are scattered on different chromosomes . Therefore , we hypothesized that a functional effect on phenotype was most likely the presence of missense mutations rather than regulatory expression changes on the causal allele . Prkdc carried two missense mutations , one defined as deleterious by SIFT . This is the same Prkdc ( R2140C ) mutation previously mapped and validated as a modifier of lymphomagenesis ( Mori et al . , 2001 ) , breast cancer ( Yu et al . , 2001 ) , and adenoma associated with ionizing radiation ( Degg et al . , 2003 ) . PRKDC ( also known as DNA-PKc ) plays a critical role in ensuring genome integrity and in cancer in general by mediating ligation of double stranded breaks in DNA ( Goodwin et al . , 2015 ) . The other gene within the interval that is a likely candidate is Arvcf ( armadillo repeat gene deleted in velo-cardio-facial syndrome ) , a catenin protein family member . This family plays an important role in the formation of adherens complexes , which are thought to facilitate communication between the inside and outside environments of a cell . This carries on the causal 129S allele a missense mutation that is defined as tolerated by SIFT , so is perhaps not as likely to be the causal gene . In sum , while we cannot in effect rule out another gene within the region , the most likely candidate is Prkdc , since it carries a missense mutation on the causal allele shown before to confer cancer susceptibility , with Acvrf and Lamp3 arguably less likely candidates . We noted that other genes involved in sensing DNA damage and repairing double stranded breaks ( e . g . PARP1 , ATM , APEX1 ) are in linkage disequilibrium with SNPs associated with melanoma risk in GWAS ( Hulur et al . , 2017 ) . We reasoned that even though PRKDC is not a GWAS hit , we could determine whether variation in its expression is correlated with expression of these GWAS genes . We examined global gene expression in non-sun exposed human skin across the GTEx cohort ( Genenetwork . org ) . Networks of the top 500 genes correlated with PRKDC were constructed at a confidence value of 0 . 9 using STRING ( https://string-db . org ) . The most significant network for molecular function was RNA binding ( p=1 . 22−37 false discovery rate ) , and in KEGG pathways DNA replication ( p=1 . 22×10−8 ) . PARP1 ( at 170 , r = 0 . 46 , p=2×10−16 ) , and APEX1 ( at 287 , r = 0 . 43 , p=8×10−14 ) were in the top 300 most significantly correlated genes with PRKDC . ATM was not in the top 500 , but its relative ATR was at number 105 ( r = 0 . 49 , p=4×10−16 ) . While these correlations are based only on gene expression across the GTEx cohort , not any other aspect of gene function , they do point to the possibility that PRKDC is associated with pathways associated with DNA double strand break repair , components of which are encoded by other genes which confer MM risk in the general population . CC-transgenic progeny strains from 70 CC strains were exposed to a single neonatal exposure then followed until MM developed ( Figure 3A ) . Median age of melanoma onset per strain was scored as the phenotype ( Figure 3B ) . Only lesions developing on the UVR-exposed dorsal surface ( where the overwhelming majority developed ) were counted . We identified a major effect QTL on mouse chr . 1 ( -Log10 ( P ) −1 interval = 187 . 8–189 . 2 Mb ) ( Figure 3C ) , a region containing 10 genes . We ascertained which genes within the interval were the best candidates by cataloguing DNA variants on the causal allele that vary between susceptible or resistant strains . The causal allele was carried by AJ and NOD . Mining the Sanger Mouse Genomes and ENCODE databases for NOD/AJ-specific variants revealed four candidates carrying variants specific to the causal allele ( Tgfb2 , Rrp15 , Spata17 , and Gpatch2 ) . Rrp15 was the only one carrying missense mutations ( Figure 3D ) , while the other three genes ( Tgfb2 , Spata17 , and Gpatch2 ) , had AJ/NOD-specific single nucleotide polymorphisms ( SNPs ) in their 5’ or 3’ UTR , or introns ( Figure 3E ) . Next we looked at the Hair Gel database to determine whether these genes were expressed in skin . All but Spata17 were , essentially ruling this gene out as a candidate ( Figure 3F ) . We then looked at skin gene expression ( Supplementary file 1 ) between AJ , NOD , and FVB , which carry the Chr . one allele , and B6 , DBA , and 129S which do not ( Figure 3G ) . There were no significantly differentially expressed genes between groups , suggesting a missense rather than a regulatory causal variant . Thus , while we cannot rule out Tgfb2 and Gpatch2 , by those criteria Rrp15 is the best candidate . The causal allele ( AJ/NOD ) carried two Rrp15 missense variants , both defined by SIFT as likely to be ‘tolerated’ . But one at amino acid 117 is most likely to be the causal mutation given that it is a Glu > Gln change , whereas the other is Ala > Val , which is likely to be silent . Rrp15 encodes a ribosomal subunit that is part of pre 40S and pre 60S subunits that is important for rRNA transcription and ribosome biogenesis . Furthermore , Rrp15 knockout in vitro causes nucleolar stress by activating the Mdm2-Trp53 axis , and subsequently a G1-S phase cell cycle blockage ( De Marchis et al . , 2005; Dong et al . , 2017 ) . Deregulation of the ribosome complex decreases the fidelity and patterns of mRNA translation and many other downstream events ( Quin et al . , 2014; Pelletier et al . , 2018 ) . Although much of the effect on cell behaviour is via altering the stabilisation of p53 by Mdm2 , this can also occur via p53-independent mechanisms ( James et al . , 2014 ) . DO mice harbor frequent recombinations throughout their genomes . We hypothesized that some DO animals ( Churchill et al . , 2012 ) would have informative recombinants of the causal AJ/NOD allele to make them useful for fine mapping the linked chr . 1 interval . We tested DNA from 314 DO mice , initially using six polymorphic markers across an 8 Mb region around Rrp15 . We selected a subgroup of 8 DO mice with apparent recombinations within the AJ/NOD alleles across the region and genotyped them using a panel of 60 SNPs ( 20 of which are the most informative and shown in Supplementary file 2 . SNPs were chosen based on their ability to discriminate the AJ/NOD allele from the other eight founders , especially those variants private for AJ , NOD , or both . While this NOD/AJ allele was scattered throughout the region in different DO mice , as expected we observed many recombinations across the chr . 1 region of interest . However there were many gaps and alleles for which we could not unambiguously call the founder haplotypes on both strands across the region ( Supplementary file 2 ) . To adequately do this , many more SNPs would have to be tested . Nonetheless we crossed each of the eight selected DO mice with Cdk4::NRAS transgenics , and studied UVR-induced MM onset in the progeny ( Figure 3H ) . Because of the limitations of our SNP-based map of the region , to genotype the DO mice we used manual Sanger sequencing to genotype more densely , in particular in the region of our candidate gene Rrp15 ( Figure 3I ) . As seen in Figure 3H , the predicted causal SNPs in Rrp15 do not segregate with fast MM onset as they did in the CC strains . We hypothesized that lack of penetrance of the Rrp15 allele was due to the introduction of additional resistance alleles elsewhere in the genome due to the high levels of recombinations and heterozygosity in the DO genomes . We tried to circumvent this by backcrossing transgenic-DO mice onto C57BL/6 . Mice from several litters were followed , and after two backcrosses we assessed UVR-induced MM age of onset , with progeny genotyped immediately around Rrp15 by Sanger sequencing . We found that the penetrance of the causal Rrp15 variant was restored on backcrossing: mice carrying the NOD/AJ alleles of Rrp15 had significantly earlier MM onset ( Figure 3J ) . Thus , the propensity for neonatal UVR to accelerate melanoma is highly dependent upon genetic background influences . In addition , there are additional potential resistance alleles which can , if present , interact with the Rrp15 susceptibility allele . We further analyzed the role of neonatal UVR in accelerating melanoma age of onset by subtracting the average age of onset of spontaneous from that of UVR-accelerated melanoma ( Figure 3K ) . This provides us with a further phenotype: the strain-specific rate of acceleration of MM onset by neonatal UVR ( Figure 3L ) . The median age-of-onset for UVR-induced MM per strain was subtracted from the median onset for spontaneous MM , to denote what we have termed the ‘effectiveness’ of UVR in inducing MM . In resistant CC strains , neonatal UVR does not accelerate MM onset at all ( e . g . HOE , BAX2 ) , or does so by a very small amount , whereas for susceptible strains MM onset was accelerated by more than 300 days ( e . g . SEH , XAJ2 ) as seen in Figure 3K and L . The phenotype is independent of pigmentation status ( albino vs pigmented ) ( Figure 3L ) . The genome scan for this trait is shown in Figure 3M . Unfortunately , we only have both spontaneous and UVR-induced onset data for 27 strains . While this does not provide enough power to detect genome-wide significant linkage to this phenotype , there was a peak of suggestive significance at chromosome 1 p , overlapping with the QTL detected for UVR-induced age of onset , containing the Rrp15 gene , with the NOD allele showing a different coefficient from the other strains . The way by which neonatal UVR accelerates MM may provide insights into early events in the initiation of this neoplasm . To search for pathways deregulated in neonatal UVR-exposed skin , we performed global gene expression chip studies with Illumina Beadchips on UVR exposed and non-exposed epidermis of wild-type FVB mice at various time-points ( Supplementary file 3 ) . The top 500 significantly up- and down- regulated genes after neonatal UVR were used to construct networks for UVR-induced gene expression changes at a confidence value of 0 . 9 using the STRING resource ( https://string-db . org ) ( Figure 4 ) . At 6 hr after UVR exposure , one can see strong significant evidence of ribosome biogenesis occurring , with downregulation of various metabolic pathways compared to control untreated skin ( Supplementary file 4 ) . At 10 hr , post-UVR metabolic activity is still suppressed , but DNA replication is either already occurring to some extent , or about to occur for repair and reconstruction of damaged cells . Upregulated p53 signaling was also observed . At 24 hr post UVR various pathways involved in cell proliferation ( ribosomes and translation , metabolism ) and reconstruction ( extracellular matrix , metabolism ) are significantly activated in the epidermis . One of the problems with using whole tissue is that one cannot discriminate in terms of cell type . As a way to put the skin gene expression into some perspective we wanted to look at skin gene expression changes at some days after UVR , since it is known that this period corresponds with an influx of immune cells into the dermis ( Zaidi et al . , 2011 ) and the influx of melanocytes into the epidermis ( Walker et al . , 2009 ) . Therefore , we separately studied gene expression in the neonatal epidermis and dermis harvested at 3d after UVR ( Figure 5A ) . There was a striking loss of immune markers in the epidermis , mostly reflecting UVR-induced migration of epidermal Langerhans cells to the dermis as expected . We also saw a strong gene expression cluster for genes involved in melanogenesis , reflecting the presence of melanocytes in the UVR-exposed epidermis , but not in control skin at the time-point 3 d after UVR . In the dermis , we observed networks reflecting increased cellular activity , in particular upregulation of kinases and interferon-induced enzymes . But in contrast to the epidermis , in the dermis we also observed a signal for myeloid cells , presumably reflecting the influx of macrophages , which occurs maximally at around this time post UVR ( Handoko et al . , 2013 ) . The critical question with respect to genes deregulated by neonatal UVR is the behavior of our candidate genes , one of which must be causal in accelerating MM . Rrp15 was the only one of the chr . 1 candidates significantly changed in expression by UVR ( Figure 5B ) . Hence in addition to the susceptible CC strains carrying a missense mutation in the putative Rrp15 causal allele , the fact that it is the only candidate that responds to neonatal UVR adds weight to it being the best candidate in the linked genomic region . Our gene expression analysis was performed on whole skin , epidermis , or dermis , rather than individual cell types , allowing signals from a number of cells types , for example keratinocytes , immune cells , and melanocytes , to be taken into account . Since Rrp15 is expressed virtually ubiquitously in the skin we do not know whether its mode of action in accelerating UVR-induced melanoma would be cell intrinsic or extrinsic . But we note that in a study of UVR-treated cultured melanocytes , ribosome metabolism was the most significantly altered pathway at 6 , 12 , and 24 hr after UVR ( López et al . , 2015 ) , as in our whole skin studies . In addition , to gain a better sense of changes in skin correlated with RRP15 expression we constructed gene networks based on the top 500 genes correlated with RRP15 expression across the sun-exposed anatomical site human skin GTEx cohort ( Figure 5B ) . As expected changes in RRP15 gene expression are correlated with ribosome biogenesis , but also many other aspects of cell behavior , including DNA damage recognition and repair , transcription , and splicing . Defects in any of these processes could in theory explain why MM is accelerated by UVR exposure not only in mouse strains which carry germline Rrp15 variants , and putatively in humans also . Since genetic background greatly influenced whether or not MM was accelerated by neonatal UVR , we studied various effects of UVR a few days after exposure on traits that may differ between susceptible and resistant strains . These included the rate of removal of UVR-induced cyclobutane pyrimidine dimers ( CPDs ) , influx of inflammatory neutrophils and macrophages , and the proliferation of melanocytes ( increase in melanocyte number in the epidermis ) ( Figure 6A ) . We again used readily-available laboratory strains ( AJ , B6 , DBA , FVB , NOD , 129 s ) , four of which are CC founders . For all six , we determined the median age of onset of MM after neonatal UVR . We chose time-points 1 , 4 , and 7 days post UVR . In C57BL/6 mice , CPDs are generally removed by d4 after UVR , certainly by d7 . Neutrophils generally infiltrate the skin by d1 after UVR and numbers decrease by d4 . Macrophages are maximally present at 4d ( 39 ) . At 4d , melanocytes have migrated to the epidermis in response to UVR , macrophages are at their maximum number , roughly concurrent with maximum epidermal melanocyte density . By the d7 post UVR time-point , the skin is ostensibly returned to normal , but in some mouse strains there may be a delay in some measures . First , across the six laboratory strains we found no differences in the capacity to remove CPDs , as measured by the proportion of skin cells carrying them post-UVR . However , there was great variation between the various strains in the size and timing of the melanocyte and myeloid cell responses after UVR ( Figure 6A ) . To ascertain whether the level of any of these responses could be correlated with age of MM onset , we compared results for the lab strains stratified in terms of whether they carry the susceptible or resistant allele at chr . 1 ( e . g . around Rrp15 ) ( Figure 6B ) . When compared in this way none of these commonly purported mechanisms by which UVR exacerbates MM were significantly associated with onset of UVR-induced MM , although we found a non-significant trend for the susceptible strains to show higher levels of neutrophil influx at d1 after UVR . As an additional assessment , we performed a correlation test between MM age of onset and the phenotypes measured and found no significant correlation , although d1 neutrophil infiltration was nearest to significance . If one accepts that the low number of strains compared ( 3 vs 3 ) makes it difficult to attain statistical significance for a mechanism with a complex milleiu of events , one could argue that neutrophil influx may have an effect on exacerbating MM . Thus we set out to determine whether depletion of neutrophil infiltration using a neutrophil-blocking anti-Ly6g antibody would influence subsequent melanoma induction in Cdk4::NRAS mice . We injected blocking antibody at 1d before UVR , the day of UVR , and 2d after UVR . First we assessed neutrophil depletion after three injections by taking skin sections and staining with neutrophil-specific anti-myeloperoxidase antibody ( Figure 6C ) . Anti-Ly6g treatment significantly depleted neutrophil influx after neonatal UVR ( Figure 6D ) . We exposed two cohorts of Cdk4::NRAS mice to anti-Ly6g anti-neutrophil depleting Ab , or PBS control respectively , then performed a longitudinal study for MM age of onset , but saw no difference in the age of onset of MM between cohorts ( Figure 6E ) . Thus , inhibition of the neutrophil influx did not suppress UVR-induced genesis .
There are likely to be multiple and interacting mechanisms which might explain how UVR exposure could initiate or accelerate different types of MM . We have used a mouse MM model to study the development of MM on many CC strain backgrounds . The Cdk4::NRAS model we used is a well-characterized model of UVR-induced MM , and in terms of an acceleration of MM onset after neonatal UVR it behaves similarly to other models which carry constitutive oncogenic mutations in melanocytes ( e . g . the Mt-Hgf model ) ( Noonan et al . , 2001 ) . But , surprisingly , we found that MM was not accelerated in a model carrying an inducible BrafV600E mutations in its melanocytes . It is known that modes of engineered oncogene expression in melanocytes can influence other phenotypes also . For example , BrafV600E induced in melanocytes in utero is incompatible with embryonic viability ( e . g . Dankort et al . , 2009 ) , whereas in a non-inducible transgenic BrafV600E models it is not ( Wurm et al . , 2012 ) . We do not know why MM was not accelerated in the inducible BrafV600E model , but one possibility is that there may be simply more activated melanocytes in the neonatal Tyr-NRASQ61K transgenics at the time of UVR exposure since NRASQ61K is expressed in utero , whereas BrafV600E is induced only for a short period of 3 days before UVR ( i . e . P1-P3 ) . We favour such an explanation rather than a mechanistic difference between the respective oncogenes ( Braf vs NRAS ) per se , since in another similar inducible BrafV600E model repeated UVR exposures significantly accelerate MM when the oncogene is activated at 4 weeks of age ( Goel et al . , 2009 ) . In a model using inducible Trp53 deletion instead of constitutive Cdk4R24C we found that UVR accelerated MM , but whether Trp53 was deleted before or after UVR was inconsequential . This could be somewhat akin to the work of the Evans laboratory who showed that although TP53 is very important in helping modulate the acute DNA damage response to radiation , its function in this period does not contribute to protecting against transformation , in fact it is only within the long lag phase leading to cancer development that TP53 acts as a tumour suppressor ( Christophorou et al . , 2006 ) . Therefore , mouse models appear to differ with respect to whether and by how much neonatal UVR accelerates MM . There is no perfect animal model for ‘generalized’ MM , but one would expect that the Cdk4::NRAS transgenic in combination with the CC should provide useful information as to how genetic background can influence UVR-induced murine MM . Age of onset of spontaneous MM across the CC was underpinned by germline variation in the Prkdc gene on chr . 16 . This is overwhelmingly the best candidate in the mapped interval as the mutation carried on the 129S susceptibility allele has been functionally validated in other mouse models of cancer ( Yu et al . , 2001; Degg et al . , 2003; Goodwin et al . , 2015 ) . Although it is known to be involved in recognition and removal of DNA damage , it is unclear whether its function in MM would be cell intrinsic or extrinsic , since it is ubiquitously expressed in skin cell types . In addition , this is quantitative genetics , and although the QTL explains a large proportion of the phenotypic variation , other cooperating loci may also exist , for instance on chr . 5 , ( Figure 3C ) , which might become statistically significant ( or disappear ) if more CC strains were tested . In terms of possible human relevance , other genes ( e . g . PARP1 , APEX1 ) involved similarly in DNA repair , and whose expression is correlated with that of PRKDC in human skin , are associated with population-based MM risk ( Duffy et al . , 2018 ) . Thus PRKDC may be an example of a modifier of MM in mice , but mediated by other genes of the same pathway in humans . But the mechanistic result of Prkdc deregulation might be similar to the effects of deregulating MM risk genes like PARP1 and APEX1 , that is , all ultimately influencing DNA repair . Hence our genetic screen appears to have revealed mechanistically relevant information , notwithstanding that the particular causal gene variants can be different between mouse and man . One can also consider Prkdc as a MM modifier gene , that is , our genetic screen detects MM modifiers in the context of a mouse model which carries a germline Cdk4 variant . In keeping with this , PRKDC was also one of the few DNA repair genes associated that modified MM risk in MM-prone families ( Liang et al . , 2012 ) , many of which carry a CDKN2A mutation , and a few CDK4 mutation . If we assume that Cdk4::NRAS mice may in some respects be ostensibly a model for familial MM , the findings of Liang et al . , 2012 further support the possible relevance of our findings in human disease . One of the most striking findings from our study was that the genetic polymorphisms modifying MM onset were very different between spontaneous and UV-induced disease ( Figure 7 ) . This is particularly notable since in this model MM is accelerated by just a single UVR exposure . The age of onset of neonatal UVR-induced MM in the transgenic-CC progeny across the CC was linked to a chr . 1 locus containing a strong candidate , Rrp15 , with a missense mutation . Taken together with the fact that of the candidates within the linked region , only Rrp15 was significantly differentially expressed between susceptible and resistant strains , only Rrp15 was upregulated in the epidermis of neonatal skin after UVR exposure ( at 24 hr ) , we deemed it the strongest candidate . In addition , gene ontology and KEGG pathway analysis of UVR-induced gene expression changes in neonatal mouse skin revealed that ribosome biogenesis is one of the major gene networks upregulated at 6 and 24 hr after UVR . One caveat here is that for the UVR-induced gene expression studies we used whole epidermis ( or dermis ) . Notwithstanding that there are differences in melanocyte and immune cell density over time between 1 and 3 d after neonatal UVR , most of the signal would have come from keratinocytes at all time- points since they are the most numerous cells in the epidermis . But our analysis of UVR-induced gene expression changes , based upon fold-change , are sensitive enough to pick up gene networks that reflect predictable changes in melanocyte and immune cell density ( although not the type of immune cells ) . Rrp15 is expressed in all cell types within the neonatal skin ( Figure 3F ) , but we do not know whether its mode of action in modifying melanocyte transformation is cell autonomous or non-cell autonomous . We assume that the upregulation of ribosomal metabolism we observe at 24 hr after UVR reflects mostly changes in keratinocytes , but very similar responses have also been found in UVR-treated cultured melanocytes ( López et al . , 2015 ) . Single cell sequencing of different skin cells may help resolve these problems , but even such methods also have their drawbacks including unwanted gene expression changes induced soon after removal of cells from their microenvironment ( Yuan et al . , 2017 ) . We have used inferences taken from a variety of sources , which are by themselves not confirmatory , to build a strong case for a role for variation in Rrp15 function influencing the propensity for UVR to accelerate MM . There is some evidence that SNPs near RRP15 are also associated with human MM in some contexts . The world melanoma genetics consortium found such evidence using a method for candidate genetic association to detect variants that may not reach genome-wide statistical significance after correction for multiple testing ( Schoof et al . , 2012; Wurm et al . , 2012 ) . There was no information on individual sun exposure in the tested cohorts . Of the 39 immune-related genes tested , SNPs near LGAL3 and TGFB2 were the most significantly associated with MM risk ( Rendleman et al . , 2013 ) . Of note , the RRP15 and TGFB2 genes are located adjacent to one another and just 7 kb apart . Which of the two is the causal gene is difficult to elucidate in the human genetics study , but our systems analysis work using the CC has allowed us to build a very strong case for Rrp15 . We discovered that genetic background dramatically influences the propensity for melanocyte transformation after UVR . As well as discovering a gene likely to be associated with this , we also examined differences in skin responses between susceptible versus resistant mouse strains . We did not observe major differences in the propensity for removal of UVR-induced CPDs , nor in melanocyte proliferation , nor macrophage influx . We observed a weak correlation with the number of skin-infiltrating neutrophils at d1 after neonatal UVR and UVR-induced MM onset . However , depleting neutrophils before and after neonatal UVR did not reduce the time of onset of MM , somewhat in line with our previous finding that depletion of macrophage infiltration also did not abrogate the MM-accelerating effect of UVR ( Handoko et al . , 2015 ) . It could be argued that by using immunohistochemical staining to assess removal of UVR-induced CPDs we could miss subtle differences between mouse strains in the repair process that may be consequential in terms of leaving a molecular memory of UVR damage in the transformed cells . But even our exome sequencing on the neonatal UVR-induced mouse melanomas does not suggest a major role for UVR-induced mutagenesis ( Mukhopadhyay et al . , 2016 ) . Researchers tend to look for measurable skin responses to UVR exposure to try to understand how melanocytes may be destabilized and ultimately transformed by DNA damage . The ‘usual suspects’ include defective DNA repair , photo-immunosuppression , inflammation , and cell proliferation , which are observable in the days following exposure . These can all enhance carcinogenesis in specific contexts . But our results suggest that earlier events in the few hours post UVR such as aberrant ribosome activity that can cause inappropriate protein expression , may be more important . It is not clear whether this would be acting during the acute damage repair period after neonatal UVR , or during the lag period leading to tumour initiation . The notion of a particular constellation of gene networks that vary between CC strains and confer resistance to the MM-accelerating effects of UVR may not be dissimilar to what occurs in amphibians , where regenerating limbs , but not non-regenerating body parts , are resistant to carcinogen-induced cancer ( Sarig and Tzahor , 2017 ) , despite the fact that both anatomical sites incur the requisite DNA damage . Particular molecular networks within skin cells in the MM-resistant strains appear to work against transformation , despite animals from all strains being exposed to high levels of UVR-induced damage . Hence in keeping with the fact that even non-cancerous human skin can carry UVR signature mutations in cancer genes ( Martincorena et al . , 2015 ) , incorrectly repaired UVR-induced DNA damage leading to somatic mutation is in essence necessary but not sufficient to exacerbate MM , and the presence of germline variants for melanoma susceptibility and resistance is very important . Biological systems ( e . g . skin and skin UVR responses ) are very complex and varied across a population of different individuals . Our systems analysis strategy has attempted to harness such complexity and in doing so has resulted in the discovery of some potentially important findings with regards UVR-induced melanoma . We have performed a genetic screen for natural genes regulating MM age of onset and found surprisingly that different genes mediate spontaneous and UVR-induced MM susceptibility ( Figure 7 ) . We have identified a strong candidate genes and potential mechanisms in both cases .
Cdk4R24C/R24C::Tyr-NRASQ61K/+ mice are previously described in Ferguson et al . ( 2010 ) . We crossed Cdk4R24C/R24C::Tyr-NRASQ61K/+ mice with breeding partners from each CC strain ( Ferguson et al . , 2015 ) . Hence all study mice are Cdk4R24C/+::Tyr-NRASQ61K/+ . All experiments were undertaken with institute animal ethics approval ( A98004M ) . Mice were sacrificed before tumors exceeded 10 mm in diameter . In some melanoma-resistant strains lymphomas developed in some mice at >400 days of age . Such mice were counted as melanoma-free at the age of death . Each phenotypic measurement is based upon at least 4 and up to 15 mice per CC strain background . p53F/F mice ( carrying floxed alleles allowing Cre-mediated excision of exons 1–10 ) were obtained from the Mouse Models for Cancer Consortium ( http://mouse . ncifcrf . gov ) . Melanocyte-specific Trp53 deletion in p53F/F/Tyr-Cre ( ER ) /Tyr-NRAS mice was induced via topical application of 8-OH-tamoxifen ( 15 mg/ml in DMSO ) at P0 , 1 and 2 . For the studies involving the BRAF model we used the inducible BRAFV600E model generated by ( 24 ) . All these mice were of FVB strain background . Collaborative cross mice were obtained from the Animal Research Council ( ARC ) in collaboration with Prof Grant Morahan of the Harry Perkins Institute of Medical Research , Perth , Australia . A/J , C57BL/6J ( B6 ) , 129/SvJ ( 129S ) , DBA , and NOD/ShiLtJ ( NOD ) were purchased from the Animal Resources Centre , Western Australia . Pups ( 3-day-old ) were exposed to a single UVB exposure from a bank of 6 Phillips TL100W 12RS UVB lamps ( Total UVB dose , 5 . 9 kJ/m2 , or an erythemally-weighted dose of 1 . 8 kJ/m2 ) UVB . We have previously described a system for visual tracking of lesions developing on the FVB Cdk4R24C/R24C::Tyr-NRASQ61K/+ mice and a histology-based staging system ( Wurm et al . , 2012 ) and mice on the various CC strain backgrounds were scored in this way . Briefly , lesions were excised after death followed by conventional histopathologic work-up with haematoxylin and eosin ( H and E ) staining . Each lesion was viewed and confirmed individually as MM by BF and GJW as described in Wurm et al . ( 42 ) . Where there was any doubt about diagnosis , tumors were stained with Trp1 and/or Sox10 . No skin tumors were observed apart from melanomas . We included in our analyses melanomas that developed on dorsal surface only . Mice were sacrificed before tumors exceeded 10 mm in diameter . The construction of the CC founder haplotypes is described in Ram and Morahan ( 2017 ) . For mapping we used a logistic regression matrix model over the reconstructed haplotypes matrix to produce genome-wide distribution of P values ( ANOVA chi-squared ) . We used a false discovery rate of p=0 . 0001 to define significant genome wide linkage . We used a custom SNP approach performed by AGRF , a custom array of 92 SNPs across the 180–190 Mb region of mouse Chr . 1 . Genotyping was performed using the Sequenom mass array system . Target fragments chosen to contain multiple SNPs were PCR-amplified , then cleaned from excess primers and nucleotides using CleanSweep PCR Purification ( Applied Biosystems , Life Technologies , Carlsbad , CA , USA ) . Sequencing was carried out using BigDye Terminator v3 . 1 Cycle Sequencing and then run on ABI Prism DNA Sequencers ( Applied Biosystems ) . The sequencing traces were compared using Multiple SeqDoc chromatogram comparison programme ( http://research . imb . uq . edu . au/seqdoc/multi . html ) . Five hundred ng of total RNA from each tumor was used as the starting material to produce cRNA , following Illumina TotalPrep RNA Amplification protocol . From each sample , 1500 ng of cRNA were hybridised to Illumina MouseWG-6 v2 . 0 Expression BeadChips ( Illumina , San Diego , CA , USA ) and then scanned . Data were extracted using Genome Studio ( Illumina , San Diego , CA , USA ) and then imported into GeneSpring GX 11 . 5 . 1 ( Agilent Technologies , Santa Clara , CA , USA ) , before subsequent analysis . Normalisation was performed using the R package LUMI , and differential analysis done using LIMMA . Multiple testing correction was carried out using the Benjamini-Hochberg procedure . Differentially expressed genes with p-adjusted values < 0 . 05 were considered significant . Skin gene expression experiments were undertaken with institute animal ethics approval ( A98004M ) . RNA was isolated using RNeasy Kit and libraries generated with Illumina mRNA kit . Sequencing was performed using Ilumina HiSeq chemistry , with 50 bp single reads . RNA-seq samples were mapped to mouse genome MM10 using STAR . Quality control metrics were computed using RNA_SeQC version 1 . 1 . 18 , and expression values estimated using RSEM version 1 . 2 . 30 . We corrected for library size by dividing each sample’s count by millions of reads mapped . We used the ‘calcnormfactors’ function from the EdgeR package to obtain TMM factors and used these to correct for differences in RNA composition . To generate interconnected networks based on correlations , gene lists were clustered using STRING ( http://string-db . org/ ) . STRING creates networks representing the best available knowledge of gene interconnections . Each protein-protein interaction is annotated with 'scores' indicating how likely an interaction should be true . Scores rank from 0 to 1 , with one being the highest confidence . A score of 0 . 5 indicates roughly every second interaction might be erroneous . Gene-gene co-expression correlations were computed as Pearson product-moment correlations ( r ) in Genenetwork . org after removing outliers . Skin from the pups at Day one after UVB radiation ( D1 ) , D4 and D7 was paraffin embedded . All immunohistochemistry staining was performed on pup skin sections ( 4 um ) with standard DAB or NovaRed . Counts were performed on multiple fields from multiple skin sections from each of >3 mice ( each field is ~1 mm in length ) . Macrophage staining used F4/80 from CD68 rat monoclonal antibody diluted to 1:400 , Abcam ab6640 , CI: A3-1 ( Cambridge , UK ) . Melanocyte nuclear staining used Sox10 from sc-17342 goat polyclonal antibody ( N-20 ) diluted to 1:200 , Santa Cruz Biotechnology ( Dallas , TX , USA ) . Cyclobutane pyrimidine dimers ( CPD ) staining was with monoclonal anti-thymine dimer , Clone H3 ( T1192 ) diluted to 1:400 , Sigma-Aldrich ( St Louis , MO , USA ) . For this CPD staining , sections were first blocked in 1% hydrogen peroxide , then incubated in 50% ethanol , 30% ethanol/0 . 02N HCl , 0 . 05N HCl , and 0 . 07N NaOh/70% ethanol , before incubating in primary antibody . Neutrophils were stained using an anti-Ly6G as primary antibody , except after neutrophil depletion , when staining was done using MPO staining ( see below ) . Anti-Ly6G was from rat anti-neutrophil monoclonal antibody Abcam ab2557: NIMP-R14 ( Cambridge , UK ) at a concentration of 1:100 . Each NOD/Cdk4R24C/+::NRAS pup was injected intraperitoneally , with either 100 ug InVivo MAb anti-mouse Ly6G ( Bio C Cell , Beverly , MA , USA ) or PBS ( as a control ) at P2 , P3 and P5 . At P3 , all pups were also given UVR treatment . As the depletion was done using anti-Ly6G , staining to differentiate between depleted and non-depleted skin was done using Myeloperoxidase staining ( Abnova rabbit anti-MPO , Taipei , Taiwan ) at a concentration of 1:75 . One-way ANOVA tests were used to determine the significant difference between means , using R . The survival of mice in each treatment group was estimated using Kaplan-Meier analysis ( PRISMTM ) , and the Log-Rank ( Mantel-Cox ) test was used to test for differences between the groups . For correlation comparisons Pearson correlation r value was calculated in PRISM along with the p-value for significant correlation . | Melanoma is a type of skin cancer . Melanoma tumors form in the skin’s pigment-producing cells or melanocytes . Growing evidence points to complex interactions between genetics and environmental exposures that contribute to the risk of developing melanoma . Ultraviolet ( UV ) radiation from the sun causes genetic mutations in melanocytes . This sun exposure interacts with genetic variations that may make people more or less vulnerable to such DNA damage . For example , genetic variations that control skin color or the cell’s ability to repair DNA , and that influence how easily people develop sunburn , all affect whether UV damage leads to melanoma . However , some forms of melanoma are not caused by sun exposure at all . Most of the genetic variations linked to melanoma have a small effect on the risk of developing the disease . So , it is unlikely that these genes alone cause melanoma . Few studies have been able to map the complex interactions between genes and the environment that lead to melanoma . So far , it has been unclear if there are different genetic mechanisms that lead to an increased risk for sun-exposure linked melanoma and non-sun linked melanoma . Now , Ferguson et al . show that variations in the genes involved in DNA repair during normal cell growth are linked to non-sun linked melanoma . Sun-linked melanoma , on the other hand , was associated with genes involved in the production of proteins in part of the cell called ribosomes . In the experiments , the effects of both UV light and various genetic variations were assessed across many different strains of mice . Mutations that impair the cell’s ability to repair UV-induced DNA damage or that contribute to excessive inflammation in response to sunburn did not increase melanoma susceptibility in these experiments . Ferguson et al . show that the amount of UV-induced DNA damage alone does not explain melanoma risk , which may not always depend on skin pigmentation . The experiments also suggest that non-UV linked melanoma is caused by a different mechanism than sun exposure-linked melanoma . Learning more about different genetic factors that affect the risk of developing different types of melanoma may help scientists develop more specific treatments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology"
] | 2019 | Different genetic mechanisms mediate spontaneous versus UVR-induced malignant melanoma |
In vertebrates , the total number of vertebrae is precisely defined . Vertebrae derive from embryonic somites that are continuously produced posteriorly from the presomitic mesoderm ( PSM ) during body formation . We show that in the chicken embryo , activation of posterior Hox genes ( paralogs 9–13 ) in the tail-bud correlates with the slowing down of axis elongation . Our data indicate that a subset of progressively more posterior Hox genes , which are collinearly activated in vertebral precursors , repress Wnt activity with increasing strength . This leads to a graded repression of the Brachyury/T transcription factor , reducing mesoderm ingression and slowing down the elongation process . Due to the continuation of somite formation , this mechanism leads to the progressive reduction of PSM size . This ultimately brings the retinoic acid ( RA ) -producing segmented region in close vicinity to the tail bud , potentially accounting for the termination of segmentation and axis elongation .
Body skeletal muscles and vertebrae form from a transient embryonic tissue called paraxial mesoderm ( PM ) . The PM becomes segmented into epithelial structures called somites , which are sequentially produced in a rhythmic fashion from the presomitic mesoderm ( PSM ) . The PSM is formed caudally during gastrulation by ingression of the PM progenitors located initially in the anterior part of the primitive streak ( PS ) and later , in the tail-bud ( Bénazéraf and Pourquié , 2013 ) . At the end of somitogenesis , the embryonic axis is segmented into a fixed species-specific number of somites which varies tremendously between species ranging from as little as ∼32 in zebrafish to more than 300 in some snakes . The somites subsequently differentiate into their final vertebral and muscular derivatives to establish the various characteristic anatomical regions of the body . Hox genes code for a family of transcription factors involved in specification of regional identity along the body axis ( Mallo et al . , 2012; Noordermeer and Duboule , 2013 ) . In mouse and chicken , the 39 Hox genes are organized in four clusters containing up to thirteen paralogous genes each . Hox genes exhibit both spatial and temporal collinearity , meaning that they are activated in a sequence reflecting their position along the chromosome and become expressed in domains whose anterior boundaries along the body axis also reflect their position in the clusters . Whether Hox genes control axis length and segment number has been controversial . Mouse mutants in which entire sets of Hox paralogs are inactivated show severe vertebral patterning defects but exhibit normal vertebral counts ( Wellik and Capecchi , 2003; McIntyre et al . , 2007 ) . In contrast , precocious expression of Hox13 genes in transgenic mice leads to axis truncation with reduced vertebral numbers ( Young et al . , 2009 ) . Furthermore , mouse null mutations for Hoxb13 or Hoxc13 result in the production of supernumerary vertebrae ( Godwin and Capecchi , 1998; Economides et al . , 2003 ) . In chicken and fish embryos , the arrest of axis elongation has been linked to the inhibition of FGF and Wnt signaling in the tail-bud which leads to the down-regulation of the transcription factor T/Brachyury and of the Retinoic Acid ( RA ) -degrading enzyme Cyp26A1 ( Young et al . , 2009; Martin and Kimelman , 2010; Tenin et al . , 2010; Olivera-Martinez et al . , 2012 ) . Downregulation of Cyp26A1 in the tail-bud ultimately leads to rising RA levels and to differentiation and death of the PM progenitors which terminates axis elongation . Premature exposure of the tail-bud to high RA levels in chicken or mouse embryos inhibits Wnt and FGF signaling and leads to axis truncation ( Tenin et al . , 2010; Olivera-Martinez et al . , 2012; Iulianella et al . , 1999 ) suggesting that the tail-bud must be protected from the differentiating action of RA . In the Cyp26A1 null mutant mice , RA-signaling reaches the tail-bud , prematurely inducing the downregulation of FGF signaling and the increase of Sox2 expression , resulting in axis truncation posterior to the thoracic level ( Abu-Abed et al . , 2001; Sakai et al . , 2001 ) . In chicken , the tail-bud starts to produce RA when explanted in culture after the 40-somite stage ( Tenin et al . , 2010 ) . This late RA signaling activity in the tail-bud is involved in the termination of segmentation and axis elongation ( Tenin et al . , 2010; Olivera-Martinez et al . , 2012 ) . At the 40-somite stage , the mRNA for Raldh2 , the RA-biosynthetic enzyme becomes expressed in the tail-bud potentially accounting for this late RA activity . What triggers this late expression of Raldh2 in the chicken tail-bud is however unknown . In vertebrates , the termination of axis elongation is accompanied by a progressive reduction in size of the PSM ( Gomez et al . , 2008; Gomez and Pourquié , 2009 ) . The shrinking of the PSM which brings the segmented region producing RA in the vicinity of the tail-bud might also contribute to the raise in RA levels in the tail-bud and possibly to the late Raldh2 activation in the tail-bud . Thus in the chicken embryo , the timing of elongation arrest ( and hence the total number of somites formed ) could be in part controlled by the kinetics of PSM shrinking . PSM size depends on the velocity of somite formation which removes cells anteriorly and on the flux of cells from the primitive streak and tail-bud generated during elongation which injects cells posteriorly . How this flux of progenitors ingressing in the PSM is regulated over time , and which genes are regulating this process remain poorly understood . Hoxb1-9 genes have been proposed to control cell ingression of paraxial mesoderm precursors from the epiblast during gastrulation ( Iimura and Pourquié , 2006 ) . However , Hoxb1-9 genes are only expressed in anterior regions of the embryo precluding their playing a role in the control of axis termination . Here , we first investigate the relationship between the speed of somite formation and of axis elongation . We show that , in the chicken embryo , activation of Abdominal B-like posterior Hox genes in the tail-bud correlates with the slowing down of axis elongation , while the speed of somitogenesis remains approximately constant . Our data indicate that a subset of progressively more posterior Hox genes , which are collinearly activated in vertebral precursors , repress Wnt and FGF activity with increasing strength , leading to a graded repression of the Brachyury/T transcription factor . This progressively reduces mesoderm ingression and cell motility in the PSM , thus slowing down the elongation process . Due to the continuation of somite formation at a steady pace , this mechanism leads to the progressive reduction of PSM size .
We measured the variations of velocities of axis elongation and somite formation in time-lapse videos of developing chicken embryos during the formation of the first 30 somites ( Video 1 ) . The velocity of somite formation shows limited variation during this developmental window ( Tenin et al . , 2010 ) ( Figure 1A , n = 4 embryos for each condition ) . In contrast , axis elongation velocity increases during the formation of the first 10 somites and then it decreases until the 25-somite stage , when it drops abruptly ( Figure 1A and Video 2 , n = 41 embryos ) . The number of PSM cells decreases over time ( Figure 1B , n = 5 embryos for each condition ) while no significant difference in cell proliferation or apoptosis in the PSM and tail-bud is observed ( Figure 1C–F , n = 4 embryos for each condition ) . Cell motility , which has been implicated in the control of axis elongation ( Bénazéraf et al . , 2010 ) , also decreased between 15 and 27 somites ( Figure 1G , n = 4 embryos for each condition ) . Thus , a parallel decrease in cell motility and in cell flux to the PSM accompanies axis elongation slow down . 10 . 7554/eLife . 04379 . 003Video 1 . Time-lapse video of an embryo from Stage 5 HH to 29 somites showing the different phases of axis elongation ( Bright-field , ventral view , anterior is up ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 00310 . 7554/eLife . 04379 . 004Figure 1 . Slowing down of axis elongation correlates with decreasing cell ingression in the PSM . ( A ) Velocity of axis elongation and of somite formation . ( B ) PSM cell number . ( C–D ) Tiling of confocal sections of 20-somite ( C ) and 25-somite ( D ) stage embryos . EdU positive cells are labeled in green , phosphorylated histone H3 ( pH3 ) in red , and nuclei in blue ( DAPI ) . ( C′ , D′ ) Higher magnification of PSM regions used to quantify the proliferation . ( C″ , D″ ) Confocal sections of parasagittal cryosections of tail-bud used to quantify cell proliferation . ( E–F ) Quantification of cell proliferation ( E ) and apoptosis ( F ) in 20–22 and 25–27 somites chicken embryos . ( G ) Cell motility in the posterior PSM . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 00410 . 7554/eLife . 04379 . 005Video 2 . Time-lapse videos showing axis elongation slow-down around the 25-somite stage . Bright-field imaging of chicken embryos at 15–17 somites ( left panel ) , 20–22 somites ( middle panel ) , and 25–27 somites ( right panel ) ( ventral view , anterior is up ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 005 Hoxb1-9 genes were shown to regulate cell flux to the PSM by controlling the timing of cell ingression from the epiblast ( Iimura and Pourquié , 2006 ) . Activation of Hox genes in the epiblast and tail-bud is collinear and occurs in two phases . First , the paralog groups 1 to 8 ( and Hoxb9 ) are quickly activated within ten hours before the first somite formation ( stage 7 HH [Hamburger and Hamilton , 1992]; Figure 2 , n = 8 embryos for each condition ) . This phase is followed by a pause during formation of the first ten somites . Then between the 10 and 40-somite stage , the posterior Hox genes corresponding to the paralog groups 9–13 ( and Hoxc8 and Hoxd8 ) become subsequently activated in a slower phase which takes almost 48 hr ( Figure 2 , n = 8 embryos for each condition ) . Hoxa13 is the first Hox13 activated at the 25-somite stage , when axis elongation slows down abruptly . Thus , there is a striking correlation between the timing of posterior Hox genes activation and the beginning of axis elongation slow down ( Figures 1A and 2 ) . 10 . 7554/eLife . 04379 . 006Figure 2 . Collinear activation of Hox genes in paraxial mesoderm precursors . ( A ) Table showing the collinear onset of Hox genes expression in the epiblast/tail-bud generated from ( B ) Chicken embryos hybridized in whole-mount with Hoxa ( blue ) , Hoxb ( yellow ) , Hoxc ( red ) , and Hoxd ( green ) probes . Each panel shows the beginning of activation of each Hox gene in paraxial mesoderm precursors in the epiblast of the anterior primitive streak or in the tail-bud . Hox probe used is indicated on the top of each panel . Anterior is up . Dorsal view . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 006 In order to test the role of posterior Hox genes on the control of cell ingression and cell motility in the developing chicken embryo , we used an in vivo electroporation technique , allowing to precisely target the paraxial mesoderm precursors in the epiblast of the anterior primitive streak ( Bénazéraf et al . , 2010 ) ( Video 3 ) . We developed a strategy allowing to overexpress two different sets of constructs in largely different populations of paraxial mesoderm cells by performing two consecutive electroporations of the paraxial mesoderm ( PM ) precursors of the epiblast of stage 4–5 HH embryos . Embryos are first electroporated on the left side of the primitive streak with a control Cherry construct , and then on the right side of the streak with a second vector expressing the yellow fluorescent protein Venus and a Hox construct ( Figure 3A ) . This strategy results in essentially different PM cells expressing the two sets of constructs with the Cherry expressing cells enriched on the left side whereas Hox expressing cells are mostly found on the right side . When no Hox construct is present in the Venus vector , the Cherry and Venus-expressing populations of cells were observed to extend from the tail-bud to the same antero-posterior level of the paraxial mesoderm indicating that they began ingressing at the same time ( Figure 3B , n = 8 embryos ) . In contrast , cells expressing Cherry were always extending more anteriorly than cells expressing Venus and one of the following posterior Hox gene: Hoxa9 , Hoxc9 , Hoxd10 , Hoxd11 , Hoxc11 , Hoxa13 , Hoxb13 , or Hoxc13 , indicating that these Hox genes can delay cell ingression of the PSM progenitors ( Figure 3B–C n > 6 for each condition and not shown ) . This simply reflects the fact that cells ingressing later become located more posteriorly . Strikingly , the effect on ingression was progressively stronger when overexpressing more 5′ genes suggesting a collinear trend ( Figure 3C ) . Inverting the order in which the constructs are electroporated did not affect the final phenotype . The distance between the anterior boundaries of the two domains was found to progressively increase with more posterior Hox genes as shown by measuring the ratio between Venus and Cherry posterior domains ( Figure 3A–C ) . Over-expression of Hoxa10 , Hoxc10 , Hoxa11 , Hoxc12 , Hoxd12 and Hoxd13 showed no difference with the control Cherry vector ( Figure 3A–C , n > 6 for each condition and data not shown ) . Using consecutive electroporation of Hoxd10 and Hoxc11 constructs labeled with Cherry and Venus , respectively , we observed that Hoxc11 has a stronger effect on ingression than Hoxd10 ( Figure 3—figure supplement 1 , n = 12 embryos ) . A similar result was observed when Hoxa13 was compared to Hoxc11 in the same assay ( Figure 3—figure supplement 1 , n = 6 embryos ) . Thus , a subset of posterior Hox genes is able to delay PSM cell ingression in a collinear manner . 10 . 7554/eLife . 04379 . 007Video 3 . Time-lapse video showing the precise targeting of PSM progenitors and the ingression of the epiblast cells to form the PSM . Bright-field ( purple ) merged with fluorescent images of PSM cell progenitors electroporated with a control H2B-Venus ( ventral view , anterior is up ) from stage 6 HH onwards . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 00710 . 7554/eLife . 04379 . 008Figure 3 . Posterior Hox genes can regulate cell ingression in a collinear fashion . ( A ) Consecutive electroporation protocol . The ratio of the green domain ( green bar , Hox expressing ) over the red domain ( red bar , control vector ) measures the ingression delay . ( B ) Embryos consecutively electroporated first with Cherry and then with Venus together with control , Hoxa9 , Hoxc11 , or Hoxb13 vectors . Arrowheads: anterior boundary of Cherry ( red ) and Venus ( green ) domains . ( C ) Ratio of Venus over Cherry domains for posterior Hox genes . Dots: electroporated embryos . Bar indicates the mean . Stars: p-value of two-tailed Student's t-test applied between the different conditions . *p < 0 . 05; **p < 0 . 01; ***p < 0 . 005 . Error bars: standard error to the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 00810 . 7554/eLife . 04379 . 009Figure 3—figure supplement 1 . The posterior Hox genes regulate cell ingression with increasing strength . ( A ) Embryos consecutively electroporated with Hoxd10-Cherry and Hoxc11-Venus ( left ) and with Hoxc11-Cherry and Hoxa13-Venus ( right ) . Arrowheads: anterior boundary of Cherry ( red ) and Venus ( green ) domains . ( B ) Ratio of Venus over Cherry domains corresponding to A . This shows that Hoxa13 retains the cell longer in the epiblast than Hoxc11 which retains the cell longer than Hoxd10 . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 009 To analyze the effect of posterior Hox genes on ingression , PM precursors were electroporated with Venus and a Hoxa13 or a control construct and harvested after 5 hr when the electroporated cells start to ingress . No ectopic expression of laminin ( Figure 4A , B–C″ ) , acetylated tubulin ( Figure 4A , D–E″ ) , or E-cadherin ( data not shown ) was observed after Hoxa13 over-expression . We compared the number of Venus-positive cells in epiblast vs primitive streak and mesoderm in embryo sections . The majority of Hoxa13-expressing cells were still found in the epiblast while control cells have ingressed into the primitive streak and mesoderm indicating that Hoxa13 delays ingression by retaining cells in the epiblast ( Figure 4F–H , n = 4 embryos for each condition ) . No up-regulation of the neural marker Sox2 was observed in the tail-bud of embryos electroporated with Hoxa13 ( Figure 4I–J , n = 8 embryos for each condition ) and very few cells were observed in the neural tube of embryos electroporated with Hox constructs ( see Figure 3B , Figure 4I–J , Figure 6A–B , Figure 7D–J and Figure 9A and Videos 4–8 ) . This indicates that the effect on ingression is not caused by the recruitment of PM precursor cells to a neural fate . Ingression of cells from the epiblast to the primitive streak occurs via an epithelium to mesenchyme transition which involves first destabilization and then complete loss of basal microtubules in these cells . This process has been shown to be regulated by a basally localized activity of the small GTPase RhoA ( Nakaya et al . , 2008 ) . In order to test if the effect of the posterior Hox genes on delaying PSM progenitors ingression could involve regulation of microtubule stability , we used a dominant negative form of RhoA ( DN-RhoA ) as a tool to destabilize basal microtubules in the epiblast ( Nakaya et al . , 2008 ) . We performed consecutive electroporations at stage 5 HH to overexpress a control Cherry vector in one population of PSM progenitors and Hoxa13 with DN-RhoA vectors in another population and allowed the embryos to develop for 20 hr . We observed that the two populations of cells reach the same anterior level ( Figure 4K , n = 5/5 embryos ) indicating that these cells ingressed at the same time . Altogether , these results suggest that Hox genes control PSM progenitors ingression through the regulation of basal microtubule stability in the epiblast . 10 . 7554/eLife . 04379 . 010Figure 4 . Epiblast cells overexpressing Hox genes do not convert to a neural fate . ( A ) Transverse section of a stage 7 HH chicken embryo labeled with phalloidin ( white ) to highlight the actin network and with laminin ( red ) to identify the epiblast basal membrane . Colored boxes indicate the different phases of differentiation of the mesoderm: epiblast ( purple ) , ingressing cells ( yellow ) , and mesoderm ( blue ) . ( B–E″ ) Transverse sections at the PSM progenitors level 5 hr after electroporation of a control Venus or of Hoxa13 . ( B-C” ) Laminin immunolabeling ( red ) after Venus ( B–B″ ) or Hoxa13 over-expression ( C–C″ ) . ( D–E″ ) Acetylated α-tubulin immunolabeling ( red ) after Venus ( D–D″ ) or Hoxa13 ( E–E″ ) over-expression . ( F–G ) Transverse cryosections of the anterior primitive streak of an embryo electroporated with Venus ( F ) or with Venus and Hoxa13 ( G ) . White arrow: cells ingressed in the primitive streak ( F ) and non-ingressed epiblast cells ( G ) . Green: Venus; red: laminin; blue: nuclei . ( H ) Quantification of ingression in embryos electroporated with control or Hoxa13-expressing constructs . ( I–J ) In situ hybridization of 2-day old chicken embryos electroporated with Venus ( I ) or Hoxa13-Venus ( J ) expressing vectors . Left panel shows Sox2 expression in the neural tube and tail-bud , and right panels show GFP immunohistochemistry . ( K ) Chicken embryo consecutively electroporated with a control ( Cherry , red ) and a mix of Hoxa13+DN-RhoA ( Venus , green ) . Arrowheads: anterior boundary of Cherry ( red ) and Venus ( green ) domains . Stars: p-value of two-tailed Student's t-test applied between the different conditions . ***p < 0 . 005 . Error bars: standard error to the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 010 We next tested the effect of over-expressing posterior Hox genes on axis elongation ( Figure 5A–C , Video 4 , n = 47 embryos ) . Over-expression of either Hoxa9 , Hoxc9 , Hoxd10 , Hoxd11 , Hoxc11 , Hoxa13 , Hoxb13 or Hoxc13 but not of Hoxa10 , Hoxc10 , Hoxa11 , Hoxc12 , Hoxd12 and Hoxd13 in PM precursors caused a significant decrease of elongation velocity ( Figure 5A–C and not shown ) . The effect of Hox genes becomes progressively stronger for more posterior genes ( Figure 5C and not shown , Video 4 ) . Therefore , the same posterior Hox genes can alter cell ingression and axis elongation with a similar collinear trend ( Figures 3C and 5C ) . The cell-autonomous control of ingression by posterior Hox genes ( Figure 4F–H ) is expected to reduce the supply of motile cells in the posterior PSM . This should slow down elongation movements and could explain why such a non-cell autonomous effect on axis elongation is observed while only 30–50% PM cells express the Hox constructs . These data suggest that a subset of posterior Hox genes controls the slowing down of axis elongation by regulating ingression of PM precursors . 10 . 7554/eLife . 04379 . 020Figure 5 . Posterior Hox genes control the axis elongation velocity in a collinear fashion . ( A–B ) Time-lapse series of chicken embryos electroporated either with control ( A ) or Hoxa13 ( B ) . Red line: position of Hensen's node . ss = somite-stage . ( C ) Velocity of axis elongation of embryos electroporated with either a control , Hoxa9 , Hoxc9 , Hoxd10 , Hoxd11 , Hoxc11 , Hoxa13 , Hoxb13 , or Hoxc13 expressing constructs . Stars: p-value of two-tailed Student's t-test applied between the different conditions . *p < 0 . 05 . Error bars: standard error to the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 02010 . 7554/eLife . 04379 . 015Video 4 . Effect of Hoxa9 , Hoxc11 , and Hoxa13 electroporation on axis elongation and ingression . Bright-field ( purple ) merged with fluorescent images of PSM cell progenitors electroporated with either a control H2B-Venus ( first panel from the left ) , Hoxa9-ires2-H2B-Venus ( second panel from the left ) , Hoxc11-ires2-H2B-Venus ( third panel ) or a Hoxa13-ires2-H2B-Venus ( right panel ) constructs ( green ) ( ventral view , anterior is up ) from Stage 6 HH onwards . Over-expression of Hoxa9 , c11 , and a13 affects ingression and axis elongation in a collinear fashion . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 015 In Drosophila and vertebrates , Hox genes expressed posteriorly can suppress the function of more anterior ones , a property termed phenotypic suppression or posterior prevalence ( Duboule and Morata , 1994 ) . We previously showed that posterior prevalence applies for the control of cell ingression by Hoxb1-9 genes ( Iimura and Pourquié , 2006 ) . To test whether this property also applies to the posterior Hox genes with an effect on axis elongation , we performed consecutive electroporations first with a mix of Hoxd10 and Hoxc11 constructs ( leading to expression in the same cells , in green Figure 6A ) and then with a mix of Hoxc11 and a control construct ( a mutated Hoxc11 unable to bind DNA ( Hoxc11mutH ) , in red Figure 6A ) . We observed that cells over-expressing the two functional Hox genes reach the same anterior position as cells over-expressing Hoxc11 and control ( Figure 6A , C , n = 10 embryos ) . Thus Hoxc11 function is dominant over Hoxd10 . Similarly , we observed dominance of Hoxa13 over Hoxc11 in the same assay ( Figure 6B , C , n = 8 embryos ) . Therefore , posterior prevalence appears to generally apply for Hox control of cell ingression in the mesoderm ( Iimura and Pourquié , 2006 ) . As a result , the effect of Hox genes on cell retention in the epiblast should become progressively stronger as more posterior genes become activated . 10 . 7554/eLife . 04379 . 011Figure 6 . Posterior prevalence of posterior Hox genes . ( A ) Embryos consecutively electroporated first with Hoxc11-Cherry + Hoxc11mutH-Cherry and with Hoxd10-Venus + Hoxc11-Venus shown 24 hr after reincubation . ( B ) Embryos consecutively electroporated first with Hoxa13-Cherry + Hoxa13mutH-Cherry and then with Hoxa13-Venus + Hoxc11-Venus shown 24 hr after reincubation . Red arrowheads: anterior boundary of Cherry-expressing cells . Green arrowheads: anterior boundary of Venus-expressing cells . ( C ) Quantification of the ratio of Venus over Cherry expressing domains for the experiments shown in A and B . Each dot corresponds to one electroporated embryo and bar indicates the mean . ( D–E ) Luciferase assay measuring Wnt/βcatenin pathway activity after over-expression of the BATLuc construct together with a Renilla-expressing vector and either ( D ) control , Hoxa9 , Hoxa13 or the combination of Hoxa9 and Hoxa13 expressing vectors . ( E ) Blow-up of the samples shown in ( D ) . ( F ) BATLuc assay with serial dilutions of the Hoxa13 plasmid ( in μg/μl on the x axis ) . ( G ) Western blot labeled with an anti-HA antibody showing embryos electroporated with Hoxa13 under the control of a doxycycline-responsive promoter activated with different doses of doxycycline ( in μg/ml ) . ( H ) BATLuc assay after Hoxa13 over-expression under the control of a doxycycline-responsive promoter activated with different doses of doxycycline ( in μg/ml on the x axis ) . Stars represent the p value of the two-tailed Student's t-test applied between the different conditions . **p < 0 . 01; ***p < 0 . 005 . Error bars represent the standard error to the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 011 Expression of anterior Hox genes in the primitive streak is maintained during the fast axis elongation phase occurring during the formation of the first ten somites , suggesting that there must be a mechanism blocking their effect on ingression during this time window ( Figure 1A ) . TALE ( Three Amino-acid Loop Extension ) family members have been shown to differentially interact with anterior and posterior Hox genes ( Chang et al . , 1995; Moens and Selleri , 2006 ) . In chicken , the only TALE gene expressed in PM precursors is Pbx1 which is detected in the primitive streak from stage 4 to 7 HH ( Figure 7A , n = 8 embryos for each condition [Coy and Borycki , 2010] ) . Electroporation of a siRNA targeting Pbx1 in the epiblast resulted in strong down-regulation of Pbx1 ( Figure 7B–C , n = 4 embryos for each condition ) . In consecutive electroporations performed first with Cherry and a control siRNA and then with Venus and a siRNA targeting Pbx1 , cells electroporated with the Pbx1 siRNA were found extending more anteriorly than control cells ( Figure 7D , K , n = 19 embryos ) . The effect of Pbx1 siRNA on ingression could be rescued by co-expressing Pbx1 ( Figure 7E , K , n = 16 embryos ) . We compared in consecutive electroporations the effect of expressing first a control siRNA with either Hoxb7 , Hoxb9 , Hoxa9 , Hoxc9 , Hoxd10 , Hoxd11 , Hoxc11 , Hoxa13 , Hoxb13 or Hoxc13 , and then the Pbx1 siRNA with the same Hox gene . Cells co-expressing Hoxb7 or Hoxb9 and the Pbx1 siRNA reached more anterior levels than cells co-expressing these Hox genes and the control siRNA ( Figure 7F–G , K , n = 10 and 15 embryos respectively ) . In contrast , cells co-expressing either a control or the Pbx1 siRNA together with a posterior Hox gene were found to extend up to the same anterior level ( Figure 7H–K and not shown , n > 8 embryos for each condition ) . Over-expression of Pbx1 in PM precursors after the 3-somite stage slowed down axis elongation ( Figure 7L and Video 5 , n = 12 embryos ) , suggesting that Pbx1 can restore the effect of anterior Hox genes on ingression during this time window . Thus , Hox-dependent control of ingression in the paraxial mesoderm requires Pbx1 for anterior but not for posterior Hox genes . 10 . 7554/eLife . 04379 . 012Figure 7 . Control of ingression of PM precursors by anterior Hox genes is dependent on Pbx1 . ( A ) Pbx1 expression during somitogenesis . Red squares: region of PM progenitors . White dashed line: level of transverse section shown in bottom left . ( B–C ) Pbx1 expression in stage 6–7 HH chicken embryos electroporated with Venus and control siRNA ( B ) or Pbx1 siRNA ( C ) . Left panels: Venus expression . ( D–J ) 2-day-old chicken embryos consecutively electroporated first with Cherry and a control siRNA and then with a Pbx1 siRNA and a Venus construct either alone ( D ) or together with Pbx1 ( E ) , Hoxb7 ( F ) , Hoxb9 ( G ) , Hoxc9 ( H ) , Hoxc11 ( I ) , Hoxa13 ( J ) . Arrowheads: anterior boundary of Cherry ( red ) and of Venus ( green ) domains . ( K ) Ratio of Venus over Cherry expressing domains . Dots: electroporated embryos . Bar indicates mean . ( L ) Effect of Pbx1 over-expression on axis elongation rate . Stars represent the p value of the two-tailed Student's t-test applied between the different conditions . ***p < 0 . 005 . Error bars: standard error to the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 01210 . 7554/eLife . 04379 . 016Video 5 . Effect of Pbx1 over-expression between the 5- and 9-somite stage . Bright-field ( purple ) merged with fluorescent images of PSM cell progenitors electroporated with either a control pBIC ( left panel ) or a Pbx1pBIC ( right panel ) construct ( green ) ( ventral view , anterior is up ) . Over-expression of Pbx1 slows down axis elongation . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 016 To identify effector targets regulated by posterior Hox genes , we electroporated epiblast PM progenitors in stage 5 HH embryos , either with a control H2B-Venus or with a Hoxa13-IRES-H2B-Venus vector , and we harvested embryos at 9 somites ( Figure 8A ) . Venus-positive cells were sorted by Fluorescence Activated Cell Sorter ( FACS ) following tail dissociation and their transcriptome was analyzed using Affymetrix microarrays ( Figure 8A , n = 2 × 2 arrays for each condition ) . The Wnt/βcatenin pathway targets , Axin2 , Fgf8 , and Sp8 were down-regulated in Hoxa13 over-expressing cells ( Table 1 , Supplementary file 1 , Figure 8B ) suggesting that posterior Hox genes might control axis elongation rate by progressively down-regulating the Wnt/βcatenin pathway . To test this hypothesis , we first performed in situ hybridizations ( ISH ) for Axin2 , Fgf8 , and T/Brachyury that show that their expression in the tail-bud is down-regulated when Hoxa13 becomes activated ( Figure 8—figure supplement 1A–D , n = 8 embryos for each condition ) . Since the ISH technique is not quantitative enough to resolve slight differences , we performed quantitative Reverse Transcription PCR ( qRT-PCR ) on micro-dissected tail-buds from 10 , 15 , 20 , and 25-somite stages for Axin2 , Fgf8 , and T/Brachyury . These experiments show a slight progressive down-regulation of these genes from the 10 to 20-somite stage followed by a significant decrease in gene expression at the 25-somite stage correlating with the slowing down of axis elongation as well as with the timing of posterior genes expression ( Figure 8C–E , n = 5 embryos for each stage ) . Co-electroporation of Hoxd10 , Hoxc11 , or Hoxa13 with βcatLEF ( which activates the Wnt/βcatenin pathway [Galceran et al . , 2001] ) rescues axis elongation ( Figure 8F–H , Videos 6–8 , n = 41 ) . We co-electroporated a Wnt/βcatenin firefly luciferase reporter ( BATLuc ) and a CMV-Renilla luciferase construct in PM progenitors together with either Venus or Hoxa9 , Hoxc9 , Hoxd10 , Hoxd11 , Hoxc11 , Hoxa13 , Hoxb13 , or Hoxc13 . These Hox genes induced a down-regulation of luciferase activity which increased in a collinear fashion ( except for Hoxd10 and Hoxd11 which showed a weaker effect ) ( Figure 8I and Figure 8—figure supplement 2A , n = 83 embryos ) . All together , these results strongly suggest that the posterior Hox genes control axis elongation by modulating Wnt/βcatenin signaling activity . When co-expressing Hoxa9 and Hoxa13 , the Wnt-repressive effect was equivalent to that of Hoxa13 , indicating that posterior prevalence also applies to Wnt repression ( Figure 6D–E , n = 30 embryos ) . By expressing various amounts of Hoxa13 , we observed that Wnt repression is independent of the quantity of protein expressed ( Figure 6G–H , n = 62 embryos ) , suggesting that Hox proteins levels are saturating in our experiments . Therefore , the same posterior Hox genes can regulate ingression , axis elongation , and Wnt signaling with strikingly similar collinear trends . 10 . 7554/eLife . 04379 . 021Figure 8 . Collinear repression of Wnt/βcatenin signaling by posterior Hox genes . ( A ) Design of the microarray experiment . ( B ) Validation by Q-RT PCR of selected Hoxa13 targets identified in the microarray experiment . ( C-E ) Q-RT PCR for ( C ) T/Brachyruy , ( D ) Axin2 , and ( E ) Fgf8 at 10 , 15 , 20 , and 25-somite stage from microdissected tail-buds . ( F–H ) elongation velocity of embryos over-expressing ( F ) Hoxd10mutH , Hoxd10 or Hoxd10+βcatLEF , ( G ) Hoxc11mutH , Hoxc11 or Hoxc11+βcatLEF , ( H ) Hoxa13mutH , Hoxa13 or Hoxa13+βcatLEF . ( I ) Luciferase assay measuring Wnt/βcatenin activity after over-expression of BATLuc together with CMV-Renilla and either control , Hoxa9 , Hoxc9 , Hoxd10 , Hoxd11 , Hoxc11 , Hoxa13 , Hoxb13 , or Hoxc13 . ( J–M ) Luciferase assay measuring Wnt/βcatenin activity after over-expression of BATLuc and CMV-Renilla and control , Hoxa13 , Hoxa13+dBC , or Hoxa13+Lrp6ΔN ( J ) , or control , Hoxa13 , Hoxa13+Wnt3a or Hoxa13+Wnt5a ( K ) , or control , Fzd2 , Hoxa13 , or Hoxa13+Fzd2 ( L ) or control and Dact2 ( M ) . Firefly luciferase intensity values have been normalized to their respective Renilla values ( RLU ) . Controls have been set to 1 . Stars: p value of the two-tailed Student's t-test applied between the different conditions . *p < 0 . 05; **p < 0 . 01; ***p < 0 . 005 . Error bars represent standard error to the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 02110 . 7554/eLife . 04379 . 022Figure 8—figure supplement 1 . The Wnt signaling is repressed when posterior Hox genes are activated . ( A–F ) In situ hybridization of 15-somite ( left panels ) and 25-somite stage ( right panels ) embryos hybridized with Hoxa13 ( A ) , Axin2 ( B ) , Fgf8 intronic ( C ) , T intronic ( D ) , Fzd2 ( E ) , and Dact2 ( F ) ( red arrowhead: tail-bud ) showing a repression of the Wnt targets and components as well as an upregulation of the Wnt inhibitor Dact2 when Hoxa13 start to be expressed in the tail-bud . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 02210 . 7554/eLife . 04379 . 023Figure 8—figure supplement 2 . Collinear repression of Wnt signaling and cell motility by posterior Hox genes . ( A ) Graph showing Figure 8I samples after removal of control and Hoxd10 and Hoxd11 ( which have a weaker effect ) to highlight the collinear trend of this set of Hox genes on Wnt repression . ( B ) Cell motility measured in the posterior PSM of embryos electroporated with H2B-Venus and either a control , Hoxb1 , Hoxa5 , Hoxc11 or Hoxa13 . Stars represent the p-value of the two-tailed Student's t-test applied between the different conditions . *p < 0 . 05; ***p < 0 . 005 . Error bars represent the standard error to the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 02310 . 7554/eLife . 04379 . 024Table 1 . List of selected genes of the Wnt and FGF pathways down-regulated or up-regulated following over-expression of Hoxa13 in tail-bud cellsDOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 024GeneAverage ( control ) Standard Dev ( control ) Average ( Hoxa13 ) Standard dev ( Hoxa13 ) Fold changeSp8949 . 9279 . 2483 . 821 . 40 . 51Fzd2139 . 710 . 778 . 58 . 60 . 56Axin2857 . 842 . 5677 . 099 . 10 . 79Dact2415 . 8134 . 4989 . 8270 . 12 . 38Cyp26a1625 . 1258102 . 9130 . 16Fgf81523 . 9159 . 3591 . 2650 . 39Etv1296 . 6113 . 2155 . 123 . 80 . 52Fgfr1145 . 95 . 8800 . 60 . 55Rasgrp31441 . 3671 . 8362 . 8218 . 80 . 2510 . 7554/eLife . 04379 . 017Video 6 . Activation of Wnt/βcatenin signaling and T over-expression rescue Hoxa13 axis elongation phenotype . Bright-field ( purple ) merged with fluorescent images of PSM cell progenitors electroporated with Hoxa13mutH-ires2-H2B-Venus ( left panel ) , Hoxa13-ires2-H2B-Venus ( second panel ) , T and Hoxa13-ires2-H2B-Venus construct ( third panel ) or βcatLEF and Hoxa13-ires2-H2B-Venus construct ( right panel ) ( green ) ( ventral view , anterior is up ) from Stage 6 HH onwards . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 01710 . 7554/eLife . 04379 . 018Video 7 . Activation of the Wnt/βcatenin pathway rescues the axis elongation phenotype due to Hoxd10 over-expression . Bright-field ( purple ) merged with fluorescent images of PSM cell progenitors electroporated with Hoxd10mutH-ires2-H2B-Venus ( left panel ) , Hoxd10-ires2-H2B-Venus ( middle panel ) or βcatLEF and Hoxd10-ires2-H2B-Venus construct ( right panel ) ( green ) ( ventral view , anterior is up ) from Stage 6 HH onwards . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 01810 . 7554/eLife . 04379 . 019Video 8 . Activation of the Wnt/βcatenin pathway rescues the axis elongation phenotype due to Hoxc11 over-expression . Brightfield ( purple ) merged with fluorescent images of PSM cell progenitors electroporated with Hoxc11mutH-ires2-H2B-Venus ( left panel ) , Hoxc11-ires2-H2B-Venus ( middle panel ) , or βcatLEF and Hoxc11-ires2-H2B-Venus construct ( right panel ) ( green ) ( ventral view , anterior is up ) from Stage 6 HH onwards . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 019 We next analyzed how Hox genes interfere with Wnt function . Hoxa13 ingression phenotype is rescued by an activated form of Lrp6 or a stabilized form of Ctbbn1 ( Figure 8J , n = 42 embryos ) but not by Wnt3a or Wnt5a ( Figure 8K , n = 30 embryos ) . This suggests that , genetically , Hox genes act on Wnt signaling at the membrane level . Over-expression of the Wnt receptor Fzd2 ( down-regulated in Hoxa13 over-expressing cells ( Figure 8B , Table 1 and Supplementary file 1 ) ) with Hoxa13 rescued Wnt repression ( Figure 8L , n = 29 embryos ) . Fzd2 is expressed in the tail-bud at 15 somites and down-regulated after 25 somites ( Figure 8—figure supplement 1E , n = 8 embryos ) . Over-expression of the Wnt pathway component Dact2 ( which is expressed in the tail-bud from 25 somites onward and up-regulated in Hoxa13 over-expressing cells [Figure 8—figure supplement 1F , n = 8 embryos , Table 1 , Supplementary file 1] ) , repressed Wnt activity ( Figure 8M , n = 9 embryos ) . In Hoxa13 over-expressing cells , the FGF receptor FGFR1 , its ligand Fgf8 , and its targets Etv1 and Cyp26A1 as well as the FGF pathway component Rasgrp3 , were down-regulated while the FGF/MAPK inhibitor , Spred2 , was up-regulated ( Figure 8B , Table 1 , Supplementary file 1 ) , indicating that Hoxa13 can also inhibit FGF signaling . This inhibition is consistent with the down-regulation of PSM cell motility observed after Hoxc11 or Hoxa13 over-expression ( Figure 8—figure supplement 2B , n = 20 embryos ) ( Bénazéraf et al . , 2010 ) . FGF down-regulation is expected since FGF and Wnt signaling reciprocally regulate each other in PM precursors ( Aulehla et al . , 2003; Naiche et al . , 2011 ) . Down-regulation of Cyp26A1 , which degrades RA , can up-regulate RA signaling leading to repression of the Wnt pathway non cell-autonomously ( Iulianella et al . , 1999; Young et al . , 2009; Martin and Kimelman , 2010 ) . Together , these data suggest that posterior Hox genes act on a gene network converging toward autonomous and non-autonomous negative Wnt regulation . The T-box transcription factor T ( aka Brachyury ) is a well-characterized Wnt target which has been shown to control cell ingression to the mesoderm ( Wilson et al . , 1995; Yamaguchi et al . , 1999 ) . Q-PCR analysis of micro-dissected tail buds shows that T expression levels decrease between 10 and 20-somite stage and then significantly drop at the 25-somite stage ( Figure 8C ) . Over-expressing T by electroporation often resulted in PM-expressing cells extending more anteriorly than control cells suggesting that they ingress earlier ( Figure 9A–B , n = 6 embryos ) . Over-expression of T together with either Hoxa9 , Hoxd10 , Hoxc11 or Hoxa13 rescued the ingression delay ( Figure 9A–B , n = 6 , 11 , 10 and 7 embryos respectively ) . T also rescued the elongation slow down observed after Hoxa13 over-expression ( Video 6 , Figure 9C , n = 4 embryos ) . A lower dose of T ( 0 . 5 μg/μl ) only led to partial rescue of the Hoxa13 phenotype ( Figure 9C , n = 4 embryos ) . Endogenous T expression is down-regulated in Hoxa13 over-expressing cells FACS-sorted from electroporated embryos ( Figure 9D , n = 2 FACS sorted cell samples for each condition ) . Over-expression of a reporter generated by fusing one kilobase of the chicken T promoter to the firefly luciferase ( cTprLuc ) together with Hoxc11 or Hoxa13 and the CMV-Renilla luciferase show T repression which is stronger for Hoxa13 ( Figure 9E , n = 19 embryos ) . Over-expression of βcatLEF leads to T up-regulation ( Figure 9F , n = 20 embryos ) and co-expression of Hoxa13 with βcatLEF totally rescues T repression ( Figure 9F , n = 20 embryos ) suggesting that Hox genes down-regulate T expression by repressing the Wnt/βcatenin pathway . Over-expressing T had no effect on BATluc activation ( Figure 9—figure supplement 1 , n = 8 embryos ) . This argues that the effect of Hox genes on epiblast ingression involves quantitative regulation of T expression levels . 10 . 7554/eLife . 04379 . 013Figure 9 . Hox genes effect on axis elongation involves Brachyury regulation downstream of the Wnt/βcatenin pathway . ( A ) Consecutive electroporation of PM precursors with Cherry and then with Venus together with T ( left panel ) , Hoxa13 ( middle ) , or a combination of the two vectors ( right ) . Arrowheads: anterior boundary of Cherry ( red ) and Venus ( green ) domains . ( B ) Ratio of Venus over Cherry domains . Dots: electroporated embryos . Bar indicates the mean . ( C ) Axis elongation velocity of embryos electroporated with control , Hoxa13 , or co-electroporated with Hoxa13 and either high or low dose of T . ( D ) Q-RT PCR quantification of T expression in control or Hoxa13-expressing PM progenitor cells . ( E–F ) Luciferase activity ( RLU ) after over-expression of cTprLuc and CMV-Renilla together with either ( E ) control , Hoxc11 or Hoxa13 or ( F ) control , βcatLEF , Hoxa13 or Hoxa13+βcatLEF . ( G–I ) Luciferase assay measuring Wnt/βcatenin activity after over-expression of BATLuc and CMV-Renilla and ( G ) Hoxa13mutH , Hoxa13dn , Hoxa13+Hoxa13mutH or Hoxa13+Hoxa13dn . ( H ) Hoxd10mutH , Hoxd10dn , Hoxd10+Hoxd10mutH or Hoxd10+Hoxd10dn , ( I ) Hoxc11mutH , Hoxc11dn , Hoxc11+Hoxc11mutH or Hoxc11+Hoxc11dn . ( J ) Luciferase assay measuring Wnt/βcatenin activity from 28-somite stage dissected tail-buds after over-expression of BATLuc and CMV-Renilla constructs and either Hoxa13mutH or Hoxa13dn , or Hoxa13mutH with Hoxc11mutH or Hoxa13dn with Hoxc11dn , or Hoxa13mutH with Hoxc11mutH and Hoxd10mutH or Hox13dn with Hoxc11dn and Hoxd10dn . ( K , L ) Q-RT PCR quantification of T , Axin2 ( K ) , and Fzd2 ( L ) expression in PM progenitors co-expressing either Hoxa13mutH with Hoxc11mutH and Hoxd10mutH or Hoxa13dn with Hoxc11dn and Hoxd10dn . Stars: p-value of the two-tailed Student's t-test applied between the different conditions . *p < 0 . 05; **p < 0 . 01; ***p < 0 . 005 . Error bars: standard error to the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 01310 . 7554/eLife . 04379 . 014Figure 9—figure supplement 1 . Overexpression of T has no effect on Wnt activity . Luciferase assay measuring Wnt/βcatenin pathway activity 20 hr after over-expression of BATLuc and Renilla constructs together with control , T , Hoxa13 , or Hoxa13+T . Stars represent the p-value of the two-tailed Student's t-test applied between the different conditions . *p < 0 . 05; ***p < 0 . 005 . Error bars represent the standard error to the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 014 A Hoxa13 truncated form is responsible for the dominant Hand-Foot-Genital syndrome in man ( Mortlock and Innis , 1997 ) . A similar truncation in the chicken homolog ( Hoxa13dn ) acts as a dominant-negative inhibiting the function of all Hox13 genes ( de Santa Barbara and Roberts , 2002 ) . When over-expressed before activation of Hox13 paralogs , Hoxa13dn had no effect on BATluc activity ( Figure 9G , n = 18 embryos ) . However , co-expression with Hoxa13 in similar conditions abolished Wnt repression ( Figure 9G , n = 18 embryos ) . Similar truncations in chicken Hoxd10 ( Hoxd10dn ) and Hoxc11 ( Hoxc11dn ) also exert a dominant-negative effect on their wild-type counterparts ( Figure 9H–I , n = 38 embryos ) . We over-expressed Hoxa13dn alone or combined with either Hoxc11dn or with Hoxc11dn and Hoxd10dn along with BATLuc and CMV-Renilla constructs in PM precursors of the streak at stage 8 HH . Embryos were harvested at the 28-somite stage when most Hox10-13 paralogs are expressed . Increasing the number of dominant-negative constructs results in a corresponding increase in luciferase activity ( Figure 9J , n = 35 embryos ) . We next co-expressed the three dominant-negative vectors Hoxd10dn , Hoxc11dn , and Hoxa13dn together , along with Venus and FACS-sorted dissociated Venus-positive cells from tail-buds of 28-somite embryos . qRT-PCR analysis of T , Axin2 , and Fzd2 in the Venus-positive cells shows up-regulation of the three genes ( Figure 9K-L , n = 4 embryos for each condition ) . All together these results argue that a subset of posterior Hox genes gradually represses Wnt/βcatenin signaling and consequently T/Brachyury in paraxial mesoderm precursors of the epiblast . This progressive repression leads to reduced cell ingression and cell motility in the PSM , resulting in a slowing down of axis elongation . In order to identify the domain of posterior Hox proteins involved in repressing T/Brachyury expression , we generated chimera proteins where the different regions ( N-terminal , homeodomain and C-terminal ) of different posterior Hox proteins are swapped with the equivalent region of Hoxa5 which has no effect on axis elongation , Wnt activity and T/Brachyury expression ( Figure 10A–B ) . Over-expression of cTprLuc along with a chimera where the homeodomain of Hoxa5 has been swapped with the one from Hoxa13 ( Hoxa5Ha13 ) does not show any repression of luciferase activity while over-expression of a chimera where the homeodomain of Hoxa13 has been swapped with the one from Hoxa5 ( Hoxa13Ha5 ) shows a strong repression of luciferase activity ( Figure 10A–B , n = 35 embryos ) suggesting that the homeodomain does not contain the major domain responsible for T/Brachyury repression . We next tested if either the N-terminal domain ( N-ter ) or the C-terminal domain ( C-ter ) is responsible for T/Brachyury repression . Overexpression of a chimera where the N-ter of Hoxa5 is swapped with the N-ter of Hoxa13 ( NHoxa13HCa5 ) shows a strong repression of luciferase activity while a chimera where the C-ter of Hoxa5 is swapped with the C-ter of Hoxa13 ( Hoxa5Ca13 ) does not show any repression ( Figure 10C–D , n = 16 embryos ) suggesting that the N-ter region of Hoxa13 contains the domain responsible for the repression of T/Brachyury . Sequence alignment of the N-terminal regions of Hoxa9 , d10 , c11 , and a13 shows little conservation at the amino acid level suggesting that it is not a conserved amino acid domain but rather a structural domain that is responsible for the repression activity of these proteins ( Figure 10—figure supplement 1 ) . We next tested if the nature of the homeodomain could have a role in refining the level of repression of T by designing chimeras where the homeodomain of Hoxc11 and Hoxa13 was replaced by the homeodomain of Hoxa5 ( Hoxc11Ha5 and Hoxa13Ha5 , respectively ) ( Figure 10E ) . With the wild-type proteins , we observe a stronger downregulation of T/Brachyury with Hoxa13 than with Hoxc11 ( Figure 9E ) . Surprisingly , when we overexpress Hoxc11Ha5 or Hoxa13Ha5 along with cTprLuc , we observe a stronger down-regulation of T/Brachyury with the chimera containing the Hoxc11 N-ter than with the one containing the Hoxa13 N-ter ( Figure 10E–F , n = 26 embryos ) suggesting that the homeodomain could be responsible for fine tuning T/Brachyury repression . Altogether our data suggest that the progressive deployment of posterior Hox genes in PM precursors during axis elongation leads to a collinear repression of the Wnt/βcatenin pathway and its target T/Brachyury . 10 . 7554/eLife . 04379 . 025Figure 10 . The N-terminal region of posterior Hox genes contains the repressive domain . ( A , C , E ) Design of the Hox chimeras . N-ter is in blue , the homeodomain in white , and the C-ter in red . ( B , D , F ) Luciferase assay measuring T/brachyury expression 20 hr after over-expression of cTprLuc and Renilla constructs together with ( B ) control , Hoxa5 , Hoxa13 , Hoxa5Ha13 , or Hoxa13Ha5 , ( D ) control , Hoxa13 , NHox13HCa5 , or Hoxa5Ca13 , ( E ) control , Hoxc11Ha5 , or Hoxa13Ha5 . Stars represent the p-value of the two-tailed Student's t-test applied between the different conditions . ***p < 0 . 005 . Error bars represent the standard error to the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 02510 . 7554/eLife . 04379 . 026Figure 10—figure supplement 1 . The N-ter region of posterior Hox genes is poorly conserved at the amino-acid level . ClustalW alignment of the N-ter region of Hoxa9 , Hoxd10 , Hoxc11 , and Hoxa13 shows poor conservation at the amino acid level . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 026
Here , we show that a subset of posterior Hox genes represses Wnt/T signaling with increasing strength showing a collinear trend . We observe a similar collinear effect of these posterior genes overexpressed in PM precursors on delaying their ingression in the PSM and on the slowing down of axis elongation . This inhibition of Wnt signaling is accompanied by a down-regulation of FGF signaling which was shown to control elongation velocity by regulating cell motility in the PSM ( Bénazéraf et al . , 2010 ) . This suggests that posterior Hox genes are involved in the slowing down of axis elongation by acting both on the flux of cells in the posterior PSM and on the motility of PSM cells . In the chicken embryo , PM precursors originate initially from the lateral epiblast which migrate toward the midline during formation of the primitive streak ( Selleck and Stern , 1991; Hatada and Stern , 1994 ) . Around stage 4 HH , somite precursors begin to ingress from the superficial epiblast of the anterior primitive streak and posterior Node region ( Psychoyos and Stern , 1996; Iimura et al . , 2007 ) . Two types of PM precursors have been identified in chicken and mouse embryos ( Wilson et al . , 2009 ) . A first set derives from the Node/primitive streak border and exhibits long-term self-renewal properties ( Selleck and Stern , 1991; Cambray and Wilson , 2002 , 2007 ) . These cells express Sox2 and Brachyury and they can contribute both to the PM ( mostly to the medial part of the somites ) and to the neural tube ( Ordahl , 1993; McGrew et al . , 2008; Tzouanacou et al . , 2009; Kondoh and Takemoto , 2012; Olivera-Martinez et al . , 2012 ) . A second set derives from the anterior portion of the primitive streak and contributes to shorter clones restricted to the PM ( Iimura et al . , 2007; McGrew et al . , 2008; Tzouanacou et al . , 2009 ) . After stage 4 HH , in the chicken embryo , the primitive streak begins to regress and after stage 13-14 HH , it becomes part of the tail-bud ( Schoenwolf , 1979 ) . At the 25-somite stage ( stage 15 HH ) , the posterior neuropore closes and the tail-bud becomes enclosed into the tail fold . During these stages , PM precursors are continuously produced first by the primitive streak and then by the tail-bud . Fate mapping of the 25-somite stage tail-bud with quail-chick chimeras and diI labeling showed that the formation of the PM follows morphogenetic movements very similar to that seen earlier at the primitive streak level during gastrulation ( Catala et al . , 1995; Knezevic et al . , 1998 ) . After this stage , the remnant of the primitive streak becomes localized ventrally to form a structure known as the Ventral Ectodermal Ridge ( VER ) ( Schoenwolf , 1979; Goldman et al . , 2000 ) . Whether cell ingression continues after posterior neuropore closure to generate the PM is not well established . Knezevic et al . reported that cell ingression from the VER stops at stage 16 HH ( 26–28 somites ) ( Knezevic et al . , 1998 ) but Ohta et al . demonstrated that ingression into the mesoderm continues in the VER up to the 40-somite stage ( stage 20 HH ) ( Ohta et al . , 2007 ) . There is also some lineage continuity at the level of the PM precursors of the Node/primitive streak border which were shown to become internalized to become part of the chordo-neural hinge in the tail-bud . DiI labeling of the late chordo-neural hinge in stage 20–22 HH ( 40–45 somites ) embryos showed that mesoderm cells are produced by this structure at late stages ( Olivera-Martinez et al . , 2012 ) . The cellular organization of the chordo-neural hinge has not been characterized and whether mesoderm production by this structure occurs through ingression movements involving an epithelium to mesenchyme transition as is seen for the production of paraxial mesoderm from the primitive streak is not established . Overall , very little is known about the movements of cells in the tail-bud after the 25-somite stage in chicken and mouse embryos . The respective contribution of the VER and the CNH to the PM at these late stages has not been characterized . If Knezevic et al . are correct , it could be that the action of posterior Hox genes on ingression ends at the 25-somite stage when Hoxa13 is first expressed . The strong effect of Hoxa13 on ingression might trigger the arrest of cell ingression leading to the slowing down of axis elongation observed at this stage . The resulting imbalance between the velocity of somitogenesis and of axis elongation could account for the progressive shortening of the PSM observed during the production of the next 25–28 somites . Alternatively , it could be that , as suggested by Ohta et al . and by Olivera-Martinez et al . , ingression continues up to the 40–45 somite stage . In this case , Hoxa13 expression at the 25-somite stage would only significantly reduce the rate of cell ingression into the PM . At the 40–43 somite stage , Hoxb13 and Hoxc13 become expressed in the tail-bud potentially terminating further ingression . Whether Hox13 genes might regulate the late ingression of PM at the level of the VER , as shown by Ohta et al . , or at the level of the CNH as proposed by Olivera-Martinez remains to be established . In both cases , however , the PSM is expected to shrink in response to Hox13 genes . Even though our experiments do not directly address the process whereby axis elongation stops , they suggest that Wnt and FGF repression in the tail-bud , which signals termination of axis formation ( Olivera-Martinez et al . , 2012 ) , could be mediated by posterior Hox genes . By reducing the flux of cells to the PSM and their motility , posterior Hox genes can indirectly control its progressive shortening . Furthermore , the inhibition of FGF and Wnt signaling which are required for the segmentation clock oscillations provides an explanation for the arrest of somite formation before the complete exhaustion of the PSM described in avians ( Bellairs , 1986; Tenin et al . , 2010 ) . In vivo , the downregulation of the FGF target Cyp26A1 downstream of Hox13 genes ( this report , Young et al . , 2009 ) would leave the tail-bud more vulnerable to the increase of RA . Whether , the raise in RA levels caused by bringing the segmented region closer is also responsible for raldh2 activation in the late tail-bud remains to be explored . In such scenarios , posterior Hox genes indirectly control the termination of axis elongation and hence the segment number in the chicken embryo . In mouse embryos , over-expression of Hox13 genes results in axis truncation posterior to the thoracic level ( Young et al . , 2009 ) . Remarkably , overexpression of Hoxa13 , b13 , and c13 from the same promoter in transgenic mice results in truncations at different antero-posterior levels ( Young et al . , 2009 ) , arguing for different truncation efficiency of the mouse Hox13 proteins . This is highly reminiscent of our observations showing different quantitative effects of the overexpression of the same three Hox13 genes in chicken embryos . Duplications and deletions of regions of the mouse Hoxd cluster lead to heterochronic expression of posterior Hoxd genes in the tail-bud yet they do not seem to alter segment numbers ( Spitz et al . , 2001; Kmita et al . , 2002; Tarchini and Duboule , 2006; Tschopp et al . , 2009 ) . This is also consistent with our observations that Hoxd genes have limited effect on axis elongation in our experiments . In transgenic mice overexpressing Hoxc13 , Wnt targets and the FGF target Cyp26A1 were also found to be down-regulated ( Young et al . , 2009 ) as observed in chicken embryos overexpressing Hoxa13 . This argues for a conserved role of posterior Hox proteins in the repression of the Wnt and FGF pathway between chicken and mouse embryos . In mouse embryos however , no strong raldh2 expression or late RA production is detected in the tail bud ( Tenin et al . , 2010 ) and axis elongation continues for a longer time resulting in tail formation . Moreover , raldh2 −/− mouse embryos which lack RA production during posterior body formation can form normal tails , suggesting that RA is not involved in axis termination in mouse ( Cunningham et al . , 2011 ) . In mouse embryos , Wilson and Beddington ( 1996 ) initially reported an arrest of ingression when the posterior neuropore closes at the 30-somite stage ( Wilson and Beddington , 1996 ) , but Cambray and Wilson , 2002 subsequently provided evidence for continued ingression of cells in the PM after this stage ( Cambray and Wilson , 2002 ) . Thus , in mouse embryos , termination of axis elongation could simply result from exhaustion of PM progenitors caused by the slowing of axis elongation triggered by posterior Hox genes acting on cell ingression and motility . Remarkably , among amniotes , many species such as lizards , rodents , or monkeys bear a long tail whereas others such as birds or humans do not . Closely related species such as monkeys and apes can differ by the presence of a tail suggesting that the genetic switch involved in the control of tail formation is quite simple . Whether this switch involves an RA-dependent elongation arrest mechanism as seen in chicken and whether this control depends on posterior Hox genes is an attractive possibility which remains to be investigated . Our work provides evidence for functional collinearity in the control of axis elongation by posterior Hox genes . Our data also suggest that our overexpression conditions are saturating ( Figure 6 ) , abolishing any effect of gene dosage of the overexpressed Hox genes . This confirms previous results published in Iimura and Pourquié , 2006 showing that Hoxb1-9 gene expression driven by promoters of different strength ( CMV , TK , and CAGGS ) leads to similar ingression phenotypes . Together , this suggests that the information driving the quantitative effects of Hox proteins on Wnt repression , cell ingression , and elongation is built in the structure of the proteins themselves rather than reflecting the actual amounts of Hox proteins present . This functional collinearity might be related to the recently described structural collinearity of binding specificities reported for fly Hox proteins ( Slattery et al . , 2011 ) . Our work suggests that low amounts of posterior Hox protein levels could be saturating in vivo . This is consistent with the analysis of paralog knock-out experiments showing that leaving only one single wild-type allele leads to a much milder phenotype than the deletion of an entire paralog group ( Wellik and Capecchi , 2003; McIntyre et al . , 2007 ) . Also , increasing Hox doses by adding an extra mouse or human HoxD cluster does not alter the vertebral formula ( Spitz et al . , 2001; Kmita et al . , 2002; Tarchini and Duboule , 2006; Tschopp et al . , 2009 ) . The fact that low levels of Hox proteins are saturating could confer great robustness to the system consistent with the extreme stability of intraspecific vertebral formula . That 8 of the 16 posterior Hox genes from all posterior paralog groups except Hox12 show an effect in the ingression , elongation , and Wnt signaling assays argue for an extreme redundancy of the system that could further explain the intraspecific robustness of the vertebral formula . We observe a trend showing an increasing strength of the effects on cell ingression , Wnt repression , and axis elongation when overexpressing progressively more 5′ Hox genes . This increasing trend can be partly accounted for by the posterior prevalence of posterior Hox genes observed in the control of ingression and in the repression of Wnt signaling for genes of different paralog groups . However , different quantitative effects were observed for genes from the same paralog groups arguing against a simple posterior prevalence model . This is in line with the result of inactivation of the entire paralogs groups such as Hox10 or Hox11 which demonstrates specific properties of each of these paralog groups arguing against a simple posterior prevalence model functioning in vertebral patterning ( Wellik and Capecchi , 2003; McIntyre et al . , 2007 ) . We identify a role for the TALE protein Pbx1 in the control of cell ingression from the epiblast into the PSM by anterior but not posterior Hox genes . TALE homeoproteins have been shown to act as co-factors able to enhance DNA binding specificity of Hox genes ( Moens and Selleri , 2006 ) . Pbx proteins bind anterior Hox proteins via a specific hexapeptide sequence ( Chang et al . , 1995 ) . The null mutation of Pbx1 in mouse leads to patterning defects of the axial skeleton but axis length appears essentially normal ( Selleri et al . , 2001 ) . In contrast , double mutants for Pbx1 and Pbx2 often show a smaller number of somites suggesting that these two genes could act redundantly in patterning the axial skeleton ( Capellini et al . , 2008 ) . In Pbx1−/−; Pbx2+/− mutants , anterior shifts of Hox expression boundaries in the paraxial mesoderm have been reported ( Capellini et al . , 2008 ) . Such shifts are consistent with a precocious ingression of cells normally fated to a more posterior identity . Genetic studies on mouse T mutants have shown that graded T activity is required for body axis formation ( Stott et al . , 1993; Wilson and Beddington , 1997 ) . Embryos with progressively lower quantities of T exhibit more severe axis truncations ( Stott et al . , 1993 ) . Similar graded truncations are also observed for Wnt3a allelic series ( Galceran et al . , 1999 ) , indicating that precise quantitative regulation of this pathway is required for completion of body axis elongation . Repression of the Wnt pathway and of T together with axis truncations was also observed in Hox13 over-expressing transgenic mice ( Young et al . , 2009 ) . Our data suggest that the gradient of T activity is established by the graded regulation of Wnt signaling by posterior Hox genes , ( Figure 11 ) thus providing a possible explanation for these complex phenotypes . At the cellular level , it argues that the Hox-dependent regulation of T levels in the epiblast is critical to control the balance between cell ingression and maintenance of a self-renewing paraxial mesoderm progenitor pool in the epiblast/tail-bud . Cell ingression requires an EMT that involves destabilization of the basal microtubules of epiblast cells followed by basal membrane breakdown ( Nakaya et al . , 2008 ) . Inhibiting Rhoa activity can rescue the ingression delay caused by Hoxa13 overexpression , suggesting that posterior Hox genes can control cell flux to the PM by acting on basal microtubule stabilization in epiblast cells . As T is also able to rescue Hoxa13 phenotype on elongation , it could act upstream of this process and the details of such a molecular pathway remain to be investigated . 10 . 7554/eLife . 04379 . 027Figure 11 . Model representing the 3 phases ( I , II , and III ) of Hox action in PM precursors in the epiblast/tail-bud during axis elongation . Model representing the 3 phases ( I , II , and III ) of Hox action in PSM precursors in the epiblast/tail-bud during body axis elongation . Anterior Hox genes ( paralogs 1–9 ) are expressed during phase I . They control cell ingression in a Pbx1-dependent manner leading to the collinear positioning of Hox genes expression domains in the anterior region of the embryo . No Hox genes are activated during phase II , allowing fast elongation of the embryonic axis . During phase III , posterior Hox genes ( paralogs 9–13 ) are collinearly activated in PSM precursors . Our data suggest that collinear activation of posterior Hox genes leads to repression of Wnt signaling and its target T/Brachyury , which progressively increases in strength . This results in a progressive arrest of cell ingression in the PSM , leading to a decrease in axis elongation rate . Since the velocity of somite formation is roughly constant , PSM size starts to decrease when elongation velocity becomes slower than that of somite formation . During this latter phase the control of cell ingression by posterior Hox genes appears to be independent of Pbx1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 027 Wnt signaling was proposed to promote the paraxial mesoderm fate at the expense of the neural fate in a population of bipotential neuro-mesodermal stem cells in the tail bud ( Martin and Kimelman , 2010 , 2012; Gouti et al . , 2014; Tsakiridis et al . , 2014 ) . Thus , the Wnt repression experienced by epiblast cells in response to posterior Hox genes overexpression could induce these cells toward a neural fate hence preventing them to ingress . However , Hoxa13 overexpression does not lead to up-regulation of the neural marker Sox2 in electroporated cells as detected by in situ hybridization and in our microarray analysis . Furthermore , electroporated cells are seen to enter the PM and do not enter the neural tube ( Figure 4 and supplementary videos ) . This therefore suggests that posterior Hox genes are unlikely to control cell ingression by promoting acquisition of a neural fate in epiblast cells . While we cannot completely rule out that a subpopulation of these cells remains in the tail-bud as Sox2-positive cells , it is unlikely that this contributes to the dramatic axis elongation slow down observed after posterior Hox genes overexpression . Thus our data suggest that disruption of the balance between epiblast and ingressing cells can be achieved by interfering with T levels either directly or indirectly by altering Wnt levels or Hox expression . The graded repressive activity of posterior Hox genes on the Wnt/T pathway might provide an evolutionary constraint that led to the selection of collinearity of posterior Hox genes ( Duboule , 2007 ) .
Fertilized chicken eggs were obtained from commercial sources . Eggs were incubated at 38°C in a humidified incubator for approximately 24 hr . Embryos were prepared for Early Chick ( EC ) cultures ( Chapman et al . , 2001 ) and then electroporated . Embryos were staged following the Hamburger and Hamilton ( HH ) table ( Hamburger , 1992 ) and by counting somites ( somite stage: ss ) . Electroporation of the paraxial mesoderm ( PM ) precursors of the epiblast was carried out as described in Bénazéraf et al . ( 2010 ) . For the axis elongation assay , the electroporation success is first monitored 3 hr after by examining the embryos under a fluorescent stereomicroscope . Only embryos successfully electroporated in the PSM progenitors ( 90–100% ) are processed for videos ( see the axis elongation measurement section ) . For the luciferase assay embryos are electroporated in the PM progenitors and examined 20 hr after electroporation ( see the Luciferase assay section ) . In both assays , we finally obtain 90–100% of the electroporated embryos showing reporter expression restricted to the paraxial mesoderm . In rare cases ( less than 10% ) , we observed few cells in the lateral plate mesoderm . These embryos were discarded . To better illustrate the accuracy of the electroporations performed , we now present a video showing the expression of a control construct electroporated in the anterior epiblast and showing the ingressing PSM cells ( Video 3 ) . To over-express two sets of constructs in different somitic precursors of the epiblast , two consecutive electroporations of the anterior primitive streak ( PS ) were carried out . Embryos were first prepared for EC culture at room temperature which pause their development . The first construct mixed with the control vector pCX-MyrCherry ( gift from X Morin ) was microinjected on one side of the PS groove and the first electroporation was carried out as described above right after injection . Only 10 s after the first electroporation , the second construct , containing the gene to over-express ( Hox , T… ) cloned in the pCAGGS-I2-MyrVenus , was then microinjected at the same level on the other side of the PS groove and immediately electroporated . This procedure targets the entire paraxial mesoderm territory of the epiblast of the anterior primitive streak on the electroporated side . Thus , in these experiments , we only track the timing of ingression of the most anterior epiblast cells which give rise to the most anteriorly localized progeny in the paraxial mesoderm . By biasing the position of the electrode on the right or on the left side of the primitive streak , we can ensure that the electroporation is biased on one side allowing easier observation of the anterior boundaries of the over-expressing cells . In most cases , however , some electroporated cells are seen on both sides despite this bias . The control Cherry vector is usually found to be more bilateral than the Hox expressing Venus-positive cells as can be seen from the pictures , which reflects an effect specific to Hox genes . Inverting the order of the electroporated plasmids in consecutive electroporation has no effect on the outcome of the experiment , and similar results are observed when the entire somitic territory of the epiblast of the anterior streak is electroporated . Following consecutive electroporation , embryos were cultured in a humidified incubator at 38°C to resume their development . 3 hr after electroporation , the embryos were screened based on fluorescence to ensure that both constructs were expressed and that the correct region was targeted for each construct . At this stage , up to 70% of the embryos are successfully electroporated . The embryos were then reincubated at 38°C until they reach the 15-somite stage ( ∼20 hr ) . As can be seen on the embryos shown in Figure 3 , Figure 6 , and Figure 9 or in the videos , virtually no cells are seen outside of the paraxial mesoderm meaning that the electroporation accurately targeted the anterior streak epiblast . Since expression of the constructs is driven by the ubiquitous CAGGS promoter , even a slight inaccuracy in positioning the electrode would result in expression in the neural tube or the lateral plate . A fluorescent stereomicroscope ( Leica M205 FA ) equipped with a color camera was used to track anterior boundaries of both control Cherry-expressing cells and mutant Venus-expressing cells . At this stage , up to 60% of the electroporated embryos are exclusively electroporated in the PSM and retained for further analyses ( ∼10% are discarded because of developmental defects due to the consecutive electroporation procedure ) . We used the ‘Measure’ plugin of ImageJ to measure the distance between the tail-bud and anterior boundary of both Venus and Cherry-expressing domains in the same embryos . We used these measures to calculate the ratio of Venus over Cherry domains . The Dot-Plot resulting from these ratios was generated using Graphpad 5 ( Prism ) . Paraxial Mesoderm ( PM ) progenitors in the anterior primitive streak were electroporated in Stage 5 HH embryos with either a control vector pCAGGS-Venus or pCAGGS-Hoxa13-IRES2-Venus and cultured in a humidified chamber at 38°C for 5 hr . Embryos were then fixed in 4% paraformaldehyde at room temperature for 40 min and immunolabeled for GFP , laminin , and DAPI as described below . Embryos were then mounted and imaged with a Zeiss 510 NLO equipped with a 20× NA 0 . 8 objective . 80 µm z-stack were acquired ( one section every 0 . 42 µm ) , and cells were scored with respect to their position in the primitive streak or in the epiblast after 3D reconstruction and optical transverse sections using Imaris software . Epiblast cells were counted as ‘non-ingressed’ and primitive streak/mesodermal cells as ‘ingressed’ . Whole mount RNA in situ hybridizations were carried out as described ( Henrique et al . , 1995 ) . Pictures of whole embryos were made using a macroscope ( Z16APOA , Leica ) with a 1× planapo objective ( Leica ) and a high resolution color camera ( DFC 420C , Leica ) . Chicken Pbx1 RNA probe is described in Coy and Borycki ( 2010 ) . The Fgf8 intronic probe is decribed in Dubrulle and Pourquié , ( 2004 ) . A 750-bp fragment of the coding sequence ( from nucleotide 301 to 1061 ) of chicken Fzd2 , the last 800 bp of the coding sequence of chicken Dact2 , intron 6 of chicken T , a 948-bp fragment of cSox2 coding sequence ( gift from B Pain ) were used as probes . Chicken Hox RNA probes were Hoxa2 ( Prince and Lumsden , 1994 ) , Hoxa3 , b3 , d4 ( gift from R Krumlauf ) , Hoxa10 , a11 , a13 , c4 , c5 , c6 , c8 , c9 , d8 , d9 , d10 , d11 , d12 , and d13 ( gift from C Tabin ) , Hoxb1 , b4 , b7 , b9 ( described in Iimura and Pourquié , 2006 ) , Hoxb8 ( gift from A Kuroiwa ) , Hoxb5 , c10 , c11 , c12 , c13 , and b13 ( cloned by PCR using ENSEMBL sequence informations ) , Hoxa4 ( chEST 427p4 , Geneservice ) , Hoxa5 ( chEST 382m24 , Geneservice ) , Hoxa6 ( chEST 338h20 , Geneservice ) , Hoxa7 ( chEST 259o11 , Geneservice ) , Hoxa9 ( chEST 333f16 , Geneservice ) , Hoxb2 ( chEST 194e4 , Geneservice ) , Hoxb6 ( ChEST147L22 , Geneservice ) , Hoxd3 ( chEST 195d1 , Geneservice ) . Full-length coding sequences for chicken Hox genes , Pbx1 , T , Wnt3a , Wnt5a , Dact2 , and mouse Fzd2 were PCR-amplified from chicken or mouse cDNAs using the proofreading Accuprime pfx DNA polymerase ( Invitrogen , Grand Island , NY ) . PCR fragments were then cloned in either , Grand Island , NY P221 ( Invitrogen ) or pENTR-D/TOPO ( Invitrogen ) to generate gateway ( Invitrogen ) entry clones . The constitutively active version of lef1 ( βcatLEF ) ( gift from R Grosschedl ) ( Galceran et al . , 2001 ) , a dominant activated form of Lrp6 ( Lrp6ΔN ) ( gift from S Aaronson ) ( Liu et al . , 2003 ) , a stabilized form of Ctbbn1 ( dBC ) ( Harada et al . , 1999 ) , and a dominant negative form or Rhoa ( DN-Rhoa ) ( Gift from P Kulesa ) was PCR-amplified and sub-cloned in pENTR-D/TOPO ( Invitrogen ) . Hoxa13 , Hoxc11 , and Hoxd10 mutated versions unable to bind DNA ( HoxmutH ) were generated by mutating amino acids 50 , 51 , and 53 of the homeodomain to alanine ( Gehring et al . , 1994 ) . When over-expressed in paraxial mesoderm precursors , these HoxmutH constructs show no effect on cell ingression , elongation velocity , and Wnt signaling ( data not shown ) . The dominant-negative forms of Hoxd10 , c11 , and a13 ( respectively Hoxd10dn , Hoxc11dn and Hoxa13dn ) were generated by inserting a stop codon instead of the amino acid 50 of the homeodomain . Chimeras for Hox genes were generated by fusion PCR . The homeodomain sequence of Hoxa13 was fused to the N-ter and C-ter of Hoxa5 to generate Hoxa5Ha13 . The homeodomain of Hoxa5 was fused to the N-ter and C-ter of Hoxa13 to generate Hoxa13Ha5 . The N-ter of Hoxa13 was fused to the homeodomain and C-ter of Hoxa5 to generate NHoxa13HCa5 . The Cter of Hoxa13 was fused to the Nter and homeodomain of Hoxa5 to generate Hoxa5Ca13 . The homeodomain of Hoxa5 was fused with either the Nter and Cter of Hoxc11 or the Nter and Cter of Hoxa13 to generate Hoxc11Ha5 and Hoxa13Ha5 , respectively . The chimeras were cloned in pENTR-D/TOPO to generate entry clones . Entry clones were then cloned in destination vectors ( depending on the experiments ) using Gateway technology ( Invitrogen ) . For consecutive electroporations and luciferase assays , a pCAGGS-IRES2-Venus-RFA destination vector was generated as follows: a yellow fluorescent protein ( YFP ) , Venus , with two sites of myristoylation that target the fluorescent protein to the membrane ( Venus , gift from K Hadjantonakis ) ( Rhee et al . , 2006 ) , was fused to an Internal Ribosomal Entry Site ( IRES2 ) ( Clontech ) by PCR . The primers used contained an EcoRI site in 5′ and a NotI site in 3′ . The EcoRI/IRES2-Venus/NotI fragment was then cloned into the EcoRI-NotI restriction sites of pCAGGS . A Gateway cassette ( RFA , Invitrogen ) was then inserted into the EcoRV site of the pCAGGS-I2-Venus , upstream of the IRES2 . For axis elongation measurements and cell tracking experiments , a pCI2HV-RFA destination vector was generated as follows: a YFP protein , Venus , was first fused to the full-length coding sequence of histone H2B to target the fluorescent protein to the nucleus ( H2B-Venus ) . The H2B-Venus PCR fragment was then fused by PCR to an IRES2 ( Clontech ) . The primers used contain an EcoRI site in 5′ and a NotI site in 3′ . The EcoRI/IRES2-H2B-Venus/NotI fragment was then cloned into the EcoRI-NotI restriction sites of pCAGGS . A Gateway cassette ( RFA , Invitrogen ) was then cloned in the EcoRV site of the pCI2HV , upstream of the IRES2 . For luciferase assay experiments , the chicken T promoter ( 1 kb upstream of the ATG ) was PCR-amplified and cloned upstream of the firefly luciferase in the pGL4 . 10 ( luc2 ) vector ( Promega ) to generate the cTprLuc reporter . Expression driven by this promoter fragment in chicken embryo recapitulates the PM expression of T ( not shown ) . The Wnt/βcatenin pathway activity reporter ( seven TCF/LEF binding sites + siamois minimal promoter ) was PCR amplified from the BAT-GAL plasmid ( Addgene plasmid 20889 ) ( Maretto et al . , 2003 ) and cloned upstream of a firefly luciferase in the pGL4 . 10 ( luc2 ) vector ( Promega ) to generate the BATLuc reporter . RNA interference experiments were performed using 21-nucleotide dsRNAs ( Dharmacon , Option A4 ) . To identify electroporated cells , siRNAs ( suspended in TE to a final concentration of 5 mg/ml ) were mixed with a pCAGGS-Venus or Cherry expression plasmid ( 1 . 0 mg/ml ) . The target sequence against chick Pbx1 was as follows: 5′- GTGTGAAATCAAAGAGAAA-3′ . As a control siRNA , we used a siRNA targeting chick Pbx1 containing two point mutations ( underlined in the sequence ) : 5′-ACACAAAGCTGAAGAAGTA-3′ that show no effect on Pbx1 expression . To monitor the Pbx1 siRNA efficiency , the anterior primitive streak of stage 4 HH embryos was electroporated with either control siRNA or Pbx1 siRNA mixed with a pCAGGS-Venus expression plasmid ( 1 . 0 mg/ml ) . Embryos were reincubated at 38°C until they reach stage 7 HH when they were harvested and processed for ISH for Pbx1 and immunofluorescence against GFP . A pBIC control vector ( derived from the pBI-tet [clontech] in which Cherry has been cloned ) ( gift from J Chal ) that allows simultaneous expression of two proteins at the same level once activated by doxycycline ( Tet-on , Clontech ) or the pBIC vector containing the full length Pbx1 along with a vector expressing the rtTA ( Clontech ) were electroporated in PSM progenitors at Stage 5 HH . Embryos were reincubated until they reach the 3-somite stage . They were then placed on imaging plates containing 0 . 5 μg/ml doxycycline for 1 hr at 38°C before starting acquisition . Axis elongation measurements were performed as described below between 5 and 9-somite stages . Electroporated embryos were cultured ventral side up on a microscope stage . We used a computer controlled , wide-field ( 10× objective ) epifluorescent microscope ( Leica DMR ) workstation , equipped with a motorized stage and cooled digital camera ( QImaging Retiga 1300i ) , to acquire 12-bit grayscale intensity images ( 492 × 652 pixels ) . For one embryo , several images at different focal planes and different fields were captured at a single time-point ( frame ) . The acquisition rate used was 10 frames per hour ( 6 min between frames ) . Image processing , including focal plane ‘collapsing’ field merging and registering , was performed to create high-resolution , 2D time-lapse sequences for cell tracking and axis elongation measures ( see Czirók et al . , 2002 , for details ) . To correct for the gradual drift of the embryo position or sudden changes due to repositioning of the microscope stage , images were registered to the embryo center . Variation of the distance between a formed somite and the node was used to determine the velocity of body axis elongation . The coordinates of the different points were determined on bright-field images of the time-lapse experiments using the cellular tracking option of ImageJ . ImageJ is a public domain , Java-based image processing program developed at the National Institutes of Health . For wild-type embryo measurements , axis elongation velocity was measured between 1 and 3 somites ( n = 8 ) , between 5 and 7 somites ( n = 8 ) , between 9 and 11 somites ( n = 6 ) , between 15 and 17 somites ( n = 5 ) , between 20 and 22 somites ( n = 6 ) , and between 25 and 27 somites ( n = 8 ) . For 15–17 somites measurements , embryos were cultured starting at 13 somites and imaged until 18 somites . For 20–22 somite measurements , embryos were cultured starting at 18 somites and imaged until 23 somites . For 25–27 somites measurements , embryos were cultured starting at 23 somites and imaged until 28 somites . For measurements of axis elongation velocity after Hox or T over-expression , electroporated embryos at stage 5 HH were cultured in a humidified incubator at 38°C for 3 hr and then placed on the microscope stage , as described above , for 18 hr . Axis elongation velocity was measured for 10 hr , starting from the 5-somite stage . Student's t-tests were applied to evaluate the differences between conditions . Cells electroporated with either a control or a Hox gene and a nuclear fluorescent protein ( H2B-Venus or H2B-GFP ) were automatically tracked using the Imaris software's cell tracking module ( version 7 . 3 . 1 ) . Cells were segmented based on nucleus size ( set at 5 μm ) and fluorescence intensity . The tracking algorithm was based on Brownian motion . Only cells in the posterior PSM were tracked for 10 hr . To substract the tissue motion to the single cell motion , the average speed of all tracked cells , that represent the tissue motion , has been substracted from the average speed of each individual cell ( as described in Bénazéraf et al . , 2010 ) . Student t-tests were applied to evaluate the differences recorded between the different conditions . Embryos were harvested at stage 5 HH and electroporated with a DNA mix containing either cTprLuc or BATLuc ( 1 μg/μl final ) , CMV-Renilla ( Promega , Madison , WI ) ( used as a control to normalize the differences of electroporation intensity between embryos [0 . 2 μg/μl final] ) , a control pCAGGS-Venus vector ( gift from K Hadjantonakis ) or a gene of interest cloned in pCAGGS-IRES2-Venus ( 5 μg/μ; final ) . Electroporated embryos were cultured in a humidified incubator at 38°C for 20 hr . Embryos were analyzed using a fluorescent microscope and only embryos showing restricted expression of Venus in the paraxial mesoderm were selected ( 90–100% of the electroporated embryos ) for luciferase assay ( between 3 and 5 embryos for each condition ) . The posterior region ( from somite 1 to tail-bud ) of the selected embryos was dissected and lysed in passive lysis buffer ( Promega ) for 15 min at room temperature . Lysates were then distributed in a 96-well plate and luciferase assays were performed using a Centro LB 960 luminometer ( Berthold Technology , France ) and the dual luciferase kit ( Promega ) following manufacturer's instructions . Raw intensity values for Firefly luciferase signal were normalized with corresponding Renilla luciferase values ( RLU ) and the control experiment was set to 1 . Student t-tests were applied to evaluate the differences between conditions . For Hox dominant-negative experiment , embryos were electroporated at st8 HH with a mix containing BATLuc , CMV-Renilla and either a Hoxa13mutH or a mix of Hoxc11mutH and Hoxa13mutH or a mix Hoxd10mutH , Hoxc11mutH and Hoxa13mutH ( in pCAGGS-I2-Venus [control condition] ) or Hoxa13dn , or a mix of Hoxc11dn and Hoxa13dn or a mix of Hoxd10dn , Hoxc11dn and Hoxa13dn ( in pCAGGS-I2-Venus [mutant condition] ) . Embryos were reincubated until they reach the 28-somite stage . The tail-bud of each embryo was dissected and used for the luciferase assay as described above . Stage 5 HH embryos electroporated with either a control pCI2HV or a pCI2HVHoxa13 vector were cultured in a humidified incubator at 38°C for 6 hr . Embryos were then selected using a fluorescence stereomicroscope based on electroporation efficiency . Selected embryos were fixed for 30 min at room temperature and then cryo-preserved in 30% sucrose in PBS at 4°C . Embryos were then transferred in a solution containing 7 . 5% gelatin and 15% sucrose in PBS and placed at 42°C . Embryos were then included in a cryosection mold and flash frozen in a dry ice-ethanol bath . 12-μm transverse cryosections of the electroporated region were prepared using a Leica CM3050 S cryostat . Sections were collected on superfrost slides and stored at −20°C . For immunocytochemistry , sections were placed in warm PBS ( 42°C ) for 5 min to remove gelatin . Sections were incubated with the primary antibody in PBS/BSA ( 2% ) /Triton ( 0 . 1% ) for 2 hr in a humidified chamber at room temperature . Slides were then washed four times for 15 min in PBS and incubated with the secondary antibody in PBS/BSA ( 2% ) /Triton ( 0 . 1% ) for 45 min in a humidified chamber . For the cell ingression assay and the labeling of the extracellular matrix ( ECM ) , we used , respectively , a rabbit anti-GFP ( abcam , #ab290 , UK ) at 1/2000 and the mouse anti-laminin ( DSHB , #3H11 , Iowa City , IA ) at 1/200 . The secondary antibodies were anti-rabbit Alexafluor488 ( Invitrogen ) and anti-mouse IgG1Alexafluor555 ( Invitrogen ) , respectively , used at 1/1000 . DAPI ( Invitrogen , 1/1000 dilution ) and an Alexafluor 633 phalloidin ( Invitrogen ) were applied at the same time as the secondary antibodies to label the nuclei and the F-actin , respectively . For tubulin labeling , we used the mouse anti-acetylated alpha-tubulin ( sigma T6793 ) at 1/1000 . The secondary antibody was an anti-mouse IgG2b Alexafluor546 ( Invitrogen ) , used at 1/1000 . Slides were mounted in Fluoromount-G ( SouthernBiotech ) and analyzed with a LSM 510 NLO inverted confocal microscope ( Carl ZEISS , Germany ) using a plan apochromat 63× ( NA 1 . 4 ) immersion ( oil ) objective ( Carl Zeiss ) . A pBIC control vector ( described above ) or the pBIC vector containing the full-length Hoxa13 with a C-terminal HA tag along with a vector containing EGFP under the CAGGS promoter and a vector expressing the rtTA ( Clontech , France ) were electroporated in the PM progenitors at stage 5 HH . A drop of 50 μl of different doses of doxycyclin ( from 50 μg/ml to 0 . 5 μg/ml ) was applied on top of the embryos immediately after electroporation and the embryos were reincubated for 20 hr . Three embryos for each condition were individually lysed following standard procedure and each lysate was loaded on a different well of an SDS-page gel . Western blot analysis was done following standard procedure . An anti-HA-HRP antibody was used to detect Hoxa13 ( Roche #12013819001 , dilution 1/1000 , Germany ) . An anti-GFP antibody ( abcam ab6556 , dilution 1/2000 ) was used to detect GFP from the pCAGGS-EGFP used as an electroporation control . An anti-β actin antibody ( Sigma A5441 , dilution 1/5000 , Germany ) was used to verify that the same amount of tissue was loaded in each well . This experiment has been repeated twice independently . A 20 μl drop of 100 µM EdU ( Click-iT EdU kit , Cat . #C10083 Invitrogen ) was applied on the posterior region of 20–22- and 25–27-somite stage embryos cultured in vitro for 45 min . Embryos were then immediately fixed in 4% paraformaldehyde ( PFA ) for 45 min at room temperature ( RT ) and were then processed as described in Warren et al . ( 2009 ) . Phospho-histone H3 ( pH3 ) ( Millipore , #06-570 , 1/1000 dilution , France ) immunolabelling was performed after the EdU reaction . Single plane sections were generated , and the PSM region was manually segmented . For the tail-bud proliferation assay , parasagittal cryosections ( 20 µm ) were made . Nuclei labeled with DAPI and EdU and/or pH3 were manually counted . Sections were imaged using a Zeiss 510 NLO and a 20× dry NA0 . 8 objective . Embryos were harvested at 20–22- and 25–27-somite stage and processed as described ( Smith and Cartwright , 1997 ) using the ApopTag Red In Situ kit ( #S7165; Millipore ) . Single plane sections were generated and the PSM region was manually segmented . For tail-bud apoptosis assay , parasagittal cryosections ( 20 µm ) were made . Nuclei labeled with DAPI and/or apoptotic labeling were manually counted . Labeled embryos were imaged using a Zeiss 510 NLO and a 20× dry NA0 . 8 objective . PM precursors of the anterior primitive streak of Stage 5 HH embryos were electroporated as previously described either with a control vector coding for a H2B-venus fusion ( pCI2HV ) or a vector coding for Hoxa13 and a H2B-venus fusion ( Hoxa13pCI2HV ) . Embryos were reincubated for 14 hr in a humidified incubator at 38°C until they reach the 9-somite stage . The region containing the PM progenitors was dissected from several embryos and pooled in a drop of PBS/FCS1% ( seven embryos per condition ) on ice . Dissected tissues were then transferred in a drop of diluted trypsin and incubated at 38°C for 10 min to allow efficient enzymatic dissociation of cells . Cell dissociation was completed mechanically by pipetting up and down . Cells were then transferred into 500 μl of PBS/FCS 1% on ice and sorted based on YFP fluorescence using a FACS DIVA ( BD technologies , France ) . For each condition , one thousand YFP+ cells were collected directly in Trizol ( Invitrogen ) and immediately frozen at −80°C . This experiment was repeated twice independently . Extraction of total RNA was performed according to manufacturer's instructions ( Trizol , Invitrogen ) . Biotinylated cRNA targets were prepared from total RNA using a double amplification protocol according to the GeneChip Expression Analysis Technical Manual: two-Cycle Target Labeling Assay ( P/N 701021 Rev . 5 , Affymetrix , Santa Clara , USA ) . Following fragmentation , cRNAs were hybridized for 16 hr at 45°C on GeneChip Chicken Genome arrays . Each microarray ( one microarray per condition = two control microarrays and 2 Hoxa13 microarrays ) was then washed and stained on a GeneChip fluidics station 450 and scanned with a GeneChip Scanner 3000 7G . Finally , raw data ( . CEL Intensity files ) were extracted from the scanned images using the Affymetrix GeneChip Command Console ( AGCC ) version 3 . 1 . CEL files were further processed with MAS5 and RMA algorithms using the Bioconductor package ( version 2 . 8 ) available through R ( version 2 . 12 . 1 ) . Probe sets were filtered based on their expression intensity value ( MAS5 value ) . Probe sets with an intensity value under 100 were discarded . Probe sets were ranked based on fold change between the intensity value of the control condition and the Hoxa13 over-expression condition . The microarrays raw data are available on the GEO website ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE38107 ) . RNAs from the microarray experiments were used as templates for cDNA synthesis using the Qantitect kit ( Qiagen ) . 3 μl of cDNA was mixed with 6 μl of 2× Lightcycler 480 SYBR green I master ( Roche ) and 1 μM of primers ( listed in Table 2 ) in a total volume of 12 μl . The Q-PCR reactions were run on a Lightcycler 480 ( Roche ) with the Lightcycler 480 . Each sample was run in duplicate and gapdh was used as a control gene . The CT values obtained for each gene were normalized against the CT value obtained for gapdh . 10 . 7554/eLife . 04379 . 028Table 2 . List of primers used for Q-RT PCRDOI: http://dx . doi . org/10 . 7554/eLife . 04379 . 028Gene nameGene referencePrimers sequence 5′→3′Size of the ampliconGapdhNM_204305 . 1F: GCTGAGAACGGGAAACTTGTG62 bpR: GGGTCACGCTCCTGGAAGATNM_204940 . 1F: CGAGGAGATCACAGCTTTAAAAATT75 bpR: TCATTTCTTTCCTTTGCGTCAAAxin2NM_204491 . 1F: GCGCAAACGATAGTGAGATATCC76 bpR: CCATCTACACTGCTGTCTGTCATTGSp8NM_001198666 . 1F: CATGGCGCACCCCTACGAGTC131 bpR: CGTTGGGGGCACGTCGATCCAFzd2NM_204222 . 1F: CCCTGCCCGCTGCACTTCAC190 bpR: CCGCTCACACCGTGGTCTCGCyp26a1NM_001001129 . 1ikF: AGGAGCCCGAGGGTGGCTACA138 bpR: TGGCAGTGGTTTCATGACCTCCAAFgf8NM_001012767 . 1F: CGCTCTTCAGCTACGTGTTCATGC108 bpR: TGGTAGGTGCGCACGAGCCEtv1NM_204917 . 1F: ATGGACCACAGATTTCGCCGCC145 bpR: TTGGACGTCCTTCCCTCGGCAFgfr1NM_205510 . 1F: CACGCTGCCCGACCAAGCTC168 bpR: GTGATGCGCGTGCGGTTGTTRasgrp3NM_001006401 . 1F: AACGGCATCTCCAAGTGGGTCCA111 bpR: GAGATGAAGGAGCTTCTGTGCAACA PM precursors of the anterior primitive streak of Stage 8 HH embryos were electroporated as previously described either with a mix of control vectors coding for Hoxd10mutH , Hoxc11mutH and Hoxa13mutH in pCAGGS-IRES2-Venus or a mix of vector coding for Hoxd10dn , Hoxc11dn and Hoxa13dn in pCAGGS-IRES2-Venus . Embryos were reincubated in a humidified incubator at 38°C until they reached the 28-somite stage . Tail-bud regions containing the PM progenitors were then dissected and pooled in a drop of PBS/FCS1% ( three embryos per condition ) on ice . Dissected tissues were then transferred into a drop of diluted trypsin and incubated at 38°C for 10 min to allow efficient dissociation of the cells . The dissociation of cells was completed mechanically using a glass micropipette by pipetting up and down . Cells were then transferred into 500 μl of PBS/FCS1% on ice and sorted based on YFP fluorescence using a FACS DIVA ( BD technologies ) . For each condition , one thousand YFP+ cells were collected directly in Trizol ( Invitrogen ) and immediately frozen at −80°C . This experiment was repeated four times independently . Extraction of total RNA was performed according to manufacturer's instructions ( Trizol , Invitrogen ) . RNAs were used as templates for cDNA synthesis using the QuantiTect kit ( Qiagen ) . 3 μl of cDNA was mixed with 6 μl of 2× Lightcycler 480 SYBR green I master ( Roche ) and 1 μM of primers ( listed in Table 2 ) in a final volume of 12 μl . The Q-RT PCR reactions were run on a Lightcycler 480 ( Roche ) . Each sample was run in duplicates , and gapdh was used as a control . The CT values obtained for each gene were normalized against the CT value obtained for gapdh . Embryos were harvested in PBS at different stages ( 10 , 15 , 20 , or 25-somite stages ) and pined using 0 . 10 mm minutiens on a silicon-coated petri dish . The tailbud region was then microdissected using a sharpened tungsten needle and care was taken to remove the endoderm and the ectoderm . Each individual tailbud was immediately transferred in 500 μl of Trizol ( Invitrogen ) in a 1 . 5 ml RNAse free tube ( Ambion ) on ice until five individual tailbuds were collected per stage ( resulting in five tubes per stage ) . Then the tubes were immediately frozen at −80°C . Extraction of total RNA was performed according to manufacturer's instructions ( Trizol , Invitrogen ) . RNAs were used as templates for cDNA synthesis using the iScript reverse transcriptase Supermix ( Biorad ) . 3 μl of cDNA was mixed with 5 μl of 2× SSoAdvanced universal SYBR green supermix ( Biorad ) and 1 μM of primers ( listed in Table 2 ) in a final volume of 10 μl . The Q-RT PCR reactions were run on a CFX384 ( Biorad ) . Each sample was run in triplicates , and gapdh was used as a control . The CT values obtained for each gene were normalized against the CT value obtained for gapdh . | In humans and other vertebrates , the number of bones ( vertebrae ) in the spine is determined early in development . The vertebrae form from blocks of tissue called somites that make segments along the body axis—a virtual line running from the head to the tail-end—of the embryo . The somites form as the embryo increases in length , with new somites forming periodically at the back near the embryo's tail-end . A family of genes called the Hox genes are involved in controlling the formation of the somites . However , it is not known whether they directly control the number of somites that form , or whether they control the length of the body of the embryo . Denans et al . studied the Hox genes in chicken embryos . The experiments suggest that the activation of some of the Hox genes in a structure called the tail-bud , which is found at the tail-end of the embryo , slow down the elongation of the body . The Hox genes achieve this by repressing the activity of a signaling pathway called Wnt so that Wnt activity in the tail-bud progressively decreases as the embryo develops . The elongation of the body stops when the levels of a molecule called retinoic acid increase in the tail-bud , which causes the loss of the stem cells that are needed to make the somites . Denans et al . 's findings suggest that Hox genes influence the timing of the halt in elongation , which in turn is important for determining the total number of somites that form . Understanding how Hox genes control the formation of the cells that will make up the somites and influence Wnt signaling is a major challenge for the future . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2015 | Hox genes control vertebrate body elongation by collinear Wnt repression |
How movements are selected is a fundamental question in systems neuroscience . While many studies have elucidated the sensorimotor transformations underlying stimulus-guided movements , less is known about how internal goals – critical drivers of goal-directed behavior – guide movements . The basal ganglia are known to bias movement selection according to value , one form of internal goal . Here , we examine whether other internal goals , in addition to value , also influence movements via the basal ganglia . We designed a novel task for mice that dissociated equally rewarded internally-specified and stimulus-guided movements , allowing us to test how each engaged the basal ganglia . We found that activity in the substantia nigra pars reticulata , a basal ganglia output , predictably differed preceding internally-specified and stimulus-guided movements . Incorporating these results into a simple model suggests that internally-specified movements may be facilitated relative to stimulus-guided movements by basal ganglia processing .
As we interact with the world , our movements are selected based on external sensory stimuli and internal variables representing action value , learned stimulus-response contingencies , and prior experiences ( Gold and Shadlen , 2007 ) . Selecting the movement associated with the most desirable outcome requires appropriately weighting each of these factors . While the neural substrates for movements based on external sensory stimuli have been the focus of much research ( Hall and Moschovakis , 2003 ) , where , how , and when internal goals influence movement selection is less well understood . The basal ganglia ( BG ) are known to be involved in motor control ( Hikosaka and Wurtz , 1989; Mink , 1996 ) , contributing to movement selection by modulating inhibition on competing downstream motor structures ( Basso and Wurtz , 2002; Di Chiara et al . , 1979; Hikosaka and Wurtz , 1985 ) . In particular , the BG have been thought to bias the selection of movements towards those associated with the highest value ( Hikosaka et al . , 2006 ) . This 'value-biasing' hypothesis is supported by much evidence showing that activity in several BG nuclei is modulated , prior to stimulus presentation , by reward expectation ( Bryden et al . , 2011; Handel and Glimcher , 2000; Hikosaka et al . , 2006; Kawagoe et al . , 1998; Sato and Hikosaka , 2002 ) such that movements toward high-value targets are disinhibited relative to movements toward low-value targets ( Hikosaka et al . , 2006; Lauwereyns et al . , 2002 ) . Anatomical evidence is consistent with a primary role for the BG in mediating the integration of value-based information into motor plans ( Bolam et al . , 2000; Gerfen and Surmeier , 2011 ) . However , movement selection may also be guided by other internal representations , such as recent movements and their outcomes ( Corrado et al . , 2005; Fecteau and Munoz , 2003; Lau and Glimcher , 2005 ) . We therefore asked whether BG activity mediates the influence of internal goals , in addition to value , on movement selection . We reasoned that , if this were the case , BG output would differ when selecting equally valuable stimulus-guided and internally-specified movements . Specifically , we would expect that internally-specified movements would be promoted relative to otherwise-identical stimulus-guided movements , just as more valuable movements have been shown to be promoted relative to otherwise-identical less valuable movements ( Hikosaka et al . , 2006; Sato and Hikosaka , 2002 ) . Notably , it is has been proposed that Parkinsonian patients exhibit more pronounced bradykinesia when initiating internally-specified than stimulus-guided movements because the latter engage pathways outside of the BG ( Glickstein and Stein , 1991 ) . However , whether the BG themselves are differentially engaged by these two types of movements has not been tested . We distinguished between these two possibilities by recording from neurons in the substantia nigra pars reticulata ( SNr ) , an output nucleus of the BG critical for orienting movements ( Basso and Sommer , 2011; Handel and Glimcher , 1999; Hikosaka and Wurtz , 1983a ) , in mice performing a behavioral task in which a sensory stimulus either was or was not informative of the rewarded direction of an orienting movement . Using a design akin to that of other recent studies ( Pastor-Bernier and Cisek , 2011; Seo et al . , 2012; Ito and Doya , 2015 ) , in alternating blocks of trials , the rewarded direction was either determined by a sensory cue or by internal representations informed by recent trial history . Critically , we designed the task such that correct movements were equally valuable in both conditions . We found that SNr activity predictably differed between these two conditions , supporting the idea that the BG mediate the influence on movement selection of internal goals . We interpret these results , in the context of a simple model of BG output ( Hikosaka et al . , 2006 ) , as suggesting that internally-specified movements may be promoted over stimulus-guided movements by BG activity .
We trained mice on a delayed-response spatial choice task comprised of interleaved blocks of 'stimulus-guided' ( SG ) trials , in which the direction of movement is selected based on a sensory stimulus ( Uchida and Mainen , 2003 ) , and 'internally-specified' ( IS ) trials , in which the direction of an otherwise-identical movement is selected based on internal representations informed by recent trial history ( see Materials and methods; Figure 1A , B ) . In each trial of the task , the mouse is presented with a binary odor mixture at a central port , waits for an auditory go cue , and moves to the left or right reward port for a water reward . In SG trials , the dominant component of the odor mixture – which varies trial by trial – determines the side at which reward will be delivered , while in IS trials , a balanced mixture of the two odors is always presented but reward is delivered at only one side throughout the block ( see Materials and methods; Figure 1B ) . Thus , while both trial types require the mouse to sample the stimulus , in SG trials the stimulus indicates that the rewarded side is determined by the odor mixture and in IS trials the stimulus indicates that the rewarded side is determined by the recent history of choices and outcomes . We found that mice were able to infer ( unsignaled ) transitions between the SG and IS blocks and switch their response mode accordingly: during SG blocks , mice were equally likely to choose the left and right port ( Figure 1C , gray boxes ) reflecting a dependence on the odor mixture ( Figure 1D ) , while during IS blocks , mice reliably returned to the same ( rewarded ) port in each trial ( Figure 1C , white boxes ) . We quantified performance in IS blocks by calculating , for each block , the percentage of correct trials and the number of error events , defined as a run of consecutive incorrect choices ( Figure 1E ) . Finally , we reasoned that if a mouse were to recognize that a given trial belonged to an IS block , it could prepare its movement in advance and would therefore be able to reach the reward port faster ( Niemi and Näätänen , 1981; Poulton , 1950; Seo et al . , 2012 ) . Indeed , across the population of sessions , we found that reaction time – defined as the time from the go cue to reward port entry – was shorter in IS trials than in the 'easy' SG trials in the same session [Figure 1F; we used only easy SG trials ( mixture ratios of 95/5 , 80/20 , 20/80 , and 5/95 ) to control for a potential dependence of reaction time on difficulty; population of sessions: p = 1 . 7 × 10−11 , paired t-test; individual sessions: IS shorter than SG in 46/108 , SG shorter than IS in 3/108 , p<0 . 05 , Wilcoxon rank-sum test; ipsiversive and contraversive trials compared separately] . Together , these data suggest that , as intended , the direction of movement in SG blocks is selected based on the stimulus while the direction of movements in IS blocks is selected based on recent trial history . We therefore utilized this behavioral assay to compare how stimulus-guided and internally-specified movements are mediated by the BG , as described below . 10 . 7554/eLife . 13833 . 003Figure 1 . Behavioral task and performance . ( A ) Timing of events in each trial . The mouse enters the odor port , receives an odor mixture , waits for the go signal , exits the odor port , moves to one of the reward ports , and receives a water reward for a correct choice . Gray box , delay epoch . ( B ) Organization of SG ( gray ) and IS ( white ) blocks within a session . All sessions start with an SG block and alternate between SG and IS blocks . In SG blocks , reward side corresponds to the dominant odor in the mixture [ ( - ) -carvone , left; ( + ) -carvone , right]; when the odors are balanced ( [ ( - ) -carvone] = [ ( + ) -carvone] ) , the probability of reward at both reward ports is 0 . 5 . In IS blocks , odors are balanced in every trial and reward is available at the same side in each trial . Thickness of horizontal lines corresponds to probability of reward . SG , stimulus guided; IS , internally specified; L , left; R , right . ( C ) Fraction of left choices across block types throughout the session . Dashed line shows an example session ( boxcar smoothed over 7 trials ) , solid black line shows mean over all sessions ( 54 , from 4 mice ) , horizontal black lines show block means , horizontal gray lines show ideal block means ( if all choices were correct ) . To account for different numbers of trials per block across sessions , trials that occur in < 60% of sessions are excluded . In SG blocks only difficult trials [ ( + ) -carvone/ ( - ) -carvone = 60/40 , 50/50 , or 40/60] are shown . ( D ) Mean performance in SG blocks over all sessions , separated by rewarded side of first IS block in the session . Lines show best fit to p=11+e−a−bx , where x is the proportion of the left odor [ ( - ) -carvone ) ] in the mixture , p is the fraction of right choices , and a and b are free parameters . While choices were slightly biased by the rewarded direction in the first IS block ( center panels ) , they were much more strongly influenced by the stimulus . ( E ) Performance in IS blocks . Histograms of percent correct choices ( top ) and number of error events ( run of consecutive incorrect choices , bottom ) across blocks over all sessions . ( F ) Mean reaction time in easy SG trials plotted against mean reaction time in IS trials in the corresponding session , separately for each direction of movement . DOI: http://dx . doi . org/10 . 7554/eLife . 13833 . 003 If the BG integrates not only value but also other internal representations , then stimulus-guided and internally-specified movements may differentially engage the BG despite being equally valuable . In this case , we would predict that BG output would depend on whether the movement was internally specified or stimulus guided , and specifically , in our task , on the degree to which recent trial history is informative of rewarded direction . To test this prediction we examined activity in the SNr , a BG output known to be involved in orienting movements ( Basso and Sommer , 2011; Handel and Glimcher , 2000 , 1999; Hikosaka and Wurtz , 1983a , 1983b ) . We recorded from 296 well-isolated left SNr neurons ( see Materials and methods; Figure 2 ) in four mice performing the task ( see Supplementary file 1 ) . Data from one example neuron are shown in Figure 3A , B , segregated by reward port selected ( ipsilateral vs . contralateral to the recording side ) and by trial type ( SG vs . IS ) . The activity of this neuron clearly depends on both movement direction and trial type . To examine these dependencies across the population of neurons we first examined firing rate during the delay epoch , defined as the time from odor valve open to the time of odor port exit ( Figure 1A , gray box ) , which most directly captures , across trial types , activity underlying selection of the direction of movement ( but since activity in IS trials may , by design , reflect direction selection even before stimulus delivery , we subsequently examine activity in other epochs ) . Based on the firing rate during this epoch in SG and IS trials , we then calculated direction preference ( see Materials and methods ) . This value ranges from -1 ( strongly 'prefers' ipsiversive ) to 1 ( strongly prefers contraversive ) , where 0 represents no preference . We found that 216/296 neurons displayed a significant direction preference ( p<0 . 05 ) during the delay epoch , with about as many preferring ipsiversive ( 94/216 ) as contraversive ( 122/216 ) movements ( Figure 3C ) . Since SNr activity has been shown to exhibit both movement-related increases and decreases ( Bryden et al . , 2011; Gulley et al . , 2002 , 1999; Handel and Glimcher , 1999; Sato and Hikosaka , 2002 ) , we next asked whether a relationship existed between direction preference and the sign of activity change during the delay epoch , relative to baseline ( see Materials and methods ) . We found that 188/296 neurons exhibited an increase in activity during this epoch , 91/296 exhibited a decrease , and 17/296 exhibited no change ( Table 1 ) , consistent with previous studies ( Bryden et al . , 2011; Gulley et al . , 2002 , 1999; Handel and Glimcher , 2000 , 1999; Sato and Hikosaka , 2002 ) . Within these groups , neurons exhibited a preference for ipsiversive , contraversive , or neither direction in roughly equal numbers ( Table 1 ) . 10 . 7554/eLife . 13833 . 004Figure 2 . Confirmation of recording sites and spike clustering . ( A ) Schematic ( left ) shows targeted recording extent ( bar ) within SNr; coronal section ( right , 3 . 3 mm caudal from bregma ) shows representative tetrode track ( arrow ) in SNr . ( B ) Left , peaks of waveforms from lead 1 plotted against peaks of waveforms from lead 3 of one tetrode for a representative recording session . Note that clustering was performed using additional features to those shown here . Red and green points show waveform peaks recorded from neurons considered to be distinct . Right , waveforms ( mean ± SD ) corresponding to red and green points . DOI: http://dx . doi . org/10 . 7554/eLife . 13833 . 00410 . 7554/eLife . 13833 . 005Figure 3 . SNr activity during the delay epoch depends on movement direction and trial type . ( A ) Rasters for an example neuron grouped by movement direction ( rows ) and trial type ( columns ) . For each raster , each row shows spikes ( black ticks ) in one trial , aligned to time of odor valve open ( red line ) and sorted by duration of delay epoch . Green ticks , times of go signal; blue ticks , times of odor port exit . Fifty pseudo-randomly selected trials are shown per group . ( B ) Peri-event histograms showing average activity , separately by direction , in stimulus-guided ( left ) and internally-specified ( right ) trials . Shading , ± SEM . Histograms are smoothed with a Gaussian filter ( σ = 15 ms ) . Ipsi . , ipsiversive; Contra . , contraversive . ( C ) Histogram of direction preferences during delay epoch across population of neurons . Arrowhead corresponds to example neuron in A . ( D ) Difference in delay-epoch firing rate between ipsiversive and contraversive trials in SG vs . IS trials in the same session , separately for ipsiversive-preferring neurons ( left subpanel , corresponding to left black bars in C ) and contraversive-preferring neurons ( right subpanel , corresponding to right black bars in C ) . Only correct trials are included; all choices on 50/50 SG trials were considered correct regardless of whether they were rewarded . Dashed lines show x = 0 , y = 0 , and x = y . Red marker corresponds to example neuron from A and B . FR , firing rate . DOI: http://dx . doi . org/10 . 7554/eLife . 13833 . 00510 . 7554/eLife . 13833 . 006Table 1 . Direction preference and activity change during delay epoch . Neurons are grouped by direction preference and whether activity in the preferred direction increased or decreased relative to baseline ( see Materials and methods ) , during the delay epoch . Numbers and percentages of grand total ( 279 ) are shown; note that 17 neurons exhibited no change in activity and are not included here . DOI: http://dx . doi . org/10 . 7554/eLife . 13833 . 006PreferenceIncreaseDecreaseTotalContraversive8832%2710%11541%Ipsiversive5219%3814%9032%Nonselective4817%269%7427%Total18867%9133%279100% We next examined whether activity during the delay epoch of direction-selective neurons ( Figure 3C , black bars ) differed between SG and IS trials , in two complementary ways . First , we examined whether the difference in activity between ipsiversive and contraversive trials depended on whether the movement was stimulus guided or internally specified . Across our population , neurons tended to show a larger difference in firing rate preceding ipsiversive and contraversive movements in IS than in SG trials [Figure 3D; ipsiversive-preferring neurons: p = 2 . 2 × 10−10 , paired t-test; contraversive-preferring neurons: p = 4 . 0 × 10−4 , paired t-test] . Second , we determined whether neurons were trial-type-dependent by comparing firing rates between SG and IS trials in which the selected movement was correct and in the preferred direction of the neuron; we then repeated this comparison for the antipreferred direction . For trials in the preferred direction , we found that the activity of approximately half of the direction-selective neurons was modulated by trial type ( 101/216; p<0 . 05 , unpaired t-test ) , with more neurons exhibiting higher activity in IS trials than SG trials ( 84/101 vs . 17/101; p = 2 . 6 × 10−11 , X2 test ) . Conversely , for trials in the antipreferred direction , we again found that the activity of approximately half of the direction-selective neurons was modulated by trial type ( 109/216; p<0 . 05 , unpaired t-test; 158/216 direction-selective neurons were modulated by trial type in at least one direction ) , but that more neurons exhibited higher activity in SG trials than IS trials ( 76/109 vs . 33/109; p = 3 . 8 × 10−5 , X2 test ) . Therefore , while we found that neurons were about equally likely to prefer upcoming ipsiversive and contraversive movements ( Figure 3C ) , their activity depended , in a predictable manner , on trial type ( Figure 3D ) . While these findings suggest that the BG differentially mediate internally-specified and stimulus-guided movements , as we had predicted , a few differences between SG and IS trials may have contributed to this observation . We therefore sought to identify these differences and determine their influence , in several ways . First , we reasoned that , if neural activity indeed reflects trial type , firing rate would systematically change during the IS block as the mouse increasingly based its movement choice on internal representations instead of the stimulus ( recall that the transitions from SG to IS blocks were unsignaled ) . To test this idea , we calculated the correlation between the trial-by-trial firing rate during the delay epoch and the extent to which the mouse had learned that its movement choice should be internally specified , estimated with a reinforcement learning algorithm ( see Materials and methods ) . We performed this analysis on the 158 neurons with firing rates that depended on both direction and trial type , separately for choices in the preferred and antipreferred direction . Figure 4A shows data from an example neuron displaying a significant correlation for trials in the preferred direction of the neuron ( r = 0 . 65 , p = 7 . 0 × 10−9 ) , and no correlation for trials in the antipreferred direction ( r = 0 . 066 , p = 0 . 60 ) . Overall , 77/158 neurons exhibited a significant correlation ( p<0 . 05 ) between firing rate and the number of consecutive correct trials for either direction [Figure 4B , C; 35/77 for trials in the preferred direction ( red circles ) , 29/77 for trials in the antipreferred direction ( blue circles ) , and 13/77 for trials in both directions ( purple circles ) ] , with more positive correlations for trials in the preferred direction ( p = 2 . 4 × 10−6 , X2 test ) and negative correlations for trials in the antipreferred direction ( p = 5 . 9 × 10−6 , X2 test ) , as we would expect given the pattern of results shown in Figure 3D . These results support the idea that SNr activity reflects the degree to which movements are selected based on internal representations . 10 . 7554/eLife . 13833 . 007Figure 4 . Activity depends on the extent to which movements are internally specified . ( A ) Firing rate during delay epoch plotted as a function of the value of the rewarded side , estimated via reinforcement learning ( Vdir , t ) , for both IS blocks in a session , for one example neuron . Each circle corresponds to one trial . ( B ) Correlations ( as in panel A ) for ipsiversive movement plotted against contraversive movement , for the population of ipsiversive-preferring neurons ( left black bars in Figure 3C ) with activity that depended on trial type ( SG vs . IS ) . Each circle corresponds to one neuron . ( C ) Same as B , for contraversive-preferring neurons ( right black bars in Figure 3C ) . Black box corresponds to example neuron from A . DOI: http://dx . doi . org/10 . 7554/eLife . 13833 . 007 We next examined the potential influence of other factors on the observed difference in neural activity during SG and IS trials . One difference between these trial types , by design , is that in IS trials the decision ( to move left or right ) is relatively easy , while in some SG trials this decision is more difficult ( Figure 1D ) . The difficulty of this decision – or an associated variable , such as uncertainty , or the estimated value of each movement direction – could , in principle , affect SNr activity . Were this the case , we would expect to observe a difference in activity between those SG trials requiring an easy discrimination ( mixture ratios of 95/5 , 80/20 , 20/80 , and 5/95 ) and those SG trials requiring a 'difficult' discrimination ( mixture ratios of 60/40 , 50/50 , and 40/60 ) , since easy trials resulted in a larger fraction of correct choices ( p = 3 . 4 × 10–27 , paired t-test; see Figure 1D ) , corresponding to a higher likelihood of reward . We therefore compared firing rate during the delay epoch between easy and difficult SG trials , separately for trials in the ipsiversive ( Figure 5A ) and contraversive ( Figure 5B ) direction , for the 216 direction-selective neurons ( Figure 3C , black bars ) . We found that the activity of some individual neurons depended on difficulty ( or an associated variable ) ( ipsiversive direction: 39/216 neurons; contraversive direction: 31/216 neurons , p<0 . 05 , 1-way ANOVA across mixture ratios , Figure 5A , B ) , as would be predicted by the value-biasing view of BG function . However , there was little overlap ( purple circles ) between this small population of difficulty-dependent neurons ( blue circles ) and those neurons that we classified as trial-type-dependent [ipsiversive direction: 21/109 trial-type-dependent , and 18/107 non-trial-type-dependent , neurons exhibited difficulty dependence ( these ratios did not differ: p = 0 . 32 , X2 test ) ; contraversive direction: 16/101 trial-type-dependent , and 15/115 non-trial-type-dependent , neurons exhibited difficulty dependence ( these ratios did not differ: p = 0 . 56 , X2 test ) ; p<0 . 05 , 1-way ANOVA across mixture ratios] . These results suggest that differences in decision difficulty , uncertainty , and the value associated with the direction of movement do not account for trial-type dependence or the differences in activity between SG and IS trials shown in Figure 3D . 10 . 7554/eLife . 13833 . 008Figure 5 . Dependence of firing rate on trial type cannot be explained by discrimination difficulty or an associated variable . ( A ) Mean normalized change from baseline ( Fc , see Materials and methods ) during delay epoch of easy vs . difficult ipsiversive SG trials of direction-selective neurons ( black bars in Figure 3C ) . Each circle corresponds to one neuron . Red circles indicate that activity differs between SG and IS trials , and does not depends on mixture ratio ( or an associated variable such as discrimination difficulty ) . ( B ) Same as A , for contraversive trials . DOI: http://dx . doi . org/10 . 7554/eLife . 13833 . 008 We then examined whether reaction time ( which differs between trial types; Figure 1F ) and the choice on the previous trial ( which , by design , is more likely to correlate with the current choice in IS blocks than SG blocks; Figure 1C ) could explain the difference in activity between SG and IS trials . Preliminary analyses of each of these factors in isolation indicated that , as opposed to discrimination difficulty or an associated variable such as value ( Figure 5 ) , they often correlated with firing rate during the delay epoch . In order to determine the relative influence of these factors , as well as other factors that correlate with firing rate – current choice ( Figure 3C ) and trial type ( Figure 3D ) – on neural activity during the delay epoch , we performed a linear regression analysis with previous choice , current choice , trial type and reaction time as predictor variables ( see Materials and methods ) . By considering all of these factors simultaneously , this analysis provides an unbiased method for determining their influence on neural activity . Across our population of neurons , the vast majority were influenced by at least one of these factors ( 281/296 , p<0 . 05 ) , and we found neurons with firing rates influenced by all possible combinations of factors ( Figure 6A ) . Consistent with the results shown in Figure 3C and D , respectively , this analysis confirms that , as the mouse is selecting its direction of movement , the activity of many SNr neurons was modulated by current choice ( 167/296 ) and trial type ( 142/296 ) . We also found that the activity of many neurons depended on reaction time ( 111/296 ) . Surprisingly , the largest fraction of neurons exhibited activity modulated by previous choice ( 188/296 ) . This is particularly interesting because this variable is critical for determining , in an IS block , which direction is associated with reward . 10 . 7554/eLife . 13833 . 009Figure 6 . SNr activity is influenced by several task-relevant factors throughout the trial . ( A ) Venn diagram showing the number of neurons whose firing rate during the delay epoch was significantly influenced ( p < 0 . 05 ) by previous choice , current choice , trial type , reaction time , and all combinations of these factors , or by no factor . ( B ) β coefficients estimated based on firing rate in 100 ms bins aligned to three different trial events for one example neuron ( reaction time coefficient not shown , for clarity ) . Shading , ± 95% confidence interval . ( C ) Fraction of neurons with a significant β coefficient corresponding to each predictor variable in each 100 ms bin , aligned as in panel B . All 296 neurons were included in this analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 13833 . 009 Given that movements can initially be selected earlier in IS than SG trials ( Figure 3A , B ) , we wondered how firing rate at other times during the trial depended on previous choice , current choice , trial type , and reaction time . We therefore extended our regression analysis to examine how firing rate is modulated by these factors during overlapping 100 ms windows throughout the trial ( see Materials and methods ) . In the example shown in Figure 6B , the activity of the neuron is modulated by previous choice ( cyan line ) – i . e . , the confidence interval ( shading ) for this coefficient does not include 0 – even before the odor is delivered ( odor valve open ) , and this influence persists until the movement is initiated ( odor port exit ) . The current choice ( black line ) does not influence neural activity until ramping up just prior to movement initiation , but then continues to exert an influence for the remainder of the trial . The trial type ( magenta line ) , meanwhile , exerts a moderate influence on the firing rate – specifically , activity is higher for IS trials – until just before movement initiation , after which this influence is diminished . Reaction time was a relatively poor predictor of firing rate ( not shown , for clarity ) . To examine the dynamics of the weights of these factors across our population of neurons , we calculated the fraction of neurons with significant weights in each time window ( Figure 6C ) . The pattern of results was similar to that shown in the example neuron ( Figure 6B ) . Before the odor is delivered , the firing rates of about half of the neurons are influenced by the previous choice . However , as the odor is sampled , the influence of the previous choice decreases and the influence of the current choice increases , with about two thirds of neurons exhibiting a significant weight for the current choice by the time the movement is initiated . Interestingly , trial type modulates the activity of about one third of neurons throughout the trial . The influence of reaction time is strongest during movement but has relatively little influence on the population ( not shown ) . These results indicate that SNr activity dynamically reflects trial type and other task-relevant variables throughout the trial , as would be expected if the BG are differentially involved in mediating stimulus-guided and internally-specified movements .
We have shown that SNr activity preceding orienting movements depends on whether the direction of movement was indicated by a stimulus or was specified by internal variables ( Figures 3 , 4 ) . While we designed the task such that correct movements were equally valuable across these conditions ( Figure 1 ) , given imperfect ( and stochastic ) choice behavior , the experienced value was not necessarily identical . However , the dependence on trial type could not be accounted for by differences in the estimated value of each movement direction – or an associated variable , such as difficulty or uncertainty in selecting the movement – between the trial types ( Figure 5 ) . In some neurons this dependence could be explained , in part , by the choice on the previous trial ( Figure 6A ) , which is informative of the rewarded direction in IS blocks . Over the course of the trial , while the influence on SNr activity of the previous choice decreased and that of the current choice increased , as might be expected given the demands of the task , the influence of trial type remained relatively constant ( Figure 6C ) . These results suggest that the SNr is differentially engaged by stimulus-guided and internally-specified movements . Previous studies in primates have shown that movement-related SNr activity was higher for memory-guided than visually-guided saccades ( Hikosaka and Wurtz , 1983a ) , and that SNr stimulation had a larger effect on memory- than visually-guided saccades ( Basso and Liu , 2007 ) . Movements selected based on a remembered stimulus can be thought of as internally specified , and in this sense our results ( Figure 3D ) are consistent with these findings and demonstrate that they generalize across species and movement types ( saccades and full-body orienting ) . However , the rewarded direction in both memory- and visually-guided trials was indicated by the stimulus , which was not the case in our IS trials , in which we sometimes observed that direction preference emerged even before stimulus delivery ( see example in Figure 3A , B ) . Further , the difference between direction preference in IS and SG trials during the delay epoch was correlated with the difference between preference in IS and SG trials during the epoch from odor port entry to odor valve open ( r = 0 . 47 , p = 4 . 8 × 10−18 ) . These results demonstrate that , in IS trials , the direction of movement was initially selected independent of the stimulus , which contributes to the difference in activity during the delay epoch that we observe between SG and IS trials . In addition , while Hikosaka and Wurtz ( 1983a ) examined only neurons that exhibited a decrease in activity around the time of contraversive saccades , we examined increasing and decreasing neurons that prefer both ipsiversive and contraversive movement ( Gulley et al . , 2002 , 1999; Handel and Glimcher , 2000; Sato and Hikosaka , 2002 ) and found that all of these groups exhibited a difference in activity between stimulus-guided and internally-specified movements . Therefore , the differences we observed between internally-specified and stimulus-guided movements extend our understanding of SNr function . Interestingly , patients with Parkinson’s disease and other BG pathologies have been reported to exhibit greater deficits in the initiation of internally-specified than visually-guided movements ( Forssberg et al . , 1984; Laplane et al . , 1984; Azulay et al . , 1999 ) . While the neural basis for this phenomenon is not well understood and remains an active area of study ( Distler et al . , 2016 ) , it has been suggested that visual cues engage ( intact ) sensorimotor pathways outside of the BG , such as the cerebellum ( Glickstein and Stein , 1991 ) . Our results suggest that differential processing of internally-specified and visually-guided movements within the BG themselves may also contribute to this clinical observation . As noted above , other studies have found that movement-related SNr activity is modulated by the relative value associated with a movement ( Bryden et al . , 2011; Sato and Hikosaka , 2002 ) , including whether the movement will be rewarded at all ( Handel and Glimcher , 2000 ) . This value dependence likely arises from dopaminergic input to the BG that is thought to convey reward-related information ( Schultz et al . , 1997 ) , and has been accounted for by a model in which , prior to stimulus presentation , reward expectation modulates striatal inputs to the SNr in order to bias downstream superior colliculus ( SC ) activity such that the most valuable movement is facilitated ( Hikosaka et al . , 2006; Wolf et al . , 2015 ) . We propose that a similar model can also explain how internally-specified movements , more generally , are facilitated ( Figure 7 ) . 10 . 7554/eLife . 13833 . 010Figure 7 . Model proposing how the observed activity of ipsiversive-preferring SNr neurons could facilitate internally-specified movements relative to stimulus-guided movements . ( A ) Line thickness corresponds to level of activity . Activity preceding stimulus-guided rightward movement . A left SNr neuron is moderately weakly active , providing moderately weak inhibition to the left SC ( superior colliculus ) . A right SNr neuron is moderately strongly active , providing moderately strong inhibition to the right SC . This pattern of activity in the SC moderately promotes rightward movement . ( B ) Activity preceding internally-specified rightward movement . Compared to A , a left SNr neuron is very weakly active , providing very weak inhibition to the left SC; and a right SNr neuron is very strongly active , providing very strong inhibition to the right SC . This pattern of activity in the SC strongly promotes rightward movement . DOI: http://dx . doi . org/10 . 7554/eLife . 13833 . 010 To illustrate this idea , consider how , given the data presented here , the relative activity between ipsiversive-preferring left and right SNr neurons would relate to an upcoming rightward movement ( we consider relative , rather than absolute , activity since this is most directly relevant to the decision – move left vs . move right – required by our task ) . Left and right SNr neurons would exhibit a larger difference in activity in IS trials than in SG trials ( Figure 3D , left ) . If ipsiversive-preferring SNr neurons primarily project to the ipsilateral SC ( Hikosaka and Wurtz , 1983b ) , then a downstream left SC neuron , the activity of which promotes rightward movement ( Felsen and Mainen , 2012; Horwitz and Newsome , 2001; Stubblefield et al . , 2013 ) will receive less inhibition from the left SNr when the movement is internally specified , thereby facilitating rightward movements that are internally specified ( Figure 7 ) . Preceding the same movement , contraversive-preferring left and right SNr neurons would also exhibit a larger difference in activity in IS trials than in SG trials ( Figure 3D , right ) . If contraversive-preferring SNr neurons comprise the 'crossed' projection to the contralateral SC ( Jiang et al . , 2003 ) , then a downstream right SC neuron would receive more inhibition from the left SNr when the movement is internally specified , again facilitating the rightward movement . However , we observed that slightly more SNr neurons prefer contraversive than ipsiversive movement ( Figure 3C ) but many fewer SNr neurons project to the contralateral than ipsilateral SC , particularly in rodents ( Beckstead et al . , 1981; Deniau et al . , 1977; Gerfen et al . , 1982; Jayaraman et al . , 1977 ) , and contraversive-preferring SNr neurons may preferentially project to non-tectal targets . Thus , SNr activity may be consistent with the facilitation of internally-specified contraversive movements . Our results therefore extend the model underlying the value-biasing view of BG function ( Hikosaka et al . , 2006 ) by suggesting that the influence of the SNr on downstream motor regions is modulated by internal representations in addition to value . In summary , we have shown that SNr activity depends on whether otherwise-identical movements are specified by internal representations of task variables or guided by an external stimulus . We suggest that this dependence may reflect a facilitation for internally-specified movements , consistent with the view that , although movements are often made in response to sensory stimuli , internal representations of priors play a critical role in guiding motor output ( Wolpert and Landy , 2012 ) . Our results are sufficiently consistent with results in primate SNr ( Handel and Glimcher , 2000 , 1999; Hikosaka and Wurtz , 1983a; Liu and Basso , 2008; Sato and Hikosaka , 2002 ) that they can inform the interpretation of previous studies ( e . g . , our proposed extensions of the model explaining the value-biasing role of the BG described above ) , while also offering novel insight into BG function . Future studies can utilize the task established here , in the experimentally-advantageous awake-behaving mouse model ( Carandini and Churchland , 2013 ) , to examine whether the difference in SNr activity preceding internally-specified and stimulus-guided movements is established by local processing or via striatal inputs ( Hikosaka et al . , 2006; Lauwereyns et al . , 2002 ) and to further elucidate how the BG control goal-directed movements .
All experiments were performed according to protocols approved by the University of Colorado School of Medicine Institutional Animal Care and Use Committee . We used male adult C57BL/6J mice ( n = 4 , determined by estimating the number of neurons required for our analyses and by the number of neurons recorded per mouse in initial experiments; aged 7–14 months at the start of experiments; Jackson Labs ) housed in a vivarium with a 12-hr light/dark cycle with lights on at 5:00 am . Food ( Teklad Global Rodent Diet No . 2918; Harlan ) was available ad libitum . Access to water was restricted to the behavioral session to motivate performance; however , if mice did not obtain ~1 ml of water during the behavioral session , additional water was provided for ~2–5 min following the behavioral session ( Smear et al . , 2011; Thompson and Felsen , 2013 ) . All mice were weighed daily and received sufficient water during behavioral sessions to maintain >85% of pre-water restriction weight . In general , mice were first trained to perform an odor-guided spatial choice task – which was comprised of 'stimulus-guided' ( SG ) trials – as described in Stubblefield et al . ( 2013 ) , and were then trained to perform 'internally-specified' ( IS ) trials . Briefly , each mouse was water-restricted and trained to interact with three ports ( center: odor port; sides: reward ports ) along one wall of a behavioral chamber ( Island Motion ) . In each trial , the mouse entered the odor port , triggering the delivery of an odor; waited 488 ± 104 ms ( mean ± SD ) for a go signal ( auditory tone ) ; exited the odor port; and entered one of the reward ports ( Figure 1A ) . Premature exit from the odor port resulted in the unavailability of reward on that trial . Odors were comprised of binary mixtures of ( + ) -carvone and ( - ) -carvone , commonly perceived as caraway and spearmint , respectively; an enantiomeric odor pair was selected to control for differences in molecular structure of odorant stimuli . In each SG trial , one of seven odor mixtures was presented via an olfactometer ( Island Motion ) : volume ( + ) -carvone/ ( - ) -carvone = 95/5 , 80/20 , 60/40 , 50/50 , 40/60 , 20/80 , or 5/95 . Mixtures in which ( + ) -carvone > ( - ) -carvone indicated reward availability only at the right port and mixtures in which ( - ) -carvone > ( + ) -carvone indicated reward availability only at the left port [we therefore refer to ( - ) -carvone as the 'left odor' ( e . g . , Figure 1D ) for simplicity] . In trials in which ( + ) -carvone = ( - ) -carvone , the probability of reward at the left and right ports , independently , was 0 . 5 . Reward , consisting of 4 μl of water , was delivered by transiently opening a calibrated water valve 10–100 ms after reward port entry . Odor and water delivery were controlled , and port entries and exits were recorded , using custom software ( available at https://github . com/felsenlab; adapted from C . D . Brody ) written in MATLAB ( MathWorks ) . Mice learned to perform SG trials within ~48 sessions ( 1 session/day ) ; detailed training stages are described in Stubblefield et al . ( 2013 ) . Mice required an additional ~5 sessions to learn to perform interleaved blocks of SG and IS trials . In every IS trial the 50/50 mixture was presented , and reward was available only at one side throughout the block . Mice were first introduced to interleaved blocks , each of which required 25 correct trials to advance to the next block . Once they performed ~70% of trials in the session correctly , the number of correct trials required per block was increased to 50 . Mice performed 5 blocks ( SG , IS , SG , IS , SG ) per session ( Figure 1B ) ; the side associated with reward switched between each IS block . Upon completing training , mice were implanted with microdrives for neural recording ( see below ) . During each of the 54 recording sessions , mice performed 321 . 81 ± 89 . 49 ( mean ± SD ) trials . Details of the surgical procedure are provided in Thompson and Felsen ( 2013 ) . Briefly , once the mouse was fully trained on the task , it was anesthetized with isoflurane and secured in a stereotaxic device , the scalp was incised and retracted , 2 small screws were attached to the skull , and a craniotomy targeting the left SNr was performed , centered at 3 . 27 mm posterior from bregma and 1 . 4 mm lateral from the midline ( Paxinos and Franklin , 2004 ) . A VersaDrive 4 microdrive ( Neuralynx ) , containing 4 independently adjustable tetrodes , was affixed to the skull via the screws , luting ( 3M ) , and dental acrylic ( A-M Systems ) . A second small craniotomy was performed in order to place the ground wire in direct contact with the brain . After the acrylic hardened , a topical triple antibiotic ointment ( Major ) mixed with 2% lidocaine hydrochloride jelly ( Akorn ) was applied to the scalp , the mouse was removed from the stereotaxic device , the isoflurane was turned off , and oxygen alone was delivered to the animal to gradually alleviate anesthetic state . Mice were administered sterile isotonic saline ( 0 . 9% ) for rehydration and an analgesic ( Ketofen; 5 mg/kg ) for pain management . Analgesic and topical antibiotic administration was repeated daily for up to 5 days , and animals were closely monitored for any signs of distress . Neural recordings were collected using four tetrodes , wherein each tetrode consisted of four polyimide-coated nichrome wires ( Sandvik; single-wire diameter 12 . 5 μm ) gold plated to 0 . 2–0 . 4 MΩ impedance . Electrical signals were amplified and recorded using the Digital Lynx S multichannel acquisition system ( Neuralynx ) in conjunction with Cheetah data acquisition software ( Neuralynx ) . Tetrode depths were adjusted approximately 23 hr before each recording session in order to sample an independent population of neurons across sessions . To estimate tetrode depths during each session we calculated distance traveled with respect to rotation fraction of the screw that was affixed to the shuttle holding the tetrode . One full rotation moved the tetrode ~250 μm and tetrodes were moved ~62 . 5 μm between sessions . The final tetrode location was confirmed through histological assessment using electrolytic lesions and tetrode tracks ( see below ) . Offline spike sorting and cluster quality analysis was performed using MClust software ( MClust-3 . 5 , A . D . Redish , et al . ) in MATLAB . Briefly , for each tetrode , single units were isolated by manual cluster identification based on spike features derived from sampled waveforms ( Figure 2B ) . Identification of single units through examination of spikes in high-dimensional feature space allowed us to refine the delimitation of identified clusters by examining all possible two-dimensional combinations of selected spike features . We used standard spike features for single unit extraction: peak amplitude , energy ( square root of the sum of squares of each point in the waveform , divided by the number of samples in the waveform ) , and the first principal component normalized by energy . Spike features were derived separately for individual leads . To assess the quality of identified clusters we calculated two standard quantitative metrics: L-ratio and isolation distance ( Schmitzer-Torbert et al . , 2005 ) . Clusters with an L-ratio of less than 0 . 70 and isolation distance greater than 6 . 5 were deemed single units , which resulted in the exclusion of 12% of the identified clusters . Although units were clustered without knowledge of interspike interval , only clusters with few interspike intervals less than 1 ms were considered for further examination . Furthermore , we excluded the possibility of including data from the same neuron twice by ensuring that both the waveforms and response properties sufficiently changed across sessions . If they did not , we conservatively assumed that we were recording from the same neuron , and only included data from one session . To verify final tetrode location we performed electrolytic lesions ( 100 μA , ~1 . 5 min per lead ) after the last recording session . One day following lesion , mice were overdosed with an intraperitoneal injection of sodium pentobarbital ( 100 mg/kg ) and transcardially perfused with saline followed by ice-cold 4% paraformaldehyde ( PFA ) in 0 . 1 M phosphate buffer ( PB ) . After perfusion , brains were submerged in 4% PFA in 0 . 1 M PB for 24 hr for post-fixation and then cryoprotected for 24 hr by immersion in 30% sucrose in 0 . 1 M PB . The brain was encased in the same sucrose solution , and frozen rapidly on dry ice . Serial coronal sections ( 60 μm ) were cut on a sliding microtome for reconstruction of the lesion site and tetrode tracks . Fluorescent Nissl ( NeuroTrace , Invitrogen ) was used to identify cytoarchitectural features of the SNr and verify tetrode tracks and lesion damage within or below the SNr . Images of SNr ( see Figure 2A ) were captured with a 10x objective lens , using an LSM 5 Pascal series Axioskop 2 FS MOT confocal microscope ( Zeiss ) . All analyses were performed in MATLAB . | An important role of the nervous system is to allow us to move around in the world . These movements are typically influenced by the goal that we want to achieve ( for example , finding food ) as well as stimuli that we sense in our environment ( for example , the smell of pizza ) . Yet we understand little about how the brain controls these sorts of goal-directed movements , even under normal conditions . This lack of basic understanding presents a big problem when it comes to treating movement disorders like Parkinson’s disease . For a long time , a collection of brain regions called the basal ganglia have been known to be important for controlling movements , although the specific role that they play in this process is not well understood . Does the brain activity that controls movements differ depending on whether the movement is made in response to a stimulus or not ? Using mice , Lintz and Felsen have now recorded the activity of individual neurons in the basal ganglia that signal to other brain regions as the animals performed a behavioral task . Different trials in the task required the mouse to make two types of otherwise-identical movements: movements based on a stimulus , and movements based on recent experiences ( and not triggered by a stimulus ) . The output activity of the basal ganglia differed under these two conditions , suggesting that the basal ganglia may play different roles in each type of movement . From the results , Lintz and Felsen could make some predictions about how the basal ganglia influence the activity of downstream regions of the nervous system that control movement . Further studies are now required to test these predictions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Basal ganglia output reflects internally-specified movements |
The stochastic multicolor labeling method ‘Brainbow’ is a powerful strategy to label multiple neurons differentially with fluorescent proteins; however , the fluorescence levels provided by the original attempts to use this strategy were inadequate . In the present study , we developed a stochastic multicolor labeling method with enhanced expression levels that uses a tetracycline-operator system ( Tetbow ) . We optimized Tetbow for either plasmid or virus vector-mediated multicolor labeling . When combined with tissue clearing , Tetbow was powerful enough to visualize the three-dimensional architecture of individual neurons . Using Tetbow , we were able to visualize the axonal projection patterns of individual mitral/tufted cells along several millimeters in the mouse olfactory system . We also developed a Tetbow system with chemical tags , in which genetically encoded chemical tags were labeled with synthetic fluorophores . This was useful in expanding the repertoire of the fluorescence labels and the applications of the Tetbow system . Together , these new tools facilitate light-microscopy-based neuronal tracing at both a large scale and a high resolution .
Neuronal circuits are the basis for brain function . Therefore , the reconstruction of neuronal wiring diagrams is key to understanding circuit function . Fluorescence imaging has been a powerful approach in visualizing the three-dimensional structure of neuronal morphology . In particular , fluorescent proteins are useful for labeling genetically-defined neuronal populations . In recent years , a number of tissue-clearing methods have been developed , and these have been optimized for use with fluorescent proteins and deep-tissue antibody staining ( Hama et al . , 2011; Chung et al . , 2013; Ke et al . , 2013; Susaki et al . , 2014; Richardson and Lichtman , 2015 ) . These new tools have expanded the scale of the available technologies for fluorescence imaging to whole-organ and whole-organism levels . It is still difficult , however , to dissect and trace an individual neuron from a brain sample labeled with a single type of fluorescent protein . One way to overcome this problem is to improve the spatial resolution . Recently , we developed a tissue-clearing agent for high-resolution three-dimensional fluorescence imaging , named SeeDB2 ( Ke et al . , 2013 ) . SeeDB2 was designed to minimize spherical aberrations , allowing for high-resolution imaging including super-resolution microscopy . In this approach , there was much improvement in the z-resolution , a critical factor for dissection of neuronal fibers crossing over along the z-axis . Similarly , expansion microscopy is also a promising new approach used to improve resolution in three-dimensional fluorescence imaging ( Chen et al . , 2015; Ku et al . , 2016; Tillberg et al . , 2016; Chang et al . , 2017 ) . Another approach to the dissection of neuronal circuits is multicolor labeling . To facilitate the dissection of individual neurons , a transgenic multicolor labeling method , Brainbow , has been developed , in which three different fluorescent proteins were expressed in a stochastic manner ( Livet et al . , 2007; Cai et al . , 2013; Loulier et al . , 2014 ) . Brainbow used the Cre-loxP system to express one of the three fluorescent protein genes stochastically in a transgene . When multiple copies of the transgene cassette are introduced , stochastic choices will result in a combinatorial expression of these three genes with different copy numbers , producing dozens of color hues . Although the Brainbow concept is powerful for discriminating between numerous neurons using light microscopy , the existing Brainbow methods are of limited use for neuronal tracing . This is because the stochastic and combinatorial expression of fluorescent proteins is possible only at low copy number ranges for the transgenes , so that the expression levels of the fluorescent proteins were not sufficiently high for bright and high-resolution imaging of axons and dendrites . Therefore , many of the previous studies were forced to use subsequent antibody staining to produce reliable neuronal tracing . In the present study , we utilized the Tet-Off system ( Tetbow ) to develop a multicolor labeling method with enhanced expression . As vector ( plasmid and virus ) -mediated gene transfer has become a versatile tool in modern neuroscience , we aimed to perform multicolor labeling using these tools . As a proof-of-concept experiment , we demonstrated the ability to trace axons of individual neurons on the scale of several millimeters in the mouse olfactory system . To improve the stability of the fluorescence labels after harsh tissue-clearing treatment , we also developed a Tetbow system with chemical tags . When combined with the advances in the growing field of tissue-clearing techniques , these new multicolor labeling strategies should facilitate neuronal tracing at higher densities and resolutions .
Earlier Brainbow methods utilized transgenic animals for stochastic multicolor labeling . They utilized the Cre-loxP system , in which DNA recombination resulted in a stochastic selection of one fluorescent protein gene out of three ( or more ) choices . When multiple copies of the Brainbow transgene were introduced into the genome , stochastic recombination produced a variety of color hues based on different copy numbers of expressed fluorescent protein genes ( collectively called XFPs ) ( Livet et al . , 2007 ) . In recent years , however , plasmid or virus vector-mediated gene transfer has become a more versatile strategy in neuroscience . We therefore tried to optimize a multicolor labeling method for vector-mediated gene transfer . Previously , an adeno-associated virus ( AAV ) -mediated Brainbow method ( AAV-Brainbow ) has been reported ( Cai et al . , 2013 ) . However , in the vector-mediated gene transfer , the Cre-loxP system is not essential for the stochastic and combinatorial expression of XFP genes ( Kobiler et al . , 2010; Weber et al . , 2011; Siddiqi et al . , 2014 ) . We can easily make color variations by introducing a mixture of three different XFP constructs: as long as the copy number of the introduced genes is small , labeled neurons will still produce a variety of colors irrespective of whether the Cre-loxP system is used or not ( Figure 1A–C ) . It should be noted , however , that color variation reduces as the copy number of the introduced genes increases . We can estimate the optimum copy number of the introduced XFP genes as follows . We considered that the number of introduced genes will follow a Poisson distribution ( Kobiler et al . , 2010 ) . When three different XFP genes are introduced at 20 copies/cell/color on average , a similar number of copies will be introduced into each neuron , and only a small degree of color variation will be generated ( Figure 1C ) . By contrast , if the copy number is too small , many of the neurons will express just one XFP gene ( Figure 1A ) . When these three genes are introduced at an average 2 copies/cell/color , much larger color variations will be produced ( Figure 1B ) . We wanted to determine the optimum copy number of the expressed XFP genes on the basis of this simulation . We plotted the color values for cells in the color-coding space after intensity normalization ( total intensity = 1 ) . In this coding space , each dimension represents the intensity of one of the three colors ( Red , Green , and Blue in pseudocolor representation ) . The mean Euclidean distance ( d ) for two randomly chosen cells was greater when the copy number was lower ( Figure 1D , E ) , but this does not necessarily mean that we can discriminate between many cells , as the two cells are more likely to become the same color when copy number is too low ( Figure 1A ) . We therefore calculated the probability that two randomly chosen cells are discriminated on the basis of a given threshold distance in the color-coding space . Here we considered that cells within the threshold distance ( d ) from a reference are indiscriminable in the color-coding space; cells outside of the threshold distance were considered discriminable from the reference ( Figure 1F ) . In our simulation , when we assumed a threshold distance of 0 . 1 , 95 . 3% of cells could be discriminated from a given cell when XFP genes were expressed at 2 copies/color/cell ( Figure 1G ) . We also found that experienced researchers can discriminate two colors separated by 0 . 1 Euclidean distances in the color-coding space at 94 . 5 ± 1 . 73% accuracy ( mean ± S . D . ; Figure 1H and Figure 1— figure supplement 1 ) . Thus , ~2 copies/color/cell is the optimum when the color hues are judged visually by human experimenters . We evaluated this prediction using the in utero electroporation of plasmid vectors into layer 2/3 cortical pyramidal neurons ( electroporated at embryonic day 15 ) . We introduced a mixture of three separate plasmid vectors encoding mTurquoise2 ( blue ) , EYFP ( green ) , and tdTomato ( red ) genes under a CAG promoter . When these plasmids were expressed at high copy numbers ( CAG High; DNA solution introduced was 1 μg/μL/plasmid; 3 μg/μL in total; see 'Materials and methods' section for details ) , only small color variations were produced ( Figure 2A , C ) . When the DNA concentrations were reduced ( CAG Low; 0 . 25 μg/μL/plasmid ) , the color variation increased ( Figure 2B ) , but the overall fluorescence levels were much reduced ( Figure 2E ) . Thus , there is a trade-off between the expression levels of fluorescent proteins and color variations , and this is the reason why the previous Brainbow methods were unable to produce color variations that were bright enough for labeling . We therefore wanted to enhance the expression levels of fluorescent proteins while maintaining the required low copy numbers of XFP genes . Recent studies indicated that the tetracycline response element ( TRE ) promoter ensures much higher expression levels than the CAG promoter when expressed with a tetracycline trans-activator ( tTA ) ( Madisen et al . , 2015; Sadakane et al . , 2015 ) . We therefore used the tTA-TRE ( Tet-Off ) system instead of a common CAG promoter . We also introduced a Woodchuck hepatitis virus posttranscriptional regulatory element ( WPRE ) sequence in the 3’′ UTR of XFP genes to improve their expression levels . Earlier Brainbow techniques used membrane-bound XFPs because unmodified XFPs labeled the somata too brightly and affected the tracing of nearby neurites ( Cai et al . , 2013 ) . However , this problem can be easily solved by minimizing spherical aberration in microscopy by using an index-matched clearing agent , SeeDB2 ( Ke et al . , 2016 ) . Furthermore , unmodified XFPs label axons and dendrites much more brightly than the membrane-bound ones ( data not shown ) . We therefore used unmodified XFPs instead of membrane-bound XFPs . These new set of constructs ( named Tetbow; Figure 2—figure supplement 1 ) achieved much higher XFP expression levels when compared with the CAG-promoter plasmids ( Figure 2B ) . When we quantified the expression levels of EYFP , we observed a six-fold increase in fluorescence levels ( Figure 2E; CAG 0 . 25 μg/μL vs . Tetbow 0 . 25 μg/μL; n = 104 and 260 , respectively , p<0 . 0001 , Kruskal Wallis with Dunn’s post hoc test ) . Tetbow allowed for much more robust multicolor labeling than the Brainbow constructs when introduced by in utero electroporation , allowing more reliable axon tracing ( Figure 2—figure supplement 2 ) . Owing to the enhanced expression , the Tetbow system achieved bright labeling ( Figure 2E ) while maintaining the color variations produced by low copy numbers of XFP genes . When CAG-XFP genes were introduced at high concentrations ( 1 μg/μL each ) , the color variations were small; by contrast , the Tetbow constructs ( 0 . 25 μg/μL ) produced much larger variations ( Figure 2D ) . In the color discrimination analysis , we confirmed that different cells are better discriminated by Tetbow than in the CAG High condition ( Figure 2F ) . We further tried to determine the optimum concentration of plasmids for Tetbow when labeled with in utero electroporation . Among the four plasmid concentrations we tested ( 0 . 05 , 0 . 1 , 0 . 25 , and 0 . 5 μg/μL each ) , color discrimination performance was comparable . Paradoxically , however , the expression levels of XFP were the highest at 0 . 25 μg/μL each , not at 0 . 5 . We therefore considered the possibility that a moderate expression level of tTA is critical for the optimum expression . When 0 . 1 , 0 . 25 , and 0 . 5 μg/μL of CAG-tTA plasmid was co-introduced with 0 . 25 μg/μL each of TRE-XFP plasmids , the highest expression level was found with 0 . 1 μg/μL CAG-tTA ( Figure 3D–F ) . Thus , too high a concentration of tTA leads to the suppression of TRE-XFP genes . Consistent with this assumption , when the WPRE sequence was added to the CAG-tTA plasmid , the expression level of TRE-XFP genes was reduced ( Figure 3—figure supplement 1 ) . Thus , the expression level of tTA needs to be moderate to achieve the highest expression of TRE-XFP genes . We described the use of SeeDB2 as a tissue-clearing agent ( Ke et al . , 2016 ) . SeeDB2 is designed to minimize spherical aberrations for glycerol ( SeeDB2G ) or oil ( SeeDB2S ) immersion objective lenses , making high-resolution 3D imaging possible . Furthermore , various fluorescent proteins were best preserved in SeeDB2 , much more so than in commercialized mounting media or other tissue-clearing agents ( Ke et al . , 2016 ) . In high-resolution imaging , we must obtain photons from a limited volume , and thus the fluorescence intensity must be sufficiently high . Therefore , the combination of Tetbow and SeeDB2 is ideal for high-resolution volumetric multicolor imaging . We introduced Tetbow constructs into Layer 2/3 neurons in the cerebral cortex using in utero electroporation . After clearing with SeeDB2G , the fluorescence levels of XFPs with Tetbow were strong enough to allow visualization of the fine detail of neuronal morphology in 3D ( Figure 4A and Video 1 ) . When we analyzed adult brain slices ( P70 ) , detailed structures of dendritic spines ( Figure 4B , Figure 4—figure supplement 1 , and Video 2 ) and axonal boutons ( Figure 4C ) were clearly visualized with Tetbow . It should be noted that we obtained these high-resolution images of synaptic structures using solely native fluorescence of XFPs , with no antibody staining . To evaluate the versatility of Tetbow , we also tested other types of neurons using in utero electroporation . For example , we were able to label brightly mitral and tufted ( M/T ) cells in the olfactory bulb with Tetbow . It is known that each glomerulus in the olfactory bulb is innervated by 20–50 M/T cells ( Ke et al . , 2013; Imai , 2014 ) . When we looked at each glomerulus , dendrites from different mitral cells were clearly visualized and were distinguishable by their different colors ( Figure 5 and Video 3 ) . Thus , Tetbow can be used to analyze the detailed dendrite wiring diagram of individual M/T cells , including ‘sister’ M/T cells , which are connected to the same glomerulus . Aqueous tissue-clearing agents are useful for large-scale three-dimensional imaging with fluorescent proteins . However , to clear lipid-rich myelinated axons completely , harsh clearing treatments , such as the use of detergents , solvents , and heating , are inevitable ( Dodt et al . , 2007; Chung et al . , 2013; Renier et al . , 2014; Susaki et al . , 2014; Murray et al . , 2015 ) . Indeed , there is a trade-off between the transparency of the tissues and damage to the tissues and fluorescent proteins ( Ke et al . , 2013; Ke et al . , 2016 ) . For example , most of the solvent-based clearing methods , such as BABB , quench fluorescent proteins . To overcome this problem , chemical tags could be a promising alternative to fluorescent proteins ( Kohl et al . , 2014; Sutcliffe et al . , 2017 ) . Genetically encoded chemical tags , such as SNAP , Halo , CLIP , and TMP form covalent bonds with their cognate substrate , and fluoresce with synthetic labels . The molecular weights of these ligands are relatively small , allowing easy penetration into thick tissues . Once they form covalent bonds with their substrates , the fluorescence remains stable even under harsh clearing conditions . We tested Tetbow multicolor labeling with three different chemical tags , SNAP , Halo , and CLIP ( Figure 6—figure supplement 1 ) . We introduced these three chemical tag genes into L2/3 cortical pyramidal neurons using in utero electroporation . Brains were fixed , and the three chemical tags were visualized with synthetic fluorescence labels: SNAP-Surface 488 , HaloTag TMR Ligand , and CLIP-Surface 647 , respectively ( Figure 6A ) . Like fluorescent proteins , chemical tags allowed for the robust multicolor labeling of neurons using the Tetbow system ( Figure 6B ) . Owing to the low molecular weight of the substrates , 1 mm-thick mouse brain samples were efficiently labeled ( Figure 6—figure supplement 2 ) . After tissue clearing with solvent-based tissue clearing agents , 3DISCO or BABB , fluorescent proteins were largely quenched ( data not shown ) ( Ke et al . , 2013 ) ; synthetic fluorophores bound to chemical tags , however , were bright and stable under these clearing conditions ( Figure 6B ) . AAVs are also becoming a versatile gene delivery tool in neuroscience . However , the size of the conventional Brainbow gene cassettes was too large to allow them to be packaged into an AAV vector ( <5 kb ) . For this reason , a previous study divided the Brainbow cassette into two separate AAVs with two XFP genes each , and employed Cre-loxP recombination ( Cai et al . , 2013 ) . As described above , however , the Cre-loxP system is not needed to achieve multicolor labeling using AAV-mediated gene expression . Using the Tetbow strategy ( i . e . multiple AAV vectors with different XFP genes ) , we can solve the size problem , and achieve improved multicolor labeling using simplified DNA constructs . We generated four separate AAV vectors carrying SYN1-tTA , TRE-mTurquoise2 , TRE-EYFP , and TRE-tdTomato genes ( Figure 7—figure supplement 1 ) . The human SYNI promoter was used to express tTA specifically in neurons . These four AAV vectors ( serotype AAV2/1 ) were injected into the cerebral cortex of adult mice ( P56 ) and analyzed two weeks later . We found a stochastic and combinatorial expression of the encoded fluorescent proteins in Layer five neurons when injected at 4 × 108 gc/mL of AAV-SYN1-tTA and 3 × 1010 gc/mL of AAV-TRE-XFPs ( Figure 7A ) . As expected from our simulation ( Figure 1 ) , the virus titer was critical for producing color variations . Higher virus titers led to reduced color variations ( Figure 7B , right ) . In the cerebral cortex , 1–3 × 1010 gc/mL of AAV-TRE-XFPs was found to be optimum on the basis of the resultant expression levels and color variations ( Figure 7C and D ) , but the optimum range may be different for different cell types ( Figure 7—figure supplement 2 ) , injection volumes , or virus serotypes . Expression levels were sufficiently high for high-resolution imaging of neuronal morphology , including the visualization of dendritic spines ( Video 4 ) . It should be noted , however , that the expression of Tetbow AAVs can lead to toxic effects on cellular functions after a prolonged incubation period , as a result of the extremely high levels of expression . For example , cortical neurons started to show an aggregation of XFPs and displayed morphological abnormalities 4 weeks after virus injection , whereas olfactory bulb neurons were best visualized at 4 weeks . Thus , the optimum incubation time may be different for different cell types . The bright multicolor labeling method , Tetbow , is particularly useful for the analysis of long-range axonal projection of a population of neurons . To test the utility of Tetbow , we focused on mitral and tufted ( M/T ) cells in the olfactory bulb , which project axons ( up to several millimeters ) to the olfactory cortex , including the anterior olfactory nucleus , olfactory tubercle , piriform cortex , cortical amygdala , and lateral entorhinal cortex . Previous studies have performed axon tracing on populations of M/T cells from glomeruli in the olfactory bulb using dye injections ( Nagayama , 2010; Sosulski et al . , 2011 ) , but these studies could not fully dissect individual axons . Other studies performed single-cell axon tracing using hundreds of serial brain sections , but these analyses were highly laborious and time-consuming ( Ghosh et al . , 2011; Igarashi et al . , 2012 ) . Owing to these limitations , we do not yet understand fully how the odor inputs into individual M/T cells are conveyed to the olfactory cortex . To examine the axonal projections of M/T cells from the olfactory bulb , we injected Tetbow AAVs into the mouse olfactory bulb . First , Tetbow AAVs were expressed in most of olfactory bulb neurons . When compared to single-color labeling , the stochastic and combinatorial expression of XFPs was helpful in improving tracing performance ( Figure 8—figure supplement 1 ) . Efficient axon tracing was further facilitated by the local injection of Tetbow AAVs ( Figure 8A , B ) . In the olfactory bulb , many types of neurons were labeled , but only M/T cells projected their axons to the olfactory cortex . To facilitate high-resolution fluorescence imaging with a limited working distance of the objective lens ( 20x , NA = 0 . 75 , WD = 0 . 66 mm ) , we flattened the olfactory cortical area onto a glass slide ( Sosulski et al . , 2011 ) . After the fixation and tissue clearing , axons of individual M/T cells were clearly visualized with XFPs in the entire area of the olfactory cortex ( Figure 8D and Video 5 ) . The unique color hue was largely preserved at the average level , if not at a single-pixel resolution , from proximal to distal part in each axon ( Figure 8E ) . Across the three image areas , the median Euclidean distance for the same neuron was 0 . 049 ( interquartile range = 0 . 063 , n = 34 pairs ) . The unique color hues and their consistency facilitated manual reconstruction of individual M/T cell axons in the olfactory cortex ( Figure 8F , Figure 8—figure supplement 2 , and Video 6 ) . Highly divergent patterns of axonal collaterals were observed among labeled M/T cell axons , complementing earlier findings ( Nagayama , 2010; Ghosh et al . , 2011; Sosulski et al . , 2011 ) .
To achieve the stochastic expression of multiple fluorescent proteins , it is important to optimize the number of genes that are introduced per cell . We introduced a limited number of copies of plasmids or virus vectors into neurons to enable the stochastic expression of XFPs based on a Poisson distribution . To enhance the expression levels of XFPs , we utilized a Tet-Off system , in which the tetracycline trans-activator ( tTA ) binds to the TRE promoter to drive the expression of genes of interest . In this way , we achieved multicolor labeling with much-improved brightness . Previously , the electroporation and virus infection of the Brainbow construct introduced just a small number of genes per cell , and as a result , the fluorescence levels were not sufficiently high for reliable neuronal tracing ( Kobiler et al . , 2010; Egawa et al . , 2013 ) ( see also Figure 2—figure supplement 2 ) . In fact , Brainbow methods often had to employ antibody staining to enhance the fluorescence signals ( Cai et al . , 2013 ) . However , our Tetbow strategy provides enhanced fluorescence , allowing for high-resolution imaging without the need for antibody staining . As a result , our Tetbow strategy has opened up a new opportunity for large-scale high-resolution neuronal tracing using tissue clearing . Our Tetbow strategy also overcomes a size-limit problem associated with AAV vectors . As we introduced tTA and TRE-XFP gene cassettes with separate AAV vectors , DNA construction was also simplified . Using our Tetbow AAVs , we could clearly visualize different neurons with different color hues at synaptic resolution . It should be noted , however , that using the correct virus titer is critical for the generation of color variations . If the virus titer is too low , many of the labeled neurons will express just one XFP gene . By contrast , if the virus titer is too high , many of the labeled neurons will express all three genes at similar levels . Recently , another group employed a similar strategy using a Tet-Off system and a newly engineered AAV for intravenous transduction; but the brightness levels that they achieved were not as good as those provided by CAG vectors ( Chan et al . , 2017 ) . To achieve the highest expression levels of XFPs , the expression level of tTA needs to be optimized ( Figure 3 ) . The Tet-Off system is also useful for controling the sparseness of XFP expression; by simply diluting the tTA vector , the expression of XFP can be sparsened without affecting color variations ( Chan et al . , 2017 ) . AAV- based Tetbow is useful for a relatively broad set of applications , including use in less genetically tractable model animals . For example , an AAV-based Tet-Off system has already been tested in the marmoset brain ( Sadakane et al . , 2015 ) . Like mice , these animals demonstrated high expression levels of fluorescent proteins when the Tet-Off system was used . Thus , our Tetbow strategy can be useful for neuronal tracing studies in various species including primates . In the present study , we also extended the Tetbow method to include chemical tags . Current tissue clearing methods are intended for the use of fluorescent proteins , but chemical tags can become a good alternative to the fluorescent proteins . First , unlike fluorescent proteins , once labeled with synthetic fluorophores , chemical tags are stable under harsh clearing conditions . This means that we can have a broader choice of clearing methods , particularly for lipid-rich and thick tissues . Chemical tags may also be useful for expansion microscopy , where the stability of fluorescent proteins has been a challenging aspect ( Chen et al . , 2015; Ku et al . , 2016; Tillberg et al . , 2016; Chang et al . , 2017 ) . The fluorescence labeling of chemical tags is much easier than the antibody staining of XFPs ( Kohl et al . , 2014 ) . Furthermore , synthetic fluorophores penetrate much deeper than antibodies because of their smaller size . Second , more choices of fluorescence spectrum and photochemical properties are available with chemical tags than with fluorescent proteins . For example , synthetic fluorophores cover the spectrum from UV to near-IR range , expanding the possible spectrum range when imaging . In addition , autonomously blinking fluorophores are very useful for localization-based super-resolution microscopy ( Uno et al . , 2014 ) . Multicolor 3D PALM/STORM imaging of cleared tissues would be an interesting possibility in the future . In recent years , several tissue-clearing methods that are optimized for fluorescence imaging have been developed , expanding the imaging scale in light-microscopy-based neuronal tracing . However , the imaging resolution was not sufficiently high for densely labeled neuronal circuits . Therefore , we could only look at population-level connectivity ( Oh et al . , 2014 ) or at a very small subset of neurons ( Economo et al . , 2016 ) with light microscopy . To examine the projection diagram of individual neurons , a single-cell barcoding and RNA sequencing approach , MAPseq , has been proposed ( Kebschull et al . , 2016 ) . Although MAPseq is a promising new tool in the study of area-to-area connectivity for hundreds of neurons ( Han et al . , 2018 ) , we cannot know the finer details of their neuronal morphology . Currently , saturated connectomics is only possible with electron microscopy ( Kasthuri et al . , 2015 ) , but the analytical throughput is currently not sufficient for the study of long-range projections . The combination of Tetbow and tissue clearing can become a useful tool to fill the gap , allowing the dissection of densely labeled neurons at a large scale . In the present study , we employed our Tetbow method to analyze axonal projection profiles for M/T cells in the mouse olfactory bulb ( Figure 8 ) . We were able to find labeled axons in all the areas of the olfactory cortices that we examined ( piriform cortex , cortical amygdala , and lateral entorhinal cortex ) , but it remains unclear whether we could completely label all the fibers to their termini . In fact , it is extremely difficult to know whether the labeled termini are really termini or are interrupted by small unlabeled gaps . Even if the second scenario is very likely , it is difficult to know which segments are in fact connected to each other , but the variable color hues can help . Single-cell labeling is advantageous to avoid such ambiguity , although the throughput is limited for such analyses ( Igarashi et al . , 2012 ) . By contrast , multicolor labeling may be useful to compare the projection patterns in the same animals . In our analysis shown in Figure 8F , we traced axons using conservative criteria: we terminated tracing when labeled axons were interrupted by small gaps . Nevertheless , the quality of axonal tracing was comparable to that in earlier studies ( Nagayama , 2010; Ghosh et al . , 2011 ) . To further improve the tracing performance ( including up to the axon termini ) , it will be important to improve labeling methods in order to fill the thin neuronal fibers completely and evenly . In this study , we employed manual neuronal tracing to Tetbow data , and were able to trace individual M/T cell axons successfully at the millimeter ( >6 mm ) scale . Color hues were consistent at a global scale ( Figure 8E ) , but not perfect in high-magnification images ( Figure 8D ) . Chromatic aberration can also be a problem in the high-resolution imaging of thick tissues . Another important challenge will be to improve the color hue consistency throughout a neuron and to develop auto-tracing software that is optimized for the multicolor-labeled neurons .
pCAG-mTurquoise2/EYFP/tdTomato were assembled by using pCAG-CreERT2 ( Addgene #14797 , RRID: Addgene_14797 , from Dr . Cepko ) , pmTurquoise2-N1 ( Addgene #60561 , RRID: Addgene_60561 , from Dr . D . Gadella ) , EYFP ( Clontech ) , and tdTomato ( a gift from Dr . R . Tsien ) . To generate a backbone vector for pTRE-XFP , an Xho-Xho fragment containing TRE-SV40 poly A was transferred from pTRE-Tight ( #631059 , Clontech ) to pBluescript II SK ( + ) ( # 212205 , Agilent ) to ensure high-copy expression in Escherichia coli . The WPRE sequence was PCR amplified from an aavCAG-pre-mGRASP-mCerulean vector ( Addgene #34910 , RRID: Addgene_34910 , a gift from Dr . J . Kim ) . The tTA sequence in pCAG-tTA and pAAV2-SYN1-tTA was derived from the tTA2 section of the pTet-Off Advanced vector ( Clontech ) . SNAPf and CLIPf tag genes were obtained from New England Biolabs ( #N9183S , #N9215S ) , and the HaloTag gene was obtained from Promega ( #G3780 ) . AAV2-Tetbow vectors were generated by modifying AAV2-miniSOG-VAMP2-tTA-mCherry ( Addgene #50970 , RRID: Addgene_50970 , from Dr . R . Tsien ) . Maps for the Tetbow plasmids are shown in figure supplements . The Tetbow plasmids and their sequences have been deposited at Addgene ( https://www . addgene . org/Takeshi_Imai/ ) with Addgene #104102–104112 . See also SeeDB Resources ( https://sites . google . com/site/seedbresources/ ) for updated information . As for the Brainbow experiments , we used AAV-EF1a-BbTagBY ( Addgene #45185 , RRID: Addgene_45185 ) and AAV-EF1a-BbChT ( Addgene #45186 , RRID: Addgene_45186 ) plasmids . All animal experiments were approved by the Institutional Animal Care and Use Committee of the RIKEN Kobe Institute and Kyushu University . ICR mice ( Japan SLC , RRID: MGI: 5652524 ) were used for in utero electroporation and C57BL/6N mice ( Japan SLC , RRID: MGI: 5658686 ) were used for AAV experiments ( age , P56-70; male ) . To obtain brain tissue , mice were i . p . injected with an overdose of nembutal ( Dainippon Sumitomo Pharma ) or somnopentyl ( Kyoritsu Seiyaku ) to produce deep anesthesia , followed by an intracardiac perfusion with 4% paraformaldehyde ( PFA ) in phosphate buffered saline ( PBS ) . Excised brain samples were post-fixed with 4% PFA in PBS at 4°C overnight . Samples were then embedded in 4% agarose and cut into slices of 220 , 500 , or 1000 μm thick with a microslicer , PRO7N ( Dosaka EM ) . The in utero electroporation of plasmids to the cerebral cortex and mitral cells was performed as described previously ( Saito , 2006; Ke et al . , 2013; Muroyama et al . , 2016 ) . Pregnant ICR mice were anesthetized with medetomidine ( 0 . 3 mg/kg ) , midazolam ( 4 mg/kg ) , and butorphanol ( 5 mg/kg ) , or with ketamine ( 100 mg/kg ) and xylazine ( 10 mg/kg ) . The uterine horns carrying embryos were exposed through a midline abdominal incision . To label L2/3 neurons in the cortex , 1 μL of plasmid solutions diluted in PBS was injected into the lateral ventricle of the embryos at E15 using a micropipette made from a glass capillary . Electric pulses ( single 10 ms poration pulse at 72 V , followed by five 50 ms driving pulses at 40–42 V with 950 ms intervals ) were delivered by a CUY21EX electroporator ( BEX ) and forcep-type electrodes ( 5 mm diameter , #LF650P5 , BEX ) . To introduce pCAG-XFP vectors , the labels pCAG-mTurquoise2 , EYFP , and tdTomato were injected into the lateral ventricle and electroporated at high-copy ( 1 μg/μL each , 3 μg/μL in total ) or low-copy ( 0 . 25 μg/μL each , 0 . 75 μg/μL in total ) numbers . pTRE-XFP vectors with pCAG-tTA were injected and electroporated at low-copy numbers ( 0 . 25 μg/μL each , 1 . 0 μg/μL in total ) when not specified; in Figure 3 , the exact plasmid concentrations are specified . To label mitral cells in the olfactory bulb , in utero electroporation was performed at E12 . Electric pulses ( single 10 ms poration pulse at 72 V , followed by five 50 ms driving pulses at 36 V with 950 ms intervals ) were delivered with forceps-type electrodes ( 3 mm diameter , #LF650P3 , BEX ) . After the electroporation , the uterine horns were placed back into the abdominal cavity , and the abdominal wall and skin were sutured . AAV vectors were generated using the AAVpro Helper Free System ( AAV1 , #6673 , Takara ) from Takara and the AAVpro 293 T cell line ( #632273 , Clontech ) following the manufacturers’ instructions . The backbone pAAV plasmid is for AAV2 . Thus , the serotype used in this study is AAV2/1 . AAV vectors were generated by AAVpro 293 T cells ( #632273 , Clontech ) . AAVpro 293T is a commercialized cell line for production of AAV vectors . We did not test for mycoplasma contamination in our hands . In our experiments , cells within 10 passages were used for virus production . AAV vectors were purified using the AAVpro Purification Kit All Serotypes ( #6666 , Takara ) . Virus titers were then determined by qPCR using the AAVpro Titration Kit ( #6233 , Takara ) and the StepOnePlus system ( ThermoFisher ) . To infect the AAV vectors , C57BL/6 mice ( P56-70 ) were anesthetized with ketamine ( 100 mg/kg ) and xylazine ( 10 mg/kg ) , and an AAV virus cocktail was injected into the brain using the nanoject II system and glass capillaries ( #3-00-203-G/XL , Drummond ) . The injection volume was 207 nL in Figure 7 , 138 nL in Figures 8 , and 207 nL ×5 different locations in Figure 8—figure supplement 1 . The final concentration of the virus cocktail was 4 × 108 gc/ml for AAV2/1-SYN1-tTA2 and 1 × 1010–1 × 1011 gc/mL each for AAV2/1-TRE-mTurquoise2-WPRE , AAV2/1-TRE-EYFP-WPRE , and AAV2/1-TRE-tdTomato-WPRE . The mice were sacrificed 2 or 4 weeks after viral injection . It should be noted that the expression of Tetbow AAV can lead to toxic effects for cellular functions after prolonged incubation because of the extremely high levels of expression . For example , cortical neurons started to show aggregation of XFPs and morphological abnormality 4 weeks after virus injection . Olfactory bulb neurons were best visualized 4 weeks after virus injection without obvious sign of toxicity . The optimum timing for the analysis may be different for different cell types . The following stereotaxic coordinates were used for AAV injection . Distance in millimeters from the Bregma for the anterior ( A ) – posterior ( P ) , and lateral ( L ) positions , and from the brain surface toward the ventral ( V ) directions are indicated . Cortical layer five neurons: P=1 . 5 , L = 1 . 5 , V = 0 . 5; hippocampal CA1 neurons: P=1 . 5 , L = 1 . 5 , V = 1; olfactory bulb granule cells: A = 4 . 5 , L = 0 . 5 , V = 0 . 5; olfactory bulb M/T cells: A = 4 . 5 , L = 0 . 5 , V = 0 . 3 . SNAP , CLIP , and HaloTags were visualized with their substrates , SNAP-Surface 488 ( #S9124S , New England Biolabs ) , HaloTag TMR Ligand ( #G8252 , Promega ) , CLIP-Surface 647 ( #S9234S , New England Biolabs ) , respectively . Brain slices of 220 or 1000 μm thickness were incubated with the substrates ( 2 μM each ) in 2 ml 2% saponin in PBS overnight . The slices were then washed with PBS ( 3 × 30 min ) . Brain-slice samples were cleared with SeeDB2G for imaging when not specified . SeeDB2G is designed for high-resolution imaging with glycerol-immersion objective lenses ( Ke et al . , 2016 ) . PFA-fixed brain samples ( embedded in agarose , cut at 220 , 500 , or 1000 μm thick ) were cleared at room temperature ( 25°C ) with a 1:2 mixture of Omnipaque 350 ( #081–106974 , Daiichi-Sankyo ) and H2O with 2% saponin ( #30502–42 , Nakalai-tesque ) for 6 hr ( 3 ml in 5 ml tube ) , a 1:1 mixture of Omnipaque 350 and H2O with 2% saponin for 6 hr , and finally Omnipaque 350 with 2% saponin overnight . Cleared samples were then mounted in SeeDB2G ( Omnipaque 350 ) on a glass slide using a 0 . 2 mm thick silicone rubber sheet ( AS ONE , #6-9085-13 , Togawa rubber ) and glass coverslips ( #0109030091 , Marienfeld , No . 1 . 5H ) ( Ke et al . , 2016 ) . A detailed step-by-step protocol has been published in bio-protocol ( Ke and Imai , 2018 ) . To quantify the fluorescence intensity of M/T cell axons ( Figure 2—figure supplement 2 ) , whole-brain samples were cleared with SeeDB2 and placed on a glass-bottomed dish for imaging and quantification . To analyze long-rage axonal projection of M/T cells ( Figure 8 ) , a right-brain hemisphere was dissected , and the dorsal part and subcortical matter was trimmed away with forceps and a scalpel . The remaining part , containing all of the olfactory cortical areas , was flattened with a 700 mm spacer and fixed with 4% PFA in PBS overnight ( Sosulski et al . , 2011 ) . Then , the sample was treated with ScaleCUBIC-1 ( 25% ( wt/wt ) urea ( #219–00175 , Wako ) , 25% ( wt/wt ) N , N , N’ , N’-tetrakis ( 2-hydroxypropyl ) ethylenediamine ( #T0781 , TCI ) , and 15% ( wt/wt ) Triton X-100 ( #12967–45 , Nakalai-tesque ) in H2O ) ( Susaki et al . , 2014 ) for 24 hr to remove lipids from the lateral olfactory tract , washed with PBS , and then cleared with SeeDB2G as described above . PFA-fixed brain samples were serially incubated in 50% , 80% , and 100% ethanol , each for 8 hr . They were then incubated in 100% ethanol for 12 hr , and then in hexane ( #17922–65 , Nakalai-tesque ) for 12 hr . Samples were cleared ( benzyl alcohol ( # 402834–100 ML , SIGMA ) :benzyl benzoate ( #025–01336 , Wako ) =1:2 ) with gentle shaking for 24 hr . PFA-fixed brain samples were serially incubated in 50% , 70% , 80% , and 100% tetrahydrofuran ( THF , # 207–17905 , Wako ) , each for 1 hr . They were then incubated in 100% THF for 12 hr , and then in dibenzyl ether ( DBE , #022–01466 , Wako ) for 3 hr ( Ertürk et al . , 2012 ) . Confocal images were acquired using an inverted confocal microscope , TCS SP8X with HyD detectors ( Leica Microsystems ) . Type G immersion ( refractive index 1 . 46 , Leica ) was used for a 20x multi-immersion objective lens ( HC PL APO 20x/0 . 75 IMM CORR CS2 , NA0 . 75 , WD 0 . 66 mm ) and a 63x glycerol-immersion objective lens ( HC PL APO 63x GLYC CORR CS2 , NA1 . 3 , WD 0 . 28 mm ) . Type F immersion ( refractive index 1 . 518 , Leica ) was used for a 40x oil-immersion objective lens ( HC PL APO 40x OIL CS2 , NA1 . 3 , WD 0 . 24 mm ) . Low-magnification images in Figure 2—figure supplement 2 were taken with a 10x dry lens ( HC PL APO 10x/0 . 40 CS , NA0 . 4 , WD 2 . 2 mm ) . Diode lasers of 442 or 448 nm , 488 nm , and 552 nm wavelength were used for mTurquoise2 , EYFP , and tdTomato , respectively . In some experiments , a white light laser was used to image mTurquoise2 , EYFP , and tdTomato . To image chemical tags , SNAP-Surface 488 , HaloTag TMR Ligand , and CLIP-Surface 647 were imaged with diode lasers of 488 nm , 552 nm , and 638 nm wavelength , respectively . Emission light was dispersed by a prism and detected by HyD detectors . Microscopy data were processed and visualized with LAS X ( RRID: SCR_013673 , Leica Microsystems ) or Imaris ( RRID: SCR_007370 , Bitplane ) . Image data were excluded from further quantitative analyses when the data contained saturated fluorescence signals . For Figures 2 and 3 , the fluorescence intensities on somata were quantified with ImageJ at every 2 . 97 μm thickness . For Figure 7 , an image of the soma was taken and analyzed at the focal depth for each cell . For Figure 8 , color values at axons were analyzed for maximum intensity projection images , as there may be a slight chromatic aberration along the z axis . After binarization , the average color value ratio was determined for each segment of each axon . Ternary plots , box plots , and statistical analyses were prepared using MATLAB ( RRID: SCR_001622 , MathWorks ) . Image data were acquired by RS , to prevent bias when they were subsequently analyzed by MNL . First , the fluorescence intensity in each channel was normalized so that the median intensity was 1 . Then , the intensity values were further normalized so that the length of the vector was 1 . Raw quantification data ( Source Data files ) are accompanied with figures . The MATLAB code and processed data for the figure panels have been deposited to GitHub ( Leiwe , 2018; copy archived at https://github . com/elifesciences-publications/TetbowCodes ) . Axon tracing was performed with Neurolucida ( RRID: SCR_001775 , MBF Bioscience ) . In Figure 8F , we focused on axons were found within 200 μm of the anterior end of the image . We terminated tracing when labeled axons were interrupted by unlabeled gaps; in such cases , we judged that we cannot be 100% sure whether the interrupted segments are connected . Thus , our tracing was performed using very conservative criteria , and thus most probably provides an underestimate of the entire wiring diagram . Axons that can be traced for >1 , 000 μm were analyzed ( 25 axons ) . Putative mitral and tufted cells were identified on the basis of the projection area ( piriform cortex vs . olfactory tubercle ) and their axonal trajectories . Brightly-labeled putative mitral and tufted cells were further analyzed in Figure 7D , E . For Figure 8—figure supplement 2 , we focused on axons that crossed the anterior plane and within the lateral half . We chose 100 axons in an unbiased manner from the dorsal to ventral direction . The criteria for successful tracing are described in Figure 8D; as these criteria are very strict , the area that is traced may be an underestimate . Color discrimination was performed visually by one experimenter , who also joined the visual color discrimination test . Custom MATLAB code ( Leiwe , 2018; copy archived at https://github . com/elifesciences-publications/TetbowCodes ) was written to quantify the ability of experienced researchers to discriminate three color space ( RGB ) in terms of 3D Euclidean space . All of the subjects ( from the Imai lab ) have experience in fluorescence imaging , understand the purpose of the test , and have trichromatic color vision . To prevent the intensity of the color from influencing the decision , the intensity was randomly varied between 50–100% for each square displayed ( Figure 1—figure supplement 1 ) . The subject was presented with a choice to discriminate between two colors at a specified Euclidean distance . Specifically , they were asked to determine whether the two columns form a cross or a parallel line . For each Euclidean distance presented , there were 100 trials per subject , with the average success rate stored . Note that this test was not intended to evaluate the color discrimination performance of people in general . Poisson distributions were calculated in MATLAB , by creating independent distributions for each XFP for each specified copy number . 200 ‘cells’ were selected for each copy number group with associated XFP values derived from the Poisson distributions . The code has been deposited to GitHub ( Leiwe , 2018; copy archived at https://github . com/elifesciences-publications/TetbowCodes ) . The raw microscopy data have been deposited to the Systems Science of Biological Dynamics ( SSBD ) database ( http://ssbd . qbic . riken . jp/ ) with a unique URL ( http://ssbd . qbic . riken . jp/set/20180901/ ) . Movies will be posted at SeeDB Resources ( https://sites . google . com/site/seedbresources/ ) . Step-by-step protocols has been posted in our website , SeeDB Resources ( https://sites . google . com/site/seedbresources/ ) . All of our published clearing methods and technical tips are also posted on this website . A detailed protocol for SeeDB2 was published in bio-protocol ( Ke and Imai , 2018 ) . | The brain is made up of millions of cells called neurons , and it is important to learn how these neurons are wired together to better understand how the brain works . To make it easier to tell individual neurons apart in samples from brains , some scientists have developed a process called Brainbow that labels individual neurons with different fluorescent colors . Scientists have also created techniques called “tissue clearing” to make a brain transparent in the laboratory . These techniques make the brain see-through enough to allow scientists to study the wiring of the brain in three dimensions . These multicolor labeling and tissue clearing techniques are very helpful for studying the brain . But they have an important limitation; the fluorescent colors are not bright enough to allow scientists to trace the long extensions called axons and dendrites that wire neurons together . As a result , tracing axons and dendrites was difficult and required cutting the brain into hundreds of thin slices . It could take several months for scientists to trace the path of a single neuron . Brighter fluorescent labeling colors would allow scientists to use high-powered microscopes to trace the entire length of a neuron in a whole brain much more quickly and easily . Now , Sakaguchi et al . have developed a bright multicolor labeling method for neurons called Tetbow . Tetbow produces more vivid colors allowing scientists to trace the wiring of neurons over long distances in the mouse brain . Sakaguchi et al . combined Tetbow with tissue clearing techniques to dissect and trace many neurons in a whole mouse brain within a few days . Neuroscientists can now use Tetbow to speed up the study of how neurons are wired in the brain . Researchers working in other fields could also use Tetbow to help track the behavior of different cells . Tetbow allows everyone to see the beautiful wiring of the brain in three dimensions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"tools",
"and",
"resources",
"neuroscience"
] | 2018 | Bright multicolor labeling of neuronal circuits with fluorescent proteins and chemical tags |
In many cases of trauma , the same environmental stimuli that become associated with aversive events are experienced on other occasions without adverse consequence . We examined neural circuits underlying partially reinforced fear ( PRF ) , whereby mice received tone-shock pairings on half of conditioning trials . Tone-elicited freezing was lower after PRF conditioning than fully reinforced fear ( FRF ) conditioning , despite an equivalent number of tone-shock pairings . PRF preferentially activated medial prefrontal cortex ( mPFC ) and bed nucleus of the stria terminalis ( BNST ) . Chemogenetic inhibition of BNST-projecting mPFC neurons increased PRF , not FRF , freezing . Multiplexing chemogenetics with in vivo neuronal recordings showed elevated infralimbic cortex ( IL ) neuronal activity during CS onset and freezing cessation; these neural correlates were abolished by chemogenetic mPFC→BNST inhibition . These data suggest that mPFC→BNST neurons limit fear to threats with a history of partial association with an aversive stimulus , with potential implications for understanding the neural basis of trauma-related disorders .
In many cases of psychological trauma , encounters with contexts and stimuli during aversive experience ( s ) are interleaved with occasions when the same stimuli are experienced without consequence . Most standard rodent assays of fear ( i . e . , threat ) memory , however , present the subject with a conditioning stimulus ( CS ) that on each occasion is paired with an aversive unconditioned stimulus ( US ) ( Fanselow and Poulos , 2005 ) . This discrepancy is pertinent to modeling traumatic memories in rodents , via back translation from human to rodent . Theoretical accounts of associative learning predict that conditioned responses to CSs with a mixed or partial reinforcement history , which render the CS uncertain or ambiguous with regard to its expected outcome , may differ in certain respects from those that are consistently reinforced . For example , as compared to fully reinforced CSs , partially reinforced CSs can be more difficult to extinguish and produce lesser conditioned responses , due to associative strength accruing to the conditioning context or through the endowment of the CS with inhibitory ( CS = no US ) properties ( Humphreys , 1939; Fitzgerald , 1963; Rawlins et al . , 1985; Rescorla , 2007; Tsetsenis et al . , 2007; Miguez et al . , 2012; Harris et al . , 2019 ) . Fear behavior that arises from partial reinforcement could involve neural circuits distinct from the well-described circuits implicated in standard ( i . e . , fully reinforced ) fear conditioning ( Pape and Pare , 2010; Bukalo et al . , 2014; Tovote et al . , 2015 ) . Two brain regions that could be important for the acquisition and expression of partially reinforced fear ( PRF ) are the medial prefrontal cortex ( mPFC , comprising , in the rodent , the prelimbic [PL] , infralimbic [IL] , and anterior cingulate [ACC] cortices ) and the bed nucleus of the stria terminalis ( BNST ) ( Lebow and Chen , 2016; Goode et al . , 2019 ) . The mPFC is engaged in experimental situations requiring integration of higher-order cues or disambiguation between conflicting cues to gate a level of response appropriate to the value of outcome ( Sharpe and Killcross , 2018; Marek et al . , 2019 ) , while the BNST has been shown to support learning when a stimulus poorly predicts threat ( Lebow and Chen , 2016; Goode et al . , 2019 ) . These structures are also anatomically connected , with a particularly dense connection between the IL and the anterior regions of the BNST ( Hurley et al . , 1991; McDonald et al . , 1999; Dong et al . , 2001; Vertes , 2004; Radley and Sawchenko , 2011; Radley et al . , 2013; Johnson et al . , 2016; Glangetas et al . , 2017; Tillman et al . , 2018; Johnson et al . , 2019 ) . Additionally , BNST-projecting IL cells are activated by ‘unpredictable’ threat in a backward conditioning paradigm ( Goode et al . , 2019 ) . Moreover , stimulation of glutamatergic mPFC inputs produces synaptic depression in the BNST ( Glangetas et al . , 2013 ) . Together , these findings suggest the mPFC and BNST might form a functional circuit regulating fear to ambiguous and uncertain threats . Here , we sought to elucidate the potential role of the mPFC and BNST and other neural circuits in PRF , using a paradigm in which a CS was paired with a footshock US on only half of the trials ( McHugh et al . , 2015; Glover et al . , 2017 ) . By combining immediate-early gene mapping , neuronal pathway tracing , in vivo chemogenetics , and a multiplexed approach combining in vivo chemogenetics and in vivo neuronal recordings , we demonstrate that the mPFC→BNST circuit negatively gates PRF .
The PRF conditioning procedure entailed presenting male C57BL/6J ( B6 ) mice with three pairings of a tone CS and a footshock US , along with three interspersed presentations of the same CS without concomitant footshock ( McHugh et al . , 2015; Glover et al . , 2017 ) . For comparison , a fully reinforced fear ( FRF ) group received 3x CS+US pairings , and a CS-only control group received 6x CS presentations without the US ( Figure 1A , B ) . Freezing increased to a similar extent over the six conditioning trials in the PRF group and over the three conditioning trials , plus the corresponding three no-trial periods , in FRF group , but did not significantly increase in the CS-only group ( analysis of variance [ANOVA] group-effect: F ( 2 , 17 ) =5 . 74 , p=0 . 0125; trial-effect: F ( 5 , 85 ) =13 . 49 , p<0 . 0001; interaction: F ( 5 , 85 ) =1 . 64 , p=0 . 1099 ) . On a retrieval test conducted in a novel context ( context B ) the following day , the PRF and FRF groups froze more than CS-only controls during pre-CS baseline and CS presentation . Notably , however , CS-evoked freezing was lower in the PRF , relative to the FRF , group ( ANOVA group-effect: F ( 2 , 17 ) =53 . 02 , p<0 . 0001; CS-effect: F ( 1 , 17 ) =216 . 90 , p=0 . 0001; interaction: F ( 2 , 17 ) =25 . 51 , p=0 . 0001 , followed by post-hoc tests: CS-only vs PRF p<0 . 0001 , CS-only vs FRF p<0 . 0001 , PRF vs FRF p=0 . 0008 ) ( Figure 1C , Figure 1—figure supplement 1 ) . These data show that B6 mice express less freezing in the PRF , as compared to FRF , procedure despite the number of CS–US pairings being equivalent in both conditions . These differences are in line with lower freezing in the PRF procedure in a mixed C57BL/6J;CBA/J;129S6/SvEvTac genetic background ( Tsetsenis et al . , 2007 ) but , indicating a degree of strain dependency of PRF , differ from data in outbred CD-1 mice , in which freezing is equivalent between PRF and FRF groups ( Glover et al . , 2017 ) . We next reasoned that an inbred strain ( S1 ) , which exhibits impaired contextual ( and cued ) fear discrimination , deficits in limiting fear following extinction and conditioned inhibition , and high fear expression in a different assay for PRF ( Camp et al . , 2012 ) , might exhibit deficits in the current PRF assay ( Camp et al . , 2009; Camp et al . , 2012; Figure 1D ) . Across conditioning trials , there was increased freezing in the PRF and FRF groups , irrespective of strain ( ANOVA group-effect: F ( 1 , 25 ) =0 . 15 . p=0 . 7023 , trial-effect: F ( 5 , 125 ) =38 . 02 , p<0 . 0001; strain-effect: F ( 1 , 25 ) =7 . 68 , p=0 . 0103; three-way interaction: F ( 5 , 125 ) =0 . 66 , p=0 . 6539 ) . On retrieval in context B , PRF B6 mice showed less CS-related freezing than their FRF counterparts , whereas freezing was equivalent in PRF and FRF S1 mice ( ANOVA strain-effect: F ( 1 , 27 ) =8 . 72 , p=0 . 0065; conditioning-type effect: F ( 1 , 27 ) =6 . 15 , p=0 . 0197; CS-effect: F ( 1 , 27 ) = 524 . 00 , p<0 . 0001; three-way interaction: F ( 1 , 27 ) =7 . 03 , p=0 . 0132 , followed by post-hoc tests: PRF vs FRF in B6 p=0 . 0038 , PRF vs FRF in S1 p=0 . 2322 ) ( Figure 1E , Figure 1—figure supplement 1 ) . The finding that S1 mice exhibit similar freezing to the PRF and FRF procedures aligns with the excessive fear shown by this strain to innocuous stimuli and following extinction ( Camp et al . , 2009; Camp et al . , 2012 ) and further illustrates the strain dependency of PRF . An earlier study by Glover et al . , 2017 found that following PRF conditioning , CD-1 mice had a higher latency to feed , as compared to a FRF group , in the novelty-suppressed feeding ( NSF ) test , an assay sensitive to anxiolytics and antidepressants ( Ramaker and Dulawa , 2017 ) . To test whether PRF had a similar effect in B6 mice , NSF was assessed under either high or low illumination levels the day after B6 mice underwent either PRF or FRF . Under high , but not low , illumination , latencies to feed were higher in the PRF and FRF groups than unconditioned controls ( ANOVA group-effect: F ( 1 , 44 ) =10 . 93 , p=0 . 0019; illumination-effect: F ( 1 , 44 ) =4 . 11 , p=0 . 0230; interaction: F ( 1 , 44 ) =2 . 83 , p=0 . 0699 , followed by post hoc tests: PRF vs Con p=0 . 0004 , FRF vs Con p=0 . 0222 ) ( Figure 1—figure supplement 2 ) . These data show that PRF conditioning increases anxiodepressive-like anxiety-like behavior under relatively aversive ( high illumination ) conditions of approach–avoidance conflict . The complex behavioral sequelae of PRF suggest that this form of fear may have different neural substrates than FRF . We , therefore , sought to identify neural correlates of PRF by quantifying the number of c-Fos+ cells , as a proxy for neuronal activity , in forebrain regions following retrieval ( for corresponding behavioral data , see Figure 1B ) . There were a higher number of c-Fos+ cells in the basolateral amygdala ( BLA ) of FRF mice ( ANOVA group-effect: F ( 2 , 17 ) =6 . 79 , p=0 . 0068 , followed by post hoc tests: FRF vs CS-only p=0 . 0038 , FRF vs PRF p=0 . 0132 ) , as compared to either PRF mice or a set of controls that had received CS-only trials during conditioning . In the paraventricular nucleus of the thalamus ( PVT ) , another region implicated in fear ( Penzo et al . , 2015 ) , c-Fos+ counts were higher in the PRF and FRF groups than controls ( ANOVA group-effect: F ( 2 , 17 ) =4 . 01 , p=0 . 0374 , followed by post-hoc tests: CS-only vs PRF p=0 . 0281 , CS-only vs FRF p=0 . 0145 ) . No group differences were evident in the lateral or medial habenula , or ventral or dorsal hippocampus ( Figure 2A–I , Figure 2—figure supplement 1 ) . In subregions of the mPFC , however , there were more c-Fos+ cells in the IL ( F ( 2 , 17 ) =8 . 21 , p=0 . 0032 , followed by post-hoc tests: CS-only vs PRF p=0 . 0009 , CS-only vs FRF p=0 . 0411 , FRF vs PRF p=0 . 0420 ) , but not the posterior ACC ( F ( 2 , 17 ) =1 . 01 , p=0 . 3862 ) of PRF and FRF mice , relative to CS-only controls . Counts in the PL were higher in PRF mice relative to controls and trended higher in the FRF group ( F ( 2 , 17 ) =3 . 60 , p=0 . 0499 , followed by post hoc tests: CS-only vs PRF p=0 . 0196 ) . The same pattern of elevated activity in the PRF group , relative to the other groups , was also evident in the BNST , though specifically in the anteroventral BNST ( avBNST ) ( F ( 2 , 17 ) =19 . 43 , p=0 . 0001 , followed by post hoc tests: CS-only vs PRF p=0 . 0001 , CS-only vs FRF p=0 . 0294 , PRF vs FRF p=0 . 0005 ) , not the anterodorsal BNST ( adBNST ) ( F ( 2 , 14 ) =1 . 38 , p=0 . 2831 ) ( Figure 2A–I ) . These findings show that retrieval of a PRF CS , despite being characterized by lower freezing than FRF , associates with a unique pattern of regional brain activation , with preferentially high activation in the IL and PL subregions of the mPFC and the avBNST . Previous studies in the rat have demonstrated a direct ( GABAergic ) input from the mPFC to the BNST that is particularly dense between the IL and avBNST ( Dong et al . , 2001 ) , but also present between the PL and avBNST ( Johnson et al . , 2016; Johnson et al . , 2019 ) . As our c-Fos data indicated activation of the IL , PL , and avBNST by PRF , we sought to verify an mPFC-to-BNST projection in mice . In a combinatorial viral tracing approach to label postsynaptic targets of mPFC neurons in the BNST ( Zingg et al . , 2017; Sengupta and Holmes , 2019 ) , a construct containing a Cre-containing anterograde trans-synaptic virus was infused into the mPFC and a Cre-dependent , synaptophysin-containing , mCherry-fused construct infused into the BNST ( Figure 2—figure supplement 2 ) . Indicative of monosynaptic input from the mPFC , mCherry labeling was apparent in BNST neurons , mainly in the ventral areas below the anterior commissure . In the rat , PL neurons form close appositions with GABAergic cells in the avBNST that in turn send efferents to the paraventricular nucleus of the hypothalamus ( PVN ) , a key mediator of responses to stress and defensive behaviors ( Johnson et al . , 2016; Johnson et al . , 2019 ) . Indicating that a similar connection is likely present in mice , inspection of our tissue revealed mCherry/synaptophysin expression originating from mPFC-innervated BNST neurons in the PVN , as well as lateral hypothalamus . A corollary to the existence of a disynaptic mPFC–BNST–PVN circuit in mice is whether the PVN in turn targets other fear-mediating regions in this species . To gain initial insight into this question , we infused a Cre-dependent , YFP-fused construct containing either channelrohodpsin2 ( ChR2 ) or synaptophysin into the PVN of oxytocin-Cre mice , to label a major population of ( oxytocin-positive ) PVN cells . This indicated labeling in the ventrolateral periaqueductal gray ( vl/PAG ) ( Figure 2—figure supplement 2 ) , a region known to regulate defensive behaviors including freezing ( Tovote et al . , 2015 ) . Together these data provide evidence of input from the mPFC to the BNST in the mouse , as well as onward connections from the BNST to the PVN and in turn possibly on to the vl/PAG . Thus , PRF engagement of the mPFC and BNST can be viewed in the context of a direct connection between these regions and their downstream access to a broader fear-regulating neural circuitry . To causally interrogate the contribution of the mPFC→BNST pathway to PRF , a retrogradely transported Cre-containing construct viral construct was infused into the BNST and a construct containing a Cre-dependent form of hM4Di ( or mCherry control ) infused into the mPFC , enabling the expression of the inhibitory DREADD in mPFC→BNST neurons to inhibit their activity , via systemic injection of clozapine N-oxide ( CNO ) , during retrieval ( Figure 3A , B ) . During conditioning , freezing increased over trials to a similar extent in all groups ( ANOVA trial-effect: F ( 5 , 145 ) =23 . 54 , p<0 . 0001; group effect: F ( 3 , 145 ) =0 . 91 , p=0 . 4467; interaction: F ( 15 , 145 ) =0 . 61 , p=0 . 08647 ) ( Figure 3—figure supplement 1 ) . Following CNO administration , CS-related freezing during retrieval was lower in PRF mice than in FRF mice expressing the control virus , replicating our earlier data . By contrast , there was no difference in freezing in mice expressing hM4Di ( ANOVA conditioning-type effect: F ( 1 , 29 ) =9 . 35 , p=0 . 0048; virus-group effect: F ( 1 , 29 ) =12 . 15 , p=0 . 0016; CS: F ( 1 , 29 ) =1331 . 02 , p<0 . 0001; three-way interaction: F ( 1 , 29 ) =6 . 58 , p=0 . 0157 , followed by post-hoc tests: mCherry PRF vs mCherry FRF p<0 . 0001 , hM4Di PRF vs hM4Di FRF p=0 . 1425 , mCherry PRF vs hM4Di PRF p=0 . 0013 , mCherry PRF vs hM4Di PRF p=0 . 7951 ) ( Figure 3C ) . Examination of the trial-by-trial freezing during retrieval indicated no significant trial-related differences in freezing , despite a trend for decreasing freezing across trials in the mCherry PRF group ( ANOVA trial-effect: F ( 5 , 145 ) =1 . 83 , p=0 . 1098; group-effect: F ( 3 , 29 ) =14 . 15 , p<0 . 0001; trial x group interaction: F ( 15 , 145 ) =1 . 04 , p=0 . 4213 ) ( Figure 3—figure supplement 1 ) . These data show that inhibition of mPFC→BNST neurons increases freezing to a PRF CS . This finding suggests that engagement of these mPFC→BNST neurons limits the expression ofto the unreliable , PRF , though it remains possible that inhibition of these neurons also produces an increase in PRF expression , which may have been masked due to high ( ceiling ) levels of freezing . The finding that inhibiting mPFC outputs to the BNST pathway increases freezing to a PRF CS implies that mPFC neurons likely encode some aspects of fear . To address this possibility , we devised an approach entailing chemogenetic inhibition of mPFC→BNST neurons ( as described above ) coupled with in vivo recordings of mPFC single-unit activity via chronically implanted electrode arrays , which we targeted at the IL ( Figure 3—figure supplement 2 ) . The average firing rate of units did not statistically differ between groups ( FRF mCherry: 4 . 10 ± 0 . 64 , FRF hM4Di: 3 . 45 ± 0 . 49 , FRF mCherry: 2 . 67 ± 0 . 66 , FRF hM4Di: 1 . 67 ± 0 . 34 ) . Aligning the single-unit data to the presentation of the CSs during retrieval revealed examples of IL units with activity time-locked to the onset of the CS ( Figure 3—figure supplement 3 ) . Units exhibiting activity >/<1 . 96 z scores from baseline in at least two 100 ms time bins within the 500 ms of CS onset were classified as CS responsive ( CS-ON ) . Overall , CS-ON units showed a significant change in neuronal activity in response to the CS ( baseline: 0 . 15 ± 0 . 35 , post-CS: 1 . 43 ± 0 . 55 , paired t-test: t ( 10 ) =6 . 51 , p<0 . 0001 ) ( Figure 3—figure supplement 3 , and for heat maps , see Figure 3—figure supplement 4 ) . Peak responses occurred within 200–300 s of CS onset and were highest in the mCherry FRF group ( Figure 3—figure supplement 3 ) . However , when the percentage of CS-ON units was calculated and compared across the conditioning and virus groups , this revealed a higher proportion of CS-ON units in the mCherry groups than in hM4Di groups for PRF mice ( Fisher’s exact test: p=0 . 0122 ) , but no differences between virus groups in the FRF mice ( Fisher’s exact test: p=0 . 6090 ) , and no difference between PRF and FRF groups , irrespective of virus group ( Fisher’s exact test in mCherry: p=0 . 2510; in hM4Di: p=1 . 000 ) ( Figure 3—figure supplement 3 ) . To examine whether IL cells were also associated with the behavior of mice during testing , their activity was aligned to episodes of freezing and those cells displaying a reliable change relative to either the onset or cessation of freezing ( i . e . , resumption of movement; >/<1 . 96 z from baseline in at least two 100 ms time bins within the 500 ms of the event ) . These units , classified as Freeze-ON and Freeze-OFF , respectively , showed a significant change in baseline-normalized activity ( Freeze-ON baseline: −0 . 81 ± 0 . 47 , post-event: −2 . 12 ± 0 . 52 , paired t-test: t ( 10 ) =4 . 60 , p=0 . 0010 , Freeze-OFF baseline: 0 . 73 ± 0 . 43 , post-event: 1 . 67 ± 0 . 40 , paired t-test: t ( 12 ) =8 . 54 , p<0 . 0001 ) ( Figure 3—figure supplement 3 , and for heat maps , see Figure 3—figure supplement 4 ) . Freeze-ON units displayed a decreased firing rate at freezing onset , which was most evident in both of the mCherry groups , while Freeze-OFF units increased firing rate at the cessation of freezing in both groups ( Figure 3—figure supplement 3 ) . When the percentage of these cell types were compared across groups , there was a higher percentage of Freeze-OFF units in the mCherry PRF group than in the hM4Di PRF group ( Fisher’s exact test: p=0 . 0024 ) , whereas there was no group difference in FRF mice ( Fisher’s exact test: p=1 . 000 ) and no difference between PRF and FRF groups in either the mCherry ( Fisher’s exact test: p=0 . 4600 ) or hM4Di ( Fisher’s exact test: p=0 . 0590 ) virus conditions ( Figure 3—figure supplement 3 ) . Conversely , there was no difference between the mCherry and hM4Di groups in the proportion of Freeze-ON units , irrespective of whether mice had undergone PRF or FRF .
Here , we sought to provide new insight into the neural substrates regulating the fear response to an uncertain/ambiguous threat . Employing an assay of partial tone+shock reinforcement in B6 mice , we found that PRF conditioning produced a lower fear response than FRF , which was associated with preferential neuronal activation in the mPFC and BNST . We also show that the mPFC and BNST formed a monosynaptic circuit that , when chemogenetically inhibited , caused a selective increase in the expression of PRF and an attendant loss of in vivo correlates of both CS onset and freezing cessation in IL units . The current findings align with and extend prior work implicating the mPFC and BNST in various situations in which there is ambiguity and uncertainty about a threat . For example , the mPFC is engaged in settings that require integration of higher-order cues to gate learned responses ( Halladay and Blair , 2015; Halladay and Blair , 2017; Sharpe and Killcross , 2018; Marek et al . , 2019 ) , or where there is conflict between excitatory and inhibitory CS associations , for instance in fear extinction ( Milad and Quirk , 2012; Bloodgood et al . , 2018; Lay et al . , 2020 ) , fear discrimination ( Grosso et al . , 2018 ) , threat/safety conditioning ( Sangha et al . , 2014; Meyer et al . , 2019 ) , and punished reward-seeking ( Burgos-Robles et al . , 2017; Halladay et al . , 2020 ) . The BNST , meanwhile , supports learning in the absence of the BLA ( Poulos et al . , 2010; Zimmerman and Maren , 2011 ) , and when a stimulus poorly predicts threat ( Lebow and Chen , 2016; Goode et al . , 2019; Bjorni et al . , 2020 ) , either because it is distal ( e . g . , predator odor; Fendt et al . , 2003; Xu et al . , 2012; Breitfeld et al . , 2015; Verma et al . , 2018; Goode et al . , 2020 ) , diffuse ( e . g . , contextual; Sullivan et al . , 2004; Kalin et al . , 2005; Duvarci et al . , 2009; Davis et al . , 2010; Luyten et al . , 2012; Jennings et al . , 2013; c . f . Haufler et al . , 2013 ) , or temporally ambiguous ( e . g . , random or sustained; Waddell et al . , 2006; Walker et al . , 2009; Hammack et al . , 2015; Daldrup et al . , 2016; Goode and Maren , 2017; Lange et al . , 2017 ) with respect to the US . These known functions of the mPFC and BNST make these structures well placed to mediate fear under conditions of partial reinforcement where the CS is experienced both with and without the US . As we show here , the mPFC and BNST form a discrete neural circuit , through a direct anatomical connection , that serves to limit the expression of partially reinforced fear . This observation is reminiscent of a recent study showing that IL neurons projecting to the avBNST are activated ( measured by c-Fos ) by a measure of unpredictable threat ( backward conditioning ) , in which US presentation precedes the CS ( Goode and Maren , 2017 ) . In conjunction with the current data , there is convergent evidence supporting a key role for the mPFC→BNST circuit in mediating fear across various measures of threat uncertainty , ambiguity , and unpredictability . The precise nature of this role remains to be fully clarified , however . One possibility is that when discrete cues are relatively poor predictors of danger , other environmental stimuli , such as context , modulate the expression of fear in a manner that recruits the mPFC to exert top-down control over the BNST . In support of this possibility , the mPFC is posited to subserve higher-order modulation of conditioned responding ( Sharpe and Killcross , 2018 ) ; indeed , a recent study in rats found that lesions of either the PL or IL impaired one measure of such modulation known as occasion setting ( Roughley and Killcross , 2019 ) . Another possible explanation for the increase in PRF caused by mPFC→BNST inhibition is that CS-alone presentations during conditioning imbues the CS with inhibitory properties that are gated by mPFC→BNST neurons during retrieval . Learned inhibition is a function attributed to the mPFC and in particular the IL . For example , pharmacological , optogenetic , and chemogenetic inhibition of the IL impairs the formation and/or retrieval of fear extinction memories ( Laurent and Westbrook , 2009; Bukalo et al . , 2015; Do-Monte et al . , 2015; Kim et al . , 2016; Lay et al . , 2020; Bukalo et al . , 2021 ) and the expression of learned safety acquired through explicit CS–US unpairing ( Sangha et al . , 2014 ) . Conversely , presentation of a safety signal during inescapable stress decreases activity ( c-Fos ) in a lateral area of BNST encompassing avBNST ( Christianson et al . , 2011 ) . Furthermore , lesioning this area reduces inappropriate fear to a non-reinforced CS in rats with high-trait anxiety ( Duvarci et al . , 2009 ) . Indeed , a growing number of lesion and functional neuroimaging studies in non-human primates and humans have implicated the BNST in the processing of uncertain threat ( see Goode et al . , 2019; Miles and Maren , 2019 ) . Together , these findings suggest that inhibitory properties of the partially reinforced CS could be signaled by the IL downstream to the BNST , thereby limiting the expression of CS-induced fear to a level appropriate to its partial reinforcement history . It is important to note in this regard that while we targeted our virus infusions to the IL and avBNST – based on prior evidence of a dense anatomical connection between these subregions – the small size and ventral location of these areas meant that viral transfection encompassed parts of the PL and adBNST . The adBNST is engaged by challenges that produce negative affect ( Centanni et al . , 2019 ) , undergoes plastic changes in response to chronic stress ( Conrad et al . , 2011 ) , and , of particular relevance here , is a target of dorsal raphe ( Marcinkiewcz et al . , 2016 ) , BLA ( Lange et al . , 2017 ) , and central amygdala ( Asok et al . , 2018 ) neurons that sustain fear responses to predictable and unpredictable threats . As such , although the adBNST is a less densely innervated by the mPFC ( Hurley et al . , 1991; McDonald et al . , 1999; Dong et al . , 2001; Vertes , 2004; Radley and Sawchenko , 2011; Radley et al . , 2013; Johnson et al . , 2016; Glangetas et al . , 2017; Tillman et al . , 2018; Johnson et al . , 2019 ) , and current as well as prior ( Goode et al . , 2019 ) c-Fos data do not indicate adBNST activation with uncertain threat , it would be premature to exclude a contribution of this area to PRF . The possible contribution of BNST-targeting PL neurons to PRF also should not be discounted . Precisely dissociating the roles of PL and IL inputs to BNST in PRF will be an interesting avenue for future work . While the PL has been ascribed a role in promoting FRF via its outputs to the BLA ( Pape and Pare , 2010; Dias et al . , 2013; Bukalo et al . , 2014; Tovote et al . , 2015 ) , recent work found that optogenetically silencing PL inputs to the avBNST increased immobility and associated stress hormone responses in the rat shock-probe burying and tail suspension tests ( Johnson et al . , 2019; see also Radley et al . , 2009 ) . These effects suggest a role for PL inputs in attenuating negative affect and as such are broadly congruent and potentially explanatory of the current data , despite important differences in methodology . In the current study , we found neuronal correlates of PRF , specifically within the IL . As in our prior studies of FRF in mice ( Fitzgerald et al . , 2014; Fitzgerald et al . , 2015 ) , a subset of IL neurons were phasically active to CS presentation during fear retrieval . Intriguingly , we also found a subset of IL neurons that displayed phasic activity during the cessation , but not onset , of freezing during fear retrieval , echoing recordings in rats that uncovered movement-related activity in IL units ( Halladay and Blair , 2015; Halladay and Blair , 2017 ) . Inhibition of BNST-projecting mPFC neurons during fear retrieval essentially abolished the CS- and Freeze-OFF-associated neuronal activity in IL neurons and , in parallel , increased freezing in the PRF group . The ability of mPFC→BNST inhibition to ablate these neuronal correlates could have arisen from the chemogenetic inhibition of BNST-projecting IL neurons , resulting in a loss-of-function in this pathway and a selective increase in PRF mice . Two caveats to this interpretation are that , firstly , inhibition-induced increases in freezing in FRF mice may have been masked by a performance ‘ceiling’ in the FRF control group and , secondly , hM4Di expression in our recording experiment was not restricted to BNST-projections within the IL , and also encompassed neurons in the PL . With regard to the broader neural circuitry in which the mPFC→BNST circuit operates to mediate PRF , using trans-synaptic tracing , we found evidence of a disynaptic mPFC→BNST→PVN circuit in mice , as has been reported in rats ( Radley and Sawchenko , 2011; Johnson et al . , 2016; Johnson et al . , 2019 ) . The PVN contains a high density of cells expressing peptide hormones implicated in stress and fear , notably corticotrophin-releasing hormone ( CRH ) and oxytocin ( Herman and Tasker , 2016; Triana-Del Río et al . , 2019 ) . Though CRH-expressing neurons in the PVN receive input from the BNST ( Colmers and Bains , 2018 ) , it is unclear whether oxytocin-producing cells do so . Interesting nonetheless , we found that mouse PVN oxytocinergic neurons strongly innervate the freezing-regulating vl/PAG , replicating earlier work in cats and naked mole rats ( Holstege , 1987; Rosen et al . , 2008 ) . Given the vl/PAG also receives input from a population of avBNST neurons that is , in turn , innervated by the PL ( Johnson et al . , 2016; Johnson et al . , 2019 ) , these tracing results suggest that in addition to directly innervating the PVN , mPFC→avBNST neurons may also have both direct and indirect ( via the PVN ) access to the vl/PAG . This positions the circuit to modulate multiple behavioral and neuroendocrine responses to PRF . In summary , the current study found that B6 mice expressed lower fear to a CS that is partially , rather than fully , reinforced with footshock . Lower PRF expression was not apparent in a mouse strain ( S1 ) deficient in fear discrimination and learned inhibition . Furthermore , c-Fos mapping revealed PRF preferentially recruited the mPFC and BNST , and neuronal tracing showed direct neuronal projections from mPFC to the BNST , with downstream connections to stress- and fear-mediating regions . Demonstrating the causal importance of the mPFC→BNST neurons , inhibiting this pathway increased PRF and abolished neuronal correlates of CS presentation and freezing cessation in the IL . Collectively , these findings provide novel insight into the neural substrates of PRF , with potential translational relevance to anxiety and trauma- and stressor-related disorders in which threats are typically ambiguous and unpredictable .
Subjects were adult male C57BL/6J ( B6 ) , 129S1/SvImJ ( S1 ) , and B6;129S-Oxttm1 . 1 ( cre ) Dolsn/J ( JAX strain 024234 ) ( Oxt-Cre ) mice obtained from the Jackson Laboratory ( Bar Harbor , ME , USA ) and were at least 8 weeks old at the time of testing . Mice were group-housed in a temperature ( 22 ± 3°C ) and humidity ( 45 ± 15% ) controlled vivarium under a 12 hr light/dark cycle ( lights on 0600 hr ) . Mice undergoing surgery for chronic implantation were single housed after surgery to prevent the implant being damaged by a cage mate . All experimental procedures were approved by the National Institute on Alcohol Abuse and Alcoholism ( NIAAA ) and Santa Clara University Animal Care and Use Committees ( SCU AWA: D18-01042 ) and followed the NIH guidelines outlined in ‘Using Animals in Intramural Research’ and the local Animal Care and Use Committees . The threat conditioning procedures were based on previous studies with slight modifications ( McHugh et al . , 2015; Glover et al . , 2017 ) . For this and all other experiments , prior to testing , mice were randomly assigned to experimental groups and habituated to handling for approximately 10 min per day for 4 days . The following procedures were used in all experiments , unless stated otherwise below . Conditioning was conducted in a 27 × 27 × 11 cm chamber with opaque metallic walls and a metal rod floor ( context A ) . The walls of the chamber were cleaned with a 79% water/20% ethanol/1% vanilla-extract solution to provide a distinctive odor – this was repeated after each session . All conditioning procedures began with a 180 s baseline period . Conditioning for FRF entailed three presentations ( 60–90 s variable inter-CS interval ) of a 30 s , 75 dB ( 50 ms rise time ) , white noise ( CS ) that co-terminated with a 2 s , 0 . 6 mA scrambled footshock ( US ) . After the final pairing for all groups , there was a 120 s no-stimulus period before the mouse was returned to the home-cage . The procedure was the same for PRF , with the exception that the CS was presented , without the US , on an additional three occasions during the intervals between the CS+US pairings ( order: CS+US; CS+US; CS-noUS; CS-noUS; CS+US; CS-noUS , 15–60 s inter-stimulus interval ) . Where a CS-only control group was included ( stated below ) , the CS was presented on six occasions , corresponding to the order and timing of the PRF group , but without any concomitant US . CS retrieval took place one day after conditioning in a novel context B , a 27 × 27 × 11 cm chamber with white Plexiglas walls ( rear wall curved ) and a solid white floor , which was housed in a different room to context A . Between each session , all surfaces of the chamber were cleaned with a 99% water/1% acetic acid solution . After a 180 s baseline period , there were six CS presentations ( 20–60 s inter-pairing interval ) . There was a 20 s no-stimulus period before the mouse was returned to the home-cage . All groups were tested in the same manner . Stimulus presentation was controlled by the Med Associates VideoFreeze system ( Med Associates , Burlington , VT , USA ) . Freezing , scored manually every 5 s ( as no visible movement except that required for breathing ) , was measured as an index of fear ( Blanchard and Blanchard , 1972 ) and converted to a percentage ( [number of freezing observations/total number of observations] x 100 ) . Mice ( assigned to FRF , PRF , and a control group exposed to the conditioning context for 1 min ) underwent conditioning and were then food-deprived for 24 hr . Subsequently , they were assessed using the NSF test for anxiodepressive-like behavior , as previously described ( Glover et al . , 2017 ) . The test apparatus was a novel , 50 cm3 white Plexiglas box with the floor covered by fresh cage substrate . A single pellet of regular home-cage food chow was placed within a plastic weigh-boat in the center of the box . Separate groups of mice underwent the test under 180 lux ( low ) and 1350 lux ( high ) illumination . The mouse was placed in a corner of the box , facing the center , and the latency to begin eating the chow was measured from a video-recording . The test ended when eating started or when 600 s had elapsed . B6 and S1 mice underwent conditioning and retrieval as described under ‘Partially versus fully reinforced threat ( standard procedure ) ’ . Mice ( assigned into FRF , PRF , and a CS-only group ) underwent conditioning and retrieval as described under ‘Partially versus fully reinforced threat ( standard procedure ) ’ . Ninety minutes after retrieval , mice were deeply anesthetized with sodium pentobarbital and transcardially perfused with ice-cold phosphate-buffered saline ( PBS , pH 7 . 4 ) followed by ice-cold 4% paraformaldehyde ( PFA ) . Brains were removed , and 50 µm coronal sections were cut on a vibratome ( Leica VT1000 S , Leica Biosystems Inc , Buffalo Grove , IL , USA ) and stored free floating in 0 . 1 M phosphate buffer ( PB ) at 4°C for <1 week . Sections were incubated successively with 10% normal goat serum and 1% bovine serum albumin in PBS-TritonX ( 0 . 3% ) for 2 hr , a mixture of rabbit anti-c-Fos ( 9F6 ) ( cat# 2250S , 1:1000 , Cell Signaling Technology , Danvers , MA , USA ) and a mouse monoclonal anti-NeuN antibody ( MAB377 , Millipore , 1:1000 ) in a dilution of 1% normal goat serum and 0 . 1% bovine serum albumin in PBS-TritonX ( 0 . 3% ) for two nights on a platform rocker at 4°C . Sections were then rinsed 3× for 10 min in PBS and incubated in anti-rabbit Alexa 488 secondary antibody ( cat# A-11034 , 1:500 , Invitrogen , Eugene , OR , USA ) and Alexa Fluor 555 anti-mouse antibody ( cat# A-21422 , 1:500 , Invitrogen ) in a dilution of 1% normal goat serum and 0 . 1% bovine serum albumin in PBS-TritonX ( 0 . 3% ) at room temperature on a platform rocker for 2 hr . Sections were rinsed in PBS 2× for 10 min and then counterstained with Hoechst 33342 ( 5 µg/mL , cat# H1399 , Thermo Fisher Scientific , Waltham , MA , USA ) in PBS . Sections were rinsed 3× for 10 min in PBS before each series . After rinsed once in 0 . 1 M PB for 10 min , serial sections were mounted onto slides , air-dried , coverslipped with aqueous mounting media ( 10 mM Tris–HCl [pH 8 . 0] [5 mL] , DABCO [cat# D27802-25G , Sigma–Aldrich] [1 . 42 g] , and glycerol [cat# 5516 , Sigma–Aldrich] [50 mL] ) , then sealed with clear nail polish . Images of all three channels ( c-Fos , NeuN , Hoechst ) for all sections were acquired using an Olympus VS120 Virtual Slide Microscope system ( Olympus , Center Valley , PA , USA , VS_ASW software ) with a 20× objective ( U Plan S Apo; 20× , NA 0 . 75 ) . The NeuN channel , in the autofocus mode , was used as a focus reference , in the autofocus mode . For image analysis , the FIJI ( https://imagej . net/Fiji ) ( Schindelin et al . , 2012 ) with VSI reader plugin ( BIOP , Zurich , Switzerland , https://c4science . ch/w/bioimaging_and_optics_platform_biop/image-processing/imagej_tools/ijab-biop_vsireader/ ) was used . A contour of each brain area ( region of interest , ROI ) was manually drawn on the Hoechst channel with reference to a mouse brain atlas ( Paxinos and Franklin , 2001 ) on the thumbnail image that covers the whole coronal sections and the full resolution image of the ROI was extracted for all channels . Counts were made in the following brain regions: PL , IL , ventromedial BNST , dorsolateral BNST , BLA , lateral and medial habenula , and ventral and dorsal hippocampus ( for cartoons depicting region definitions and example images , see Figure 2 , Figure 2—figure supplement 1 ) . For each brain region , cell counts were conducted ( blind to test group ) in two to four sections from each hemisphere , for a total of six data points per region per mouse . It was unnecessary to correct for double counting because sections were non-consecutive . The ROIs were transferred to the c-Fos channel , and the mean number of c-Fos positive cells per 0 . 25 mm2 within the ROI was quantified in a semi-automated manner using a custom-written macro . Mice were placed in a stereotaxic alignment system ( Kopf Instruments , Tujunga , CA , USA ) and kept under isoflurane anesthesia . AAV1-hSyn Cre-WPRE-hGH ( titer: 3 . 5 × 1013 GC/mL , plasmid# 55637 , obtained from Addgene , Cambridge , MA , USA , and packaged by Vigene Biosciences , Rockville , MD , USA ) was unliterally infused ( 0 . 15 µL ) into the PFC , and AAV5-Ef1a-DIO-eYFP ( titer: 2 . 1 × 1012 , obtained from the UNC Vector Core ) was unilaterally infused ( 0 . 15 µL ) into the BNST of the same hemisphere . The coordinates for BNST infusions were Medial/Lateral = ±0 . 80 , Dorsal/Ventral = −4 . 15 , Anterior/Posterior = +0 . 03 . The coordinates for mPFC infusions were Medial/Lateral = ±1 . 30 ( 20° angle ) , Dorsal/Ventral = −3 . 00 , Anterior/Posterior = +2 . 00 . Four weeks later , mice were terminally anesthetized with sodium pentobarbital ( 50–60 mg/kg ) . Brains were removed and initially suspended in 4% PFA overnight and then at 4°C in 0 . 1 M PB for 1–2 days . The general histological procedures were also the same as described under ‘Regional patterns of fear-related c-Fos activity’ , with the exception that sections were successively immunostained with rabbit anti-DsRed ( 1:200 dilution , cat# 632496 , Takara Bio , Mountain View , CA , USA ) and Alexa Fluor 555 Goat Anti-Rabbit ( 1:500 dilution , cat# A-21428 , Thermo Fisher Scientific ) , PBS ( 9 mL ) , 10% Triton X-100 ( 0 . 3% final ) ( 300 µL ) , and blocking buffer ( as above; 1 mL ) , and incubated on a platform rocker for 2 hr ( 20°C ) . The sections were mounted onto slides and allowed to dry . Far-Red nuclear staining dye ( four to five drops of NucRed Dead 647 ReadyProbe Reagent , Thermo Fisher Scientific , and 0 . 1 M phosphate buffer [2 . 5 mL] ) was pipetted onto the mounted sections . After 15–20 min , the excess solution was suctioned using a benchtop aspirator . Once sufficiently dried , the slides were coverslipped using the same aqueous mounting media . Fluorescent images were taken with a Zeiss ( LSM 700 , Carl Zeiss Microscopy , Thornwood , NY , USA ) confocal microscope under a Plan-Apochromat 10x/0 . 8 M27 objective . Oxt-Cre mice were placed in a stereotaxic alignment system ( Kopf Instruments ) under isoflurane anesthesia . Either a viral vector-containing ChR2 , fused to GFP ( AAV2-EF1a-DIO-hChR2 ( E123T/T159C ) -EYFP , titer: 6 . 10 × 1012 vp/mL , Addgene plasmid#35509 , obtained from the UNC Vector Core , Chapel Hill , NC , USA ) or a vector-containing synaptophysin , fused to GFP ( AAV8 . 2-hEF1a-DIO-synaptophysin-EYFP , titer: 2 . 1 × 1013 vg/mL , generously provided by Dr . R . Neve , Massachusetts General Hospital , Belmont , MA , USA ) was bilaterally infused ( 0 . 15 µL ) into the PVN ( coordinates ML = ±1 . 50 mm ( 15° angle ) , DV = −4 . 87 mm , and AP = +0 . 75 mm , relative to bregma ) . Five weeks later , mice were deeply anesthetized with sodium pentobarbital and transcardially perfused with PBS followed by 4% PFA . Coronal ( 50 µm thick ) sections were prepared by vibratome ( VT1000S; Leica ) . The general histological procedures were also the same as described under ‘Regional patterns of fear-related c-Fos activity’ , with the exception that sections were successively immunostained with chicken anti-GFP ( 1:5000 dilution , cat# ab13970 , Abcam , Cambridge , UK ) and anti-chicken Alexa 488 secondary antibody ( 1:500 dilution , cat# ab150169 , Abcam ) . Images were taken with a fluorescence microscopy ( VS120; U Plan S Apo; 20× , NA 0 . 75; Olympus ) . Mice were placed in a stereotaxic alignment system ( Kopf Instruments ) and kept under isoflurane anesthesia . rAAV2-retro-Ef1a-Cre ( titer: 1 . 0 × 1013 gc/mL , obtained from the Salk Institute , La Jolla , CA , USA ) was bilaterally infused targeting the BNST ( 0 . 15 µL/hemisphere ) . Additionally , either AAV8-hSyn-DIO-hM4D ( Gi ) -mCherry-WPRE ( titer: 2 . 25 × 1013 gc/mL , obtained from the Massachusetts General Hospital Gene Delivery Technology Core , Cambridge , MA , USA ) or AAV8 . 2-hEF1-DIO-mCherry-WPRE ( titer: 2 . 13 × 1013 vg/mL , obtained from the Massachusetts General Hospital Gene Delivery Technology Core ) was bilaterally infused targeting the IL ( 0 . 15 µL/hemisphere ) . Each infusion was done over 10 min using a Hamilton syringe and 33-gauge needle . The needle was left in place for a further 5 min to ensure diffusion . The coordinates for mPFC and BNST infusions were as described above . Four weeks after surgery , mice underwent conditioning and retrieval testing as described under ‘Partially versus fully reinforced threat ( standard procedure ) ’ . Thirty minutes prior to the retrieval test , CNO ( 0 . 01 mg/mL/kg ) was injected intraperitoneally . After the completion of testing , mice were terminally anesthetized with sodium pentobarbital ( 50–60 mg/kg ) . Brains were removed and suspended in 4% PFA overnight and then at 4°C in 0 . 1 M PB for 1–2 days . Coronal sections ( 50 μm thick ) were cut with a vibratome ( Leica VT1000 S , Leica Biosystems Inc ) and coverslipped with Vectashield HardSet mounting medium with DAPI ( Vector Laboratories , Inc , Burlingame , CA , USA ) . Sections were imaged using an Olympus BX41 microscope ( Olympus America Inc , Center Valley , PA , USA ) . Mice without viral ( i . e . , mCherry ) expression in the region of interest were removed from the analysis . Mice were placed in a stereotaxic alignment system ( Kopf Instruments ) and kept under isoflurane anesthesia . rAAV2-retro-Ef1a-Cre ( titer: 1 . 0 × 1013 GC/mL , obtained from the Salk Institute ) was bilaterally infused targeting the BNST ( 0 . 25 µL/hemisphere ) . In addition , either AAV8-hSyn-DIO-hM4D ( Gi ) -mCherry-WPRE ( titer: 2 . 25 × 1013 GC/mL , obtained from the Massachusetts General Hospital Gene Delivery Technology Core ) or AAV8 . 2-hEF1-DIO-mCherry-WPRE ( titer: 2 . 13 × 1013 vg/mL , obtained from the Massachusetts General Hospital Gene Delivery Technology Core ) was bilaterally infused targeting the IL ( 0 . 25 µL/hemisphere ) . Infusions were done over 10 min using a Hamilton syringe and 33-gauge needle . The needle was left in place for a further 5 min to ensure diffusion . The coordinates for mPFC and BNST infusions were as described above . During the same surgery , a microelectrode array ( two rows of eight electrodes with 35 µm electrode spacing and 200 µm row spacing [Innovative Neurophysiology , Durham , NC , USA] ) was unilaterally ( hemisphere counterbalanced ) targeting the IL ( array center: ML = ±0 . 30 mm , DV = −2 . 70 mm , AP = +1 . 75 mm ) and affixed to the skull with dental cement . Five weeks after surgery , mice were habituated to the recording tethers for 30 min in their home-cage for two consecutive days prior to behavioral testing . Mice underwent conditioning and retrieval testing using the standard procedure described above , with the exception that retrieval was conducted in a 30 cm diameter clear acrylic cylinder with an open top to accommodate the cable connecting the head-stage . Thirty minutes prior to the retrieval test , CNO ( 0 . 01 mg/mL/kg ) was intraperitoneally injected . Electrophysiological and behavioral recordings were acquired using SpikeGadgets main control unit and Trodes software ( SpikeGadgets , San Francisco , CA , USA ) . Unit recordings were carried out using 16-channel digitizing head-stages , sampled at 30 kHz . Behavioral videos were scored offline by an experimenter blind to conditions . Single units were sorted manually using Offline Sorter v3 . 0 ( Plexon Inc , Dallas , TX , USA ) and analyzed using NeuroExplorer , version 5 ( Nex Technologies , Colorado Springs , CO , USA ) as previously described ( Halladay and Blair , 2015; Halladay et al . , 2020 ) . Unit data were aligned to CS and freezing events . Freezing was manually scored ( by an experimenter blind to experimental group ) and resultant time stamps aligned with the neuronal data . Freezing onset was defined as a transition from movement to no visible movement except that required for breathing . Freezing cessation was defined as a transition from freezing to movement . To determine whether units were responsive to the CS , data during a 500 ms window following the start of the CS for each unit were binned in 100 ms bins and normalized to a 1 s baseline defined as the 10 bins immediately prior to the start of the CS . Units with at least two bins of the same sign in the 500 ms following the start of the CS with a value of >1 . 96 ( p<0 . 05 ) were considered significantly different from baseline and classified as CS responsive . Unit responsiveness to freezing onset and freezing cessation were analyzed similarly , with the exception that the baseline was shifted from −2 to −1 ( rather than −1 to 0 ) s prior to start of an onset or cessation event to ensure that event and the baseline were temporally separate . On completion of testing , mice were anesthetized with 2% isoflurane and a current stimulator ( S48 Square Pulse Stimulator , Grass Technologies , West Warwick , RI , USA ) that delivered 2 s of 40 µA DC current through each electrode to make a small marking lesion . The next day , mice were overdosed via an intraperitoneal injection of 150 mg/kg Euthasol ( Henry Schein , Melville , NY , USA ) and perfused intracardially with PBS followed by 4% PFA . Brains were left in 4% PFA overnight , then transferred to a 30% sucrose PBS solution for cryoprotection . Coronal sections ( 50 µm thick ) were cut on a cryostat ( Leica Biosystems Inc ) and mounted onto slides . Tissue was stained with DAPI ( Sigma–Aldrich ) and imaged using a Keyence BZ-X800 fluorescence microscope ( Keyence Corporation of America , Itasca , IL , USA ) . Mice without viral ( i . e . , mCherry ) expression in the mPFC or correct electrode placement in the IL were removed from the analysis . Differences in freezing and c-Fos counts were analyzed using ANOVA followed by Dunn’s post hoc tests . Differences in z scored single-unit values were analyzed using paired t-tests . Differences in the percentage of recorded units responsive to the CS onset , freezing onset , and freezing cessation were analyzed using non-parametric Fisher’s exact tests . The threshold for statistical significance was set at p<0 . 05; significance values are shown up to p<0 . 0001 . | While walking home alone late one night , you hear footsteps behind you . Your heart starts to beat faster as you wonder whether someone might be following you . Being able to identify and evade threats is essential for survival . A key part of this process is learning to recognize signals that predict potential danger: the sound of footsteps behind you , for example . But many such cues are unreliable . The person behind you might simply be heading in the same general direction as you . And if you spend too much time and energy responding to such false alarms , you may struggle to complete other essential tasks . To be useful , responses to cues that signal potential threats must thus be proportionate to the likelihood that danger is actually present . By studying threat detection in mice , Glover et al . have identified a brain circuit that helps ensure that this is the case . Two groups of mice learned to fear a tone that predicted the delivery of a mild footshock . In one group of animals , the tone was followed by a shock on every trial ( it was said to be ‘fully reinforced’ ) . But in the other group , the tone was followed by a shock on only 50% of trials ( ‘partially reinforced’ ) . After training , both groups of mice froze whenever they heard the tone – freezing being a typical fear response in rodents . But the animals trained with the partially reinforced tone showed less freezing than their counterparts in the fully reinforced group . Moreover , freezing in response to the partially reinforced tone was accompanied by activity in a specific neural pathway connecting the frontal part of the brain to an area called the bed nucleus of the stria terminalis . Inhibiting this pathway made mice respond to the partially reinforced tone as though it had been reinforced on every trial . This suggests that activity in this pathway helps dampen responses to unpredictable threat cues . In people with anxiety disorders , cues that become associated with unpleasant events can trigger anxiety symptoms , even if the association is unreliable . The findings of Glover et al . suggest that reduced activity of circuits that constrain excessive responses to threats might contribute to anxiety disorders . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2020 | A prefrontal-bed nucleus of the stria terminalis circuit limits fear to uncertain threat |
Kinesins are a superfamily of microtubule-based ATP-powered motors , important for multiple , essential cellular functions . How microtubule binding stimulates their ATPase and controls force generation is not understood . To address this fundamental question , we visualized microtubule-bound kinesin-1 and kinesin-3 motor domains at multiple steps in their ATPase cycles—including their nucleotide-free states—at ∼7 Å resolution using cryo-electron microscopy . In both motors , microtubule binding promotes ordered conformations of conserved loops that stimulate ADP release , enhance microtubule affinity and prime the catalytic site for ATP binding . ATP binding causes only small shifts of these nucleotide-coordinating loops but induces large conformational changes elsewhere that allow force generation and neck linker docking towards the microtubule plus end . Family-specific differences across the kinesin–microtubule interface account for the distinctive properties of each motor . Our data thus provide evidence for a conserved ATP-driven mechanism for kinesins and reveal the critical mechanistic contribution of the microtubule interface .
Kinesins are a large family of microtubule ( MT ) -based motors that play important roles in many cellular activities including mitosis , motility , and intracellular transport ( Vale , 2003; Hirokawa and Noda , 2008; Hirokawa et al . , 2010 ) . Their involvement in a range of pathological processes also highlights their significance as therapeutic targets and the importance of understanding the molecular basis of their function ( Mandelkow and Mandelkow , 2002; Greber and Way , 2006; Henry et al . , 2006; Stokin and Goldstein , 2006; Liu et al . , 2012b ) . Kinesins are defined by their motor domains that contain both the MT and ATP binding sites . Three ATP binding motifs—the P-loop , switch I , switch II–are highly conserved among kinesins ( Sablin et al . , 1996 ) , myosin motors , and small GTPases ( Vale , 1996 ) . Kinesins also share a conserved mode of MT binding ( Woehlke et al . , 1997; Alonso et al . , 1998 ) such that MT binding , ATP binding , and hydrolysis are functionally coupled for efficient MT-based work . A number of kinesins drive long distance transport of cellular cargo ( Hirokawa et al . , 2010; Soppina et al . , 2014 ) with dimerisation allowing them to take multiple 8 nm ATP-driven steps toward MT plus ends ( Svoboda et al . , 1993 ) . Their processivity depends on communication between the two motor domains , which is achieved via the neck linker that connects each motor domain to the dimer-forming coiled-coil ( Hackney , 1994; Rice et al . , 1999; Tomishige and Vale , 2000; Clancy et al . , 2011 ) . In the presence of MTs , ATP binding stimulates neck linker association ( docking ) with the motor domain towards the MT plus end , while ATP hydrolysis and MT release cause neck linker undocking ( Rice et al . , 1999; Vale and Milligan , 2000; Skiniotis et al . , 2003; Asenjo et al . , 2006 ) ; thus , the neck linker is required for both intra-dimer communication and directionality . However , even when the role of the motor N-terminus in reinforcing neck linker movement via cover neck bundle ( CNB ) formation is considered ( Hwang et al . , 2008; Khalil et al . , 2008 ) , the contribution of neck linker docking to the force generating mechanism ( s ) of these kinesins remains uncertain ( Rice et al . , 1999; Vale and Milligan , 2000; Rice et al . , 2003 ) . New insights into the conformational rearrangements of these motors when bound to MTs are essential to reveal how they produce force . The high resolution X-ray structures of a range of kinesin motor domains have established a major communication route from the nucleotide binding site via helix-α4 ( the so-called relay helix ) to the neck linker , such that alternate conformations of helix-α4 either block or enable neck linker docking ( Vale and Milligan , 2000; Kikkawa et al . , 2001 ) . However , the neck linker conformation seen in these MT-free structures is not always correlated to the nucleotide bound ( Vale and Milligan , 2000; Grant et al . , 2007 ) . Cryo-electron microscopy ( cryo-EM ) has played a major role in elucidating several aspects of MT-bound kinesin mechanochemistry ( Sosa and Milligan , 1996; Sosa et al . , 1997; Rice et al . , 1999; Skiniotis et al . , 2003; Hirose et al . , 2006; Kikkawa and Hirokawa , 2006; Sindelar and Downing , 2007 , 2010; Goulet et al . , 2012 , 2014 ) . Despite these contributions , and despite recent advances in the study of kinesin–tubulin complexes using X-ray crystallography ( Gigant et al . , 2013 ) , several outstanding questions concerning kinesin mechanochemistry remain . Specifically , the mechanism by which MT binding stimulates the kinesin ATPase and in particular enhances Mg-ADP release by several orders of magnitude is not clear ( Hackney , 1988; Ma and Taylor , 1997; Sindelar , 2011 ) . Although several speculative models have been proposed , an unambiguously interpretable structure of nucleotide-free MT-bound kinesin is currently lacking and is clearly critical in establishing how such transitions are achieved . Such a structure would also provide key insights into how ATP binding is coupled to both neck linker docking and force generation . To address these major questions , we describe the MT-bound mechanochemical cycles of two plus-end directed human kinesin motor domains , a kinesin-1 ( Kif5A ) and a kinesin-3 ( Kif1A ) , using cryo-EM structure determination at subnanometer resolution . Kinesin-1s ( Kin1 ) and kinesin-3s ( Kin3 ) are both important neuronal plus-end directed transport motors ( Hirokawa et al . , 2009b ) , but recent data suggest that Kin3 rather than Kin1 motors specifically are involved in long distance transport ( Soppina et al . , 2014 ) . Their motor domains share 41% sequence identity , but profoundly different mechanochemistries—in which Kin1 dimers take processive steps and Kin3 monomers diffuse along MT tracks—have been proposed for these motors ( Hirokawa et al . , 2009a; Sindelar , 2011 ) . Thus , we wanted to investigate these differences and compare the motors side by side . The high quality of our reconstructions , coupled to flexible fitting , enables new insights into the kinesin mechanism . In particular , nucleotide-free reconstructions for both motor domains reveal a conserved mechanism , whereby MT binding stimulates changes at the nucleotide-binding site favouring Mg-ADP release and conformationally primes the motor to receive Mg-ATP . We also show that relatively small structural transitions occur at the nucleotide-binding site on Mg-ATP binding , but that these lead to larger scale conformational changes and neck linker docking . Structural analysis of two different transport kinesins allows a direct comparison of their conserved mechanochemical features and identification of attributes that confer distinctive properties on each motor .
We calculated MT-bound Kin3 reconstructions and pseudo-atomic models in four different nucleotide states: ( 1 ) Mg-ADP; ( 2 ) no nucleotide ( NN ) , using apyrase treatment; ( 3 ) Mg-AMPPNP ( a non-hydrolysable ATP analogue ) ; and ( 4 ) Mg-ADPAlFx ( an ATP hydrolysis transition state mimic ) , consistent with the previously described tight association of the Kin3 motor domain with MTs throughout its ATPase cycle ( Tables 1 and 2 , Figure 1—figure supplements 1 , 2; Okada and Hirokawa , 2000 ) . We also calculated three Kin1 reconstructions and pseudo-atomic models: ( 1 ) no nucleotide ( NN ) , ( 2 ) Mg-AMPPNP , and ( 3 ) Mg-ADPAlFx ( Tables 1 and 2 , Figure 1—figure supplements 1 , 2 ) . Steady-state ATPase activities of the proteins that we used for our cryo-EM reconstructions ( Table 3 ) show that the catalytic turnover of these motors are similar , but that the KmMT of Kin3 is ∼250× lower than Kin1 . These values are broadly consistent with previous reports and also with our ability to form complexes for structure determination ( Woehlke et al . , 1997; Okada and Hirokawa , 1999; Sindelar and Downing , 2010 ) . The conformations of both Kin3 and Kin1 in Mg-AMPPNP and Mg-ADPAlFx states were indistinguishable from each other at the resolution of our reconstructions ( global RMSD: Kin3 ADPAlFx/AMPPNP = 0 . 7 Å; Kin1 ADPAlFx/AMPPNP = 0 . 6 Å ) as had been previously shown in other studies of transport kinesins ( Kif5B; Sindelar and Downing , 2010; Gigant et al . , 2013 ) . Thus , for simplicity , we describe here one Mg-ATP-analogue ( ‘Mg-ATP-like’ ) reconstruction for each kinesin ( Kin3: Mg-ADPAlFx; Kin1: Mg-AMPPNP ) . Views of the alternative Mg-ATP-like reconstructions for each kinesin are shown in figure supplements . 10 . 7554/eLife . 03680 . 003Table 1 . Data set size and estimated reconstruction resolutionsDOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 003Kinesin and nucleotide stateNumber of AUFSCt 0 . 5 ( 0 . 143 ) FSCtrue 0 . 5 ( 0 . 143 ) Rmeasure 0 . 5 ( 0 . 143 ) EMDB accession numberKin3-Mg-ADP181 , 3117 . 9 ( 6 . 3 ) 8 ( 7 ) 8 . 1 ( 7 . 5 ) EMD-2768Kin3-NN187 , 5387 . 4 ( 6 . 3 ) 7 . 5 ( 6 . 3 ) 7 . 8 ( 6 . 9 ) EMD-2765Kin3-Mg-AMPPNP97 , 8778 . 1 ( 6 . 9 ) 8 . 2 ( 7 . 0 ) 8 ( 7 . 3 ) EMD-2766Kin3-Mg-ADPAlFx156 , 8457 . 9 ( 6 . 8 ) 8 . 3 ( 7 . 0 ) 8 ( 7 . 3 ) EMD-2767Kin1-NN168 , 9748 . 2 ( 7 . 2 ) 8 . 3 ( 7 . 4 ) 8 . 3 ( 7 . 3 ) EMD-2769Kin1-Mg-AMPPNP186 , 3297 . 3 ( 6 . 0 ) 7 . 5 ( 6 . 5 ) 7 . 7 ( 6 . 9 ) EMD-2770Kin1-Mg-ADPAlFx65 , 5729 ( 7 . 3 ) 9 . 1 ( 7 . 7 ) 9 . 1 ( 8 . 1 ) EMD-2771For each reconstruction , the motor domain and nucleotide state , number of asymmetric units ( AU ) in the final reconstruction , the resolutions at a cut-off of 0 . 5 and 0 . 143 estimated by standard FSC ( FSCt ) and that corrected with the HRnoise substitution test ( FSCtrue ) ( Chen et al . , 2013 ) and by Rmeasure ( Sousa and Grigorieff , 2007 ) and the EMDB accession number are given . 10 . 7554/eLife . 03680 . 004Table 2 . Calculation of pseudo-atomic modelsDOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 004Kinesin and nucleotide stateModels usedCCC initial modelCCC final modelPDB codeKin3-Mg-ADP1VFZ ( Nitta et al . , 2004 ) 0 . 660 . 684uxs1I5S ( Kikkawa et al . , 20014AQW ( Goulet et al . , 2012 ) Kin3-NN1VFZ/1I5S/4HNA ( Gigant et al . , 2013 ) /4AQW0 . 630 . 684uxoKin3-Mg-AMPPNP1VFV ( Nitta et al . , 2004 ) 0 . 720 . 754uxp4HNAKin3-Mg-ADPAlFx1VFV/4HNA0 . 740 . 754uxrKin1-NN1BG2 ( Kull et al . , 1996 ) /4HNA/4AQW0 . 710 . 734uxtKin1-Mg-AMPPNP4HNA0 . 730 . 764uxyKin1-Mg-ADPAlFx4HNA0 . 690 . 724uy0A set of starting models were used for each nucleotide state of each motor . Flexible fitting and further refinement were performed using Flex-EM and Modeller ( see ‘Materials and methods’ ) . Global CCCs of models with their respective reconstructions were calculated using the Fit In Map tool in Chimera . PDB accession codes for the final models are also shown . 10 . 7554/eLife . 03680 . 005Table 3 . Steady-state MT-activated ATPase parameters of our Kin3 and Kin1 motor domain constructsDOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 005Kin3 ( Kif1A ) Kin1 ( Kif5A ) kcat ( s−1 ) 43 . 4 ± 1 . 034 . 2 ± 5 . 7K0 . 5ATP ( μM ) 30 ± 1025 ± 5K0 . 5 MT ( nM ) 53 . 7 ± 5 . 712 , 745 ± 4041 All our reconstructions have , as their asymmetric unit , a triangle-shaped motor domain bound to an αβ-tubulin dimer within the MT lattice ( Figure 1 ) . The structural comparisons below are made with respect to the MT surface , which , at the resolution of our structures ( ∼7 Å , Table 1 ) , is the same ( CCC > 0 . 98 for all ) . As is well established across the superfamily , the major and largely invariant point of contact between kinesin motor domains and the MT is helix-α4 , which lies at the tubulin intradimer interface ( Figure 1C , Kikkawa et al . , 2001 ) . However , multiple conformational changes are seen throughout the rest of each domain in response to bound nucleotide ( Figure 1D ) . Below , we describe the conformational changes in functionally important regions of each motor domain starting with the nucleotide-binding site , from which all other conformational changes emanate . 10 . 7554/eLife . 03680 . 006Figure 1 . Overview of MT-bound kinesin motor domain cryo-EM reconstructions . ( A ) Example cryo-EM image of kinesin-decorated MT ( Kin1-Mg-AMPPNP ) ; blue arrows indicate individual Kin1 motor domains . ( B ) Example of cryo-EM reconstruction of 13 protofilament , kinesin-decorated MT ( Kin1-Mg-AMPPNP ) ; blue arrows indicate individual Kin1 motor domains , and the dotted red box shows an asymmetric unit . A single protofilament is indicated along with the position of the lattice seam . ( C ) Example of an individual asymmetric unit ( Kin1-Mg-AMPPNP ) , contoured to show secondary structural elements . ( D ) Two views , related by 180° , of an exemplar pseudo-atomic model ( Kin1-Mg-AMPPNP ) calculated using our cryo-EM reconstruction . The major mechanochemical elements discussed in the text are colour-coded as indicated in the key . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 00610 . 7554/eLife . 03680 . 007Figure 1—figure supplement 1 . Resolution estimation for cryo-EM reconstructions . For each reconstruction , three Fourier Shell Correlation ( FSC ) curves are plotted: standard FSCt ( blue ) between two half data sets , FSCn ( noise substitution cutoff 10 Å , red ) and FSCtrue ( green , see Chen et al . , 2013 ) . ( A ) Kin3-Mg-ADP-MT , ( B ) Kin3-NN-MT , ( C ) Kin3-Mg-AMPPNP-MT , ( D ) Kin3-Mg-ADPAlFx-MT , ( E ) Kin1-NN-MT , ( F ) Kin1-Mg-AMPPNP-MT , ( G ) Kin1-Mg-ADPAlFx-MT . Dotted lines indicate estimated resolution by FSCtrue at 0 . 143 ( considered appropriate for FSCtrue ) and 0 . 5 criteria . The overall good agreement between FSCt and FSCtrue curves demonstrates that minimal over-fitting occurred during refinement of the cryo-EM data . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 00710 . 7554/eLife . 03680 . 008Figure 1—figure supplement 2 . Local assessment of fit quality of the pseudo-atomic models within the cryo-EM density . Following flexible fitting of each kinesin motor domain , the local fit quality of specific elements was calculated . ( A and B ) NN cryo-EM density for ( A ) Kin3 and ( B ) Kin1 is shown with their respective docked pseudo-atomic model colour-coded according to segment based cross correlation coefficient ( SCCC , see colour key; Pandurangan et al . , 2014 ) . ( C and D ) Heat map showing the quality of the local fit for specific elements of the motor domain in different nucleotide states for ( C ) Kin3 and ( D ) Kin1 . The colour ( see key ) denotes the SCCC score as calculated with TEMPy ( Farabella et al . , 2014 ) . This analysis shows the quality of the fits and provides confidence in our interpretation of conformational changes in these regions . In particular , it shows that loop9 and loop11 have similar ( good ) quality of fit compared to the α-helices , apart from loop11 in the Kin3-Mg-ADP reconstruction , for which cryo-EM density was not seen . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 008 The nucleotide-binding site ( Figure 2 ) has three major elements: ( 1 ) the P-loop ( brown ) is visible in all our reconstructions; ( 2 ) loop9 ( yellow , contains switch I ) undergoes major conformational changes through the ATPase cycle; and ( 3 ) loop11 ( red , contains switch II ) that connects strand-β7 to helix-α4 , the conformation and flexibility of which is determined by MT binding and motor nucleotide state . The presence or absence of density for nucleotide in the nucleotide-binding site in each reconstruction ( Figure 2 and Figure 2—figure supplement 5 ) is consistent with the well-established sample preparation methods used ( see ‘Materials and methods’ ) . In the Kin3-Mg-ADP reconstruction , the N-terminal half of helix-α4 lies at the back of the nucleotide-binding site where its N-terminal end is partially flexible ( Figure 2A ) . ∼50% of the adjacent loop11 is not visible presumably also due to flexibility , and density for this loop is only visible close to the P-loop at the edge of the motor's central β-sheet . In contrast , density corresponding to loop9 is clearly defined: the 4-turn helix-α3 is broken by a single residue , before two further helical segments are seen , one of which coordinates Mg-ADP , together with switch II ( Coureux et al . , 2003; Hirose et al . , 2006; Kull and Endow , 2013 ) . The conformations of loop9 and loop11 in this reconstruction are thus essentially the same as is seen in the Kin3-Mg-ADP crystal structure ( Kikkawa et al . , 2001 ) . 10 . 7554/eLife . 03680 . 009Figure 2 . Conserved conformations at the nucleotide-binding pocket in Kin3s and Kin1s . ( A–C ) The nucleotide-binding pocket of MT-bound Kin3 reconstructions ( shown as blue transparent density ) in ( A ) Mg-ADP , model shown in light blue; the arrowhead indicates residual flexibility in the helix-α4 N-terminus and the region of loop11 for which density is missing is depicted by a dotted red line; ( B ) no nucleotide ( NN ) , model shown in mid-blue; density connects the C-terminal helical turn of loop11 with the MT ( arrow ) , density corresponding to the rest of loop11 is seen ( chevron ) and density now connects the extended loop 9 and the P-loop ( arrowhead ) ; ( C ) Mg-ADPAlFx , model shown in dark blue; the C-terminal helical turn of loop11 has moved away from the MT ( arrow ) and strong density is seen connecting it , helix-α4 and loop9 around the bound nucleotide . ( D–E ) The nucleotide-binding pocket of MT-bound Kin1 reconstructions ( shown as green transparent density ) in ( D ) no nucleotide , model shown in light green; density connects the C-terminal helical turn of loop11 with the MT ( arrow ) , density corresponding to the majority of loop11 is seen ( chevron ) and density now connects the extended loop 9 and the P-loop ( arrowhead ) ; ( E ) Mg-AMPPNP , model shown in dark green; the C-terminal helical turn of loop11 has moved away from the MT ( arrow ) and strong density is seen connecting it , helix-α4 and loop9 around the bound nucleotide . In all reconstructions , density for the motor domain was contoured to an equivalent volume . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 00910 . 7554/eLife . 03680 . 010Figure 2—figure supplement 1 . Conserved conformations at the nucleotide-binding pocket in Kin3 and Kin1 alternative ATP-like states . ( A ) The nucleotide-binding pocket of the MT bound Kin3-Mg-AMPPNP ( blue transparent density and navy blue model ) . ( B ) The nucleotide-binding pocket of the MT bound Kin1-Mg-ADPAlFx reconstruction ( green transparent density and olive green model ) . The major features are shared by all the ATP-like reconstructions: in Kin3-Mg-AMPPNP the C-terminal helical turn of loop11 has moved away from the MT ( arrow ) and strong density ( arrowhead ) is seen connecting it , helix-α4 and loop9 around the bound nucleotide . The Kin1-Mg-ADPAlFx reconstruction is lower resolution ( FSCtrue , 0 . 143 = 7 . 7 ) , which may explain why residual density connects the C-terminal helical turn of loop11 with the MT ( arrow ) ; however strong density is seen connecting it , helix-α4 and loop9 around the bound nucleotide . In all reconstructions , density for the motor domain was contoured to an equivalent volume . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 01010 . 7554/eLife . 03680 . 011Figure 2—figure supplement 2 . Coordination of Mg-ADP cluster by loop9 and loop11 . ( A ) Sequence alignment of Kin3 and Kin1 highlighting conserved Mg-water ‘cap’ coordinating residues ( magenta squares above residue letters ) in loop9 ( yellow shading ) and near loop11 ( red shading ) . ( B ) The crystal structure of Kin3-Mg-ADP ( Kif1A; PDB 1I5S; Kikkawa et al . , 2001 ) showing the side chains of the residues ( Kin3: Arg203 , Ser214 , Ser215 , Asp248 ) indicated in panel A . Putative hydrogen bonds ( displayed with FindHBond Chimera plugin ) between these residues and the Mg-water cap are shown as solid magenta lines . Water molecules and Mg are shown as red and green spheres respectively . We propose that MT-triggered displacement of loop9 leads to destabilization of the Mg-water cap and consequent Mg-ADP release from the nucleotide pocket . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 01110 . 7554/eLife . 03680 . 012Figure 2—figure supplement 3 . Conserved residues involved in MT-mediated stimulation of Mg-ADP release . ( A ) Sequence alignment of Kin3 and Kin1 highlighting residues likely to be important in MT-mediated stimulation of Mg-ADP release . Residues involved in MT sensing and stabilization of loop11 are indicated by purple squares above residue letters ( Kin3 residue number ) , whereas those involved in communication between loop11 ( at the MT ) and loop9 ( water-Mg-ADP coordination ) are indicated by magenta squares . Loop9 is indicated by yellow shading , loop11 by red shading , and the P-loop by brown shading . ( B and C ) Location of these residues in the NN-MT-bound models of ( B ) Kin3 ( mid blue ) within the equivalent reconstruction ( blue transparent density ) and ( C ) Kin1 ( light green ) within the equivalent reconstruction ( green transparent density ) , contoured at equivalent volumes . We propose that MT binding reduces the conformational freedom of loop11 , stabilizing a helical turn that involves Kin3 Ala255 ( Kin1 Val238 ) and Ala260 ( Kin1 Ala244 ) , and Kin3 Thr258 ( Kin1 Thr242 ) above α-tubulin's H3ʹ . Kin3 helix-α4 Asn272 ( Kin1 Asn256 ) sits at the interface of α-tubulin and loop11 , likely interacting with both ( Gigant et al . , 2013 ) and presumably stabilizing loop11 . Kin3 loop11 Arg254 ( Kin1 Lys238 ) may help stabilize loop11 through its interaction with the acidic tip of α-tubulin's H12 ( Gigant et al . , 2013 ) . Communication between loop11 and loop9 likely occurs via a salt bridge between Kin3 loop Glu253 ( Kin1 Glu237 ) and loop9 Arg216 ( Kin1 Arg204 ) as reported in hydrolysis-competent conformation ATP-like crystal structures ( Chang et al . , 2013; Gigant et al . , 2013; Parke et al . , 2010 ) . Kin3 helix-α4 Glu267 ( Kin1 Glu251 ) also interacts with loop9 Arg216 ( Kin1 Arg204 ) , an interaction that also involves loop7 Tyr150 ( Kin1 Tyr139; Liu et al . , 2012a ) . Evidence for these residues involvement in MT-mediated Mg-ADP release is provided by structural and biochemical studies and disease-causing patient mutations ( *Nitta et al . , 2008; †Woehlke et al . , 1997; ‡Yun et al . , 2001; §Ebbing et al . , 2008; ¶Song and Endow , 1998; \\Liu et al . , 2012a ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 01210 . 7554/eLife . 03680 . 013Figure 2—figure supplement 4 . Structural routes of communication between the nucleotide-binding pocket and helix-α6 for mechanochemical coupling . ( A ) Sequence alignment of Kin3 and Kin1 highlighting residues involved in communication from the nucleotide-binding pocket to helix-α6 . Residues involved in loop9-loop11 communication are indicated by magenta squares above residue letters and loop11-helix-α6 communication by orange squares above residue letters . Residue numbers for Kin1 ( Kif5A ) are indicated . Loop9 is indicated by yellow shading , loop11 by red shading , and the P-loop by brown shading . ( B ) The crystal structure of tubulin dimer-bound Kin1-Mg-ADPAlFx ( Kif5B; PDB 4HNA ) focusing on the residues indicated in panel A . Residue numbers for Kif5A are indicated . The close association of loop9 and loop11 in ATP-like crystal structures ( Chang et al . , 2013; Gigant et al . , 2013; Parke et al . , 2010 ) involves backbone hydrogen bonds between loop9 Asn197 and loop11 Thr242 , and also involves Met198 . Residues in loop11 ( Lys241 , Lys238 in Kin1 , Arg264 in Kin3 ) interact with the base of helix-α6 ( Asn310 , Glu313 in Kin1 , Asn337 , Glu340 in Kin3 ) . P-loop residues in Kin1 ( Tyr85 , Gln87; Kin3 Tyr96 , Gln98 ) also interact with helix-α6 . We propose that these interactions will form in the transition from NN to Mg-ATP bound ( Figure 2 ) and will contribute to mechanical transmission ( Figure 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 01310 . 7554/eLife . 03680 . 014Figure 2—figure supplement 5 . Occupancy of the nucleotide pocket . Similar views of the nucleotide-binding pocket aligned on the P-loop are shown for each reconstruction , with the corresponding model fitted into density; ( A ) Kin3-Mg-ADP , ( B ) Kin3-NN , ( C ) Kin3-Mg-AMPPNP , ( D ) Kin3-Mg-ADPAlFx , ( E ) Kin1-NN , ( F ) Kin1-Mg-AMPPNP , ( G ) Kin1-Mg-ADPAlFx . The presence or absence of density in the nucleotide-binding pocket is consistent with the sample preparation used for each reconstruction and supports their interpretation . ( H ) The Kin3-Mg-ADP model is shown in the Kin3-NN reconstruction , clearly demonstrating the lack of density in the nucleotide-pocket to accommodate Mg-ADP ( arrow ) and supporting our assignment of this structure as nucleotide-free . The opacity of all reconstructions in this figure has been increased in order to more clearly illustrate the boundary of the EM density compared to the docked model . The contouring is the same as in all other figures . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 014 In the Kin3-NN reconstruction ( Figure 2B ) , the N-terminus of helix-α4 is fully stabilised , while the C-terminal portion of loop11 adopts a helical turn that forms a new contact with α-tubulin that likely contributes to the strengthened motor domain-MT interaction in the NN state ( Nakata and Hirokawa , 1995 ) . Density corresponding to the rest of loop11 is now also fully visible , such that switch II is seen running from the β-sheet core past the P-loop . Loop9 has undergone a large conformational change: helix-α3 now terminates after four turns and the resulting elongated conformation of loop9 forms a finger-like extension that reaches towards the nucleotide pocket and the new helical turn in loop11 . Density connects this extended form of loop9 and the N-terminus of helix-α4; density also connects the P-loop and loop9 ( as previously described for Kif5B; Sindelar and Downing , 2007; Sindelar , 2011 ) . The Kin1-NN reconstruction shows a very similar configuration at the nucleotide-binding site ( Figure 2D ) . This arrangement of the nucleotide–binding loops in both motors is striking because even in the absence of bound nucleotide , the loops adopt a conformation related ( but not identical ) to that formed when Mg-ATP is bound ( Parke et al . , 2010; Chang et al . , 2013; Gigant et al . , 2013 ) . That is , MT-stimulated Mg-ADP release appears to conformationally prime the switch loops for Mg-ATP binding . The similarity of these reconstructions supports the idea of a conserved mechanism of: ( 1 ) MT-induced Mg-ADP release ( Figure 2—figure supplement 3 ) and ( 2 ) MT priming of the conformation of the nucleotide-binding pocket to receive Mg-ATP in both Kin1s and Kin3s . Because of this conformational priming , structural changes in the nucleotide-binding site upon ATP-binding are comparatively small when the NN and Mg-ATP-reconstructions are compared ( Figure 2B–E , Figure 2—figure supplement 1 ) . In both Kin3 and Kin1 , loop9 now reaches further into the nucleotide-binding pocket to cradle the Mg-ATP mimic , enclosing it in a catalytically competent conformation and forming continuous density with the nucleotide and P-loop ( Figure 2C , E ) . The C-terminus of loop11 retains a helical turn conformation similar to that observed in the nucleotide free reconstructions . Density for the N-terminus of loop11 runs from the core β-sheet past the P-loop and the γ-phosphate mimic . Importantly , however , in comparison to the nucleotide-free reconstruction , the loop11 helical turn shows reduced contact with tubulin and has moved toward loop9 and helix-α6 ( see arrow , Figure 2C , E ) . The ‘pincer-like’ movement of the switch loops is associated with formation of a prominent connection of density between them and is consistent with a ‘phosphate tube’ structure similar to that described recently for other kinesins ( Parke et al . , 2010; Sindelar and Downing , 2010; Chang et al . , 2013; Gigant et al . , 2013 ) . We note that , although the structure of the mammalian Kin1 Kif5A bound to MT has not previously been determined , our Kif5A reconstruction displays the major features seen in the recently published tubulin dimer-bound Kif5B Mg-ADPAlFx X-ray structure and to previous Mg-ATP analogue Kif5B cryo-EM reconstructions ( Sindelar and Downing , 2007 , 2010; Gigant et al . , 2013 ) . Overall , in response to the presence of γ-phosphate , loop9 and loop11 draw closer to each other and to helix-α6 in both motors . This movement also reduces the density that connects loop11 with the MT . As shown in Figure 2 , the N-terminus of helix-α6 is closely associated with elements of the nucleotide-binding site suggesting that its conformation alters in response to different nucleotide states . In addition , because the orientation of helix-α6 with respect to helix-α4 controls neck linker docking ( Vale and Milligan , 2000; Kikkawa et al . , 2001 ) , and because helix-α4 is held against the MT during the ATPase cycle , conformational changes in helix-α6 control movement of the neck linker . In the Kin3-Mg-ADP reconstruction , helix-α6 contacts α-tubulin as was previously reported ( Figure 3A , arrowhead; Kikkawa and Hirokawa , 2006 ) ; this interaction is likely to involve basic residues conserved in Kin3 ( discussed below ) and negatively charged residues in the N-terminal region of α-tubulin H12 . The small β-sheet composed of strands-β1a , b , c ( β-sheet1abc ) lies on top of helix-α6 and above the MT surface; this β-sheet is situated roughly perpendicular to the core β-sheet of the motor domain and contains the characteristically extended Kin3 loop2 . In the Kin3-Mg-ADP state , the orientation of helix-α6 with respect to helix-α4 ensures both that helix-α6 cannot fully extend and the neck linker is undocked; this is indicated , first , by a lack of density between helix-α4 and helix-α6 , and second by a lack of density along the core β-sheet ( Figure 3—figure supplement 3A ) . The neck linker is mainly invisible and presumably disordered , consistent with previous reports ( Rice et al . , 1999; Skiniotis et al . , 2003 ) . However , some density that probably corresponds to the N-terminus of the neck linker is visible extending from the C-terminus of helix-α6 , suggesting its flexible conformations are directed largely towards the MT minus end ( Figure 3A , arrow and Figure 3—figure supplement 3A ) . Density that is likely to correspond to the Kin3 N-terminus is also visible but no single conformation can be distinguished . 10 . 7554/eLife . 03680 . 015Figure 3 . Conserved conformational changes of helix-α6 alter MT connectivity and allow neck linker docking on Mg-ATP binding . ( A–C ) View of helix-α6 and the neck linker ( in fuchsia ) of MT-bound Kin3 reconstructions ( shown as blue transparent density ) in ( A ) Mg-ADP , model shown in light blue , ( B ) no nucleotide ( NN ) , model shown in mid-blue , ( C ) Mg-ADPAlFx , model shown in dark blue; ( D–E ) View of helix-α6 and the neck linker ( in fuchsia ) of MT-bound Kin1 reconstructions ( shown as green transparent density ) in ( D ) no nucleotide , model shown in light green , ( E ) Mg-AMPPNP , model shown in dark green . In Mg-ADP ( Kin3 ) and NN states ( both motors ) , helix-α6 contacts the surface of α-tubulin ( arrowhead ) and its orientation with respect to helix-α4 ensures that the neck linker cannot dock . Regions of density at the C-terminal end of helix-α6 likely representing conformers of the N-terminal portion of the neck linker are observed ( arrows ) , although the majority is not visible , presumably due to flexibility . In both motors , peeling of the motor domain β-sheet core away from helix-α4 upon Mg-ATP binding allows rotation and extension of helix-α6 , drawing it away from the MT surface ( arrowhead ) , and allowing it to occupy the space between helix-α4 and the β-sheet core . The neck linker docks towards the MT plus end ( arrow ) and forms the CNB with the N-terminus ( in orange ) . In all reconstructions , density for the motor domain was contoured to an equivalent volume . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 01510 . 7554/eLife . 03680 . 016Figure 3—figure supplement 1 . Conserved conformation of helix-α6 allows neck linker docking on Mg-ATP binding in Kin3 and Kin1 alternative ATP-like states . ( A ) View of helix-α6 and the neck linker ( in fuchsia ) of MT bound Kin3-Mg-AMPPNP ( blue transparent density and navy blue model ) . ( B ) View of helix-α6 and the neck linker ( in fuchsia ) of MT bound Kin1-Mg-ADPAlFx reconstruction ( green transparent density and olive green model ) . The major features are shared by all the ATP-like reconstructions: in both motors , peeling of the motor domain β-sheet core on Mg-ATP binding allows rotation and extension of helix-α6 , drawing it away from the MT surface ( arrowhead ) . The neck linker docks towards the MT plus end ( arrow ) and forms the CNB with the N-terminus ( in orange ) . In all reconstructions , density for the motor domain was contoured to an equivalent volume . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 01610 . 7554/eLife . 03680 . 017Figure 3—figure supplement 2 . Tilting of the core β-sheet on Mg-ATP binding in Kin1 and Kin3 causes peeling of the β-sheet from the C-terminus of helix-α4 to allow movement and extension of helix-α6 and neck linker docking . In each panel , a stripped-down depiction of each pseudo-atomic model is presented showing helix-α4 , adjacent loops ( shown for orientation ) and the core β-sheet , viewed from the MT minus end . ( A ) MT bound Kin3-NN; ( B ) MT bound Kin3-ATP-like; ( C ) MT bound Kin3-NN; ( D ) Kin-ATP-like . In each case , the distance between the backbone Cα of conserved residues at the helix-α4 C-terminus and the immediately overlying β-sheet region were measured in Chimera ( indicated in pink ) . The tilt of each β-sheet upon ATP-analogue binding was calculated by measuring the change in angle between helix-α4 and the β-sheet using the Axes/Planes/Centroids tool in Chimera . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 01710 . 7554/eLife . 03680 . 018Figure 3—figure supplement 3 . Conserved conformational changes of helix-α6 relative to helix-α4 control neck-linker docking along the core β-sheet when Mg-ATP binds . ( A–D ) View towards the MT with the plus end towards the top of MT-bound Kin3 reconstructions ( shown as blue transparent density ) in ( A ) Mg-ADP , model shown in light blue , ( B ) no nucleotide ( NN ) , model shown in the mid-blue , ( C ) Mg-AMPPNP , model shown in navy blue , and ( D ) Mg-ADPAlFx , model shown in dark blue; ( E–G ) Same view of MT-bound Kin1 reconstructions ( shown is green transparent density in E ) no nucleotide ( NN ) , model shown in light green , ( F ) Mg-AMPPNP , model shown in dark green , ( G ) Mg-ADPAlFx , model shown in olive green . In Mg-ADP/NN states of Kin3 ( A and B ) and the NN state of Kin1 ( E ) helix-α6 terminates before helix-α4 leaving a gap ( chevrons ) . Additional regions of density ( arrows ) at the helix-α6 C-terminus likely represent conformers of the initial portion of the neck linker ( fuchsia ) , most of which is invisible and presumably flexible . However , in AMPPNP/ADPAlFx states of both Kin3 ( C and D ) and Kin1 ( F and G ) , tilting of the motor domain allows helix-α6 to extend , closing the gap between helix-α4 and allowing neck linker docking , for which extra density is seen alongside the core β-sheet ( arrowheads ) . Neck linker docking allows CNB formation with the N-terminus ( orange ) . In all reconstructions , density for the motor domain was contoured to an equivalent volume . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 018 In the Kin3-NN reconstruction , contact between helix-α6 and α-tubulin remains fixed , although the C-terminal end of helix-α4 is disconnected from the MT at its junction with the helix-α6 C-terminus ( Figure 3B ) . The relative orientation of these helices ensures that the neck linker remains undocked and flexible; this is again indicated by the gap separating these helices and by density extending from the C-terminus of helix-α6 , similar to that described in the Mg-ADP state ( Figure 3B and Figure 3—figure supplement 3B ) . The flexible distribution of the N-terminus is also unaltered . The Kin1-NN reconstruction shows an overall similar configuration in the region of helix-α6 , with its neck linker undocked and flexible and its N-terminus disordered ( Figure 3D and Figure 3—figure supplement 3E ) . However , some family specific differences are apparent , both within the motor domain structure and at the motor–MT interface ( Figure 3D ) . For example , in Kin1 β-sheet1abc appears more compact than in Kin3 because loop2 and loop3 are shorter . In Kin1 helix-α6 , differences are present in the charged residues compared to Kin3 ( described in more detail below ) and , perhaps as a consequence , the C-terminus of Kin1 helix-α6 is connected by less density to the MT surface compared to Kin3 ( Figure 3B , D , arrowhead ) . Thus , relatively limited conformational changes appear to accompany Mg-ADP release in the vicinity of helix-α6 and the neck linker . This is despite the previously described significant rearrangement of the switch loops at the nucleotide-binding site on the other side of the domain ( Figure 2 ) . However on Mg-ATP binding , a major conformational change of helix-α6 is observed in both motors ( Figure 3C , E; Figure 3—figure supplement 1 ) . Compared to the NN reconstructions , helix-α6 and β-sheet1abc have together lifted and rotated away from the MT surface . In the Mg-ATP-like reconstructions , a hydrophobic cavity forms above helix-α4 ( Kikkawa et al . , 2001 ) because the central β-sheet has peeled away from its C-terminal end ( see Figure 3C , E; and Figure 3—figure supplements 2 and 3C , D , F , G ) , helix-α6's C-terminus extends by a turn and inserts into this cavity . In the Kin3-Mg-ATP-like reconstruction , as a result of the repositioning of helix-α6 , only a narrow bridge of density connects its N-terminal end with α–tubulin ( Figure 3C , arrowhead ) . This N-terminal end is more negatively charged than the C-terminal end of helix-α6 that was in contact with the MT surface prior to Mg-ATP binding . In Kin1 , density for helix-α6 disconnects from the MT surface altogether ( Figure 3E , arrowhead ) . Importantly , in both motors , this structural reorganisation allows the neck linker to extend towards the MT plus end and dock along strand-β8 of the central β-sheet ( Figure 3C , E and Figure 3—figure supplement 3C , D , F , G ) ( Rice et al . , 1999 ) . The N-termini of both motors are also directed towards the MT plus end , lying across the docked neck linker to form the CNB ( Figure 3—figure supplement 3C , D , F , G and Figure 4C , E ) ( Hwang et al . , 2008; Khalil et al . , 2008 ) . Thus , concerted conformational changes involving a number of structural elements appear to contribute to movement of helix-α6 and neck linker docking . 10 . 7554/eLife . 03680 . 020Figure 4 . Nucleotide-independent interactions between the kinesin motor domain and the MT surface . ( A–C ) View from the MT plus end of the motor domain-MT interface in MT-bound Kin3 reconstructions ( shown as blue transparent density ) in ( A ) Mg-ADP , model shown in light blue , ( B ) no nucleotide ( NN ) , model shown in mid-blue , ( C ) Mg-ADPAlFx , model shown in dark blue , in which the CNB is formed between the neck linker ( fuchsia ) and N-terminus ( orange ) . The N-terminus of loop12 ( light pink ) extends helix-α4 by a turn but the central , lysine-rich portion of this loop is not visible ( dotted pink line ) , nor is the β-tubulin CTT ( arrowhead ) with which it is known to interact . Loop8/strand-β5 forms a clear connection to the MT surface ( arrow ) . ( D–E ) The same view of the motor domain-MT interface in MT-bound Kin1 reconstructions ( shown as green transparent density ) in ( D ) no nucleotide , model shown in light green , ( E ) Mg-AMPPNP , model shown in dark green , in which the CNB is formed between the neck linker ( fuchsia ) and N-terminus ( orange ) . The shorter Kin1 loop12 is clearly visualised and contacts the MT surface while loop8/strand-β5 are not connected by density to the MT surface ( arrow ) . In all reconstructions , density for the motor domain was contoured to an equivalent volume . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 02010 . 7554/eLife . 03680 . 021Figure 4—figure supplement 1 . Conserved conformations at the kinesin motor domain and the MT surface in Kin3 and Kin1 alternative ATP-like states . ( A ) View from the MT plus end of the motor domain-MT interface in the MT bound Kin3-Mg-AMPPNP ( blue transparent density and navy blue model ) . ( B ) View from the MT plus end of the motor domain-MT interface in the MT bound Kin1-Mg-ADPAlFx reconstruction ( green transparent density and olive green model ) . The major features are shared by all the ATP-like reconstructions: the CNB is formed between the neck linker ( fuchsia ) and N-terminus ( orange ) . The N-terminus of loop12 ( light pink ) extends helix-α4 by a turn but the central , lysine-rich portion of this loop is not visible ( dotted pink line ) , nor is the β-tubulin CTT ( arrowhead ) with which it is known to interact . Loop8/strand-β5 forms a clear connection to the MT surface ( arrow ) . The Kin1-Mg-ADPAlFx reconstruction is lower resolution ( FSCtrue , 0 . 143 = 7 . 7 ) , which may explain why residual density connects Loop8/strand-β5 and the MT surface , which is not the case in the Kin1-Mg-AMPPNP reconstruction ( Figure 4E ) . In all reconstructions , density for the motor domain was contoured to an equivalent volume . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 021 These analyses show that in both Kin1 and Kin3 , the same , small conformational changes at the nucleotide binding site on Mg-ATP binding have large structural consequences elsewhere . One important aspect of transmission of this mechanochemical information is that a stable interaction with the MT is sustained . Our data show that several structural elements form apparently invariant contacts with the MT ( primarily β-tubulin ) in all the nucleotide states we examined . In the Kin3 reconstructions , density corresponding to helix-α4 runs across the whole motor domain–MT interface ( Figure 4A–C ) . At its C-terminal end , density corresponding to the N-terminal portion of the extended Kin3 loop12 sequence is stabilised as a helical turn ( Figure 4A–C , pink ) . However , density corresponding to the middle Kin3-characteristic Lys-rich portion of this loop ( the so-called K-loop ) is not visible in any nucleotide state ( Figure 4A–C , pink dotted line ) . This suggests that this highly basic middle section of loop12 remains mobile even while close to the MT surface ( discussed below ) . The C-terminal end of Kin3 loop12 , on the other hand , is visible and is stabilised by interaction with β-tubulin . Loop12 leads into an interconnected region of contacts between the MT surface and the motor , composed of helix-α5 along with loop8/strand-β5 . These elements do not alter their interaction with the MT in the different nucleotide states calculated ( Figure 4A–C; Figure 4—figure supplement 1 ) . The Kin1 reconstructions show the same structural components at the motor domain–MT interface , which are also invariant in the different nucleotide states ( Figure 4D , E ) . In the Kin1 reconstructions—as with Kin3—helix-α4 forms a major contact at the tubulin intradimer interface and adopts a conserved orientation relative to the MT ( Figure 4D , E ) . However , the C-terminus of the Kin1 helix-α4 is shorter by one turn compared to Kin3 because its loop12 is shorter and also lacks the lysine cluster characteristic of Kin3s ( compare e . g . Figure 4B , D ) . Density corresponding to the Kin1 loop12 connects directly to helix-α5 at the MT interface ( Figure 4D , E; Figure 4—figure supplement 1 ) . However , in contrast to Kin3 , there is no density in our reconstructions connecting Kin1 loop8/strand-β5 and the MT surface ( Figure 4D , E ) . A key conformational change in the motor domain following Mg-ATP binding is peeling of the central β-sheet from the C-terminus of helix-α4 increasing their separation ( Figure 3—figure supplement 2 ) ; this is required to accommodate rotation of helix-α6 and consequent neck linker docking ( Figure 3B–E ) . Peeling of the central β-sheet has previously been proposed to arise from tilting of the entire motor domain relative to static MT contacts , pivoting around helix-α4 ( the so-called ‘seesaw’ model; Sindelar , 2011 ) . Specifically , this model predicts that the major difference in the motor before and after Mg-ATP binding would be the orientation of the motor domain with respect to helix-α4 ( Vale and Milligan , 2000 ) . Globally , the conformations of both Kin1 and Kin3 in our reconstructions are consistent with motor domain tilting of 12–15° on Mg-ATP binding ( Figure 3B–E , Figure 3—figure supplement 2 ) . In both motors , subtle flexure of the central β-sheet itself is also apparent on Mg-ATP binding ( Figure 5—figure supplement 1 ) such that loop7 and the bottom of strand-β3 that connects to the P-loop are not superimposable . Differences in the β-sheet when comparing the Kin3-Mg-ADP and Kin3-NN models are even smaller in comparison ( Figure 5—figure supplement 1A ) . In myosin , the equivalent structural region undergoes substantial β-sheet flexure on nucleotide release ( backbone RMSD > 3 . 2 Å , Figure 5—figure supplement 1D; Coureux et al . , 2003; Reubold et al . , 2003 ) . However , our data provide no evidence of significant flexing in the kinesin β-sheet that has been proposed to accompany Mg-ADP release ( Kull and Endow , 2013 ) . Furthermore , although the slight β-sheet bending that occurs when Mg-ATP binds may contribute to force generation as previously suggested ( Gigant et al . , 2013 ) , it cannot , by itself , account for the peeling of the β-sheet that allows neck linker docking . If motor domain tilt was sufficient to account for the mechanochemical transmission that takes place on Mg-ATP binding , superposition of the β-sheets of the NN and Mg-ATP structural states would be predicted to bring the motor domains into alignment ( apart from helix-α4 and the nucleotide-invariant MT contacts ) . However , such a superposition shows large residual differences in multiple regions of the motor domain ( Figure 5A , B; depicted as RMSDs between each pair of NN/Mg-ATP models ) . This clearly demonstrates that the β-sheet tilting that occurs in the transition from NN to Mg-ATP is not sufficient to describe the conformational changes in either Kin3 or Kin1 . This is further emphasized when the Kin3 and Kin1 NN pseudo-atomic models are superimposed on the β-sheets of their respective ATP-like docked models and compared to the Mg-ATP-like cryo-EM reconstructions ( Figure 5C , D ) . Various parts of the NN models protrude from the density for the ATP-like reconstructions illustrating the poor fit , agreeing with the RMSD calculations , and further supporting their tilt-independent movements ( Figure 5C , D compare to Figure 2C , E ) . At the nucleotide-binding site , this analysis highlights that movement of loop9 around the bound Mg-ATP is large compared to motor domain tilting . Similarly , while loop11 retains a similar conformation before and after Mg-ATP binding , it does not tilt along with the core β-sheet but instead moves towards the motor domain core ( see Figure 5—figure supplement 2 ) . In addition , helix-α2a and loop5 above the nucleotide-binding site , and helix-α0 below the nucleotide-binding site , accommodate Mg-ATP binding in both motors ( Figure 5A , B ) . Some structural changes are seen in helix-α1 , whereas the β-sheet1abc shows clear conformational differences; family-specific loop insertions in loop2 and loop3 particularly exaggerate these movements in Kin3 ( Figure 5C ) . The expected extension of helix-α6 and neck-linker docking is also highlighted by this analysis . However , it is also apparent that helix-α6 movement cannot be described purely by motor domain tilt , because it also undergoes a translational shift towards the MT plus end , as was recently proposed for Kin1 ( Gigant et al . , 2013 ) . The improved resolution of our reconstructions thus allows us to conclude that the conformational changes that underlie force generation in both Kin1 and Kin3 involve: ( 1 ) motor domain tilting relative to static MT contacts , but also ( 2 ) more complex sets of movements that accommodate Mg-ATP binding and bring about mechanical amplification . 10 . 7554/eLife . 03680 . 022Figure 5 . Transmission of force generation across the motor domain on Mg-ATP binding . ( A and B ) Conformational changes relative to superposition of the core β-sheet of Kin3 ( A ) and Kin1 ( B ) showing the RMSDs due to Mg-ATP binding coloured from yellow ( no change ) to pink ( large change ) , depicted on the Mg-ATP-like structures . Note , because the core β-sheet moves relative to helix-α4 , which is held at the MT interface , alignment of the β-sheet artificially shows large displacements of helix-α4 and other nucleotide-invariant MT contacts at the back of this view . ( C and D ) Comparison of the nucleotide-binding site before and after Mg-ATP binding in Kin3 ( C ) and Kin1 ( D ) . In each case , the NN model is depicted within the Mg-ATP cryo-EM density and shows that the regions of the largest RMSDs ( pink in panels A and B ) correspond to regions of the models that clearly do not fit in the density , that is , that undergo conformational changes when Mg-ATP binds . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 02210 . 7554/eLife . 03680 . 023Figure 5—figure supplement 1 . Limited β-sheet flexure during kinesin ATPase cycle compared to myosin5 . Superposition of the core β-sheets of motor domains in different nucleotide states reveals subtle differences at their edges , indicating β-sheet flexure at each transition . On the left of each panel , the core β-sheets of ( A ) Kin3-Mg-ADP-MT and Kin3-NN-MT , ( B ) Kin3-NN-MT and Kin3-Mg-ADPAlFx-MT , ( C ) Kin1-NN-MT and Kin1-Mg-AMPPNP-MT models are shown superimposed , viewed from the MT minus end . ( D ) For comparison Myosin5-NN ( PDB 1OE9 ) and Myosin5 Mg-ADP-BeFx ATP-like ( PDB 1W7J ) crystal structures are shown superimposed , where β-sheet flexure has been shown to occur ( Coureux et al . , 2003; Reubold et al . , 2003 ) . Arrowheads indicate the tip of loop7 and arrows indicate strand-β3 ( which connects to the P-loop ) , or the structurally equivalent region in the Myosin motor domain ( indicated with * ) . On the right of each panel , the corresponding RMSDs of each overlay are shown , displayed using a scale from 0 ( yellow ) to pink ( 3 . 2 Å ) . The motor domain MT minus end is to the left and plus end , that contains the flexible loop10 , to the right . ( A ) Kin3 Mg-ADP release: maximum loop7 RMSD ∼1 . 6 Å; ( B ) Kin3 Mg-ATP binding: loop7 , RMSD ∼2 . 5 Å , strand-β3: RMSD ∼1 . 7 Å; ( C ) Kin1 Mg-ATP binding: loop7 , RMSD ∼1 . 8 Å , strand-β3: RMSD ∼1 . 2 Å; ( D ) Myosin5 Mg-ADP release: loop7* maximum RMSD ∼3 . 3 Å ( Coureux et al . , 2003; Reubold et al . , 2003 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 02310 . 7554/eLife . 03680 . 024Figure 5—figure supplement 2 . Pincer-like closure of loop9 and loop11 contributes to motor domain tilt when ATP binds . ( A ) MT binding and Mg-ADP release in the Kin3-NN-MT , viewed from the MT minus end , induce an ordered loop9 and loop11 conformation; ( B ) ATP-binding induces loop9 and loop11 to move together contributing to motor domain tilting towards the bound nucleotide , thereby enabling neck linker docking . ( C and D ) The same conformational changes are seen in Kin1 . Red and yellow arrows represent the ‘pincer’-like movement of loop9 and loop11 towards each other that produces the new density connection between them . Tilting of the motor domains relative to helix-α4 is indicated with orange-curved arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 024 Despite high structural and mechanistic similarity between Kin3 and Kin1 , contacts across the motor domain–MT interface are likely to contribute to differences in these motors' transport properties ( Figure 6 ) . One major difference is the presence of a Lys-rich insertion in Kin3 loop12 ( the ‘K-loop’ ) ( Figure 6A , pink shading ) ( Okada and Hirokawa , 1999 ) . In Kin3s , loop12 mediates 1D diffusion of ADP-bound monomeric and dimeric Kin3s along MTs via flexible , electrostatic interactions with the acidic C-terminal tails ( CTTs ) of tubulin ( Okada and Hirokawa , 1999 , 2000; Kikkawa et al . , 2000; Soppina et al . , 2014 ) . The K-loop also enhances the initial interaction between Kin3 dimers and their track prior to processive stepping ( Soppina and Verhey , 2014 ) . In addition , whereas the catalytic turnover of Kin3 compared to Kin1 monomers are similar ( our data in Table 3 and e . g . , Okada and Hirokawa , 2000 ) , steady state ATPase assays show that the KmMT of Kin3 is several hundred times lower than Kin1 , a difference that depends partly on the K-loop ( Okada and Hirokawa , 2000 ) . Since the KmMT is indicative of the MT affinity of ADP-bound kinesin ( Woehlke et al . , 1997 ) , this is consistent with the role of the Kin3 loop12 in enhancing the association of Mg-ADP Kin3s with MTs ( Okada and Hirokawa , 1999 , 2000; Kikkawa et al . , 2000; Soppina and Verhey , 2014 ) . 10 . 7554/eLife . 03680 . 019Figure 6 . Comparison of Kin3 and Kin1 . ( A ) Sequence alignment of Kin3 ( Kif1A ) and Kin1 ( Kif5A ) motor domains showing secondary structural elements within the domains , annotated according to sequence and charge conservation . Elements depicted in other panels are underlined . ( B ) Longitudinal slice through the Kin3-NN model viewed from the front showing the MT contact elements and the underlying structural regions in αβ-tubulin . ( C ) MT binding surface of Kin3-NN model viewed from the MT surface ( 180° rotated compared to B ) annotated by sequence identity ( black ) between Kin3 and Kin1 and sequence insertions ( green ) . Structural elements in the MT are removed in this view to most clearly show elements in the motor domain . ( D ) MT binding surface of Kin3-NN model showing the differences in charge ( blue: Kin3 more acidic than Kin1; red: Kin3 more basic than Kin1 ) ; same view as in C . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 019 There is no density corresponding to the K-loop—nor of the tubulin CTTs with which it is proposed to interact—in any of our Kin3 reconstructions ( Figure 4A–C ) . Given that density corresponding to Kin1 loop12 ( Figure 4D , E ) and Kin3 loops of equivalent size ( e . g . loops 2 and 3 [7 and 8 residues respectively] , Figure 3A–C ) are clearly visualised , this suggests that this region of Kin3 is structurally heterogeneous , and therefore invisible in the context of our averaging methods . The K-loop may be intrinsically flexible due to its sequence , consistent with its role in mediating 1D diffusion . In addition , the lack of structural detail in this region could be due to the biochemical heterogeneity ( different isoforms and post-translational modifications ) of the CTTs of the bovine tubulin used in our experiments . Our structures imply that conformational flexibility of the K-loop persists throughout the motor's ATPase cycle but more information from future experiments is needed to clarify the contribution of this region to motor function . However , the K-loop is reported to account for only a 10-fold enhancement of MT association of monomeric Kin3s over Kin1s ( Okada and Hirokawa , 1999 , 2000 ) , implying that other regions of the Kin3 motor domain also contribute . Our data show clear structural differences between Kin1 and Kin3 at the interface of the acidic tip of α-tubulin H12 with helix-α6 , especially in the Mg-ADP/NN reconstructions ( Figure 3 ) . In addition , more subtle differences in the distribution of charged residues in loop11 and helix α4's N-terminus would be predicted to influence MT affinity ( Figure 6D ) . Sequence divergence in loop8/strand-β5 was previously proposed to enable discrimination of post-translational modification in α-tubulin CTTs by Kin3 compared to Kin1 ( Konishi and Setou , 2009 ) . A direct role for recognition of the α-tubulin CTT is unlikely given its distance from loop8/strand-β5 . However , differences in connectivity between this region of the motor domain and β-tubulin when comparing Kin1 and Kin3 ( Figure 4 ) could contribute to differences in their apparent overall affinity . Intriguingly , recent data show that the K-loop does not contribute to the super-processive stepping properties of Kin3 dimers ( Soppina and Verhey , 2014 ) . Although a number of motor parameters could in principle contribute to processivity ( e . g . , coordination between dimer motor domains via the NL [Clancy et al . , 2011] ) , our structures suggest that other regions of the Kin3–MT interface may also influence functional differentiation of these motors including super-processivity ( Figure 6C , D ) .
Kinesin mechanochemistry and the extent of mechanistic conservation within the motor superfamily are open questions , critical to explain how MT binding , and ATP binding and hydrolysis drive motor activity . Our structural characterisation of two transport motors now allows us to propose a model that describes the roles of mechanochemical elements that together drive conserved MT-based motor function ( Figure 7 ) . 10 . 7554/eLife . 03680 . 025Figure 7 . Model of conserved MT-bound kinesin mechanochemistry . Loop11/N-terminus of helix-α4 is flexible in ADP-bound kinesin in solution , the neck linker is also flexible while loop9 chelates ADP . MT binding is sensed by loop11/helix-α4 N-terminus , biasing them towards more ordered conformations . We propose that this favours crosstalk between loop11 and loop9 , stimulating ADP release . In the NN conformation , both loop11 and loop9 are well ordered and primed to favour ATP binding , while helix-α6—which is required for mechanical amplification–is closely associated with the MT on the other side of the motor domain . ATP binding draws loop11 and loop9 closer together; causing ( 1 ) tilting of most of the motor domain not contacting the MT towards the nucleotide-binding site , ( 2 ) rotation , translation , and extension of helix-α6 which we propose contributes to force generation , and ( 3 ) allows neck linker docking and biases movement of the 2nd head towards the MT plus end . DOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 025 In the Mg-ADP-bound kinesin , association with the MT surface is experienced directly by loop11 and the N-terminus of helix-α4 biasing their conformations towards more structured states . Full stabilisation of these elements is not achieved until Mg-ADP is released , and the additional contacts with the MT surface may in particular serve to nucleate the single turn helix in loop11 . This is consistent with the well-documented role of loop11 in sensing MT attachment and triggering Mg-ADP release via interactions with α-tubulin ( Woehlke et al . , 1997; Yun et al . , 2001; Ebbing et al . , 2008; Uchimura et al . , 2010 ) . Loop9 does not directly contact the MT before or after Mg-ADP release but dramatically changes conformation , unfurling , and extending around the nucleotide-binding site . The structured conformations of loop11 and the N-terminus of helix-α4 are sterically compatible with the conformations of loop9 before and after Mg-ADP release—that is , no clashes are seen in either case . However , the extended conformation of loop9 and the ordered conformations of helix-α4/loop11 are likely to be mutually stabilising due to formation of additional contacts , and thereby mediate communication between the nucleotide and MT-binding sites ( Woehlke et al . , 1997; Yun et al . , 2001; Farrell et al . , 2002; Ebbing et al . , 2008; Nitta et al . , 2008 ) . Critically , however , the water network coordinating Mg-ADP is stabilized exclusively by the retracted helical conformation of loop9 ( Figure 2—figure supplement 2 ) . The transition towards the extended conformation of loop9 promotes Mg-ADP release by destabilisation of Mg coordination ( Nitta et al . , 2008 ) . These structural rearrangements therefore indicate that sequential conformational changes of the switch loops in the presence of MTs stimulate Mg-ADP release , the rate-limiting step of motors in solution ( Hackney , 1988 ) . These rearrangements allow formation of a nucleotide-free motor that is strongly bound to its MT track ( Nakata and Hirokawa , 1995 ) , at least in part due to additional contacts formed between loop11 and the MT . Conformational changes at the nucleotide-binding site that lead to Mg-ADP release also appear to prime the kinesin motor domain for Mg-ATP binding . However , the primed conformation clearly does not lead to neck linker docking in the absence of Mg-ATP , contrary to previous predictions ( Nitta et al . , 2008 ) . Multiple strands of evidence suggest that the neck linkers of transport kinesins in solution explore both docked and undocked conformations independent of the nucleotide state ( Rice et al . , 1999; Nitta et al . , 2008; Scarabelli and Grant , 2013 ) . Thus , tight MT binding is critical in strongly biasing neck linker conformation in the absence of nucleotide such that it will be undocked and , in our reconstructions , directed albeit flexibly towards the MT minus end . Interaction of helix-α6 with α-tubulin's H12 ( Uchimura et al . , 2010 ) may therefore help to prevent neck linker docking in the absence of nucleotide , despite changes in the conformations of the switch loops at the active site . Mg-ATP binding does not cause large rearrangements of the nucleotide-binding site of MT-bound motor domains . However , the presence of the pre-hydrolysis γ-phosphate of Mg-ATP is critical for the pincer-like movement of loop11 and loop9 towards each other . Along with formation of strong additional contacts between these loops , the helix-α4 N-terminus and the P-loop ( see Figure 2—figure supplement 4 and Parke et al . , 2010; Chang et al . , 2013; Gigant et al . , 2013 ) , this new local connectivity induces the larger rearrangements that cause neck linker docking . The resulting conformational changes cannot be described only as a tilt of the motor domain relative to static contacts with the MT including helix-α4; in addition to β-sheet tilting , multiple changes across the domain reinforce mechanical amplification and neck linker docking when Mg-ATP binds . The resolution of our reconstructions also allows us to detect subtle distortion of the central β-sheet edges on Mg-ATP binding . However , arguably the most important consequences of Mg-ATP binding are the changes—extension , tilting , and translation—in helix-α6 that allow neck linker docking . This conformation is stabilised by contacts between its N-terminus and elements in the nucleotide-binding pocket ( see Figure 2—figure supplement 4 and Parke et al . , 2010; Chang et al . , 2013; Gigant et al . , 2013 ) . Neck linker docking is essential for both defining the directionality of kinesin motility and mediating head–head tension to ensure processive dimer stepping ( Rice et al . , 1999; Tomishige and Vale , 2000; Vale and Milligan , 2000; Skiniotis et al . , 2003; Clancy et al . , 2011; Sindelar , 2011 ) , but whether docking itself can generate the force required for kinesin stepping has been questioned ( Rice et al . , 2003 ) . Thus , the structural basis of ATP-dependent force generation remains a matter of debate in the field ( Visscher et al . , 1999; Cross and McAinsh , 2014 ) . The conformational changes associated with helix-α6 during the ATPase cycle—in which contacts with the MT formed in the ADP/NN state are broken as Mg-ATP-dependent rotation pulls it away from the MT surface—reinforce neck linker movements and may also contribute to mechanical amplification and force generation . The translation/extension of helix-α6 into the hydrophobic cavity that is created by β-sheet tilting when Mg-ATP binds may ensure that this tilting is not reversed . Intriguingly , mutagenesis of residues at the helix-α6/neck linker junction has a profound effect on the activity of kinesin monomers ( Case et al . , 2000 ) , pointing to the importance and likely conservation of structural transitions in this region ( Case et al . , 1997 ) . Importantly , movement of helix-α6 also relieves steric blocking of neck linker docking and presumably biases the mobile neck linker trajectory . In collaboration with the motor N-terminus , formation of the CNB reinforces the plus end directionality of this bias . Thus , we propose that the helix-α6 is a key mechanical element within the kinesin motor domain , and that its Mg-ATP-dependent movement is essential to plus-end directed stepping . Once the neck linker has docked , ATP hydrolysis occurs , ensuring efficient coupling between kinesin stepping , Mg-ATP binding and hydrolysis ( Schnitzer et al . , 2000; Hahlen et al . , 2006 ) . A detailed reaction mechanism for hydrolysis has been proposed based on the conformations of loop9 and loop11 ( a so-called ‘phosphate tube’ ) with Mg-ATP-analogue bound ( Parke et al . , 2010 ) . Consistent with MT binding being important in the catalytic enhancement of kinesins ( Ma and Taylor , 1997 ) , this hydrolysis competent configuration of the switch loops is rarely seen in Mg-ATP-analogue kinesin structures in the absence of MTs ( e . g . , Kikkawa et al . , 2001; Nitta et al . , 2004; Cochran et al . , 2009 , with Parke et al . , 2010; Chang et al . , 2013 being the notable exceptions ) ; those in complex with tubulin always adopt this configuration ( Sindelar and Downing , 2010; Goulet et al . , 2012; Gigant et al . , 2013 ) . On Mg-ADP release , loop9 and loop11 are stabilized into conformations quite close to catalytically competent ones . This suggests that the conformational changes triggered by MT binding that lead to MT-stimulated ADP release also contribute to setting up the catalytic site for ATP hydrolysis . Thus , a subset of mutations in MT-sensing residues in loop11 or which decouple MT affinity and ADP-release also affect MT-stimulated ATP-hydrolysis ( Woehlke et al . , 1997; Song and Endow , 1998; Yun et al . , 2001; Ebbing et al . , 2008; Uchimura et al . , 2010 ) . Following hydrolysis and phosphate release , we would predict that the Mg-ADP remaining in the catalytic site causes retraction of loop9 , subsequent destabilization of loop11 and the helix-α4 N-terminus , leading to track detachment . This model allows several previously proposed hypotheses , in particular concerning MT-stimulated Mg-ADP release , to be excluded . Mechanisms that involve MT-induced ‘opening’ of the nucleotide pocket , disordering of the switch loops around the nucleotide pocket to destabilise Mg-ADP coordination , or in which loop9 extends into the nucleotide pocket to perturb the P-loop and eject Mg-ADP ( Yun et al . , 2001; Kikkawa and Hirokawa , 2006; Sindelar and Downing , 2007; Nitta et al . , 2008; Sindelar , 2011 ) are not supported by our observations that: ( 1 ) both loop9 and loop11 move towards the nucleotide-binding pocket on Mg-ADP release , ( 2 ) these loops adopt well-defined and conserved conformations that are clearly visualised after Mg-ADP release , and ( 3 ) the conformation of these loops does not sterically interfere with nucleotide binding or disrupt the P-loop . Another prominent idea is that a significant twist of the core β-sheet caused by MT attachment would promote Mg-ADP release analogous to the equivalent release step in myosin ( Coureux et al . , 2003; Hirose et al . , 2006; Kull and Endow , 2013 ) . However , comparison of our Kin3-Mg-ADP and Kin3-NN reconstructions ( Figure 5—figure supplement 1A ) does not support β-sheet twist as a mechanism for Mg-ADP release in kinesins . The structural elements involved in these mechanochemical transitions are extremely well conserved amongst kinesins , and it is likely that the mechanisms we describe are utilised by all superfamily members . We previously characterised the MT-bound ATPase cycle of human kinesin-5 ( Kin5 , Goulet et al . , 2012 , 2014 ) . Although the resolutions of those cryo-EM reconstructions ( ∼10 Å ) do not provide the level of detail of the current work , many of our current hypotheses are consistent with a conserved mechanochemistry , specifically conformational coupling of loops9 and 11 to bring about MT-induced Mg-ADP release and Mg-ATP induced neck linker docking . Superimposed on this conserved mechanochemistry , family-specific modifications were also detected; most strikingly for Kin5 , these include the proposed role of the Kin5-extended loop5 in controlling nucleotide binding and the stiffer properties of the Kin5 neck linker that undergoes an order-to-order transition on Mg-ATP binding . Family-specific insertions elsewhere in the motor domain are likely to have other modifying roles , such as Kin3's loop12 , which enhances the initial interaction between these highly processive motors and their tracks ( Soppina and Verhey , 2014 ) . A tantalising hint of how insertions in loop2 may be coupled to MT depolymerisation in for example kinesin-13s ( Desai et al . , 1999; Moores et al . , 2002; Asenjo et al . , 2013 ) and kinesin-8s ( Varga et al . , 2006; Peters et al . , 2010 ) is provided by its proximity to the MT surface and the mechanical amplifier helix-α6 , and by its large displacement on Mg-ATP binding . Future studies at high resolution will provide further insights into the ways this conserved mechanochemistry is modified in diverse functional contexts within the kinesin superfamily .
A human kinesin-1 ( Kin1 ) construct ( Kif5A , residues 1–340 , in pET151-D-TOPO [Invitrogen , Carlsbad , CA with a TEV protease-cleavable N-terminal His6-tag] ) was expressed recombinantly in Escherichia coli and purified using cobalt affinity chromatography . The His6-tag was removed by cleavage with TEV protease , and the untagged protein was buffer exchanged into BrB20 buffer ( 20 mM PIPES , 2 mM MgCl2 , 1 mM EGTA , 2 mM DTT , pH 6 . 8 ) . A human kinesin-3 ( Kin3 ) construct ( Kif1A , residues 1–361 , in pFN18a ( with a TEV protease-cleavable N-terminal Halo-tag and a C-terminal His6-tag [a kind gift from Prof Christopher A Walsh's laboratory , Harvard Medical School] ) was expressed recombinantly in E . coli and purified using nickel affinity chromatography and size exclusion chromatography ( GE Healthcare Life Science , UK , Superdex 75 ) . The N-terminal Halo-tag was removed by cleavage with TEV protease , the sample was dialyzed into storage buffer ( 20 mM HEPES , pH 7 , 150 mM NaCl , 1 mM TCEP , 5 mM MgCl2 , and 0 . 1 mM ADP ) and concentrated . Note that this construct contains the native Kin3 ( Kif1A ) sequence , as opposed to several previous studies where a chimeric protein with substitution of its neck linker with that of the kinesin-1 Kif5C ( Kikkawa et al . , 2001; Nitta et al . , 2004; Kikkawa and Hirokawa , 2006; Nitta et al . , 2008 ) . The steady-state MT-activated ATPase activities of our motor constructs were determined by measuring phosphate production with a commercially available kit ( EnzChek , Molecular Probes , Eugene , OR ) . Assays contained 10 nM motor domain and a minimum of fourfold molar excess of paclitaxel-stabilised MTs in 50 mM K-acetate , 25 mM HEPES , 5 mM Mg-acetate , 1 mM EGTA , pH 7 . 5 at 20°C . The dependence of rates of inorganic phosphate production on [MT] and [ATP] was fitted with a Michaelis–Menten relationship ( Table 3 ) . Bovine tubulin ( Cytoskeleton Inc , Denver , CO ) at a final concentration of 50 μM in MT polymerization buffer ( 100 mM MES , pH 6 . 5 , 1 mM MgCl2 , 1 mM EGTA , 1 mM DTT , 5 mM GTP ) was polymerized at 37°C for 1 hr . 1 mM paclitaxel ( Calbiochem , San Diego , CA ) in DMSO was then added , and the sample was incubated at 37°C for a further hour . MTs were diluted in BrB20 to a final concentration of 5 μM . Kin1 and Kin3 were diluted in BrB20 containing either 2 mM of AMPPNP , ADP , ADP + AlF4 , or apyrase ( 10 units/ml ) , according to established protocols ( Hirose and Amos , 2007; Sindelar and Downing , 2007 , 2010; Fourniol and Moores , 2011 ) , and warmed to room temperature 10 min prior to complex formation . The final concentrations used to visually achieve full decoration in the various nucleotide states are shown in Table 4 . C-flat holey carbon grids ( Protochips , Raleigh , NC ) with 2 μm holes and 4 μm spacing were glow-discharged in air . 4 μl drops of MT then Kin1 or Kin3 samples were added and blotted in sequential fashion using a Vitrobot plunge-freezing device ( FEI Co . , Hillsboro , OR ) operating at 25°C and 100% humidity and vitrified in liquid ethane . 10 . 7554/eLife . 03680 . 026Table 4 . Final protein concentrations used for cryo-EM sample preparationDOI: http://dx . doi . org/10 . 7554/eLife . 03680 . 026Kinesin and nucleotide state[MT] ( μM ) [Motor domain] ( μM ) Kin3 MgADP510Kin3 NN55Kin3 Mg-AMPPNP55Kin3 Mg-ADP . AlFx55Kin1 NN5100Kin1 Mg-AMPPNP550Kin1 Mg-ADP . AlFx550Kin1 samples required higher concentrations than Kin3 to achieve good MT occupancy . Images of MT-kinesin complexes were collected using a 4k × 4k CCD camera ( Gatan Inc . , Pleasanton , CA ) on a FEI Tecnai G2 Polara operating at 300 kV with a calibrated magnification of 100 , 000× and a final sampling of 1 . 5 Å/pixel . A defocus range of 0 . 4–3 . 5 μm and an electron dose of ∼20 e−/Å2 were used . Images were screened manually to remove those with drift and/or objective astigmatism , contamination , and not containing at least one fully decorated and straight 13 protofilament MT . Kinesin-decorated straight 13 protofilament MT segments were manually boxed using Eman suite's Boxer ( Ludtke et al . , 1999 ) and input to a set of custom-designed semi-automated single-particle processing scripts using Spider ( Frank et al . , 1996 ) and Frealign ( Grigorieff , 2007 ) as described previously ( Sindelar and Downing , 2007 , 2010 ) , with minor modifications during local refinement . The phi-angle and thus seam location is determined in pseudo-symmetrical 13 protofilament MTs using projection matching in Spider ( Frank et al . , 1996 ) . Once approximate alignment parameters are determined and manually verified ( based on known values for the MT lattice ) , local refinement and CTF correction is performed in Frealign ( Grigorieff , 2007 ) . Eight rounds of refinement were undertaken and a negative B-factor of −400 was applied to the output reconstruction of round five to escape local minima in the search space; no B-factor was applied in the following three rounds to reduce possible over-fitting ( http://grigoriefflab . janelia . org/forum ) . The angular distribution was isotropic for all data sets and the final reconstructions of the asymmetric unit ( αβ-tubulin heterodimer + kinesin motor domain ) were generated using 13 protofilament MT pseudo-symmetry . All final maps were assessed for possible over-fitting during refinement using a high-resolution noise-substitution test ( Chen et al . , 2013 ) . Final estimated resolutions for each reconstruction are reported in Table 1 and FSC curves are shown in Figure 1—figure supplement 1 . Band-pass filtering of these reconstructions using a Fermi temperature of 0 . 04 was performed in Spider ( Frank et al . , 1996 ) between frequencies of 15–6 Å ( except for K1 Mg-ADPAlFx-MT reconstruction , where 15–7 Å was used ) . 50 initial atomic models of each motor domain ( in each nucleotide state ) were built using Modeller v9 . 12 ( Sali and Blundell , 1993 ) based on multiple template structures ( see Table 2 ) . Initial fitting of each model into the respective maps was done using the Chimera fit_in_map tool ( Goddard et al . , 2007 ) . The best model was selected based on a combination of the cross correlation coefficient ( CCC ) between each model and the density map and a statistical potentials score ( zDOPE; Shen and Sali , 2006 ) . Each map was box-segmented around the motor domain , and the EM density for the tubulin was masked out ( using Chimera volume eraser tool ) . The best fits were further refined with Flex-EM following a multistep optimisation protocol relying on simulated annealing molecular dynamics and a conjugate-gradients minimization applied to a series of subdivisions of the structure into rigid bodies ( Topf et al . , 2008 ) as identified by RIBFIND ( Table 2; Pandurangan and Topf , 2012 ) . In order to analyse subtle conformational changes occurring in various regions of the domain in the different nucleotide states , the quality of the final fits was assessed locally with TEMPy ( Farabella et al . , Unpublished ) using the segment based cross-correlation coefficient ( SCCC , Figure 1—figure supplement 2 ) ( Pandurangan et al . , 2014 ) . | The interior of a cell is a hive of activity , filled with proteins and other items moving from one location to another . A network of filaments called microtubules forms tracks along which so-called motor proteins carry these items . Kinesins are one group of motor proteins , and a typical kinesin protein has one end ( called the ‘motor domain’ ) that can attach itself to the microtubules . The other end links to the cargo being carried , and a ‘neck’ connects the two . When two of these proteins work together , flexible regions of the neck allow the two motor domains to move past one another , which enable the kinesin to essentially walk along a microtubule in a stepwise manner . To take these steps along microtubules , each kinesin motor domain in the pair must undergo alternating cycles of tight association and release from their tracks . This cycle is coordinated by binding and breaking down a molecule called ATP , which also provides the energy needed to take the next step . How the cycle of loose and tight microtubule attachment is coordinated with the release of the breakdown products of ATP , and how the energy from the ATP molecule is converted into the force that moves the motor along the microtubule , has been unclear . Atherton et al . use a technique called cryo-electron microscopy to study—in more detail than previously seen—the structure of the motor domains of two types of kinesin called kinesin-1 and kinesin-3 . Images were taken at different stages of the cycle used by the motor domains to extract the energy from ATP molecules . Although the two kinesins have been thought to move along the microtubule tracks in different ways , Atherton et al . find that the core mechanism used by their motor domains is the same . When a motor domain binds to the microtubule , its shape changes , first stimulating release of the breakdown products of ATP from the previous cycle . This release makes room for a new ATP molecule to bind . The structural changes caused by ATP binding are relatively small but produce larger changes in the flexible neck region that enable individual motor domains within a kinesin pair to co-ordinate their movement and move in a consistent direction . This mechanism involves tight coupling between track binding and fuel usage and makes kinesins highly efficient motors . The structures uncovered by Atherton et al . reveal a mechanism that links microtubule binding , the energy supplied to the motor domain and the force that moves the kinesin along a microtubule . Future work will clarify whether the key features observed in the motor domains of kinesin-1 and kinesin-3 are also found in other types of kinesin motors . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Conserved mechanisms of microtubule-stimulated ADP release, ATP binding, and force generation in transport kinesins |
Resveratrol has beneficial effects on aging , inflammation and metabolism , which are thought to result from activation of the lysine deacetylase , sirtuin 1 ( SIRT1 ) , the cAMP pathway , or AMP-activated protein kinase . In this study , we report that resveratrol acts as a pathway-selective estrogen receptor-α ( ERα ) ligand to modulate the inflammatory response but not cell proliferation . A crystal structure of the ERα ligand-binding domain ( LBD ) as a complex with resveratrol revealed a unique perturbation of the coactivator-binding surface , consistent with an altered coregulator recruitment profile . Gene expression analyses revealed significant overlap of TNFα genes modulated by resveratrol and estradiol . Furthermore , the ability of resveratrol to suppress interleukin-6 transcription was shown to require ERα and several ERα coregulators , suggesting that ERα functions as a primary conduit for resveratrol activity .
Many beneficial effects on human health have been described for resveratrol ( ( E ) -5- ( p-hydroxystyryl ) resorcinol ) , including prevention of skin and colorectal cancer , protection from metabolic and cardiovascular disease , neuroprotection , and general anti-inflammatory effects . Efficacy associated with resveratrol use has been attributed to activation of the lysine deacetylase , Sirtiun 1 ( SIRT1 ) ( Baur and Sinclair , 2006 ) , the cAMP pathway , or the AMP-activated protein kinase ( AMPK ) ( Park et al . , 2012; Price et al . , 2012; Tennen et al . , 2012 ) . Resveratrol is also a phytoestrogen that modulates estrogen receptor ( ER ) -mediated transcription ( Gehm et al . , 1997; Bowers et al . , 2000 ) , though only a small percent of published papers consider ER as a potential mediator of the complex pharmacology of resveratrol . The estrogenic role of resveratrol is important because a variety of resveratrol-sensitive tissues are ER-positive , and the two ER subtypes in mammals , ERα and ERβ , exhibit different tissue-specific expression profiles ( Bookout et al . , 2006 ) . Specifically , effects of resveratrol on ER include anti-inflammatory effects such as protection from trauma-hemorrhage-induced injury and suppression of Interleukin-6 ( IL-6 ) expression in the liver , intestine , and cardiovascular system ( Yu et al . , 2008 , 2010 , 2011b ) . However , in contrast to other ERα agonist , resveratrol does not induce proliferation of mammary or uterine tissues ( Turner et al . , 1999 ) , allowing it to be taken as a dietary supplement . The structural and molecular mechanisms for this pathway selective signaling are not known . The roles of resveratrol as a stimulant of SIRT1 and ER signaling have been presented as distinct mechanisms . However , dissection of these mechanisms of action is complicated by physical and functional interactions between ERα and SIRT1 , where: ( i ) ERα is a SIRT1 substrate ( Kim et al . , 2006; Ji Yu et al . , 2011 ) , and ( ii ) SIRT1 functions as an ER coregulator required for the oncogenic effects of estrogens in breast cancer ( Elangovan et al . , 2011 ) . Further , SIRT1 also deacetylates NF-κB subunits to inhibit expression of inflammatory genes ( Rothgiesser et al . , 2010 ) , and ERα also inhibits NF-κB signaling ( Cvoro et al . , 2006; Nettles et al . , 2008a , 2008b; Saijo et al . , 2011; Srinivasan et al . , 2013 ) . Thus , understanding the anti-inflammatory actions of resveratrol requires careful dissection of its ER-mediated vs non ER-mediated effects , and the role of SIRT1 . ER activates transcription in response to estradiol ( E2 ) , and a wide cast of other estrogenic compounds , including steroids , phytoestrogens , and environmental estrogens , by either direct binding to DNA , or tethering to DNA-bound transcription factors ( Cicatiello et al . , 2004; DeNardo et al . , 2007 ) . Transactivation via direct binding of ER to estrogen response elements ( EREs ) has been well studied , and it involves ER-mediated recruitment of transcriptional coregulators , including coactivators and corepressors ( Shang et al . , 2000; Metivier et al . , 2003 ) . These coregulators remodel chromatin , regulate post-translational modification ( PTM ) of histones and non-histone substrates , and control assembly of transcription-initiation and transcription-elongation complexes at target gene promoters ( Bulynko and O'Malley , 2010; Perissi et al . , 2010 ) . Coregulator function in ER-mediated transcription is consistent with a hit-and-run model where one coregulator complex lays down PTMs and changes the chromatin and coregulator environment so as to increase affinity for the next coregulator complex ( Shang et al . , 2000; Fletcher et al . , 2002; Metivier et al . , 2003 ) . In contrast , ER-mediated repression of inflammatory genes has been less extensively studied . ER represses transcription through a tethering mechanism called transrepression , via interaction with NF-κB and activator protein-1 complexes . Only a few key coregulators involved in this process have been identified ( Cvoro et al . , 2006; Yu et al . , 2007; Nettles et al . , 2008b; Saijo et al . , 2011 ) . Moreover , the mechanism through which resveratrol modulates the inflammatory response is poorly understood . In a screen for ERα ligands that inhibit IL-6 production , we found that resveratrol was among the most efficacious ( Srinivasan et al . , 2013 ) , prompting us to explore this mechanism . To address the question of how resveratrol regulates IL-6 without stimulating proliferation , we examined the roles of ERα , SIRT1 , and a cast of coregulators . Resveratrol inhibited IL-6 expression via ERα , which was recruited to the IL-6 promoter where it altered the recruitment profile of coregulators , including SIRT1 , and reduced acetylation of p65 NF-κB , which is required for transcriptional activation . Unexpectedly , there was a marked diversity of coregulators required for signal integration , where many display distinct roles in TNFα vs ERα signaling .
Resveratrol , which has a non-steroidal chemical structure ( Figure 1A ) , profiled as a partial agonist in ER-positive MCF-7 breast cancer cells , stimulating 3xERE-luciferase reporter activity with about 30% efficacy relative to E2 ( Figure 1B ) . To assess the effect of resveratrol on MCF-7 cell proliferation , cells in steroid-depleted media were treated for 7 days with several ER ligands including resveratrol . Unlike E2 , resveratrol did not stimulate cell proliferation ( Figure 1C ) . 10 . 7554/eLife . 02057 . 003Figure 1 . Effects of resveratrol on the canonical ERα proliferative pathway . ( A ) Chemical structures of E2 and resveratrol . ( B ) Luciferase assay of MCF-7 cells transfected with 3xERE-luciferase reporter and stimulated with 10 nM E2 or 10 μM resveratrol . ( C ) Steroid-deprived MCF-7 cells were treated with 10 nM E2 , 10 μM 4-hydroxytamoxifen ( TAM ) , 10 μM ICI182 , 780 , or 10 μM resveratrol . After 7 days , cell number was determined with a standard curve . ( D ) Mammalian two-hybrid assays with ERα and the coactivators SRC1-3 . HEK293 cells were transfected with Gal4 SRC1-3 fusions , ERα-VP-16 , and the 5xUAS-luciferase reporter for 24 hr . Cells were treated with 10 nM E2 , 10 μM TAM , 10 μM ICI , or 10 μM resveratrol for 24 hr and processed for luciferase activity . Data are presented in panels B–D as mean ± SEM . ( E ) Resveratrol-induced recruitment of ERα to the GREB1 promoter . Occupancy of GREB1 by ERα was compared by ChIP assay in MCF-7 cells that were steroid deprived for 3 days , treated with 10 nM E2 or 10 μM resveratrol , and fixed after 0 , 15 , 30 , or 45 min ( mean ± s . e . m . n = 2 ) . ( F ) Resveratrol reduced SRC3 but not SRC2 recruitment at the GREB1 promoter . Occupancy of GREB1 by SRC2 and SRC3 were examined by ChIP assay in MCF-7 cells treated as described in panel A . Average promoter occupancies are shown as fold changes ( mean ± SEM n = 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 003 Steroid receptor coactivators , SRC1 , SRC2 , and SRC3 , are primary mediators of ERα activity , and they provide a scaffold for recruitment of other coregulators such as p300 and CBP ( Chen et al . , 2000; Wong et al . , 2001; Huang and Cheng , 2004 ) . Despite their overlapping functions , SRCs play disparate roles in normal mammary gland development , with SRC3 , and to some extent SRC1 , contributing to growth ( Xu and Li , 2003 ) . In MCF-7 cells , SRC3 is selectively required for E2-induced proliferation ( Karmakar et al . , 2009 ) . When compared to E2 in a mammalian two-hybrid assay , resveratrol induced full association of ERα with SRC2 , but reduced interaction with SRC1 or SRC3 in a mammalian two-hybrid assay ( Figure 1D ) , which we propose does is not a sufficient interaction to support the proliferative response . This idea is further supported by chromatin immunoprecipitation ( ChIP ) assays examining recruitment of these factors to a canonical ERα binding site in the GREB1 gene , a gene required for estrogen-induced cell proliferation ( Rae et al . , 2005; Sun et al . , 2007 ) . We found that resveratrol induced less SRC3 recruitment than was observed upon E2 treatment , but induced comparable SRC2 and ERα recruitment ( Figure 1E , F ) . Thus , the lack of proliferative signal is consistent with ligand-selective coregulator recruitment by resveratrol-bound ERα and the disparate roles of the SRCs in the proliferative response . Together with anti-inflammatory effects described below , these results indicate that resveratrol acts as a pathway-selective ERα agonist . ERα coordinates a wide range of physiologic events outside of reproductive tissues , including modulation of brain function , cardiovascular and bone health , metabolic functions in the liver and muscle , remodeling of the immune system , and coordination of the inflammatory response in ERα-target tissues ( Nilsson et al . , 2011 ) . TNFα or Toll-like receptor agonists such as lipopolysaccharide ( LPS ) trigger rapid translocation of NF-κB transcription factors from the cytoplasm into the nucleus , causing activation of inflammatory genes such as IL-6 via direct binding of NF-κB to κB response elements , recruitment of transcriptional coactivators , and assembly of transcription-initiation and transcription-elongation complexes at target gene promoters ( Ben-Neriah and Karin , 2011 ) . ER-mediated suppression of inflammatory genes can occur by inhibition of NF-κB translocation or DNA binding , or through a transrepression mechanism involving recruitment of ERα to the cytokine promoters via protein–protein interactions ( Ghisletti et al . , 2005; Cvoro et al . , 2006; Nettles et al . , 2008b; Saijo et al . , 2011 ) , a mechanism that is also evident with anti-inflammatory effects of the glucocorticoid receptor ( Uhlenhaut et al . , 2013 ) . For detailed mechanistic studies , we decided to focus on IL-6 , whose suppression by ERα ligands in MCF-7 cells has remained robust and consistent over time ( Srinivasan et al . , 2013 ) , unlike others genes such as monocyte chemoattractant protein-1 ( MCP-1 ) , whose inhibition has been variable ( not shown ) . Treatment of MCF-7 cells with TNFα increased secretion of IL-6 protein , and E2 or resveratrol inhibited this response ( Figure 2A ) . The full ERα antagonist , faslodex/fulvestrant/ICI 182 , 780 ( ICI ) reverses resveratrol-dependent inhibition of IL-6 production by these cells ( Srinivasan et al . , 2013 ) ; thus ERα mediates resveratrol-directed inhibition . Similar ERα-mediated effects were observed in mouse RAW2645 . 7 macrophages stimulated with LPS ( Figure 2B ) , which again were reversed by ICI . To fully characterize the role of resveratrol and ER in coordinating the inflammatory response , MCF-7 cells were treated with TNFα and either E2 or resveratrol , and gene expression was analyzed using Affymetrix cDNA microarrays . Notably , almost all of the resveratrol-modulated genes were also E2 regulated ( Figure 2C , D ) , supporting an ER-mediated mechanism of action . Interestingly , genes that were modulated by ERα ligand in the same direction as TNFα were more sensitive to E2 than resveratrol ( Figure 2C ) . In contrast , resveratrol had a greater impact in opposing TNFα activity ( Figure 2D ) . The set of genes that were regulated in opposite directions at least twofold by E2 vs resveratrol was less than 0 . 5% of total; thus nearly all of the effects of resveratrol were ERα-mediated in this context . 10 . 7554/eLife . 02057 . 004Figure 2 . Resveratrol represses inflammatory genes through ER . ( A ) MCF-7 cells were plated into charcoal-stripped phenol red free media and treated for 24 hr with 1 ng/ml TNFα ±10 μM E2 , or 10 μM resveratrol . Secreted IL-6 protein was measured from the media using AlphaLISA . Mean ± SEM of biological triplicates are shown . ( B ) RAW264 . 7 macrophages were treated as in panel A , and stimulated with LPS as indicated . Mean ± SEM from biological triplicates are shown . ( C and D ) Steroid-deprived MCF-7 cells were treated for 4 hr with 10 ng/ml TNFα alone or in combination with 10 nM E2 or 10 μM resveratrol . Total RNA was reverse transcribed and analyzed using Affymetrix Genechip microarrays . Transcripts showing >twofold changes in expression upon TNFα stimulation were classified as indicated . Summary of genes regulated ( C ) in the same direction or ( D ) in opposite directions by TNFα and ER ligands are shown . ( E ) Steroid-deprived MCF-7 were pre-treated for 1 hr with ethanol vehicle or 1 μM ICI , and then treated as indicated with 10 ng/ml TNFα , 10 nM E2 , and 10 μM resveratrol for 2 hr . Total RNA reverse-transcribed and analyzed by qPCR for the indicated mRNAs . Mean ± SEM of a representative experiment of biological duplicates are shown . ( F and G ) IL-6 mRNA levels in steroid-deprived MCF-7 cells pre-treated with vehicle or 10 μg/ml CHX for 1 hr and stimulated TNFα , E2 , and resveratrol as in panel D for 3 hr were analyzed by qPCR . Levels in the control samples ( first bar ) of each graph were arbitrarily set to 1 . Mean ± SEM of a representative experiment are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 00410 . 7554/eLife . 02057 . 005Figure 2—figure supplement 1 . Resveratrol represses IL-6 in a dose-dependent manner . Steroid-deprived MCF-7 cells were pretreated with ethanol vehicle or increasing doses of ICI for 1 hr , and then stimulated with 10 ng/ml TNFα in combination with increasing doses of E2 or RES for 2 hr . Relative IL-6 mRNA levels were determined by qPCR . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 00510 . 7554/eLife . 02057 . 006Figure 2—figure supplement 2 . Resveratrol represses inflammatory genes through ER . Steroid-deprived T47D cells were pretreated for 1 hr with ethanol vehicle or 1 μM ER antagonist ICI , and then treated as indicated with 10 ng/ml TNFα , 10 nM E2 and 10 μM resveratrol for 2 hr . Relative mRNA levels were determined by qPCR . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 00610 . 7554/eLife . 02057 . 007Figure 2—figure supplement 3 . Resveratrol represses IL-6 in cycloheximide-treated cells . IL-6 mRNA levels in steroid-deprived T47D cells pre-treated with vehicle or 10 μg/ml CHX for 1 hr and stimulated TNFα , E2 , and resveratrol as in Figure 2—figure supplement 2 for 3 hr were analyzed by qPCR . Levels in the control samples ( first bar ) of each graph were arbitrarily set to 1 . Mean ± SEM of a representative experiment are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 007 Confirmation of the ligand-modulated , TNFα-induced gene expression profile with qPCR showed that IL-6 , prostaglandin E receptor 4 ( PTGER4 ) , and TNF receptor superfamily member 11b ( TNFRSF11B ) were TNFα-induced , and equally suppressed by E2 or resveratrol ( Figure 2E ) . Importantly , the effects of resveratrol on expression of these inflammatory genes were fully reversed by ICI in both MCF-7 ( Figure 2E , Figure 2—figure supplement 1 ) , and T47D breast cancer cells ( Figure 2—figure supplement 2 ) , demonstrating that ERα mediates resveratrol-dependent repression of these genes . Other genes such as Rho-associated , coiled-coil containing protein kinase 1 ( ROCK1 ) exhibited E2-selective repression ( Figure 2E ) , consistent with the array data showing some E2-selective genes . To determine if resveratrol and E2 repress IL-6 indirectly , via transcriptional regulation of another protein that regulates NF-κB activity ( Auphan et al . , 1995; Scheinman et al . , 1995; King et al . , 2013 ) , cells were pre-treated with vehicle or the protein-synthesis inhibitor , cycloheximide ( CHX ) . In both MCF-7 and T47D cells , CHX led to super-induction of IL-6 mRNA ( Figure 2F , Figure 2—figure supplement 3 ) , which is a hallmark of CHX response ( Faggioli et al . , 1997; Hershko et al . , 2004 ) . However , CHX did not affect repression of TNFα-induced IL-6 expression by E2- or resveratrol ( Figure 2G , Figure 2—figure supplement 3 ) . Thus , resveratrol and E2 do not require de novo protein synthesis for this repression . Collectively , these results suggest that resveratrol modulates the inflammatory response through a direct , ERα-mediated transrepression mechanism , which we further verify with ChIP assays , below . Upon agonist binding , the ERα LBD undergoes a conformational change that allows helix 12 to dock across helix 11 and helix 3 ( Figure 3—figure supplement 1 ) , thereby forming a coactivator-binding surface called activation function 2 ( AF2 ) ( Brzozowski et al . , 1997; Shiau et al . , 1998; Warnmark et al . , 2002 ) . Importantly , removal of helix 12 from this position reveals a longer groove that binds an extended peptide motif found in transcriptional corepressors , such as NCoR and SMRT ( Heldring et al . , 2007 ) . Further , antagonists can reposition helix 12 out of the active conformation , and stimulate recruitment of corepressors to this extended groove , or position helix 12 to block both coactivators and corepressors to the AF2 surface ( Shiau et al . , 1998; Figure 3—figure supplement 1 ) . By binding to the LBD , ER ligands may also facilitate recruitment of coactivators to another major coregulator-binding site in the unstructured amino-terminal domain of ERα , called AF1 ( Webb et al . , 1998; Nettles and Greene , 2005 ) . In fact , the agonist activity of tamoxifen is mediated by AF1 in tissues with higher expression of coactivators that bind preferentially to that region ( McInerney and Katzenellenbogen , 1996; Shang and Brown , 2002 ) . These different potential signaling mechanisms were reviewed in Nettles and Greene ( 2005 ) . The DNA-binding domain also contributes to AF2-mediated receptor activity through unknown mechanisms , further complicating matters ( Meijsing et al . , 2009; Srinivasan et al . , 2013 ) . In addition , coactivator recruitment to AF2 is also affected by partial agonists , which subtly reposition helix 11 to disrupt proper docking of helix 12 in its active position ( Nettles et al . , 2008a ) . Thus the AF2 surface represents a nexus for ligand-mediated control of both recruitment of coregulators to the LBD and allosteric signaling to other domains . To define the structural basis for the selective anti-inflammatory property of resveratrol , the ERα LBD was crystallized in complex with resveratrol and the SRC2 nuclear receptor-interacting domain peptide containing an LxxLL motif ( Figure 3A , Table 1 ) . Unlike E2 , which binds in a single orientation ( Brzozowski et al . , 1997; Warnmark et al . , 2002; Figure 3B ) , resveratrol binds to ERα in two different orientations in one subunit of the dimer , shown as conformers #1 and #2 ( Figure 3C ) . Conformer #1 shows the canonical para phenol of resveratrol mimicking the A-ring of E2 , whereas in conformer #2 , this is flipped . Also unexpectedly , in the other subunit of the dimer , resveratrol bound predominantly with the resorcinol group mimicking the A-ring of E2 , in conformer #2 . To our knowledge , this is the first example of a ligand-bound ER structure that does not have a para phenol moiety in that position . 10 . 7554/eLife . 02057 . 008Figure 3 . ERα adopts a resveratrol-specific conformation . ( A ) Crystal structure of ERα LBD in complex with resveratrol . The LBD is shown as a ribbon diagram with one monomer colored gray and the other cyan , except for helix 12 ( h12 ) , colored magenta . The receptor-interacting peptide of SRC2 ( coral tube ) docks at the AF2 surface . ( B ) Structure of E2-bound ERα shows that the A-ring forms a hydrogen-bonding network that is conserved among steroid receptors . PDB ID: 1ERE . ( C ) Binding orientations of resveratrol . Compared to E2 , resveratrol binds in two distinct orientations . Conformer #1 shows the expected binding orientation , with the phenol mimicking the A-ring of E2 . In contrast , the ‘flipped’ conformer #2 with the resorcinol mimicking the A-ring of E2 was unexpected and predominant . Hydrogen bonds ( dashes ) and residues that contact the resveratrol molecule are shown . ( D ) 19F-NMR of F-resveratrol . The inset shows a narrow peak in the spectrum of F-resveratrol in buffer ( half-height line width = 27 Hz ) , while the broad peak for F-resveratrol bound to ERα LBD ( modeled in orange ) fits best to two NMR resonances ( colored red and blue ) , consistent with two distinct binding modes . ( E–G ) Crystal structure of the ERα LBD in complex with the control compound i . e . , an A-CD ring estrogen ( gray ) , was superposed on the resveratrol-bound structure ( cyan ) . In panel E , resveratrol ( green ) shifts h3 Met343 to disrupt the normal packing of the h11–h12 loop , shifting the position of V534 by 2 . 5 Å . In panel F , resveratrol-induced shift in h3 is transmitted allosterically via ERαV355 and ERα I358 to SRC2 L693 within its 690LxxLL694 motif . Panel G shows the resveratrol-induced rotation of the SRC2 peptide . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 00810 . 7554/eLife . 02057 . 009Figure 3—figure supplement 1 . Crystal structures of the ERα LBD in complex with E2 and 4-hydroxytamoxifen ( TAM ) . In the E2-bound conformation , h12 lies across h11 and h3 , thus allowing the LxxLL motif peptide of the coactivator , SRC2 to bind at the AF2 surface . In contrast , TAM directly relocates h12 , which in turn occludes the AF2 surface . PDB IDs: 1GWR and 3ERT . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 00910 . 7554/eLife . 02057 . 010Figure 3—figure supplement 2 . Chemical structures of F-resveratrol and the A-CD ring estrogen used as a structural control . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 01010 . 7554/eLife . 02057 . 011Figure 3—figure supplement 3 . Deconvolution of NMR signal from F-resveratrol bound to ERα . One or two peaks were fit to the F-resveratrol signal . The two-peak fit is significantly better than the one peak fit ( Bayesian Information Criterion , or BIC , score for two peak fit is 96 , 550 while that for the one peak fit is 99 , 165; the lower BIC score indicates a significantly better fit ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 01110 . 7554/eLife . 02057 . 012Figure 3—figure supplement 4 . F-RES also binds ERα in two orientations . Crystal structure of the ERα LBD in complex with F-RES showing ligand-binding orientations observed within the pockets . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 01210 . 7554/eLife . 02057 . 013Figure 3—figure supplement 5 . Electron density maps of resveratrol and F-resveratrol within the ERα ligand-binding pocket . The 2Fo-Fc maps ( blue ) were contoured at 1σ , while Fo-Fc difference maps ( red and green ) were contoured at 3σ to indicate where the model is wrong . Red indicates clashes , while green indicates omissions . Densities observed after molecular replacement and autobuild , and before ligand docking ( None ) ; after docking and refinement with the obvious ligand conformer ( Add #2 ) , or both ligand conformers ( Add both ) ; and upon shaking coordinates by 1 Å , and refinement with simulated annealing after removal of either ligand conformer ( Remove #1 or Remove #2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 01310 . 7554/eLife . 02057 . 014Figure 3—figure supplement 6 . Electron density maps of SRC2 peptides docked at the AF2 surface . The electron density maps ( shown as described in Figure 3—figure supplement 5 ) of the SRC2 peptide obtained from the resveratrol- and A-CD ring estrogen-bound structures , upon shaking coordinates by 1 Å , and refinement with simulated annealing , with or without removing the peptide . The peptide is shown in both cases for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 01410 . 7554/eLife . 02057 . 015Table 1 . Data collection and refinement statistics for new ERα structuresDOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 015LigandResveratrolF-resveratrolA-CD ring estrogenPDB ID4PP64PPP4PPSData collection Space groupP 1 21 1P 1 21 1P 1 21 1 a , b , c ( Å ) 56 . 04 , 84 . 67 , 58 . 4254 . 19 , 81 . 93 , 58 . 4756 . 11 , 84 . 19 , 58 . 48 α , β , γ ( ° ) 90 . 0 , 108 . 32 , 90 . 090 . 0 , 110 . 86 , 90 . 090 . 0 , 108 . 35 , 90 . 0 Resolution ( Å ) 33 . 7–2 . 2 ( 2 . 28–2 . 20 ) 46 . 3–2 . 7 ( 2 . 78–2 . 69 ) 33 . 5–1 . 9 ( 2 . 00–1 . 93 ) Number of reflections22 , 678 ( 944 ) 11 , 884 ( 481 ) 38 , 369 ( 3443 ) I/σ12 . 6 ( 2 . 9 ) *22 . 5 ( 1 . 7 ) 27 . 7 ( 2 . 1 ) Rmerge0 . 07 ( 0 . 21 ) 0 . 09 ( 0 . 45 ) 0 . 05 ( 0 . 45 ) Completeness ( % ) 86 . 14 ( 36 . 35 ) 88 . 46 ( 35 . 98 ) 98 . 68 ( 89 . 10 ) Multiplicity2 . 5 ( 1 . 5 ) 6 . 3 ( 5 . 8 ) 3 . 5 ( 2 . 0 ) Refinement Number of non-H atoms Protein384037104014 Ligands515436 Water30736323 Rwork/Rfree16 . 79/22 . 2218 . 38/23 . 9017 . 38/20 . 15 Ramachandran favored ( % ) 999598 Ramachandran outliers ( % ) 0 . 211 . 10 Wilson B-factor17 . 3144 . 1227 . 03 Average B-factor All atoms26 . 766 . 136 . 1 Protein26 . 466 . 436 . 1 Water29 . 742 . 940 . 4 RMS deviations Bond lengths ( Å ) 0 . 0080 . 0110 . 002 Bond angles ( ° ) 1 . 031 . 260 . 61* ( Highest-resolution shell ) . We previously described the binding of ligand in multiple poses as ‘dynamic ligand binding’ , as it was associated with a ligand’s ability to stabilize different conformations of ERα . In this model , a dynamically binding ligand perturbs the conformational ensemble such that there are discrete populations of stable conforms each associated with a specific binding pose , where each receptor can undergo a conformational change as it re-binds the ligand . Further , we showed that this phenomenon is a structural mechanism for partial agonist activity ( Nettles et al . , 2008a; Bruning et al . , 2010; Srinivasan et al . , 2013 ) , which has now also been shown with a G protein-coupled receptor ( Bock et al . , 2014 ) . Lastly , ERα ligands that exhibit this so called dynamic binding profile showed greater anti-inflammatory activity than matched controls that bound in a single pose ( Srinivasan et al . , 2013 ) . To assess whether this dynamic ligand binding occurs in solution , we used F19 NMR , which established dynamic ligand binding to PPARγ ( Hughes et al . , 2012 ) and ERα ( Srinivasan et al . , 2013 ) . Our previous work established that fluorinated ligands display the expected line broadening in F19 NMR signal upon binding to proteins . However , there were characteristic differences in matched isomeric ligands that bound in either a single orientation , or multiple orientations to their respective proteins . The ligands that bound in a single orientation displayed a single broadened peak , while the ligands that bound in more than one orientation displayed either multiple broadened peaks , or a single , asymmetrically shaped peak that was best modeled as two overlapping peaks ( Hughes et al . , 2012; Srinivasan et al . , 2013 ) . Here , we synthesized resveratrol with a fluorine substitution at the meta position on the phenol to generate F-resveratrol ( Figure 3—figure supplement 2 ) , and examined binding of F-resveratrol to ERα . F-resveratrol alone showed a single sharp peak; however when bound to ERα , it displayed a very broad peak that fit best to two peaks ( Figure 3D , Figure 3—figure supplement 3 ) , indicating multiple binding modes . This dynamic binding was corroborated by the crystal structure of an F-resveratrol-bound ERα complex ( Table 1 ) that was best fit with two ligand-binding orientations similar to those displayed by resveratrol ( Figure 3—figure supplements 4 , 5 ) . The crystal structure of the ERα LBD in complex with an A-CD ring estrogen ( Figure 3—figure supplement 2 ) , which has the typical phenolic A-ring but like resveratrol , does not have an adjacent B-ring , showed the same space group and crystal packing as the resveratrol-bound ERα structure ( Table 1 ) . This was therefore used as a control agonist structure . Compared to the typical phenolic A-ring of the A-CD ring estrogen , the resorcinol group of resveratrol induced a shift in helix 3 via a hydrogen bond with the backbone of Leu387 ( Figure 3C ) . In turn , the shift in helix 3 disrupts the loop that connects helix 11 to helix 12 , which in solution should destabilize helix 12 in the agonist conformation . This impact on helix 12 is visualized by the 2 . 5 Å shift in the positioning of the γ-carbons of Val534 ( Figure 3E ) . Helix 12 does not participate in crystal packing , so this change is ligand driven . Notably , the shift in helix 3 also alters binding of the SRC2 peptide at the AF2 surface . Leu693 of the SRC2 peptide binds helix 3 between Val355 and Ile358 , and is shifted by 1 . 6 Å in the resveratrol-bound structure ( Figure 3F ) , inducing an overall rotation of the peptide ( Figure 3G ) . The electron density for the peptides allowed clear visualization of this rotation ( Figure 3—figure supplement 6 ) . Here , the coactivator peptide participates in crystal packing , but the crystals are in the same space group and show the same crystal packing interactions . Thus the rotation of the peptide occurs despite being held in place by an adjacent molecule . In summary , resveratrol induced several unique structural perturbations in ERα , including shifts in the helix 11–12 loop , which should modulate helix 12 dynamics , and direct remodeling of the coactivator-binding surface , which could contribute to an altered receptor–coactivator interaction profile and the lack of a proliferative signal . To determine if the resveratrol-induced structural changes at the AF2 surface directly affect binding of SRC peptides to the ERα LBD in vitro , a non-competitive FRET assay was performed using a fixed amount of GST-ERα LBD and ligands , and increasing doses of fluorescein-tagged SRC peptides . For each SRC family protein , the highest affinity LxxLL motif peptide from SRC1 , SRC2 , or SRC3 , for the E2-bound ERα LBD complex , was selected for further comparisons ( Figure 4—figure supplement 1 ) . Although the EC50 of the SRC2 peptide could not be determined accurately due to a lack of plateau on the curve , the EC50s of all three SRC peptides were comparable for the resveratrol-bound ERα LBD ( Figure 4A ) . This suggests that the coactivator-selectivity profile of resveratrol-bound ERα requires important regions outside the ERα LBD and SRC peptides . 10 . 7554/eLife . 02057 . 016Figure 4 . Resveratrol alters the binding of coregulator peptides to the ERα LBD . ( A ) E2- and resveratrol-induced binding of SRC1 , SRC2 , and SRC3 peptides to the ERα LBD were compared using LanthaScreen assay performed at fixed ligand concentrations , with increasing doses of SRC peptides . Mean ± SEM ( n = 3 ) are shown . *EC50 could not be determined accurately since the saturating SRC2 peptide dose is unclear . ( B and C ) Hierarchical clustering of coregulator peptide-binding at the AF2 surface induced by 1 μM E2 , ( B ) 100 μM resveratrol or ( C ) 1 μM A-CD ring estrogen , was performed using the quantitative in vitro assay , MARCoNI . MI >0 suggests ligand-induced recruitment , while MI <0 suggests ligand-dependent dismissal of a peptide compared to DMSO vehicle . The black bracket shows a cluster of E2 dismissed peptides that are not dismissed by resveratrol . See Figure 4—figure supplement 2 for more details . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 01610 . 7554/eLife . 02057 . 017Figure 4—figure supplement 1 . SRC peptides . The SRC peptides shown exhibited the highest E2-induced binding to ERα and were therefore used for the LanthaScreen assay . The sequence bound to the AF2 surface in the resveratrol- and A-CD ring crystal structures is underlined . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 01710 . 7554/eLife . 02057 . 018Figure 4—figure supplement 2 . Details of proteomic comparison of ligand-induced binding of coregulator peptides using MARCoNI . Statistically significant changes relative to vehicle were identified by Student's t-test . *p<0 . 05 , **p<0 . 01 or ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 018 To test if the altered AF2 surface was also apparent in solution , we analyzed ligand-induced binding of over 150 distinct , nuclear receptor-interacting , coregulator peptide motifs to the ERα LBD , including those derived from both coactivators and corepressors , using the microarray assay for real-time coregulator-nuclear receptor interaction ( MARCoNI ) ( Aarts et al . , 2013 ) . Hierarchical clustering of the peptide-binding results showed that compared to E2 , resveratrol showed similar patterns of recruitment , but with reduced binding of most coactivator peptides to the ERα LBD ( Figure 4B , Figure 4—figure supplement 2 ) . However , there was a subset of peptides that were dismissed by E2 , including several from the NCoR corepressor , which resveratrol failed to dismiss . In contrast , the A-CD ring estrogen and E2 had similar effects on coactivator peptide recruitment to , or dismissal from the ERα LBD ( Figure 4C , Figure 4—figure supplement 2 ) , consistent with a fully functional AF2 surface . Taken together , these results suggest that resveratrol binds the ERα LBD , and induces an altered AF2 surface , which reduces affinity for most peptides , but enables selectivity in the context of full-length receptor and coregulators , as shown by our mammalian two hybrid and ChIP data . ERα uses a large array of coregulators to activate transcription ( Bulynko and O'Malley , 2010; Lupien et al . , 2009; Metivier et al . , 2003 ) , but much less is known about the requirements for ERα-mediated transrepression . To identify factors required for E2 and resveratrol-dependent repression of IL-6 , we undertook a small-scale siRNA screen , targeting over 25 factors including estrogen receptors and known ERα-interacting coregulators . ERα knockdown blocked inhibition of IL-6 expression by both E2 and resveratrol , unlike siRNA against ERβ or the estrogen-binding G protein-coupled receptor , GPR30 ( Figure 5A , Figure 5—figure supplements 1 , 2 ) , confirming that ERα mediates both E2- and resveratrol-dependent repression of IL-6 . 10 . 7554/eLife . 02057 . 019Figure 5 . Molecular requirements for resveratrol- and E2-mediated suppression of IL-6 . ( A–E ) MCF-7 cells were transfected with the indicated siRNAs and steroid-deprived for 48 hr . The cells were then treated with 10 ng/ml TNFα and 10 μM resveratrol or 10 nM E2 for 2 hr . IL-6 mRNA levels were compared by qPCR , and are shown relative to the cells treated with ethanol vehicle and control siRNA . The small-scale siRNA screen was repeated three different times . Mean ± SEM are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 01910 . 7554/eLife . 02057 . 020Figure 5—figure supplement 1 . Molecular requirements for resveratrol- and E2-mediated suppression of IL-6 . MCF-7 cells were transfected with the indicated siRNAs and steroid-deprived for 48 hr . The cells were then treated with 10 ng/ml TNFα and 10 μM RES or 10 nM E2 for 2 hr . Total RNA was analyzed by qPCR . Changes IL-6 mRNA levels are shown relative to the cells treated with ethanol vehicle and siRNA control . The small-scale siRNA screen was repeated three different times . The mean ± SEM are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 02010 . 7554/eLife . 02057 . 021Figure 5—figure supplement 2 . Effect of siRNAs on target mRNA levels . MCF-7 cells were transfected with the indicated siRNAs , steroid-derived for 48 hr and analyzed by qPCR for the indicated mRNAs . Mean +SEM of duplicate experiments are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 021 Knockdown of SRC1 , SRC2 , and SRC3 by RNA-interference revealed that these coregulators play distinct but overlapping roles in controlling IL-6 expression . SRC1 and SRC3 knockdown led to an increase in IL-6 mRNA in cells treated with either vehicle or TNFα , indicating a general role in repressing IL-6 transcription ( Figure 5B ) . SRC3 knockdown also blocked E2- and resveratrol-mediated suppression , suggesting an additional role for SRC3 in integrating TNFα and ERα signaling . By contrast , SRC2 knockdown markedly reduced TNFα-directed induction of IL-6 transcripts , demonstrating that it is required for coactivation of TNFα induction of this gene . However , SRC2 knockdown also demonstrated that SRC2 is required for repression of IL-6 by E2 or resveratrol ( Figure 5B ) , suggesting that these ER ligands switched SRC2 function from that of a coactivator to a corepressor . This is similar to the context-dependent role of SRC2/GRIP1 in glucocorticoid action ( Rogatsky et al . , 2002 ) , and the gene-specific role of silencing mediator for retinoid and thyroid hormone receptors ( SMRT/NCOR2 ) , which corepresses some ERα-target genes , while being required for activation of others ( Peterson et al . , 2007 ) . Knockdown studies also established that the acetyltransferase , CBP , was also required for suppression of IL-6 induction by E2 and resveratrol , whereas p300 and pCAF were rather required for TNFα-induced expression of IL-6 ( Figure 5C ) , suggesting that CBP and p300 play opposing roles in repression vs activation of the same gene . Finally , knockdown of macro domain protein 1 ( LRP16 ) , a coactivator for both ERα and NF-κB ( Han et al . , 2007; Wu et al . , 2011 ) , also dampened the TNFα response , but did not affect suppression of IL-6 by either E2 or resveratrol ( Figure 5—figure supplement 1 ) . We also tested the roles of various components of complexes that harbor dedicated corepressors , including nuclear receptor corepressor ( NCoR/NCOR1 ) , SMRT , repressor element 1-silencing transcription corepressor 1 ( RCOR1/CoREST ) , and ligand-dependent nuclear receptor corepressor ( LCoR ) . CoREST functions as a scaffold protein that associates with several histone-modifying enzymes , including lysine-specific demethylase 1 ( LSD1 ) , euchromatic histone methyltransferase 1 ( GLP ) and 2 ( G9a ) , C-terminal binding protein 1 ( CtBP1 ) as well as histone deacetylases HDAC1 and HDAC2 ( Shi et al . , 2003 , 2004 ) . Knockdown of CoREST blocked suppression of IL-6 by both E2 and resveratrol ( Figure 5D ) , demonstrating that CoREST is required for ERα-mediated repression of IL-6 . In contrast , knockdown of LSD1 , HDAC1 , HDAC2 , HDAC3 , G9a , GLP , and several other ERα-interacting corepressors including SMRT , NCoR , LCoR , and CtBP1 ( Fernandes et al . , 2003; Garcia-Bassets et al . , 2007 ) , had no effect on suppression of IL-6 by either E2 or resveratrol ( Figure 5D , Figure 5—figure supplement 1 ) . However , knockdown of HDAC2 siRNA globally raises expression of IL-6 , as did knockdown of CoREST . Collectively , these findings suggest that CoREST is a dedicated corepressor required for ERα-mediated transrepression , but that it also has a more general role in limiting IL-6 expression . Knockdown of SIRT1 or SIRT2 had little or no effect on the suppression of IL-6 by the ER ligands ( Figure 5E ) . Indeed , SIRT1 siRNA slightly raised the expression of IL-6 in cells treated with vehicle but not TNFα . However , two other proteins known to associate with SIRT1 , nicotinamide mononucleotide adenylyltransferase 1 ( NMNAT1 ) and deleted in breast cancer 1 ( DBC1 ) ( Zhang et al . , 2009; Yu et al . , 2011a ) , contributed to ligand-dependent repression of IL-6 ( Figure 5E ) . In contrast , depletion of poly ADP-ribose polymerase 1 ( PARP1 ) , which also interacts with SIRT1 ( Rajamohan et al . , 2009; Zhang et al . , 2012 ) , had no obvious effect on ERα-mediated repression of IL-6 ( Figure 5—figure supplement 1 ) . Thus ERα requires a distinct , functionally diverse cohort of coregulators to mediate ligand-dependent transrepression of IL-6 . Although knockdown studies suggested that SIRT1 is not required for resveratrol-mediated suppression of IL-6 , resveratrol is best known as a SIRT1 activator . Further , the lack of phenotype in a screening mode could reflect a number of alternatives for any of the individual siRNAs , including functional redundancies , lack of sufficient knockdown , and slow protein turnover . Consequently , we wanted to assess the contribution of other resveratrol signaling pathways to resveratrol-dependent repression of IL-6 ( Figure 6A ) . Recently , some of these effects were shown to occur via resveratrol binding to and inhibiting cAMP-specific phosphodiesterases ( PDEs ) that hydrolyze and deplete cAMP ( Park et al . , 2012; Tennen et al . , 2012 ) . Thus , resveratrol elevates cellular cAMP levels and stimulates the canonical cAMP-signaling network downstream of catecholamine and glucagon signals that activate protein kinase A ( PKA ) and the CREB transcription factor , as well as rapid AMP-activated protein kinase ( AMPK ) signaling . In turn , AMPK drives production of NAD+ , the cofactor required for SIRT1 deacetylase activity , although signaling in the opposite direction has also been reported , where resveratrol stimulates SIRT1 via an unknown mechanism to activate AMPK ( Price et al . , 2012 ) . 10 . 7554/eLife . 02057 . 022Figure 6 . Resveratrol does not repress IL-6 through the cAMP or AMPK pathways . ( A ) Resveratrol stimulates ERα activity and inhibits cAMP-specific phosphodiesterases ( PDEs ) to activate cAMP SIRT1 , and AMPK . The small molecule compounds i . e . , the adenylyl cyclase activator forskolin ( FSK ) and the PDE inhibitor rolipram ( ROL ) used to further dissect this signaling network are shown in blue . ( B ) Resveratrol increases intracellular NAD+ levels . Average intracellular NAD+ concentrations were determined in MCF-7 cells treated with resveratrol for 5 min . Unpaired Student's t test ( mean ± SEM , n = 6 ) was used to determine statistical significance . *p=0 . 006 . ( C ) IL-6 mRNA levels in steroid-deprived MCF-7 cells transfected with the indicated siRNAs and treated as described in Figure 5 , were compared by qPCR . Mean ± SEM of representative biological duplicates are shown . ( D ) Steroid-deprived MCF-7 cells were treated with 10 ng/ml TNFα , 10 nM E2 , 10 μM resveratrol , 10 μM FSK , and 25 μM ROL as indicated for 2 hr . Relative IL-6 mRNA levels were compared by qPCR . Mean ± SEM of representative biological duplicates are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 02210 . 7554/eLife . 02057 . 023Figure 6—figure supplement 1 . Resveratrol represses IL-6 in cells dorsomorphin-treated cells . Steroid-deprived MCF-7 cells were pre-treated with DMSO control or 1 μM Dorsomorphin ( DOS ) for 30 min , and then treated as indicated with TNF , E2 and RES for 2 hr . Relative IL-6 mRNA levels were determined by QPCR . Shown are mean ± SEM of representative biological duplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 02310 . 7554/eLife . 02057 . 024Figure 6—figure supplement 2 . Effect of siRNAs on the mRNA levels of AMPK catalytic subunits . MCF-7 cells were transfected with the indicated siRNAs , steroid-derived for 48 hr and analyzed by qPCR for the indicated mRNAs . Mean +s . e . m of duplicate experiments are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 024 Resveratrol increased intracellular NAD+ levels , and this increase was statistically significant at a resveratrol dose of 100 μM ( Figure 6B ) . These findings suggest that the PDE/cAMP and AMPK pathways for NAD+ production are active in this context , but at higher doses of resveratrol than required for anti-inflammatory effects through ERα . This raises the possibility that resveratrol represses IL-6 via both ERα- and PDE-mediated mechanisms ( Figure 6A ) . However the AMPK inhibitor , Dorsomorphin , did not affect repression of IL-6 by E2 or resveratrol ( Figure 6—figure supplement 1 ) . Further , knockdown of both catalytic subunits of AMPK increased IL-6 expression globally , but did not affect resveratrol-mediated repression of IL-6 ( Figure 6C , Figure 6—figure supplement 2 ) . Finally , activation of the cAMP pathway with forskolin , or the PDE inhibitor rolipram , increased IL-6 expression ( Figure 6D ) , demonstrating categorically that activation of this pathway does not inhibit IL-6 expression . To further probe the mechanism of ERα-mediated transrepression , chromatin immunoprecipitation ( ChIP ) assays were used to compare protein recruitment and accumulation of PTMs at the IL-6 promoter . TNFα led to recruitment of ERα and the p65 NF-κB subunit , which were unaffected by E2 or resveratrol ( Figure 7A ) . As a control , ChIP using pre-immune rabbit IgG showed no changes in promoter occupancy ( Figure 7—figure supplement 1 ) . In addition , resveratrol alone did not induce recruitment of ERα to IL-6 promoter ( Figure 7—figure supplement 2 ) , as we have previously reported for the effects of E2 on several inflammatory genes ( Nettles et al . , 2008b ) . TNFα also led to accumulation of p65 acetylated at Lys310 ( p65 K310-ac ) , a PTM catalyzed by p300 that is essential for full transcriptional activity ( Chen and Greene , 2004 ) , while resveratrol and E2 reduced p65 K310-ac levels ( Figure 7A ) . 10 . 7554/eLife . 02057 . 025Figure 7 . ERα orchestrates ligand-dependent coregulator exchange at the IL-6 promoter . ( A–D ) Occupancy of the indicated factors at the IL-6 promoter were compared by ChIP assay in steroid-deprived MCF-7 cells treated with 10 ng/ml TNFα alone or in combination with 10 nM E2 or 10 μM resveratrol , and fixed after 0 , 15 , 30 , and 45 min ( mean ± SEM n = 3 ) ( E ) Effect of ICI on promoter occupancy was determined by ChIP assay in steroid-deprived MCF-7 cells were pretreated with vehicle or 1 μM ICI for 1 hr , stimulated with 10 ng/ml TNFα plus 10 μM resveratrol , and fixed after 15 , 30 , or 45 min . Average promoter occupancies are shown as fold changes ( mean ± SEM n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 02510 . 7554/eLife . 02057 . 026Figure 7—figure supplement 1 . Control ChIP assay . As a control , ChIP assay was performed using pre-immune rabbit IgG to compare MCF-7 cells treated with 10 ng/ml TNFα and 10 μM RES and 10 nM E2 for 0 , 15 , 30 and 45 min . Average promoter occupancies are shown as fold changes ( mean ± SEM n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 02610 . 7554/eLife . 02057 . 027Figure 7—figure supplement 2 . Without TNFα , RES does not induce recruitment of ERα to the IL-6 promoter . ChIP assay using ERα antibody was performed in steroid-deprived MCF-7 cells treated with only 10 μM RES for 0 , 15 , 30 and 45 min . Average promoter occupancies are shown as fold changes ( mean ± SEM n = 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 02710 . 7554/eLife . 02057 . 028Figure 7—figure supplement 3 . RES induces ERα and SIRT1 recruitment at the TFF1/pS2 promoter . Steroid-deprived MCF-7 cells were treated with 10 μM RES , and fixed after the indicated periods . Average promoter occupancies were determined by ChIP assay using ERα and SIRT1 antibodies and are shown as fold changes ( mean ± SEM n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 02810 . 7554/eLife . 02057 . 029Figure 7—figure supplement 4 . ICI increased p65 K310-ac levels at the IL-6 promoter . Steroid-deprived MCF-7 cells were pretreated with or without 1 μM ICI for 1 hr before a 45-min treatment with 10 ng/ml TNFα and 10 μM RES and 10 nM E2 as indicated . Average promoter occupancies were determined by ChIP assay using p65 K310-ac antibody and are shown as fold changes ( mean ± SEM n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 02910 . 7554/eLife . 02057 . 030Figure 7—figure supplement 5 . ICI did not increase recruitment of pCAF , p300 and SIRT2 . Effect of ICI on promoter occupancy was determined by ChIP assay in steroid-deprived MCF-7 cells were pretreated with vehicle or 1 μM ICI for 1 hr , stimulated with 10 ng/ml TNFα plus 10 μM resveratrol , and fixed after 15 , 30 or 45 min . Average promoter occupancies are shown as fold changes ( mean ± SEM n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 030 Resveratrol and E2 also reduced the recruitment of several coregulators . TNFα led to recruitment of pCAF , followed by p300 , while E2 and resveratrol delayed recruitment of pCAF , and inhibited recruitment of p300 ( Figure 7B ) , consistent in their ability to reduce levels of p65 K310-ac levels and IL-6 expression . TNFα signaling also led to recruitment of SIRT1 , followed by SIRT2 , and resveratrol and E2 inhibited recruitment of both sirtuins at the IL-6 promoter ( Figure 7B ) . However , this was not the case at the estrogen-induced pS2 promoter , where resveratrol and E2-induced ERα and SIRT1 recruitment ( Figure 7—figure supplement 3 ) . It is noteworthy that SRC2 was required for coactivation by TNFα and for suppression of IL-6 by ER ligands , but its recruitment was similar across signals ( Figure 7C ) . Resveratrol and E2 also modulated the recruitment of coregulators that showed some ligand-dependent differences . TNFα evicted CoREST and SRC3 , whereas E2 and resveratrol led to recruitment of both factors to the IL-6 promoter ( Figure 7D ) . Resveratrol induced less recruitment of CoREST and SRC3 than E2 , consistent with the reduced recruitment of SRC3 by resveratrol-bound ERα in the context of full-length proteins ( Figure 1D , F ) . Resveratrol and E2 also augmented recruitment of SMRT , CBP and NMNAT1 , and the effects of resveratrol were slightly greater than E2 ( Figure 7D ) . To determine if ERα mediated these events at the IL-6 promoter , ChIP assays were performed in MCF-7 cells pre-treated with vehicle or the ER antagonist , ICI , and then treated with TNFα and resveratrol for an interval that showed a maximal effect , as determined from Figure 5A–D . ICI reduced recruitment of key coregulators , including CoREST , SRC3 , CBP and NMNAT1 ( Figure 7E ) , as well as coregulators such as SMRT and SIRT1 that were not required for resveratrol-dependent suppression of IL-6 . ICI also increased p65 K310-ac levels ( Figure 7E , Figure 7—figure supplement 4 ) , consistent with higher NF-κB activity . It is interesting that ICI did not stimulate recruitment of pCAF , p300 , or SIRT2 ( Figure 7—figure supplement 5 ) , suggesting that ICI can block ligand-induced activity , but does not mimic the unliganded receptor . Overall , the data demonstrate that resveratrol mediates repression of IL-6 by orchestrating an ERα- and ligand-dependent exchange of a number of distinct coregulators that are required for the integration of steroidal and inflammatory signaling pathways ( Figure 8 ) . 10 . 7554/eLife . 02057 . 031Figure 8 . Proposed model for ERα-mediated transrepression of IL-6 . In MCF-7 cells stimulated with TNFα , the p65 NF-κB subunit binds the IL-6 promoter and mediates recruitment of many coregulators including p300 , which acetylates p65 at Lys310 , to drive transactivation of IL-6 . In these cells , TNFα also induces recruitment of ERα to this site via a tethering mechanism . In response to E2 or resveratrol , ERα undergoes a conformational change , dismisses the set of coregulators including p300 , and recruits a set that contains SRC3 , CoREST , and other key coregulators required to inhibit p65 acetylation and repress IL-6 . DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 031
We demonstrate that resveratrol is a pathway-selective ERα ligand that modulates the inflammatory response without stimulating proliferation , by binding dynamically to the receptor , inducing an altered AF2 coactivator-binding site , and regulating the recruitment of a cast of coregulators at the IL-6 locus . There is a large body of literature on resveratrol-mediated suppression of IL-6 , as part of the inflammatory response in a variety of tissues , including liver , microglia , gut , and cardiovascular system , which are all ERα-positive tissues ( Csiszar et al . , 2008; Pfluger et al . , 2008; Lu et al . , 2010; Singh et al . , 2010 ) . There is also evidence that the in vivo effects of resveratrol on the inflammatory response require ERs ( Yu et al . , 2008 ) , but through previously unknown mechanisms . In this study , we show that the effects of resveratrol are ERα-dependent , and that resveratrol alters recruitment of the coregulators associated with ERα , thereby establishing ERα as the primary target for resveratrol modulation of the inflammatory response . Our results support the concept that subtle modulation of receptor–coregulator interactions is sufficient to drive highly divergent phenotypes . This is shown by the reduced interaction of resveratrol-bound ERα with SRC3 in a mammalian two-hybrid assay , and reduced ERα-mediated recruitment of SRC3 to both the estrogen-stimulated gene , GREB1 , and the estrogen-repressed gene , IL-6 . While the original report of resveratrol as an ERα ligand described it as a superagonist ( Gehm et al . , 1997 ) , many subsequent reports have described it as a partial agonist , and non-proliferative in the breast and uterus ( Ashby et al . , 1999; Turner et al . , 1999; Bowers et al . , 2000; Bhat et al . , 2001; Xu and Li , 2003; Karmakar et al . , 2009 ) . Further , there have been a number of clinical trials of resveratrol in humans , without reports of feminization ( Tome-Carneiro et al . , 2013 ) . We found that pathway-selective resveratrol action was associated with changes in the AF2 surface of the LBD , but not differences in affinity between the short LxxLL motif peptides derived from different members of the SRC family . Instead , the determinants of SRC-binding selectivity may be just C-terminal to the ordered part of the receptor-interaction domain ( Scheinman et al . , 1995 ) , may lie further outside the SRC regions tested ( Leo and Chen , 2000 ) , and might involve the other functional domains of ERα outside the LBD . In fact , SRC2 also interacts with ERα via the AF1 coactivator-binding site located in the unstructured N-terminus of the receptor ( Norris et al . , 1998 ) . The peptide profiling experiments show that resveratrol generally lowers affinity for recruited peptides , but display a defect in dismissal of peptides bound to the unliganded LBD , thus demonstrating a change in the shape of the AF2 surface in solution . Also , functional analysis of ERα domains suggests that the DNA-binding domain plays a vital role in resveratrol-induced ERα activity ( Srinivasan et al . , 2013 ) . These data support the idea that inter-domain communication and binding of coactivators to multiple ERα domains is an important aspect of this anti-inflammatory-selective signaling mechanism . Resveratrol belongs to a newly discovered class of compounds that can bind to ERα in two different orientations . With either the phenol or the resorcinol group forming the conserved hydrogen bond with helix 3 , the ensemble of receptors will display a mixture of conformers , including potentially dimers with different combinations of binding modes . Importantly , we previously showed that binding of ligands in two flipped orientations could stabilize the receptor in either the active or inactive conformations , generating partial agonist activity ( Bruning et al . , 2010 ) . Further , those compounds could be modified to titrate the relative balance of stabilizing the active vs inactive protein conformations . Ligand dynamics as an allosteric control mechanism represents a new principle in drug design that has since been observed with PPARγ ( Hughes et al . , 2012 ) , dihydrofolate reductase ( Carroll et al . , 2011 ) , and more recently a mechanism to generate partial agonists for G protein-coupled receptors ( Bock et al . , 2013 ) . In addition , we found that dynamic binding of ligands also contributes to pathway-selective signaling , which like resveratrol , was selectively anti-inflammatory ( Srinivasan et al . , 2013 ) . Thus , the multiple binding modes for resveratrol may contribute to its reduced gene activation signal and lack of a proliferative effect . Signaling from estrogens or pro-inflammatory cues involves spatio-temporal coordination of complex transcriptional activation programs ( Shang et al . , 2000; Metivier et al . , 2003; Medzhitov and Horng , 2009 ) . Kinetic ChIP assays at a single locus are an important addition to genome-scale ChIP studies , and they have revealed that signal integration can involve shifts in the timing of chromatin association . For example , pCAF recruitment to the IL-6 promoter is dynamically regulated in a distinct fashion by different signaling curves . Likewise , our results suggest that estrogen- and resveratrol-dependent attenuation of the inflammatory response is not simply a blockade of a single signaling pathway , but requires ERα-mediated orchestration of complex transcriptional repression programs . At the IL-6 promoter , one aspect of this repression program involves recruitment of SRC3 and CBP , ligand-dependent dismissal of p300 , and loss of p65 K310-ac , which is required for full transcriptional activity and which could be directed by p300 ( Chen and Greene , 2004 ) . Our results support a model where resveratrol-bound ERα mediates recruitment of an SRC3/CBP complex and blocks the TNFα-induced recruitment of p300 and pCAF , thereby blocking acetylation of p65 ( Figure 8 ) . The initial description of coregulators as either coactivators or corepressors has evolved with the understanding that they have more context-specific effects . The opposing effects of CBP and p300 , and the different roles of SRC2—coactivating TNFα induction of IL-6 , but corepressing ERα-mediated signaling on the same gene—support this idea . The disparate roles of SRCs are also striking and unexpected , as all three played some role in repressing IL-6 . SRC1 and SRC3 played ligand-independent roles , while SRC2 and SRC3 were more specifically required for repression by E2 and resveratrol . These differences are likely due to the different transient , multi-protein complexes formed by these promiscuous coregulators ( Stenoien et al . , 2001; Jung et al . , 2005; Malovannaya et al . , 2011 ) . For example , the mouse ortholog of SRC2 , called GRIP1 , was found to have an additional role in glucocorticoid-mediated repression of inflammatory genes ( Rogatsky et al . , 2002 ) , which mapped to a binding site for a trimethyltransferase , Suv4-20h1 , an enzyme that represses glucocorticoid receptor activity ( Chinenov et al . , 2008 ) . A similar context-dependent activity is also seen with the corepressor , SMRT , which is required for activation of some ERα-target genes ( Peterson et al . , 2007 ) . The preferred association of resveratrol-bound ERα with SRC2 is also intriguing , given roles of both resveratrol and SRC2 in metabolic regulation ( York and O'Malley , 2010 ) . Interestingly , recruitment of p65 and ERα were largely insensitive to E2 and resveratrol , suggesting that these ligands change the conformation of ERα at the promoter to dictate the shape of the AF2 surface and modulate recruitment of SRCs and other coregulators , but also to change the structure of proteins such as SRC2 , which shows changes in function despite similar recruitment profiles ( Figure 8 ) . Other coregulators , such as the scaffold CoREST , form biochemically stable complexes ( Shi et al . , 2005; Malovannaya et al . , 2011 ) , which may provide a less flexible platform for signal integration , but which brings together a dedicated group of effector enzymes . The lack of phenotypes from targeting the enzyme components of the CoREST complex does not necessarily indicate that these targets are not involved in ERα-mediated transrepression , as the siRNA screen showed variable knockdown , and target-specific optimizations might be required to reveal their effects . Moreover , this may also reflect functional redundancy , for example of HDAC1 and HDAC2 , or G9a and GLP . However , HDAC2 siRNA increased basal expression of IL-6 , suggesting that these subunits are required to restrain IL-6 expression in a TNFα- and ERα-independent manner , consistent with previous ChIP-array studies in MCF-7 cells , which suggest that IL-6 is a target of an LSD1/CoREST/HDAC complex ( Wang et al . , 2009 ) . The ability to perturb and track many coregulators in parallel illustrates that multiple determinants contribute to a single phenotype such as IL-6 expression , similar to the different coregulator requirements of estrogen-induced genes ( Won Jeong et al . , 2012 ) . While knockdown of SIRT1 had no major effect on IL-6 expression in breast cancer cells , ERα-driven control of the association of SIRT1 with chromatin contributes to SIRT1 activity in other contexts ( Elangovan et al . , 2011; Yu et al . , 2011a ) . Indeed , we show here that ERα ligand can direct SIRT1 to a canonical ERE of an estrogen-induced gene , pS2 , while blocking TNFα-induced recruitment of SIRT1 to the IL-6 promoter . Further , several approaches established that activation of the cAMP or AMPK pathways were not required for resveratrol-directed suppression of IL-6 , and in fact , forskolin strongly induced IL-6 expression . Thus , resveratrol regulates SIRT1 through several possible mechanisms , including via ERα , as established here . This polypharmacology likely accounts for the unique health benefits of resveratrol in different preclinical models . For example , in the muscle the beneficial metabolic effects of resveratrol may be via ERα-directed induction of Glut4 and increased glucose uptake ( Deng et al . , 2008 ) , up-regulation of cAMP signaling ( Park et al . , 2012 ) , PGC-1α expression ( Pfluger et al . , 2008 ) , mitochondrial biogenesis ( Price et al . , 2012 ) , and activation of the AMPK ( Patel et al . , 2011; Price et al . , 2012 ) or PPARγ ( Ge et al . , 2007 ) . Thus , dissecting the effects of resveratrol require consideration of several potential signaling pathways , as well as tissue context . This work advances our understanding of resveratrol , which acts through ERα to modulate the inflammatory response , without the proliferative effects of estradiol . Therefore , this work will impact future medicinal chemistry efforts to improve the potency or efficacy of resveratrol .
MCF-7 and T47D cells were cultured in growth medium containing Dulbecco's minimum essential medium ( DMEM ) ( Cellgro by Mediatech Inc , Manassas , VA ) plus 10% fetal bovine serum ( FBS ) ( Hyclone by Thermo Scientific , South Logan , UT ) , and 1% each of nonessential amino acids ( NEAA ) ( Cellgro ) , Glutamax and Penicillin-streptomycin-neomycin ( PSN ) antibiotics mixture ( Gibco by Invitrogen Corp . Carlsbad , CA ) and maintained at 37°C and 5% CO2 . For each experiment , MCF-7 cells are seeded in growth medium for 24 hr . The medium was then replaced with steroid-free medium containing phenol red-free DMEM plus 10% charcoal/dextran-stripped ( cs ) FBS , and 1% each of NEAA , Glutamax and PSN , and the cells were incubated at 37°C for 48–72 hr before treatment . The cells were pre-treated with 1 μM ICI 182 , 780 ( ICI ) or 1 μM in solution AMPK inhibitor compound C/Dorsomorphin ( DOS ) ( Calbiochem , EMD Millipore Corp . Billerica , MA ) . The cells were treated simultaneously with the following , unless otherwise indicated: 10 ng/ml human tumor necrosis factor alpha ( TNFα; Invitrogen ) , and 10 nM E2 , 10 μM resveratrol ( RES ) , 25 μM Rolipram ( ROL ) , or 10 μM Forskolin ( FSK ) ( Sigma–Aldrich Inc . , St . Louis , MO ) . MCF-7 cells were transfected with a widely used 3xERE-luciferase reporter and luciferase activity was measured as previously described ( Wang et al . , 2012 ) . MCF-7 cells were placed in 384-well plates containing phenol red-free growth media supplemented with 5% charcoal-dextran sulfate-stripped FBS , and stimulated with ER ligands the next day , using a 100 nl pintool Biomeck NXP workstation ( Beckman Coulter , Inc . ) . After 3 days , the treatments were repeated . The number of cells/well was determined using CellTitre-Glo reagent ( Promega Corp . , Madison , WI ) as previously described ( Srinivasan et al . , 2013 ) , 7 days after the initial treatment . HEK293-T cells were transfected with ERα-VP16 and either GAL4-SRC1 , GAL4-SRC2 or GAL4-SRC3 and GAL4-UAS-Luciferase using TransIT LT1 transfection reagent ( Mirus Bio LLC , Madison , WI ) , processed and analyzed as previously described ( Srinivasan et al . , 2013 ) . Aliquots of media conditioned by stimulated MCF-7 and RAW264 . 7 macrophages were respectively analyzed using human IL-6 or mouse Il-6 AlphaLISA no-wash ELISA kits ( PerkinElmer , Inc . , Shelton , CT ) , as previously described ( Srinivasan et al . , 2013 ) . Total RNA was isolated using the RNeasy Mini kit ( Qiagen Inc . , Valencia , CA ) and submitted to the Scripps-Florida genomics core for cDNA microarray analysis using Affymetrix Genechip Human Gene ST arrays . For high-throughput , quantitative RT-PCR ( qPCR ) , 1 μg of total RNA per sample was reverse-transcribed in a 20-μl reaction using a High capacity cDNA kit ( Applied Biosystems , Carlsbad , CA ) . 1 μl of the resulting cDNA mixture was amplified in a 10 μl reaction using gene-specific primers ( Tables 2 and 3 ) in 1x Taqman or SYBR green PCR mixes ( Applied Biosystems ) . Data were analyzed using the ΔΔCT method as previously described ( Bookout and Mangelsdorf , 2003 ) and GAPDH expression as an endogenous control ( Product number: 4333764F; Applied Biosystems ) . 10 . 7554/eLife . 02057 . 032Table 2 . Gene-specific qPCR primersDOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 032GeneForward ( 5′-3′ ) Reverse ( 5′-3′ ) CTBP1CTCAATGGGGCTGCCTATAGGGACGATACCTTCCACAGCADBC1GATCCACACACTGGAGCTGATGGCTGAGAAACGGTTATGGG9aCTTCAGTTCCCGAGACATCCCGCCATAGTCAAACCCTAGCGLPGCTCGGGTTTGACTATGGAGCAGCTGAAGAGCTTGCCTTTGPR30CTGACCAAGGAGGCTTCCAGCTCTCTGGGTACCTGGGTTGHDAC1AAGGAGGAGAAGCCAGAAGCGAGCTGGAGAGGTCCATTCAHDAC2TCCAAGGACAACAGTGGTGAGTCAAATTCAGGGGTTGCTGHDAC3AGAGGGGTCCTGAGGAGAACGAACTCATTGGGTGCCTCTGLCoRCTCTCCAGGCTGCTCCAGTAACCACTCCGAAGTCCGTCTLRP16AGCACAAGGACAAGGTGGACCTCCGGTAGATGTCCTCGTCLSD1GGCTCAGCCAATCACTCCTATGTTCTCCCGCAAAGAAGAROCK1CCACTGCAAATCAGTCTTTCCATTCCACAGGGCACTCAGTCSIRT2TTGGATGGAAGAAGGAGCTGCATCTATGCTGGCGTGCTCSRC1CACACAGGCCTCTACTGCAATCAGCAAACACCTGAACCTGSRC2AGCTGCCTGGAATGGATATGAACTGGCTTCAGCAGTGTCASRC3GTGGTCACATGGGACAGATGTCTGATCAGGACCCATAGGC10 . 7554/eLife . 02057 . 033Table 3 . Inventoried TaqMan gene expression assays ( Applied Biosystems ) DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 033GeneAssay IDAMPKα1Hs01562315_m1AMPKα2Hs00178903_m1CBPHs00231733_m1CoRESTHs00209493_m1ERαHs01046812_m1ERβHs01100353_m1IL-6Hs00174131_m1NCoRHs01094540_m1NMNAT1Hs00978912_m1P300Hs00914223_m1PARP1Hs00242302_m1PCAFHs00187332_m1SIRT1Hs01009006_m1SMRTHs00196955_m1 The ERα ligand-binding domain containing an Y537S mutation was expressed in E . coli and purified as previously described ( Nettles et al . , 2008a ) . The protein solution was mixed with resveratrol and a receptor-interacting SRC2 peptide , and allowed to crystallize at room temperature . X-ray diffraction data on the crystal was collected at Stanford Synchrotron Radiation Lightsource beam line 11-1 . The structure was solved via automated molecular replacement and rebuilding of the genistein-bound ERα ( PDB 2QA8 ) ( Nettles et al . , 2008a ) , using the PHENIX software suite ( Adams et al . , 2010 ) . Ligand docking was followed by series of ExCoR and rebuilding as previously described ( Nwachukwu et al . , 2013 ) . F-resveratrol was added to dilute ERα ligand binding domain ( Y537S ) in 15 ml of buffer ( 20 mM Tris pH 8 . 0 , 150 mM NaCl , 5% glycerol , 15 mM BME ) , then concentrated , and 10% D2O added for a final protein concentration of 260 μM with 0 . 2% DMSO-d6 . 19F NMR was performed on a 700 MHz Bruker NMR spectrometer ( 19F @ 659 MHz ) without proton decoupling . Spectra were referenced to KF in buffer ( set to 0 ppm ) using a thin coaxial tube insertion . SRC peptide binding to the ERα ligand-binding domain ( LBD ) was examined using the LanthaScreen time-resolved fluorescence resonance energy transfer ( FRET ) ERα Coactivator Assay kit ( Invitrogen Corporation , Carlsbad , CA ) , as previously described ( Choi et al . , 2011 ) , but run in agonist mode . Specifically , 3 . 5 nM ERα-LBD-GST , 5 nM Terbium-tagged anti-GST antibody , fluorescein-tagged SRC peptides , and E2 or resveratrol were placed in triplicates in a 384-well plate , mixed , and incubated at room temperature for 1 hr in the dark . The FRET signals emitted upon excitation at 340 nm were read at 520 nm and 495 nm , and the emission ratio ( 520/495 ) from each well was calculated . Microarray assay for real-time nuclear receptor coregulator interaction ( MARCoNI ) was performed as previously described ( Aarts et al . , 2013 ) . In short , a PamChip peptide micro array with 154 unique coregulator-derived NR interaction motifs ( #88101; PamGene International ) was incubated with His-tagged ERα LBD in the presence of 10 μM E2 or A-CD ring estrogen , 100 μM resveratrol , or solvent only ( 2% DMSO , apo ) . Receptor binding to each peptide on the array was detected using fluorescently labeled His-antibody , recorded by CCD and quantified . Per compound , three technical replicates ( arrays ) were analyzed to calculate the log-fold change ( modulation index , MI ) of each receptor–peptide interaction vs apo . Significance of this modulation was assessed by Student's t test . MCF-7 cells were placed in a 24-well plate at a density of 50 , 000 cells/well for 24 hr . The next day , cells were transfected with 100 nM siRNAs ( Tables 4 and 5 ) using X-tremeGENE siRNA transfection reagent ( Roche Applied Science , Indianapolis , IN ) . For each well , a 25-μl mixture containing 2 . 5 μl X-tremeGENE + 22 . 5 μl Opti-MEM ( Invitrogen ) was added to a 25-μl solution of siRNA + Opti-MEM , mixed and incubated at room temperature for 20 min , and then added to cells in 0 . 45 ml Opti-MEM . After 6 hr , the media was replaced with steroid-free media and left for 48 hr before ligand stimulation . 10 . 7554/eLife . 02057 . 037Table 4 . Flexitube siRNAs ( Qiagen ) DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 037siRNAGene IDCatalog No . AMPKα15562SI02622228AMPKα25563SI02758595CoREST23 , 186SI03137435CtBP11487SI03211201DBC157 , 805SI00461846GLP79 , 813SI02778923G9a10 , 919SI00091189ERα2099SI02781401ERβ2100SI03083269GPR302852SI00430360HDAC13065SI02663472HDAC23066SI00434952HDAC38841SI00057316LCOR84 , 458SI00143213LRP1628 , 992SI00623658LSD123 , 028SI02780932NMNAT164 , 802SI04344382P3002033SI02622592PARP1142SI02662996SIRT123 , 411SI04954068SIRT222 , 933SI02655471SRC18648SI00055342SRC210 , 499SI00089509SRC38202SI0008936910 . 7554/eLife . 02057 . 038Table 5 . ON-TARGETplus SMARTpool siRNAs ( Thermo Scientific Dharmacon , Lafayette , CO ) DOI: http://dx . doi . org/10 . 7554/eLife . 02057 . 038siRNAGene IDCatalog No . PCAF8850L-005055-00CBP1387L-003477-00NCoR9611L-003518-00SMRT9612L-020145-00 AllStars negative control ( siControl , Qiagen Inc . ) . MCF-7 cells were seeded in a 24-well plate at a density of 50 , 000 cells/well in growth medium for 24 hr . The medium was then replaced with steroid-free medium for 48 hr . The cells were stimulated with the indicated doses of resveratrol . After 5 min , the cells were washed with cold PBS , disrupted in 100 μl NAD extraction buffer , and analyzed using the EnzyChrom NAD+/NADH Assay kit ( BioAssay Systems , Hayward , CA ) . MCF-7 cells in a 12- or 24-well plate were fixed , and washed with cold 1X PBS . 400 μl/well of cold lysis buffer was added to the cells which were then incubated at 4°C for 1 hr . Whole cell lysates were transferred to a 1 . 5-ml tube for sonication . For each IP , 100 μl aliquots of sonicated lysate was mixed with antibody and 25 μl Dynabeads protein G ( Invitrogen ) to make a 200 μl lysis buffer mixture that was rotated for 24 hr at 4°C . The precipitate was washed sequentially in previously described low salt , high salt , and LiCl buffers ( Nwachukwu et al . , 2007 ) and twice in 1x TE buffer , after which the crosslinks were reversed . DNA fragments were isolated using QIAquick PCR purification kit ( Qiagen ) , and analyzed by qPCR using Taqman 2x PCR master mix and a custom FAM-labeled promoter probes ( Applied Biosystems ) . | Resveratrol is a compound found in significant quantities in red wine , grapes , and peanuts . Many health benefits have been linked to it , including protecting against certain types of cancer and reducing the risk of cardiovascular disease . How resveratrol could produce these very different effects is unknown , but evidence is emerging that it is involved in a wide range of biological processes . However , the ability of resveratrol to bind with , and activate , proteins called estrogen receptors has largely been overlooked . These receptors have a range of roles . For example , estrogen receptors fight against inflammation by preventing the transcription of the gene that encodes a signaling protein called interleukin-6 . However , estrogen receptors do not work alone: other molecules called coregulators interact with them and alter how effectively they can prevent gene expression . Resveratrol has also been associated with anti-inflammatory effects , particularly in tissues that contain large numbers of an estrogen receptor called ERα , though this connection has been little studied . Nwachukwu et al . now reveal that the two are linked—the anti-inflammatory response of resveratrol relies on it being bound to ERα . This binding changes the shape of the receptor in a way that controls which coregulator molecules help it to regulate transcription . Additionally , this binding complex does not produce the cancer-causing side effects often associated with activated ERα . Nwachukwu et al . also found that the effect of resveratrol on the inflammatory response depends on specific other coregulators being present . The role of ERα in enhancing and activating resveratrol's effects is important because resveratrol has poor bioavailability in humans , and so it is not easily absorbed into the bloodstream . This makes it difficult for someone to get a dose high enough to produce beneficial effects . Further research targeting ERα may produce similar beneficial compounds to resveratrol , but with improved bioavailability . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Resveratrol modulates the inflammatory response via an estrogen receptor-signal integration network |
Vesicle fusion is mediated by an assembly of SNARE proteins between opposing membranes , but it is unknown whether transmembrane domains ( TMDs ) of SNARE proteins serve mechanistic functions that go beyond passive anchoring of the force-generating SNAREpin to the fusing membranes . Here , we show that conformational flexibility of synaptobrevin-2 TMD is essential for efficient Ca2+-triggered exocytosis and actively promotes membrane fusion as well as fusion pore expansion . Specifically , the introduction of helix-stabilizing leucine residues within the TMD region spanning the vesicle’s outer leaflet strongly impairs exocytosis and decelerates fusion pore dilation . In contrast , increasing the number of helix-destabilizing , ß-branched valine or isoleucine residues within the TMD restores normal secretion but accelerates fusion pore expansion beyond the rate found for the wildtype protein . These observations provide evidence that the synaptobrevin-2 TMD catalyzes the fusion process by its structural flexibility , actively setting the pace of fusion pore expansion .
SNARE-mediated membrane fusion comprises a series of mechanistic steps requiring both protein-protein as well as protein-lipid interactions . Protein-protein interactions involving SNARE proteins in the fusion process have been explored in great detail ( Jahn and Fasshauer , 2012; Sudhof and Rothman , 2009 ) , but the functional role of SNARE-lipid interplay has remained enigmatic . Previous studies provided conflicting views on the requirement of proteinaceous membrane anchors of SNARE proteins for efficient neurotransmitter release or vacuole-vacuole fusion ( Chang et al . , 2016; Fdez et al . , 2010; Grote et al . , 2000; Pieren et al . , 2015; Rohde et al . , 2003; Wang et al . , 2004; Zhou et al . , 2013 ) . Even more unclear is how a proteinaceous TMD may regulate the membrane fusion process . Experiments in reduced model systems have suggested that lipidic SNARE-anchors are inefficient in driving proper fusion between artificial liposomes ( McNew et al . , 2000 ) , cells expressing ‘flipped’ SNAREs ( Giraudo et al . , 2005 ) , or between liposomes and lipid nanodiscs ( Bao et al . , 2015; Shi et al . , 2012 ) . However , these experiments were unable to track kinetic intermediates en route to fusion ( e . g . priming , triggering or fusion pore expansion ) leaving the questions unanswered whether and if so , at which step TMDs of SNARE proteins may regulate fast Ca2+-triggered exocytosis and membrane fusion ( Fang and Lindau , 2014; Langosch et al . , 2007 ) . In comparison to other single-pass transmembrane proteins , SNARE TMDs are characterized by an overrepresentation of ß-branched amino acids ( e . g . valine and isoleucine , ~40% of all residues [Langosch et al . , 2001; Neumann and Langosch , 2011] ) , which renders the helix backbone conformationally flexible ( Han et al . , 2016; Quint et al . , 2010; Stelzer et al . , 2008 ) . In an α-helix , non-ß-branched residues like leucine can rapidly switch between rotameric states , which favor van der Waals interactions with their i ± 3 and i ± 4 neighbors , thereby forming a scaffold of side chain interactions that defines helix stability ( Lacroix et al . , 1998; Quint et al . , 2010 ) . Steric restraints acting on the side chains of ß-branched amino acids ( like valine and isoleucine ) instead favor i ± 4 over i ± 3 interactions leading to local packing deficiencies and backbone flexibility . In vitro experiments have suggested that membrane-inserted short peptides mimicking SNARE TMDs ( without a cytoplasmic SNARE motif ) exhibit a significant fusion-enhancing effect on synthetic liposomes depending on their content of ß-branched amino acids ( Hofmann et al . , 2006; Langosch et al . , 2001 ) . Furthermore , simulation studies have shown an inherent propensity of the SNARE TMDs or the viral hemagglutinin fusion peptide to disturb lipid packing , facilitating lipid splay and formation of an initial lipid bridge between opposing membranes ( Kasson et al . , 2010; Markvoort and Marrink , 2011; Risselada et al . , 2011 ) . Here , we have investigated the functional role of the synaptobrevin-2 ( syb2 ) TMD in Ca2+-triggered exocytosis by systematically mutating its core residues ( amino acid positions 97–112 ) to either helix-stabilizing leucines or flexibility–promoting ß-branched isoleucine/valine residues . In a gain-of-function approach TMD mutants were virally expressed in v-SNARE deficient adrenal chromaffin cells ( dko cells ) , which are nearly devoid of exocytosis ( Borisovska et al . , 2005 ) . By using a combination of high resolution electrophysiological methods ( membrane capacitance measurements , amperometry ) and molecular dynamics simulations , we have characterized the effects of the mutations in order to delineate syb2 TMD functions in membrane fusion . Our results indicate an active , fusion promoting role of the syb2 TMD and suggest that structural flexibility of the N-terminal TMD region catalyzes fusion initiation and fusion pore expansion at the millisecond time scale . Thus , SNARE proteins do not only act as force generators by continuous molecular straining , but also facilitate membrane merger via structural flexibility of their TMDs . The results further pinpoint a hitherto unrecognized mechanism wherein TMDs of v-SNARE isoforms with a high content of ß-branched amino acids are employed for efficient fusion pore expansion of larger sized vesicles , suggesting a general physiological significance of TMD flexibility in exocytosis .
To study the potential impact of structural flexibility of the syb2 TMD on fast Ca2+-dependent exocytosis , we substituted all core residues of the syb2 TMD with either leucine , valine or isoleucine ( Figure 1A ) and measured secretion as membrane capacitance increase in response to photolytic uncaging of intracellular [Ca]i . Replacing the syb2 TMD by a poly-leucine helix ( polyL ) strongly reduced the ability of the syb2 mutant to rescue secretion in v-SNARE deficient chromaffin cells ( Figure 1B ) . Indeed , a detailed kinetic analysis of the capacitance changes revealed that both components of the exocytotic burst , the rapidly releasable pool ( RRP ) and the slowly releasable pool ( SRP ) , were similarly diminished , and the sustained rate of secretion was reduced , but no changes in exocytosis timing were observed ( Figure 1B ) . The similar relative decrease in both , the RRP and the SRP component , could indicate that the polyL mutation interferes with upstream processes like the priming reaction leading to impaired pool formation and reduced exocytosis competence . By studying SNARE complex assembly with recombinant proteins , we found that the polyL variant affects neither the rate nor the extent of SNARE complex formation ( Figure 1—figure supplement 1 ) . This renders the possibility unlikely that the mutant syb2 TMD allosterically affects the upstream SNARE motif leading to altered interaction with its cognate SNARE partners . Thus , the secretion deficiency in polyL expressing cells is not due to impaired SNARE complex formation , i . e . by causing changes in vesicle priming , but rather reflects defective vesicle fusion . 10 . 7554/eLife . 17571 . 003Figure 1 . Helix stabilizing amino acids in the syb2 TMD diminish secretion . ( A ) Schematic representation of syb2 and corresponding TMD mutants ( polyL , polyV , polyI , polyLV ) . ( B–E ) Mean flash-induced [Ca2+]i levels ( top panels ) and corresponding CM responses ( middle panels ) of dko cells expressing syb2 wt , polyL , polyV , polyI or polyLV mutants . The polyL mutation reduced RRP and SRP size as well as sustained rate of release , whereas other substitutions of the TMD core residues with valine , isoleucine or a combination of leucine and valine fully restored exocytosis ( bottom panels ) . The kinetics of release τRRP , τSRP and the secretory delay are unchanged for all mutants . Arrow indicates flash . Data are represented as mean ± SEM and numbers of cells are indicated within brackets . ***p<0 . 001 , Mann Whitney U test versus syb2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 00310 . 7554/eLife . 17571 . 004Figure 1—figure supplement 1 . The poly-L mutant forms SDS-resistant SNARE complexes like the wildtype protein . Time-dependent SNARE complex formation between GST-syb2 ( A ) or GST-syb2-polyL ( B ) and their SNARE partners syntaxin 1 ( amino acids 1–262 ) and SNAP25 ( amino acids 1–206 ) . SNARE complexes were formed by mixing approximately equal molar amounts ( ~5 µM ) of the proteins and incubating at 25°C for the indicated times . The ability of SNARE proteins to form SDS-resistant complexes was analyzed by SDS-PAGE . Exemplary Coomassie-stained SDS gels are shown . ( C ) Quantification of SDS-resistant SNARE complex binding at different times after mixing the proteins . The polyL mutation affects neither the rate nor the extent of SNARE complex formation rendering the possibility unlikely that the altered TMD allosterically affects the upstream SNARE motif to interact with its cognate SNARE partners . Data were collected from four independent experiments and are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 00410 . 7554/eLife . 17571 . 005Figure 1—figure supplement 2 . Substitution of conserved amino acids within the syb2-TMD does not affect vesicle fusion . ( A–C ) Mean flash-induced [Ca2+]i levels ( top panels ) and corresponding CM responses ( middle panels ) of dko cells expressing syb2 or V101A ( pink ) or V112A ( purple ) or G100L ( light pink ) mutants . Numbers of cells analyzed are indicated within the brackets . Neither magnitudes of RRP and SRP nor kinetics of release are changed for the mutant proteins ( bottom panel , Mann Whitney U Test versus syb2 ) . ( D ) Mean capacitance response upon intracellular perfusion with 19 µM free Ca2+ for the indicated groups ( left panel ) . Total ∆CM and amperometric event frequency measured over 120 s ( middle panel ) from 68 dko+syb2 , 25 dko+V101A , 22 dko+V112A and 21 dko+G100L cells . The amperometric event frequency scales proportionally to ΔCM in syb2 or its mutant variants expressing dko cells ( right panel ) , indicating that alterations in CM are due to changes in granule exocytosis and that mutants of syb2 do not cause premature closure of fusion pore . Continuous lines indicate linear regression . ( E ) Properties of the main amperometric spikes , displayed as cumulative frequency distribution for syb2 ( 8005 events ) , V101A ( 2885 events ) , V112A ( 2481 events ) , G100L ( 2386 events ) show unchanged spike charge , amplitude , rise time and half width in the mutant variants compared to control . Data are represented as mean ± SEM , one-way analysis of variance versus control . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 00510 . 7554/eLife . 17571 . 006Figure 1—figure supplement 3 . Syb2 and its TMD mutants are sorted to chromaffin granules with similar efficiency . ( A ) Exemplary immunostainings ( imaged with SIM ) for ceb ( green ) and syb2 ( red ) in littermate control chromaffin cells and syb2 ko cells expressing syb2 or one of the TMD variants ( top to bottom: polyL , polyV , polyI , polyL-Ct and polyL-Nt ) . Cells were imaged within the footprint area to minimize the contribution of ER/Golgi-derived fluorescence in virus-transfected cells . Syb2 fluorescence signals in littermate control cells were excited with five-fold higher laser power than in virus-transfected cells ( x5 ) . The merged images and their magnified view display a clear colocalization between ceb and syb2 ( or the TMD mutants ) , as also illustrated in the corresponding line scans ( magnified view , dashed lines; pixel size , 40 nm ) . ( B ) Mander’s weighted colocalization of syb2 or its TMD mutants to endogenous ceb indicates a similar colocalization coefficient ( ~80% ) for all the tested mutant variants . The total fluorescence intensity of single syb2 immunopositive puncta is similar for syb2 and the TMD mutants and five-fold higher than the littermate control ( wt ) cells . Images were thresholded to values 6xSD of the background fluorescence to isolate discrete regions of interest ( ROIs ) . Similar-sized ROIs in wildtype cells ( 0 . 0558 ± 0 . 0024 µm²; 9 cells; 1042 ROIs ) and dko cells expressing syb2 ( 0 . 0552 ± 0 . 0019 µm²; 10 cells; 1087 ROIs ) or the TMD mutants: polyL ( 0 . 0517 ± 0 . 0031 µm²; 10 cells; 2011 ROIs ) , polyV ( 0 . 0527 ± 0 . 0028 µm²; 15 cells; 2021 ROIs ) , polyI ( 0 . 0562 ± 0 . 0022 µm²; 12 cells; 1981 ROIs ) , polyL-Ct ( 0 . 0551 ± 0 . 0026 µm²; 10 cells; 970 ROIs ) , polyL-Nt ( 0 . 0610 ± 0 . 0019 µm²; 11 cells; 1025 ROIs ) were analyzed . ( C ) Exemplary fluorescence profiles of discrete syb2 immunopositive puncta analyzed by z-stacking reveal singular fluorescence peaks consistent with the vesicular origin of the immunosignals . ( D ) Exemplary images of a dko+gfp , ceb ko ( ctrl , littermate control ) +gfp and dko cells overexpressing either syb2 or the mutated v-SNAREs proteins . In dko cells expressing syb2 or its mutant variants , signals are much stronger than in ctrl+gfp cells and are visualized after adjustment of the camera’s exposure time ( dko+gfp and ctrl+gfp , 3 . 9 s; dko+v-SNARE , 0 . 31 s ) . ( E ) Mean total fluorescence intensity of dko+gfp , ctrl+gfp and dko cells expressing syb2 or the indicated mutants ( determined 5 . 5 hr after transfection ) . Note that expression of syb2 or its mutants in dko cells leads to ~ten fold increase in protein level when compared with the wt signal . Data are normalized to the immunosignal of dko+syb2 cells . Data are represented as mean ± SEM . ***p<0 . 001 , one-way analysis of variance versus control . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 006 In contrast , replacing the core residues of the syb2 TMD with either a poly-valine ( polyV ) or poly-isoleucine ( polyI ) helix resulted in mutants that support exocytosis like the wildtype protein ( Figure 1C , D ) . Thus , substitution of a substantial amount of amino acids within the syb2 TMD with either type of ß-branched residue is tolerated without affecting secretion ( Figure 1A , C , D ) . Since both , polyV and polyI mutants can functionally replace the wildtype protein , it seems likely that membrane fusion does not critically depend on conserved key residues at specific positions within the syb2 TMD . To substantiate this hypothesis , we substituted single highly conserved TMD amino acids , the G100L , or those residues that remain unchanged in the polyV mutant ( syb2 V101A and syb2 V112A , Figure 1A ) . None of these mutations interfered with the Ca2+-triggered secretion response ( Figure 1—figure supplement 2 ) . Moreover a variant , in which all TMD core residues were substituted by an alternating sequence of leucine and valine ( denoted polyLV ) in order to match the ~50% ß-branched amino acid content of the syb2 TMD , also rescued secretion like the wildtype protein ( Figure 1E ) . In control experiments we further confirmed by epifluorescence and high resolution structured illumination microscopy ( SIM ) that the syb2 TMD mutant proteins were correctly sorted to chromaffin granules and expressed with similar efficiency as the wildtype protein ( Figure 1—figure supplement 3 ) . The strong functional differences seen in Ca2+-triggered exocytosis when replacing the TMD core by leucines and isoleucines ( or valines , respectively ) are remarkable , given that these aliphatic amino acids hardly deviate in their physicochemical properties regarding hydrophobicity ( Kyte-Doolittle scale: Leu 3 . 8 , Ile 4 . 5 , Val 4 . 2 ) and side chain volume ( Leu 168 Å , Ile 169 Å , Val 142 Å ) . However , an attractive explanation for the different secretory effects of the amino acids is delivered by their different side chain mobility ( Leu > Ile/Val ) , thereby influencing side chain to side chain interactions and TMD back bone dynamics ( Quint et al . , 2010 ) , as will be further explored below ( Figure 4 ) . Taken together , the combined set of mutant phenotypes supports the view that the changes in overall structural flexibility of the TMD , rather than a requirement of specific residues at key positions , determine the exocytotic response by changing vesicle fusogenicity , pointing to an active role of the v-SNARE TMD in membrane fusion . Analysis of tonic secretion ( evoked by continuous intracellular perfusion with solution containing 19 µM free calcium ) with simultaneous membrane capacitance ( CM ) measurements and carbon fiber amperometry independently confirmed our observations that the polyL mutant diminishes exocytosis , whereas the polyI and polyV variants support secretion at wildtype levels ( Figure 2A–D ) . The close correlation between the results of both types of secretion measurements for syb2 and its mutant variants ( slope: syb2 0 . 18 events/fF , r² = 0 . 97; polyL 0 . 17 events/fF , r² = 0 . 97; polyV 0 . 17 events/fF , r² = 0 . 95; polyI 0 . 17 events/fF , r² = 0 . 94 ) shows that the observed CM changes are due to alterations in exocytosis of catecholamine-containing granules . They further render the possibility unlikely that mutant-mediated changes of the CM signal are due to premature closure of the fusion pore and interference with subsequent vesicle endocytosis ( Deak et al . , 2004; Rajappa et al . , 2016; Xu et al . , 2013 ) . 10 . 7554/eLife . 17571 . 007Figure 2 . Modifying the number of ß-branched residues in the syb2 TMD changes the kinetics of cargo discharge . ( A ) Schematic representation of syb2 and its TMD mutants ( polyL , polyV , polyI ) . ( B ) Exemplary recordings of CM and amperometry for dko cells expressing syb2 or the polyL mutant ( dashed lines indicate cell opening initiating intracellular perfusion with 19 µM free Ca2+ ) . ( C ) Mean capacitance responses over 120 s . ( D ) Total ∆CM after 120 s ( top ) and amperometric event frequency ( bottom ) averaged from the indicated number of cells . ( E ) Properties of the main amperometric spike , displayed as cumulative frequency distribution for the indicated parameters . Exemplary amperometric events with similar charge for the indicated groups are shown ( right ) . ( F ) The polyL mutant prolonged and the polyV or polyI mutations accelerated spike kinetics ( causing corresponding changes in spike amplitude ) without affecting quantal size . Values are given as mean of median determined from the parameter’s frequency distribution for each cell . Data were collected from cells/events measured for syb2 ( 61/7275 ) , polyL ( 22/1018 ) , polyV ( 18/2431 ) , polyI ( 30/2789 ) . Only cells with >20 events were considered . Data are represented as mean ± SEM . **p<0 . 01 , ***p<0 . 001 , one way analysis of variance versus syb2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 007 Carbon fiber amperometry allows for resolution of discrete phases of transmitter discharge from single vesicles , comprising a prespike signal that reflects transmitter release through the narrow initial fusion pore and a main amperometric spike that coincides with bulk release ( Albillos et al . , 1997; Bruns and Jahn , 1995; Chow et al . , 1992 ) . The polyL variant not only lowered the frequency of exocytotic events but also profoundly slowed transmitter release from the vesicle , compatible with the phenotype of a fusion mutant . The release events were characterized by a decreased amplitude and increased rise-time as well as half-width of the amperometric signal ( Figure 2E , F ) . In clear contrast , expression of either polyV or polyI variant accelerated catecholamine release compared to controls , as indicated by significantly higher spike amplitudes , reduced rise-times and half-width values ( Figure 2E , F ) . Evidently , modifying the content of ß-branched amino acids within the TMD causes correlated changes in spike waveform , even producing a gain-of-function phenotype in pore expansion kinetics for TMDs enriched in ß-branched residues . Moreover , TMD mutations also affected the prespike signal and its current fluctuations , which report transient changes in neurotransmitter flux through the early fusion pore ( Kesavan et al . , 2007 ) . The polyL mutation prolonged the expansion time of the initial fusion pore , lowered its current amplitude and diminished fluctuations in the signal time-course compared with the wildtype protein ( Figure 3 ) . The polyV and polyI variants shortened prespike duration , increased its amplitude , and current fluctuations . Taken together , the polyL and poly I/V mutants oppositely affect both , the prespike and the spike phase of transmitter discharge , implying that conformational properties of the syb2 TMD govern the fusion process from the opening of the nascent fusion pore to its final expansion . 10 . 7554/eLife . 17571 . 008Figure 3 . PolyL and polyV ( or I ) mutations oppositely alter the kinetics of prespike signals . ( A ) Exemplary prespike events and analysis of their current fluctuations ( highlighted area ) during transmitter discharge through a narrow pore . Deflections of the current derivative ( red trace ) above the threshold ( dashed lines = ± 4 SD of base line noise ) were counted as fluctuations ( blue trace ) . The displayed events have a similar total charge and 50%–90% rise time ( dko+syb2: 477fC , 280 µs; dko+polyL: 458fC , 240 µs; dko+polyV: 462fC , 200 µs; dko+polyI: 483fC , 200 µs ) , indicating that the different fluctuation behavior is not due to differences in diffusional broadening of the current signals . ( B ) PolyL mutation and polyV or I mutations oppositely altered the amplitude and kinetics of the prespike event , without changing its charge . ( C ) The average fluctuation frequency ( sum of positive and negative fluctuations ) of all events with an amplitude >7 pA as well as of events with spike rise times <340 ms ( minimizing a potential distortion of the signal time course by diffusional broadening ) decreased for the polyL mutant and increased for the polyV/I mutants . Mean rms noise of the current derivative during the prespike signal , serving as threshold-independent parameter of fusion pore jitter , confirms mutant protein-mediated changes in fusion pore dynamics . ( D–E ) Cumulative frequency distributions for fluctuation frequency and rms noise of current derivative from dko+syb2 ( black ) , dko+polyL ( red ) , dko+polyV ( green ) and dko+polyI cells ( blue ) . Data were collected from the indicated number of events ( cells ) and are represented as mean ± SEM . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , one way analysis of variance versus syb2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 008 Our mutational analysis suggested that changes in the conformational properties of the TMD can cause characteristic fusion defects , thereby indicating a TMD-based mechanism supporting exocytosis . To further investigate this mechanism we studied the structure and dynamics of the TMD mutants using molecular dynamics simulations of the C-terminal region of syb2 ( residues 71–116 ) . Based on the X-ray crystallographic structure ( Stein et al . , 2009 ) , syb2 and its mutant variants were embedded in an asymmetric membrane ( mimicking the physiological lipid composition of synaptic vesicles [Sharma et al . , 2015; Takamori et al . , 2006] ) and structural flexibility was calculated from the root mean square fluctuation ( RMSF ) of the backbone atoms for each peptide ( Figure 4 ) . The results show that conformational flexibility of the TMD region is significantly lowered in the polyL and increased in the polyV variant compared with the wildtype protein . Similarly , changes in the root mean square displacement ( RMSD ) of the Cα-atoms relative to an ideal α-helix ( syb2 0 . 104 ± 0 . 004 nm; polyL 0 . 067 ± 0 . 003 nm , p<0 . 001; polyV 0 . 139 ± 0 . 005 nm , p<0 . 001 , one-way analysis of variance versus syb2 ) are paralleled by alterations in α-helix content of syb2 TMD ( 79 ± 0 . 56% ) and its variants ( polyL 83 ± 0 . 56% , and polyV 65 ± 5 . 6% ) . Taken together , changing the frequency of ß-branched residues within the syb2 TMD substantially varies conformational flexibility , which clearly correlates with alterations in the kinetics of the nascent fusion pore as well as in the spike waveform . Overall , these data provide strong evidence that structural features of v-SNARE TMDs are crucial for Ca2+-triggered exocytosis , enabling TMDs to actively promote the fusion process . 10 . 7554/eLife . 17571 . 009Figure 4 . Conformational flexibilities of syb2 and its mutant variants . ( A ) Snapshot from the atomistic simulation for inserted syb2 ( residues 71–116 ) in an asymmetric , self-assembled membrane ( cytoplasmic leaflet , top; intravesicular leaflet , bottom ) with the protein backbone depicted in cartoon representation . The phosphate atoms of lipid and the hydroxyl carbon of cholesterol are shown in the Van der Waals representation . Other atoms of the lipids are shown as grey lines ( water molecules are not shown for clarity ) . Different lipid moieties are depicted according to the colour code shown below . Note the asymmetric lipid composition of the bilayer . ( B ) Root mean square fluctuations ( RMSF ) of C-α atoms derived from 40 ns simulation runs for syb2 and the mutants relative to the average structure of the corresponding peptide . Flexibility of the TMD region ( residues 97–112 ) was determined from non-overlapping 10 ns periods during the last 30 ns of the trajectories for each simulation . The polyV region of the mutant showed on average an increased flexibility ( 0 . 0644 ± 0 . 0029 nm , p<0 . 001 ) , while the polyL region ( 0 . 0374 ± 0 . 0017 nm , p<0 . 05 ) showed a decreased jitter when compared with syb2 ( 0 . 0472 ± 0 . 0022 nm , one-way analysis of variance versus syb2 ) . Error bars are s . e . m . of averages calculated from the values of the individual 10 ns windows ( n = 9 for wt , n = 6 for mutants ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 009 Membranes first fuse with their outer leaflets , transiting through a hemifused state , before complete merger ( continuity of both leaflets ) is reached . To study whether structural flexibility is required throughout the entire TMD region or preferentially in one leaflet , we selectively exchanged either half of the syb2 TMD with leucine residues ( Figure 5A ) . For tonic secretion ( intracellular perfusion with high Ca2+-containing solution , Figure 5B ) and synchronized exocytosis ( photolytic uncaging of intracellular Ca2+ , Figure 5—figure supplement 1 ) , we found that leucine substitution within the N-terminal half of the TMD ( amino acid 97–104 , polyL-Nt , spanning the outer leaflet of the vesicle membrane ) failed to fully rescue exocytosis . A similar replacement of amino acids in the corresponding C-terminal half ( amino acids 105–112 , polyL-Ct , spanning the inner leaflet of the vesicle membrane ) was without any effect when compared with the wildtype protein ( Figure 5B and Figure 5—figure supplement 1 ) . Furthermore , an exchange of the N-terminal amino acids with ß-branched isoleucines ( polyI-Nt ) rescued secretion like the wildtype protein . 10 . 7554/eLife . 17571 . 010Figure 5 . Structural flexibility of the N-terminal TMD region catalyzes fusion initiation and fusion pore dilation . ( A ) Schemes of syb2 and corresponding TMD mutants ( polyL-Nt , polyL-Ct , polyLV , polyI-Nt ) . ( B ) Mean capacitance changes in response to intracellular perfusion with 19 µM free Ca2+ in the indicated groups . Total ∆CM ( top ) and amperometric event frequency ( bottom ) measured over 120 s show that only polyL-Nt mutant fails to rescue normal exocytosis . Data are averaged from the indicated number of cells . ( C ) Properties of the main amperometric spikes displayed as cumulative frequency distribution for the indicated parameters and color coded according to ( A ) . ( D ) Exemplary amperometric events with similar charge but altered release profile for the indicated syb2 variant . ( E ) PolyL-Nt mutation slowed the spike waveform ( reduced amplitude , increased rise time , and half width ) while polyI-Nt increased the amplitude and decreased the rise time and half width . Values are given as mean of median determined from the indicated parameter’s frequency distribution for each cell . Data were collected from cells/events measured for syb2 ( 83/9054 ) , polyL-Nt ( 19/951 ) , polyL-Ct ( 18/2684 ) , polyLV ( 25/3576 ) , polyI-Nt ( 21/2057 ) . Only cells with >20 events were considered . Data are represented as mean ± SEM . ***p<0 . 001 , one-way analysis of variance versus syb2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 01010 . 7554/eLife . 17571 . 011Figure 5—figure supplement 1 . Conformational flexibility of the N-terminal region of syb2 TMD supports LDCV fusion . ( A–C ) Mean flash-induced [Ca2+]i levels ( top panels ) and CM responses ( middle panels ) of dko+polyL-Nt , dko+polyL-Ct and dko+polyI-Nt versus the corresponding control ( dko+syb2 ) . RRP and SRP size as well as sustained rate of release and kinetics of release ( τRRP , τSRP and delay ) are plotted in the corresponding bottom panels . Evoked secretion is only reduced in the polyL-Nt mutant , while the kinetics of release is unchanged for all mutants ( bottom panels ) . Numbers of cells for each group are indicated within the brackets . Data are represented as mean ± SEM . ***p<0 . 001 , Mann Whitney U test versus syb2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 01110 . 7554/eLife . 17571 . 012Figure 5—figure supplement 2 . The N-terminal region of syb2 TMD controls kinetics and fluctuations of the early fusion pore . ( A ) The polyL-Nt mutant decreases the amplitude and increases the duration of the prespike signal , displayed as cumulative frequency distribution ( upper panel ) and cell weighted averages ( lower panel ) for the indicated parameters . polyI-Nt mutant oppositely affects the prespike parameters , whereas the polyL-Ct and the polyLV do not affect fusion pore behavior . None of the mutants changes the prespike charge . ( B ) The polyL-Nt reduces and the polyI-Nt enhances fusion pore dynamics , measured as fluctuation frequency and rms noise of the current derivative during the prespike signal . Data were collected from cells/events measured for syb2 ( 83/6517 ) , polyL-Nt ( 19/625 ) , polyL-Ct ( 18/2147 ) , polyLV ( 25/2538 ) , polyI-Nt ( 21/1613 ) . Data are represented as mean ± SEM . **p<0 . 01 , ***p<0 . 001 , one-way analysis of variance versus syb2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 012 Detailed analysis of amperometric events with respect to spike ( Figure 5C–E ) and prespike properties ( Figure 5—figure supplement 2 ) showed that exchanging the N-terminal half of the syb2 TMD with either leucine or isoleucine sufficed to reproduce the altered fusion pore behavior seen with an overall exchange of the TMD residues ( compare Figures 2 and 5 ) . For parameters describing the main spike kinetics we found a clear proportionality between the speed of catecholamine discharge and the number of ß-branched residues near the N-terminal end of the TMD helix ( rise-time , r2= 0 . 95; half-width r2 = 0 . 96 , Figure 6A , B ) . Moreover , the pre-spike duration is progressively shortened by increasing the fraction of ß-branched residue in the N-terminal half of the TMD ( r2 = 0 . 93 ) ( Figure 6C ) . Following the same line , we also found a strong correlation between the frequency of pre-spike current fluctuations of the mutant variants and their valine/isoleucine-content ( r2 = 0 . 96 ) . Evidently , increasing or decreasing the number of ß-branched amino acids in the TMD oppositely controls conformational flexibility in the TMD helix ( Figure 4 ) and the rate of cargo release from single vesicle . These findings provide strong evidence for a mechanistic link between TMD flexibility and the kinetics of fusion pore expansion . However , deviating from this pattern , total secretion in Ca2+-infusion and Ca2+-uncaging experiments was not further potentiated beyond the wildtype response by increasing the fraction of ß-branched residues within the N-terminal half of the TMD ( Figure 6D ) . Most likely , docking and priming reactions become rate-limiting ( Sorensen , 2009 ) , thereby preventing the total release to exceed wildtype levels . 10 . 7554/eLife . 17571 . 013Figure 6 . Speed of cargo release is systematically correlated with the number of β-branched amino acids in the N-terminal region of the syb2 TMD . ( A ) Schemes of syb2 and corresponding mutants depicting the fraction of β-branched amino acids in the N-terminal region of the TMD ( underlined ) . ( B–C ) Increasing the fraction of β-branched amino acids accelerates the rate of cargo release ( spike ) as well as the dynamics of the nascent fusion pore ( prespike ) . ( D ) Tonic and synchronous secretion are reduced with the loss of ß-branched amino acids but cannot be further potentiated by enriching ß-branched amino acids in the TMD N-terminal region when compared with syb2 . ( E ) Hypothetical models illustrating how conformational flexibility of the syb2 TMD ( specifically of the N-terminal region ) enhances lipid splay to promote intermembrane contact ( PM , plasma membrane , VM , vesicle membrane ) during fusion initiation and lowers negative membrane curvature ( outer leaflet ) to facilitate pore expansion . Data are represented as mean ± SEM . ***p<0 . 001 , one-way analysis of variance between indicated groups . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 013 Taken together , the syb2 TMD possesses an inherent functional polarity , with the N-terminal region being more important for fusogenicity than the C-terminal side . These observations agree well with previous coarse-grained models of SNARE-mediated fusion events ( Risselada et al . , 2011 ) , suggesting a similar directionality of SNARE TMDs in perturbing lipid packing ( enhancing lipid splaying ) preferentially in the cytoplasmic leaflets and , thereby , facilitating the first hydrophobic encounter for forming a lipid bridge between opposing membranes ( Figure 6E ) . Similarly , TMD backbone dynamics within the outer leaflet of the fusion pore neck may lower its high membrane curvature , driving fusion pore expansion ( Figure 6E ) . A partial rescue of synaptic transmission has previously been observed in cortical syb2-/- neurons expressing an acylated syb2-CSP fusion protein lacking the TMD ( Zhou et al . , 2013 ) . This finding has been interpreted as evidence that v-SNARE TMDs are functionally interchangeable with lipidic membrane anchors . Opposing this view , a recent study showed that the same lipid-anchored syb2 provides little support for spontaneous synaptic transmission ( Chang et al . , 2016 ) . We also found that this acylated syb2-CSP fusion protein was largely inefficient in reconstituting Ca2+-triggered exocytosis in chromaffin cells ( 21% of syb2 , Figure 7A , B ) , albeit showing similar expression levels and sorting to granules as the wildtype protein ( Figure 7—figure supplement 1 ) . Interestingly , while expression of syb2-CSP raises secretion significantly over the level of the dko ( 2% of syb2 ) , the phenotype is still more severe than the secretion deficits seen with the polyL variant ( 35% of syb2 , Figure 2C , D ) , reconfirming that the proteinaceous membrane anchor provides an autonomous facilitating function in Ca2+-triggered exocytosis . Furthermore , like the polyL mutant , the lipid-anchored syb2 prolonged the time course of transmitter discharge during the spike phase ( without changing the event charge ) ( Figure 7C ) and even more strongly slowed down kinetics of the early fusion pore ( Figure 7D–F ) . Collectively , these results highlight the important role of the proteinaceous syb2 membrane anchor in membrane fusion , generally facilitating fusion initiation and pore expansion . Our data obtained with the acylated syb2-CSP fusion protein appear to deviate from the previously reported results by Zhou et al . ( 2013 ) , wherein the mutant protein significantly rescued synaptic transmission compared to a syb2-RST-mVenus construct serving as control . However , as reported by Chang et al . ( 2016 ) , the syb2-RST-mVenus construct does not support wildtype like fusion both , in neurons and neuroendocrine cells due to the presence of a positively charged arginine residue in the C-terminal end of the syb2 TMD . Indeed , a previous study has shown that insertion of charged residues in the C-terminal end of the syb2 TMD impairs the release response ( Ngatchou et al . , 2010 ) . Consequently , the reduced ability of the syb2-RST-mVenus construct to rescue neuronal exocytosis may have led to an overestimation of the acylated syb2-CSP response providing an explanation for the apparently discrepant results . 10 . 7554/eLife . 17571 . 014Figure 7 . Lipid anchored syb2 failed to support normal secretion from chromaffin cells . ( A ) Mean capacitance responses upon intracellular perfusion with 19 µM free Ca2+ in the indicated groups . ( B ) Total ∆CM as well as amperometric event frequency measured over 120 s show that the lipid-anchored syb2 ( ΔTMD-CSP ) restored secretion above levels of dko cells but largely failed to support exocytosis like the wildtype protein indicating the functional necessity of a proteinaceous membrane anchor for unperturbed fusion . Data are averaged from the indicated number of cells . ANOVA followed by Kruskal-Wallis post hoc test was performed . ( C ) Properties of the amperometric spike phase , displayed as cell weighted averages show that the ΔTMD-CSP mutant decreased the spike amplitude while increasing the spike rise time and half width . ( D ) Effects of the ΔTMD-CSP mutation on the indicated prespike parameters . ( E–F ) Prespike fluctuations and rms noise of the current derivative are significantly reduced compared to control . Data were collected from events ( cells ) : dko+syb2 4267 ( 36 ) ; dko+ΔTMD-CSP 1078 ( 34 ) and are represented as mean ± SEM . ***p<0 . 001 , Mann Whitney U test versus syb2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 01410 . 7554/eLife . 17571 . 015Figure 7—figure supplement 1 . Lipid-anchored syb2 ΔTMD-CSP mutant exhibit similar expression and sorting to granules like the wildtype protein . ( A ) Exemplary images of a double knock-out ( dko ) +gfp , cellubrevin ko ( ctrl , littermate control ) +gfp and dko cells overexpressing either syb2 ( dko+syb2 ) or mutant protein ( dko+ΔTMD-CSP ) . Immunosignals were visualized after appropriate adjustment of the camera’s exposure time ( dko+gfp and cntl+gfp , 4 . 2 s; dko+v-SNARE , 0 . 31 s ) . ( B ) Mean total fluorescence intensity of dko+gfp , ctrl+gfp , dko+syb2 and dko+ΔTMD-CSP averaged from the indicated number of cells ( determined 5 . 5 hr after transfection ) . Expression of syb2 or its mutant in dko cells leads to ~ten fold increase in protein level when compared with the control signal ( ctrl-GFP ) . Data are normalized to the immunosignal of dko+syb2 cells . ( C ) Exemplary SIM images for ceb ( green ) and syb2 ( red ) in wildtype chromaffin cells and syb2 ko cells expressing syb2 or ΔTMD-CSP mutant . Syb2 fluorescence signals in wildtype cells were excited with five-fold higher laser power than in virus-transfected cells ( x5 ) . The merged images and their magnified inset display a clear colocalization between ceb and syb2 ( or the ΔTMD-CSP mutant ) , as also illustrated in the corresponding line scans ( magnified view , dashed lines; pixel size , 40 nm ) . ( D ) Mander’s weighted colocalization of syb2 or its TMD mutants to endogenous ceb indicates a similar colocalization coefficient ( ~77% ) for syb2 and ΔTMD-CSP mutant . Numbers of cells analyzed are indicated in the bar . ( E ) The fluorescence intensity of discrete immunopositive puncta is similar for syb2 and the mutant protein and five-fold higher than the littermate control ( wt ) cells . For analysis , images were thresholded to values 6xSD of the background fluorescence to isolate discrete regions of interest . Data are represented as mean ± SEM . *p<0 . 05 , ***p<0 . 001 , one-way analysis of variance between indicated groups ( B and E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 01510 . 7554/eLife . 17571 . 016Figure 7—figure supplement 2 . β–branched amino acids in the syb2 TMD regulate synaptic vesicle fusion . ( A ) Exemplary AP evoked EPSCs of wildtype hippocampal autaptic neurons ( black , n = 23 ) and syb2 ko neurons ( blue , n = 7 ) or syb2 ko neurons expressing the syb2 mutant variants ( polyL , red , n = 13 and polyV , green , n = 12 ) . ( B1–B3 ) Loss of syb2 abolishes evoked release ( syb2 ko ) . Expression of the polyV mutant in syb2ko neurons rescues EPSC amplitude and charge , while the polyL mutant significantly reduces the evoked amplitude and charge compared to control . The time to reach the maximum EPSC amplitude ( TTP ) is unaffected by the mutants when compared to wt . ( C ) Whole cell responses from wildtype ( black , n = 22 ) , syb2 ko ( blue , n = 4 ) , syb2 ko + polyL ( red , n = 9 ) and syb2 ko + polyV ( green , n = 11 ) neurons evoked by application of 500 mM hypertonic sucrose solution . ( D ) The readily releasable pool charge ( RRP , as quantified from hypertonic stimulation shown in C ) is fully ( polyV ) or only partially ( polyL ) restored to levels of wt cells . ( E ) Exemplary recordings of mEPSCs activity from syb2 ko neurons expressing syb2 , polyL or polyV mutant . ( F ) The polyL mutant reduces the mEPSC frequency compared to wildtype and ko+syb2 while polyV fully rescues release . ( G ) Amplitude of the quantal events is not changed in the TMD mutants compared to controls . Data for ( E–F ) are collected from the following number of cells: wildtype ( 38 ) , syb2 ko ( 8 ) , ko+syb2 ( 52 ) , ko+polyL ( 23 ) and ko+polyV ( 21 ) . ANOVA followed by Kruskal-Wallis post hoc test was performed . ( H ) Representative confocal images of syb2 ko neurons expressing either syb2 or its mutant variants . To illustrate the synaptic localization of the syb2 signal , neurons were co-stained for the synaptic vesicle marker protein Synaptophysin . ( I ) Quantification of the immunosignals for syb2 and its mutants . The average fluorescence intensity of syb2 is unchanged for the syb2 variants and reaches the levels observed for wildtype neurons . Note that the syb2 staining is absent in syb2 ko neurons ( wt , n = 18; syb2 ko , n = 14; syb2 ko+syb2 , n = 16; syb2 ko+polyL , n = 15; syb2ko+polyV , n = 18 ) . ( J ) The synapse density ( determined from the number of synaptophysin positive puncta per 50 µm neurite length ) is indistinguishable in wt cells or the syb2ko neurons expressing either syb2 or the mutant variants . Data are represented as mean ± SEM . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , one-way analysis of variance versus control . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 016 In any case , since SNARE-mediated fusion of SSVs might mechanistically deviate from granule secretion in neuroendocrine cells , we also analyzed the impact of our syb2 TMD mutants on fast glutamatergic release in autaptic cultures of syb2-/- hippocampal neurons . Viral expression of the polyV mutant rescued evoked synaptic transmission to the level of wildtype cells , while the polyL mutant largely failed to support neurotransmitter release ( Figure 7—figure supplement 2A–D ) , which is reminiscent of our findings in neuroendocrine cells . Immunofluorescence analyses confirmed that polyL and polyV mutant proteins were indeed targeted to synaptic vesicles with comparable efficiency as the wildtype protein ( Figure 7—figure supplement 2E–G ) . To test whether the flexibility of the syb2 TMD is also critical for quantal signaling , we recorded spontaneous excitatory postsynaptic currents ( mEPSCs ) in the presence of 1 µM TTX using mass cultures of hippocampal neurons . Compared with the wildtype syb2 protein , expression of the polyL mutant in syb2 ko neurons significantly reduced the frequency of spontaneous events , whereas the polyV mutant fully rescued spontaneous release ( ko+syb2: 1 . 46 ± 0 . 24 Hz , n = 52; ko+polyL: 0 . 49 ± 0 . 12 Hz , n = 23; ko+polyV: 1 . 42 ± 0 . 31 , n = 21 ) . Notably , the frequency of mEPSCs recorded for the polyL mutant is more than 20fold higher compared to syb2 ko neurons ( 0 . 02 ± 0 . 005 Hz , n = 8 ) , emphasizing the gain-of-function phenotype of the TMD mutant . In contrast , we failed to detect significant alterations in the mean amplitude of the polyL or polyV-mediated mEPSCs compared with the wildtype controls . Potential changes in the release profile of small synaptic vesicles ( SSVs ) may be masked by dendritic filtering of the receptor-mediated response . Moreover , release from SSVs might be less dependent on TMD-mediated acceleration of fusion pore expansion due to the high curvature of the vesicle , as will be discussed below . Taken together , comparable deficits in exocytosis are observed for the TMD mutants in neurons as well as neuroendocrine cells indicating similar structural requirements for v-SNARE TMDs to initiate fusion .
The membrane-bridging interactions of SNARE proteins bring vesicle and plasma membrane into close apposition and mediate membrane fusion , but the mechanistic events leading to the formation and control of the exocytotic fusion pore have remained unknown . Here , we studied whether the syb2 TMD serves mechanistic functions beyond the passive membrane-anchoring of the force-generating SNARE complex . By systematically changing the structural flexibility of the syb2 TMD ( Figure 4 ) , we observed secretion phenotypes that highlight the functional impact of the TMD at various stages of membrane fusion . Our experiments provide first evidence that the syb2 TMD plays an active role in Ca2+-triggered exocytosis , acting as a crucial catalyst for membrane merger and fusion pore . These observations raise the important question how TMDs actually contribute to the fusion mechanism . While the TMD variants tested here leave the stimulus-secretion coupling unchanged ( Figure 1 ) , mutant syb2 variants designed to dissipate the force transfer between SNARE motif and TMD have been reported to clearly prolong the exocytotic delay ( Kesavan et al . , 2007 ) , providing independent evidence for a distinctive and autonomous function of the TMD in membrane fusion . Neither the overall expression level nor colocalization analyses with the intrinsic marker protein cellubrevin delivered evidence for inefficient sorting of the different mutant proteins to chromaffin granules , thus attributing potential fusion deficits to changes in TMD-mediated function ( Figure 1—figure supplement 3 ) . An exciting interpretation of our data is that TMD mutations may change protein-lipid interactions during Ca2+-dependent fusion by altering the conformational dynamics of the helical backbone . Indeed , previous NMR-studies have shown that increasing the content of ß-branched amino acids of TMD mimic peptides profoundly enhanced lipid mobility and wobbling of lipid head groups ( Agrawal et al . , 2010 , 2007 ) . Hydrophobic nucleation events , in which lipid tails from opposite membranes initially interconnect the adjacent leaflets , have been identified as a highly energy-demanding step en route to fusion ( Kasson et al . , 2010; Risselada et al . , 2011; Smirnova et al . , 2010 ) . The reduced fusogenicity of vesicles in chromaffin cells expressing either rigid polyL or lipid anchored syb2 variants indicates that the interplay between flexible SNARE TMDs and surrounding lipids could promote hydrophobic tail protrusion and thereby fusion initiation ( Figure 6 ) . Indeed , our results are supported by previous in vitro work , showing that isolated SNARE TMDs ( Langosch et al . , 2001 ) or syb2-juxtamemembrane region-TMD constructs ( Tarafdar et al . , 2015 ) facilitate liposome-liposome fusion . A similar dependence on protein-lipid interactions for membrane fusion has previously been observed with viral fusogens ( Kasson et al . , 2010; Tamm et al . , 2003 ) , suggesting that Ca2+-triggered exocytosis and viral fusion engage common mechanisms to drive membrane fusion . Given that the formation of an initial lipid stalk is generally observed in direct vicinity of SNARE TMDs in MD simulations ( Risselada et al . , 2011 ) , TMD-rigidifying mutations ( e . g . polyL mutation ) may lower the probability of lipid splay and thereby produce more unsuccessful fusion attempts with vesicles arrested in a trapped state prior to membrane merger . As fusion mutants are usually expected to slow down stimulus secretion coupling , a scenario where vesicles are led into a trapped state would explain why helix-rigidifying mutations do not alter the kinetics of the exocytotic burst component ( Figure 1 ) . Regardless of the exact underlying mechanism , our results support a model wherein conformational flexibility of a proteinaceous v-SNARE TMD is required to surmount the energy barrier for initial membrane merger . Fast and efficient discharge of bulky cargo molecules from large secretory vesicles is bound to expansion of the exocytotic fusion pore . Since bilayer bending mechanics allow pores of smaller vesicles to expand more rapidly ( Alvarez de Toledo et al . , 1993; Chizmadzhev et al . , 1995; Zhang and Jackson , 2010 ) , increasing the content of ß-branched amino acids in the v-SNARE TMD can ease the expansion of a lipidic pore for larger vesicles as they fuse . Our results show that systematically changing the number of helix-destabilizing , ß-branched valine or isoleucine residues in the syb2 TMD leads to correlated changes in fusion pore behavior . In particular , increasing the number of ß-branched residues within the N-terminal half of the TMD causes an unprecedented gain-of-function phenotype , wherein fusion pore dilation is even accelerated beyond the rate found for the wildtype protein , emphasizing the key role of structural dynamics of the syb2 TMD in membrane fusion . Both , substitution of the syb2 TMD with a lipid anchor or with rigidifying leucine residues strongly slowed down kinetics of transmitter discharge , demonstrating the inherent propensity of the syb2 TMD to promote fusion pore expansion . An attractive explanation for this phenomenon could be that structural flexibility within the N-terminal half of the syb2 TMD counters the highly negative curvature of the membrane’s outer leaflet to drive expansion of the narrow fusion pore neck ( Figure 6E ) . Similarly , Ca2+-bound synaptotagmin-1 ( syt1 ) induces positive curvature to the cytoplasmic leaflets of the fusing membranes ( Hui et al . , 2009; Martens et al . , 2007 ) and thereby may destabilize the early fusion pore ( Dhara et al . , 2014 ) . In this context , it stands to reason that SNARE force-mediated membrane straining ( Kozlov et al . , 2010 ) and TMD-mediated lipid perturbation together with syt1’s ability to bend membranes are synergistic mechanisms that provide mutual reinforcement to form a nascent lipid bridge between membranes and to drive subsequent pore expansion . As an alternative hypothesis , membrane-spanning v- and t-SNARE TMDs have been proposed to form channel structures that are aligned in a stacked manner to generate a gap junction-like pore through the vesicular membrane and plasma membrane ( Bao et al . , 2015; Chang et al . , 2015; Han and Jackson , 2005 ) . However , this concept of a proteinaceous fusion pore is difficult to reconcile with our observation that an acylated syb2-CSP fusion protein lacking the TMD can still significantly raise secretion over dko levels ( Figure 7A , B ) . Furthermore , TMD variants furnishing hydrophobic , identical residues can rescue ( polyLV , Figure 5 and Figure 5—figure supplement 2 ) or even speed up ( polyV and polyI ) transmitter discharge , albeit these helices neither exhibit any polarity nor asymmetry with respect to the side-chain volume of residues that could generate different surfaces of the putative proteinaceous pore . Homotypic TMD-TMD interactions have been implicated in fusion between vacuoles ( Hofmann et al . , 2006 ) and may be involved in a supramolecular assembly of SNARE proteins that precedes the hemifusion state along the fusion pathway ( Lu et al . , 2008 ) . However , considering the phenotypes within our set of different TMD mutants ( G100L , polyV , polyI , polyLV , polyL-Ct , polyL-Nt ) , we found that neither key residues for syb2 TMD dimerization ( G100 , Figure 1—figure supplement 2 ) ( Fdez et al . , 2010 ) , nor those that comprise the interacting helical face of the TMDs ( L99 , C103 , I106 , I110 , [Laage and Langosch , 1997; Roy et al . , 2004; Tong et al . , 2009] ) , play a significant role for membrane fusion or fusion pore expansion . In this context , it is noteworthy that neither deletion nor substitutions of membrane-proximal tryptophane ( Trp ) residues within the juxtamembrane domain ( JMD ) of syb2 ( Borisovska et al . , 2012 ) were found to alter the tonic secretion response or fusion pore properties as observed here with the TMD mutants ( Figure 5 ) . Thus , it is unlikely that TMD mutants interfere with functions of the Trp moiety which influences the electrostatic surface potential by controlling the JMD position at the membrane-water interface ( Borisovska et al . , 2012 ) . Moreover , the fully zippered cis-SNARE complex ( all SNAREs in one membrane ) also establishes several stabilizing interactions between the TMDs of syb2 ( I98 , L99 , I102 , I106 ) and syntaxin-1 ( syx1 ) ( Stein et al . , 2009 ) that might be compromised by mutating the TMD core residues . Yet , several lines of evidence render the possibility unlikely that such a scenario is responsible for functional deficits observed with the TMD mutants . First , the complete substitution of syx interacting residues in syb2 TMD with valine ( or isoleucine ) had no effect on the total secretion and even accelerated fusion pore expansion compared with the wildtype protein . Secondly , the polyV ( Figure 2 ) and the polyLV ( Figure 5 ) mutants exhibited different kinetics of fusion pore expansion , even though the crucial amino acids I98 , I102 and I106 were substituted by valine residues in both mutant variants . These results counter the view that perturbations of ‘lock and key’ like protein-protein interactions between syb and syx TMD are responsible for the functional effects of the TMD mutants . Third , neither short insertions of amino acids ( e . g . 2 residue KL insertion ) nor insertion of 2 helix breaking proline residues , immediately upstream of the syb2 TMD ( Kesavan et al . , 2007 ) , which should interfere with N- to C-terminal zipping of SNAREs into the bilayer spanning helical bundle , were found to affect overall secretion or fusion pore properties . Even a 5 amino acid insertion had no functional consequences on fusion pore dynamics ( Kesavan et al . , 2007 ) . These results together with the strong fusion deficits observed for polyL and polyL-Nt mutants suggest thatconformational flexibility of the syb2 TMD ( within the cytoplasmic leaflet of the membrane ) rather than defined protein-protein interactions upon progressive zipping of syb2/syx TMDs facilitates secretion and fusion pore expansion . Thus , heterodimerization between v- and t-SNARE TMDs likely succeeds but does not promote fusion pore opening and expansion . Notably , single point mutations ( G100L , V101A , V112A , Figure 1—figure supplement 2 ) , may similarly change structural flexibility of the TMD . Yet , given the observed proportionality between the number of ß-branched amino acids and fusion pore parameters ( Figure 6 ) , they are not expected to detectably affect fusion pore dynamics . The increasing energy barrier for larger vesicles to overcome bilayer bending within their nascent fusion pores is documented in amperometric recordings , showing that larger vesicles form more stable initial fusion pores ( i . e . longer prespike duration , [Alvarez de Toledo et al . , 1993; Chizmadzhev et al . , 1995; Zhang and Jackson , 2010] ) . The observed systematic dependency of fusion pore dynamics on the number of ß-branched amino acids in the syb2 TMD raises the question whether structural flexibility of TMDs indeed varies among other v-SNARE isoforms and thus could facilitate cargo release in the context of diverse physiological processes . Interestingly , v-SNARE isoforms responsible for exocytosis of differentially-sized secretory vesicles show a considerable degree of variability regarding the content of ß-branched amino acids within the N-terminal half of their TMDs ( Table 1 ) . v-SNARE proteins , like VAMP7 and VAMP8 , contain more than 70% ß-branched amino acids within this TMD region and thereby are well-suited for exocytosis of large zymogen granules and mast cell vesicles facilitating rapid pore expansion and release of their bulky cargo molecules such as interferon ( Krzewski et al . , 2011 ) and hexoaminidase ( Lippert et al . , 2007; Wang et al . , 2004 ) . Others , like cellubrevin ( VAMP3 ) or syb2 ( VAMP2 ) , with an intermediate content of ß-branched amino acids ( 33% and 44% , respectively ) , are responsible for exocytosis of smaller-sized vesicles such as chromaffin granules ( Borisovska et al . , 2005 ) , cytotoxic T-cell lytic granules ( Matti et al . , 2013 ) or small-synaptic vesicles ( SSV ) ( Schoch et al . , 2001 ) , whereas syb1 with only 22% ß-branched amino acids preferentially mediates SSV exocytosis to release classical neurotransmitters at the NMJ ( Li et al . , 1996; Liu et al . , 2011 ) . Thus , the number of helix-destabilizing ß-branched amino acids within the N-terminal half of different v-SNARE TMDs appears to be evolutionary adapted to the size of vesicles to catalyze fusion pore expansion and facilitate bona fide cargo release . Such a mechanism could also tip the balance between an expanding or non-expanding fusion pore , on the one hand ensuring efficient discharge of bulky cargo molecules from large vesicles and on the other hand favoring release of small cargo as well as rapid recycling of SSVs by reducing the likelihood of complete merger with the plasma membrane . 10 . 7554/eLife . 17571 . 017Table 1 . TMD sequence alignment of exocytotic v-SNARE variants . Amino acid residues comprising the putative TMD regions of the indicated v-SNARE variants are colored red . Note the different number and percentage of ß-branched amino acids ( valine or isoleucine , bold ) in the N-terminal half of the TMD as quantified on the right . Vesicle diameters are taken from the following references for small synaptic vesicles ( Takamori et al . , 2006 ) , chromaffin granules ( Borisovska et al . , 2005 ) , cytotoxic T-cell lytic granules ( Ming et al . , 2015 ) , insulin granules ( Fava et al . , 2012 ) , mast cell granules ( Alvarez de Toledo et al . , 1993 ) , zymogen granules ( Nadelhaft , 1973 ) and sequences were obtained from UniProt database . DOI: http://dx . doi . org/10 . 7554/eLife . 17571 . 017vesicle sizevesicle type ( diameter ) v-SNARE isoform ( M . musculus ) Transmembrane domainN term . C term . no . / % of V or I in the N-terminussmallsmall synaptic vesicles ( 40 nm ) Synaptobrevin 1 93KNCK MMIMLGAIC AIIVVVIVI YFFT118 2 / 22inter-mediatesmall synaptic vesicles ( 40 nm ) chromaffin ( 120 nm ) , lytic ( 250 nm ) and insulin ( 240 nm ) granulesCellubrevin 78KNCK MWAIGISVL VIIVIIIIV WCVS103 3 / 33Synaptobrevin 2 91KNLK MMIILGVIC AIILIIIIV YFST116 4 / 44largemast cell and zymogen granules ( 500–800 nm ) VAMP7 185KNIKLTIIIIIVSIV FIYIIVSLLCGGFTW215 8 / 73VAMP8 72KNVK MIVIICVIV LIIVILIIL FATG97 7 / 77 Overall , our results unmask an active role of the proteinaceous TMD in membrane fusion that clearly goes beyond simple membrane anchoring and may be used to optimize release from differentially sized vesicles . ß-branched amino acids are key determinants for the fusogenic role of the v-SNARE TMD , most likely promoting the conformational dynamics of the TMD helix , which may perturb the packing of the surrounding phospholipids and thereby facilitate first intermembrane contact as well as fusion pore expansion . Taken together , SNARE proteins do not only act as force generators by continuous molecular straining on membranes , but also catalyze membrane merger via structural flexibility of their TMDs .
Experiments were performed on embryonic mouse chromaffin cells prepared at E17 . 5–E18 . 5 from double-v-SNARE knock-out mice ( dko cells; Synaptobrevin-/-/Cellubrevin-/- , [Borisovska et al . , 2005] ) or syb2 knock-out mice ( syb2 ko; Synaptobrevin-/- [Schoch et al . , 2001] ) . Preparation of adrenal chromaffin cells was performed as described before ( Borisovska et al . , 2005 ) . Recordings were done at room temperature on 1–3 days in culture ( DIC ) and 4 . 5–5 . 5 hr after infection of cells with virus particles . Autaptic cultures of hippocampal neurons were prepared at E18 from syb2 knock-out mice , as described previously ( Bekkers and Stevens , 1991; Guzman et al . , 2010; Schoch et al . , 2001 ) . Recordings were performed at room temperature on days 11–15 of culture . For expression in chromaffin cells , cDNAs encoding for syb2 and its TMD mutants were subcloned into the viral plasmid pSFV1 ( Invitrogen , San Diego , CA ) , upstream of an internal ribosomal entry site ( IRES ) controlled open reading frame that encodes for enhanced green fluorescent protein ( EGFP ) . EGFP expression ( excitation wavelength 477 nm ) was used as a reporter to identify infected cells . Mutant constructs were generated by PCR using the overlap expansion method ( Higuchi et al . , 1988 ) . All mutations were confirmed by DNA sequence analysis ( MWG Biotech , Germany ) . Virus cDNA was linearized with restriction enzyme SpeI and transcribed in vitro by using SP6 RNA polymerase ( Ambion , USA ) . BHK21 cells were transfected by electroporation ( 400V , 975 µF ) with a combination of 10 µg syb2 ( wildtype/ mutant ) and pSFV-helper2 RNA . After 15 hr incubation ( 31°C , 5% CO2 ) , virions released into the supernatant were collected by low speed centrifugation ( 200 g , 5 min ) , snap-frozen and stored at -80°C ( Ashery et al . , 1999 ) . For transfection of neurons , cDNAs encoding for syb2 and its mutants were subcloned into pRRL . sin . cPPT . CMV . WPRE lentiviral transfer vector ( Follenzi et al . , 2000 ) , which contains a cPPT sequence of the pol gene and the posttranscriptional regulatory element of woodchuck hepatitis virus ( Follenzi et al . , 2002 ) . To identify transfected cells , syb2 proteins were expressed as fusion constructs with the monomeric red fluorescent protein ( mRFP ) linked to the C-terminal domain of syb2 via a 9aa linker ( GGSGGSGGT ) . Mutant constructs were cloned analogous to the methods described above were verified by DNA sequence analysis . Lentiviral particles were produced as previously described ( Guzman et al . , 2010 ) . Briefly , a 85% confluent 75 cm2 flask of 293FTcells ( Invitrogen ) was transfected with 10 µg of the transfer vector , and 5 µg of each helper plasmid ( pMDLg/pRRE , Addgene #12251; pRSV-Rev , Addgene #12253; pMD2 . G , Addgene #12259 ) using a standard CaCl2-PO4 transfection protocol . Medium was exchanged 8 hr after transfection , viral particles were harvested after 48–72 hr , concentrated using a centrifugal device ( 100 kDa Molecular weight cutoff; Amicon Ultra-15; Millipore ) and immediately frozen and stored at -80°C . Primary neurons were transfected with 300 µl of viral suspension ( 1DIC ) . Whole-cell membrane capacitance measurements and photolysis of caged Ca2+ as well as ratiometric measurements of [Ca2+]i were performed as described previously ( Borisovska et al . , 2005 ) . The extracellular Ringer's solution contained ( in mM ) : 130 NaCl , 4 KCl , 2 CaCl2 , 1 MgCl2 , 30 glucose , 10 HEPES-NaOH , pH 7 . 3 , 320 mOsm . Ratiometric [Ca]i measurements were performed using a combination of fura2 and furaptra ( Invitrogen ) excited at 340 nm and 380 nm . The composition of the intracellular solution for flash experiments was ( in mM ) : 110 Cs-glutamate , 8 NaCl , 3 . 5 CaCl2 , 5 NP-EGTA , 0 . 2 fura-2 , 0 . 3 furaptra , 2 MgATP , 0 . 3 Na2GTP , 40 HEPES-CsOH , pH 7 . 3 , 310 mOsm . The flash-evoked capacitance response was approximated with the function: f ( x ) = A0 + A1 ( 1−exp[−t/t1] ) + A2 ( 1−exp[−t/ t2] ) + kt , where A0 represents the cell capacitance before the flash . The parameters A1 , t1 , and A2 , t2 , represent the amplitudes and time constants of the rapidly releasable pool and the slowly releasable pool , respectively ( Rettig and Neher , 2002 ) . The stimulus-secretion delay was defined as the time between the flash and the intersection point of the back-extrapolated fast exponential with the baseline . Production of carbon fiber electrode ( 5 μm diameter , Amoco ) and amperometric recordings with an EPC7 amplifier ( HEKA Elektronik ) were done as described before ( Bruns , 2004 ) . For Ca2+ infusion experiments , the pipette solution contained ( in mM ) : 110 Cs-glutamate , 8 NaCl , 20 DPTA , 5 CaCl2 , 2 MgATP , 0 . 3 Na2GTP , 40 HEPES-CsOH , pH 7 . 3 , 310 mOsm ( 19 μM free calcium ) . Amperometric current signals were filtered at 2 kHz and digitized gap-free at 25 kHz . Amperometric events with a charge ranging from 10 to 5000 fC and peak amplitude >4 pA were selected for frequency analysis , while an amplitude criterion of >7 pA was set for the analysis of single spike characteristics . For fluctuation and rms noise analyses prespike signals with durations longer than 2 ms were considered and the current derivative was additionally filtered at 1 . 2 kHz . Fluctuations exceeding the threshold of ± 6 pA/ms ( ~4 times the average baseline noise ) were counted . The number of suprathreshold current fluctuations divided by the corresponding prespike signal duration determines the fluctuation frequency . Whole-cell voltage-clamp recordings of synaptic currents were obtained from isolated autaptic or mass cultures of hippocampal neurons . All experiments include measurements from >3 different culture preparations and were performed on age-matched neurons derived from mice of the same litter . Intracellular solution contained ( in mM ) : 137 . 5 K-gluconate , 11 NaCl , 2 MgATP , 0 . 2 Na2GTP , 1 . 1 EGTA , 11 HEPES , 11 D-glucose , pH 7 . 3 . Extracellular solution contained ( in mM ) 130 NaCl , 10 NaHCO3 , 2 . 4 KCl , 2 CaCl2 , 2 MgCl2 , 10 HEPES , 10 D-glucose , pH 7 . 3 , 295 mOsm . To minimize the potential contribution of GABAergic currents the reversal potential of chloride-mediated currents was adjusted to the holding potential . Neurons were voltage-clamped at −70 mV ( without correction for the liquid junction potential , V LJ 9 . 8 mV ) with an EPC10 amplifier ( HEKA Electronic ) under control of Pulse 8 . 5 program ( HEKA Electronic ) and stimulated by membrane depolarizations to +10 mV for 0 . 7 ms every 5 s ( 0 . 2 Hz ) . Cells with an average access resistance of 6–12 MΩ and with 70–80% resistance compensation were analyzed . Current signals were low-pass filtered at 2 . 9 kHz ( four pole Bessel filter EPC10 ) and digitized at a rate of 10 or 50 kHz . The readily releasable pool ( RRP ) was determined by a 5 s application of hypertonic sucrose solution ( 500 mM sucrose ) using a gravity-fed fast-flow system ( Bruns , 1998 ) . To accurately calculate the RRP size , the integral of current flow caused by a hypertonic solution was corrected by subtracting the amount of steady-state refilling and exocytosis that occurred during hypertonic challenges ( Stevens and Wesseling , 1999 ) . For recordings of spontaneous mEPSCs , mass cultures of hippocampal neurons were bathed in Ringer’s solution containing 1 µM tetrodotoxin ( TTX ) . To determine the mEPSC properties with reasonable fidelity events with a peak amplitude >15 pA ( ~5 times the S . D . of the background noise ) and a charge criterion >25 fC were analyzed using a commercial software ( Mini Analysis , Synaptosoft , Version 6 . 0 . 3 ) . SNAP-25 ( amino acids 1–206 ) and Syntaxin 1a ( amino acids 1–262 ) were expressed with an N-terminal 6-histidine tag ( His6 ) in the E . coli strain BL21DE3 and purified using nickel-nitrilotriacetic acid-agarose ( Qiagen , Hilden , Germany ) . Recombinant variants of syb2 ( amino acids 1–116 ) and syb2-polyL were expressed as N-terminal tagged GST fusion proteins ( pGEX-KG-vector ) in the E . coli strain BL21DE3 and purified using glutathione-agarose according to the manufacturer’s instructions . All column elutes were analyzed for integrity and purity of the expressed proteins by SDS-PAGE and staining with Coomassie blue . SNARE complexes were formed by mixing equal molar amounts ( ~5 µM ) of the proteins and incubating at 25°C for the indicated times ( Figure 1—figure supplement 1 ) . The binding buffer contained ( in mM ) : 100 NaCl , 1 DTT , 1 EDTA , 0 . 5% Triton X-100 , 20 Tris ( pH 7 . 4 ) . Assembly reactions were stopped by adding 5xSDS sample buffer . The ability of SNARE proteins to form SDS-resistant complexes was analyzed by SDS-PAGE ( without boiling the samples ) and Coomassie blue staining of protein bands . Chromaffin cells were processed 3 . 5 hr after virus infection for immunolabeling as described previously ( Borisovska et al . , 2012 ) . An affinity purified mouse monoclonal antibody against syb2 ( clone 69 . 1 , antigen epitope amino acid position 1–14 , kindly provided by R . Jahn , MPI for Biophysical Chemistry , Göttingen , Germany ) and a rabbit polyclonal antibody against ceb ( TG-21 , synaptic system ) were used for the immunocytochemical analysis . For epifluorescence microscopy , a Zeiss AxioVert 200 microscope was used , digital images ( 8 bit encoded ) were acquired with a CCD camera and AxioVert Software ( Zeiss , Germany ) and analyzed with ImageJ software version 1 . 45 . The total intensity of the fluorescent immunolabel was determined from ( area of interest comprising the outer cell perimeter – area of interest comprising the cell nucleus ) . To determine the localization and sorting of the mutant syb2 variants in large dense core vesicles , high resolution structured illumination microscopy ( SIM ) was employed . Cells were imaged through a 63x Plan-Apochromat ( NA , 1 . 4 ) oil-immersion objective on the stage of a Zeiss Axio Observer with excitation light of 488 and 561 nm wavelengths . The ELYRA PS . 1 system and ZEN software 2011 ( Zeiss ) were used for acquisition and processing of the images for SIM . Properties of syb2-fluorescent puncta in z-stacks were analyzed with the software package ImageJ , version 1 . 45 . After threshold subtraction , Mander’s weighted colocalization coefficients were determined from the sum of syb2 pixels intensities that colocalizes with ceb , divided by the overall sum of syb2 pixels intensities ( Bolte and Cordelieres , 2006 ) . Therefore MSyb2 = ΣSyb pixel intensity ( coloc . ceb pixel ) / ΣSyb pixel intensity ( Manders et al . , 1993 ) . For immunostaining of the hippocampal neurons , cells were processed on 13 DIC as described for the chromaffin cells . Neurons were imaged with confocal microscope ( LSM 510; Carl Zeiss ) using the AxioVision 2008 software ( Carl Zeiss ) and a 100x , 1 . 3 NA oil objective at room temperature . Images were analyzed with the software package ImageJ ( version 1 . 45 ) and SigmaPlot 8 . 0 ( Systat Software , Inc . ) . Immunopositive spots were determined using a threshold-based detection routine , with the threshold adjusted to the background signal of the neuronal process . Immunosignals were quantified as mean fluorescent intensity per puncta . For the analysis of synaptic density , synaptophysin-positive puncta were counted along 50 μm length of a neuronal process . The atomistic structure of the C-terminal region of syb2 ( residues 71–116 ) was obtained from the X-ray crystallographic structure ( Stein et al . , 2009 ) and the missing C-terminal residue of syb2 ( residue 116 ) was added using Modeller ( Sali and Blundell , 1993 ) . The insertion of the transmembrane domain of syb2 in an asymmetric bilayer was carried out using a self-assembly procedure described elsewhere ( Sharma et al . , 2015 ) . Briefly , the atomistic structure was converted into a coarse-grained ( CG ) representation using a PerlScript file adapting the Martini coarse-graining method ( Monticelli et al . , 2008 ) . The CG protein was positioned at the center of a box with dimensions of 10 × 10 × 11 nm along with overlapping boxes of randomly placed cytoplasmic ( CP ) lipids and intravesicular ( IV ) lipids . The composition of CP lipids was 22 Palmitoyl-Oleoyl-PS ( POPS ) , 76 Palmitoyl-Oleoyl-PE ( POPE ) and 92 cholesterol ( CHOL ) molecules and IV lipids was 22 Palmitoyl-sphingomyelin ( PPCS ) , 66 Palmitoyl-Oleoyl-PC ( POPC ) and 52 CHOL . The resulting lipid box was filled with CG water and an appropriate number of Na+ ions were added to preserve electro-neutrality . This was followed by 1000 steps of energy minimization using steepest descent algorithm after which ten production runs were carried out , each for 200 ns , using a time step of 2 fs . The effective time sampled in the production runs was therefore 2 μs . The CG simulations were analyzed for tilt angle of syb2 transmembrane domain and location of WW domain with respect to the phosphate groups of the membrane . The martini force field employs secondary structure constraints that do not allow changes in conformational states . Based on the CG analyses a representative structure was chosen and converted to an atomistic ( AT ) representation using a reverse transformation protocol ( Wassenaar et al . , 2014 ) . Starting from the reverse-transformed atomistic wildtype structure of the syb2 C-terminal domain in the membrane , residues 97–112 were mutated to either Leu or to Val residues using Modeller to generate the respective mutants . The generated atomistic representations of syb2 , polyL and polyV mutants were initially equilibrated for 2 ns with position restraints on the backbone heavy atoms using a harmonic force constant of 1000 kJ mol–1 nm–2 . After this short equilibration , three 40 ns long production simulations were performed for the wildtype starting from different random velocities . For the mutants , two independent 40 ns long simulations were carried out . All analyses were done on the last 30 ns simulation , unless mentioned otherwise . The lipid models used in the CG simulations , POPS , POPE , POPC , PPCS and the cholesterol model were CG Martini models and simulated using Martini force field ver . 2 . 0 and standard martini simulation parameters with a time step of 20 fs ( Monticelli et al . , 2008 ) . The AT system was described using Slipids ( Stockholm lipids ) ( Jambeck and Lyubartsev , 2012 ) for lipids , AMBER99SB-ILDN ( ff99SB-ILDN ) ( Lindorff-Larsen et al . , 2010 ) for protein and the waters were described using TIP3P ( Jorgensen et al . , 1983 ) . A time step of 2 fs was used for AT simulations . All bonds were constrained using the LINCS algorithm . The bonds in water were constrained using the analytical SETTLE method ( Miyamoto and Kollman , 1992 ) . The pressure was kept constant at one atmosphere by a Parrinello-Rahman barostat ( Parrinello and Rahman , 1981 ) with a coupling constant of 10 . 0 ps and an isothermal compressibility of 4 . 5 × 10–5 bar–1 . A semi-isotropic coupling scheme was employed where the pressure in the xy plane ( bilayer plane ) is coupled separately from the z direction ( bilayer normal ) . The Nosé-Hoover thermostat ( Hoover , 1985; Nose , 1984 ) was used to maintain a constant temperature ( 323 K ) with a coupling constant of 0 . 5 ps . Electrostatic interactions were calculated at every step with the particle-mesh Ewald method ( Essmann et al . , 1995 ) with real-space cutoff of 1 . 0 nm . The van der Waals interactions were cut off at 1 . 4 nm . All simulations were carried out with the GROMACS package version 4 . 6 , ( Pronk et al . , 2013 ) . Analyses were performed by using utilities within the GROMACS package . The secondary structure analyses were carried out using the dictionary of secondary structure of proteins ( DSSP ) method ( Kabsch and Sander , 1983 ) . Values are given as mean ± SEM ( standard error of mean ) unless noted otherwise in the figure legends . To determine statistically significant differences , one-way analysis of variance and a Tukey–Kramer post hoc test were used , if not stated otherwise . | Neurons signal to other cells by releasing chemicals known as neurotransmitters . The neurotransmitters are stored in the neuron in small membrane-bound compartments called vesicles . When a neuron receives an electrical impulse , this ultimately triggers the vesicles to fuse with the cell membrane and release their contents into the gap between the neurons . This process is known as exocytosis . Other cells called neuroendocrine cells , which can receive signals from neurons , also undergo exocytosis to release chemicals into the bloodstream . A group of membrane-bound proteins called SNAREs help a vesicle to fuse with the cell membrane . SNARE proteins are embedded in both the vesicle and cell membrane , and force them into close proximity . When the two membranes make contact , a small channel called the fusion pore forms and expands to release the vesicle’s contents out of the cell . Synaptobrevin-2 is a SNARE protein found in the vesicle membrane . The part of the protein that sits in the membrane is called the transmembrane domain; however , it is not clear whether this domain plays any role in membrane fusion . The transmembrane domain of synaptobrevin-2 is rich in certain amino acids that are thought to make it flexible , thereby allowing it to bend and tilt in the membrane . Dhara , Yarzagaray et al . altered these amino acids in such a way that made this domain either more or less flexible than in the normal protein . The results show that in both neurons and a type of neuroendocrine cell called chromaffin cells , exocytosis is significantly reduced and the fusion pores open more slowly when the transmembrane domain is less flexible . By contrast , exocytosis occurs normally when the transmembrane domain is more flexible; however , the fusion pore expands more rapidly than normal . These results suggest that the flexibility of the transmembrane domain of synaptobrevin-2 promotes membrane fusion and sets the pace at which the fusion pore expands . It is likely that the transmembrane domain disturbs the surrounding membrane in a way that enables these events to happen . Further work is needed to address whether this is the case . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | v-SNARE transmembrane domains function as catalysts for vesicle fusion |
Orphans are genes restricted to a single phylogenetic lineage and emerge at high rates . While this predicts an accumulation of genes , the gene number has remained remarkably constant through evolution . This paradox has not yet been resolved . Because orphan genes have been mainly analyzed over long evolutionary time scales , orphan loss has remained unexplored . Here we study the patterns of orphan turnover among close relatives in the Drosophila obscura group . We show that orphans are not only emerging at a high rate , but that they are also rapidly lost . Interestingly , recently emerged orphans are more likely to be lost than older ones . Furthermore , highly expressed orphans with a strong male-bias are more likely to be retained . Since both lost and retained orphans show similar evolutionary signatures of functional conservation , we propose that orphan loss is not driven by high rates of sequence evolution , but reflects lineage-specific functional requirements .
Orphans are genes with limited phylogenetic distribution and represent a considerable fraction ( up to 30% ) of the gene catalog in all sequenced genomes ( Khalturin et al . , 2009 ) . Studies conducted in different eukaryotes showed that orphans emerge at high rates ( Domazet-Loso et al . , 2007; Wissler et al . , 2013 ) . While gene duplication and exaptation from transposable elements often result in orphan genes ( Toll-Riera et al . , 2009 ) , they also originate frequently de novo from non-coding DNA ( Cai et al . , 2008; Heinen et al . , 2009; Knowles and McLysaght , 2009; Wu et al . , 2011; Yang and Huang , 2011; Xie et al . , 2012; Neme and Tautz , 2013; Wu and Zhang , 2013 ) , probably through intermediate proto-genes ( Carvunis et al . , 2012 ) . Compared to evolutionary conserved genes , orphans are overall shorter ( Lipman et al . , 2002 ) , fast evolving ( Domazet-Loso and Tautz , 2003 ) , have lower and more tissue-restricted expression ( Lemos et al . , 2005 ) . Moreover , they often show testis-biased expression ( Levine et al . , 2006; Begun et al . , 2007 ) , probably due to frequent origination in testis ( Kaessmann , 2010 ) . In Drosophila the rate of orphan emergence is particulary high ( Domazet-Loso et al . , 2007 ) and many orphans become quickly essential ( Chen et al . , 2010 ) . Although the function of only a few orphan genes has been studied , it has been proposed that orphans might serve an important role in speciation and adaptation to different environments ( Khalturin et al . , 2008; Khalturin et al . , 2009; Colbourne et al . , 2011 ) . The high rate of orphan origination would predict an increase in gene content over time . However , gene content in eukaryotes is remarkably stable compared to genome size , as highlighted by Tautz and Domazet-Loso ( 2011 ) . To solve this paradox Tautz and Domazet-Loso proposed that orphans have only a short lifetime ( ‘rapid-turnover’ hypothesis ) ( Tautz and Domazet-Loso , 2011 ) . Thus , although orphans are continuously created , most of them might be lost in a relatively short evolutionary time . Relaxed selective costraints in orphans ( Cai and Petrov , 2010 ) might also contribute to the high rate of orphan loss . Moreover , since orphans are typically identifed by the comparison of distantly related species , their evolutionary stability has been so far neglected . This contrasts the comprehensive analysis of evolutionary patterns of gains and losses of non-orphan genes ( Hahn et al . , 2007 ) . In this study , several partially interrelated factors affect gene loss , including gene expression levels , number of protein–protein interactions , gene dispensability , and rate of sequence evolution ( Krylov et al . , 2003; Cai and Petrov , 2010 ) . This study focuses for the first time on the evolutionary stability of orphan genes . We investigate the factors contributing to orphan loss and find that orphan age , male-biased gene expression , and microsatellite content are correlated with orphan stability . Surprisingly , differences in evolutionary rates cannot explain orphan loss and we propose that orphan loss is driven by lineage-specific evolutionary constraints . Overall , orphan genes are lost at a significantly higher rate than non-orphan genes , supporting the ‘rapid-turnover’ hypothesis .
Orphans are commonly detected by BLASTing the genes of a given organism against a set of outgroup species ( Domazet-Loso and Tautz , 2003; Toll-Riera et al . , 2009 ) . A BLASTP cutoff of 10−3–10−4 was found to be optimal to maximize sensitivity and specificity in Drosophila ( Domazet-Loso and Tautz , 2003 ) . To identify orphans we used a BLASTP cutoff of 10−4 combined with a TBLASTN cutoff of 10−4 , to exclude genes with unannotated orthologs in other species . Following these criteria , we searched in Drosophila pseudoobscura for genes with no sequence conservation in 10 Drosophila species outside the Drosophila obscura group ( Figure 6—figure supplement 1 ) . In total , we identified 1152 orphans , corresponding to 7% of all the D . pseudoobscura genes . Our estimate is slightly lower than a previous one ( Zhang et al . , 2010 ) , due to our different filtering procedure , but still consistent with a high rate of orphan gain in Drosophila ( Domazet-Loso and Tautz , 2003; Domazet-Loso et al . , 2007; Zhou et al . , 2008; Wissler et al . , 2013 ) . Our data clearly indicate that orphan genes are subject to purifying selection , as they show several hallmarks of functional protein-coding sequences ( Figures 1 , 2 ) . A comparison of orphan genes preserved between D . pseudoobscura and D . affinis resulted in a distribution of dN/dS significantly lower than 1 with a median of 0 . 44 ( Figure 1—figure supplement 1 , one-sided Wilcoxon signed-rank test , p<1 . 0 × 10−15 ) , as expected for protein-coding sequences . Moreover , dN/dS for orphans is significantly lower ( Mann–Whitney test , p=2 . 7 × 10−14 ) than dN/dS calculated on a random set of intergenic regions with the same length distribution of orphans ( see ‘Materials and methods’ , section ‘Evolutionary rates’ ) ( Figure 1A ) . Consistent with this , we also found orphans to be more conserved than intergenic regions ( Figure 1B , Figure 1—figure supplement 2 ) . The codon usage bias of orphans is intermediate to that of old genes and intergenic regions ( Figure 1C ) . 10 . 7554/eLife . 01311 . 003Figure 1 . Orphans are subject to purifying selection . ( A ) dN/dS of D . pseudoobscura and D . affinis orthologs . dN/dS is lowest for old genes , but also orphan genes have dN/dS smaller than one . A comparison of orphans and intergenic regions shows that dN/dS for orphans is significantly smaller ( Mann–Whitney test , p=9 . 5 × 10−10 ) , indicating purifying selection on orphan genes . Intergenic regions were of similar length and chromosomal position as the orphan genes . ( B ) Sequence similarity in HSPs obtained from BLASTing D . pseudoobscura genes against the D . affinis genome . Orphans are more conserved than intergenic regions ( Mann–Whitney test , p=0 . 00238 ) and less conserved than old genes ( Mann–Whitney test , p<1 . 0 × 10−15 ) . ( C ) Codon usage was measured by the Codon Adaptation Index ( Sharp and Li , 1987 ) . The codon usage of orphans is significantly higher than that of intergenic regions ( Mann–Whitney test , p<1 . 0 × 10−15 ) indicating that orphans are subject to purifying selection . In comparison to old genes , orphans have a significantly lower codon usage bias ( Mann–Whitney test–p<1 . 0 × 10−15 ) . Overall , all three analyses demonstrate that orphans are not annotation artifacts , but evolutionary conserved genes . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 00310 . 7554/eLife . 01311 . 004Figure 1—figure supplement 1 . Distribution of dN/dS for orphan genes . Most orphans have dN/dS lower than 1 , consistent with the hypothesis of purifying selections acting on these genes . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 00410 . 7554/eLife . 01311 . 005Figure 1—figure supplement 2 . Conservation of orphans in the obscura group . Sequence similarity of old genes , orphans and random intergenic region obtained from BLASTing D . pseudoobscura genes against the genomes of D . lowei ( A ) , D . miranda ( B ) and D . persimilis ( C ) . Orphans are significantly more conserved than random intergenic regions in D . lowei ( Mann–Whitney test , p=0 . 00857 ) , D . miranda ( Mann–Whitney test , p<0 . 00034 ) and D . persimilis ( Mann–Whitney test , p<2 . 8 × 10−13 ) . These results are consistent with purifying selection acting on orphans . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 00510 . 7554/eLife . 01311 . 006Figure 2 . pN/pS for old genes , orphans , and intergenic regions . Orphans show a pN/pS intermediate between old genes and intergenic regions . Nevertheless , pN/pS is significantly smaller for orphans compared to intergenic regions ( Mann–Whitney test , p<1 . 0 × 10−15 ) , indicating coding purifying selection acting on orphans . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 006 To further test for purifying selection acting on orphans , we used a polymorphism dataset of 45 strains being re-sequenced for the third chromosome of D . pseudoobscura ( ‘Materials and methods’ ) . We calculated the ratio of synonymous to non-synonymous polymorphism ( pN/pS ) , since it provides an indication of purifying selection . We found that pN/pS for orphans is significantly lower compared to intergenic regions ( Mann–Whitney test , p=0 . 02182 ) ( Figure 2 ) , and significantly greater for old genes ( Mann–Whitney test , p<1 . 0 × 10−15 ) , consistent with purifying selection operating on orphans . In agreement with studies in other species ( Domazet-Loso and Tautz , 2003; Toll-Riera et al . , 2009; Wolf et al . , 2009; Cai and Petrov , 2010; Capra et al . , 2010; Carvunis et al . , 2012 ) , we also find that orphan genes are shorter ( median length for orphans = 344 bp , median length for old genes = 1470 bp ) , have a lower GC content ( median GC content for orphans = 0 . 54 , median GC content for old genes = 0 . 55 ) , are expressed at lower levels ( expression in D . pseudoobscura males: mean expression for orphans = 29 FPKM , mean expression for old genes = 41 FPKM ) than old genes ( Figure 3 ) . Using CD-Hit ( Li and Godzik , 2006 ) , we found the fraction of genes with a paralog ( >90% protein similarity ) to be similar for orphans ( 6 . 9% ) and old genes ( 6 . 4% ) . Orphans are more enriched in microsatellites , also consistent with previous findings in vertebrates ( Toll-Riera et al . , 2012 ) and rice ( Guo et al . , 2007 ) . Furthermore , unlike mammals ( Toll-Riera et al . , 2009 ) , none of the D . pseudoobscura orphans was found to be associated with transposable elements ( see ‘Transposons detection’ ) . 10 . 7554/eLife . 01311 . 007Figure 3 . Comparison of orphans and genes conserved among 10 Drosophila species outside of the obscura group . Orphans differ from old genes in various features: ( A ) gene length ( B ) GC content , ( C ) dN/dS ( D ) percentage of microsatellites in coding sequence ( E ) Codon Adaptation Index ( F ) gene expression level in D . pseudoobscura males ( G ) gene expression level in D . pseudoobscura females ( H ) sex-biased expression . Orphans are shorter ( Mann–Whitney test , p<1 . 0 × 10−15 ) , have lower GC content ( Mann–Whitney test , p=3 . 9 × 10−7 ) , lower codon usage bias ( Mann–Whitney test , p<1 . 0 × 10−15 ) , lower expression ( Mann–Whitney test , p<1 . 0 × 10−15 ) , higher proportion of microsatellites ( Mann–Whitney test , p=1 . 8 × 10−4 ) and higher dN/dS ( Mann–Whitney test , p<1 . 0 × 10−15 ) compared to old genes . Moreover , orphans are more enriched in male-biased genes compared to old genes ( χ2-test , p<1 . 0 × 10−15 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 007 The distribution of orphans is heterogeneous across chromosomes ( χ2-test , p<1 . 0 × 10−15 ) , with the X chromosome having the highest fraction of orphans . In the obscura group , the two X-chromosome arms have a different evolutionary history . XL corresponds to Muller’s element A and is homologous to the X chromosome in D . melanogaster . XR , however , has been recently derived from an autosome ( Muller’s element D , 3L in D . melanogaster ) . Analyzing the old-X and neo-X chromosomes separately , we observed a striking difference in the number of orphans despite similar chromosome sizes , with the old-X responsible for the excess of X-linked orphan genes , and the neo-X showing a similar number of orphans as the autosomes ( Figure 4 ) . For each chromosomal arm , we computed genomic features in 100 kb windows to correlate them with the difference in orphan content between old-X and neo-X . We found that average GC content , microsatellite density , transposon density , and length of intergenic regions differ between the two chromosomal arms ( Figure 5 ) . 10 . 7554/eLife . 01311 . 008Figure 4 . Chromosomal distribution of old genes and orphan genes . Orphans are overrepresented on the old-X . The number of orphan genes on the neo-X ( XR ) is significantly lower than on the old-X ( XL ) ( χ2-test , p<1 . 0 × 10−15 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 00810 . 7554/eLife . 01311 . 009Figure 5 . Comparison of genomic features among autosomes , old-X and neo-X . ( A ) GC content in 100 kb windows , ( B ) Microsatellite density in 100 kb windows , ( C ) Transposon density in 100-kb windows , ( D ) Length of intergenic regions , ( E ) Recombination rate . GC content is significantly greater on the neo-X compared to old-X for 10 kb windows ( Mann–Whitney test , p=0 . 00020 ) , but not for 100 kb windows ( Mann–Whitney test , p=0 . 1092 ) . Microsatellite density is significantly higher on the neo-X for both windows of 10 kb ( Mann–Whitney test , p=1 . 9 × 10−12 ) and 100 kb ( Mann–Whitney test , p=0 . 00025 ) . Transposon density is significantly lower on the neo-X for both windows of 10 kb ( Mann–Whitney test , p<1 . 0 × 10−15 ) and 100 kb ( Mann–Whitney test , p=4 . 6 × 10−12 ) . Intergenic regions are significantly shorter on the neo-X compared to the old-X ( Mann–Whitney test , p=7 . 4 × 10−9 ) . Recombination rate does not differ significantly between old-X and neo-X ( Mann–Whitney test , p=0 . 629 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 009 We hypothesized that this pronounced difference between the two chromosome arms might reflect a different history of X-linkage . If orphan genes emerge at a higher rate on the X-chromosome ( Levine et al . , 2006 ) , the shorter history of X-linkage on the neo-X could explain the paucity of orphans on the neo-X compared to old-X . In this case , the difference in orphan number between old-X and neo-X chromosomes should date back to the time before the origin of the neo-X , with a similar number of orphans originating after the creation of the neo-X . We therefore used the genomic sequences of five members of the D . obscura group ( D . pseudoobscura [Richards et al . , 2005] , D . miranda [Zhou and Bachtrog , 2012] , and the de novo assembled D . persimilis , D . lowei , and D . affinis ) to date the origin of the orphan genes to different ancestral nodes in the phylogenetic tree of these species ( Beckenbach et al . , 1993 ) . We distinguished five groups of genes: old genes ( non orphans ) and four different orphan age classes ( Figure 6 ) . Surprisingly , we observed a consistent paucity of orphans on XR relative to XL across all age classes ( Figure 7 ) . This persistent difference in orphan number between XL and XR in all age classes suggests that X-linkage is not sufficient to explain the enrichment of orphans on XL . We conclude that the former autosome differs from the ancestral X chromosomal arm by a yet unidentified feature that affects the emergence of new orphans . 10 . 7554/eLife . 01311 . 010Figure 6 . Orphan gain and losses in the Drosophila obscura group . Schematic phylogenetic tree of the Drosophila obscura group species according to Beckenbach et al . ( 1993 ) with D . melanogaster as outgroup . Genes conserved between D . pseudoobscura and 10 non-obscura Drosophila species correspond to age class 5 ( old genes ) . For each age class the number of gene gains is shown in black . Orphans lost at a given branch are indicated in red . Note that losses at internal branches cannot be calculated , since all the orphans are present in D . pseudoobscura . Losses in D . affinis cannot be unambiguously assigned due to the absence of an additional obscura outgroup . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 01010 . 7554/eLife . 01311 . 011Figure 6—figure supplement 1 . Schematic tree of the Drosophila species analyzed in this study . The tree includes the 12 Drosophila species from FlyBase ( Clark et al . , 2007 ) plus three additional members of the obscura group ( D . affinis , D . lowei , and D . miranda ) . The obscura group is highlighted in magenta . The species corresponding to the black subtrees were used as outgroups in the orphan detection pipeline ( see ‘Materials and methods’ ) . Divergence times for the 12 Drosophila species are taken from Table 3 in Obbard et al . ( 2012 ) ( estimates based on mutation rate ) ; for D . affinis and D . miranda from Gao et al . ( 2007 ) ; for D . lowei from Beckenbach et al . ( 1993 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 01110 . 7554/eLife . 01311 . 012Figure 7 . Chromosomal distribution of orphans of different age classes . In each age class orphans are underrepresented on the neo-X ( XR ) compared to old-X ( XL ) ( Age class 4: χ2-test , p=6 . 3 × 10−9; age class 3: χ2-test , p=4 . 4 × 10−5; age class 2: χ2-test , p=0 . 00590; age class 1: χ2-test , p=0 . 00876; D . pseudoobscura specific: χ2-test , p=0 . 00030 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 012 The analysis of orphans that have putatively lost their function via the acquisition of a stop codon or a frame shift causing insertion/deletion ( pseudogenized/lost orphans ) reveals another interesting feature of the XL–XR fusion . The oldest orphans in our dataset ( age class 4 ) show a pronounced excess of pseudogenized orphans on XR in D . affinis and D . miranda ( Figure 8A ) . This trend was not observed for orphans that emerged on XR after the XL–XR fusion ( Figure 8B , C ) , nor for old genes ( Figure 8D ) and is not due to an increased rate of orphan gain on XR ( Figure 9 ) . Since the oldest orphans ( age class 4 ) on XR are a mixture of autosomal ( i . e . , before the fusion ) and sex-chromosomal ( i . e . , after the fusion ) orphans , we speculate that the high rate of pseudogenization of orphans on the XR may reflect the new X-linkage of previously autosomal orphans . A previous study ( Meisel et al . , 2009 ) found that the XR chromosome has experienced a burst of gene duplications to autosomes after its creation . It is plausible that after the conversion of the XR from autosome to sex-chromosome , orphans might have been duplicated to autosomes , whereas the XR ancestral copy would have become pseudogenized . To test this hypothesis , we looked for evidence of gene duplications for the orphans lost on the XR at node 4 ( Figure 6 ) . We aligned the sequences of these genes in D . lowei and D . miranda to the respective genomes using BLASTN ( cutoff 10−5 ) . Upon manual inspection of the alignments , we found that only 1 out of 21 genes in D . miranda ( gene ID: GA23486 ) and 1 out of 14 genes in D . lowei ( gene ID: GA23807 ) had a second hit on an autosome covering at least 50% of the length of the query gene . Other genes either produced a single best hit on the XR chromosome or spurious short hits on other chromosomes ( data not shown ) . Thus , we conclude that duplication of orphans cannot explain the excess of pseudogenized orphans on XR . Nevertheless , our analysis clearly indicates that the emergence of the neo-X chromosome influenced the orphan dynamics on XR , affecting rates of both gain and loss , thus we excluded this chromosome arm for our analyses of the rate of orphan turnover . 10 . 7554/eLife . 01311 . 013Figure 8 . Orphans predating the XL-XR fusion are preferentially lost on the neo-X . For three terminal branches ( D . lowei , D . miranda , and D . persimilis ) the fraction of lost genes for each age class is shown . Each autosome and both X-chromosome arms are shown in different color . At node 4 , where the neo-X originated , we observed the highest rate of orphan pseudogenization on the neo-X ( A ) . Notably , this effect is not seen for younger orphans ( B and C ) neither for old genes ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 01310 . 7554/eLife . 01311 . 014Figure 9 . No change in orphan gain on the neo-X chromosome . The percentage of orphan genes on the neo-X chromosome remains constant through time ( indicated by age classes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 014 For each age class , we determined the number of pseudogenized orphans ( Tautz and Domazet-Loso , 2011 ) . In the D . persimilis lineage , orphan pseudogenization can be studied for three different age classes . If orphans of all age classes were functionally equivalent , no difference in the rate of orphan pseudogenization would be expected . We observe , however , that the fraction of orphan pseudogenes decreases with age ( Figure 10 ) . The D . miranda lineage also shows a higher loss of young orphan genes . The relatively small number of observations , however , precludes statistical testing of this trend . Overall , orphan genes are lost significantly more often than old genes ( Fisher’s exact test , p=3 . 3 × 10−8 ) , consistent with the rapid turnover hypothesis . The unequal conservation of orphans of different age classes is also apparent after normalizing by coding sequence length ( Figure 11 ) , to account for the fact that longer coding sequences ( CDS ) have a greater chance of acquiring ORF-disrupting mutations . When looking at the distribution of premature termination codons ( PTC ) along the open reading frame ( ORF ) of all genes , we observed that PTCs are enriched at the beginning and at the end of the ORF ( Figure 12 ) , consistent with previous results in D . melanogaster ( Lee and Reinhardt , 2012 ) and D . pseudoobscura ( Hoehn et al . , 2012 ) . Since ORF-disrupting mutations occuring at the end of the ORF might have little impact on gene function , we redefined pseudogenes by considering only ORF-disrupting mutations localized in the first half of the ORF and confirmed that orphans of age class 3 are lost more often than those of age class 4 ( Figure 13 ) . Age class 2 was intermediate , most likely not reflecting a biological phenomenon , but due to a high sampling variance associated with the small number of observations ( 9 orphans ) . Finally , the pattern is also robust to a more conservative criterion for ortholog assignment ( see ‘Annotation of the obscura species’ , Figure 14 ) . 10 . 7554/eLife . 01311 . 015Figure 10 . Young orphan genes are more likely to be lost . The barplot shows the fraction of orphans that has acquired a frameshift or premature stop codon ( i . e . , lost function ) . For D . lowei , D . miranda , and D . persimilis , the fraction of lost orphans is shown for different age classes . Orphans are more likely to be lost than old genes . Both the D . miranda and D . persimilis lineage show that younger orphans are more likely to lose function than older ones . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 01510 . 7554/eLife . 01311 . 016Figure 11 . Young orphan genes are more likely to be lost: accounting for CDS length . To test if the short CDS of orphans affects the pattern that young orphans are more likely to lose function , we normalized the percentage of losses by the median CDS length of genes at that node . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 01610 . 7554/eLife . 01311 . 017Figure 12 . Distribution of premature stop codons ( PTCs ) along the ORF for all genes containing PTCs . PTCs are enriched at the beginning and at the end of the ORF in each species . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 01710 . 7554/eLife . 01311 . 018Figure 13 . Young orphan genes are more likely to be lost: considering only frameshifts and premature stop codons occurring in the first half of the ORF . We repeated the analysis shown in Figure 10 by considering only frameshifts and premature stop codons occurring in the first half of the ORF to define a conservative set of pseudogenes , since disrupting mutations occurring at the end of the ORF are likely to have little impact on the gene function . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 01810 . 7554/eLife . 01311 . 019Figure 14 . Young orphan genes are more likely to be lost: the conservative set of orthologs . We repeated the analysis shown in Figure 10 by restricting it to orthologs for which at least one flanking gene is identified in the same contig ( see ‘Annotation of the obscura species’ ) . Due to the substantially reduced number of orphans in the older age classes , we combined age class 3 and 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 019 To determine features associated with the differences in disabling mutations among orphans from different age classes , we contrasted orphans lost in D . lowei and/or D . persimilis ( lost orphans ) vs orphans conserved in all the obscura species ( conserved orphans ) . Genes in both classes evolve at the same rate , are of similar length , and have similar codon usage bias ( Figure 15A–E ) . Conserved orphans have a higher GC content , contain fewer microsatellites , are expressed at a higher level and are more male-biased ( Figure 15B , D–F , G , H ) compared to lost orphans . Conserved orphans tend to increase their expression level as they become older ( Figure 16A ) , whereas the opposite pattern is true for lost orphans ( Figure 16B ) . 10 . 7554/eLife . 01311 . 020Figure 15 . Features of conserved orphans vs lost orphans measured in D . pseudoobscura . ( A ) Gene length ( B ) GC content , ( C ) dN/dS ( D ) percentage of microsatellites in coding sequence ( E ) Codon Adaptation Index ( F ) gene expression levels in D . pseudoobscura males ( G ) gene expression levels in D . pseudoobscura females ( H ) sex-biased expression . Gene length ( Mann–Whitney test , p=0 . 7235 ) and evolutionary rates ( Mann–Whitney test , p=0 . 5835 ) are not significantly different between conserved and lost orphans . Lost orphans have higher GC content ( Mann–Whitney test , p=0 . 00325 ) , lower expression in D . pseudoobscura males ( Mann–Whitney test , p=0 . 00012 ) and females ( Mann–Whitney test , p=0 . 00230 ) and a higher microsatellite content ( Mann–Whitney test , p=0 . 00049 ) compared to conserved orphans . Lost orphans are enriched in unbiased genes compared to conserved orphans ( χ2-test , p=0 . 02611 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 02010 . 7554/eLife . 01311 . 021Figure 16 . Conserved and lost orphans differ in their gene expression pattern . Expression intensity and sex bias in D . miranda for orphans conserved in all the obscura species ( conserved orphans ) vs orphans that pseudogenized in D . lowei and/or D . persimilis ( lost orphans ) . Expression is calculated in males for orphans of age classes 3 and 4 . Expression level increases with age for conserved orphans ( A ) , while it decreases for lost orphans ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 021 Orphan genes are frequently expressed in the testis ( Levine et al . , 2006; Begun et al . , 2007 ) and have a male-biased gene expression pattern ( Metta and Schlötterer , 2008 ) . This pattern could be generated by pervasive gene expression in testis , which facilitates the functional recruitment of non-specific expression ( Kaessmann , 2010 ) . Another explanation is that expression in testis does not require a complex architecture of regulatory modules ( Sassone-Corsi , 2002; Kleene , 2005; Kaessmann , 2010 ) , so that fewer substitutions are required to obtain a functional regulatory module for expressing a novel gene in testis compared to other tissues . We scrutinized these explanations by comparing the fraction of male-biased genes among orphan genes from different age classes . Unexpectedly , the fraction of male-biased genes increases with the age of the orphan genes ( Figure 17 ) . This increase of male-biased orphans among the older age classes is the result of a preferential loss of orphans with an unbiased gene expression ( Figure 18 ) . To confirm that male-biased gene expression is associated with orphan retention rather than emergence , we analyzed the sex-bias in D . miranda for orphans with and without an open reading frame . Consistent with the gene expression pattern in D . pseudoobscura , we found that lost orphans have a significantly lower male-bias in D . miranda ( Figure 19 ) . We conclude that the previously reported male-biased gene expression of orphan genes is not the result of a preferential recruitment of male-biased transcripts , nor do orphans gradually acquire male-biased gene expression . Rather , male-biased orphans are more likely to be retained . 10 . 7554/eLife . 01311 . 022Figure 17 . The proportion of male-biased orphans increases with age . Sex-biased expression was measured in D . pseudoobscura for orphans belonging to different age classes and for old genes ( age class 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 02210 . 7554/eLife . 01311 . 023Figure 18 . Conservation of orphans is correlated with male-biased gene expression . Orphans with male-biased gene expression in D . pseudoobscura were grouped into classes according to expression bias strength . The fraction of conserved orphans in each bin shows a significant positive correlation with expression bias ( Spearman’s rho = 0 . 811 , p=0 . 02692 ) . This correlation suggests that orphans with a more pronounced male-biased expression tend to persist longer than less male-biased orphans . No similar trend was seen for female-biased orphans ( Spearman’s rho = 0 . 78262 , p=0 . 1176 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 02310 . 7554/eLife . 01311 . 024Figure 19 . Comparison of strength of sex-biased gene expression for conserved and lost orphans in D . miranda . A sex-biased gene expression larger than zero indicates a higher gene expression intensity in males than in females ( male-biased gene expression ) . Conserved orphans have significantly higher male-biased expression than lost orphans ( Mann–Whitney test , p=0 . 03158 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 024
Our study provides the missing link to understand orphan dynamics . Until now , orphan evolution was primarily studied on long phylogenetic branches . Although this approach is well suited to discover new orphans , it does not allow tracing the evolution of orphans . Previous studies showed a high rate of orphan gain , which is not reflected in an increase in gene number . To resolve this apparent paradox , it has been postulated that orphans must be lost at a high rate as well ( Tautz and Domazet-Loso , 2011 ) . In this study , we used the framework of closely related species in the obscura group to study the patterns of orphan gain and losses . We show that orphans not only emerge at high rates , but that they are also rapidly lost ( Figure 10 ) . Interestingly , most losses ( ∼76% ) were due to disabling mutations rather than deletions of the orphan gene . Although under equilibrium conditions the number of losses balances the number of orphan gains , here , we observed a surplus of orphan gains ( Figure 6 ) . We caution that this discrepancy probably does not imply an increase of gene number , but rather reflects the limited evolutionary time to acquire mutations . Using a rather conservative criterion for disabling mutations , either premature stop codons or frameshift indels , we have probably not identified all orphans that have lost their function . Furthermore , we do not account for the possibility of loss of function due to changes in gene regulation . Importantly , codon usage bias , dN/dS values and sequence conservation clearly suggest that orphan genes are functionally constrained and these constraints do not differ among orphans that are conserved in the obscura group and those that lost function in at least one species of the group . Hence , it may be possible that orphan loss is stochastic and reflects weak purifying selection . Nevertheless , lost orphans differ in some aspects from conserved ones . Orphans that are lost contain more microsatellite stretches and have a lower , less sex-biased gene expression than retained ones . Furthermore , we also found that the rate of orphan loss decreases with orphan age , a result consistent with orphans serving a functional role only temporarily . Previous work suggested that orphans are important for adaptation to novel environments ( Khalturin et al . , 2009; Colbourne et al . , 2011 ) , but it is also possible that orphans contribute to stabilize new connections in gene networks ( Capra et al . , 2010; Warnefors and Eyre-Walker , 2011 ) and become obsolete once such new connections have been optimized . Our data suggest that orphans become quickly functional , which is reflected in their codon usage bias , dN/dS ratio and sequence conservation . The chromosomal translocation resulting in the neo-X chromosome provides another interesting perspective on the evolution of orphan genes . Despite the fact that the neo-X is now fully dosage compensated ( Abraham and Lucchesi , 1974 ) , and has obtained a similar base composition as the XL ( Gallach et al . , 2007 ) , we noted that the translocation resulted in a preferential loss of orphan genes on the neo-X . Since this pattern is restricted to orphans that most likely originated before the chromosomal fusion , we argue that the change in chromosomal environment has affected the function of orphan genes , most likely via expression differences . We speculate that the selective advantage conferred by these orphans has diminished , which resulted in a higher loss rate . Interestingly , the elevated rate of orphan loss after the neo-X formation seems to be still ongoing . This differential loss of orphan genes point in a similar direction as the observation that the gene composition of the neo-X has been altered by gene duplication ( Meisel et al . , 2009 ) . Hence , both ( orphan ) gene loss and duplication contribute to fast gene content remodeling on a newly formed sex chromosome in Drosophila .
An individual species sample of D . affinis ( stock number 140120141 . 02 ) was ordered from the Drosophila Species Stock Center ( https://stockcenter . ucsd . edu/info/welcome . php ) and sequenced on the Illumina GAIIx following the paired-end library preparation protocol ( version Illumina 1 . 7 ) in two runs ( run 1: read length = 101 bp , insert size = 230 bp; run 2: read length = 101 bp , insert size = 550 bp ) . Short genomic reads for D . lowei ( accessions SRX091466 and SRX091467 ) and D . persimilis ( accession SRX091471 ) were downloaded from the Sequence Read Archive ( http://www . ncbi . nlm . nih . gov/sra ) . The genome of D . miranda was downloaded from NCBI ( GenBank Assembly ID GCA_000269505 . 1 ) . The genome of D . pseudoobscura was downloaded from FlyBase ( release 2 . 23 ) . Reads for D . affinis , D . lowei , and D . persimilis were trimmed using the Perl script trim_fastq . pl ( parameters –quality-threshold 20 −−min-length 40 ) from PoPoolation ( Kofler et al . , 2011 ) . For each species , a de novo assembly ( parameters: min-contig-length 200 ) was performed using CLC Genomics Workbench 4 . 6 ( http://www . clcbio . com/products/clc-genomics-workbench/ ) , followed by scaffolding with nucmer ( parameters: –c 30 –g 1000 –b 1000 –l 15 ) against the D . pseudoobscura genome . Average coverage per assembled genome was calculated by realigning the reads against the contigs of the respective species with Bowtie 2 . 1 . 0 ( parameters: --very-fast ) and selecting only reads with mapping quality >20 . The annotation of D . affinis , D . lowei , D . miranda , and D . persimilis is based on orthology to D . pseudoobscura using Exonerate 2 . 2 . 0 ( parameters: -model protein2genome–bestn 1 -showtargetgff ) , by aligning the longest isoform of D . pseudoobscura proteins extracted from a recent re-annotation of D . pseudoobscura ( Palmieri et al . , 2012 ) to the genomes of D . affinis , D . lowei , D . miranda , and D . persimilis . For each gene , the best unambiguous hit was retained . To remove non-informative hits , we also required a minimum fraction of the gene to be recovered . Since the sequence conservation of orthologs decreases with divergence time , the expected length of the ortholog depends strongly on the phylogenetic distance between query and subject sequence . To apply consistent criteria for all species , we empirically determined the expected fraction of a gene with sequence homology . Based on genes that are conserved between D . pseudoobscura and the 10 Drosophila species outside the obscura clade ( old genes ) ( Figure 6—figure supplement 1 ) , we determined the distribution of the fraction of the genes that could be aligned . As cutoff the value we used the 5th percentile of the distribution of aligned protein length of old genes . This resulted in a threshold of 47% for D . affinis , 52% for D . lowei , 59% for D . miranda and 53% for D . persimilis . Hence , only orphan orthologs that showed a fraction of aligned coding sequence higher than the empirically determined cutoffs were retained . In addition to this ortholog set , we generated an alternative , more conservative ortholog set . For this one , at least one of the flanking genes of D . pseudoobscura was required to be in synteny with the respective orthologs in D . affinis , D . miranda and D . persimilis . D . lowei was not considered in the synteny analysis since most of the genes in this species are flanked by genomic gaps , due to the shorter contig length of the D . lowei assembly ( Table 1 ) , which caused many contigs to contain only a single gene ( Table 2 ) , thus precluding proper synteny assignments . Assembly and annotation of all the species are available at http://popoolation . at/affinis_genome , http://popoolation . at/lowei_genome , http://popoolation . at/miranda_genome and http://popoolation . at/persimilis_genome . Detailed annotation statistics for each gene are available at http://dx . doi . org/10 . 5061/dryad . hq564 ( Palmieri et al . , 2014 ) . 10 . 7554/eLife . 01311 . 025Table 1 . De novo assembly statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 025D . affinisD . loweiD . persimilisNumber of contigs28 , 946106 , 46517 , 387N759 , 4781 , 21810 , 359N5025 , 1603 , 23024 , 172N2549 , 0627 , 35749 , 047Minimum length121162147Maximum length216 , 90387 , 164204 , 742Average length5 , 1831 , 3887 , 736Total bp150 , 030 , 247147 , 756 , 871134 , 501 , 523Average coverage51 X92 X44 XThe D . miranda genome was available at NCBI , thus no de novo assembly was made for this species . 10 . 7554/eLife . 01311 . 026Table 2 . Orthology annotation statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 01311 . 026D . affinisD . loweiD . mirandaD . persimilisTotal genes14 , 28714 , 95215 , 28214 , 995Genes with frameshifts/PTC*1 , 2331 , 2661 , 171898Mean number of genes per contig3 . 41 . 6–3 . 4Median number of genes per contig21–2Maximum number of genes per contig3524–37*PTC = Premature termination codons . D . pseudoobscura proteins corresponding to the longest isoform for each gene were aligned using BLASTP ( E < 10−4 ) and TBLASTN ( E < 10−4 ) against the published proteomes and genomes of 10 Drosophila species outside the obscura group ( D . melanogaster , D . simulans , D . sechellia , D . erecta , D . yakuba , D . ananassae , D . willistoni , D . mojavensis , D . virilis , D . grimshawi ) . Genes without BLAST hits and without annotated orthologs in FlyBase ( gene orthologs release 09-2011 ) were classified as orphans . Illumina reads for 45 D . pseudoobscura strains were downloaded from NCBI ( Sequence Read Archive , accession SRP017196 ) . Reads were trimmed using PoPoolation ( Kofler et al . , 2011 ) and a total of 3 . 5 million reads was randomly extracted for each strain and combined into a single FASTQ file . The combined reads were treated as a Pool-Seq dataset and mapped to the FlyBase D . pseudoobscura genome release 2 . 23 with BWA ( Li and Durbin , 2009 ) ( parameters -o 1 -n 0 . 01 -l 200 -e 12 -d 12 ) on a hadoop cluster using DistMap ( Pandey and Schlötterer , 2013 ) . From the resulting BAM file , PCR duplicates were removed with Picard ( http://picard . sourceforge . net ) using the tool MarkDuplicates . jar ( parameters REMOVE_DUPLICATES = true , VALIDATION_STRINGENCY = SILENT ) . Proper-pairs with mapping quality >20 were extracted with samtools ( version 0 . 1 . 18 ) ( Li et al . , 2009 ) . Indels were detected with PoPoolation using the script identify-genomic-indel-regions . pl ( parameters--min-count 2 --indel-window 5 ) and masked from the reference genome prior to SNP calling . Coverage was subsampled to 50X for all the chromosomes . Only SNPs on the 3rd chromosome were considered in all analyses , since a balancer chromosome was used to extract the 3rd chromosome , precluding an unbiased polymorphism analysis for the remaining chromosomes . SNPs were called with the PoPoolation script Variance-sliding . pl ( parameters--min-coverage 10 --min-count 2 --max-coverage 500 --min-qual 20 --window-size 500 --step-size 500 --fastq-type sanger--pool-size 45 ) . Orphan gains and losses ( pseudogenizations ) were inferred by Dollo parsimony . Based on the phylogenetic tree of Figure 6 , a gene was assigned as gained at a given node if an intact ortholog was present in both external branches of the subtree corresponding to that node . For example , a gene having an intact ORF in D . lowei but not in D . affinis was classified as gained at node 3 ( Figure 6 ) . A gene was considered to be lost at a terminal branch if at least one ORF-disrupting mutation ( frameshift/premature stop codon ) was present in the gene at that branch and two intact ORFs were detected at both external leaves ( Wang et al . , 2006 ) . The relatively high coverage of our assemblies ( Table 1 ) makes unlikely that disrupting mutations are sequencing errors . In D . affinis for instance , only 8 genes had an average coverage lower than 20x . A gene was considered as completely deleted in a species if no ortholog was detected in that species and no BLASTP ( E < 10−4 ) or TBLASTN ( E < 10−4 ) hit was found . Deletions were not considered into analyses of gene turnover , since they cannot be distinguished from missing annotations . Four RNA-Seq datasets of D . pseudoobscura males and females ( strains ps94 and ps88 from the ArrayExpress database—accession E-MTAB-1424 ) , together with two RNA-Seq samples of D . miranda males and females from the Sequence Read Archive ( accessions SRX106024 , SRX106025 ) , were used for expression analysis . For each sample , reads were trimmed using PoPoolation ( Kofler et al . , 2011 ) and aligned to the genome of the respective species with GSNAP version 2012-07-12 ( Wu and Nacu , 2010 ) ( parameters: –N 1 ) . Only proper pairs mapping unambiguously to one position were retained . Expression in FPKM was calculated with Cufflinks version 1 . 2 . 1 ( parameters: -F 0 . 10 –j 0 . 15 –I 300000 ) . For D . pseudoobscura sex-bias was calculated using the package DESeq ( Anders and Huber , 2010 ) , treating the strains as two biological replicates for each sex and applying an FDR = 0 . 1 . Differential expression between D . miranda males and females was calculated for both species using the log2 fold change on the normalized expression counts using the normalization protocol implemented in the R package DESeq ( Anders and Huber , 2010 ) version 1 . 10 . 1 . Codon usage bias was calculated using the R package seqinr ( function cai ) based on the D . pseudoobscura codon usage table downloaded from http://www . kazusa . or . jp/codon/cgi-bin/showcodon . cgi ? species=7237 . Coding sequences of D . pseudoobscura and D . miranda orthologs without frameshifts/stop codons were aligned using PRANK ( Loytynoja and Goldman , 2005 ) ( parameters: –codon ) . To test for purifying selection on orphans , dN/dS was compared between orphans and a set of randomly selected intergenic regions . This set was generated as follows: ( 1 ) we identified the intergenic regions from the D . pseudoobscura annotation from Palmieri et al . ( 2012 ) , ( 2 ) for each CDS belonging to an orphan gene we extracted all the intergenic regions longer than that CDS , ( 3 ) we randomly selected one intergenic region and we extracted from that a random subregion with the same length of a given orphan CDS , ( 4 ) this procedure was repeated for all orphan CDS , resulting in a set of intergenic regions with the same length distribution as orphan CDSs . These regions were aligned with BLASTN ( cutoff 10−5 ) to the D . affinis genome and for each region the best hit was kept and realigned with PRANK ( default parameters ) to the D . pseudoobscura query sequence . Each alignment was truncated at the 5’end to get an alignment length , which is a multiple of 3 . Internal stop codons were replaced by Ns . The ratio of the rates of nonsynonymous and synonymous substitutions per gene ( dN/dS ) was measured using Markov models of codon evolution and maximum likelihood methods implemented in PAML ( Yang , 2007 ) . To shed light on the differences in orphan number between XL and XR , different features were compared among old-X , neo-X , and autosomes in D . pseudoobscura ( unassembled contigs were not considered in this analysis ) : ( A ) GC content was calculated with the R package seqinr for 100 kb sliding windows along each chromosome ( B ) microsatellite density was calculated using SciRoKo 3 . 4 ( Kofler et al . , 2007 ) ( parameters: -mode mmfp–l 15 –r 3 –s 15 –p 5 –seedl 8 –seedr 3 –mmao 3 ) for 100 kb sliding windows along each chromosome; ( C ) transposon density was estimated with RepeatMasker 3 . 2 . 9 ( parameters: –q–gff -nolow–norna–species drosophila ) for 100 kb sliding windows along each chromosome; ( D ) length of intergenic regions were calculated using BEDTools ( -complement ) by interval subtraction between genome and gene coordinates; ( E ) recombination rates for different windows were taken from McGaugh et al . ( 2012 ) . Microsatellites were detected on the transcript sequences of the longest isoform for each D . pseudoobscura gene using the tool SciRoKo 3 . 4 ( Kofler et al . , 2007 ) ( parameters: -mode mmfp–l 15 –r 3 –s 15 –p 5 –seedl 8 –seedr 3 –mmao 3 ) . Genomic annotation of transposons was performed in D . pseudoobscura using RepeatMasker 3 . 2 . 9 ( parameters: –q–gff -nolow–norna–species drosophila ) . Only transposons longer than 50 bp and not overlapping with microsatellites ( see ‘Microsatellite detection’ ) were retained . We required for an orphan to contain a full transposon sequence in one of its exons in order to classify it as associated with a transposon . | New genes are added to most genomes on a steady basis . A new gene can either begin as a copy of an existing gene from elsewhere in the genome , or is created entirely ‘from scratch’ from a DNA sequence that had not previously encoded for a protein . New genes that are not found in other related species are called orphan genes—and these genes can account for up to 30% of all the genes in the well-studied genomes . However , for reasons that are not fully understood , the total number of genes in most genomes remains fairly constant despite these regular additions . Now , Palmieri et al . have investigated this paradox by following the evolutionary fate of orphan genes in a small group of related species of fruit fly . Palmieri et al . discovered that most orphan genes are very short-lived , even though they showed clear signals of carrying out important functions . Most orphan genes died out quickly due to mutations that made them unable to be expressed as functional proteins , and a small number were deleted entirely from the genome . Unexpectedly , new orphan genes were more likely to die out than those that had been around for a while . Palmieri et al . also found that the expression levels of orphan genes determined their probability of dying with those genes that were expressed to the highest levels being most likely to persist longer . Furthermore , genes that were expressed more in males than in females were also less likely to die . The next challenge will be to identify the mechanisms that determine which orphan genes survive and which do not . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"evolutionary",
"biology"
] | 2014 | The life cycle of Drosophila orphan genes |
Millions of neurons drive the activity of hundreds of muscles , meaning many different neural population activity patterns could generate the same movement . Studies have suggested that these redundant ( i . e . behaviorally equivalent ) activity patterns may be beneficial for neural computation . However , it is unknown what constraints may limit the selection of different redundant activity patterns . We leveraged a brain-computer interface , allowing us to define precisely which neural activity patterns were redundant . Rhesus monkeys made cursor movements by modulating neural activity in primary motor cortex . We attempted to predict the observed distribution of redundant neural activity . Principles inspired by work on muscular redundancy did not accurately predict these distributions . Surprisingly , the distributions of redundant neural activity and task-relevant activity were coupled , which enabled accurate predictions of the distributions of redundant activity . This suggests limits on the extent to which redundancy may be exploited by the brain for computation .
Neural circuits relay information from one population of neurons to another . This relay involves successive stages of downstream neurons reading out the activity of upstream neurons . In many cases , the same activity in the downstream population can be produced by different population activity patterns in the upstream population , a phenomenon termed neural redundancy . Redundancy is ubiquitous in neural computation , from sensory input to motor output . For example , during a task where subjects need to discriminate the color of a stimulus while ignoring its orientation ( Mante et al . , 2013 ) , population activity patterns corresponding to the same color but different orientations are read out equivalently , and are therefore redundant . There is mounting evidence that redundancy in readouts may provide various computational benefits . For example , neural redundancy may allow us to prepare movements without executing them ( Kaufman et al . , 2014; Elsayed et al . , 2016 ) , enable stable computation despite unstable neural dynamics ( Driscoll et al . , 2017; Druckmann and Chklovskii , 2012; Murray et al . , 2017 ) and allow the central nervous system to filter out unwanted noise ( Moreno-Bote et al . , 2014 ) . To fully utilize the proposed benefits of neural redundancy , the population activity should be allowed to freely vary , as long as the readout of this activity remains consistent with task demands . This would allow the population activity to perform computations that are not reflected in the readout . However , a commonly held assumption is that neural activity might also be constrained by energetics: All things being equal , if two population activity patterns are read out equivalently , the brain should prefer the pattern that requires less energy to produce ( Laughlin et al . , 1998; Barlow , 1969; Levy and Baxter , 1996 ) . These two lines of reasoning raise the following questions: What principles guide the production of redundant neural activity patterns ? Are there constraints on which redundant activity patterns can be produced ? If so , this may limit the extent to which neural circuits can exploit the proposed computational benefits of redundancy . Redundancy has been studied extensively in motor control ( Lashley , 1933; Bernstein , 1967 ) , albeit in terms of muscular redundancy rather than neural redundancy . During arm movements , different combinations of muscle activity can lead to the same arm kinematics , meaning these different muscle activity patterns are redundant . Previous work on this muscle redundancy problem has identified two principles guiding the selection of redundant muscle activity . First , because muscle contraction requires energy in the form of ATP , the selected muscle activity should require minimum energy relative to the other redundant options ( Thoroughman and Shadmehr , 1999; Huang et al . , 2012; Fagg et al . , 2002 ) . Second , a minimal intervention strategy has been proposed in which subjects control only the aspects of muscle activity that influence the task outcome , and allow for variability in the aspects of muscle activity that do not influence the task outcome ( Scholz and Schöner , 1999; Todorov and Jordan , 2002; Valero-Cuevas et al . , 2009 ) . To generate movements , the brain not only needs to deal with muscle redundancy , but also neural redundancy , which has been less studied . One way in which neural redundancy can arise is when there are more elements ( neurons or muscles ) upstream than downstream . During arm movements , the activity of around thirty muscles in the arm and hand is controlled by tens of thousands of neurons in the spinal cord ( Gray , 1918; Feinstein et al . , 1955 ) . Those neurons are in turn influenced by millions of neurons in the primary motor cortex and other motor areas ( Ettema et al . , 1998; Lemon , 2008 ) . Thus , the neural control of arm movement is redundant ( Figure 1A ) , in that different population activity patterns can generate the same movement ( Rokni et al . , 2007; Ajemian et al . , 2013 ) . Can the principles of muscular redundancy inform our understanding of neural redundancy ? A common challenge in studying neural redundancy is that it is typically not known which neural activity patterns are redundant , because we do not know how downstream neurons or muscles read out information . In this study we overcome this problem by leveraging a brain-computer interface ( BCI ) , in which the activity of dozens of neurons is read out as movements of a cursor on a computer screen ( Figure 1B ) ( Taylor et al . , 2002; Carmena et al . , 2003; Hochberg et al . , 2006; Ganguly and Carmena , 2009; Gilja et al . , 2012; Hauschild et al . , 2012; Sadtler et al . , 2014 ) . A key advantage of a BCI is that the readout of the population activity ( termed the BCI mapping ) is fully known and defined by the experimenter ( Golub et al . , 2016 ) . This allows us to determine precisely the redundant population activity patterns , which are those that move the cursor in exactly the same way . To illustrate this , consider a simplified example where the activity of two neurons controls a 1D cursor velocity ( Figure 1C ) . The two dark green activity patterns produce the same cursor movement ( v1 ) , and the two light green patterns produce a different movement ( v2 ) . We can decompose any population activity pattern into two orthogonal components: output-potent activity and output-null activity ( Figure 1C , black axes ) ( Kaufman et al . , 2014; Law et al . , 2014 ) . The output-potent component determines the cursor’s movement , whereas the output-null component has no effect on the cursor . Two population activity patterns are redundant if they have the same output-potent activity , but different output-null activity ( e . g . the dark green square and circle on the 'v1' dotted line in Figure 1C ) . The question we address here is , which redundant population activity patterns are preferred by the nervous system ? To answer this , we assessed the distribution of output-null activity produced during each cursor movement ( Figure 1D ) , and compared it to what we would expect to observe under each of several candidate hypotheses for explaining neural redundancy . We trained three Rhesus macaques to perform a brain-computer interface task in which they controlled the velocity of a cursor on a computer screen by volitionally modulating neural activity in primary motor cortex . To understand the principles guiding the selection of redundant neural activity , we compared the observed distributions of output-null activity to those predicted by three different hypotheses . The first two hypotheses we considered were inspired by studies of muscle redundancy . First , by analogy to minimum energy principles ( Thoroughman and Shadmehr , 1999; Huang et al . , 2012; Fagg et al . , 2002 ) , neural activity may minimize unnecessary spiking ( Barlow , 1969; Levy and Baxter , 1996 ) . Second , by analogy to the minimal intervention strategy ( Scholz and Schöner , 1999; Todorov and Jordan , 2002; Valero-Cuevas et al . , 2009 ) , output-null activity might be uncontrolled ( i . e . output-potent activity is modified independently of output-null activity ) because neural variability in this space has no effect on cursor movement . Third , we considered the possibility that the distribution of redundant activity may be coupled with the task-relevant activity , so that producing particular activity patterns in output-potent dimensions requires changing the distribution of activity in output-null dimensions . We tested all hypotheses in terms of their ability to predict the distribution of output-null activity , given the output-potent activity . Hypotheses were tested within the space in which the population activity naturally resides , termed the intrinsic manifold ( Sadtler et al . , 2014 ) . The results of Sadtler et al . ( 2014 ) indicate that neural activity cannot readily leave this manifold , and more recent results demonstrate that neural activity is further constrained by a neural repertoire within the intrinsic manifold ( Golub et al . , 2018 ) . However , a repertoire defines only a set of population activity patterns , and not how often different activity patterns within the repertoire are produced . Therefore , to understand the principles governing the selection among redundant population activity patterns , we focused on predicting the distribution of redundant population activity within the intrinsic manifold and neural repertoire . We found strong evidence for the third hypothesis , that redundant activity is coupled with task-relevant activity . This indicates that neural redundancy is resolved differently than muscular redundancy . Furthermore , the output-null space should not be thought of as a space in which neural activity can freely vary to carry out computations without regard to the output-potent activity . Instead , the distribution of output-null activity is constrained by the corresponding output-potent activity . If the required output-potent activity is defined by the task demands , this can constrain how the output-null activity can vary , and correspondingly the computations that can be carried out in the output-null space .
Previous work in motor control has found that subjects select muscle activations that minimize energy use , that is , subjects tend not to make movements with more stiffness or muscular co-contraction than necessary to complete the task ( Thoroughman and Shadmehr , 1999; Fagg et al . , 2002; Huang et al . , 2012 ) . We tested whether an analogous principle might hold true at the level of neurons ( Figure 2A , Minimal Firing hypothesis ) . Because spiking incurs a metabolic cost ( Laughlin , 2001; Laughlin et al . , 1998 ) , we first considered the hypothesis that among all the population activity patterns that produce the same cursor movement , the subject will select the one requiring the fewest spikes ( Barlow , 1969; Softky and Kammen , 1991; Levy and Baxter , 1996 ) . To predict the distribution of output-null activity under this hypothesis , at each time step we found the population activity pattern that would produce the observed cursor movement with the fewest spikes across all recorded neurons ( see Materials and methods ) . This means population activity will have minimal variability in output-null dimensions , because spiking in these dimensions does not affect cursor movement . In Figure 2A , the orange square depicts the activity pattern nearest zero spikes/s ( gray square ) among all activity patterns that would produce the same cursor movement ( black dotted line ) . This would produce a delta distribution of output-null activity , where the delta would be located at the predicted value ( orange square ) . To make this prediction more realistic , we incorporated Poisson spiking noise . In addition , for this hypothesis and those following , we ensured that all predictions were physiologically plausible ( i . e . firing rates were between zero and the maximum rates observed in the experiment; see Materials and methods ) . We constructed histograms of the output-null activity predicted by the Minimal Firing hypothesis by pooling over all time steps in which the cursor moved in a similar direction ( e . g . 0° , 45° , etc . ) ( Figure 2B , orange ) . We compared these predicted distributions to the observed distributions of output-null activity measured for that movement direction during the experiment ( Figure 2B , black ) . Figure 2C depicts these histograms for the same session across eight different cursor directions ( rows ) , in three of the eight output-null dimensions ( columns ) . For visualization , we applied principal components analysis ( PCA ) to display the output-null dimensions ordered by the amount of shared variance in the output-null activity . To assess how well the Minimal Firing hypothesis predicted the observed output-null activity , we computed the absolute error between the predicted and observed histograms . These errors were averaged across histograms for all eight cursor directions and eight output-null dimensions in a given session . We normalized the errors so that a perfect match between the observed and predicted histograms would result in an error of 0% , while complete mismatch between the predicted and observed histograms would yield an error of 100% ( see Materials and methods ) . We found that the predictions of the Minimal Firing hypothesis differed from the observed activity by 73 . 2% ±1 . 3% ( mean ± SE ) across sessions . One possible explanation as to why these predictions were so different from the observed activity is that minimal energy principles in the brain may not equate to minimal spiking . Perhaps a more relevant constraint is not how far the activity is away from zero firing , but rather how far the activity is from a different level of activity , such as the mean firing rate for each neuron . This alternative version of a minimal energy hypothesis ( Figure 2D , Minimal Deviation hypothesis ) predicts that among all the population activity patterns that produce the same cursor movement , subjects select the one with the smallest deviation from some baseline population activity pattern . For each session , we identified the population activity pattern that would minimize the output-null prediction error across cursor directions in a cross-validated fashion ( see Materials and methods ) ( Figure 2E ) . This hypothesis yielded an average histogram error of 30 . 9% ±1 . 2% ( mean ± SE ) across sessions . While this represents a substantial improvement over the Minimal Firing hypothesis ( paired t-test of histogram errors in each session , p<0 . 001 ) , the predicted distributions of output-null activity still show clear discrepancies from the observed distributions ( Figure 2F ) . Thus , we sought a hypothesis that could better predict the observed distributions of output-null activity . It has been shown that muscle activity exhibits more variability in output-null dimensions than in output-potent dimensions ( Scholz and Schöner , 1999; Todorov and Jordan , 2002; Valero-Cuevas et al . , 2009 ) . An explanation of this variability asymmetry is the ‘minimal intervention’ principle ( Todorov and Jordan , 2002; Valero-Cuevas et al . , 2009; Diedrichsen et al . , 2010 ) , which states that while variability in output-potent dimensions should be corrected to ensure task success , variability in output-null dimensions can be left uncorrected because it does not lead to deficits in task performance . While this principle has been used to explain muscle activity , here we investigate whether it also explains neural activity . This hypothesis , that output-null activity will be ‘uncontrolled’ and have high variability , is in contrast to the minimal firing hypotheses , which predict that output-null activity will have low variability . The idea that neural activity may be selected according to a minimal intervention principle does not , by itself , specify the form of the distribution in output-null dimensions . We therefore considered two specific forms of uncontrolled hypotheses . First , we supposed that if all values of output-null activity are equally likely , then output-null activity would have a uniform distribution with bounds determined by each neuron’s physiological range ( Figure 3A , Uncontrolled-uniform ) . We emphasize that the minimal intervention principle does not specify a candidate distribution , and so we consider this particular hypothesis as a limiting case , where output-null activity has maximum entropy within bounds on minimum and maximum activity . At each time step , we sampled the output-null activity from a uniform distribution within ranges observed experimentally ( see Materials and methods ) . This procedure predicts that the output-null activity is selected independently of the current output-potent activity , reflecting the minimal intervention principle . However , note that the extent of the uniform distribution depends on the physiological range of each neuron , and so the predicted distributions of output-null activity vary slightly with the cursor direction ( Figure 3B–C ) ( e . g . the length of the green bar in Figure 3A depends on the output-potent activity ) . As before , for visualization we ordered the eight output-null dimensions by the amount of shared variance explained in the recorded activity , and displayed the first three of these output-null dimensions ( Figure 3C ) . Because these three dimensions were rotated along the dimensions of highest variance , the predicted histograms are mound-shaped rather than uniformly distributed ( see Materials and methods ) . The predictions of the Uncontrolled-uniform hypothesis differed from the observed output-null activity by 56 . 6% ±1 . 1% ( mean ± SE ) across sessions . In the second variant of this hypothesis , we considered a non-uniform distribution of output-null activity . If the natural variability of output-null activity is truly unmodified , then the distribution of activity observed in the same dimensions when a subject was controlling a different ( previous ) BCI mapping should have the same distribution under the current mapping ( Figure 3D , Uncontrolled-empirical ) . Thus , under this hypothesis we construct an empirical distribution of output-null activity , which we form by projecting all of the population activity that the subject produced under the previous mapping onto the output-null dimensions of the current BCI mapping ( see Materials and methods ) . At each time step , we sampled from this empirical distribution of output-null activity independently of the output-potent activity , again reflecting the minimal intervention principle ( Figure 3D ) . We checked that combining the output-null and output-potent activity resulted in physiologically plausible population activity ( see Materials and methods ) . If it did not , then we re-sampled a different output-null activity pattern until the combination resulted in physiologically plausible population activity . Due to this resampling , the predicted distributions of output-null activity vary slightly with the cursor direction ( Figure 3E–F ) . The histograms of the predictions differed from the observed data by only 23 . 8% ±0 . 8% ( mean ± SE ) across sessions , which is the lowest error of all hypotheses considered so far . This suggests that previously observed population activity ( in this case , recorded during use of a different BCI mapping ) offers greater predictive power of the selection of output-null activity than a priori predictions such as those of the Minimal Firing , Minimal Deviation , and Uncontrolled-uniform hypotheses . Thus far , the hypothesis that best predicts the observed output-null activity is the one that uses previously observed activity to generate its predictions ( Uncontrolled-empirical ) . This motivated us to consider more refined hypotheses that make use of this previously observed activity to generate predictions . We first considered the hypothesis that in order to produce a desired movement , the subject selects neural activity as if he were still using the previous mapping , and corrects this activity only to ensure task success ( Figure 4A , Persistent Strategy ) . Conceptually , when the subject wants to move the cursor in a particular direction using the current BCI mapping , he starts with the population activity patterns that he used to move the cursor in that direction under an earlier mapping ( Figure 4A , light blue shading ) . Because this activity will not move the cursor in the same way that it did under the previous mapping , this activity is modified along the output-potent dimensions of the current mapping ( Figure 4A , red arrows ) , reflecting the minimal intervention principle ( Todorov and Jordan , 2002; Valero-Cuevas et al . , 2009; Diedrichsen et al . , 2010 ) . This is similar to the Uncontrolled-empirical hypothesis in that we assume activity in output-null dimensions can be corrected independently of the activity in output-potent dimensions . However , instead of sampling from the entire distribution of previously observed output-null activity at each time step , here we only sample from the subset of this activity observed when subjects needed to move the cursor in the same direction as the current time step . The predictions of this hypothesis ( Figure 4B–C ) differed from the observed output-null activity by 17 . 4% ±0 . 7% ( mean ± SE ) across sessions . The principle of minimal intervention posits that output-null activity can change independently from output-potent activity . Here we examine this assumption in detail . Previous work has found that the characteristic ways in which neurons covary ( i . e . the dimensions of the intrinsic manifold ) persist even under different BCI mappings , perhaps owing to underlying network constraints ( Sadtler et al . , 2014 ) . All hypotheses we consider here are evaluated within the intrinsic manifold , and thus respect these constraints on population variability . Because the dimensions of the intrinsic manifold capture the variability among the neurons , it is plausible that the activity along different dimensions of the intrinsic manifold can vary independently , consistent with the minimal intervention principle . By contrast , in the next hypothesis we consider the possibility that activity along different dimensions exhibit dependencies . We considered the hypothesis that the distribution of activity in output-null dimensions would be predictably coupled with the activity in output-potent dimensions , even under a different BCI mapping when those dimensions were not necessarily potent and null . Under this hypothesis ( Figure 4D , Fixed Distribution ) , given the output-potent activity , the distribution of the corresponding output-null activity remains the same as it was under a different BCI mapping ( Figure 4D , blue frequency distribution ) , even if this activity was not output-null under the other mapping . This hypothesis predicts that neural activity patterns are ‘yoked’ across dimensions , such that producing particular activity in output-potent dimensions requires changing the distribution of activity in output-null dimensions . The histograms of output-null activity predicted by the Fixed Distribution hypothesis were a striking visual match to the recorded activity , and accurately predicted the dependence of these distributions on the cursor direction ( Figure 4E–F ) . Overall , these predictions differed from the observed output-null activity by only 13 . 4% ±0 . 5% ( mean ± SE ) across sessions . The Fixed Distribution hypothesis yielded a lower histogram error than all other hypotheses across sessions from three different animals ( Figure 5A ) . In total , the Fixed Distribution hypothesis had the lowest histogram error in 41 of 42 sessions . The histogram error metric does not explicitly capture the degree to which hypotheses predicted the mean output-null activity , or any correlations that exist across output-null dimensions . We therefore assessed how well the predictions captured the mean and covariance of observed data in all output-null dimensions jointly ( see Materials and methods ) . In agreement with our findings for histogram error , the mean ( Figure 5B ) and covariance ( Figure 5C ) of output-null activity was best predicted by the Fixed Distribution hypothesis , with an average mean error of 23 . 5 ± 1 . 4 spikes/s ( mean ± SE ) and an average covariance error of 1 . 4 ± 0 . 1 ( mean ± SE in arbitrary units; see Materials and methods ) . These error metrics offer further evidence that the Fixed Distribution hypothesis provides a good match to the output-null distribution , as measured by the agreement between the first and second moments of the two distributions . Because these error metrics rely on a limited number of trials , they should not be compared relative to zero error . We estimated the smallest histogram , mean , and covariance errors achievable by any hypothesis , given the limited number of samples available to estimate the true output-null distributions ( see Materials and methods , and gray regions in Figure 5 ) . The errors of Fixed Distribution were exceedingly close to the lowest achievable error given the number of samples available ( see Materials and methods ) . Next , we found that the Fixed Distribution hypothesis achieved the lowest prediction errors among all hypotheses when data for each monkey was considered individually ( Figure 5—figure supplement 1 ) . We repeated our analyses to predict output-null activity produced during the first mapping using activity observed during the second mapping ( Figure 5—figure supplement 2 ) . We also predicted output-null activity using the actual BCI mapping rather than the animal’s internal model to define the output-null dimensions ( Figure 5—figure supplement 3 ) . Both analyses yielded results similar to those in Figure 5 . So far we have shown that the Fixed Distribution hypothesis provides a better explanation for the structure of output-null activity than hypotheses incorporating constraints on firing rates or the minimal intervention principle . We next sought stronger evidence for the Fixed Distribution hypothesis by assessing our predictions in the particular dimensions of population activity where it is least likely to hold . Because cursor velocity is a two-dimensional quantity , all but two dimensions of population activity for each BCI mapping are output-null . Thus , given two different BCI mappings , most dimensions will be output-null under both mappings , and so most components of the population activity have no reason to change from one mapping to the other . Therefore , we assessed whether our results held in dimensions of population activity that were output-potent during the first mapping , but output-null during the second mapping ( see Materials and methods ) . These are the dimensions in which one would expect to see the most changes in the population activity between the first and second mappings . Our hypotheses make distinct predictions about how the variance of activity should change if a dimension is output-potent under the first mapping and becomes output-null under the second mapping . For example , according to the Minimal Firing and Minimal Deviation hypotheses , the variance of activity will collapse in dimensions that are output-null because unnecessary spiking is undesirable . Thus , if a dimension becomes output-null , variance in this space should exhibit a marked decrease . On the other hand , the Uncontrolled hypotheses predict that , when conditioned on the cursor movement , variance will expand when the activity is output-null . This occurs because variability in this dimension will no longer affect cursor movement , and would therefore no longer need to be suppressed . Finally , the Fixed Distribution hypothesis posits that the same distributions of output-null activity will be observed regardless of whether a dimension was previously output-potent or output-null , and so this hypothesis predicts that there will be little to no change in the variance of activity in a particular dimension under the two mappings . We asked whether the variance of population activity decreased , increased , or remained the same in dimensions that changed from being output-potent to output-null ( Figure 6A ) . Critically , we computed the variance of activity after first binning by the corresponding angle in the output-potent dimensions of the second mapping . This was done so that the neural activity in each bin would all result in similar cursor movements under the second mapping , and is identical to the procedure used previously to assess the errors of the hypotheses’ predictions . Notably , binning in this way means that each bin may contain activity corresponding to different cursor movements under the first mapping , and so one might expect that in each bin the activity recorded under the first mapping would be more heterogeneous than the activity recorded under the second mapping . We observed that the variance of population activity recorded under the first and second mappings was remarkably similar in the dimensions that changed from output-potent to output-null , even though these activity patterns usually corresponded to different cursor movements under the two mappings ( Figure 6B ) . Thus , the variance of activity did not change much when an output-potent dimension became output-null , in agreement with the predictions of the Fixed Distribution hypothesis . To quantify these observations , we computed the average change in variance in each session ( see Materials and methods ) . Across sessions , we found that the variance of observed activity showed a small but significant decrease when it became output-null ( Figure 6C , ‘Data’ ) ( t-test , p<0 . 001 ) . This is in contrast to the predictions of the Minimal Firing and Minimal Deviation hypotheses , which predicted much larger decreases . The observed change in variance lies closest to the predictions of the Fixed Distribution hypothesis . In fact , we observed that the Fixed Distribution hypothesis also predicted a slight decrease in variance in dimensions that became output-null ( Figure 6C , ‘Fixed Distribution’ ) ( t-test , p<0 . 001 ) . This slight predicted change in variance occurs because the distributions of activity in the output-potent dimensions of the second mapping are different under the first and second mappings . Because the Fixed Distribution hypothesis predicts a fixed conditional distribution of output-null activity given the output-potent activity , slightly different sets of output-potent activity will result in a slightly different distribution of the corresponding output-null activity . These analyses show that , contrary to the predictions of the minimal firing and uncontrolled hypotheses , the variance of population activity did not change dramatically in dimensions that were output-potent under the first mapping and output-null under the second mapping . We also assessed whether the reverse was true—if the variance of activity changed in dimensions that began as output-null and became output-potent . To measure this , we repeated the above analyses after predicting output-null activity produced during the first mapping using the activity observed under the second mapping ( as in Figure 5—figure supplement 2 ) . We found that the activity showed little to no change in variance in these dimensions ( t-test , p>0 . 5 ) , in agreement with the predictions of Fixed Distribution ( Figure 6—figure supplement 1 ) . Importantly , the agreement between the observed output-null activity and the predictions of the Fixed Distribution hypothesis in these analyses indicates that our ability to accurately predict the distribution of output-null activity is not merely a result of most activity being output-null under both mappings . Instead , the distribution of output-null activity remains consistent with the Fixed Distribution hypothesis even in the output-null dimensions that were previously output-potent . In Figure 6C , the observed output-null activity showed a larger decrease in variance than the predictions of the Fixed Distribution hypothesis , at least in the 2D subspace of output-null activity that was output-potent during the first mapping . This slight decrease in variance is in the direction of the predictions of Minimal Firing and Minimal Deviation . If this decrease in variance is to be explained by Minimal Firing or Minimal Deviation principles , we would expect that the observed mean output-null activity would also move in the direction of the predictions of Minimal Firing and Minimal Deviation , relative to what is predicted by Fixed Distribution . To see if this was the case , we first computed the distance of the observed mean output-null activity from the mean predicted by Minimal Deviation for each movement direction , and compared this to the distance of the mean output-null activity predicted by Fixed Distribution from the mean predictions of Minimal Deviation ( Figure 6—figure supplement 2A ) . We did not find evidence that the observed mean output-null activity was closer to the mean predicted by Minimal Deviation than was the mean predicted by Fixed Distribution ( one-sided Wilcoxon signed rank test , p>0 . 5; see Figure 6—figure supplement 2B and Materials and methods ) . Repeating the analysis with Minimal Firing instead of Minimal Deviation yielded similar results ( one-sided Wilcoxon signed rank test , p>0 . 5 ) . Thus , while we observed a slight decrease in the variance of output-null activity in dimensions that changed from output-potent to output-null , we did not find any evidence that the mean output-null activity moved in the direction of the predictions of Minimal Firing or Minimal Deviation .
Recent work has suggested that neural redundancy may be exploited for various computations ( Druckmann and Chklovskii , 2012; Kaufman et al . , 2014; Moreno-Bote et al . , 2014; Elsayed et al . , 2016; Driscoll et al . , 2017; Murray et al . , 2017 ) . However , if the activity in output-null dimensions is constrained by the output-potent activity , then this may limit the ability of output-null activity to perform computations without affecting the readout . Here , we studied neural redundancy in the primary motor cortex using a BCI , where it is known exactly which population activity patterns are redundant , meaning they produce an identical cursor movement . We generated predictions of the distributions of output-null neural activity for subjects performing a BCI cursor control task , and compared them to the distributions observed in our experiments . We found that hypotheses inspired by minimal firing and minimal intervention principles , drawn from theories of muscle coordination , did not accurately predict the observed output-null activity . Instead , we found that the distribution of output-null activity was well predicted by the activity in the two output-potent dimensions . This coupling between the output-potent and output-null activity implies that , when output-potent activity is used to satisfy task demands , there are constraints on the extent to which neural circuits can use redundant activity to perform additional computations . Our results indicate that the way in which neural redundancy is resolved is different from how muscle redundancy is resolved . There have been several prevalent proposals for how muscle redundancy is resolved , including minimal energy , optimal feedback control ( OFC ) , and habitual control . Models incorporating minimal energy principles have helped to explain observed gait ( McNeill Alexander and McNeill , 2002 ) and arm reaches ( Thoroughman and Shadmehr , 1999; Huang et al . , 2012; Fagg et al . , 2002; Farshchiansadegh et al . , 2016 ) . By analogy , it has been proposed that the brain may prefer an ‘economy of impulses’ ( Barlow , 1969; Softky and Kammen , 1991; Levy and Baxter , 1996 ) , resolving neural redundancy by minimizing the production of action potentials . However , we found that minimal energy principles in terms of firing rates do not play a dominant role in the selection of output-null neural activity . Given that metabolic activity can decrease without corresponding changes in firing rates ( Picard et al . , 2013 ) , the brain may implement minimal energy principles without influencing the way neural redundancy is resolved . OFC posits that motor control signals are selected to minimize a cost function that depends on task requirements and other factors , such as effort or delayed reward . OFC models have been widely used to explain muscle activity during motor tasks ( Todorov , 2004; Scott , 2004; Diedrichsen et al . , 2010 ) . Our results for neural activity differ in two important respects from OFC predictions with standard cost functions involving task requirements and effort . First , those implementations of OFC predict that variability in task-irrelevant dimensions should be higher than variability in task-relevant dimensions , a concept often referred to as the ‘uncontrolled manifold’ ( Scholz and Schöner , 1999 ) . We found that the variability of neural activity did not increase in dimensions that went from being task-relevant to task-irrelevant ( Figure 6C ) . Second , those implementations of OFC predict a ‘minimal intervention’ strategy , whereby activity in task-relevant dimensions is corrected independently of activity in task-irrelevant dimensions ( Todorov and Jordan , 2002; Valero-Cuevas et al . , 2009; Diedrichsen et al . , 2010 ) . Three of the hypotheses we tested incorporate this minimal intervention principle: Uncontrolled-uniform , Uncontrolled-empirical , and Persistent Strategy . None of these hypotheses predicted neural activity in task-irrelevant dimensions as accurately as did the Fixed Distribution hypothesis , which predicts that the distributions of task-relevant and task-irrelevant activity are yoked . Overall , our work does not rule out the possibility that OFC is appropriate for predicting neural activity . First , it may be possible to design a cost function such that OFC predictions are consistent with the findings presented here . Second , one could consider applying OFC with the control signal being the input to M1 ( e . g . PMd activity ) , rather than the control signal being M1 activity ( as we have done here ) or muscle activity ( where OFC has been traditionally applied ) . This could induce coupling between the output-potent and output-null dimensions of the M1 activity , and thereby yield predictions that are consistent with the findings presented here . It has also been proposed that muscle recruitment is habitual rather than optimal , such that muscle recruitment under altered dynamics is a rescaled version of that under normal control ( de Rugy et al . , 2012 ) . The results for habitual control are similar to what we found for neural activity , in that ( 1 ) we could predict activity from previously observed activity , and ( 2 ) we observed a tight coupling of the distributions of task-relevant and task-irrelevant activity ( in contrast to minimal intervention ) . However , the results for habitual control are different from our findings in that we found that subjects appear to use the same distribution of activity in each of two different BCI mappings , whereas different ( overlapping ) subsets of muscle activation patterns were used under different conditions in de Rugy et al . ( 2012 ) . Given how many dimensions of population activity there are ( in this case , 10 ) , it is somewhat surprising that conditioning on only the two output-potent dimensions could provide so much explanatory power for predicting the distribution in the remaining neural dimensions . This suggests that many of the dimensions of population activity are coupled , that is , changing the activity along some dimensions may also lead to changes along other dimensions , even though those dimensions are mutually orthogonal . During arm movement control , output dimensionality and presumably the neural dimensionality are larger than in our BCI setup . We speculate that during arm movements , many of the null dimensions will remain coupled with the potent dimensions , thereby yielding results similar to what we found here . Future work could examine whether animals can be trained to uncouple dimensions , as well as the effects of larger output-potent dimensionality on redundancy , by repeating our analyses with a higher-dimensional effector , such as a multiple degree-of-freedom robotic limb ( e . g . Wodlinger et al . , 2015 ) . The results presented here are related to , and go beyond , those in Golub et al . ( 2018 ) . Although the two studies analyzed data from the same experiments , they ask distinct questions . Golub et al . ( 2018 ) focused on explaining the changes in population activity underlying behavioral learning . By contrast , in the present work we seek to determine the constraints on activity in the task-irrelevant ( i . e . output-null ) dimensions . In other words , while Golub et al . ( 2018 ) focused on explaining the changes leading to behavioral learning , we focus here on the principles other than behavior that constrain population activity . As a result , all hypotheses we consider in the present work make predictions consistent with the observed behavior in the output-potent dimensions . Golub et al . ( 2018 ) found that the amount of learning animals showed was consistent with a fixed neural repertoire of population activity patterns being reassociated to control the second BCI mapping . The repertoire of population activity refers to the set of population activity patterns that were observed , whereas here we focused on the distribution , which describes how often the animals produced different activity patterns . In other words , the finding of a fixed repertoire is a statement about the support of the distribution of population activity , whereas here we found that the distribution of population activity can be predicted in output-null dimensions , given the output-potent activity . Because many different distributions of neural activity can be constructed from a fixed repertoire , the present results represent a stronger constraint on population activity than that shown in Golub et al . ( 2018 ) . Indeed , the majority of the hypotheses we tested were consistent with a fixed neural repertoire , and thus cannot be disambiguated based on our prior work . This is evidenced by the predicted distributions largely overlapping with the support of the actual data distributions ( Figures 2–4 ) . The two hypotheses that were not fully consistent with a fixed repertoire are the Minimal Firing and Uncontrolled-uniform hypotheses . However , in the context of predicting the distribution of activity in redundant dimensions , these hypotheses represent interesting cases worth considering ( i . e . where population activity either obeys minimal firing constraints , or that the output-null activity is fully unstructured , respectively ) , and so we included these hypotheses to cover these possibilities . It is interesting to consider the relationship between arm movements and BCI cursor movements ( Orsborn et al . , 2014; Vyas et al . , 2018 ) . If the dimensions responsible for moving the arm overlap with both the output-potent and output-null dimensions of the BCI , this might explain the coupling we observe between the output-potent and output-null dimensions . However , in these experiments , the animal’s arm was not moving during BCI control ( see Extended Data Figure 5 in Sadtler et al . , 2014 ) . Thus , the activity we study here resides within the arm’s output-null dimensions . This implies that in our recordings the arm’s output-potent dimensions do not overlap with either the output-potent or the output-null dimensions of the BCI , and so arm movements ( or the lack thereof ) are unlikely to explain the coupling we observed between the output-potent and output-null dimensions of the BCI . Overall , being unaware of extra output-potent dimensions would likely make the predictions of the Fixed Distribution hypothesis worse , not better . The reason for this is as follows . The Fixed Distribution hypothesis predicts that the distribution of activity in output-null dimensions depends upon the corresponding output-potent activity . Under this hypothesis , the more we know of the output-potent activity , the better we can predict the output-null distribution . If there is an output-potent dimension that we have not accounted for in our analyses , accounting for this dimension would likely improve our predictions . The fact that we were able to accurately predict the output-null distributions ( 13% histogram error on average , with the lowest possible error being 7% ) without knowing all the potent dimensions is then evidence that these extra potent dimensions , if they exist , would not provide substantial additional predictive power . In this work , we define a set of population activity patterns as redundant if they all result in the same readout in downstream areas . This definition of redundancy comes from early work on motor control ( Bernstein , 1967; Sporns and Edelman , 1993 ) , where it was noted that different motor signals can result in the same movement kinematics . This is related to but distinct from the information-theoretic definition of redundancy ( Schneidman et al . , 2003; Latham et al . , 2005; Averbeck et al . , 2006 ) . In the information-theoretic case , redundancy describes the extent to which correlations among neurons limit decoding accuracy for different stimuli . This is distinct from the type of redundancy studied here , defined as the existence of multiple population activity patterns corresponding to the same readout . For example , by the information-theoretic definition , a system may have no redundancy ( e . g . the population activity allows one to perfectly decode the encoded variable ) , but there may still be multiple population activity patterns that refer to this same encoded variable . We found that the distribution of output-null activity could be well predicted using activity recorded under a different BCI mapping . Two factors of our experimental design are particularly relevant when interpreting this result . First , we used a balanced center-out task design in which subjects made roughly equal numbers of movements in each direction . If we had , for example , required far more leftward than rightward movements , this would have altered the distribution of joint activity and skewed the estimates of output-null activity during the second mapping . Second , this study focused on short timescales , where we predicted output-null activity within one to two hours of subjects learning a new BCI mapping . On this timescale , the motor system must be able to rapidly learn a variety of different mappings between neural activity and behavior , and thus , a variety of different sets of redundant activity . An interesting avenue for further research would be to determine if the constraints we observe on neural redundancy remain over longer timescales . Given repeated practice with the same BCI mapping across days and weeks ( Ganguly and Carmena , 2009 ) , it is possible that there are different and perhaps fewer constraints on neural redundancy than what we found here . We have tested six specific hypotheses about how neural redundancy is resolved . These hypotheses cover a spectrum of how strongly the activity in output-null dimensions is constrained , with the minimal firing hypotheses being the most constrained , the minimal intervention hypotheses being the least constrained , and the Fixed Distribution hypothesis lying in between . Although the hypotheses we tested are not exhaustive , the best hypothesis ( Fixed Distribution ) yielded predictions of the distributions of output-null activity whose marginal histograms differed from the data by only 13% on average ( Figure 4F ) , where we estimated the lowest error possible to be 7% on average . Further improvements to the prediction accuracy may be possible by incorporating additional constraints , such as dynamics ( Shenoy et al . , 2013 ) . It should be stressed that our focus here was on predicting the distribution of output-null activity . Future work can assess whether output-null activity can be predicted on a time-step-by-time-step basis . The central premise of the null space concept is that some aspects of neural activity are read out by downstream areas ( output-potent ) while other aspects are not ( output-null ) ( Kaufman et al . , 2014 ) . This idea is related to the study of noise correlations , where it was recognized that activity fluctuations that lie outside of a stimulus encoding space ( i . e . ‘stimulus-null’ ) are not detrimental to the stimulus information encoded by the neurons ( Averbeck et al . , 2006; Moreno-Bote et al . , 2014 ) . Studies have also shown that structuring neural activity in an appropriate null space can allow for multiplexing of different types of information ( Mante et al . , 2013; Raposo et al . , 2014 ) , as well as stable behavior ( Leonardo , 2005; Rokni et al . , 2007; Ajemian et al . , 2013 ) and stable working memory ( Druckmann and Chklovskii , 2012; Murray et al . , 2017 ) in the presence of time-varying neural activity . Additionally , the existence of output-null dimensions in the motor system may facilitate motor learning ( Moorman et al . , 2017; Ranganathan et al . , 2013; Singh et al . , 2016 ) or allow for motor preparation ( Kaufman et al . , 2014; Elsayed et al . , 2016 ) or novel feedback processing ( Stavisky et al . , 2017 ) without causing overt movement . Our work suggests that there may be limits on the extent to which output-null activity might be leveraged for neural computation . The coupling we observe between the distributions of output-null and output-potent activity suggests that output-null activity is not modified independently of output-potent activity . This coupling may cause activity fluctuations in a stimulus-null space to influence the downstream readout , or limit one’s ability to plan the next movement without influencing the current movement . Moving forward , an important direction for understanding the computations performed by different brain areas is to find out which aspects of the neural activity are read out ( Pagan et al . , 2013; Kaufman et al . , 2014 ) and to understand how the dependencies like those identified in this study impact the computations being performed .
Experimental methods are described in detail in both Sadtler et al . ( 2014 ) and Golub et al . ( 2018 ) . Briefly , we recorded from the proximal arm region of primary motor cortex ( M1 ) in three male Rhesus macaques using implanted 96-channel microelectrode arrays ( Blackrock Microsystems ) . All animal care and handling procedures conformed to the NIH Guidelines for the Care And Use of Laboratory Animals and were approved by the University of Pittsburgh’s Institutional Animal Care and Use Committee . The population spiking activity in each non-overlapping 45 ms bin was computed as the number of threshold crossings on each channel . In each session , 85–94 neural units were recorded ( 25 sessions from monkey J , six sessions from monkey L , 11 sessions from monkey N ) . These sessions were analyzed previously in Golub et al . ( 2018 ) . Data from monkeys J and L were first presented in Sadtler et al . ( 2014 ) . The average firing rate of the neural units per session was 50 ± 8 , 42 ± 4 , and 55 ± 14 spikes/s ( mean ± s . d . ) for monkeys J , L , and N , respectively . Each session began with a block of calibration trials . The calibration procedure for monkey J involved either passive observation of cursor movement , or closed-loop BCI cursor control using the previous day’s BCI mapping . For monkeys L and N , we used a closed-loop calibration procedure that gradually stepped from passive observation to closed-loop control , as described in Sadtler et al . ( 2014 ) . We then applied factor analysis ( FA ) to the spike counts recorded during these calibration trials to identify the 10D linear subspace ( i . e . the ‘intrinsic manifold’ ) that captured dominant patterns of co-modulation across neural units ( Churchland et al . , 2010; Harvey et al . , 2012; Sadtler et al . , 2014; Athalye et al . , 2017 ) . We then estimated the factor activity , zt∈R10×1 , as the posterior expectation given the observed spike counts , ut∈Rq×1 , where q is the number of neural units: ( 1 ) zt=L⊤ ( LL⊤+Ψ ) −1 ( ut−d ) Here , L , Ψ , and d are FA parameters estimated using the expectation-maximization algorithm , where Ψ is constrained to be a diagonal matrix . The factor activity , zt , can be interpreted as a weighted combination of the activity of different neural units . We refer to zt as a ‘population activity pattern . ’ We next orthonormalized zt so that it had units of spike counts per time bin ( Yu et al . , 2009 ) , using the following approach . In our FA model , L defines a mapping from low-dimensional factor space to the higher-dimensional neural space . Because the columns of L are not orthonormal , the factor activity does not have the same units ( spikes counts per time bin ) as the neural activity . However , we can fix this by finding an orthonormal basis for the columns of L ( Yu et al . , 2009 ) . To do this , we apply the singular value decomposition , yielding L=USV⊤ , where U∈ℝq×10 and V∈R10×10 have orthonormal columns and S∈R10×10 is diagonal . Then , we can write Lzt=U ( SV⊤zt ) =Uz~t . Because U has orthonormal columns , z~t=SV⊤zt has the same units ( spike counts per time bin ) as ut . For notational simplicity , we refer to z~t as zt throughout . The values in zt appear larger than those expected for a single neuron because this value tends to grow with the total number of neural units . Over the course of each experiment , animals used two different BCI mappings ( see ‘Behavioral task’ below ) . Each BCI mapping translated the resulting moment-by-moment factor activity ( zt ) into a 2D cursor velocity ( vt ) using a Kalman filter: ( 2 ) vt=Avt−1+Bzt+c For the first BCI mapping , A∈R2×2 , B∈R2×10 , and c∈R2×1 were computed from the Kalman filter parameters , estimated using the calibration trials . For the second BCI mapping , we changed the relationship between population activity and cursor movement by randomly permuting the elements of zt before applying Equation 2 . This permutation procedure can be formulated so that Equation 2 still applies to the second BCI mapping , but with an updated definition of B ( Sadtler et al . , 2014 ) . Each animal performed an 8-target center-out task by modulating its M1 activity to control the velocity of a computer cursor . Each session involved two different BCI mappings . The first mapping was chosen to be intuitive for the animal to use . The animal used this first mapping for 200–400 trials , after which the mapping was changed abruptly to a second BCI mapping . The second mapping was initially difficult for the animal to use , and the animal was given 400–600 trials to learn to use the second mapping . Both mappings were chosen to be within the animal’s instrinic manifold , mappings that we found in previous work could be readily learned within one session ( Sadtler et al . , 2014 ) . At the beginning of each trial , a cursor appeared in the center of the workspace , followed by the appearance of one of eight possible peripheral targets ( chosen pseudorandomly ) . For the first 300 ms of the trial , the velocity of the cursor was fixed at zero . After this , the velocity of the cursor was controlled by the animal through the BCI mapping . If the animal acquired the peripheral target with the cursor within 7 . 5 s , he received a water reward , and the next trial began 200 ms after target acquisition . Otherwise , the trial ended , and the animal was given a 1 . 5 s time-out before the start of the next trial . The data analyzed in this study were part of a larger study involving learning two different types of BCI mapping changes: within-manifold perturbations ( WMP ) and outside-manifold perturbations ( OMP ) ( Sadtler et al . , 2014 ) . We found that animals learned WMPs better than OMPs . Because we need animals to show stable cursor control under both mappings , we only analyzed WMP sessions in this study . Among the WMP sessions , we further selected those in which the animal learned stable control of the second mapping ( 42 selected and 12 discarded ) . This was important because performance with the second mapping was generally not as good as with the first mapping ( Figure 1—figure supplement 1 ) , and we wanted to ensure that any potential results were not due to incomplete learning of the second mapping ( see also ‘Internal model estimation’ below ) . We further sub-selected from each session only those trials which exhibited stable behavioral performance , using a metric defined below . This was done to ensure that we were analyzing trials for which animals used a consistent strategy for selecting activity patterns . We included sessions in which there existed a block of at least 100 consecutive trials that showed both substantial learning of the second mapping and consistent behavior . To identify trials showing substantial learning , we computed the running mean of the target acquisition time ( on correct trials only ) , smoothed with a 100-trial boxcar shifted one trial at a time . The smoothed acquisition time for a trial corresponded to the average acquisition time within a 100-trial window centered on that trial . We then normalized these values so that 1 corresponded to the largest acquisition time in the first 50 trials using the second mapping , and 0 corresponded to the smallest acquisition time in the subsequent trials using the second mapping . We defined trials showing substantial learning as those with normalized acquisition times below 0 . 5 . Next , to identify trials with consistent behavior , we computed the running variance of the target acquisition time . This was computed by taking the variance of the smoothed acquisition time above in a 100-trial boxcar , shifted one trial at a time . We then normalized these variances so that 1 corresponded to the largest variance in the first half of trials using the second mapping , and 0 corresponded to the smallest variance in any trial using the second mapping . We defined trials showing stable behavior as those with normalized variance below 0 . 5 . We then identified blocks of consecutive trials that passed both of these criteria , joining blocks if they were separated by no more than 10 trials . We then selected the longest such block of at least 100 trials for our analyses . If no such block of trials was found , we excluded that session from our analyses . This procedure resulted in the 42 sessions across three monkeys that we included in our analyses . We analyzed only successful trials . To avoid analyzing time steps with potentially idiosyncratic cursor control , we also ignored portions of the trial when the cursor was closer than 50 mm or more than 125 mm away from the origin . We repeated our analyses without the latter exclusion and obtained quantitatively similar results . When an animal uses a BCI mapping , its internal conception of the BCI mapping can differ from the actual BCI mapping , even during proficient control ( Golub et al . , 2015 ) . As a result , the animal’s conception of output-potent versus output-null dimensions can be different from those defined by the actual BCI mapping . To control for this possibility , we evaluated our predictions based on the animal’s internal conception of the output-null dimensions , rather than the actual output-null dimensions of the BCI mapping . This is particularly important for the second mapping , but we also did this for the first mapping . We used a method ( Internal Model Estimation , IME ) that we developed previously for estimating the animal’s internal model of the BCI mapping ( Golub et al . , 2015 ) , with the exception that here we apply the model directly to the factor activity ( zt ) as opposed to the neural activity ( ut ) , as was done in Golub et al . ( 2015 ) . The main idea of the IME framework is that the animal generates neural activity consistent with aiming straight to the target through an internal model of the BCI mapping . Due to natural visual feedback delay , the animal cannot exactly know the current cursor position , and thus aims from an internal estimate of the current cursor position . The internal estimate of the cursor position is a feedforward prediction based on previously issued neural activity and the most recently available visual feedback . Figure 5—figure supplement 4A shows a single-trial BCI cursor trajectory ( black ) , along with the animal’s internal belief ( red ‘whisker’ ) about how cursor position ( red dots ) evolved from the cursor position known from the most recently available visual feedback . The final segments of the trajectories reflect the same neural activity , which produces the actual cursor velocity ( black arrow ) through the actual BCI mapping , or the animal’s intended cursor velocity ( red arrow ) through the animal’s internal model . The animal’s velocity command viewed through the internal model points closer toward the target than the actual movement of the BCI cursor , corresponding to a smaller angular error . Across sessions , the animals’ angular errors when using the second BCI mapping did not usually return to the original level of error that the animal achieved under the first mapping ( Sadtler et al . , 2014 ) ( Figure 5—figure supplement 4B ) . However , when viewed through the animals’ internal models of the BCI mappings , angular errors during the second mapping were more similar to those observed during the first mapping ( Figure 5—figure supplement 4C ) . Thus , the internal model helps to control for possible incomplete learning of the second mapping . We used IME to obtain the animal’s internal model of the BCI mapping ( in the form of A , B , c in Equation 2 ) , which yielded a corresponding set of cursor velocities ( vt ) , cursor-target angles ( θt ) , and bases for the output-potent and output-null dimensions of each mapping ( see N and R below ) that we used in our offline analyses . The results reported in the main text are based on these quantities obtained from IME . When we analyzed the data without using IME ( i . e . using the actual output-null dimensions of the BCI mapping ) , all of the results we report still held ( Figure 5—figure supplement 3 ) . In Equation 2 , the matrix B∈ℝ2×10 linearly projects a 10-dimensional input ( factor activity ) to a 2-dimensional output ( cursor velocity ) . Thus , for any given cursor velocity ( vt ) there are multiple values of factor activity ( zt ) that would produce it . These multiple values of factor activity are all behaviorally equivalent , and we refer to their existence as ‘neural redundancy . ’ Mathematically , it is useful to consider the null space , Nul ( B ) , and the row space , Row ( B ) , of the matrix B . The critical property of Nul ( B ) is that for any element y∈Nul ( B ) ⊆ℝ10 , we have Bx=B ( x+y ) for all x∈ℝ10 . In other words , any change in activity within the null space of B has no effect on the cursor movement produced . On the other hand , to achieve a particular cursor velocity ( vt ) , there is exactly one x∈Row ( B ) such that Bx=vt . Thus , the activity in the row space of B uniquely determines the cursor movement . To find a basis for Row ( B ) and Nul ( B ) , we took a singular value decomposition of B=USVT , where the diagonal elements of S were ordered so that only the first two values were nonzero . Then , we let R∈ℝ10×2 be the first two columns of V , and N∈ℝ10×8 be the remaining eight columns . The columns of N and R are mutually orthonormal and together form an orthonormal basis for the 10-dimensional space of factor activity . This allowed us to decompose the factor activity zt at each time step into two orthogonal components: ( 1 ) activity in the row space of B that affects the cursor velocity , which we call the output-potent activity ( ztr∈ℝ2 ) ; and ( 2 ) activity in the null space of B that does not affect the cursor movement , which we call the output-null activity ( ztn∈ℝ8 ) : ( 3 ) zt=Nztn+Rztrwhereztn:=N⊤zt , ztr:=R⊤zt Note that all behaviorally equivalent activity will have the same output-potent activity ( ztr ) , but can differ in output-null dimensions . Thus , for time steps with similar cursor movements , the subject’s choice of 8D output-null activity ( ztn ) describes how the subject selected activity from a set of behaviorally equivalent options . Because the cursor velocity ( vt ) at each time step is a combination of output-potent activity and the cursor velocity at the previous time step ( see Equation 2 ) , output-potent activity can be thought of as driving a change in the cursor velocity . Note that in the depictions of hypotheses in Figure 1 , Figure 2 , Figure 3 , and Figure 4 , we used vt=Bzt instead of Equation 2 for clarity . Our goal for each experiment was to predict the distribution of observed output-null activity during the second mapping across time steps corresponding to a given cursor movement direction ( defined as the angle of vt in Equation 2 ) . In the context of the center-out task , we assumed that cursor movements in the same direction but with different speeds were still behaviorally equivalent to the animal . This is supported by previous work that found substantially more direction-related information than speed-related information in both single-unit and population activity in M1 ( Golub et al . , 2014 ) . For this reason we assessed the output-null distribution in bins of cursor movement direction rather than cursor velocity ( i . e . direction × speed ) . All hypotheses generated predictions of the distribution of output-null activity observed while animals used the second BCI mapping , unless otherwise noted . To generate predictions of the distributions of output-null activity , we made predictions of the output-null activity at each time step . This allowed us to ensure that our predictions were consistent with the cursor kinematics observed during the experiment . We then aggregated the predictions across all time steps during the experiment with a similar cursor movement direction . In all cases , the predicted output-null activity respected the intrinsic manifold ( Sadtler et al . , 2014 ) , because the output-null activity lies in an 8-dimensional subspace of the 10-dimensional intrinsic manifold . To generate a prediction of the output-null activity for a particular time step ( ztn ) , each hypothesis had access to three sources of information recorded during the experiments . First , all hypotheses used the observed output-potent activity ( ztr ) , in order to ensure that every prediction was physiologically plausible ( see below ) . Second , all hypotheses except for the Minimal Firing hypothesis utilized factor activity recorded during use of the first BCI mapping to form their predictions of output-null activity . Finally , the Persistent Strategy hypothesis also utilized the current position of the cursor relative to the target , defined as the cursor-target angle ( θt ) . We ensured that all predictions of output-null activity ( z^tn ) corresponded to physiologically plausible neural activity ( u^t ) . By ‘physiologically plausible’ we mean that the neural activity was non-negative , and no greater than the maximum number of spikes ( per 45 ms time step ) observed for that neural unit during trials using the first BCI mapping ( umax ) . To enforce the constraint , we either incorporated the constraint 0≤u^t≤umax directly in the optimization problem ( Minimal Firing hypothesis ) , or rejected predictions of neural activity that fell outside of the constraint ( all other hypotheses ) . In the latter case , we combined the predicted output-null activity with the observed output-potent activity at that time step to form the predicted factor activity ( z^t ) . We then converted this value to neural activity using the FA generative model: ( 4 ) u^t:=Lz^t+d If this neural activity was not physiologically plausible , we attempted to generate a new prediction of z^tn according to the hypothesis . This was possible because all hypotheses incorporated some form of sampling to generate their predictions . If this procedure failed even after 100 attempts to generate a physiologically plausible prediction , we skipped making a prediction for that time step . This happened for less than 1% of all time steps . For each session , we evaluated the predicted output-null distributions of the above hypotheses in terms of how well they matched the observed output-null distributions for all time steps with similar cursor movements . To do this , we first grouped time steps by their corresponding cursor velocity into eight non-overlapping bins of cursor movement directions ( 0∘±22 . 5∘ , 45∘±22 . 5∘ , . . . , 315∘±22 . 5∘ ) . We then evaluated the accuracy of the predictions for each cursor movement direction . For consistency , all predictions were evaluated in terms of factor activity . The Minimal Firing hypothesis generated its predictions in terms of neural activity , and we converted these predictions to factor activity using Equation 1 . We sought to assess whether the variance of population activity changed in dimensions that became output-null under the second mapping . To do this , we identified the subspace of activity that was output-potent under the first mapping , but output-null under the second mapping . As before , let the columns of N be a basis for the null space of the second mapping . Now let the columns of R1 be a basis for the row space of the first mapping . Then the space spanned by the columns of ( NN⊤ ) R1∈ℝ10×2 describes the activity that would move the cursor during the first mapping but would not move the cursor during the second mapping . Let S∈ℝ10×2 be an orthonormal basis for ( NN⊤ ) R1 , which we obtained by performing a singular value decomposition . Now let Z∈ℝ10×n be a matrix of n factor activity patterns . To measure the amount of variance of Z in the subspace spanned by the columns of S , we computed Trace ( Cov ( Z⊤S ) ) ∈ℝ . To assess how the variance of activity changes when it becomes irrelevant to cursor control , we grouped the time steps based on the cursor movement angle under the second mapping , for activity recorded under both the first and second mappings . First conditioning on the movement angle under the second mapping is consistent with our earlier analyses , when comparing the predicted and observed output-null distributions . To compute the cursor movement angle through the second mapping for activity recorded under the first mapping , we used the terms of Equation 2 not involving the cursor velocity at the previous time step ( i . e . we computed vt=Bzt+c ) . For consistency , we recomputed the cursor movement angle for activity recorded under the second mapping in the same way . Let Z1 and Z2 be the factor activity in the same cursor movement angle bin recorded during the first and second mappings , respectively . We then computed the ratio of variance R as follows: ( 16 ) R=log ( Trace ( Cov ( Z2⊤S ) ) Trace ( Cov ( Z1⊤S ) ) ) The sign of R specifies whether the variance of activity increased ( R>0 ) or decreased ( R<0 ) when that activity became irrelevant to cursor control under the second mapping . We took the average of this ratio across all cursor movement direction bins to compute a ratio for each session . To compute this ratio for the predictions of our hypotheses , as in Figure 6C , we substituted Z2 with the predictions of our hypotheses , i . e . by combining their predicted output-null activity with the observed output-potent activity under the second mapping . We also repeated the above analyses on our predictions of output-null activity produced during the first mapping using the activity observed under the second mapping , as shown in Figure 5—figure supplement 2 and Figure 6—figure supplement 1 . This was done by swapping the roles of the first and second mappings in the above analysis description . For each cursor direction on each session , we computed the distance from the mean observed output-null activity to the mean predicted by the Minimal Deviation hypothesis , where the distance was computed as the ℓ2 norm between the two 8D mean vectors . We then compared this distance to the distance between the mean predicted by Fixed Distribution and the mean predicted by Minimal Deviation ( Figure 6—figure supplement 2 ) . If the latter distance was consistently smaller than the former , this would be evidence that the observed mean output-null activity had moved towards the predictions of Minimal Deviation , relative to what was predicted by Fixed Distribution . We did not find evidence that this was the case ( one-sided Wilcoxon signed rank test , p>0 . 5 ) , suggesting that the mean observed output-null activity was not closer to Minimal Deviation than expected under Fixed Distribution . We repeated the same analysis using the mean predicted by Minimal Firing instead of Minimal Deviation , and reached the same results ( one-sided Wilcoxon signed rank test , p>0 . 5 ) . | When you swing a tennis racket , muscles in your arm contract in a specific sequence . For this to happen , millions of neurons in your brain and spinal cord must fire to make those muscles contract . If you swing the racket a second time , the same muscles in your arm will contract again . But the firing pattern of the underlying neurons will probably be different . This phenomenon , in which different patterns of neural activity generate the same outcome , is called neural redundancy . Neural redundancy allows a set of neurons to perform multiple tasks at once . For example , the same neurons may drive an arm movement while simultaneously planning the next activity . But does performing a given task constrain how often different patterns of neural activity can be produced ? If so , this would limit whether other tasks could be carried out at the same time . To address this , Hennig et al . trained macaque monkeys to use a brain-computer interface ( BCI ) . This is a device that reads out electrical brain activity and converts it into signals that can be used to control another device . The key advantage of a BCI is that the redundant activity patterns are precisely known . The monkeys learned to use their brain activity , via the BCI , to move a cursor on a computer screen in different directions . The results revealed that monkeys could only produce a limited number of different patterns of brain activity for a given BCI cursor movement . This suggests that the ability of a group of neurons to multitask is restricted . For example , if the same set of neurons is involved in both planning and performing movements , then an animal’s ability to plan a future movement will depend on the one it is currently performing . BCIs can help patients who have suffered stroke or paralysis . They enable patients to use their brain activity to control a computer or even robotic limbs . Understanding how the brain controls BCIs will help us improve their performance and deepen our knowledge of how the brain plans and performs movements . This might include designing BCIs that allow users to multitask more effectively . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Constraints on neural redundancy |
During embryonic development , cell type-specific transcription factors promote cell identities , while epigenetic modifications are thought to contribute to maintain these cell fates . Our understanding of how genetic and epigenetic modes of regulation work together to establish and maintain cellular identity is still limited , however . Here , we show that DNA methyltransferase 3bb . 1 ( dnmt3bb . 1 ) is essential for maintenance of hematopoietic stem and progenitor cell ( HSPC ) fate as part of an early Notch-runx1-cmyb HSPC specification pathway in the zebrafish . Dnmt3bb . 1 is expressed in HSPC downstream from Notch1 and runx1 , and loss of Dnmt3bb . 1 activity leads to reduced cmyb locus methylation , reduced cmyb expression , and gradual reduction in HSPCs . Ectopic overexpression of dnmt3bb . 1 in non-hematopoietic cells is sufficient to methylate the cmyb locus , promote cmyb expression , and promote hematopoietic development . Our results reveal an epigenetic mechanism supporting the maintenance of hematopoietic cell fate via DNA methylation-mediated perdurance of a key transcription factor in HSPCs .
Generating and maintaining stable cell fates is crucial for normal development . Many of the key cell types generated during embryonic development must be maintained throughout adult life to support normal organ and tissue function and homeostasis . Hematopoietic stem and progenitor cells ( HSPCs ) are formed during early development and are responsible for maintaining the lifelong supply of blood cells , but the factors necessary for their generation and maintenance are still not well understood . HSPCs arise during embryogenesis from hemogenic endothelial cells located in the ventral floor of the trunk dorsal aorta ( Bertrand et al . , 2010a; Kissa and Herbomel , 2010; Boisset et al . , 2010; Murayama et al . , 2006 ) that enter the circulation to seed hematopoietic organs and tissues and eventually give rise to all differentiated blood lineages ( Bertrand et al . , 2010a; Kissa and Herbomel , 2010; Murayama et al . , 2006 ) . The emergence of HSPCs from aortic endothelium is regulated by a Notch-Runx1-Cmyb signaling pathway ( Burns et al . , 2005; Clements et al . , 2011; Richard et al . , 2013; Bertrand et al . , 2010b; Rowlinson and Gering , 2010 ) . Studies in mice and zebrafish have shown that the Runx1 transcription factor is required downstream from Notch signaling and upstream from Cmyb for HSPC development from the endothelium , although once HSPCs are specified , Runx1 transcription is down-regulated and it is not necessary for HSPC fate maintenance ( Chen et al . , 2009; Liakhovitskaia et al . , 2009; Sood et al . , 2010; Kissa and Herbomel , 2010; Tober et al . , 2013; Lam et al . , 2009 ) . The transcription factor Cmyb is also required for HSPC formation downstream from Runx1 . However , unlike Runx1 , Cmyb expression persists in HSPC and its function is required not only for differentiation of HSPCs but also for their subsequent maintenance ( Mukouyama et al . , 1999; Mucenski et al . , 1991; Soza-Ried et al . , 2010; Zhang et al . , 2011 ) . It is not clear how transient expression of Runx1 leads to the long-lasting expression and persistent functional role for Cmyb and maintenance of HSPC identity . Epigenetic regulatory mechanisms have been implicated in long-term maintenance of gene expression ( Deaton and Bird , 2011 ) and differentiated cell fate ( Hu et al . , 2012; Mohn et al . , 2008; Lay et al . , 2015; Wu et al . , 2010 ) , and we hypothesized that they might play a role in HSPC fate maintenance during the Runx1-independent phase . A number of different mechanisms have been described for epigenetic regulation of gene expression , including post-translational modification of histones by the covalent addition of acetyl , methyl , and other groups to specific histone amino acid residues ( Greer and Shi , 2012 ) . DNA methylation is another well-studied and important mechanism for epigenetic gene regulation ( Suzuki and Bird , 2008 ) . In eukaryotes , DNA methyltransferases ( DNMTs ) add a methyl group to the 5 position of cytosine residues in DNA , typically on cytosines in CpG dinucleotides . CpGs are often grouped in clusters called 'CpG islands' ( CGIs ) , which are themselves frequently found within or in the 5’ regions of genes ( Deaton and Bird , 2011; Suzuki and Bird , 2008 ) . DMNTs are classified as either 'maintenance' DNMTs , responsible for preserving existing DNA methylation after every cellular DNA replication cycle , or 'de novo' DNMTs , which place new methyl 'marks' on DNA . Mammalian DNMT1 acts as a maintenance enzyme and has high affinity for binding to hemimethylated DNA . Other mammalian DNMTs , including the DNMT3 family enzymes DNMT3A and DNMT3B , act as de novo methyltransferases and establish initial DNA methylation patterns ( Goll and Bestor , 2005; Xu et al . , 2010; Cheng and Blumenthal , 2008 ) . Genomic imprinting by DNA methylation has a well-documented role in reducing gene expression in both plants and animals ( Reik and Dean , 2001 ) , for example , during X-chromosome inactivation in mammals ( Tada et al . , 2000; Gendrel and Heard , 2014 ) or repression of imprinted allele expression during Arabidopsis embryonic development ( Jullien and Berger , 2009 ) . DNA methylation can also act as a positive regulator for gene expression , particularly when methylation occurs in gene bodies as opposed to promoter or enhancer regions ( Baubec et al . , 2015; Yang et al . , 2014 ) . In mouse stem cells , DNMT3B selectively binds to and methylates gene body regions , excluding promoter and enhancer regions . DNMT3B binding is found preferentially in gene body regions of actively expressed genes ( Baubec et al . , 2015 ) . In a recent elegant study performed in the chick , DNMT3A was shown to act as an epigenetic switch repressing neural and promoting neural crest fate by binding to and methylating the SOX2 and SOX3 promoters and inhibiting their expression ( Hu et al . , 2012 ) . These findings highlight the important role that epigenetic regulation plays in establishing cell fate and the diverse functions DNMTs play in regulating gene expression during early embryonic development . Here , we examine the molecular mechanisms responsible for maintaining HSPC fate independent of Runx1 function in the zebrafish . We find that the de novo DNA methyltransferase Dnmt3bb . 1 functions downstream from Runx1 to maintain cmyb gene expression and HSPC cell fate . Our findings reveal a previously unknown epigenetic mechanism regulating HSPC fate maintenance .
Using whole-mount in situ ( WISH ) hybridization , we find that around 36 hpf selected cells in the developing zebrafish dorsal aorta specifically express DNA methyltransferase 3bb . 1 ( dnmt3bb . 1 ) ( Figure 1a , b ) , a gene most closely related to human DNMT3B ( Figure 1—figure supplement 1a ) . As noted above , DNMT3B functions as a de novo DNMT in humans , adding new methyl 'marks' to cytosine residues in DNA . Dnmt3bb . 1-positive cells are present in the ventral floor of the dorsal aorta ( Figure 1c , d ) ( Takayama et al . , 2014; Thisse , 2001 ) ) , and double WISH confirms that dnmt3bb . 1 is co-expressed with cmyb ( Figure 1e , f ) , a known marker for developing HSPCs ( Gering and Patient , 2005 ) . dnmt3bb . 1 is expressed specifically in HSPCs in the developing trunk , although expression is also present in portions of the eye and a few other selected head tissues during early development ( Aanes et al . , 2011 ) ( Figure 1—figure supplement 1b–e ) . As noted above , a Notch-Runx1-Cmyb pathway regulates the endothelial to HSPC transition ( Figure 1g ) ( Burns et al . , 2005; Chen et al . , 2009; Kissa and Herbomel , 2010 ) . To determine whether dnmt3bb . 1 acts in conjunction with this pathway , we examined the expression of dnmt3bb . 1 using WISH and quantitative reverse transcriptase-polymerase chain reaction ( qRT-PCR ) after manipulating runx1 or notch ( Figure 1h–m ) . Expression of dnmt3bb . 1 is strongly reduced in runx1 morpholino-injected animals or runx1W84X mutants ( Sood et al . , 2010 ) ( Figure 1h , i , m ) , while over-expression of runx1 by injection of runx1 mRNA results in increased expression of dnmt3bb . 1 in the axial vasculature ( Figure 1i , j , m ) . Similarly , expression of dnmt3bb . 1 is also strongly reduced in Notch-deficient mindbomb ( mib ) mutants ( Figure 1k , l , m ) . These results show that dnmt3bb . 1 is expressed by developing HSPCs downstream from the established Notch-Runx1 pathway for HSPC specification ( Burns et al . , 2005; Chen et al . , 2009; Kissa and Herbomel , 2010 ) , and that its expression is lost when HSPCs are not properly specified . 10 . 7554/eLife . 11813 . 003Figure 1 . The DNA methyltransferase3bb . 1 gene is expressed in developing hematopoietic stem and progenitor cells and is regulated by hematopoietic stem and progenitor cell ( HSPC ) -specific pathways . ( a ) Camera lucida drawing of a 24 hpf zebrafish embryo with a red box noting the approximate region of the trunk shown in in situ hybridization images . ( b , c ) Whole-mount in situ hybridization of a 36 hpf zebrafish trunk probed for dnmt3bb . 1 , showing expression in the ventral floor of the dorsal aorta ( arrows in panel c ) . Yellow inset box in panel b indicates the magnified area shown in panel c . ( d ) Diagram corresponding to panel c showing dorsal aorta ( red ) , cardinal vein ( blue ) , and dnmt-positive cells in the floor of the dorsal aorta ( yellow ) . ( e , f ) Double in situ hybridization of a 36 hpf zebrafish trunk probed for dnmt3bb . 1 ( e , blue ) and c-myb ( f , red ) . The c-myb positive HSPC progenitors also stain for dnmt3bb . 1 ( arrows ) . ( g ) A Notch- Runx- c-myb pathway regulates HSPC emergence in the zebrafish . ( h–j ) In situ hybridization of 36 hpf runx1 morpholino-injected ( h ) , control ( i ) , or runx1 mRNA-injected ( j ) zebrafish trunks probed for dnmt3bb . 1 , showing that dnmt3bb . 1 expression is reduced by runx1 knockdown ( asterisks ) and increased by runx1 overexpression ( arrows ) . ( k , l ) In situ hybridization of 36 hpf wild type ( k ) or mind bomb mutant ( l ) zebrafish trunks probed for dnmt3bb . 1 , showing that dnmt3bb . 1 expression seen in wild-type siblings ( arrows in panel k ) is strongly reduced in Notch-deficient mind bomb mutants ( asterisks in panel l ) . ( m ) Quantitative reverse transcriptase-polymerase chain reaction ( qRT-PCR ) analysis of dnmt3bb . 1 transcript levels in 36 hpf ( i ) control , ( ii ) runx1W84X mutant , ( iii ) runx1 mRNA injected , ( iv ) mind bomb ( mib ) wild type sibling ( wild type sibling of mib mutants ) , and ( v ) mib mutant zebrafish embryos . dnmt3bb . 1 levels are normalized to the reference gene elf1α and to levels in controls . All graphs show mean ± standard error of the mean ( SEM ) and are representative of three biological replicates . Scale bars = 100 µm in b , h-l and 50 µm in c , f . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 00310 . 7554/eLife . 11813 . 004Figure 1—figure supplement 1 . Dnmt3bb . 1expression analysis and homology . ( a ) Phylogenetic tree showing zebrafish Dnmt3bb . 1 is the closest homolog of human DNMT3B . ( b–d ) WISH showing non-hematopoietic expression of dnmt3bb . 1 , b , cmyb , c , and dnmt3bb . 1 and cmyb , d , in the heads of 36 hpf zebrafish . ( e ) relative levels of dnmt3bb . 1 in embryos at different developmental stages . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 004 To test whether Dnmt3bb . 1 function is necessary for hematopoietic development , we generated genetic mutants in the dnmt3bb . 1 locus using TALEN technology ( Dahlem et al . , 2012 ) . We obtained a dnmt3bb . 1y258 mutant allele encoding a polypeptide prematurely truncated at amino acid 172 ( P166fs ) due to an 11 nucleotide deletion ( Figure 2—figure supplement 1a–e ) . Morphologically , two to five dpf dnmt3bb . 1y258 mutants are indistinguishable from their wild-type siblings ( Figure 2a , b ) . At 36 hpf , cmyb expression in HSPCs in the ventral aorta is similar in dnmt3bb . 1y258 mutant embryos and their wild type siblings ( Figure 2c , d ) . However , by 72 and 96 hpf cmyb expression is strongly reduced in the caudal hematopoietic tissue ( CHT ) of dnmt3bb . 1y258 mutants ( Figure 2e–h ) , as confirmed by genotyping of blindly scored embryos ( Figure 2—figure supplement 1e ) . Loss of cmyb is also accompanied by strong reduction in myeloid marker l-plastin ( Herbomel et al . , 1999 ) in the CHT at 72 hpf ( Figure 2i , j ) and lymphoid marker rag1 ( Willett et al . , 1997 ) in the 5 dpf thymus ( Figure 2k , l ) . We observed similar hematopoietic defects in wild type embryos injected with any one of three different dnmt3bb . 1-targeting morpholinos , with progressive loss of cmyb expression ( Figure 2—figure supplement 2a–f ) as well as reduced expression of l-plastin ( Figure 2—figure supplement 2g , i ) and erythroid marker gata1 ( Figure 2—figure supplement 2k–n ) in the 72 hpf CHT , and reduced expression of rag1 ( Figure 2—figure supplement 2h , j ) , ikaros , and lck:gfp ( Figure 2—figure supplement 2o–r ) in the 5 dpf thymus . Quantitative RT-PCR confirmed progressive loss of cmyb and strong reduction in subsequent l-plastin and rag1 expression in Dnmt3bb . 1-deficient animals ( Figure 2m , n ) . Strong increase in active caspase 3 in HSPC in the trunks of 48 hpf Dnmt3bb . 1-deficient animals indicated that HSPCs are undergoing apoptosis ( Figure 2—figure supplement 3a–c ) . Despite dramatic hematopoietic defects , Dnmt3bb . 1-deficient animals appeared otherwise unaffected , with normal vascular patterning and vascular gene expression ( Figure 2—figure supplement 3d–i ) . These results suggest that loss of dnmt3bb . 1 activity leads to a specific defect in HSPCs , with progressive loss of cmyb expression and decreased numbers of HSPCs . 10 . 7554/eLife . 11813 . 005Figure 2 . Dnmt3bb . 1is necessary for hematopoietic gene expression . ( a , b ) Transmitted light images of 72 hpf control ( a ) and dnmt3bb . 1y258 ( b ) mutant zebrafish , showing absence of developmental delay or gross morphological abnormalities in dnmt3bb . 1y258 mutants . ( c–h ) Whole mount in situ hybridization of 36 hpf ( c , d ) , 72 hpf ( e , f ) and 96 hpf ( g , h ) wild type ( WT ) sibling ( c , e , g ) or dnmt3bb . 1y258 ( d , f , h ) mutant animals , probed for cmyb . Cmyb is expressed in dnmt3bb . 1 mutant at 36 hpf ( arrows ) but this expression is strongly reduced by 72 and 96 hpf ( arrowheads ) . ( i–l ) Whole mount in situ hybridization of 72 hpf tails probed for l-plastin ( i , j ) and 5 dpf heads probed for rag1 ( k , l ) from WT sibling ( i , k ) or dnmt3bb . 1 ( j , l ) mutant animals . Expression of both l-plastin and rag1 is present in controls ( arrows ) but strongly reduced in dnmt3bb . 1 mutant animals ( arrowheads ) . ( m ) Quantitative RT-PCR analysis of cmyb transcript levels in control ( blue columns ) or dnmt3bb . 1 ( red columns ) morpholino injected animals at 36 , 48 , and 72 hpf . Transcript levels are normalized to the reference gene elf1α and to levels in 36 hpf control morphants . ( n ) Quantitative RT-PCR analysis of 3 dpf l-plastin and 5 dpf rag1transcript levels in control ( blue columns ) or dnmt3bb . 1 ( red columns ) morpholino-injected animals . Transcript levels are normalized to the reference gene elf1α and to levels in controls . All graphs in panels i and j show mean ± SEM , are representative of three biological replicates . Scale bars = 150 µm in a , b , 50 µm in c , d , 100 µm in e-l . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 00510 . 7554/eLife . 11813 . 006Figure 2—figure supplement 1 . TALENmediated Dnmt3bb . 1 mutation . ( a ) Schematic diagram showing the position of the TALEN target site within exon 6 of dnmt3bb . 1 . ( b ) TALEN pair binding sites and spacer between them in the WT dnmt3bb . 1 locus . The spacer region includes a Sac I restriction enzyme site . ( c ) dnmt3bb . 1y258 mutation with an 11 nucleotide deletion resulting in loss of the Sac I site . ( d ) The dnmt3bb . 1y258 mutation causes a frame shift after amino acid 166 that results in a truncated polypeptide of 172 amino acids lacking the DNA methylase domain as well as other important functional domains . ( e ) A representative genotyping gel containing PCR products digested with SacI , amplified from genomic DNA prepared from individual 72 hpf WT , dnmt3bb . 1y258 heterozygous , and dnmt3bb . 1y258 homozygous mutant embryos after blind scoring for cmyb expression by in situ hybridization . Eight out of 32 embryos scored showed weak expression of cmyb at 72 hpf , and all eight were homozygous mutants while the rest were wild type or heterozygotes . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 00610 . 7554/eLife . 11813 . 007Figure 2—figure supplement 2 . Dnmt3bb . 1 morpholino phenocopies TALEN-induced mutant defects . ( a ) Schematic diagram showing the splice acceptor morpholino target site at the dnmt3bb . 1 intron 3-Exon 4 boundary ( red line ) and positions of primers used to detect splicing changes ( blue arrows ) . ( b ) RT-PCR gel image showing loss of correctly spliced dnmt3bb . 1 transcripts in splice acceptor morpholino-injected animals . ( c–f ) WISH of 36 hpf trunks ( c , d ) and 72 hpf tails ( e , f ) from control ( c , e ) and dnmt3bb . 1 ( d , f ) morpholino injected animals , probed for cmyb . cmyb is expressed in dnmt3bb . 1 morphants at 36 hpf ( arrows ) but the expression is lost by 72 hpf ( arrowheads ) . ( g–j ) WISH of 72 hpf tails probed for l-plastin ( g , i ) and 5 dpf heads probed for rag1 ( h , j ) from control ( g , h ) or dnmt3bb . 1 ( i , j ) morpholino-injected animals . ( k–n ) WISH of 5 dpf embryos probed for gata1 expression in control ( k , m ) and dnmt3bb . 1 ( l , n ) morpholino injected embryos . m , n , higher magnification images of highlighted areas from k , l . ( o ) Quantitative RT-PCR analysis at 5 dpf lck , ikaros and eGFPtranscript levels in WT siblings ( red columns ) , dnmt3bb . 1 mutants ( green columsn ) , control morpholino ( blue columns ) or dnmt3bb . 1 ( yellow columns ) morpholino-injected animals . Transcript levels are normalized to the reference gene elf1α and to levels in controls . ( p , q ) WISH of 5 dpf embryos probed for ikaros expression in control ( p ) and dnmt3bb . 1 ( q ) morpholino injected embryos . Expression of l-plastin , rag1 , gata1 and ikaros is present in controls ( arrows ) but reduced in dnmt3bb . 1 morpholino-injected animals ( arrowheads ) . ( r , s ) Confocal images of 5 dpf Tg ( lck:gfp ) embryos injected with control ( q ) or dnmt3bb . 1 morpholinos . Scale bars 50 µm in d , 100 µm in f , i , j , n , p , 25 μm in r . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 00710 . 7554/eLife . 11813 . 008Figure 2—figure supplement 3 . HSPCapoptosis but no gross morphological and vascular defects in in dnmt3bb . 1 morphants . ( a , b ) Immunohistochemical staining of the trunks of 48 hpf control ( a ) and dnmt3bb . 1 ( b ) morpholino-injected Tg ( cmyb:gfp ) transgenic zebrafish with anti-GFP ( green ) and anti-caspase3 ( red ) antibodies , showing increased numbers of caspase3-positive HSPC in dnmt3bb . 1-deficient animals . ( c ) Quantitation of the percentage of GFP ( cmyb ) positive HSPCs that are also caspase3-positive in 48 hpf control or dnmt3bb . 1 morpholino-injected Tg ( cmyb:gfp ) transgenic zebrafish . Columns show mean ± SEM , and statistical significance calculated by paired t-test is shown ( p = 0 . 0012 ) . ( d , e ) Transmitted light images of whole 36 hpf control ( d ) and dnmt3bb . 1 ( e ) morpholino-injected zebrafish , showing absence of developmental delay or gross morphological abnormalities in dnmt3bb . 1 morphants . ( f , g ) Confocal images of trunk EGFP fluorescence in 36 hpf control ( f ) or dnmt3bb . 1 ( g ) morpholino-injected Tg ( fli1a:egfp ) y1 transgenic zebrafish , showing normal trunk vessel patterning in dnmt3bb . 1 morphants . ( h , i ) Whole mount in situ hybridization of the trunks of 36 hpf control ( h ) or dnmt3bb . 1 ( i ) morpholino-injected zebrafish probed for vecdn , showing normal expression in dnmt3bb . 1 morphants . All images are lateral views , rostral to the left . Scale bars 25 µm in b , 100 µm in e and 50 µm in g , i . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 008 To examine whether changes in DNA methylation correlate with hematopoietic defects resulting from the loss of dnmt3bb . 1 , we isolated HSPCs from 36 hpf control or dnmt3bb . 1-deficient animals using a previously reported double-transgenic Fluorescence Activated Cell Sorting ( FACS ) method ( Bertrand et al . , 2010a ) ( Figure 3a ) . DNA and RNA were prepared simultaneously from HSPC-enriched samples for global analysis of DNA methylation and gene expression using Reduced Representation Bisulfite Sequencing ( RRBS ) ( Meissner et al . , 2005 ) and RNAseq , respectively ( Figure 3a , Figure 3—figure supplement 1a ) . RNAseq analysis ( and subsequent qPCR validation of selected genes ) showed that the expression of a variety of hematopoietic genes was reduced in dnmt3bb . 1-deficient HSPCs , while many endothelial and apoptosis pathway genes were increased relative to control HSPCs ( Figure 3b , c , Figure 3—figure supplement 1b ) . Ingenuity pathway analysis ( www . qiagen . com/ingenuity ) confirmed elevated VEGF/angiogenesis and apoptosis pathway gene expression indicating lack of endothelial to hematopoietic transition and induction of apoptosis in dnmt3bb . 1-deficient HSPCs ( Figure 3d ) . However , RRBS analysis revealed that very few of the hematopoietic or endothelial genes with significantly altered expression displayed significantly decreased DNA methylation ( Figure 3—source data 1 ) . cmyb was one of the most affected and highly correlated of the hematopoietic genes that showed both significantly reduced expression in dnmt3bb . 1-deficient HSPCs and significantly reduced methylation of an adjacent promoter or gene body CpG island ( Figure 3e , Figure 3—source data 1 ) . Direct sequencing of PCR-amplified fragments from bisulfite treated DNA confirmed reduced methylation in the cmyb intron 1 CpG island in dnmt3bb . 1-deficient HSPCs ( Figure 3f ) , but no change in the runx1 exon 3 CpG island ( Figure 3g ) , or in CpG islands from a number of other hematopoietic and endothelial genes showing significantly altered expression in dnmt3bb . 1-deficient HSPC ( Figure 3—figure supplement 1c , d ) . These results suggest that cmyb is a target of dnmt3bb . 1-mediated regulation in HSPCs . 10 . 7554/eLife . 11813 . 009Figure 3 . Dnmt3bb . 1is necessary for epigenetic regulation of hematopoietic gene expression . ( a ) Schematic diagram showing the experimental procedure for the isolation of GFP/mApple double-positive HSPCs from cmyb:GFP , kdrl:mApple double transgenic zebrafish and use of DNA and RNA from these HSPC for next-gen sequending projects ( b ) RNAseq analysis showing differentially expressed endothelial and hematopoietic genes in dnmt3bb . 1 deficient HSPCs . ( c ) qRT-PCR analysis of selected genes from the RNAseq analysis . ( d ) Ingenuity Pathway Analysis ( IPA ) of the RNAseq data showing top pathways up- or down-regulated in dnmt3bb . 1-deficient HSPCs . ( e ) Top five HSC-expressed genes with the most significantly reduced DNA methylation and expression from bisulfite and RNAseq analysis , respectively , as well as the top 2 endothelial genes . ( f , g ) Bisulfite sequencing analysis of DNA methylation at the cmyb intron 1 ( f ) and runx1 exon 3 ( g ) CpG islands in DNA isolated from control ( 'Ctrl MO' , top ) and dnmt3bb . 1 ( 'Dnmt3bb . 1 MO' , bottom ) morpholino-injected HSPCs , showing strongly reduced methylation of the cmyb CpG island but not the runx1 CpG island in dnmt3bb . 1-deficient animals . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 00910 . 7554/eLife . 11813 . 010Figure 3—source data 1 . Hematopoieticand Vascular genes with reduced methylation and altered expression in dnmt3bb . 1 morpholino-injected animals . DNA methylation ( p<0 . 0001 ) and RNA ( p<0 . 001 ) expression fold changes for hematopoietic ( left ) and vascular ( right ) genes showing the most significant reduction in methylation in the RRBS dataset and change in RNA expression in the RNAseq dataset from FACS-enriched ( as in Figure 3a ) HSPC from dnmt3bb . 1 morpholino- vs . control morpholino-injected animals . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 01010 . 7554/eLife . 11813 . 011Figure 3—figure supplement 1 . Analysisof RNA seq and DNA methylation . ( a ) Schematic diagram depicting the workflow for global RNAseq and DNA methylation ( RRBS ) analysis of control or dnmt3bb . 1 deficient HSPCs . ( b ) RNA seq analysis showing upregulation of cell death genes in dnmt3bb . 1-deficient HSPCs . ( c , d ) Bisulfite sequencing of CpG islands from hematopoietic ( c ) and endothelial ( d ) genes identified in the RNAseq analysis including hematopoietic genes gata3 , lmo2 , erg , bmp4 and runx1 , c , and endothelial genes dll4 , etv2 , cldn5b and cdh5 , d . None of the genes showed changes in methylation of their CpG islands . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 011 To further validate our findings regarding methylation of hematopoietic genes using an independent method , we identified 319 genes that were annotated as being expressed during definitive hematopoiesis from a comprehensive database of zebrafish gene expression patterns ( http://zfin . org ) . 57 of these genes were found to have a total of 72 CpG islands in their gene bodies ( we concentrated on gene body CpG islands since these are associated with active gene expression [Challen et al . , 2012; Wu et al . , 2010; Trowbridge and Orkin , 2012; Gao et al . , 2011] ) . We carried out bisulfite restriction analysis to determine whether these 72 CpG islands are methylated in blood cells , using genomic DNA isolated from circulating blood cells obtained after amputating the tails of 5–7-day-old zebrafish ( Figure 4a ) , and determined that 10 of the 72 CpG islands are methylated in blood cells . To determine whether methylation of these CpG islands was dnmt3bb . 1-dependent , we carried out bisulfite sequencing on each of the 10 methylated CpG islands using 5 dpf blood cell genomic DNA isolated from either control or dnmt3bb . 1 morpholino-injected zebrafish . Only one CpG island showed dnmt3bb . 1-dependent changes in methylation , the cmyb intron 1 CpG island ( Figure 4b ) . No change was detected in the methylation of the runx1 exon 3 CpG island ( Figure 4c ) , or in any of the other eight hematopoietic genes containing methylated CpG islands in blood cells ( Figure 4d–k ) . These results support the findings of our RRBS analysis above , suggesting that cmyb is a key proximal hematopoietic target of dnmt3bb . 1 . 10 . 7554/eLife . 11813 . 012Figure 4 . Measurementof DNA methylation by bisulfite sequencing of gene body CpG islands from all known hematopoietic genes . ( a ) Schematic diagram showing the method used to collect blood from 5–7 dpf embryos for bisulfite analysis . ( b–k ) Bisulfite sequencing of blood cell genomic DNA from 5–7 dpf control ( top ) and dnmt3bb . 1 ( bottom ) morpholino-injected animals . The results are shown for ten different sequences identified in a genome-wide screen for gene body CpG islands methylated in blood cells: cmyb ( b ) , runx1 ( c ) , CCAT/enhancer binding protein 1-Exon 1 ( d ) , Kruppel like factor 4-Exon 3 ( e ) , Nuclear receptor co-repressor-2-Exon 37 ( f ) , RALBP1 associated domain containing 2-Exon 2 ( g ) , short stature homeobox-Exon 5 ( h ) , solute carrier family 12-Exon 1 ( i ) and solute carrier family 8–5’ UTR ( j ) , lamin a-Exon 7 ( k ) . Open circles represent unmethylated and filled circles represent methylated cytosine residues from CpG dinucleotides . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 012 To examine whether Dnmt3bb . 1 can promote cmyb gene expression in endothelial cells , we drove mosaic , pan-endothelial expression of a Dnmt3bb . 1-GFP fusion protein by injecting a Tol2 ( kdrl:dnmt3bb . 1-gfp ) transgene ( Figure 5a and Figure 5—source data 1 ) into zebrafish embryos , and assayed for expression of cmyb in Dnmt3bb . 1-GFP positive cells . In wild type animals , cmyb is expressed in a limited number of endothelial precursors found in the ventral floor of the dorsal aorta in the trunk . In Tol2 ( kdrl:dnmt3bb . 1-gfp ) transgene-injected animals Dnmt3bb . 1-GFP-positive cells were consistently also cmyb positive , and double–positive cells were found throughout the vasculature , not only in their normal location in the trunk ( Figure 5b , c , d ) , but also in ectopic locations such as the cranial vasculature ( Figure 5b , e , f ) . We observed similar induction of cmyb expression in dnmt3bb . 1-gfp expressing cells when the transgene was injected into runx1 mutant animals lacking endogenous cmyb expression in HSPCs ( Figure 5b , g , h ) , showing that the forced expression of dnmt3bb . 1 can promote cmyb expression in endothelial cells even in the absence of runx1 function . 10 . 7554/eLife . 11813 . 013Figure 5 . Dnmt3bb . 1is sufficient for cmyb gene expression in the endothelium . ( a ) Schematic diagram of the I-Sce1 ( kdrl:dnmt3bb . 1-gfp ) transgene used for the pan-endothelial expression of a dnmt3bb . 1-gfp fusion protein . ( b ) Camera lucida drawing of a 24 hpf zebrafish embryo with red , blue and green boxes noting the approximate regions of the head , trunk and trunk/tail shown in in situ hybridization images in panels c-h . ( c , e ) Double whole-mount in situ hybridization of the head ( e ) and trunk ( c ) of 36 hpf I-Sce1 ( kdrl:dnmt3bb . 1-gfp ) transgene-injected zebrafish probed for gfp ( blue ) and cmyb ( red ) . ( d , f ) fluorescence images of in situ hybridization corresponding to panels c and e , respectively . In panels c-e , endothelial cells expressing both the gfp transgene and cmyb are noted with arrows , while normal trunk HSC expressing only cmyb but not gfp are noted with arrowheads . ( g , h ) Double whole-mount in situ hybridization of the trunk/tail of a 36 hpf runx1 mutant ( g ) and a 36 hpf runx1 mutant injected with I-Sce1 ( kdrl:dnmt3bb . 1-gfp ) transgene ( h ) , probed for cmyb ( blue ) and gfp ( red ) . The cmyb gene is not expressed in runx1 mutants , but injection of dnmt3bb . 1-egfp fusion protein into runx1 mutants results in the appearance of cmyb/gfp double-positive cells . Scale bars 25 µm in e , f and 100 µm in h . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 01310 . 7554/eLife . 11813 . 014Figure 5—source data 1 . Mosaic expression of dnmt3bb . 1:gfp in endothelial cells induces cmyb expression . ( I , II ) Counts of numbers of 36 hpf ISceI ( kdrl:gfp ) - ( I ) or ISceI ( kdrl:dnmt3bb . 1-gfp ) - ( II ) injected wild type animals with GFP transgene ( purple ) and cmyb ( red ) double-positive cells in either the trunk or head . ( III ) Counts of numbers of 3 dpf ISce1 ( kdrl:dnmt3bb . 1-gfp ) -injected runx1 mutant animals with GFP ( purple ) and cmyb ( red ) double-positive cells in the trunk . No cmyb-positive cells are observed in the trunks of un-injected runx1 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 014 Since endothelial cells might have intrinsic preconditioning making them more amenable to differentiating into HSPCs , we examined whether cmyb and other hematopoietic genes could also be induced in non-endothelial cells by forced overexpression of dnmt3bb . 1 . We mosaically mis-expressed Dnmt3bb . 1-GFP fusion protein in cells in the early zebrafish embryo , well before the initiation of normal hematopoietic development , by heat-shocking Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene-injected animals at the blastula stage ( 3 hpf ) and then assaying expression of cmyb and downstream hematopoietic lineage markers at the shield stage ( 6 hpf ) by WISH and qRT-PCR ( Figure 6a , b ) . As expected , heat-shocked control Tol2 ( hsp70:gfp ) transgene-injected animals do not express cmyb , rag1 , or l-plastin ( Figure 6c-e , k ) . In contrast , heat shocked animals expressing Dnmt3bb . 1-GFP fusion protein from the injected Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene showed robust mosaic expression of cmyb , rag1 , and l-plastin ( Figure 6f-h , k ) . Dnmt3bb . 1-GFP-injected embryos also express lymphoid markers ikaros and rag2 , as well as hemoglobin ae1 ( hbae1 ) , a marker of both primitive and definitive erythroid cells ( Jin et al . , 2009 ) , showing induction of erythroid lineage genes also occurs ( Figure 6—figure supplement 1a–h ) , but they do not express endothelial markers cadherin 5 ( cdh5 ) and ets-related protein ( etsrp , also known as etv2 ) ( Figure 6—figure supplement 1i–l ) . This suggests that dnmt3bb . 1 specifically promotes hematopoietic development without inducing mesoderm , endothelial precursors , or ‘hemangioblasts’ . To further investigate the ability of dnmt3bb . 1 to promote functional properties of HSPC in naive blastula cells , we examined whether cells from Tol2 ( hsp70:dnmt3bb . 1-gfp ) -injected donors would home to the thymus when transplanted into host larvae . We dissociated 5 hpf heat shocked WT , Tol2 ( hsp70:dnmt3bb . 1-gfp ) - or control Tol2 ( hsp70:gfp ) -injected blastula embryos and transplanted the dissociated cells into the circulation of 48 hpf Tg ( cmyb:gfp ) ( North et al . , 2007 ) or Tg ( lck:egfp ) transgenic ( Langenau et al . , 2004 ) hosts by intravenous injection ( Figure 6l ) . On day 5 , 17/30 ( 57% ) of the host animals that received cells from heat shocked Tol2 ( hsp70:dnmt3bb . 1-gfp ) donors showed donor cells ( red fluorescent , rhodamine/biotin-dextran-labeled ) homing to the thymus , as compared to only 11/38 ( 29% ) of host animals that received cells from control heat -shocked Tol2 ( hsp70:gfp ) donors ( Figure 6n–q ) . To be conservative , we decided to include counts of cell numbers in the thymus and heart/aortic arches only for the animals that showed transplanted cells in the thymus ( Figure 6q ) . The thymus-positive dnmt3bb . 1-gfp transplanted hosts had approximately twice as many cells in the thymus compared to the thymus-positive gfp control transplanted hosts , while the number of intravenously injected cells that were lodged non-specifically in the heart and aortic arches was comparable . 10 . 7554/eLife . 11813 . 015Figure 6 . Dnmt3bb . 1is sufficient for HSPC and downstream lineage gene expression . ( a ) Schematic diagram of the Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene used for ubiquitous heat shock-inducible expression of dnmt3bb . 1-gfp fusion protein . ( b ) Camera lucida drawing showing the experimental protocol employed for early embryonic induction of dnmt3bb . 1 . One cell-stage embryos are injected with Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene +/- cmyb morpholino ( MO ) , raised to 3 hpf ( mid-blastula stage ) , heat shocked , allowed to further develop to 6 hpf ( shield stage ) , then either fixed for whole mount in situ hybridization ( WISH ) or collected to prepare RNA for RT-qPCR . ( c–j ) Whole mount in situ hybridization of 6 hpf control Tol2 ( hsp70:gfp ) transgene-injected ( c-e ) , Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene-injected ( f-j ) , or Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene plus cmyb morpholino ( MO ) -injected ( i , j ) zebrafish embryos , probed for cmyb ( c , f ) , rag1 ( d , g , i ) , or l-plastin ( e , h , j ) . ( k ) Quantitative RT-PCR analysis of cmyb , l-plastin , and rag1transcript levels in positive control untreated 3 dpf ( for cmyb and l-plastin ) or 5 dpf ( for rag1 ) larvae ( blue columns ) , negative control shield stage ( 6 hpf ) embryos injected with a Tol2 ( hsp70:egfp ) transgene and heat shocked at 3 hpf ( red columns ) , or shield stage ( 6 hpf ) embryos injected with an Tol2 ( hsp70:dnmt3bb . 1-egfp ) transgene and heat shocked at 3 hpf ( green columns ) . ( l ) Schematic drawing showing the experimental protocol employed for early embryonic induction of dnmt3bb . 1-gfp and subsequent transplantation of cells into the circulation . One cell-stage embryos were injected with Tol2 ( hsp70:egfp ) or Tol2 ( hsp70:dnmt3bb . 1-egfp ) transgenes , raised to 3 hpf ( mid-blastula stage ) , heat shocked , and allowed to develop to 5 hpf , at which point the embryos were dissociated and cell suspensions was prepared . Dissociated cells were injected into the circulation of 48 hpf host larvae using a borosilicate needle without filament . ( m ) Camera lucida drawing of a 72 hpf zebrafish embryo with a green box noting the approximate regions containing the thymus . ( n–p ) Confocal images of a 5 dpf Tg ( lck:GFP ) cz1 host animal showing transplanted cells colonizing the thymus . Transmitted light image ( n ) , green fluorescent lck:gfp positive host thymus ( n , o ) , and red fluorescent rhodamine-dextran positive donor cells populating the thymus ( p ) . ( q ) Quantitation of the number of control GFP and dnmt3bb . 1-GFP expressing cells in the thymus and heart/aortic arch region . Images in panels c-f , r , s are dorsal views and panels n-p are lateral views . Scale bars = 150 µm in c-j , 25 µm in n , 10 µm in o . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 01510 . 7554/eLife . 11813 . 016Figure 6—figure supplement 1 . Dnmt3bb . 1induces hematopoietic but not endothelial gene expression in early embryos . ( a-l ) Whole mount in situ hybridization of 6 hpf control Tol2 ( hsp70:gfp ) ( a , c , e , g , i , k ) or Tol2 ( hsp70:dnmt3bb . 1-gfp ) ( b , d , f , h , j , l ) transgene-injected zebrafish embryos , probed for hbae1 ( a , b ) , ikaros ( c , d ) , lck ( e , f ) , rag2 ( g , h ) , cdh5 ( i , j ) and etsrp ( k , l ) All panels show animal pole views . Scale bar 150 µm in l . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 016 We carried out additional experiments to further examine whether cmyb is a proximal target of dnmt3bb . 1 . Bisulfite sequencing reveals that the cmyb intron 1 CpG island becomes methylated in DNA from heat -shocked mosaic Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene-injected shield stage animals , but not in DNA from heat shocked control Tol2 ( hsp70:gfp ) transgene-injected animals ( Figure 7a ) . A separate cmyb 5’ CpG island that is not normally methylated in HSPC in wild type animals does not become ectopically methylated in heat-shocked Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene-injected shield stage animals ( Figure 7b ) . Chromatin immunoprecipitation ( ChIP ) of DNA from heat-shocked Tol2 ( hsp70:gfp ) or Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene-injected or Tg ( hsp70:tcf-gfp ) germline transgenic ( Lewis et al . , 2004 ) shield-stage animals ( Figure 7c ) using anti-GFP antibodies shows that dnmt3bb . 1-GFP ( but not Tcf-GFP ) binds to two independent DNA sequences from the cmyb intron 1 CpG island ( Figure 7d , rows 1 and 2 ) . Dnmt3bb . 1-GFP does not bind to non-CpG island cmyb DNA from intron 1 or exon 5 , or to CpG island DNA from the runx1 or ntla gene loci ( Figure 7d , rows 3–6 ) . As an additional ChIP control , Tcf-GFP ( but not Dnmt3bb . 1-gfp ) shows specific association with sox3 DNA sequences containing known TCF binding sites ( Lee et al . , 2006 ) ( Figure 7d , row 7 ) , but does not associate with any dnmt3bb . 1 DNA sequence . ChIP using anti-methyl cytosine ( Figure 7e ) confirms methylation of the cmyb intron 1 CpG island in animals expressing dnmt3bb . 1-GFP but not GFP alone ( Figure 7e , rows 1 and 2 ) , as well as ‘constitutive’ methylation of the runx1 and ntla CpG islands ( Figure 7e , rows 5 and 6 ) . Together , these results suggest that Dnmt3bb . 1 methylates the cmyb intron 1 CpG island to maintain the expression of cmyb in HSPCs , and that loss of Dnmt3bb . 1 results in failure to maintain cmyb expression and loss of HSPCs . 10 . 7554/eLife . 11813 . 017Figure 7 . Dnmt3bb . 1induces methylation of cmyb exon 1 CpG island DNA in early embryos . ( a , b ) Bisulfite sequencing of genomic DNA from 6 hpf control Tol2 ( hsp70:gfp ) transgene- or Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene-injected embryos to assess methylation of the cmyb intron 1 CpG island ( a ) or cmyb 5’ CpG island ( b ) . Open circles represent unmethylated cytosine residues and filled circles represent methylated cytosine residues from CpG dinucleotides . Methylation of the intron 1 CpG island is detected in dnmt3bb . 1-gfp expressing embryos , but not in control gfp-expressing embryos . ( c–e ) Chromatin immunoprecipitation ( CHIP ) using anti-GFP or anti-methylcytosine antibodies to detect Dnmt3bb . a-GFP ( or TCF-GFP control ) fusion protein binding or cytosine methylation in CpG islands ( CGI ) , respectively . ( c ) Schematic diagram showing the experimental procedure used for CHIP analysis , ( d ) Anti-GFP CHIP with either no added chromatin ( negative control; column 1 ) or with chromatin from either hsp:dnmt3bb . 1-GFP injected embryos ( column 2 ) or from hsp:TCF-GFP injected embryos ( column 3; CHIP positive control ) . Column 4 is 10% of input chromatin from hsp:dnmt3bb . 1-GFP injected embryos . ( e ) Anti-methylcytosine CHIP with either no added chromatin ( negative control; column 1 ) or using chromatin from hsp:GFP injected embryos ( column 2; negative control for dnmt3bb . 1-gfp methylation ) or from hsp:dnmt3bb . 1-GFP -injected embryos ( column 3 ) . Column 4 is 10% of input chromatin from hsp:dnmt3bb . 1-GFP injected embryos . Primers used for the CHIP experiments in both e and f were: ( 1 ) cmyb Intron 1 CpG Island DNA , site #1 , ( 2 ) cmyb Intron 1 CpG Island DNA , site #2 , ( 3 ) cmyb Exon 5 DNA ( non-CpG island ) , ( 4 ) cmyb Intron 1 DNA ( non-CpG island ) , ( 5 ) runx1 Exon 3 CpG island DNA , ( 6 ) ntla CpG island DNA , and ( 7 ) sox3 upstream TCF binding site DNA ( non-CpG ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 017 Previous studies have shown that Cmyb function is necessary and sufficient for the development of definitive blood cell lineages and for the expression of downstream lineage genes ( Soza-Ried et al . , 2010; Zhang et al . , 2011; Mukouyama et al . , 1999 ) . In zebrafish , a small proportion of animals homozygous null for either cmyb or runx1 survive to adulthood . These ‘escapers’ show severe reduction in blood cell number in circulation and in the anterior kidney ( a blood-forming organ in adult zebrafish ) . In a similar fashion , we find that 10–12% of homozygous dnmt3bb . 1 mutants also survive to adulthood ( Figure 8a , b ) , and the anterior kidneys in these surviving mutants also exhibit severely reduced hematopoietic cellularity ( Figure 8c–f ) . The circulating erythrocytes that are present in adult mutants are mostly malformed ( Figure 8g–i ) . Interestingly , sorted HSPCs from dnmt3bb . 1-deficient 48 hpf embryos also display malformations including abnormally shaped , bi-lobed or even multiple distinct nuclei ( Figure 8j , k ) . Again , these results are similar to findings with cmyb mutants ( Soza-Ried et al . , 2010 ) , supporting the idea that Dnmt3bb . 1 exerts its effects on HSPC fate maintenance and hematopoietic gene expression via Cmyb . 10 . 7554/eLife . 11813 . 018Figure 8 . Hematopoieticdefects in adult dnmt3bb . 1y258 Mutants . ( a , b ) Wild type sibling ( a ) and dnmt3bb . 1y258homozygous mutant ( b ) adult zebrafish . ( c ) Representative low-magnification sagittal section through a dnmt3bb . 1y258 homozygous mutant adult zebrafish , showing the head kidney ( marked an arrow and surrounded by a dashed line ) , a major site of hematopoiesis in adult zebrafish . ( d , e ) Representative higher magnification images of sagittal sections through the head kidneys of wild type sibling ( d ) and dnmt3bb . 1y258 mutant ( e ) adult zebrafish , stained with hematoxylin/eosin . ( f ) quantitation of the percentage of the hematopoietic ( non-kidney tubule ) area that is acellular in head kidney sections from wild type sibling ( left column ) and dnmt3bb . 1y258 mutant ( right column ) adult zebrafish . ( g , h ) Giemsa -stained blood smears from wild type sibling ( g ) and dnmt3bb . 1y258 mutant ( h ) adult zebrafish . ( i ) quantitation of the adult erythroid cell circularity in WT and dnmt3bb . 1y258 mutant ( see Materials and methods ) , measured from blood smears such as those in panels g and h . ( j , k ) Giemsa stained FACS-sorted HSPC ( as in Figure 3a ) from 36 hpf dnmt3bb . 1 ( j ) or control ( k ) morpholino-injected zebrafish . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 01810 . 7554/eLife . 11813 . 019Figure 8—figure supplement 1 . Cmybinduces hematopoietic but not endothelial gene expression in early embryos . ( a ) Schematic diagram of the Tol2 ( hsp70:cmyb-2A-mCherry ) transgene used for ubiquitous heat shock-inducible expression of cmyb . ( b–q ) WISH of 6 hpf control Tol2 ( hsp70:mCherry ) ( b , d , f , h , j , l , n , p ) or Tol2 ( hsp70:cmyb-2A-mCherry ) ( c , e , g , i , k , m , o , q ) transgene-injected zebrafish embryos , probed for hbae1 ( b , c ) , rag1 ( d , e ) , l-plastin ( f , g ) , ikaros ( h , i ) , lck ( j , k ) , rag2 ( l , m ) , cdh5 ( n , o ) , and etsrp ( p , q ) All panels show animal pole views . Scale bar 150 µm in q . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 019 To examine whether ectopic Dnmt3bb . 1 and Cmyb expression in the early embryo results in similar induction of downstream hematopoietic gene expression , we drove cmyb expression in the blastula by the injection of a Tol2 ( hsp70:cmyb-2A-mCherry ) transgene ( Figure 8—figure supplement 1a ) . Like overexpression of dnmt3bb . 1-gfp in early embryos , heat shock induction of embryos injected with Tol2 ( hsp70:cmyb-2A-mCherry ) leads to ectopic expression of hbae1 , rag1 , l-plastin , ikaros , lck and rag2 ( Figure 8—figure supplement 1b–m ) , but not endothelial markers cdh5 or etsrp ( Figure 8—figure supplement 1n–q ) . To more directly examine whether cmyb is required downstream from dnmt3bb . 1 for induction of hematopoietic lineage genes , we co-injected the Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene together with a previously published cmyb morpholino ( Grabher et al . , 2011 ) and heat shocked as described above . We saw dramatic reduction in l-plastin and rag1 in dnmt3bb . 1-GFP over-expressing embryos co-injected with cmyb morpholino compared to embryos injected with Tol2 ( hsp70:dnmt3bb . 1-gfp ) transgene alone ( Figure 6g , h , i , j ) . Together , these results confirm that cmyb function is both necessary and sufficient for the expression of hematopoietic lineage genes , and further shows that the expression of hematopoietic genes cannot be induced by dnmt3bb . 1 in the absence of cmyb . This suggests that cmyb is a key locus regulated by dnmt3bb . 1 during hematopoietic differentiation .
In this manuscript , we show that an epigenetic regulatory factor promotes the maintenance of hemtopoietic cell fate downstream from an established genetic pathway required for the specification of HSPCs The de novo DNA methyltransferase dnmt3bb . 1 , the closest zebrafish ortholog of the mammalian DNA methyltransferase DNMT3B , is expressed specifically in cmyb-positive HSPC emerging from hemogenic endotheium in the ventral floor of the dorsal aorta . HSPC specification is controlled by a genetically programmed Notch-Runx1-Cmyb pathway ( Burns et al . , 2005; Gering and Patient , 2005; Bresciani et al . , 2014; Duncan et al . , 2005; Bertrand et al . , 2010b; Kumano et al . , 2003 ) . Runx1 mutants and runx1 morpholino-injected zebrafish display reduced expression of dnmt3bb . 1 , as do Notch-deficient mind bomb mutants . Since runx1 and mind bomb mutant embryos have strongly reduced numbers of HSPCs , reduced dnmt3bb . 1 expression may reflect reduced HSPC numbers rather than the loss of direct gene regulation . Conversely , overexpression of runx1 leads to increased expression of dnmt3bb . 1 in hemogenic endothelium . Since mouse Dnmt3b is also expressed by developing CD34+ HSPCs in the endothelium of early mouse embryos ( Watanabe et al . , 2004 ) , and is up-regulated by Oct4 in direct conversion of fibroblasts into multi-potent blood progenitors in cell culture ( Szabo et al . , 2010 ) , this suggests that the role of these orthologous DNMT proteins may be conserved during early hematopoietic development in vertebrates . Our loss- and gain-of-function data show that dnmt3bb . 1 is required for HSPC maintenance . In dnmt3bb . 1 mutants or dnmt3bb . 1 morpholino-injected animals HSPC are initially specified but subsequently undergo apoptosis with mutant embryos showing gradual reduction in cmyb+ HSPCs . Mutants and morphants also show later defects in downstream blood lineages , as evidenced by the reduced expression of specific markers of the myeloid , lymphoid , and erythroid lineages . These loss-of-function studies show that dnmt3bb . 1 is necessary for HSPC induction; additional gain-of–function experiments we performed show that dnmt3bb . 1 is also sufficient to drive hematopoietic development . Ectopic mosaic overexpression of Dnmt3bb . 1 activates hematopoietic gene expression in non-hemogenic endothelium , and even in ‘naive’ cells in the early blastula well before normal emergence of HSPC or specification of any blood cell lineages . Previous studies have shown that cd41:gfp transgene-positive HSPC isolated from transgenic zebrafish larvae populate the thymus upon transplantation ( Bertrand et al . , 2008 ) . Transplanted cells from dnmt3bb . 1-overexpressing blastulae also preferentally ‘home’ to the thymus and to the kidney ( data not shown ) , suggesting that these dnmt3bb . 1-overexpressing cells are recapitulating at least one functional property of HSPCs . The remarkable capacity of dnmt3bb . 1 to activate hematopoietic gene expression in non-hematopoietic cells appears to reflect a fairly direct conversion towards the hematopoietic lineage rather than a more general induction of mesoderm or hemogenic endothelium , since we do not observe ectopic expression of endothelial markers in dnmt3bb . 1-overexpressing blastulae , nor do we detect ectopic expression of other markers of mesoderm ( unpublished results ) . Another epigenetic factor , the histone demethylase Kdm2b , was recently shown to be able to promote the conversion of fibroblasts into iPS cells on its own ( Liang et al . , 2012 ) , reinforcing the idea that epigenetic factors can facilitate reprogramming independent of key specification factors ( such as Oct4 , Klf2 and Sox2 during iPS conversion ) . Further investigation into the capacity of dnmt3bb . 1 and its closest mammalian ortholog DNMT3B to promote the conversion of somatic cells into blood cells , and the potential usefulness of this for generation of blood cells from non-hematopoietic progenitors in vitro , will be of great interest . Our results strongly suggest that Dnmt3bb . 1 exerts its hematopoietic effects via Cmyb . Combined whole-genome analysis of gene expression and DNA methylation in HSPC isolated from double-transgenic zebrafish embryos by flow cytometry shows that cmyb is one of the most significant genes with both reduced expression and reduced DNA methylation in dnmt3bb . 1-deficient HSPC . Expression of number of hematopoietic genes is increased or decreased in dnmt3bb . 1-deficient HSPCs , but only a handful of these genes display significantly reduced methylation in dnmt3bb . 1-deficient animals and cmyb is one of the most highly significant of these . We do not detect significant changes in DNA methylation in runx1 , gata2 , scl , ikaros , lck , rag1 , l-plastin , or pu . 1 , suggesting that other HSPC and hematopoietic lineage genes are not direct targets of dnmt3bb . 1 . The importance of Cmyb downstream from Dnmt3bb . 1 was further confirmed by an unbiased DNA methylation survey of CpG islands associated with 319 genes annotated as ‘hematopoietic’ in a zebrafish gene expression database ( ZFIN ) . Only 10 of these genes have methylated gene body CpG islands in blood cells , and of these ten only the cmyb gene shows reduced methylation ( of its intron 1 CpG island ) in dnmt3bb . 1-deficient animals . We were able to validate that the intron 1 CpG island is indeed a bona fide target of dnmt3bb . 1 binding and DNA methylation using chromatin immunoprecipitation ( ChIP ) . ChIP with an anti-GFP antibody shows dnmt3bb . 1-GFP binds specifically to the cmyb intron 1 CpG island , and an anti-Methyl-C antibody confirms specific methylation of this CpG island by dnmt3bb . 1-GFP . The mechanisms targeting DNA methyltransferases to some but not other CpG islands is still poorly understood . Class 3 de novo DNMTs are known to interact with a number of different transcription factors in vitro ( Brenner et al . , 2005; Fuks et al . , 2001; Suzuki et al . , 2006 ) , and these interactions are thought to influence their genomic targeting specificity ( Jurkowska et al . , 2011 ) . However , these interactions and their functional consequences on gene expression have not been validated in vivo ( Hervouet et al . , 2009 ) . Besides transcription factors , DNMTs are also known to interact with other epigenetic regulators , mainly histone- modifying enzymes . In zebrafish , dnmt3bb . 2 ( previously known as Dnmt3 ) has been shown to interact with H3K9 methyltransferase G9a and regulate lef1 expression ( Rai et al . , 2010 ) . It is possible that a variety of genetic and epigenetic factors may cooperate with dnmt3bb . 1 to facilitate its interaction with the cmyb locus . Our functional data also strongly implicate cmyb as a key target regulated by dnmt3bb . 1 during hematopoietic development . Loss of Dnmt3bb . 1 results in phenotypes that closely mirror those resulting from loss of cmyb . Like previously reported zebrafish cmyb mutants ( Soza-Ried et al . , 2010 ) , a small proportion ( 10–15% ) of dnmt3bb . 1 mutants survive past 6–9 weeks . Although these dnmt3bb . 1 mutant ‘escapers’ appear externally normal , they have hypoplasia of the anterior kidney marrow and abnormal red blood cell morphology – again , phenotypes extremely similar to those seen in surviving adult cmyb mutants ( Soza-Ried et al . , 2010 ) . In addition , overexpression of cmyb in the early blastula results in induction of hematopoietic gene expression virtually indistinguishable from that seen upon overexpression of dnmt3bb . 1 . Finally , dnmt3bb . 1-induced expression of hematopoietic lineage genes in blastulae can be prevented by knock-down of cmyb , showing that cmyb is necessary for dnmt3bb . 1-mediated hematopoietic gene expression . Taken together , our results strongly suggest that dnmt3bb . 1 promotes continued hematopoietic development by methylating cmyb gene body DNA and maintaining activation of the cmyb locus . This provides an explanation for how transient expression of the HSPC-specific Runx1 transcription factor , which is required for HSPC specification during early development , can lead to the continued expression of cmyb required for prolonged maintenance of hematopoietic cell fate ( Figure 9 ) . Runx1 is required to activate expression of cmyb in HSPC , and it can do so in the absence of Dnmt3bb . 1 . However , Runx1 also activates the expression of Dnmt3bb . 1 , which ‘marks’ cmyb gene body DNA to promote cmyb activation . As Runx1 expresion is down-regulated , these cmyb gene body DNA marks help to maintain expression of cmyb in the absence of Runx1 . 10 . 7554/eLife . 11813 . 020Figure 9 . Modelfor regulation of HSPC cell fate during development . ( a ) Notch-Runx1 signaling controls specification of HSPCs in the ventral wall of the dorsal aorta during early embryogenesis . This signaling pathway initiates HSPC expression of both the key transcription factor Cmyb and the epigenetic regulator Dnmt3bb . 1 . ( b ) As development proceeds Runx1 expression is down-regulated and continued maintenance of active Cmyb expression in HSPC depends on Dnmt3bb . 1–mediated Cmyb gene body DNA methylation , ensuring maintenance of HSPC cell fate during the Runx1-independent phase . DOI: http://dx . doi . org/10 . 7554/eLife . 11813 . 020 There are previous precedents for epigenetic regulation targeting one or a small subset of genes to facilitate or reinforce a specific developmental cell fate decision . In avian embryos , expression of the related DNMT3A gene along the neural plate boundary promotes specification of neural crest by methylating CpG islands in the Sox2 and Sox3 promoter regions and repressing expression of these neural specification genes , acting as a ‘molecular switch’ promoting neural crest cell fate ( Hu et al . , 2012 ) . In the case of zebrafish dnmt3bb . 1 , HSPC maintenance appears to be promoted by gene body , not promoter , methylation of cmyb , and cmyb expression is activated by DNA methylation , not repressed . Gene body methylation has been reported in many different organisms , and it is often associated with active gene expression ( Tweedie et al . , 1997; Zemach et al . , 2010 ) , which is thought to support by promoting proper RNA splicing and by helping to maintain chromatin structure for active gene expression ( Suzuki and Bird , 2008; Jones , 2012 ) . Interestingly , a recent study showed that DNMT3B , the mammalian ortholog of zebrafish dnmt3bb . 1 , preferentially binds to and methylates gene body DNA ( but not promoter or enhancer DNA ) of actively transcribed genes in mouse embryonic stem cells ( Baubec et al . , 2015 ) , reinforcing the idea that this particular DNA methyltransferase preferentially promotes gene body DNA methylation and gene activation . Previous studies have implicated DNA methyltransferases in adult hematopoiesis , with some studies reporting that DNA methylation promotes HSPC renewal and/or expansion ( Challen et al . , 2012; Tadokoro et al . , 2007 ) and others reporting that it promotes lineage differentiation ( Broske et al . , 2009; Trowbridge et al . , 2009 ) . A recent zebrafish study reported that the maintenance DNA methyltransferase 1 facilitates HSPC formation via C/ebpa ( Liu et al . , 2015 ) . DNA methylation is also strongly implicated in hematopoietic malignancies; indeed , inhibitors of DNA methylation are currently used therapeutically in this context ( Flotho et al . , 2009 ) . Our results highlight an important new role for DNA methyltransferases during the earliest stages of HSPC emergence in the developing embryo . The molecular mechanisms involved in generation and maintenance of HSPC from hemogenic endothelium are still not well understood , and our findings provide evidence that DNA methylation-mediated epigenetic modifications downstream from Runx1 help maintain HSPC fate during definitive hematopoiesis . DNMT3B , the closest mouse ortholog of zebrafish dnmt3bb . 1 , is also expressed in developing HSPC during embryogenesis , suggesting the developmental roles of these proteins may be conserved ( Watanabe et al . , 2004 ) . Another recent report suggests that histone and chromatin modifications may also help maintain HSPC cell fate during mouse embryogenesis ( Tober et al . , 2013 ) . Together , these findings suggest that a number of different epigenetic processes may contribute to generating HSPC cell fate during embryogenesis . It will be interesting to further examine whether and how epigenetic regulators interact with various hematopoeitic transcription factors regulating HSPC and hematopoietic lineage-specific development .
Zebrafish lines used in this study are Tg ( cmyb:GFP ) zf169 ( North et al . , 2007 ) , Tg ( fli1:eGFP ) y1 ( Lawson and Weinstein , 2002 ) , mibta52b ( Itoh et al . , 2003 ) , runx1w84X ( Sood et al . , 2010 ) , Tg ( lck:GFP ) cz1 ( Langenau et al . , 2004 ) , Tg ( kdrl:mApple ) y278 , dnmt3bb . 1y258 ( an allele carrying premature stop codon at amino acid 172 ) , dnmt3bb . 1b1237 ( an allele carrying premature stop codon at amino acid 42 ) , and the EK wild type line . Generation and genotyping of dnmt3bb . 1y258 and dnmt3bb . 1b1237 are described below . Zebrafish were mated , staged ( Kimmel et al . , 1995 ) and raised ( Westerfield , 2000 ) as previously described . A pair of TALEN constructs targeting Exon 6 of dnmt3bb . 1 were designed using previously published methods ( Dahlem et al . , 2012 ) . 100 pg of in vitro synthesized RNAs for each arm were injected at one cell stage to induce mutations at the target site . F1 lines were generated from F0 injected founders showing high efficiency of somatic mutations . Carriers were identified by PCR genotyping and subsequently sequenced to identify the mutations . An allele with an 11-nucleotide deletion resulting in a premature stop codon was identified ( dnmt3bb . 1y258 ) and used for further analysis . For genotyping , PCR primers were designed spanning the TALEN target site and the resulting amplicons were subjected to SacI restriction enzyme digest . The SacI site is destroyed in the mutant allele due to the 11-nucleotide deletion . Genotyping of in situ stained embryos were done using the same PCR primers and conditions . For PCR , template genomic DNA was isolated using standard DNA extraction buffer ( http://zfin . org/zf_info/zfbook/chapt9/9 . 3 . html ) . A full length dnmt3bb . 1 zebrafish clone was obtained from Open Biosystems . Tol2 ( hsp:mCherry ) , Tol2 ( hsp:cmyb-2A-mCherry ) , Tol2 ( hsp:gfp ) , Tol2 ( hsp:dnmt3bb . 1-gfp ) , and ISceI ( kdrl:dnmt3bb . 1-gfp ) transgene constructs were generated using Gateway technology ( Kwan et al . , 2007 ) . Dnmt3bb . 1 and cmyb coding sequences were amplified using sequence specific primers with attB sites . The resulting PCR products were cloned into a pDONR vector ( Thermo Fisher Scientific , Waltham , MA ) to generate middle entry vectors using BP clonase enzyme ( Thermo Fisher Scientific , Waltham , MA ) . Correct plasmids determined by sequencing were used to generate final destination clones using LR clonase enzyme ( Thermo Fisher Scientific ) . Antisense riboprobes were generated using Roche DIG and FITC labeling mix . Full length or EST clones were restriction digested and transcribed using appropriate enzymes . Synthetically capped runx1 mRNA was synthesized using mMessage mMachine kit ( Thermo Fisher Scientific ) and injected as previously published ( Burns et al . , 2005 ) . mRNA , DNA constructs and morpholinos were injected at onecell stage embryos using a pressure injector as described previously ( Gore et al . , 2011 ) . All morpholino sequences used in this study are listed in methods supplementary file 1A . In situ hybridization and immunohistochemistry were carried out as described ( Gore et al . , 2011 ) . For double WISH , we followed modifications as described previously ( Lauter et al . , 2011 ) . For detecting cell death , primary polyclonal anti-active Caspase 3 ( C4748 , Sigma-Aldrich , St . Louis , MO ) and monoclonal anti-GFP antibodies were used at 1:500 and 1:200 dilutions , respectively . Primary antibodies were detected using anti-rabbit Alexa-543 and anti-mouse Alexa-488 labeled secondary antibodies at 1:400 dilution . After staining , embryos were cleared in Vectashield mounting media ( Vector Laboratories , Burlingame , CA ) and imaged using a Leica TCS-SP5 II confocal microscope . Cell sorting was performed as described previously ( Bertrand et al . , 2010a ) . Briefly , double transgenic embryos derived from carriers of Tg ( cmyb:GFP ) ;Tg ( kdrl:mApple ) were injected with control or dnmt3bb . 1 MO and raised to 48 hpf . Embryos were dissociated using Ttrypsin-EDTA and passed through a 70 µM strainer to collect a single cell suspension . GFP and mApple double-positive HSPCs were isolated by forward scatter using a FACS ARIA ( Becton Dickinson , Franklin Lakes , NJ ) . Sorted HSPC were pelleted and DNA and RNA were co-isolated using a Zymo Duet DNA/RNA miniprep kit ( Zymo Research , Irvine , CA ) . DNA and RNA were used for bisulfite sequencing ( RRBS; see below ) and RNA seq analysis respectively . RRBS libraries were prepared as described previously ( Gu et al . , 2011 ) . 62 ng of Genomic DNA was digested with MspI and TaqαI overnight . Digested DNA was purified and end -repaired and bar-coded methylated adapters were ligated . DNA was size-selected using AMPure beads between 100–350 bp . The resulting DNA was bisulfite treated and PCR amplified to generate final libraries . Library quality was verified using HPLC and a bioanalyzer . We generated two separate sets of libraries and sequenced them on Illumina and SOLiD platforms . Raw single-end sequence data was trimmed for quality and adapter sequence using Trimmomatic software , trimming leading or trailing bases at quality < 5 as well as using a sliding window requiring a 4 bp average of quality > 15 . Reads trimmed below 50 bp were discarded . Following trimming , reads were aligned using Bismark software against a bisulfite converted Zv9 reference genome using non-directional alignment , allowing for up to 2 bp mismatch in the seed region . Prepared Bismark reports were parsed for regions in which coverage was ≥ 10x in both experimental conditions ensuring a strong comparison could be drawn . Using defined gene coordinates , regions overlapping gene bodies were combined and conversions as well as non-conversions were summed to provide an overall bisulfite conversion rate per-gene . The ratio of conversions was then tested for change between conditions using a two-proportion z-test testing the hypothesis that the proportion of bisulfite conversion had altered between conditions . Reads from two independent sequencing rounds were pooled and aligned to the zebrafish genome . Two independent sequencing libraries were generated using SMARTer and TruSeq RNA library preparation kit . Paired-end reads were trimmed using trimmomatic and aligned to zebrafish genome using TOPHAT , differential expression analysis was performed using Cufflinks with FPKM . The PARTEK genomic suite was used to calculate differentially expressed genes by ANOVA between two control and dnmt3bb . 1 deficient HSPC RNA . Absolute fold change in DNA methylation and RNA expression are shown in ( Figure 3—source data 1 ) Ingenuity pathway alysis was carried out onRNA seqdata to identify major signaling pathways and networks affected in differentially expressed genes . For Dnmt3bb . 1 targets , common genes with reduction in DNA methylation from RRBS and RNA seq analysis were tabulated . The RNA seq data were submitted to GEO under accession number GSE74929 . Genes expressed by blood cells were identified in the ZFIN zebrafish gene expression database ( http://zfin . org/cgi-bin/webdriver ? MIval=aa-ZDB_home . apg ) using search terms blood , HSCs , ventral wall of dorsal aorta , and all associated search terms . 319 genes were identified whose expression in blood had been confirmed by in situ staining or PCR . Each of these 319 gene loci were then further analyzed using the UCSC genome browser or WashU epigenetics browser to determine whether there were CpG islands in their gene bodies . The gene body was defined as 5’UTR , all exons and introns and 3’ UTR . 57 genes with 72 CpG islands in their gene bodies were identified . To screen for the methylation of these 72 CpG islands , we designed bisulfite PCR primers using MethPrimer ( Li and Dahiya , 2002 ) : http://www . urogene . org/cgi-bin/methprimer/methprimer . cgi . To predict methylation of CpG islands by Rrestriction enzyme digestion patterns , we developed a program ( http://bisulfite . nichd . nih . gov ) to computationally convert all Cs to Ts and predict changes in digestion patterns using specific restriction enzymes . Using this information , we subjected PCR amplified CpG islands to digestion with the appropriate restriction enzymes and the resulting products were resolved on 2 . 5% Metaphor agarose . Ten CpG islands showed evidence of methylation . These ten CpG islands were subjected to bisulfite sequencing ( see below ) using genomic DNA prepared from control and dnmt3bb . 1 morpholino-injected animals to determine whether the methylation of any of these CpG islands was dependent on runx1 function . All the PCR primers used in this analysis are given in the methods Supplementary file 1C . Genomic DNA was isolated from FACS sorted HSPCs or from circulating cells isolated after tail clip at day 5–7 . Cells were collected in 1X PBS and genomic DNA was isolated using cell lysis buffer ( 10 mM Tris-HCl pH8 , 5 mM EDTA , 1% SDS and 20 μg/ml Proteinase K ) followed by phenol:chloroform extraction and ethanol precipitation . 1 μg of DNA was bisulfite converted using EZ DNA methylation-Lightning kit from Zymo research , following manufacturers instructions . Bisulfite converted DNA was PCR amplified using primers designed by MethPrimer ( Li and Dahiya , 2002 ) : http://www . urogene . org/cgi-bin/methprimer/methprimer . cgi and are listed in the methods Supplementary file 1B . One Taq Hotstart 2X master mix standard buffer ( New England Biolabs , Ipswich , MA ) or ZymoTaq DNA polymerase ( Zymo Research ) enzymes were used for amplifying bisulfite converted DNA . PCR amplicons were purified and cloned using pCRII-TOPO TA cloning kit . Mini-prepped colonies were sequenced using T7 or SP6 sequencing primers . Sequencing results were analyzed using QUMA ( Kumaki et al . , 2008 ) : http://quma . cdb . riken . jp/ . Total cellular RNA was isolated from 20–30 embryos per treatment using Trizol reagent or by Zymo Duet DNA/RNA kit . RNA was DNaseI treated to remove trace amounts of genomic DNA and phenol:chloroform extracted . Equal amounts of RNA were converted into cDNA using ThermoScript RT-PCR system ( Thermo Fisher Scientific ) . The resulting cDNA was used for either RT-PCR or qRT-PCR using Sso Advanced SYBRGreen PCR mix ( BioRad , Hercules , CA ) on a CFX96 Real Time PCR system . Primers used in qRT-PCR are listed in the Supplementary file 1B . Tol2 ( hsp:gfp ) or Tol2 ( hsp:dnmt3bb . 1-gfp ) constructs were injected at the one -cell stage and induced at 3 hpf by a 37°C heat shock for 30 min . Embryos were harvested at 5 hpf for CHIP analysis . CHIP was performed using anti-GFP ( Developmental studies hybridoma bank: DSHB-GFP-12A6 ( Sanchez et al . , 2014 ) and Thermo Fisher Scientific: A11122 ) and anti-Methyl Cytosine ( Zyme Research: A3001 and Calbiochem , San Diego , CA; Cat no: NA81 ) antibodies ( Nair et al . , 2011 ) using the EAZY CHIP chromatin immunoprecipitation kit ( Millipore , Temecula , CA; Cat no 17–371 ) as described by Ro and Dawid ( Ro and Dawid , 2011 ) . Immunoprecipitated DNA was eluted using chelex beads and used in PCR analysis . PCR primers used in this study are described in Supplementary file 1B . Cell transplants were carried as described previously , with minor modifications ( Traver et al . , 2003 ) . Donor embryos were injected at onecell stage with Tol2 ( hsp70:gfp ) or Tol2 ( hsp70:dnmt3bb . 1-gfp ) plasmids with rhodamine- and biotin-dextran as lineage tracers . At 3 hpf , embryos were heat shocked to induce transgene expression at 37°C for 30 min . Embryos were allowed to recover at 28°C until 5 hpf . At 5 hpf , manually dechorionated embryos were lysed in yolk dissociation buffer ( 55 mM NaCl , 1 . 8 mM KCl , 1 . 25 mM NaHCO3 ) and spun at 300 g for 1 min at room temperature . The cell pellet was washed twice using 1X PBS without calcium and agnesium and re-suspended in the same buffer with 5% FBS . Borosilicate needles without filaments were back-filled using cell suspension and pressure injected in the cardinal vein of a 48 hpf host embryo . Roughly , 20–25 cells were transplanted in each host embryo . Transplanted host embryos were allowed to grow until 5 dpf , then fixed and stained for biotin using Streptavidin Alexa Fluor 546 ( S-11225 ) or Alexa Fluor 488 ( S-11223 ) from Thermo Fisher Scientific . Embryos were imaged using an Olympus FV1000 confocal and labeled cells were counted in the thymus and in the aortic arch/heart region . We performed three independent rounds of transplants in EK WT and lck-GFP transgenic backgrounds . In total , 38 and 30 embryos were analyzed for gfp alone and dnmt3bb . 1-gfp transplanted cells , respectively . Thymus localizing donor cells and heart and aortic arch localizing donor cells were counted in the 11/38 gfp alone and 17/30 dnmt3bb . 1-gfp transplanted host embryos . Adult zebrafish sagittal sections , H&E staining , and blood smears were prepared as described earlier ( Traver et al . , 2003; 2004; Sood et al . , 2010 ) . | The cells in our blood are constantly being replaced with new cells that are produced by stem cells called hematopoietic stem and progenitor cells ( or HSPCs for short ) . The HSPCs form early on in the development of the embryo and continue in the same role throughout the life of the animal . A gene called runx1 is required for HSPCs to form , but is not required for these cells to maintain their role ( cell identity ) in the long term . In mice , this gene is only expressed for a brief period of time as the HSPCs form , and is switched off in the mature stem cells . Another gene called cmyb – which is switched on by runx1 – is also required for HSPCs to form . However , unlike runx1 , cmyb continues to be expressed in mature HSPCs and is required to maintain HSPC identity . It is not known how the temporary activation of runx1 causes the long-term expression of cmyb . One possible explanation is that the cmyb gene may be subject to a process called DNA methylation . This process is carried out by enzymes called DNA methyltransferases and can have long-term effects on the expression of genes by modifying the structure of the DNA that encodes them . Here , Gore et al . investigate the role of a particular DNA methyltransferase in the formation of HSPCs in zebrafish embryos . The experiments show that this enzyme is activated in developing HSPCs in response to an increase in runx1 expression . The loss of this enzyme’s activity reduces both the amount that cmyb is methylated and its level of expression , which results in a gradual decline in the number of HSPCs in zebrafish . Further experiments show that if the DNA methyltransferase is artificially activated in cells that don’t normally form blood cells , these cells change their identity to do so . This switch is accompanied by methylation of cmyb and an increase in its expression . Gore et al . ’s findings reveal that the temporary activation of runx1 triggers the production of an enzyme that methylates cmyb to maintain the identity of HSPCs . Future studies should help to reveal exactly how runx1 promotes DNA methylation , and whether this process can be harnessed to promote HSPC formation for research or medical treatments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2016 | Epigenetic regulation of hematopoiesis by DNA methylation |
Plasma membrane lipid composition must be maintained during growth and under environmental insult . In yeast , signaling mediated by TOR Complex 2 ( TORC2 ) -dependent protein kinase Ypk1 controls lipid abundance and distribution in response to membrane stress . Ypk1 , among other actions , alleviates negative regulation of L-serine:palmitoyl-CoA acyltransferase , upregulating production of long-chain base precursors to sphingolipids . To explore other roles for TORC2-Ypk1 signaling in membrane homeostasis , we devised a three-tiered genome-wide screen to identify additional Ypk1 substrates , which pinpointed both catalytic subunits of the ceramide synthase complex . Ypk1-dependent phosphorylation of both proteins increased upon either sphingolipid depletion or heat shock and was important for cell survival . Sphingolipidomics , other biochemical measurements and genetic analysis demonstrated that these modifications of ceramide synthase increased its specific activity and stimulated channeling of long-chain base precursors into sphingolipid end-products . Control at this branch point also prevents accumulation of intermediates that could compromise cell growth by stimulating autophagy .
A eukaryotic plasma membrane ( PM ) is a complex structure composed of many protein ( Sachs and Engelman , 2006 ) and lipid ( Simons and Sampaio , 2011 ) species arranged with a high degree of compositional and spatial organization . The levels and distributions of lipids are clearly important for PM processes—from solute transport ( Divito and Amara , 2009 ) , to endocytosis ( Platta and Stenmark , 2011 ) , to receptor function and signal transduction ( Groves and Kuriyan , 2010 ) . It is essential , therefore , for a cell to maintain proper lipid balance . How cells sense alterations in PM organization in response to developmental cues or environmental stresses and adjust the rates of the reactions that maintain lipid homeostasis are vital mechanisms to understand . We ( Roelants et al . , 2011 ) and others ( Berchtold et al . , 2012; Niles and Powers , 2012 ) have shown that dynamic changes in the function of Target of Rapamycin ( TOR ) Complex 2 ( TORC2 ) and in a downstream protein kinase that it stimulates , Ypk1 ( and its paralog Ypk2 ) , are important for cell survival in response to membrane stress . Ypk1 ( and Ypk2 ) are two members of the AGC kinase family ( Pearce et al . , 2010 ) and orthologs of mammalian SGK1 ( Casamayor et al . , 1999 ) . In Saccharomyces cerevisiae , Ypk1 ( and Ypk2 ) are activated at a basal level by phosphorylation of a Thr in the activation loop by the eisosome-associated protein kinase Pkh1 ( and its paralog Pkh2 ) ( Casamayor et al . , 1999; Roelants et al . , 2002 ) . The ortholog responsible for this reaction in mammalian cells is PDK1 ( Mora et al . , 2004 ) . However , this level of activity is not sufficient for Ypk1 to permit survival when cells are challenged with certain membrane perturbants ( Roelants et al . , 2004 ) . Under such conditions of membrane stress , TORC2 stimulates Ypk1 and Ypk2 function by phosphorylating at least two C-terminal sites ( Kamada et al . , 2005; Roelants et al . , 2011; Niles et al . , 2012 ) . Constitutively-active Ypk1 and Ypk2 alleles bypass the need for functional TORC2 , indicating that Ypk1 and Ypk2 are solely responsible for executing all the essential downstream functions of TORC2 ( Kamada et al . , 2005; Roelants et al . , 2011; Niles et al . , 2012 ) . How membrane stress is communicated to TORC2 is a question of on-going research . It has been reported that membrane stress caused by inhibition of sphingolipid synthesis or membrane stretch ( induced by hypotonic shock ) causes two PH domain-containing proteins ( Slm1 and Slm2 ) to relocalize from eisosomes to a separate PM region that contains TORC2 and leads to increased TORC2 activation of Ypk1 and Ypk2 ( Berchtold et al . , 2012 ) , purportedly because Slm1 and Slm2 recruit Ypk1 and Ypk2 to TORC2 ( Niles et al . , 2012 ) . However , other evidence indicates that Avo1 ( ortholog in other organisms is Sin1 ) is the subunit of the TORC2 complex primarily responsible for substrate recognition of Ypk1 and its orthologs , including association of S . cerevisiae Ypk2 with Avo1 ( Liao and Chen , 2012 ) , Gad8 with Sin1 in fission yeast ( Ikeda et al . , 2008; Kataoka et al . , 2014 ) , and SGK1 with mSin1 in mammalian cells ( Jacinto et al . , 2006; Yang et al . , 2006; Lu et al . , 2011; Liu et al . , 2013 ) . Regardless of the actual mechanism by which the activity of the TORC2-Ypk1 signaling module is affected by these assaults on the PM , it clearly sets in motion processes that allow the cells to cope appropriately with these stresses . Several targets of Ypk1 have already been identified that shed light on how TORC2-Ypk1 signaling reprograms cellular processes to cope with PM stress . Ypk1 phosphorylates and negatively regulates Fpk1 , a protein kinase responsible for activation of PM-localized aminophospholipid flippases ( Roelants et al . , 2010 ) , thereby fine-tuning the phosphatidylethanolamine content of the inner leaflet of the PM bilayer . Ypk1 also phosphorylates and alleviates the inhibitory function of two endoplasmic reticulum ( ER ) -localized proteins ( Orm1 and Orm2 ) that impede the function of the first enzyme unique to sphingolipid biosynthesis , L-serine:palmitoyl-CoA acyltransferase ( SPT ) ( Roelants et al . , 2011; Berchtold et al . , 2012; Sun et al . , 2012 ) , thereby increasing the rate of formation of the long-chain base precursor ( phytosphingosine ) to yeast sphingolipids . Ypk1 also phosphorylates and inhibits one of the two isoforms ( Gpd1 ) of glycerol-3-phosphate dehydrogenase , a primary source of glycerol-3P for production of glycerophospholipids ( Lee et al . , 2012b ) . In response to hyperosmotic shock , and in contrast to other PM stressors , TORC2 activity is markedly decreased , preventing Ypk1-mediated inhibition of Gpd1 [which also gets upregulated transcriptionally ( Ansell et al . , 1997 ) ] , thereby greatly increasing synthesis of glycerol-3P , which is dephosphorylated to produce glycerol , an innocuous osmolyte that yeast cells accumulate as a means to counteract the increase in external osmotic pressure ( Lee et al . , 2012b ) . Control of all these reactions already made it clear that TORC2-Ypk1 signaling is a central regulator of PM lipid homeostasis . To gain further insight into how the TORC2-Ypk1 signaling axis contributes to PM homeostasis and potentially other cellular processes , we devised a three-step procedure to screen in an unbiased and genome-wide manner for additional candidate Ypk1 substrates whose physiological relevance we could then evaluate . As described here , we first used bioinformatics to search the yeast proteome for presumptive targets , then applied a genetic method to narrow down the hits to likely , functionally important substrates , and then used biochemical analysis both in vitro and in vivo to validate the best prospects . Although we report here the general outcomes of this screen , we focus mainly on our discovery and demonstration that both of the catalytic subunits of the ER-localized ceramide synthase complex ( Schorling et al . , 2001; Vallée and Riezman , 2005 ) are bona fide targets of Ypk1 . These findings provide new insight into how TORC2-initiated signaling regulates flux through the sphingolipid pathway and how this specific control mechanism is important for survival when sphingolipid synthesis is compromised . In addition , we show that this regulation of the ceramide synthase complex is important for preventing accumulation of pathway intermediates that would otherwise compromise cell growth by triggering an inappropriate autophagic response .
We devised a three-step strategy ( Figure 1A ) to pinpoint bona fide cellular targets of Ypk1 , utilizing bioinformatics to predict potential Ypk1 substrates , then an in vivo genetic test involving a novel variation of the synthetic dosage lethality method to winnow the list to likely candidates , and finally biochemical analysis in vitro to confirm whether the identified gene product serves as a direct substrate of Ypk1 . The physiological relevance of Ypk1-dependent modification of each protein on the resulting final list could then been evaluated . 10 . 7554/eLife . 03779 . 003Figure 1 . A three-part screen to identify likely Ypk1 substrates . ( A ) The three-part screening strategy to identify Ypk1 substrates is shown schematically as a flow chart . Numbers indicate the number of hits/considered genes at each step in the screen . ( B ) The bioinformatic approach towards identifying Ypk1 substrates is schematized as a flowchart with each filter as a box . Genes were first filtered by MOTIPS on the basis of having likely phosphorylatable Ypk1 motifs . Subsequently , substrates were filtered by having many Ypk1 motifs or having a Ypk1 site known to be phosphorylated in published data sets . Lastly , genes were filtered by requiring the gene to have a published chemical sensitivity like Ypk1 does , or a published interaction with Ypk1 , Ypk1 regulators ( TORC2 or PP2A ) or sphingolipid biosynthetic machinery . ( C ) A possible explanation for Ypk1 synthetic dosage lethality interactions is shown . Normally , the cell has enough kinase activity to buffer overexpression of a substrate ( Substrate 2 ) , so that essential substrates are regulated and normal growth is unperturbed . However , concurrent decrease in kinase activity coupled with substrate overexpression causes loss of regulation of essential substrate ( s ) ( Substrate 1 ) leading to observable growth defects . ( D ) ypk1-as ypk2Δ ( yAM135–A ) cells were transformed with PGAL1-GFP ( negative control ) , PGAL1-Orm1 or PGAL1-Orm2 ( known Ypk1 substrates , positive SDL controls ) plasmids . Overnight cultures were then serially diluted onto either dextrose ( to repress substrate overexpression ) or galatose ( to induce substrate overexpression ) containing media with increasing concentrations of the Ypk1-as inhibitor 3-MB-PP1 . ( E ) GST-Orm1 ( 1–85 ) ( pFR203 ) and GST-Fps1 ( 1–255 ) ( pBT6 ) were purified from E . coli and incubated with [γ-32P]ATP and Ypk1-as , purified from S . cerevisiae , in the absence or presence of 3-MB-PP1 . The products were then resolved by SDS/PAGE and analyzed as described in ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 03779 . 003 For initial bioinformatic search of the yeast proteome , we developed a position-weighted consensus sequence logo ( Figure 1B ) for the preferred Ypk1 phospho-acceptor site based on two primary criteria: ( a ) the known Ypk1 sites in five , validated in vivo targets ( Fpk1 , Fpk2 , Orm1 , Orm2 , and Gpd1 ) ( Roelants et al . , 2010 , 2011; Lee et al . , 2012b; Niles et al . , 2012; Sun et al . , 2012 ) ; and , ( b ) the sequence preference displayed by Ypk1 for phosphorylation of synthetic peptides in vitro ( Casamayor et al . , 1999; Mok et al . , 2010 ) . All demonstrated substrates either in vivo or in vitro contain Arg at positions −5 and −3 with respect to the phosphorylated Ser ( or Thr ) ; thus , these positions were invariant in the search motif . Given that nearly all the verified sites within known in vivo targets possess a hydrophobic residue ( V , I , F ) at position +1 , the search motif gave preference to sites with a hydrophobic residue at the +1 position . We then took advantage of the existing MOTIPS motif analysis package ( Lam et al . , 2010 ) to identify those S . cerevisiae gene products that contain occurrences of the search logo . Several authentic Ypk1 substrates ( e . g . , Fpk1 , Orm1 , and Orm2 ) contain multiple Ypk1 phosphorylation sites . Thus , we filtered our search further by prioritizing candidates containing multiple matches to the search logo , as predicted by MOTIPS . However , to avoid disregarding potential Ypk1 substrates with a single predicted match to the sequence logo , we also considered MOTIPS hits wherein there was existing evidence in the PhosphoGRID database ( Sadowski et al . , 2013 ) indicating that a predicted site is phosphorylated in vivo . To narrow down the list of potential substrates further , we chose to pursue those gene products containing matches to the sequence logo for which there was existing information in the literature suggesting a phenotypic relationship to known Ypk1-dependent processes: ( a ) a loss-of-function mutation in the candidate gene exhibits elevated sensitivity to agents ( aureobasidin A , caspofungin , and/or myriocin ) toward which a ypk1Δ mutant is also sensitive ( Hillenmeyer et al . , 2008 ) ; ( b ) the candidate gene product is reported to be involved in a genetic or biochemical interaction with Ypk1 , as curated in YeastMine ( Balakrishnan et al . , 2012 ) ; ( c ) the candidate gene product is connected in some way to known Ypk1 regulators ( e . g . , TORC2 , PP2A ) ; and/or , ( d ) the candidate gene product is involved in a known Ypk1-regulated process ( e . g . , sphingolipid metabolism ) ( for further details , see ‘Materials and methods’ ) . Reassuringly , our approach identified three known Ypk1 substrates ( Fpk1 , Orm1 and Orm2 ) ; absence of Fpk2 and Gpd1 from the list generated solely by the MOTIPS search criteria arose from the fact that these substrates contain only a single predicted site that is not presently recorded in PhosphoGRID . For this reason , we also restored for consideration additional gene products that contain a single match to the consensus sequence logo that YeastMine indicated are involved in processes in which Ypk1 is implicated . The resulting candidates , grouped via cellular process on the basis of current GO Slim terminology ( http://www . geneontology . org/GO . slims . shtml ) , are cataloged in Table 1 , and represent fewer than 100 gene products out of the approximately 6000 protein-coding genes in the yeast genome ( Lin et al . , 2013 ) [although the number of authentic open-reading-frames undergoes constant revision ( http://www . yeastgenome . org/cache/genomeSnapshot . html ) ] . 10 . 7554/eLife . 03779 . 004Table 1 . Known Ypk1 substrates and potential substrates predicted by MOTIPS listed under GO Slim terms*DOI: http://dx . doi . org/10 . 7554/eLife . 03779 . 004GeneMOTIPS sitesChemical Sensitivity/YeastMine Interaction ( s ) SDL scoreYpk1 dosage rescueIn vitro substrateKnown Ypk1 Substrates GPD1/YDL022W*24PYeastMine+++N/A+ FPK1/YNR047W37 , 200 , 244 , 436 , 481YeastMine+N/A+ FPK1 ( D621A ) [Kinase-dead mutant]37 , 200 , 244 , 436 , 481YeastMine++N/A+ ORM1/YGR038W52 , 53YeastMine+++N/A+ ORM2/YLR350W47 , 48YeastMine++++N/A+Cytoskeleton Organization AVO1/YOL078W552P , 597 , 1078YeastMine−N/AN/A AVO2/YMR068W273P , 305 , 407YeastMine−N/AN/A BEM2/YER155C83 , 168 , 1810 , 1813Myr , YeastMine−N/AN/A BNI1/YNL271C119 , 1138P , 1533AbA , Casp , YeastMine−N/AN/A CDH1/YGL003C51 , 195 , 213PAbA , YeastMineTOXIC−N/A ENT1/YDL161W160PYeastMine−N/AN/A GIC2/YDR309C90 , 312 , 345PMyr , AbA−N/AN/A LSB3/YFR024C-A262PYeastMine−N/AN/A PAL1/YDR348C49P , 391 , 436AbA , Casp++++N/A− SLA1/YBL007C445 , 447P , 449P , 477YeastMine−N/AN/A TSC11/YER093C19P , 97 , 188YeastMineN/AN/AN/A YHR097C58 , 288P , 294PMyr+++N/A+/− YSC84/YHR016C274 , 374PMyr , YeastMine−N/AN/ABiological Process Unknown COM2/YER130C251 , 370 , 380Myr , YeastMine−N/AN/A ECM3/YOR092W*312 , 350Myr , AbA , YeastMine−N/AN/A ICS2/YBR157C14 , 95 , 136 , 172PMyr−N/AN/A JIP4/YDR475C348 , 352 , 592 , 649Myr , AbAN/AN/AN/A KKQ8/YKL168C83 , 113 , 144 , 146 , 212 , 293YeastMine−N/AN/A RTS3/YGR161C*30 , 238PYeastMine−N/AN/A SEG1/YMR086W56 , 118P , 217 , 634 , 752PMyr+N/AN/A YDR186C334P , 540P , 542P , 620 , 715PYeastMine−N/AN/A YHR080C401 , 513 , 667YeastMine−N/AN/A YNR014W54 , 115 , 156 , 197YeastMine+++N/A+ YPK3/YBR028C72 , 73 , 90PMyr , Casp−N/AN/ATranscription from RNA Polymerase II Promoter EPL1/YFL024C24 , 28 , 61YeastMine−N/AN/A FKH1/YIL131C404Myr , YeastMineTOXIC−− GAL11/YOL051W1003PMyr , YeastMine−N/AN/A HCM1/YCR065W*80AbA , YeastMine−N/AN/A HOT1/YMR172W*387 , 520 , 586Myr−N/AN/A RLM1/YPL089C*20MyrTOXIC−N/A SMP1/YBR182C*20 , 107YeastMineTOXIC++ SSN2/YDR443C608PMyr−N/AN/A YHP1/YDR451C*180 , 182Myr−N/AN/AMitotic Cell Cycle BCK2/YER167W12 , 38 , 373 , 430YeastMineTOXIC−N/A ESC2/YDR363W114 , 143 , 145YeastMine−N/AN/A PTK2/YJR059W59P , 82 , 91 , 171 , 275Myr , YeastMine+N/AN/A SET3/YKR029C236 , 405 , 428YeastMine−N/AN/A SWI4/YER111C816PMyr , YeastMine−N/AN/A VHS2/YIL135C316 , 318 , 325PMyr , Casp−N/AN/A ZDS1/YMR273C*78 , 370AbA , YeastMine−N/AN/A ZDS2/YML109W183 , 267 , 345YeastMine−N/AN/AProtein Phosphorylation HAL5/YJL165C17P , 217P , 233YeastMine−N/AN/A KIN1/YDR122W652 , 791P , 879 , 986PYeastMine+N/AN/A KIN2/YLR096W665P , 818 , 1020Myr−N/AN/A KSP1/YHR082C594 , 827P , 884PMyr , AbA , YeastMineTOXIC−N/A NPR1/YNL183C125P , 255P , 257P , 317PYeastMine++++N/A− SKY1/YMR216C383PMyr , YeastMine−N/AN/A YAK1/YJL141C128P , 206 , 240Myr , Casp , YeastMine−N/AN/A YPL150W371P , 452 , 890YeastMine−N/AN/ALipid Metabolic Process ADR1/YDR216W180 , 230P , 756Myr , AbA−N/AN/A CDC1/YDR182W*9N/A−N/A+ CKI1/YLR133W14P , 25P , 30PMyr , AbA , YeastMine−N/AN/A GPT2/YKR067W27 , 652Myr+++N/A+ LAC1/YKL008C*23 , 24Myr , YeastMine+++N/A+ LAG1/YHL003C24PMyr , YeastMine+++N/A+ LCB3/YJL134W16PMyr , YeastMine−N/A+Cellular Ion Homeostasis and Transport AVT3/YKL146W55 , 59P , 172 , 174Myr , AbA , YeastMine−N/AN/A CCH1/YGR217W†146 , 148 , 347YeastMine−N/A+ FPS1/YLL043W147 , 181 , 185 , 570PMyr , YeastMine+++++N/A+ MNR2/YKL064W165 , 620 , 621 , 826AbA−N/AN/A NHA1/YLR138W*544 , 830Myr , YeastMine−N/AN/A PPZ1/YML016C122 , 203 , 250PMyr , YeastMineTOXIC−N/ATranslation DED1/YOR204W84 , 576PYeastMine−N/AN/A HCR1/YLR192C*223Myr , AbA , YeastMine−N/AN/A HEF3/YNL014W*898Myr , AbA , YeastMine−N/AN/A RPL3/YOR063W24P , 337Myr , AbA , YeastMine−N/AN/A SUI2/YJR007W*58Myr−N/AN/A TEF1/YPR080W*72PMyr−N/AN/ACell Wall Organization or Biogenesis BPH1/YCR032W1334 , 1336 , 1963Casp−N/AN/A CSR2/YPR030W61 , 103 , 525 , 987Myr−N/AN/A ROM2/YLR371W76P , 193p , 396YeastMine−N/AN/A SSD1/YDR293C164P , 482P , 503Myr , AbA , YeastMineTOXIC−N/AGolgi Vesicle Transport BRE5/YNR051C398PMyr , YeastMine+++N/A− EXO84/YBR102C76 , 313 , 494 , 554YeastMine−N/AN/A MUK1/YPL070W173 , 184P , 185PMyr+++N/A+/− RGP1/YDR137W220 , 364P , 450 , 452YeastMine−N/AN/ASignaling IRA2/YOL081W882 , 884 , 1578 , 1745 , 3069YeastMineN/AN/AN/A GIS3/YLR094C*249 , 333Myr , AbA−N/AN/A MDS3/YGL197W757 , 824 , 842 , 851 , 1204Myr , AbA , YeastMine+++N/A− SYT1/YPR095C277P , 410 , 728YeastMineN/AN/AN/ADNA Replication CDC13/YDL220C314 , 333PYeastMineTOXIC−N/A CTI6/YPL181W155 , 216PMyr , YeastMine−N/AN/A RIM4/YHL024W93 , 429 , 525 , 607YeastMine−N/AN/AEndocytosis ALY2/YJL084C166P , 201 , 225 , 803Myr−N/AN/A ROD1/YOR018W563 , 617 , 807 , 823Myr+++N/A+/− ROG3/YFR022W425 , 584 , 718YeastMine−N/AN/AOther FRT1/YOR324C167 , 201 , 203 , 228P , 385Myr , YeastMine−N/AN/A HER1/YOR227W28P , 102p , 157PMyr , AbATOXIC−+ YSP2/YDR326C326 , 518 , 1237Myr , YeastMine+++N/A+RNA Catabolic Process JSN1/YJR091C174 , 275P , 600YeastMine−N/AN/A PUF2/YPR042C55 , 143 , 246 , 902MyrN/AN/AN/ACytokinesis CYK3/YDL117W159 , 207P , 746AbA , YeastMine+++N/A−Chromosome Segregation DSN1/YIR010W240 , 250PYeastMine−N/AN/APeroxisome Organization PEX31/YGR004W432PYeastMine++N/A+Pseudohyphal Growth PAM1/YDR251W471 , 553P , 625Myr , AbA , Casp , YeastMine−N/AN/AResponse to Starvation ATG21/YPL100W191 , 237PMyr+++N/A+*Genes that are not bioinformatically predicted Ypk1 substrates , but contain Ypk1 motifs and were included in this study are marked with an asterisk . SDL assay results are listed for each bioinformatically predicted Ypk1 substrate . The scoring system reports growth phenotypes of the ypk1-as ypk2Δ strain transformed with the indicated PGAL1-SUBSTRATE plasmid upon overexpression on galactose with varying levels of 3-MB-PP1-imposed Ypk1-as inhibition . A growth phenotype is defined as at least 1 serial dilution spot less growth than YCpLG-GFP control at the given 3-MB-PP1 concentration . A strong growth phenotype is defined as no growth at the given 3-MB-PP1 concentration . ( +++++ ) indicates a growth phenotype with no 3-MB-PP1 . ( ++++ ) is a strong growth phenotype on 1 μM 3-MB-PP1 . ( +++ ) indicates a growth phenotype on 1 μM 3-MB-PP1 . ( ++ ) is defined as no phenotype on 1 μM 3-MB-PP1 , but a strong growth phenotype on 2 μM 3-MB-PP1 . ( + ) indicates no phenotype on 1 μM 3-MB-PP1 , but a detectable growth phenotype on 2 μM 3-MB-PP1 . ( − ) indicates no growth phenotype at any concentration of 3-MB-PP1 tested . TOXIC indicates overexpression of the putative substrate on galactose-containing medium was deleterious to growth even in the wild-type control strain ( BY4741 ) . These toxic genes were then tested for Ypk1 dosage rescue ( for details , see ‘Materials and methods’ ) ; here , ( + ) indicates that Ypk1 overexpression could at least partially rescue the overexpression toxicity of the indicated gene and ( − ) indicates that Ypk1 overexpression could not rescue the overexpression toxicity . Lastly , the results of testing the indicated purified predicted Ypk1 target as a substrtate in the in vitro protein kinase assay with purified Ypk1-as; here , ( + ) indicates that Ypk1-as- dependent ( 3-MB-PP1 inhibitable ) incorporation was detectable for the substrate at a level comparable to incorporation into the positive control [the known Ypk1 substrate , GST-Orm1 ( 1–85 ) ]; ( +/− ) indicates that readily detectable Ypk1-as-dependent incorporation was found , but at a level lower than that seen for an equivalent amount of GST-Orm1 ( 1–85 ) protein . ( N/A ) indicates that the indicated gene product was not tested in the indicated assay . †The SDL assay was performed with a plasmid constitutively overexpressing CCH1 under the TDH3 promoter [pBCT-CCH1H , ( Iida et al . , 2007 ) ] , as our efforts to generate a PGAL1-CCH1 vector failed . As the secondary filter for the candidates recognized bioinformatically , we developed an in vivo approach to identify those gene products that manifested an expected hallmark of protein-substrate interaction . We reasoned that under conditions where the level of activity of an essential kinase , like Ypk1 , is near-limiting for normal growth , high-level over-expression of an authentic substrate might tie up the available pool of active enzyme and prevent efficient phosphorylation of other cellular substrates necessary for growth and/or viability ( Figure 1C ) . This scheme is a novel variation on a genetic approach referred to as synthetic dosage lethality ( SDL ) ( Sopko et al . , 2006; Sharifpoor et al . , 2012 ) . To limit Ypk1 activity , we used ypk1-as ypk2Δ cells , which express from the YPK1 locus an analog-sensitive allele , Ypk1 ( L424A ) ( Roelants et al . , 2011; Niles et al . , 2012 ) , and titrated down its activity by addition of a low concentration of an efficacious inhibitor , 1- ( tert-butyl ) -3- ( 3-methylbenzyl ) -1H-pyrazolo[3 , 4-d]pyrimidin-4-amine ( 3-MB-PP1 ) ( Burkard et al . , 2007 ) that has no effect on wild-type cells ( see Figure 2C ) . To achieve high-level over-expression , each bioinformatic hit was expressed from the galactose-inducible GAL1 promoter on a CEN plasmid . As proof of concept , we used two known Ypk1 substrates , Orm1 and Orm2 , as positive controls and GFP , which is not a Ypk1 substrate ( data not shown ) , as a negative control . In the absence of limiting the activity of Ypk1 ( L424A ) with inhibitor , overexpression of neither Orm proteins nor GFP was deleterious to cell growth ( Figure 1D , left panels , compare lower to upper ) . However , in the presence of a low dose of 3-MB-PP1 , overexpression of Orm1 and Orm2 on galactose medium prevented cell growth , whereas overexpression of GFP did not ( Figure 1D , middle and right panels , compare lower to upper ) . We were able to test the majority of the candidates ( 90/96 ) that arose in the bioinformatic search in this same fashion [however , 10/90 caused toxicity upon over-expression even in wild-type cells and , hence , could not be scored] . 10 . 7554/eLife . 03779 . 005Figure 2 . Lac1 and Lag1 , subunits of ceramide synthase , were identified by the screen . ( A ) A diagram of the membrane topology of Lac1 and Lag1 derived from ( Kageyama-Yahara and Riezman , 2006 ) ; Lac1 and Lag1 are experimentally determined to have eight transmembrane domains . The N terminus of these proteins is cytosolic and therefore accessible for Ypk1 phosphorylation . ( B ) Diagram of yeast de novo sphingolipid biosynthesis derived from ( Dickson , 2008 ) . Metabolites appear as boxes and enzymes as ovals . Metabolites in green are those directly produced or derived from ceramide synthase while those in red are alternative products at the ceramide synthase branch point . ( C ) SDL results for Lac1 and Lag1 . ypk1-as ypk2Δ ( yAM135–A ) or wildtype ( BY4741 ) cells were transformed with PGAL1-GFP ( negative control ) , PGAL1-Lac1 or PGAL1-Lag1 plasmids . The SDL assay was then performed as described in ‘Materials and methods’ . ( D ) GST-Lac1 ( 1–76 ) ( pAX131 ) , GST-Lag1 ( 1–80 ) ( pFR29 ) , GST-Lac1 ( 1–76 ) ( S23A S24A ) ( pAX132 ) and GST-Lag1 ( 1–80 ) ( S23A S24A ) ( pAX133 ) were purified from E . coli and Ypk1-as kinase assays were performed as in Figure 1D . DOI: http://dx . doi . org/10 . 7554/eLife . 03779 . 005 Those candidates that , like Orm1 and Orm2 , exhibited toxicity only on galactose medium and only when Ypk1 ( L424A ) activity was limited in the presence of 3-MB-PP1 , but not when inhibitor was absent , were designated SDL hits ( Table 1 , column 4 ) . Moreover , use of a series of 3-MB-PP1 concentrations allowed for quantification of the strength of the SDL effect ( from + to ++++ ) . In one case ( Fps1 ) , a marked SDL effect was observed upon overexpression in the ypk1-as ypk2Δ cells in the absence of chemical inhibition; we considered this a valid SDL hit because GAL promoter-driven over-expression of Fps1 was not growth inhibitory in wild-type ( YPK1+ YPK2+ ) cells . Thus , as summarized in Table 1 ( column 4 ) , a significant fraction ( 20/90 ) of the candidates identified bioinformatically that were tested in this fashion , but far from all , displayed an SDL phenotype . In this regard , it is important to note that all known Ypk1 substrates tested ( Gpd1 , Fpk1 , Orm1 and Orm2 ) displayed an SDL phenotype , whereas nearly 80% of the bioinformatic hits , like GFP , did not . Consistent with the view that the SDL phenotype could arise from the over-expressed target serving as a decoy substrate that titers a limited pool of active Ypk1 away from acting on its essential substrates , we observed that over-expressed catalytically-inactive Fpk1 caused an SDL phenotype equivalent to or stronger than wild-type Fpk1 ( Table 1 ) . If such SDL phenotypes reflect occlusion of a limited pool of enzyme by over-expressed substrate , then , conversely , co-overexpression of Ypk1 or even of a kinase-dead allele Ypk1 ( K376A ) ( driven from the MET25 promoter ) might rescue the toxicity . Indeed , the deleterious effect of Smp1 over-expression was rescued by co-overexpression of either Ypk1 or Ypk1 ( K376A ) ( Table 1 ) , suggesting that the SDL phenotype of over-expressed Smp1 also arises from titration of a limited amount of Ypk1 away from essential substrates . Lastly , to determine whether the gene products that displayed an SDL phenotype are indeed substrates for Ypk1 , we incubated those ( 17/20 ) that we were able to successfully express and purify as recombinant proteins or protein fragments ( as GST fusions ) from Escherichia coli with [γ-32P]ATP and Ypk1 ( L424A ) , which was highly purified from yeast cells as described in ‘Materials and methods’ . We chose to use the analog-sensitive allele , even though it is only about 50% as active as wild-type Ypk1 ( Roelants et al . , 2011 ) , because ablation of activity in the presence of 3-MB-PP1 allowed us to confirm that any 32P incorporation into substrate observed was due to phosphorylation by Ypk1 ( L424A ) itself ( and not due to some other protein kinase that might be present in the preparation ) . All known in vivo substrates of Ypk1 ( Fpk1 , Fpk2 , Gpd1 , Orm1 and Orm2 ) display robust incorporation ( as judged by autoradiography ) in this in vitro assay ( Roelants et al . , 2010 , 2011; Lee et al . , 2012b ) . Therefore , in testing each candidate , an appropriate positive control , Orm1 ( 1–85 ) was included ( Figure 1E ) , which also allowed comparison between independent assays . Gratifyingly , 12/17 ( 70% ) of the SDL hits tested in this manner displayed readily detectable and Ypk1-specific phosphorylation ( Table 1 , right column ) . Moreover , the results of such assays clearly demonstrated that Ypk1 is not a ‘promiscuous’ enzyme . For example , Fps1 has three possible N-terminal Ypk1 sites ( RPRGQT147T , RRRSRS181R and RSRATS185N ) and one C-terminal site . The C-terminal site is a Ypk1 target ( data not shown ) ; however , none of the N-terminal motifs serves as a Ypk1 phospho-acceptor site ( Figure 1E ) , most likely because each lacks a hydrophobic residue at +1 . Thus , we considered it very likely that the dozen candidates identified bioinformatically that also displayed an SDL phenotype and served as Ypk1 substrates in vitro ( highlighted in bold in Table 1 ) would be functionally important Ypk1 substrates in vivo . To validate this conclusion and confirm that these candidates are indeed physiologically relevant Ypk1 targets , we chose to characterize Lac1 and Lag1 , two of the dozen candidates ( Table 1 ) , because they are the catalytic subunits of the ceramide synthase complex and might further our understanding about how sphingolipid production is regulated by the TORC2-Ypk1 signaling axis . Lac1 ( 418 residues ) and Lag1 ( 411 residues ) are apparent paralogs at the primary sequence level ( 69% identity , 77% similarity ) ( Byrne and Wolfe , 2005 ) and are polytopic integral proteins in the ER membrane with the predicted Ypk1 site in each protein residing near its N-terminus ( Figure 2A ) . It has been shown that the N-termini of these proteins are exposed to the cytosol ( Kageyama-Yahara and Riezman , 2006 ) . Along with a small accessory subunit Lip1 ( 150 residues ) , Lac1 and Lag1 are demonstrated constituents of the ceramide synthase complex ( Schorling et al . , 2001; Vallée and Riezman , 2005 ) , which catalyzes N-acylation of the free amino group on the long-chain base ( LCB ) , mainly phytosphingosine in yeast , using fatty acyl-CoA as the acyl donor , thereby forming phytoceramide ( Figure 2B ) . Genetically , Lac1 and Lag1 appear to play overlapping functional roles; lac1Δ or lag1Δ single mutants are viable , whereas a lac1Δ lag1Δ double mutant is reportedly either inviable ( Jiang et al . , 1998 ) or extremely slow growing ( Barz and Walter , 1999; Schorling et al . , 2001; Vallée and Riezman , 2005 ) . The ceramide synthase reaction lies at an important branch point in the sphingolipid metabolic network ( Figure 2B ) because de novo synthesis of ceramides both consumes LCBs and prevents conversion of LCBs to their 1-phosphorylated derivatives ( LCBPs ) . Thus , the rate of ceramide synthesis is tightly coupled to the levels of both LCBs and LCBPs ( Kobayashi and Nagiec , 2003 ) ; and , moreover , the balance between ceramides and total LCBs and LCBPs affects growth rate in both fungi ( Kobayashi and Nagiec , 2003 ) and mammalian cells ( Spiegel and Milstien , 2003 ) . By virtue of their position in the pathway , Lac1 and Lag1 are situated to be important regulators of this balance . Among the bioinformatically predicted substrates , both Lac1 and Lag1 displayed a readily detectable SDL phenotype ( Figure 2C ) and both GST-Lac1 ( 1–76 ) and GST-Lag1 ( 1–80 ) served as in vitro substrates for Ypk1 , albeit with the phosphorylation of Lac1 being reproducibly much more robust than Lag1 ( Figure 2D ) . Site-directed mutagenesis confirmed that the Ypk1-mediated phosphorylation of both substrates occurred exclusively at the predicted phospho-acceptor site ( s ) , specifically Ser23 and Ser24 in both proteins ( Figure 2D ) . To determine whether both Lac1 and Lag1 are phosphorylated in vivo in a Ypk1-dependent manner and at their Ypk1 consensus sites , integrated 3xHA- or 3xFLAG-tagged versions of each protein and its corresponding S23A S24A mutant were expressed in yeast and extracts of the cells were analyzed by phosphate-affinity SDS-PAGE ( Phos-tag gels ) ( Kinoshita et al . , 2009 ) . In this separation technique , the more highly phosphorylated the protein , the more its migration is retarded . Both Lac1 ( Figure 3A , left ) and Lag1 ( Figure 3A , right ) migrated as two species , and the slower of the two could be attributed to phosphorylation because it was eliminated if the sample was pre-treated with calf intestinal phosphatase . This slower mobility species represented phosphorylation at S23 and S24 because the band was also eliminated in Lac1 ( S23A S24A ) and Lag1 ( S23A S24A ) mutants ( Figure 3A ) . We noted that phosphatase treatment , even of the Lac1 ( S23A S24A ) and Lag1 ( S23A S24A ) mutants , resulted in appearance of a third , even faster migrating species , presumably due to removal of a phosphorylation ( s ) elsewhere in these proteins , consistent with indirect evidence that Lac1 and Lag1 might be subject to casein kinase II ( yeast Cka2 ) -dependent modification ( Kobayashi and Nagiec , 2003 ) . 10 . 7554/eLife . 03779 . 006Figure 3 . Ypk1 phosphorylates Lac1 and Lag1 at S23 and S24 in vivo . ( A ) 3xHA-Lac1 ( yAM165–A ) 3xHA-Lac1 ( S23A S24A ) ( yAM166–A ) , 3xFLAG-Lag1 ( yAM159–A ) and 3xFLAG-Lag1 ( S23A S24A ) ( yAM163–A ) strains were grown to mid-exponential phase in YPD . Cells were then harvested and whole-cell extracts were prepared . Extracts were split and one fraction was then treated with calf intestinal phosphatase . The extracts resolved by Phos-tag SDS-PAGE and immunoblotted with anti-HA , -FLAG or -Pgk1 ( loading control ) antibodies . P-Lac1 and P-Lag1 indicate the band corresponding to S23 S24 phosphorylation . * indicates a non-specific band that appears in HA blots of yeast whole cell extracts . ( B ) Wildtype ( BY4741 ) , ypk1Δ ( JTY6142 ) and ypk2Δ ( yAM120–A ) strains were transformed with a plasmid centromeric plasmid encoding 3xHA-Lac1 expressed under control of its endogenous promoter ( pAX136 ) . Cells were grown to mid-exponential phase and then treated with a sublethal dose of myriocin ( 0 . 625 μM ) or methanol ( vehicle ) for 2 hr . Cell extracts were then prepared , resolved by Phos-tag SDS-PAGE and blotted as above in ( A ) . ( C ) TOR2 ( yKL04 ) or tor2-as ( yKL05 ) strains were transformed with a 3xHA-Lac1 expressing plasmid ( pAX136 ) . Cells were grown to mid-exponential phase and treated with 1 μM BEZ-235 for the indicated times . Cell extracts were then prepared , resolved by Phostag SDS-PAGE and blotted as above in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03779 . 006 In agreement with their relative efficacies as in vitro substrates ( Figure 2D ) , we found that , reproducibly , the majority of Lac1 was present in the cell as the slower mobility isoform , whereas the opposite was true for Lag1 ( Figure 3A ) . Because the behavior of Lac1 gave us greater sensitivity of detection , and for the sake of conciseness , some of our subsequent findings are illustrated with data for Lac1 only . However , all experiments were repeated with both proteins with virtually identical results and conclusions . To further confirm that Ypk1 is the protein kinase responsible for phosphorylation at Ser23 Ser24 in vivo , plasmids encoding 3xHA-tagged Lac1 and Lag1 were introduced by DNA-mediated transformation into wild-type , ypk1Δ , and ypk2Δ strains . Cultures of the resulting cells were grown in the absence or presence of a sub-lethal dose of myriocin , a condition which several previous studies have shown activates TORC2- and Ypk1-mediated signaling ( Roelants et al . , 2011; Berchtold et al . , 2012; Sun et al . , 2012 ) , and the resulting extracts were analyzed on Phos-tag gels . As observed previously for two other bona fide substrates , Orm1 and Orm2 ( Roelants et al . , 2011 ) , absence of Ypk1 totally abrogated the appearance of phosphorylated Lac1 ( Figure 3B ) and phosphorylated Lag1 ( data not shown ) , whereas elimination of Ypk2 had no effect . Thus , Ypk1 is the paralog solely responsible for the observed in vivo phosphorylation of Lac1 and Lag1 at Ser23 and Ser24 . Furthermore , under conditions that stimulate TORC2 activity , nearly all of the Lac1 ( Figure 3B ) and much more of the Lag1 ( data not shown ) were converted to the phosphorylated isoform indicating that TORC2 activation is relayed to ceramide synthase via Ypk1 . To further confirm that TORC2 function is essential for Ypk1-mediated phosphorylation of ceramide synthase , Lac1 phosphorylation was monitored in TOR2 cells and in a tor2-as mutant after addition of a specific tor2-as inhibitor , BEZ-235 ( Kliegman et al . , 2013 ) . TORC2 inhibition markedly reduced Lac1 phosphorylation within 20 min in tor2-as cells , but not in the otherwise isogenic control cells ( Figure 3C ) , confirming that TORC2 activity is necessary for Ypk1-mediated ceramide synthase phosphorylation . Thus , our screening approach was successful in revealing two , previously uncharacterized , cellular targets of the TORC2-Ypk1 signaling axis . The fact that impeding LCB production with a sub-lethal dose of the SPT inhibitor myriocin stimulated Ypk1-mediated Lac1 and Lag1 phosphorylation suggested that this modification is important for their physiological function . As one means to confirm that reduction in sphingolipid biosynthesis capacity results in up-regulation of Lac1 and Lag1 phosphorylation , we subjected the pathway to blockade near its end by treating the cells expressing integrated 3xHA-tagged Lac1 or Lag1 with aureobasidin A ( Heidler and Radding , 1995 ) , an antibiotic that prevents formation of complex sphingolipids in yeast by inhibiting Aur1 ( phosphatidylinositol:ceramide phosphoinositol transferase ) ( Nagiec et al . , 1997 ) . As observed for treatment with myriocin , the amount of phosphorylated Lac1 ( Figure 4A ) and Lag1 ( data not shown ) was markedly increased in response to treatment with aureobasidin A . 10 . 7554/eLife . 03779 . 007Figure 4 . Enhanced Ypk1 phosphorylation of Lac1 and Lag1 under sphingolipid and heat stress . ( A ) 3xHA-Lac1 ( yAM165–A ) cells were grown to early exponential phase in YPD . Cultures were then treated with sublethal doses of myriocin ( 0 . 625 μM ) or methanol ( vehicle ) or aureobasidin A ( 1 . 8 μM ) or ethanol ( vehicle ) for 2 hr . ( B ) 3xHA-Lac1 ( yAM165–A ) cells were grown to exponential phase in YPD at 30°C . A sample of this culture was then harvested . The remaining culture was then moved to 42°C to initiate heat shock and samples were harvested at the indicated time points . Whole cell extracts were prepared from each sample , resolved by Phos-tag SDS-PAGE , and immunoblotted as in Figure 3 . P-Lac1 indicates the band corresponding to S23 S24 phosphorylation . * indicates a non-specific band that appears in HA blots of yeast whole cell extracts . ( C ) LAC1 LAG1 ( yAM205–A ) , Lac1SSAA Lag1SSAA ( yAM207B ) and Lac1SSEE Lag1SSEE ( yAM210 ) were grown to exponential phase in YPD . Serial dilutions of each culture were made and spotted on YPD plates containing vehicle or the indicated concentration of myriocin or aureobasidin A . Cells were allowed to grow for 3 days at 30°C prior to imaging . ( D ) 3xHA-Lac1::HIS3 3xFLAG-Lag1::LEU2 ( yAM168 ) 3xHA-Lac1 ( S23A S24A ) ::HIS3 3xFLAG-Lag1 ( S23A S24A ) ::LEU2 ( Lac1SSAA Lag1SSAA ) ( yAM184 ) and 3xHA-Lac1 ( S23E S24E ) ::HIS3 3xFLAG-Lag ( S23E S24E ) 1::LEU2 ( Lac1SSEE Lag1SSEE ) ( yAM192–A ) cells were grown to mid-exponential phase and then treated with 1 . 0 μM myriocin or 18 . 2 nM aureobasidin A for 2 hr . Whole cell extracts were prepared from each sample , resolved by SDS-PAGE , and immunoblotted as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 03779 . 007 Another perturbation that has been shown to transiently up-regulate TORC2-Ypk1-mediated signaling is heat shock ( Sun et al . , 2012 ) . Consistent with that response , it has been shown previously that heat shock leads to a transient increase in sphingolipid production and that sphingolipid production is important for heat shock survival ( Jenkins et al . , 1997; Cowart et al . , 2003 ) . Moreover , measurement of pathway intermediates and mathematical modeling also suggested that a sharp spike of increased ceramide synthase activity may occur after heat shock ( Chen et al . , 2013 ) . Therefore , we subjected the same cells to heat shock and monitored Lac1 and Lag1 phosphorylation at various times thereafter . Within 5 min , the amount of phosphorylated Lac1 ( Figure 4B ) and Lag1 ( data not shown ) increased markedly , but was back to the resting level by 30 min . If these changes in phosphorylation state at Ser23 and Ser24 in Lac1 and Lag1 are important for the metabolic adjustments that the cell needs to adapt appropriately , then preventing phosphorylation at these sites should impair cell survival . To test this prediction , we integrated as the sole source of ceramide synthase , mutant versions of Lac1 and Lag1 in which Ser23 and Ser24 were mutated to Ala and , hence , cannot be phosphorylated under any circumstances . As a control , we also generated integrated versions of Lac1 and Lag1 in which Ser23 and Ser24 were mutated to Glu , to mimic conversion of the entire population to the phosphorylated state , a response that we showed can be achieved for the wild-type protein ( see , for example , Figure 3B ) . Indeed , we found that the cells co-expressing Lac1 ( S23A S24A ) and Lag1 ( 23A S24A ) grew detectably less well when challenged with either myriocin or aureobasidin A than either otherwise isogenic wild-type cells or cells co-expressing Lac1 ( S23E S24E ) and Lag1 ( S23E S24E ) ( Figure 4C ) . These growth differences could not be attributed to differences in the level of expression of these proteins , as immunoblot analysis demonstrated the wild-type and mutant ceramide synthase subunits were present in equivalent amounts under the conditions tested ( Figure 4D ) . Thus , Ypk1-dependent phosphorylation of these sites in Lac1 and Lag1 is functionally important for cell survival in response to the stress of limiting sphingolipid biosynthesis . Furthermore , the fact that the Lac1 ( S23E S24E ) Lag1 ( S23E S24E ) strain phenocopied a LAC1+ LAG1+ strain under conditions that promote phosphorylation of Lac1 and Lag1 indicates that these mutations generated effective phosphomimetic alleles . As a means to delineate what cellular phosphatase is responsible for counteracting the Ypk1-mediated phospho-regulation of Lac1 and Lag1 , plasmid-borne 3xHA-tagged Lac1 was expressed in a collection of deletion strains lacking each of the non-essential protein phosphatase genes or their associated factors , and the phosphorylation state of Lac1 was assessed using Phos-tag gels . By this approach , we were unable to find any phosphatase-deficient mutant that exhibited a significant increase in the amount of phosphorylated Lac1 compared to control cells ( data not shown ) . However , considerable genetic evidence indicates a strong connection between Ca2+ signaling and sphingolipid biosynthesis ( Beeler et al . , 1998 ) . Moreover , it has been reported previously that TORC2-Ypk-dependent regulation of sphinglipid biosynthesis is antagonized by the action of the Ca2+/calmodulin-dependent protein phosphatase calcineurin ( also known as phosphoprotein phosphatase 2B ) , although the level at which the phosphatase acted was unknown ( Aronova et al . , 2008 ) . Hence , we conducted the converse experiment by stimulating the cells expressing 3xHA-tagged Lac1 acutely with 0 . 2 M CaCl2 , a condition known to robustly activate calcineurin in yeast ( Stathopoulos-Gerontides et al . , 1999 ) . Within <10 min , we found total abrogation of phospho-Lac1 ( Figure 5A , left ) and total abrogation of phospho-Lag1 ( data not shown ) in wild-type cells , whereas in otherwise isogenic cna1Δ cna2Δ mutants ( which lack both calcineurin catalytic subunit paralogs ) a substantial portion of the phosphorylated species remained ( Figure 5A , right ) . 10 . 7554/eLife . 03779 . 008Figure 5 . Activation of calcineurin leads to rapid dephosphorylation of Lac1 and Lag1 without affecting Ypk1 function . ( A ) Wildtype ( BY4741 ) or cna1Δ cna2Δ ( JTY5574 ) strains were transformed with a centromeric plasmid encoding 3xHA-Lac1 expressed under control of its endogenous promoter ( pAX136 ) . Cultures were grown to mid-exponential phase in selective media and then treated with 200 mM CaCl2 for 10 min to activate calcineurin . Cultures were then harvested and whole cell extracts were prepared from each sample , resolved by Phos-tag SDS-PAGE , and immunoblotted as in Figure 3 . ( B ) 3xHA-Lac1 ( yAM165-A ) cells were transformed with a centromeric plasmid encoding the hyperactive TORC2 independent Ypk1D242A allele expressed under its own promoter ( pFR273 ) or vector ( pRS316 ) . Cultures were grown in selective media to mid-exponential phase before treatment with 200 mM CaCl2 for 10 min . Cultures were then harvested and whole cell extracts were prepared from each sample , resolved by Phos-tag SDS-PAGE , and immunoblotted as above . ( C ) 3xFLAG-Ypk1 ( YDB379 ) cells were grown to mid-exponential phase in YPD . Cultures were treated with or without 200 mM CaCl2 for 10 min . Cells were then harvested and whole extracts prepared . Extracts were resolved by SDS-PAGE and blotted with anti-pSGK ( T256 ) , which recognizes Ypk1 T504 activation loop phosphorylation ( Roelants et al . , 2010 ) and anti-FLAG antibody ( Ypk1 loading control ) . The blot is representative of triplicate samples and the quantitation of the ratio of pT504/FLAG from these replicates is shown below the blot . DOI: http://dx . doi . org/10 . 7554/eLife . 03779 . 008 These results suggested that calcineurin may directly reverse the phoshorylations introduced into Lac1 and Lag1 by Ypk1 . However , it is also the case that two of the ancillary subunits associated with TORC2 , Slm1 and Slm2 , are demonstrated calcineurin-binding proteins ( Bultynck et al . , 2006; Tabuchi et al . , 2006 ) and that calcineurin action appears to oppose TORC2-dependent signaling ( Mulet et al . , 2006; Daquinag et al . , 2007; Berchtold et al . , 2012 ) . Hence , it was possible , therefore , that the observed loss of phospho-Lac1 and -Lag1 might be due to a Ca2+-stimulated and calcineurin-mediated reduction in their TORC2-Ypk1-dependent phosphorylation , rather than to direct action of calcineurin on phospho-Lac1 and -Lag1 . To rule out the former possibility , we examined Lac1 and Lag1 phosphorylation in cells expressing constitutively-active Ypk1 ( D242A ) , which we have demonstrated bypasses the need for its TORC2-dependent activation ( Roelants et al . , 2011 ) . We found that Ca2+ addition still led to nearly complete Lac1 and Lag1 dephosphorylation in these cells ( Figure 5B ) . Furthermore , as judged by immunoblotting with a phospho-site specific antibody , Pkh1- and Pkh2-dependent phosphorylation of the activation loop in Ypk1 ( Figure 5C ) was not diminished in cells stimulated with Ca2+ , confirming that there was no calcineurin-mediated decrease in the amount of active Ypk1 present . Therefore , it seems clear that calcineurin dephosphorylates the sites in Lac1 and Lag1 phosphorylated by Ypk1 directly , rather than through down-regulation of Ypk1 function . Given the phenotypic evidence that TORC2-Ypk1-dependent modulation of Lac1 and Lag1 is physiologically important ( Figure 4C ) , we next conducted biochemical analysis to determine how Ypk1-mediated phosphorylation affects the function of Lac1 and Lag1 in ceramide synthesis . Toward that end , we first used LC-MS to monitor the levels of LCBs and LCBPs extracted from equivalent numbers of cells from exponentially-growing cultures of wild-type cells or otherwise isogenic cells expressing as the sole source of Lac1 and Lag1 either the non-phosphorylatable Lac1 ( S23A S24A ) and Lag1 ( S23A S24A ) mutants or the phosphomimetic Lac1 ( S23E S24E ) and Lag1 ( S23E S24E ) mutants . Relative to wild-type cells , the Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) cells accumulated significantly more PHS ( Figure 6A , top left ) , as well as more dihydrosphingosine ( DHS ) ( Figure 6A , top right ) , which is a more minor LCB in yeast ( note the difference in scale of the ordinate ) , whereas the Lac1 ( S23E S24E ) Lag1 ( S23E S24E ) cells displayed a level of both PHS and DHS that was somewhat lower , and a similar trend was observed even for PHS-1P , an even less abundant metabolite ( Figure 6A , bottom left ) . The level of DHS-1P ( Figure 6A , bottom right ) was so low as to make reliable measurement difficult , but no differences between strains could be detected . Given that the ceramide synthase complex is responsible for the conversion of LCBs into ceramides , these findings indicate that , in the absence of Ypk1-mediated phosphorylation , the rate of LCB utilization is significantly reduced , consistent with the conclusion that Ypk1-dependent modification of Lac1 and Lag1 promotes ceramide synthase function . 10 . 7554/eLife . 03779 . 009Figure 6 . Ypk1 phosphorylation of Lac1 and Lag1 stimulates ceramide synthase activity . ( A ) Cultures of LAC1 LAG1 ( yAM205–A ) , Lac1SSAA Lag1SSAA ( yAM207B ) and Lac1SSEE Lag1SSEE ( yAM210 ) strains were grown to mid-exponential phase and then harvested . Sphingolipids were extracted and analyzed as described in ‘Materials and methods’ . Values represent the mean of three independent experiments ( each performed in triplicate ) and error bars represent SEM . ( B ) Triplicate exponentially-growing cultures of LAC1 LAG1 ( yAM205–A ) , Lac1SSAA Lag1SSAA ( yAM207–B ) and Lac1SSEE Lag1SSEE ( yAM210 ) were grown overnight and duplicate cultures were diluted to OD600 = 1 . 0 . Complex sphingolipids were labeled and analyzed by thin layer chromatography ( TLC ) as in ‘Materials and methods’ . A representative TLC plate is shown with the origin at the bottom of the image . The assigned identity of species as IPCs and MIPCs was confirmed by pharmacological or genetic inhibition of the production of these species in control cultures ( data not shown ) . Quantification of total complex sphingolipids was performed in ImageJ by integrating the Phosphorimager screen intensity across the lane for each sample and normalized to 1 for the wild-type ceramide synthase samples . ( C ) Upper , ceramide synthase was purified by FLAG immunopurification from 3xHA-Lac1::HIS3 3xFLAG-Lag1::LEU2 ( yAM168 ) 3xHA-Lac1 ( S23A S24A ) ::HIS3 3xFLAG-Lag1 ( S23A S24A ) ::LEU2 ( Lac1SSAA Lag1SSAA ) ( yAM184 ) and 3xHA-Lac1 ( S23E S24E ) ::HIS3 3xFLAG-Lag ( S23E S24E ) 1::LEU2 ( Lac1SSEE Lag1SSEE ) ( yAM192–A ) cells . Immunoprecipitates were then split into three fractions and in vitro ceramide synthase assays ( 60 min reactions ) were performed in triplicate . A small sample of each ceramide synthase assay was resolved by SDS-PAGE and immunoblotted . The signal intensity quantified from the immunoblot was then used to normalize ceramide synthase activity in each sample . Lower , ceramide synthase was immunopurified from 3xHA-Lac1::HIS3 3xFLAG-Lag1::LEU2 ( yAM168 ) 3xHA-Lac1 ( S23A S24A ) ::HIS3 or 3xFLAG-Lag1 ( S23A S24A ) ::LEU2 ( Lac1SSAA Lag1SSAA ) ( yAM184 ) cells as above except cultures were treated with 1 . 0 μM myriocin or methanol ( vehicle ) prior to harvesting . Values represent the mean of three independent experiments ( each performed in triplicate ) and error bars represent SEM . Statistical significance of values ( Student's t test ) : *p = <0 . 05 , **p = <0 . 009; and , ***p < 0 . 0009 . DOI: http://dx . doi . org/10 . 7554/eLife . 03779 . 009 As an independent means to measure flux through the sphingolipid pathway , and because all complex sphingolipids in yeast contain inositol-phosphate , an equivalent number of cells of the same three strains in mid-exponential phase were pulse-labeled , in triplicate , with [32P]PO4−3 , and the acidic sphingolipids extracted and analyzed by thin-layer chromatography . Strikingly , the amount of complex sphingolipids generated during the pulse was reproducibly higher in the Lac1 ( S23E S24E ) Lag1 ( S23E S24E ) cells than in the wild-type controls , and Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) generated levels of complex sphingolipids lower than the wild-type controls ( Figure 6B ) . These findings are again consistent with the conclusion that Ypk1-mediated phosphorylation of Lac1 and Lag1 stimulates the production of the ceramide precursors to complex sphingolipids . Ypk1-mediated phosphorylation could stimulate the ceramide synthase reaction in vivo by stabilizing Lac1 and Lag1 thereby increasing their steady-state level , by enhancing their association with the small ancillary subunit Lip1 , and/or by direct activation . Immunoblotting of exponentially-growing cultures expressing the 3xHA-tagged versions of wild-type Lac1 and Lag1 and the Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) and Lac1 ( S23E S24E ) Lag1 ( S23E S24E ) indicated no discernible difference in their steady-state level ( see Figure 4D ) . Likewise , in cells co-expressing the same proteins and FLAG-tagged Lip1 ( gift of Howard Riezman , Univ . of Geneva ) , we observed no difference in the efficiency of Lip1 co-immunoprecipitation between wild-type Lac1 and Lag1 and either the Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) or Lac1 ( S23E S24E ) Lag1 ( S23E S24E ) mutants ( data not shown ) . These results suggested that Ypk1-mediated phosphorylation may directly enhance the catalytic efficiency of Lac1 and Lag1 . To test this possibility directly , 3xFLAG-tagged versions of Lac1 and Lag1 were immunopurified from detergent-solubilized microsomes isolated from exponentially-growing cells and equivalent amounts of the resulting protein assayed in vitro , monitoring the formation of ceramide from PHS and steroyl-CoA by LC-MS . No product was observed in the absence of added steroyl-CoA ( data not shown ) , and product formation was reduced by 85% in the presence of PHS and steroyl-CoA if 1 μM australifungin , a demonstrated and specific ceramide synthase inhibitor ( Mandala and Harris , 2000 ) , was added ( data not shown ) , confirming that the reaction measured was catalyzed by ceramide synthase . We found that the specific activity of the Lac1 ( S23E S24E ) Lag1 ( S23E S24E ) complex was reproducibly ∼2 higher than either the Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) mutant or the wild-type complex ( Figure 6C , upper panel ) , similar to the degree of difference in sphingolipid metabolites between these same strains that we measured by other means ( Figure 6A , B ) . In two independent trials ( each performed in triplicate ) , the identical trend was found when microsomes from these same cells were assayed directly ( i . e . without detergent solubilization and enzyme enrichment by immunoprecipitation [data not shown] ) . Additionally , we found that immunopurified ceramide synthase from wild-type cultures treated with myriocin had higher activity than ceramide synthase prepared from untreated cells ( Figure 6C , lower panel ) . Furthermore , this increase in ceramide synthase activity in response to myriocin treatment was not observed in Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) cells ( Figure 6C , lower panel ) , consistent with TORC2-Ypk1 signaling increasing ceramide synthase activity by phosphorylation at these residues . These findings are also in agreement with a reported ∼two-fold decrease in the rate of ceramide production by ceramide synthase complex isolated from TORC2-deficient yeast ( Aronova et al . , 2008 ) . Hence , we conclude that Ypk1 phosphorylation directly increases the catalytic activity of ceramide synthase . The observed increase in ceramide synthase activity in response to Ypk1-mediated phosphorylation could serve two roles that are not mutually exclusive: ( i ) to produce more ceramide and the derived complex sphingolipids; and , ( ii ) to prevent inadvertent accumulation of LCBs and the derived LCBPs when TORC2-driven Ypk1 activation stimulates metabolic flow into the sphingolipid pathway by alleviating Orm1- and Orm2-imposed inhibition of SPT ( Roelants et al . , 2011 ) . We reasoned that if the latter were one of the important physiological functions of Ypk1-dependent stimulation of ceramide synthase , then TORC2 activation would be detrimental to cells in which the Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) mutant is the sole source of this enzyme . To mimic TORC2-stimulated elevation of Ypk1 activity , the constitutively-active Ypk1 ( D242A ) allele ( hereafter referred to as Ypk1* ) was expressed from the YPK1 promoter on a CEN plasmid in either wild-type cells or the Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) mutant . Indeed , compared to the empty vector control , expression of Ypk1* was well tolerated by cells containing wild-type Lac1 and Lag1 , but deleterious to the growth of the Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) mutant cells , whether measured on agar plates ( Figure 7A , top ) or in liquid culture ( Figure 7A , bottom ) . 10 . 7554/eLife . 03779 . 010Figure 7 . Failure of Ypk1 to upregulate ceramide synthase causes LCBP accumulation that triggers autophagy . ( A ) LAC1 LAG1 ( yAM205–A ) or Lac1SSAA Lag1SSAA ( yAM207–B ) were transformed with PYPK1-Ypk1D242A ( shown as Ypk1* ) ( pFR273 ) or empty vector pRS316 ( EV ) . Transformants were grown to exponential phase in synthetic complete medium and then diluted to OD600 = 0 . 1 and grown in microtiter plates ( lower ) or on agar plates ( upper ) . For liquid cultures , each was grown in at least quadruplicate and the error bars indicate the SEM of replicates at each time point . ( B ) Cells from ( A ) were grown to mid-exponential phase in selective synthetic complete media and then harvested . Sphingolipids were extracted and analyzed as described in ‘Materials and methods’ . Values represent the mean of three independent experiments ( each performed in triplicate ) and error bars represent SEM . ( C ) LAC1 LAG1 lcb4Δ ( yAM237 ) or Lac1SSAA Lag1SSAA lcb4Δ ( yAM238–A ) were transformed with Ypk1* ( pFR273 ) and growth experiments performed as in ( A ) . ( D ) Liquid growth assays were performed as in ( A ) for LAC1 LAG1 ( yAM205–A ) , Lac1SSAA Lag1SSAA ( yAM207–B ) , Lac1SSEE Lag1SSEE ( yAM210 ) , LAC1 LAG1 lcb3Δ ( yGT12 ) , Lac1SSAA Lag1SSAA lcb3Δ ( yGT13 ) and Lac1SSEE Lag1SSEE lcb3Δ ( yGT14 ) strains . ( E ) LAC1 LAG1 ( yAM205–A ) or Lac1SSAA Lag1SSAA ( yAM207–B ) or LAC1 LAG1 lcb3Δ ( yGT12 ) strains were transformed with Ypk1* ( pFR273 ) or pRS316 ( EV ) and additionally PTPI1-GFP-Atg8 . Growing cultures treated with vehicle or 2 μg/ml rapamycin for 2 hr and then harvested and whole extracts prepared . Extracts were resolved by SDS-PAGE and blotted with anti-GFP to detect GFP-Atg8 and free GFP arising from GFP-Atg8 autophagic processing and anti-Pgk1 antibody . The blot is representative of triplicate samples and the quantitation of the ratio of free GFP/Pgk1 from these replicates is shown below the blot . ( F ) LAC1 LAG1 atg1Δ ( yAM239–A ) or Lac1SSAA Lag1SSAA atg1Δ ( yAM240–A ) sensitivity to Ypk1* was measured as in ( A ) . ( G ) Cells from ( F ) were grown to mid-exponential phase in selective synthetic complete media and then harvested . Sphingolipids were extracted and analyzed as described in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03779 . 010 Given that accumulation of LCBPs has been shown to be toxic to yeast cell growth ( Kim et al . , 2000 ) , we reasoned that the most likely metabolic perturbation responsible for the observed decrease in growth in the Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) cells expressing Ypk1* was the build-up of LCBs and derived LCBPs . Consistent with this conclusion , we found a reproducible and statistically significant increase in the LCBP level in Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) cells , compared to LAC1+ LAG1+ controls , and a further increase in Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) cells expressing Ypk1* ( Figure 7B ) . If accumulation of LCBPs in Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) co-expressing Ypk1* is indeed responsible for the poor growth , then reduction in LCBP synthesis by elimination of the gene LCB4 , which encodes the major LCB kinase ( Nagiec et al . , 1998 ) , should alleviate the growth inhibition . As expected , introduction of an lcb4Δ null mutation into the Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) mutant suppressed the growth inhibitory effect of Ypk1* ( Figure 7C ) . Conversely , and consistent with toxicity arising from accumulation of LCBPs when Lac1 and Lag1 cannot be stimulated by Ypk1 , we found that the poor growth phenotype of cells lacking the gene ( LCB3 ) encoding the major LCBP phosphatase ( Mandala et al . , 1998 ) , was markedly exacerbated by introduction of the Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) alleles , but not by introduction of the Lac1 ( S23E S24E ) Lag1 ( S23E S24E ) alleles ( Figure 7D ) . In fact , and strikingly , presence of the Lac1 ( S23E S24E ) Lag1 ( S23E S24E ) allele afforded nearly complete rescue of the slow-growth phenotype of the lcb3Δ mutation ( Figure 7D ) . Consistent with the toxicity of LCBPs when LCB utilization by ceramide synthase is inefficient , an elo3 mutation , which prevents efficient formation of C26-CoA ( an acyl chain found in yeast ceramides ) , was synthetically lethal with lcb3Δ ( Kobayashi and Nagiec , 2003 ) . Finally , it has been reported that aberrant increases in LCBP level impede growth by triggering autophagy even under nutrient-rich conditions ( Zimmermann et al . , 2013 ) . In agreement with the poor growth arising from LCBP-evoked induction of autophagy , we found , first , that Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) cells expressing Ypk1* exhibited a readily detectable increase in basal GFP-Atg8 processing , comparable to that in lcb3Δ cells ( but , of course , much less than that occurring when cells were treated with the starvation mimetic rapamycin ) ( Figure 7E ) . Second , we found that preventing autophagy by ablating the gene ATG1 , which encodes a protein kinase necessary for induction of autophagophore formation and its elongation ( Papinski and Kraft , 2014 ) , provided substantial rescue of the growth debilitating effect of Ypk1* expression in Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) cells ( Figure 7F , compare to Figure 7A ) . This rescue was not due to an indirect effect of the absence of Atg1 on LCBP level because introduction of the atg1Δ mutation did not prevent the observed hyper-accumulation of LCBP in Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) cells expressing Ypk1* ( Figure 7G ) . Collectively , these results indicate that , aside from stimulating ceramide synthesis per se , another physiologically important role of Ypk1-dependent ceramide synthase activation is , at least in part , to prevent hyper-accumulation of LCBPs and thereby avoid inadvertent induction of autophagy under nutrient-sufficient conditions ( Figure 8 ) . 10 . 7554/eLife . 03779 . 011Figure 8 . TORC2-Ypk1 signaling globally activates sphingolipid synthesis , selectively directs flux toward ceramide metabolites , and prevents LCBP cross-talk to the autophagy pathway . ( A ) Diagram of yeast de novo sphingolipid biosynthesis shown is derived from ( Dickson , 2008 ) . Enzymes are in ovals . Metabolites are in boxes . Increased color intensity indicates level of metabolite increase in response TORC2-Ypk1 activation . TORC2-Ypk1 signaling globally activates de novo sphingolipid biosynthesis via derepression of the SPT complex ( Roelants et al . , 2011; Berchtold et al . , 2012; Sun et al . , 2012 ) , potentially increasing levels of all metabolites . However , Ypk1 also upregulates ceramide synthesis via phosphorylation of Lac1 and Lag1 , thus primarily directing this increased flux towards ceramides and away from LCBs and LCBPs . ( B ) In the absence of Ypk1 mediated ceramide biosynthesis regulation , increased sphingolipid flux raises LCB and LCBP levels . This slows cell growth by activating autophagy . Thus , TORC2-Ypk1 signaling not only activates sphingolipid biosynthesis in response to stress , but also insulates this flux towards ceramides to prevent metabolite mediated crosstalk to the autophagy machinery . DOI: http://dx . doi . org/10 . 7554/eLife . 03779 . 011
Our approach identified Lac1 and Lag1 as potential Ypk1 targets , and our subsequent characterization demonstrated unequivocally that both Lac1 and Lag1 are bona fide Ypk1 substrates in vivo and that their Ypk1-dependent modification is biologically important for optimal modulation of sphingolipid metabolism . These findings validate the ability of our methods for discovery of physiologically relevant protein kinase substrates . Phospho-acceptor site motif pattern-matching alone , although useful in identifying kinase substrates in some cases ( Yaffe et al . , 2001; Manning et al . , 2002; Mah et al . , 2005; Holt et al . , 2007; Linding et al . , 2007; Rennefahrt et al . , 2007; Gwinn et al . , 2008; Hutti et al . , 2009 ) , can yield a large number of false positives . This possibility was a significant concern for us because certain features of the basophilic Ypk1 motif are shared with other protein kinases ( Mok et al . , 2010 ) . As a means to avoid this problem , we devised a novel SDL-based genetic approach to apply as a secondary filter to parse the bioinformatically selected candidates further . Although not the only possible explanation for detecting an SDL hit , one mechanism our genetic method should be able to assess is whether the candidate gene product displays a primary characteristic of a true substrate , namely the ability to compete with other substrates for association with Ypk1 . A genuine Ypk1 substrate should , when highly over-expressed , sequester a large fraction of the available pool of active kinase , and thus impede its actions on its essential cellular targets . In the absence of Ypk2 , over-expression of an authentic Ypk1 substrate should therefore be deleterious for growth when the activity of an analog-sensitive Ypk1 allele is reduced with a selective inhibitor . The fact that 70% of the SDL hits were indeed substrates for Ypk1-mediated phosphorylation in vitro verified that our use of this genetic method as a secondary filter to pick out true substrates from the list of bioinformatically identified candidates was well justified . In this regard , overexpression of Fpk1 ( KD ) , a catalytically-inactive non-essential Ypk1 substrate , yielded a readily detectable SDL phenotype . Thus , using the protocol we devised , SDL may be a generally useful way to identify substrates and perhaps other binding partners of protein kinases , not just those directly connected to any output phenotype being measured . Other genetic approaches have also been useful in identifying physiologically relevant targets of Ypk1 . For example , a transposon insertion that suppressed the growth defect of a ypk1-ts ypk2Δ strain at an otherwise non-permissive temperature initially identifed Orm2 as a potential Ypk1 substrate ( Roelants et al . , 2002 ) , a finding later corroborated by us ( Roelants et al . , 2011 ) and others ( Berchtold et al . , 2012; Liu et al . , 2012; Niles et al . , 2012; Sun et al . , 2012 ) . Similarly , Smp1 , which we confirmed here is a likely Ypk1 substrate , was first identified as a potential Ypk1 target because it was isolated as a dosage suppressor of the temperature-sensitive phenotype of ypk1-ts ypk2Δ cells ( Roelants et al . , 2002 ) . Smp1 is a transcription factor ( de Nadal et al . , 2003 ) that mediates iron toxicity ( Lee et al . , 2012a ) and , similarly , Ypk1 is required for iron toxicity ( Lee et al . , 2012a ) . Thus , TORC2-Ypk1 signaling might be mechanistically coupled to iron metabolism by modulation of Smp1 transcriptional output . Likewise , other situations where chemical genetics can be applied have proven useful in gleaning what aspects of cell function are controlled by a protein kinase . For example , mutagenesis of yeast Tor2 to confer susceptibility to a chemical inhibitor and thereby selectively inhibit TORC2 action has been achieved recently ( Kliegman et al . , 2013 ) . This tool was combined with a coilection of deletion mutants to identify what processes , when eliminated , are especially deleterious to cell growth and survival when TORC2 action ( and presumably Ypk1 activity ) is limiting . This analysis suggested some connection between TORC2 action and the pentose-phosphate pathway ( Kliegman et al . , 2013 ) , in keeping with the growth-promoting roles of both TORC1 and TORC2 and the demand for NADPH in many cellular anabolic reactions . Similarly , use of TOR inhibitors has implicated TORC2-Ypk1 signaling in regulation of actin filament formation that is somehow required for yeast cell survival in response to low levels of DNA damage ( Shimada et al . , 2013 ) . By contrast , the chemical genetic approach in our SDL method is quite different , in that it scores the deleterious effect arising from overexpression of a gene product ( increased protein dosage ) rather than from the total absence of a gene product , when activity of the kinase of interest is limited by inhibitor . Theoretically , under our conditions , we should also have been able to observe synthetic dosage rescue ( ‘SDR’ ) ; however , we found no such examples . In any event , our method independently identified nearly all of the previously known in vivo substrates of Ypk1 , as well as nearly a dozen genuine Ypk1 targets , including Lac1 and Lag1 , that have only been pinpointed by our three-tiered method . Thus , our SDL screening technique provides a complementary approach for identifying substrates of a protein kinase above and beyond those accessible through genetic interactions between the kinase and single-gene deletions or other genetic schemes . Many of the gene products identified by our screen are involved in processes that the TORC2-Ypk1 signaling axis is already known to regulate . For example , Ypk1 regulates glycerol-3-phosphate production via phosphorylation and inhibition of glycerol-3-phosphate dehydrogenase Gpd1 ( Lee et al . , 2012b ) . Interesting , a very likely Ypk1 target that met all of the criteria in our screen is Gpt2 , sn-glycerol-3-phosphate 1-acyltransferase ( Zheng and Zou , 2001 ) , an enzyme that esterifies glycerol-3-phosphate as the first step in glycerolipid formation . In this same regard , as another very likely Ypk1 target we also identified Fps1 , a membrane channel ( aquaglyceroporin ) that regulates efflux of glycerol ( a glycerol-3-phosphate-derived metabolite ) ( Luyten et al . , 1995 ) . These results strengthen the conclusion that TORC2-Ypk1 signaling is intimately involved in modulating the level of the precursor to both glycerophospholipids and the osmolyte glycerol ( Lee et al . , 2012b ) . Ypk1 action has been implicated in regulation of both fluid phase and receptor-mediated endocytosis ( deHart et al . , 2002 ) . In this regard , we identified as a very likely Ypk1 target an endocytic adaptor , the α-arrestin Rod1 , which is necessary for ubiquitinylation-triggered internalization of nutrient permeases ( Lin et al . , 2008; Becuwe et al . , 2012 ) and the pheromone receptor Ste2 ( Alvaro et al . , 2014 ) . Thus , as it does for PM lipids , TORC2-Ypk1 signaling may modulate PM protein composition via this α-arrestin , a possibility we are pursuing . Ypk1 function has also been implicated in regulating production of reactive oxygen species ( ROS ) ( Niles et al . , 2014 ) , but an as yet undefined mechanism . In our screen , we found Ysp2 , a protein that regulates mitochondrial morphology and ROS levels ( Sokolov et al . , 2006 ) , as a very likely Ypk1 substrate , possibly providing insight into the molecular basis of the connection between TORC2-Ypk1 signaling and ROS levels . Similarly , several other prospects that were identified by our screen as very likely Ypk1 substrates remain to be validated . Such candidates include Muk1 , a GEF for yeast Rab 5-type small GTPases ( Vps21 , Ypt52 , and Ypt53 ) involved in vesicle-mediated Golgi body-to-endosome trafficking ( Paulsel et al . , 2013 ) , suggesting that Ypk1 may also control switches that direct the flow of lipids . Muk1 is also intriguing for another potential reason . In Schizosaccharomyces pombe , a Rab 5-like GTPase ( Ryh1 ) and its Muk1-like GEF were identified in a screen for TORC2 activators ( Tatebe et al . , 2010 ) . Thus , if Ypk1-mediated phosphorylation inhibits Muk1 function , it could represent a negative feedback mechanism exerted on TORC2; conversely , if Ypk1-mediated phosphorylation stimulates Muk1 function , it could represent a mechanism for self-reinforcing maintenance of TORC2 activity and , thus , a high level of activated Ypk1 . Clearly , by further investigating the physiological relevance of these and other remaining candidates much new biology may be learned . Indeed , several gene products of totally unknown function , such as Yhr097c and Ynr014w , as well as gene products ( e . g . , Atg21 and Pex31 ) not previous linked to either TORC2 or Ypk1 , if validated , may provide new mechanistic insight into additional cellular processes regulated by TORC2-Ypk1 signaling . As we have demonstrated here , TORC2-dependent Ypk1-mediated phosphorylation of Lac1 and Lag1 stimulates the function of the ceramide synthase complex . Consistent with our findings , a previous study found a consistent decrease in ceramide synthase activity in microsomal fractions isolated from cells in which TORC2 had been inactivated ( and thus Ypk1 activity was presumably reduced ) ( Aronova et al . , 2008 ) , although indirect effects of the loss of TORC2 function on ceramide synthase activity could not be ruled out . Our findings make it clear that the role of TORC2 is to promote the Ypk1-dependent phosphorylation of Lac1 and Lag1 subunits of this enzyme . However , the precise molecular mechanism by which this post-translational modification stimulates this enzyme is still not completely clear . In this regard , it has been shown that mammalian ceramide synthase activity increases upon heterodimerization of different catalytic subunit isoforms ( Laviad et al . , 2012 ) . However , as judged by co-immunoprecipitation , we found no difference in the state of Lac1-Lag1 association between the wild-type proteins and either our Lac1 ( S23A S24A ) Lag1 ( S23A S24A ) or Lac1 ( S23E S24E ) Lag1 ( S23E S24E ) mutants ( data not shown ) . As mentioned in Results , we found no difference in the steady-state level of these same complexes or in their content of Lip1 , a non-catalytic component of the complexes also essential for ceramide synthase activity ( Vallée and Riezman , 2005 ) . Thus , understanding of how phosphorylation of Ser23 and Ser24 in Lac1 and Lag1 stimulate ceramide synthase activity may require detailed structural information , which will be challenging to obtain for these polytopic integral membrane proteins . Although the enzymic steps that carry out sphingolipid biosynthesis have been largely elucidated , much less was known , until recently , about regulation of these enzymes ( Breslow and Weissman , 2010; Breslow , 2013 ) . The first insight came when it was demonstrated ( Roelants et al . , 2011; Berchtold et al . , 2012; Sun et al . , 2012 ) that , in response to PM stresses , including treatment with myriocin and aureobasidin A , TORC2-Ypk1 signaling is activated and alleviates inhibition of the SPT complex by phosphorylating the negative regulatory proteins Orm1 and Orm2 ( Figure 8 ) . As a direct consequence , the rate of de novo production of the LCB precursor to sphingolipids is increased . Although TORC2 signaling had been implicated in promoting synthesis of ceramide , the product of LCB N-acylation ( Aronova et al . , 2008 ) , it was unknown whether that role was simply the result of TORC2-Ypk1-dependent stimulation of SPT function and the resulting increase in LCB supply . As we demonstrated here , Ypk1-mediated phosphorylation of Lac1 and Lag1 also increases in response to both myriocin and aureobasidin A , suggesting that ceramide synthesis per se , and not simply general elevation of LCB levels , is important for allowing the cells to cope with the effects of these antibiotics . Indeed , collectively , the findings we describe here demonstrate unequivocally that , in addition to up-regulation of SPT , TORC2-Ypk1 exerts direct control on the ceramide synthase step of the sphingolipid biosynthetic pathway by phosphorylating and stimulating the function of the Lac1 and Lag1 subunits of the ceramide synthase complex . Moreover , as we also demonstrated , the ceramide synthase reaction represents an important branch point in sphingolipid biosynthesis ( Figures 2B and 8 ) . LCBs produced by the SPT reaction can either be converted to ceramides or become phosphorylated by LCB kinase Lcb4 ( and its paralog Lcb5 ) to form LCBPs ( Nagiec et al . , 1998 ) . Accumulation of LCBPs has been shown to be toxic to yeast cell growth ( Kim et al . , 2000 ) , as we have also confirmed here , at least in large part because , as is now known , these metabolites trigger inappropriate induction of autophagy ( Zimmermann et al . , 2013 ) . Thus , the rate of ceramide production must be properly adjusted to maintain the pool of LCBs and derived LCBPs at a non-deleterious level , in agreement with evidence in yeast and other organisms that ceramides and LCBPs generally play antagonist roles and must be maintained in the proper dynamic balance ( Kobayashi and Nagiec , 2003; Spiegel and Milstien , 2003; Kihara et al . , 2007; Dickson , 2008; Breslow and Weissman , 2010; Bikman and Summers , 2011 ) . Hence , the function we have discovered and described here for TORC2-Ypk1 in stimulating ceramide synthase promotes utilization of the increased LCB generated upon TORC2-Ypk1-mediated up-regulation of SPT . This metabolic control has multiple clear-cut physiological benefits to the cell: ( a ) directing flow in the sphingolipid pathway toward complex sphingolipids to populate the PM barrier; ( b ) reduction of the level of potentially toxic LCBPs; and , ( c ) avoidance of inappropriate induction of autophagy under nutrient-sufficient conditions ( Figure 8 ) . Indeed , our results indicate that a significant role for the coordination exerted by TORC2-Ypk1 between the level of SPT activity and the level of ceramide synthase activity is to prevent metabolic ‘cross-talk’ to the autophagy pathway . If this function of TORC2-Ypk1 signaling is important , then it should be conserved . Indeed , alignments of the primary structures of Lac1 and Lag1 homologs predicted from sequenced fungal genomes , from Saccharomyces sensu stricto species to the very distantly related Ustilago maydis , nearly all contain a basophilic sequence that matches the Ypk1 phospho-acceptor site consensus and is located , as in S . cerevisiae Lac1 and Lag1 , in their N-terminal cytosolic extensions , suggests that Ypk1-related protein kinases may regulate ceramide synthase across essentially all fungal species . Consistent with this suggestion , a genetic interaction between a Ypk1 homolog ( YpkA ) and a Lac1/Lag1-like ceramide synthase component ( BarA1 ) has been reported in Aspergillus nidulans ( Colabardini et al . , 2013 ) . There is also evidence that Orm1 and Orm2 , when phosphorylated at unique sites by protein kinase Npr1 , promotes steps in the sphingolipid pathway that lead to more complex sphingolipids ( Shimobayashi et al . , 2013 ) . In contrast to Ypk1 , which is activated by TORC2 , Npr1 is inhibited by TORC1 ( MacGurn et al . , 2011 ) . Thus , this control mechanism will only be exerted under conditions that inactivate TORC1 , such as amino acid starvation ( Loewith and Hall , 2011 ) , a condition that presumably requires adjustment of both PM lipid and protein composition to maximize the cell's ability to scavenge and assimilate nutrients . Under the same condition , autophagy is induced because , like Npr1 , TORC1 also negatively regulates the autophagy-inducing protein kinase Atg1-Atg13 complex ( Alers et al . , 2014 ) . Conversely , under nutrient sufficient conditions , TORC1 is active , and phosphorylates and stimulates protein kinase Sch9 . Interestingly , Sch9 action should act in concert with TORC2-Ypk1 signaling to help keep the levels of LCBs and LCBPs low and ceramides high . This is likely because Sch9 promotes transcriptional repression of genes ( YDC1 and YPC1 ) that encode ceramidases and inhibits a phosphosphingolipid phospholipase C ( Isc1 ) that hydrolyzes complex sphingolipids ( Swinnen et al . , 2014 ) . Such complex multi-component controls may be a general feature of signaling modalities that interface with biosynthetic pathways that have intermediates , like LCBPs , that are not inocuous , but are themselves bioactive metabolites . Prior studies had suggested that calcineurin negatively regulates sphingolipid production via effects on the function of the ancillary TORC2 subunits , Slm1 and Slm2 ( Bultynck et al . , 2006; Mulet et al . , 2006; Tabuchi et al . , 2006; Daquinag et al . , 2007 ) , although the molecular connection between Slm1 and Slm2 and sphingolipid biosynthesis was unclear . Subsequently , it was observed that , in cells lacking the regulatory subunit ( Cnb1 ) , there was an increase in C26-containing ceramides , suggesting that calcineurin somehow antagonizes TORC2-dependent signaling ( Aronova et al . , 2008 ) . We found that calcineurin negatively regulates ceramide synthesis by directing the dephosphorylation of Lac1 and Lag1 . This Ca2+-activated calcineurin-dependent dephosphorylation occurred even in cells expressing a TORC2-independent constitutively-active Ypk1 allele . Moreover , we showed here that calcineurin does not affect Pkh1- ( and Pkh2- ) mediated phosphorylation of the activation loop of Ypk1 , and we demonstrated previously that presence or absence of calcineurin does not alter TORC2-mediated phosphorylation of Ypk1 or cause any substantial change in Ypk1 specific activity ( Roelants et al . , 2011 ) . Thus , direct down-modulation of either TORC2 or Ypk1 by calcineurin cannot account for the negative regulation it exerts on sphingolipid biosynthesis . What our findings now make clear is that calcineurin negatively regulates the sphingolipid pathway at the level of ceramide synthesis , at least in large part , by direct dephosphorylation of the stimulatory phosphorylations in the ceramide synthase subunits Lac1 and Lag1 that are installed by TORC2-Ypk1 signaling . Calcineurin recognizes substrates via a docking motif ( PxIxIT or variants thereof ) , typically also accompanied quite a distance upstream by a secondary docking site ( LxVP ) ( Roy and Cyert , 2009 ) . However , these sites can be quite degenerate; for example , the more hydrophobic variant PVIVIT is much more potent in recruiting calcineurin when it is used to replace the native ‘PxIxIT’ sequences in either the transcription factor Crz1 ( PIISIQ ) ( Roy et al . , 2007 ) or the endocytic adaptor Aly1 ( PILKIN ) ( O'Donnell et al . , 2013 ) . In both Lac1 and Lag1 , there is a similar hydrophobic sequence located at the identical position in both proteins ( 355PIVFVL360 ) . Whether this or any other degenerate match represents a calcineurin-binding site remains to be determined . In conclusion , our screening approach has provided a number of new insights into how the TORC2-Ypk1 axis modulates sphingolipid homeostasis and has uncovered a significant number of candidate substrates that will very likely shed further light on other aspects of cellular function that are regulated by TORC2-Ypk1 signaling .
All S . cerevisiae strains used in this study are listed in Table 2 . Strains were constructed using standard yeast genetic manipulations ( Burke et al . , 2005 ) . For all strains constructed , integration of the desired DNA fragment into the correct genomic loci was confirmed by PCR using an oligonucleotide complementary to the integrated DNA fragment and an oligonucleotide complementary to genomic sequence at least 150 bases away from the integration site . 10 . 7554/eLife . 03779 . 012Table 2 . Saccharomyces cerevisiae strains used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 03779 . 012StrainGenotypeSource/referenceBY4741MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0Research Genetics , Inc . BY4742MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0Research Genetics , Inc . yAM135-ABY4741 Ypk1 ( L424A ) ::URA3-ypk2Δ::KanMX4This studyJTY6142BY4741 ypk1Δ::KanMX4Research Genetics , Inc . yAM120-ABY4741 ypk2Δ::KanMX4This studyyAM159-ABY4741 3xFLAG-Lag1::LEU2This studyyAM163-ABY4741 3xFLAG-Lag1 ( S23A S24A ) ::LEU2This studyyAM165-ABY4742 3xHA-Lac1::HIS3This studyyAM166-ABY4742 3xHA-Lac1 ( S23A S24A ) ::HIS3This studyJTY5574BY4741 cna1Δ::KanMX4 cna2Δ::KanMX4M . S . Cyert , Stanford Univ . YDB379BY4741 Ypk1-3xFLAG::natNT2J . S . Weissman , Univ . of California , San FranciscoyAM205-ABY4742 Lac1::LEU2 Lag1::LEU2This studyyAM207-BBY4742 Lac1 ( S23A S24A ) ::LEU2 Lag1 ( S23A S24A ) ::LEU2This studyyAM210BY4742 Lac1 ( S23E S24E ) ::LEU2 Lag1 ( S23E S24E ) ::LEU2This studyyGT12BY4742 LYS2+ Lac1::LEU2 Lag1::LEU2 lcb3Δ::natNT2This studyyGT13BY4742 LYS2+ Lac1 ( S23A S24A ) ::LEU2 Lag1 ( S23A S24A ) ::LEU2 lcb3Δ::natNT2This studyyGT14BY4742 LYS2+ Lac1 ( S23E S24E ) ::LEU2 Lag1 ( S23E S24E ) ::LEU2 lcb3Δ::natNT2This studyyAM168BY4741 3xHA-Lac1::HIS3 3xFLAG-Lag1::LEU2This studyyAM184BY4741 3xHA-Lac1 ( S23A S24A ) ::HIS3 3xFLAG-Lag1 ( S23A S24A ) ::LEU2This studyyAM192-ABY4741 MET15+ 3xHA-Lac1 ( S23E S24E ) ::HIS3 3xFLAG-Lag1 ( S23E S24E ) ::LEU2This studyyKL4BY4741 TOR2+::HygrKristin Leskoske , this labyKL5BY4741 Tor2 ( L2178A ) ::HygrKristin Leskoske , this lab All plasmids used in this study ( except the library of PGAL1 based overexpression plasmids for synthetic dosage lethality [SDL] screening ) are listed in Table 3 . All plasmids were constructed and maintained in E . coli using standard laboratory methods ( Green and Sambrook , 2012 ) . For SDL screening , the entire open reading frame of each predicted and known Ypk1 substrate was amplified by PCR from BY4741 genomic DNA and ligated into the multiple cloning site of YCpLG ( CEN , PGAL1 , LEU2 ) , generating a vector allowing galactose inducible overexpression of each substrate . All constructs generated in this study were confirmed by sequence analysis covering all promoter and coding regions in the construct . 10 . 7554/eLife . 03779 . 013Table 3 . Plasmids used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 03779 . 013PlasmidDescriptionSource/referencepGEX6P-1GST tag , bacterial expression vectorGE Healthcare , Inc . pGEX4T-1GST tag , bacterial expression vectorGE Healthcare , Inc . YCpLGCEN , LEU2 , PGAL1 vector ( Bardwell et al . , 1998 ) BG18052 µm , URA3 , PGAL1 , C-terminal tandem affinity ( TAP ) tag vectorOpen Biosystems , Inc . pRS313CEN , HIS3 , vector ( Sikorski and Hieter , 1989 ) pRS316CEN , URA3 , vector ( Sikorski and Hieter , 1989 ) pRS416CEN , URA3 , vector ( Sikorski and Hieter , 1989 ) pBC111CEN , LEU2 , vector ( Iida et al . , 2007 ) CHp282pRS416 PMET25-GFPChau Huynh , this laboratorypLB215pRS416 PMET25-Ypk1 ( Niles et al . , 2012 ) pAX53pRS416 PMET25-Ypk1 ( K376A ) This studypAX50BG1805 Ypk1 ( L424A ) This studypFR203pGEX4T-1 Orm1 ( 1-85 ) ( Roelants et al . , 2011 ) pBT6pGEX6P-1 Fps1 ( 1-255 ) This studypBT7pGEX6P-1 Fps1 ( 531-669 ) This studypBT12pGEX6P-1 Smp1This studypAX55pGEX6P-1 Lcb3 ( 1-79 ) This studypAX56pGEX6P-1 Cdc1 ( 1-41 ) This studypAX58pGEX6P-1 Her1 ( 1-224 ) This studypAX59pGEX6P-1 Rts3This studypAX62pGEX6P-1 Fkh1This studypAX63pGEX6P-1 Yhp1This studypAX66pGEX6P-1 YNR014WThis studypAX67pGEX6P-1 YHR097CThis studypAX94pGEX6P-1 Mds3 ( 545-1016 ) This studypAX131pGEX4T-1 Lac1 ( 1-76 ) This studypAX132pGEX4T-1 Lac1 ( 1-76 ) ( S23A S24A ) This studypFR291pGEX4T-1 Lag1 ( 1-80 ) This studypAX133pGEX4T-1 Lag1 ( 1-80 ) ( S23A S24A ) This studypAX134pGEX6P-1 Muk1 ( 1-305 ) This studypAX215pGEX6P-1 Cyk3This studypAX223pGEX6P-1 Gpt2 ( 1-35 ) This studypAX224pGEX6P-1 Gpt2 ( 570-743 ) This studypAX225pGEX6P-1 Bre5This studypAX226pGEX6P-1 Npr1 ( 1-437 ) This studypAX227pGEX6P-1 Pal1This studypAX228pGEX6P-1 Ysp2 ( 97-665 ) This studypAX229pGEX6P-1 Ysp2 ( 1072-1282 ) This studypAX230pGEX6P-1 Atg21This studypAX231pGEX6P-1 Pex31 ( 250-462 ) This studypAX136pRS313 PLAC1-3xHA-Lac1This studypFR273pRS316 PYPK1-Ypk1 ( D242A ) ( Roelants et al . , 2011 ) pAX250pRS313 PTPI1-GFP-Atg8This studypBCT-CCH1HpBC111 PTDH3-Cch1 ( Iida et al . , 2007 ) A Nx20 position weight matrix defining Ypk1 phosphoacceptor site specificity was made merging previously published data sets defining Ypk1 primary sequence specificity ( Casamayor et al . , 1999; Mok et al . , 2010 ) . This position weight matrix was then used with MOTIPS ( Lam et al . , 2010 ) to identify proteins with likely phosphorylated occurrences of this motif . Yeastmine ( Balakrishnan et al . , 2012 ) was used to identify all S . cerevisiae genes that showed genetic interactions with Ypk1 , Ypk2 , any component of the sphingolipid biosynthetic pathway or any component of known Ypk1 regulators ( TORC2 and PP2A ) . Genes that showed significant growth phenotypes on myriocin , aureobasidin A and caspifungin ( all compounds that cause severe growth phenotypes in ypk1Δ strains ) were identified from the literature ( Hillenmeyer et al . , 2008 ) . Lastly , MOTIPS predicted Ypk1 phosphorylation sites were compared to Phosphogrid ( Sadowski et al . , 2013 ) to identify those sites known to be phosphorylated in vivo . To be considered a potential Ypk1 substrate in this study , a protein had to have a myriocin , aureobasidin or caspofungin phenotype or a Yeastmine identified genetic interaction and: ( a ) 4 or more MOTIPS predicted sites , ( b ) 3 sites with at least one above a MOTIPS likelihood score of 0 . 7 or a Phosphogrid identified site or ( c ) 1–2 sites with a MOTIPS likelihood score of 0 . 7 and a Phosphogrid identified site . We also considered for further analysis a limited number of proteins that did not meet these criteria , but which still contained Ypk1 motifs . These are indicated in Table 3 . For SDL screening , yAM135-A and BY4741 strains were both transformed with each SDL plasmid . Transformants were then cultured overnight in SC media containing 2% raffinose and 0 . 2% sucrose . 10-fold serial dilutions of overnight cultures starting from OD600 = 1 . 0 were then made in sterile water and spotted onto SC solid media with 2% galactose ( to induce protein expression ) or 2% dextrose ( no protein expression ) . These plates also contained 1:1000 DMSO , 1 μM 3-MB-PP1 or 2 μM 3-MB-PP1 to inhibit to varying degrees Ypk1-as kinase activity in yAM135-A . Serially spotted cultures were allowed to grow in the dark at 30°C for 3 days . Plates were then scanned on a flatbed scanner and growth phenotypes were assessed and scored . For all SDL overexpression constructs that caused toxicity upon overexpression in WT BY4741 strains , we also performed Ypk1 dosage rescue growth assays . PMET25-Ypk1 or PMET25-Ypk1 ( K367A ) ( kinase dead ) plasmids were co-transformed into BY4741 with the toxic SDL plasmid . Transformants were then cultured overnight in SC media containing 2% raffinose and 0 . 2% sucrose . 10-fold serial dilutions of overnight cultures starting from OD600 = 1 . 0 were then made in sterile water and spotted onto SC solid media with 2% galactose ( to induce protein expression ) or 2% dextrose ( no protein expression ) , in the presence of absence of methionine to drive Ypk1 overexpression . Serially spotted cultures were allowed to grow in the dark at 30°C for 3 days . Plates were then scanned on a flatbed scanner and growth phenotypes were assessed to determine if Ypk1 overexpression could rescue toxicity of overexpression of the SDL protein . For broth growth assays , exponential phase cultures growing in rich YP ( Burke et al . , 2005 ) media with 2% dextrose were diluted to OD600 = 0 . 1 . 100 μl of each culture was placed in a well in a 96 well plate with vehicle or drug at the indicated concentration . Cultures were grown with orbital shaking at 30°C in a Tecan Infinite M-1000 PRO plate reader ( Tecan Systems Inc . , San Jose , CA ) for 24 hr . Absorbance measurements were taken every 15 min . Absorbance values were converted to OD600 values using a standard curve of absorbance values of cultures at known OD600 taken on the same plate reader . To purify Ypk1-as kinase , pAX50 ( 2 μ , PGAL1-Ypk1-as-TAP , URA3 ) transformed yAM135-A yeasts were diluted to OD600 = 0 . 125 in 3 l of SC 2% raffinose 0 . 2% sucrose and grown shaking at 30°C to mid-exponential phase . Expression was induced for ∼18 hr by the addition of 2% galactose . Cells were harvested by centrifugation and frozen in liquid nitrogen . The cells were then lysed cryogenically using Mixer Mill MM301 ( Retsch , Düsseldorf , Germany ) . The lysate was resuspended at 2 ml/g in TAP-B ( 50 mM Tris-Cl pH 7 . 5 , 200 mM NaCl , 1 . 5 mM MgOAc , 1 mM DTT , 2 mM NaVO4 , 10 mM NaF , 10 mM Na-PPi , 10 mM β-glycerol phosphate , 1× complete protease inhibitor [Roche , Basel , Switzerland] ) . The lysate was clarified by centrifugation at 15×kg for 20 min . Clarified lysate was then further centrifuged at 100×kg for 1 hr and then brought to 0 . 15% NP-40 using 10% NP-40 detergent stock . Ypk1-as-TAP fusion was then affinity purified from the lysate using IgG-agarose resin ( GE Healthcare ) . The resin was extensively washed with Protease 3C Buffer ( 50 mM Tris-Cl pH 7 . 5 , 200 mM NaCl , 1 . 5 mM MgOAc , 1 mM DTT , 0 . 01% NP-40 , 10% Glycerol , 2 mM NaVO4 , 10 mM NaF , 10 mM Na-PPi , 10 mM β-glycerol phosphate ) and then resuspended in 1 ml Protease 3C Buffer . Ypk1-as was eluted by the addition of 80 U Prescission Protease ( GE Healthcare , Little Chalfont , UK ) for 5 hr at 4°C . Protease 3C was removed by the addition of glutathione-agarose ( GE Healthcare ) . Putative Ypk1 substrates were expressed as N terminal GST fusions in BL21 E . coli . 1 l cultures were grown at 37°C to mid-exponential phase and then induced at room temperature for 4 hr with 0 . 5 mM IPTG . Cells were harvested by centrifugation and the fusion proteins were purified by affinity chromatography using glutathione-agarose and standard procedures . For kinase assays 0 . 25 μg of Ypk1-as kinase was incubated with purified GST-substrate fusion in Kinase Assay Buffer ( 50 mM Tris-Cl pH 7 . 5 , 200 mM NaCl , 10 mM MgCl2 , 0 . 1 mM EDTA ) with 2 μCi [γ-32P]ATP at 30°C in the presence of absence of 10 μM 3-MB-PP1 for 30 min . Reactions were terminated by the addition of SDS/PAGE sample buffer containing 6% SDS followed by boiling for 5 min . Labeled proteins were resolved by SDS/PAGE and analyzed Coomassie blue staining and autoradiography with Phosphorimager plates ( Molecular Dynamics , Sunnyvale , CA ) on a Typhoon imaging system ( GE Healthcare ) . Cell extracts were made by alkaline lysis followed by trichloroacetic acid precipitation as previously described ( Westfall et al . , 2008 ) . To resolve Lag1 and Lac1 phosphorylated species , 15 μl of TCA extract was resolved by SDS-PAGE ( 8% acrylamide , 35 µM Phos-tag [Wako Chemicals USA , Inc . , Richmond , VA] , 35 µM MnCl2 at 160 V ) . The gel was then transferred to nitrocellulose and incubated with primary antibody in Odyssey buffer ( Licor Biosciences Inc . , Lincoln , NE ) , washed , and incubated with IRDye680-conjugated anti-mouse IgG ( Licor Biosciences ) in Odyssey buffer with 0 . 1% Tween-20 and 0 . 02% SDS . Blots were imaged using an Odyssey infrared scanner ( Licor Biosciences ) . Primary antibodies and dilutions used in this study were: 1:1000 mouse anti-HA ( Covance Inc . , Princeton , NJ ) , 1:10000 mouse anti-FLAG ( Sigma–Aldrich , St . Louis , MO ) , 1:500 mouse anti-GFP ( Roche ) , 1:500 rabbit anti-pSGK ( T256 ) ( to recognize Ypk1 phosphorylated at T504 ) ( Santa Cruz Biotechnology , Dallas , TX ) and 1:10000 rabbit anti-Pgk1 ( our laboratory ) . For phosphatase treatment of cell extracts , TCA extracts were made as above and the precipitated proteins were solubilized in 100 µl Solubilization buffer ( 50 mM Tris-Cl pH 8 . 0 , 150 mM NaCl , 2% β-mercaptoethanol , 2% SDS ) . These extracts were then diluted with 900 µl CIP Dilution buffer ( 50 mM Tris-Cl pH 8 . 0 , 150 mM NaCl , 11 . 1 mM MgCl2 ) . 150 U calf intestinal phosphatase ( New England Biolabs , Ipswich , MA ) was then added and incubated for 2 hr at 37°C . Proteins were then TCA precipitated again and resolved by SDS-PAGE as above . Complex sphingolipids were analyzed by thin layer chromatography . Cultures of strains in mid-exponential phase were adjusted to OD600 = 1 . 0 and 2 ml cultures were labeled with 100 µCi of [32P] PO4–3 and cells allowed to grow for 3 hr . Lipids were extracted and resolved as previously described ( Hanson and Lester , 1980; Momoi et al . , 2004 ) with minor modifications . The cell pellet was washed twice with 2 ml water and treated with 5% trichloroacetic acid for 20 min on ice . Pellets were extracted twice with 0 . 75 ml of ethanol/water/diethyl-ether/pyridine/NH4OH ( 15:15:5:1:0 . 018 ) at 60°C for 1 hr . Glycerophospholipids in the extract were hydrolyzed by treating with 0 . 1 M monomethylamine at 50°C for 1 hr after which the base was neutralized by addition of 12 µl of glacial acetic acid . Lipids were extracted with 1 ml chloroform , 0 . 5 ml methanol and phases separated with addition of 1 ml water . For some samples 1 ml of 4 N NaCl was used , in order to facilitate the separation of the aqueous and organic phases . The organic layer was dried under vacuum and resuspended in 50 µl of chloroform/methanol/water ( 16:16:5 ) and resolved on a silica gel TLC plate with chloform/methanol/4 . 2 N NH4OH ( 9:7:2 ) . Radioactivity on the TLC plate was visualized with a Phosphorimager screen and Typhoon imaging system . LCBs and LCB-1Ps levels were monitored by liquid chromatography-mass spectrometry- Overnight cultures were adjusted to OD600 = 0 . 2 and allowed to grow to an OD600 = 1 . 0 . Cell pellets from 10 ml of this culture equivalent to 10 OD600 units was used for analysis . C17-sphingosine ( Avanti Polar Lipids , Alabaster , AL ) , ( 5 nmol ) was added to all samples as an internal standard . Lipids were extracted as described above and the final dried lipid extract was dissolved in methanol/water/formic acid ( 79:20:1 ) , centrifuged to remove insoluble material and an aliquot of this material was injected into the HPLC for analysis . Lipid extracts were analyzed using an Agilent 1200 liquid chromatograph ( LC; Santa Clara , CA ) that was connected in-line with an LTQ Orbitrap XL mass spectrometer equipped with an electrospray ionization source ( ESI; Thermo Fisher Scientific , Waltham , MA ) . The LC was equipped with a C4 analytical column ( Viva C4 , 150 mm length × 1 . 0 mm inner diameter , 5 µm particles , 300 Å pores , Restek , Bellefonte , PA ) and a 100 µl sample loop . Solvent A was 99 . 8% water/0 . 2% formic acid and solvent B was 99 . 8% methanol/0 . 2% formic acid ( vol/vol ) . Solvents A and B both contained 5 mM ammonium formate . The sample injection volume was 85 µl ( partial loop ) . The elution program consisted of isocratic flow at 30% B for 2 min , a linear gradient to 65% B over 0 . 1 min , a linear gradient to 100% B over 4 . 9 min , isocratic flow at 100% B for 4 min , a linear gradient to 30% B over 0 . 1 min , and isocratic flow at 30% B for 18 . 9 min , at a flow rate of 170 µl/min . The column and sample compartments were maintained at 40°C and 4°C , respectively . The injection needle was rinsed with a 1:1 methanol/water ( vol/vol ) solution after each injection . The column exit was connected to the ESI probe of the mass spectrometer using PEEK tubing ( 0 . 005ʺ inner diameter × 1/16ʺ outer diameter , Agilent ) . Mass spectra were acquired in the positive ion mode over the range m/z = 250 to 1200 using the Orbitrap mass analyzer , in profile format , with a mass resolution setting of 100 , 000 ( at m/z = 400 , measured at full width at half-maximum peak height ) . In the data-dependent mode , the five most intense ions exceeding an intensity threshold of 30 , 000 counts were selected from each full-scan mass spectrum for tandem mass spectrometry ( MS/MS ) analysis using pulsed-Q dissociation ( PQD ) . MS/MS spectra were acquired in the positive ion mode using the linear ion trap , in centroid format , with the following parameters: isolation width 3 m/z units , normalized collision energy 40% , and default charge state 1+ . A parent mass list was used to preferentially select ions of interest for targeted MS/MS analysis . To avoid the occurrence of redundant MS/MS measurements , real-time dynamic exclusion was enabled to preclude re-selection of previously analyzed precursor ions , using the following parameters: repeat count 2 , repeat duration 30 s , exclusion list size 500 , exclusion duration 180 s , and exclusion width 0 . 1 m/z unit . Data were analyzed using Xcalibur software ( version 2 . 0 . 7 SP1 , Thermo ) and LIPID MAPS Online Tools ( Fahy et al . , 2007 ) . Exact masses of precursor ions in positive ion ( M + H+ ) state obtained from LIPID MAPS Online Tools were as follows- phytosphingosine- 318 . 3003 , dihydrosphingosine-302 . 3053 , phytosphingosine-1 phosphate-398 . 2666 and dihydrosphingosine-1 phosphate- 382 . 2717 . The ions were further confirmed by tandem MS and identification of the fragments; product ions for the phytosphingosine headgroup was 282 . 3 and dihydrosphigosine headgroup was 266 . 4 . Yeast expressing 3xFLAG-Lag1 ( wild-type and phospho-site mutations ) were grown at 30°C to mid-exponential phase and microsomes were prepared as described previously ( Schorling et al . , 2001 ) and resuspended in B88 buffer ( 20 mM HEPES-KOH pH 6 . 8 , 150 mM KAc , 5 mM MgOAc , and 250 mM sorbitol ) . Ceramide synthase was then immunopurified from these microsomes using anti-FLAG agarose ( Sigma ) as described previously ( Vallée and Riezman , 2005 ) . Prior to assembling ceramide synthase reactions , a small alioquot of each ( 15 μl ) immunoprecipitate was taken and proteins were resolved by SDS-PAGE and immunoblotted with anti-FLAG to determine the relative amount of immunopurified ceramide synthase in each reaction . Recovered ceramide levels for each reaction were normalized to the amount of ceramide synthase determined by this procedure . Ceramide synthesis reaction was assembled in B88 buffer with 5 mg/ml BSA , in a total reaction volume of 0 . 1 ml containing , 47 . 5 µl anti-FLAG agarose immunoprecipitates , 50 µM phytosphingosine and 0 . 1 mM of steroyl ( C18 ) - CoA . Reactions were incubated at 30°C for 1 hr . Reactions were terminated by addition of 0 . 4 ml methanol/CHCl3 ( 2:1 ) and further extracted with 0 . 4 ml CHCl3 and 0 . 4 ml water . The CHCl3 phase was separated by centrifugation and removed and dried under vacuum . The dried lipids resuspended in methanol/water/formic acid ( 79/20/1 ) and an aliquot analyzed by LC-MS as described above . The product from the reaction , C18-phytoceramide was identified by the exact mass of its precursor ion at 584 . 5612 and its product ion upon fragmentation at 282 . 3 . Additionally , authentic C18-phytoceramide ( Matreya , LLC , Pleasant Gap , PA ) standard was used to obtain a standard curve in order to establish that values from ceramide synthase reaction were in the linear range of estimation . | Cells are enclosed by a plasma membrane that separates and protects each cell from its environment . These membranes are made of a variety of proteins and fatty molecules called lipids , which are carefully organized throughout the membrane . When cells experience stresses such as heat or excessive pressure , the plasma membrane changes to help protect the cell . In particular , more of a group of lipids called sphingolipids are incorporated into the membrane under stress conditions . In yeast cells , a protein called Ypk1 plays an important role in protecting the cell from stress . Ypk1 controls the activity of a number of proteins that are responsible for balancing the amounts of different types of lipids in cell membranes . The combined action of these Ypk1-dependent proteins leads to the remodelling of the cell membrane to protect against stress . While several proteins that work with Ypk1 are known , some of the changes that serve to protect the plasma membrane cannot be explained by the action of these proteins alone . To provide a more comprehensive picture of how Ypk1 helps cells to respond to changes in the environment , Muir et al . developed a new approach that combines biochemical , genetic and bioinformatics techniques to survey the yeast genome for proteins that could be Ypk1 targets . Muir et al . first produced a list of potential candidate proteins by searching for proteins with features similar to known Ypk1 targets , and then considered those that are known to be involved in processes that also involve Ypk1 . To filter the potential targets further , Muir et al . performed experiments in yeast cells to see which proteins prevented normal cell growth if they were over-produced . Further experiments investigating which of these proteins interact with Ypk1 when purified identified 12 new proteins that are most likely targets of the Ypk1 protein . Two of these newly identified Ypk1 target proteins form part of an enzyme complex called ceramide synthase , which produces a family of waxy lipid molecules from which more complex sphingolipids are built . Muir et al . discovered that during stress , Ypk1 enhances the activity of the ceramide synthase enzyme , which increases lipid production and the amount of sphingolipid deposited in the cell membrane . If this process is interrupted at any stage , cells struggle to survive under stress conditions . The other candidate proteins identified by Muir et al . remain to be validated and characterized as Ypk1 targets . Nevertheless , the techniques used have conclusively identified some new Ypk1 targets and could also be applied to similar searches for proteins targeted in other biological processes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2014 | TORC2-dependent protein kinase Ypk1 phosphorylates ceramide synthase to stimulate synthesis of complex sphingolipids |
Asynchronous transmission plays a prominent role at certain synapses but lacks the mechanistic insights of its synchronous counterpart . The current view posits that triggering of asynchronous release during repetitive stimulation involves expansion of the same calcium domains underlying synchronous transmission . In this study , live imaging and paired patch clamp recording at the zebrafish neuromuscular synapse reveal contributions by spatially distinct calcium sources . Synchronous release is tied to calcium entry into synaptic boutons via P/Q type calcium channels , whereas asynchronous release is boosted by a propagating intracellular calcium source initiated at off-synaptic locations in the axon and axonal branch points . This secondary calcium source fully accounts for the persistence following termination of the stimulus and sensitivity to slow calcium buffers reported for asynchronous release . The neuromuscular junction and CNS neurons share these features , raising the possibility that secondary calcium sources are common among synapses with prominent asynchronous release .
Physiological studies have pointed increasingly to a central role played by asynchronous release in mediating synaptic transmission ( Goda and Stevens , 1994; Lu and Trussell , 2000; Hefft and Jonas , 2005; Iremonger and Bains , 2007; Best and Regehr , 2009 ) . At most synapses , the asynchronous contribution to release is smaller than the synchronous component , but becomes more prominent with repetitive stimulation ( Atluri and Regehr , 1998; Lu and Trussell , 2000; Hefft and Jonas , 2005 ) . Little is known about the functional significance , but the most widely held idea is that it provides persistent transmitter release under conditions where phase locking to the action potential is not required ( Atluri and Regehr , 1998; Hefft and Jonas , 2005; Best and Regehr , 2009 ) . At the zebrafish neuromuscular junction ( NMJ ) , asynchronous release may augment release during bouts of prolonged swimming , where release probability may be severely compromised . Synchronous and asynchronous modes of release are thought to arise from different calcium dynamics surrounding the presynaptic release zones . According to this idea , synchronous release is triggered by highly localized calcium transients resulting from the opening of calcium channels near the active zones ( Adler et al . , 1991; Stanley , 1993; Neher , 1998; Meinrenken et al . , 2002; Augustine et al . , 2003; Eggermann et al . , 2012 ) , whereas a more slowly decaying component of intracellular calcium is proposed to account for the asynchronous release ( Rahamimoff and Yaari , 1973; Goda and Stevens , 1994; Cummings et al . , 1996; Atluri and Regehr , 1998; Chen and Regehr , 1999 ) . The slow accumulation of calcium during repeated stimulation can potentially account for both the delayed onset and persistence ( Lu and Trussell , 2000; Hefft and Jonas , 2005 ) . Much of the evidence for this idea rests on the observation that the slow calcium buffer EGTA can block asynchronous release while leaving synchronous release intact ( Adler et al . , 1991; Cummings et al . , 1996; Atluri and Regehr , 1998; Lu and Trussell , 2000 ) . The fact that the faster calcium buffer BAPTA is required to inhibit synchronous release has been interpreted to represent a more restricted calcium domain that is in close vicinity to the calcium sensors underlying exocytosis ( Neher , 1998; Eggermann et al . , 2012 ) . There has been little consideration of alternative secondary sources of calcium , despite evidence for their involvement in synaptic transmission ( Collin et al . , 2005; Berridge , 2006 ) . In one recent case however , an unexpected source of calcium for asynchronous release was reported in the form of an unusual voltage dependent calcium channel type that provides persistent calcium entry ( Few et al . , 2012 ) . Our investigation into possible separate sources of calcium in zebrafish motor neurons was prompted by our observation that a block of calcium entry through the presynaptic P/Q calcium channels fully inhibited synchronous release , leaving asynchronous release intact . Zebrafish NMJ offers a unique opportunity to explore the temporal and spatial relationships between calcium entry and the two release modes through combining paired recording and live calcium imaging . We now present evidence for a source of off-synapse calcium that triggers the initiation of asynchronous release . This novel calcium source for asynchronous release predicts many of the central features for this mode including sensitivity to calcium buffers , persistence following termination of the stimulus , and non-phase locking to the presynaptic action potential .
Paired patch clamp recordings from the caudal primary motor neuron ( CaP ) and target fast skeletal muscle showed a stereotypic transition from exclusively synchronous to principally asynchronous transmission when stimulated at frequencies greater than 20 Hz ( Figure 1A , B; Wen et al . , 2010 ) . Behaviorally evoked contractures of zebrafish axial muscle correspond to frequencies between 20 Hz and 100 Hz , so we continue to use the latter stimulus frequency as the benchmark for our studies . At 100 Hz , greater than 95% of the synaptic responses were phase locked to the presynaptic action potential during the first second of stimulation ( Figure 1A1 , B; Wen et al . , 2010 ) . The onset of asynchronous release occurred after the first second of stimulation , displaying a time-dependent increase in overall contribution during the ensuing stimulation ( Figure 1A2 , B ) . The release was quantitated as charge transfer by integrating the EPCs for each consecutive second of the stimulation , and synchronous and asynchronous events were separated on the basis of their timing to the action potential ( Wen et al . , 2010 ) . The release associated with synchronous vs asynchronous events showed a time-dependent transition , and overall each of the two modes accounted for approximately half of the total synaptic transmission ( Figure 1B; Wen et al . , 2010 ) . The amplitudes of the asynchronous events were indistinguishable from the spontaneous synaptic events measured in the absence of stimulation ( Figure 1C ) , consistent with each representing individual quanta . The synchronous release is strictly dependent on P/Q-type calcium channel function ( Figure 1D; Wen et al . , 2013 ) . Inhibiting P/Q calcium channels with 1 µM ω-conotoxin GVIA nearly abolished synchronous release ( Figure 1D1 , E ) . Unexpectedly , asynchronous release remained intact in the ω-conotoxin GVIA treated fish , but with a greatly delayed onset compared to the control fish ( Figure 1D2 , E ) . Quantitatively , over 85% of release seen in ω-conotoxin GVIA fish was associated with the asynchronous mode , and this likely represents an underestimate because of the resolution of our analysis ( Figure 1E ) . Similar to the control , the amplitude of the asynchronous events was indistinguishable from spontaneous miniature events ( Figure 1F ) . The motility mutant line tb204a has greatly compromised P/Q calcium channel function but is not a complete null ( Wen et al . , 2013 ) . Accordingly , the synchronous release was reduced but not eliminated completely ( Figure 1G1 ) , leaving asynchronous release intact ( Figure 1G2 ) . Quantitation of the time-dependent contributions showed both reduced synchronous release and delayed onset of asynchronous release for tb204a compared to control ( Figure 1H ) . Once again , the amplitude of the late asynchronous event class was indistinguishable from the spontaneous events measured in the absence of stimulation ( Figure 1I ) , as well as those asynchronous events recorded from control ( Figure 1C ) and ω-conotoxin GVIA-treated ( Figure 1F ) fish . 10 . 7554/eLife . 01206 . 003Figure 1 . Asynchronous synaptic transmission remains intact in the P/Q calcium channel mutant tb204a and following treatment of wild-type fish with ω-conotoxin GVIA . ( A–C ) A representative paired recording from untreated wild-type fish . ( A ) Voltage clamp traces of EPCs in response to 20 s , 100 Hz stimulation of the motor neuron . Expanded views with both action potentials and associated postsynaptic EPCs showing early synchronous ( A1 ) and mixed synchronous and asynchronous release at the peak of release ( A2 ) . ( B ) Quantitation of the time-dependence of synchronous ( blue ) , asynchronous ( red ) and total ( black ) synaptic charge integrals determined using the methods described in Wen et al . ( 2010 ) . ( C ) Comparison of the stimulus evoked asynchronous event amplitudes recorded during the last 10 s of stimulation ( black fill ) and spontaneous synaptic current amplitudes ( gray fill , 402 events from 17 cells ) . The distributions are fit by a Gaussian function with means corresponding to 637 pA and 556 pA . ( D–F ) A representative paired recording from fish treated with 1 µM ω-conotoxin GVIA . ( D ) Traces of EPCs with expanded views showing near elimination of synchronous release ( D1 ) and intact asynchronous release ( D2 ) in ω-conotoxin GVIA-treated fish . ( E ) Time course of release for the recording shown in D . ( F ) Comparison of its asynchronous event amplitude ( black fill ) and the same spontaneous synaptic current amplitudes used for 1C and 1I ( gray fill ) . Events during the last 5 s of stimulation were included in the analysis . The mean value from a Gaussian fit for ω-conotoxin GVIA-treated fish was 620 pA . ( G–I ) A representative paired recording from the mutant line tb204a . ( G ) Traces of action potentials and EPCs from a homozygous tb204a mutant showing greatly reduced synchronous release ( G1 ) and intact late asynchronous release ( G2 ) . ( H ) The time course of release for the recording shown in G . ( I ) Comparison of its asynchronous event amplitudes ( black fill ) and the spontaneous synaptic current amplitudes ( gray fill ) . Events during the last 5 s of stimulation were included in the analysis . The mean value from a Gaussian fit for the mutant was 601 pA . Red circles in ( A ) , ( D ) , and ( G ) mark the peaks of synchronous events . All experiments were performed with 5 mM EGTA in the intracellular solution . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 003 When expressed as the time required to reach peak response during the stimulus train , the values were largest for ω-conotoxin GVIA-treated , smallest for control , and intermediate for tb204a mutant fish ( Figure 2A , B ) . The time to peak release for ω-conotoxin GVIA-treated fish and tb204a mutant were much more variable than seen in control fish , but despite the variability both were significantly prolonged when compared to control ( p<0 . 001; Figure 2B ) . 10 . 7554/eLife . 01206 . 004Figure 2 . Overall comparisons of release time course between control , ω-conotoxin GVIA-treated and tb204a mutant fish . ( A ) The time courses for paired recordings from wild-type ( black ) , tb204a mutant ( green ) and ω-conotoxin GVIA-treated ( red ) fish . The total release was expressed as the integrated charge for each consecutive second of the recording and normalized for comparison . ( B ) Scatter plot for recordings from wild type ( black; 3 . 4 ± 0 . 5 s , n = 15 ) , tb204a mutant ( green; 5 . 2 ± 1 . 0 s , n = 9 ) and ω-conotoxin GVIA-treated ( red; 10 . 5 ± 2 . 4 s , n = 13 ) fish comparing the time to peak release . Bars indicate the mean value and SD for each group . Asterisks indicate p<0 . 001 . All experiments were performed with 5 mM EGTA in the intracellular solution . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 004 Evidence for delayed asynchronous release in the presence of ω-conotoxin GVIA was next examined using an optical indicator of exocytosis . For this purpose , we recorded stimulus-driven fluorescence changes in a transgenic line of fish expressing synaptopHluorin in the spinal motor neurons . Exocytosis was signaled by an increase in fluorescence that results from exposure of the vesicular synaptopHluorin protein to neutral pH ( Miesenbock et al . , 1998 ) . Fluorescence was monitored in a single image plane that was selected prior to stimulation on the basis of distribution and number of boutons . For this purpose each CaP neuron was filled with the fluorescent Alexa Fluor 647 dye via the patch pipette ( Figure 3A , fill ) . Stimulation of CaP motor neurons for 10 s at 100 Hz resulted in a robust fluorescence increase that was largely restricted to the boutons associated with synapses ( Figure 3A , spH ) . The synaptic location of the regions of interest ( ROIs ) was determined by means of the postsynaptic receptor label α−bungarotoxin ( Figure 3A , α-btx and overlay ) . Next , stimulus-driven changes in synaptopHluorin fluorescence were determined . Images were captured at 33 ms intervals in a single plane that contained candidate ROIs . Post stimulus , images were corrected by background subtraction and then each ROI was quantitated as the ratio of delta fluorescence to prestimulus fluorescence ( ΔF/F0 , Figure 3B ) . This corrects for the expression of the pHluorin on the surface and in subcellular compartments other than vesicles . The fluorescence for control ROIs rose immediately at the start of 100 Hz stimulation and reached a plateau during the subsequent few seconds of stimulation ( gray traces , Figure 3B ) . However , the fluorescence signal in ω-conotoxin GVIA-treated neurons showed greatly delayed onset of fluorescence for comparable ROIs ( color traces , Figure 3B ) . Additionally , the synaptopHluorin signal was reduced in the ω-conotoxin GVIA-treated fish ( Figure 3B ) . Comparisons of the onset of fluorescence increase for overall ROIs were made by measuring the time required to reach 50% maximal intensity from the initiation of stimulation . The distribution showed non-overlapping values for control and ω-conotoxin GVIA-treated fish ( Figure 3C ) . These data agree well with those obtained by paired recordings showing delayed onset of asynchronous synaptic transmission in the absence of P/Q type calcium channel function ( Figures 1 and 2 ) . 10 . 7554/eLife . 01206 . 005Figure 3 . Delayed release in ω-conotoxin GVIA-treated fish is also observed by means of the exocytotic indicator synaptopHluorin . ( A ) Images taken from a single focal plane of the CaP motor neuron terminals showing the motor neuron fill with Alexa Fluor 647 ( gray ) , postsynaptic labeling with α-btx ( red ) , peak stimulus-induced synaptopHluorin fluorescence ( green ) and α-btx/synaptopHluorin overlay . These images are shown for a single control ( left ) and ω-conotoxin GVIA-treated ( right ) fish . The scale bar corresponds to 10 μm . ( B ) Stimulus-driven fluorescence change , expressed as ΔF/F0 for the three representative boutons shown for control ( gray ) and ω-conotoxin GVIA treated ( colored ) synaptopHluorin motor neurons . The synaptopHluorin signal was measured at each ROI during the time course of 100 Hz stimulation ( indicated by the black bar in B ) . ( C ) The histogram showing time required to reach 50% maximal fluorescence increase for boutons of control ( gray , 127 boutons from 8 fish ) and ω-conotoxin GVIA-treated fish ( red , 55 boutons from 4 fish ) . Experiments were performed with 5 mM EGTA in the intracellular solution . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 005 The delayed onset of asynchronous release seen with paired recording and synaptopHluorin imaging prompted the exploration into spatial changes in stimulus-driven calcium levels . For this purpose , calcium indicators Fluo-4 or Fluo-5F were loaded ( Figure 4 , Fluo-4 ) along with Alexa Fluor 647 for post experimental three-dimensional morphological reconstruction of the CaP neuron ( Figure 4 , fill ) . As with synaptopHluorin measurements , calcium imaging required acquisition speed that restricted measurements to a single image plane . With a 40× objective the plane of focus usually included the soma , axon initial segment , a large region of axon with the major branch point , and a field of synaptic boutons . Presynaptic ROIs were established on the basis of postsynaptic α-btx labeling ( Figure 4 , α-btx ) . When stimulated at 100 Hz , time-dependent increases in fluorescence included soma , axon and boutons ( Figure 4 , Fluo-4 ) . This location of distal fluorescence co-localized principally with the α-btx ( Figure 4 , merge ) . 10 . 7554/eLife . 01206 . 006Figure 4 . Stimulus-driven calcium signals in CaP motor neuron terminals occurred at synaptic boutons . The fill corresponds to a maximal intensity projection image of the motor neuron filled with Alexa Fluor 647 . An arrowhead indicates the soma . A single plane of focus in the filled neuron showing postsynaptic α-btx label , peak Fluo-4 calcium signal and merged α-btx and Fluo-4 signal . The scale bar corresponds to 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 006 This approach was then used to determine time-dependent changes in calcium for both control ( Figure 5A , B ) and ω-conotoxin GVIA-treated ( Figure 5C , D ) fish in response to 100 Hz stimulation . Fluorescence change was determined for each ROI and expressed as the ratio of the green fluorescence jump ( ΔG , Ca2+ signal ) to red fluorescence ( R , fill ) . This method allowed us to normalize the fluorescence change to differences in cell volume ( Figure 5B , D ) . In control neurons , ΔG/R within the axon initial segment and distal boutons showed no significant difference in the timing of signal onset ( Figure 5A , B ) . By contrast , in fish treated with ω-conotoxin GVIA , the fluorescent signal onset was delayed for all regions except the axon initial segment ( Figure 5C , D ) . Moreover , the delay was longer with greater distance from the soma ( Figure 5C , D ) . The time of onset for the calcium signal was considerably briefer than the asynchronous release shown in Figure 1 because of the differences in EGTA concentrations used . The differences between 0 . 5 mM and 5 mM EGTA on the signal onset are later dealt with in the results . The entire videos from which the static images in Figure 5 were extracted are available as Video 1 and Video 2 . 10 . 7554/eLife . 01206 . 007Figure 5 . Calcium signal onset is delayed at boutons in ω-conotoxin GVIA-treated fish . ( A and C ) Sample images of 100 Hz stimulus evoked Fluo-5F fluorescence increases taken at 1 s intervals for 4 ROIs for control ( A ) and ω-conotoxin GVIA-treated ( C ) fish . The scale bar corresponds to 20 μm . ( B and D ) The stimulus-driven fluorescence increases associated with each color-coded ROI in control ( B ) and ω-conotoxin GVIA-treated ( D ) fish . The fluorescence was baseline subtracted and the increase was expressed as ΔG/R . Black bars in ( B ) and ( D ) indicate the timing of 100 Hz stimulation . Experiments were performed with 0 . 5 mM EGTA in the intracellular solution . The entire videos for A and C are available as Video 1 and Video 2 respectively . For each video the timing of stimulation is indicated by the dot . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 00710 . 7554/eLife . 01206 . 008Video 1 . Control videoDOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 00810 . 7554/eLife . 01206 . 009Video 2 . GVIA videoDOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 009 The delayed onset of calcium indicator signal in boutons of ω-conotoxin GVIA-treated fish pointed to a process involving delayed rise in intracellular calcium at the boutons following stimulation at the soma . However , it was necessary to exclude the possibility that the calcium indicator was competing with the endogenous calcium handling in the cell . To address this possibility , we compared two calcium indicators , 100 μM Fluo-4 ( kd = 345 nM ) to 100 μM Fluo-5F ( kd = 2 . 3 µM ) , which have approximately sevenfold different binding constants for calcium ( Figure 6 ) . Comparisons were made on the basis of the onset of calcium signal in the boutons in both control and ω-conotoxin GVIA-treated fish . With both Fluo-4 and Fluo-5F the time to 20% rise was significantly longer in ω-conotoxin GVIA treated fish compared to control fish ( Figure 6 ) . Importantly , there was no significant difference between the dyes in ω-conotoxin GVIA-treated fish ( Figure 6 ) . The time to 20% rise was marginally different ( p=0 . 01 ) between the dyes in control experiments , potentially reflecting the increased time required for the lower affinity calcium indicators to measure the calcium rise . 10 . 7554/eLife . 01206 . 010Figure 6 . The delayed rise of the calcium signal is similar for two different affinity calcium indicators Fluo-4 and Fluo-5F . Comparisons of the time to reach 20% peak stimulated fluorescence in ω-conotoxin GVIA-treated ( shaded fill ) and control CaP ( no fill ) motor neurons . In control fish , Fluo-4 onset was 0 . 18 ± 0 . 16 s ( n = 64 boutons from 4 fish ) , and Fluo-5F onset was 0 . 47 ± 0 . 22 s ( n = 91 boutons from 6 fish ) . For ω-conotoxin GVIA-treated fish , Fluo-4 onset was 1 . 98 ± 0 . 84 s ( n = 69 boutons from 6 fish ) , and Fluo-5F onset was 2 . 14 ± 0 . 42 ( n = 85 boutons from 7 fish ) . Experiments were performed with 0 . 5 mM EGTA in the intracellular solution . Asterisks indicate p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 010 To determine whether the delay in calcium signal onset correlated with the physical distance of the boutons from the proximal axon or soma , we combined calcium imaging with morphological reconstruction of the motor neuron . For this purpose each motor neuron was reconstructed as a three-dimensional image using Imaris filament software on the basis of the dye fill ( Figure 7C ) . A 63× objective was substituted to obtain greater detail of the synaptic boutons resulting in exclusion of the soma from the field of view . Therefore , a reference point for distance zero was chosen that was approximately 80 μm from the soma where the axon has traversed the notochord ( Figure 7C , arrow head ) . The time-dependent changes in fluorescence were first determined for a series of ROIs corresponding to hot spots at different distances from the reference point ( Figure 7A , B ) . The rise in fluorescence was delayed with greater distance from the soma ( Figure 7B ) . When expressed as time to 20% rise vs distance from the reference point , a striking quasi-linear relationship was observed for the boutons ( Figure 7D ) . This relationship was determined for all of the recordings performed in ω-conotoxin GVIA-treated neurons ( Figure 7E ) . The mean velocity of travel obtained in boutons determined for the cumulative data was 74 μm/s in ω-conotoxin GVIA-treated fish . By contrast , the relationships obtained for control neurons showed no dependence on distance in keeping with velocity of the propagating action potential ( Figure 7E ) . In ω-conotoxin GVIA-treated neurons the distance dependence was not observed for the proximal axon , suggesting that the signaling underlying calcium propagation in these regions may be different from that of the distal terminals ( Figure 7D , E ) . 10 . 7554/eLife . 01206 . 011Figure 7 . Distance-dependent delay in Ca2+ rise in ω-conotoxin GVIA-treated CaP boutons . ( A ) Sample images of Fluo-5F fluorescence taken at 0 . 5 s intervals during 100 Hz stimulation . 5 ROIs are shown at different distances from the reference point at the ventral edge of the notochord . ( B ) The time course of fluorescence change , expressed as ΔG/R , for each of the ROIs shown in A . The black bar indicates the duration of stimulation . ( C ) Imaris Filament Tracer 3D reconstruction of the same motor neuron based on z-stacks of the Alexa Fluor 647 fill , with the ROIs in A and B overlaid . An arrowhead indicates the reference point for distance measurements . Scale bars in A and C correspond to 10 μm . ( D ) The time required for each ROI to reach 20% of peak as a function of the distance from the reference point . The distance measurements for each ROI were determined on the basis of Imaris 3D reconstruction . Colored symbols correspond to the individually colored ROIs shown in the A–C . The data points from the boutons , excluding the first distance measurement , were fit by a line with a slope corresponding to 57 µm/s . ( E ) Scatter plot of distance-dependent Ca2+ rise for 61 ROIs in ω-conotoxin GVIA-treated neurons ( n = 4 fish , colored markers ) and 47 ROIs in control ( n = 3 fish , gray markers ) . Each neuron was reconstructed using Imaris filament software to obtain the physical distances . Example cell in A–D is shown with green markers . Measurement was obtained with Fluo-5F and 0 . 5 mM EGTA in the intracellular solution . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 011 An alternative possibility to calcium propagation as causal to the delayed release and apparent wave is the existence of a proximal to distal gradient in endogenous calcium buffering strength . For example , the delayed appearance of calcium in the distal boutons might reflect a slower rise in free calcium due to greater calcium handling at the boutons . The calcium signals shown in Figures 5 and 7 suggest that such differences may exist . The signal in the axon initiation zone does not decline during maintained stimulation , whereas the distal signal appears to relax slightly during the stimulation . To test whether slow time-dependent calcium accumulation could account for the delayed rise in distal terminals , paired recordings were made using a ‘killswitch’ protocol . The stimulation was terminated either prior to ( Figure 8A ) or coincident with ( Figure 8B ) the onset of asynchronous release in ω-conotoxin GVIA-treated fish . In seven such recordings with 5 mM intracellular EGTA , the asynchronous release persisted well beyond termination of the stimulus . To test whether the calcium signal also persists after stimulus termination , we used the killswitch protocol during calcium imaging in the presence of 1 µM ω-conotoxin GVIA . The calcium signal , measured using Fluo-4 in the presence of 0 . 5 mM EGTA , also exhibited distance-dependent onset with the signal in the distal boutons peaking following termination of the stimulus ( Figure 8C ) . Because the stimulus was terminated prematurely in both paired recordings and calcium measurements , continued accumulation would not occur and could not account for the delayed onset of asynchronous release and the calcium signal . Instead , the findings strongly support the involvement of a propagating calcium signal that was triggered by stimulation at the soma . 10 . 7554/eLife . 01206 . 012Figure 8 . The delayed asynchronous release and Ca2+ rise in ω-conotoxin GVIA-treated fish is not due to slow local calcium accumulation at the distal boutons . ( A and B ) Two paired recordings for killswitch experiments are shown with expanded insets ( boxed region ) . ( A ) In this example , the motor neuron stimulation ( top trace ) was terminated prior to the sudden onset of asynchronous release ( bottom trace ) . ( B ) In this recording , the stimulus was terminated at the onset of asynchronous transmission , showing the persistence of the release . Both the experiments in ( A ) and ( B ) were performed with 5 mM EGTA in the intracellular solution . ( C ) An example calcium imaging experiment showing that the fluorescence signal peaked after the termination of the 100 Hz stimulation at 4 s . Left: an overlay of single imaging plane with the dye fill ( red ) and the Fluo-4 signal ( green ) . The individual color coded ROIs were used to generate the associated ΔG/R vs time plot ( right panel ) . The black bar shows the timing of stimulation , with a dashed line indicating the end the stimulation . This experiment was performed with 0 . 5 mM EGTA in the intracellular solution . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 012 As final evidence linking the arrival of calcium to asynchronous release we performed simultaneous paired recording and calcium imaging from ω-conotoxin GVIA-treated fish ( Figure 9 ) . This further required dye fill of both target muscle and neuron in order to identify the relevant boutons for imaging and measurement of calcium signal . The muscle cell was loaded with Alexa Fluor 555 and the CaP neuron was loaded with Alexa Fluor 647 ( Figure 9A , E ) . Stimulus-driven calcium increases ( Figure 9B , F ) were coincident with the postsynaptic asynchronous release measured by means of paired recordings ( Figure 9C , G ) . The coincidence between the onset of the calcium signal and asynchronous release was further compared on the basis of synaptic charge entry ( Figure 9D , H ) . In three separate simultaneous recordings , the onset time for the calcium signal and asynchronous release , both measured at 50% rise , was within 0 . 4 ± 0 . 3 s of each other . Additionally , a large difference in delay of onset was seen between experiments performed with 0 . 5 mM EGTA ( Figure 9A–D ) and 5 mM EGTA ( Figure 9E–H ) in the internal solution , prompting further investigation into the dependence of the calcium propagation on calcium buffering . 10 . 7554/eLife . 01206 . 013Figure 9 . Simultaneous paired recordings and calcium imaging in ω-conotoxin GVIA-treated fish . The examples shown compare the effects of 0 . 5 mM ( A–D ) vs 5 mM ( E–H ) intracellular EGTA . ( A ) and ( E ) Image of the CaP dye filled with Alexa Fluor 647 ( green ) and target muscle filled with Alexa Fluor 555 ( red ) . Multiple synaptic boutons contacting the target muscle cell are visible as yellow varicosities ( scale bar = 10 μm ) . An enlarged view of the fill ( A1 and E1 ) and peak Fluo-5F calcium response ( A2 and E2 ) are shown for the color coded boutons . ( B ) and ( F ) The associated ΔG/R plots for each of the boutons in A and E as a function of time during 20 s , 100 Hz stimulus ( indicated by black bar ) , which began at time 0 . ( C ) and ( G ) The associated patch clamp recording showing the motor neuron action potential ( top ) and postsynaptic EPCs ( bottom ) during the 20 s , 100 Hz stimulation . The scale bars correspond to 40 mV , 1 nA , and 2 s . ( D ) and ( H ) An overlay plot showing the coincidence between the onset of the mean ΔG/R fluorescence for all ROIs ( blue with gray SD ) and the onset of asynchronous release ( red ) . Release was shown as integrated synaptic charge entry for each consecutive half-second of stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 013 A hallmark of asynchronous release is inhibition by concentrations of the slow calcium buffer EGTA that do not affect synchronous transmission . Using the advantages offered by both calcium imaging ( Figure 10A , B ) and paired recordings of synaptic transmission ( Figure 10C–E ) , the effect of intracellular EGTA was determined for both the control and ω-conotoxin GVIA-treated fish . For ω-conotoxin GVIA-treated fish , calcium measurement from ROIs corresponding to boutons showed two to threefold greater delay in signal onset with 5 mM EGTA ( red ) compared to 0 . 5 mM EGTA ( blue , Figure 10A; Table 1 ) . The time to reach 20% maximal fluorescence as a function of distance from either the soma or axon reference point was compared for the two EGTA concentrations . Overall , the mean values obtained using 5 mM EGTA were significantly larger at all distances than those obtained using 0 . 5 mM EGTA for toxin treated fish ( Figure 10B , p<0 . 001 ) . Comparing distance-dependent travel of calcium in ω-conotoxin GVIA-treated neurons shows a quasi-linear relationship for the distal boutons ( Figures 7 and 10B ) . Fitting distance vs rise in the regions corresponding to boutons to a linear relationship yielded mean slope values corresponding to 74 μm/s and 35 μm/s for 0 . 5 mM and 5 mM EGTA respectively , consistent with a slower propagating rate in the presence of greater Ca2+ buffer ( Figure 10B ) . At both the EGTA concentrations , the arrival of the calcium signal in the axonal regions appeared to be independent of distance from the stimulation , despite the large differences in onset time . These data are indicative of a faster velocity of travel along these regions of the neuron . In control fish , the distance-dependence of calcium signal onset was not seen ( Figure 10B ) . However , 5 mM EGTA slowed the time to 20% rise in calcium signal compared to 0 . 5 mM EGTA , reflecting the competition between the indicator and the calcium buffer ( Table 1 , p<0 . 001 ) . 10 . 7554/eLife . 01206 . 014Figure 10 . Increasing the intracellular concentration of the calcium buffer EGTA further delays both the calcium signal in boutons and the onset of asynchronous release . ( A ) Sample traces comparing the time-dependent Fluo-5F fluorescence increases during stimulation for selected boutons of CaP motor neurons dialyzed with the indicated EGTA concentrations . ( B ) Cumulative data comparing the effects of 0 . 5 mM ( blue; 61 boutons from 4 fish ) and 5 mM ( red; 44 boutons from 3 fish ) EGTA on the time required to reach 20% maximal fluorescence change in boutons of ω-conotoxin GVIA-treated fish . The distances from the reference point were determined using Imaris filament reconstruction and binned ( 30 μm bin size ) . Control neuron dialyzed with 0 . 5 mM EGTA ( black; 47 boutons from 3 fish ) is shown for comparison . ( C ) Sample patch clamp recordings of muscle EPCs performed at three different EGTA concentrations in ω-conotoxin GVIA-treated fish . The action potentials are not shown but the stimulus lasted 35 s for the bottom trace and 20 s for the other traces . ( D ) The associated integrated synaptic currents as a function of time with 100 Hz stimulation beginning at time 0 . Only a single example is shown for 25 mM EGTA because this concentration blocked most transmission in other trials . ( E ) Comparisons of time to reach peak release for all recordings made using 0 . 5 mM ( 4 . 8 ± 0 . 4 s , n = 8 ) and 5 mM ( 10 . 5 ± 2 . 4 s , n = 13 ) EGTA in ω-conotoxin GVIA-treated fish . Data set for 5 mM EGTA was duplicated from Figure 2 for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 01410 . 7554/eLife . 01206 . 015Table 1 . Comparison of two different intracellular EGTA concentrations on Ca2+ signaling and synaptic transmission . The values indicated are all in units of s . Calcium imaging measurements were performed using Fluo-4 indicator . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 015Control+ ω-conotoxin GVIACa2+ imaging ( 20% rise ) paired recording ( time to peak ) Ca2+ imaging ( 20% rise ) paired recording ( time to peak ) 0 . 5 mM EGTA0 . 18 ± 0 . 16 ( 4 fish , 64 boutons ) 2 . 88 ± 0 . 35 ( n = 8 ) 1 . 98 ± 0 . 84 ( 6 fish , 68 boutons ) 4 . 88 ± 0 . 35 ( n = 8 ) 5 mM EGTA0 . 71 ± 0 . 47 ( 3 fish , 38 boutons ) 3 . 36 ± 0 . 50 ( n = 15 ) 5 . 42 ± 1 . 95 ( 4 fish , 56 boutons ) 10 . 54 ± 2 . 37 ( n = 13 ) Paired recordings provided an independent means of comparing the effects of different EGTA concentrations on the onset of asynchronous transmitter release ( Figure 10C–E ) . In the presence of ω-conotoxin GVIA , where release is exclusively asynchronous , the time to peak release was greater than twofold larger in the presence of 5 mM EGTA compared to 0 . 5 mM EGTA ( Table 1 ) . Sample traces showed the greatly delayed onset of asynchronous release with 5 mM EGTA compared to 0 . 5 mM EGTA ( Figure 10C ) . Inclusion of 25 mM EGTA in the electrode eliminated the majority of synaptic transmission in ω-conotoxin GVIA-treated fish in all but the single recording shown , where the onset of asynchronous release is further delayed compared with those recorded in 5 mM EGTA ( Figure 10C ) . Quantitative comparisons of delay were determined on the basis of normalized integrated postsynaptic charge entry vs time from the start of stimulation ( Figure 10D ) . The overall data for all of the recordings at 0 . 5 mM and 5 mM EGTA shows a highly significant difference in the time to peak release ( Figure 10E ) . In control recordings the time to peak release was slightly higher for 5 mM EGTA ( Table 1 , p=0 . 01 ) . The mechanisms underlying calcium entry and propagation were explored using pharmacological blockers of calcium channels and calcium-operated stores in ω-conotoxin GVIA-treated fish . Treatment of these fish with the non-specific calcium channel blocker Cd2+ produced ambiguous results . The high concentrations ( 500 μM ) required to inhibit the propagating calcium signals also inhibited the sodium channels responsible for action potential propagation , so firm conclusions could not be drawn . The rather non-specific T-type calcium channel inhibitors Mibefradil ( 5 µM ) together with Ni2+ ( 100 µM ) effectively reduced the calcium signal in the axon and boutons as did the L-type blocker Nimodipine ( 50 µM ) . However , other L-type blockers Nitrendipine ( 25 µM ) , Nifedipine ( 50 µM ) and Isradipine ( 50 µM ) were without effect . Similarly , the R-type Ca2+ channel blocker SNX 482 ( 200 nM ) was completely ineffective . The calcium stores inhibitor thapsigargin ( 1 µM ) and the IP3 receptor antagonists 2-APB ( 50 µM ) and Xestospongin-C ( 5 µM ) were also without effect . Both ruthenium red ( 400 µM ) and ryanodine ( 100 µM ) , inhibitors of the ryanodine receptors , gave variable results , only in some cases appearing to increase the arrival time for calcium in the boutons . Pharmacological treatment with tetrodotoxin ( TTX ) provided the clearest support for active propagation of the calcium signal . Prolonged depolarization of the CaP motor neuron in the presence of 1 μM TTX resulted in a calcium propagation that was clearly followed from soma to boutons without failure ( Video 3; Figure 11A ) . In most recordings , multiple propagating signals were observed in the boutons ( Video 3; Figure 11B ) and occasionally the signal back-propagated to the soma ( Video 4 ) rendering difficult the measurements of velocity . However , in the case shown in Figure 11 ( Video 3 ) , the forward propagating velocity for the first propagating signal corresponded to 3 μm/s . Overall estimates fall between 3–10 μm/s that is considerably slower than those obtained with ω-conotoxin GVIA-treated neuron ( Figures 7 and 10 ) . This difference points to a dominant role of the axonal action potential in coordinating the calcium signaling in the axon initiation zone and the axon proper . 10 . 7554/eLife . 01206 . 016Video 3 . TTX1DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 01610 . 7554/eLife . 01206 . 017Figure 11 . Depolarization-induced regenerative calcium wave in the presence of TTX . ( A ) Peak response of the Fluo-4 calcium signal following depolarization from −80 mV to +50 mV for 2 min in the presence of 1 μM TTX . The color coded ROIs are indicated . ( B ) Measurements of ΔG/R from the ROIs indicated in ( A ) show that calcium propagation traveled in two successive , distinct movements into the boutons . Each data point represents the flattened stack of 15 sections , each 1 μm thick , acquired with 1 s intervals between stacks . Experiment was performed with 0 . 5 mM EGTA in the intracellular solution . ( C ) Time to reach 50% of the first peak was plotted against distance from the soma for the example . ROIs shown in A and B are indicated with colors and additional ROIs not shown are black . The data points were fit to a line . The video for this recording and a second example recording are available as Video 3 and Video 4 respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 01710 . 7554/eLife . 01206 . 018Video 4 . TTX2DOI: http://dx . doi . org/10 . 7554/eLife . 01206 . 018
Paired motor neuron/target fast skeletal muscle recordings have shown that repetitive stimulation of the CaP motor neuron results in a transition from purely synchronous to mixed synchronous/asynchronous release after a delay that is dependent on stimulus frequency ( Wen et al . , 2010 ) . The studies on CNS neurons generally ascribe asynchronous release to persistent calcium resulting from opening of the highly localized presynaptic calcium channels ( Goda and Stevens , 1994; Cummings et al . , 1996; Atluri and Regehr , 1998; Chen and Regehr , 1999; Lu and Trussell , 2000 ) . At the zebrafish NMJ , this would be the ω-conotoxin GVIA-sensitive P/Q type calcium channel that is essential for synchronous release ( Wen et al . , 2013 ) . We tested this idea using paired recordings by either eliminating presynaptic calcium entry through blockade with ω-conotoxin GVIA or by means of a mutant line ( tb204a ) with functionally compromised P/Q type calcium channels ( Wen et al . , 2013 ) . We found that a greatly delayed asynchronous component was still present under conditions where the synchronous release calcium channel was eliminated . Using calcium indicator dyes we identified a source of calcium for asynchronous release that , under conditions of P/Q type calcium channel inhibition , originated in the axons and branch points and appeared to propagate into the synaptic boutons . The delayed arrival into the distal boutons , sites of synaptic interaction based on α-btx labeling , accounted well for the delayed onset of asynchronous release . Moreover , paired recordings performed along with calcium imaging confirmed the coincidence between the arrival of the calcium signal and the onset of delayed asynchronous synaptic transmission . The delay in the onset of asynchronous release was longest in ω-conotoxin GVIA-treated fish , intermediate in mutant tb204a fish and shortest in wild-type fish . This rank order was inversely related to the levels of P/Q type channel function , pointing to a likely contribution by these channels to asynchronous release as well . Accordingly , the early onset of asynchronous release in control recordings , compared to ω-conotoxin GVIA-treated fish , would result from the combined action of the local calcium entry and calcium arrival from a distal source . The synergistic action is fully compromised in ω-conotoxin GVIA-treated fish but only partially compromised in mutant fish . A model with separate , but interactive calcium sources , challenges the prevailing view that asynchronous release results exclusively from an expansion of the highly restricted calcium domains surrounding the same calcium channels that govern synchronous release ( Borst and Sakmann , 1996; Meinrenken et al . , 2002 ) . This model would predict that blocking the source of calcium for synchronous release would also inhibit the asynchronous release . Support for an expanding domain comes principally from the differential sensitivity of synchronous vs asynchronous release to slow calcium buffers ( Neher , 1998; Eggermann et al . , 2012 ) . The slow calcium buffer EGTA obliterates asynchronous release at concentrations that are ineffective on synchronous release ( Adler et al . , 1991; Cummings et al . , 1996; Atluri and Regehr , 1998; Lu and Trussell , 2000 ) . By contrast , blockade of synchronous release generally requires use of the fast buffer BAPTA ( Adler et al . , 1991 ) . Augustine notes that ‘this criteria has been used successfully to probe local calcium signaling with the conclusion that high BAPTA/EGTA efficacy points toward nanodomain and equal efficacy implicates diffuse calcium signals or microdomains’ ( Augustine et al . , 2003 ) , an approach widely used as a determinant of calcium channel-calcium sensor coupling ( Eggermann et al . , 2011 ) . It is the case with zebrafish NMJ that asynchronous release , as well as the calcium propagation , is highly sensitive to the slow buffer EGTA . We interpret these findings to mean that in zebrafish the differential sensitivity reflects , to a large extent , the involvement of bulk cytoplasmic changes stemming from distal calcium as opposed solely to highly restricted calcium domains entering through the P/Q type calcium channels . This provides an alternative interpretation to the BAPTA/EGTA sensitivity metric but does not exclude additional contribution by the expanding zone of calcium formed by local P/Q channel openings . The source of the axonal calcium entry underlying the propagation and the mechanism by which calcium propagates in a regenerative fashion are both unknown . The pharmacological results point to no clear source of either intracellular or extracellular calcium as required for the initiation or propagation . Effects seen using the calcium channel blockers , Mibefradil , Ni2+ and Nimodipine suggest that activation of calcium channels is involved , but the pharmacological profile does not fit any specific isoform . One impression resulting from these experiments is that the propagation is so potent in the branch points and locations near boutons , that there may be a large safety factor that requires near extinction of calcium before propagation is terminated . In this case , the pharmacological block would have to be nearly complete before the calcium propagation was inhibited . Our results from TTX experiments provide some insights into the process of calcium propagation . In the absence of action potentials , Ca2+ propagation can be elicited with prolonged depolarization , but the apparent velocity is slow . These data are compatible with a model in which Ca2+ propagation is less dependent on coordination by the action potential in the boutons than in the axons . Accordingly , an action potential traveling down the axon would instantaneously initiate Ca2+ propagation at the branching point , whereas the spread of depolarization would take much longer in the presence of TTX . The pharmacological conditions under which the calcium propagation is best revealed are non-physiological . Instead , the distance over which calcium needs to propagate under physiological conditions is certainly less due to participation of the action potential . Additionally , as pointed out earlier the P/Q channels do participate in the calcium generation and the secondary sources for calcium are also near the boutons , minimizing the need for long distance propagation . However , all evidence points to a central involvement of this secondary source of calcium in the delayed onset of asynchronous release . Our finding of a robust propagating calcium signal in the motor neuron was quite unexpected . Published accounts of propagating calcium signals like those shown in our study have been largely limited to postsynaptic cells ( Ross , 2012 ) . However , there is some precedence for ryanodine-sensitive calcium stores involvement in synaptic transmission at guinea pig sympathetic nerve terminals and frog NMJ ( Smith and Cunnane , 1996; Narita et al . , 1998 ) . Our findings take on additional significance due to our ability to assign a functional role to this secondary source of calcium . Consequently , we have a clearer understanding of the delayed onset and persistent release properties associated with asynchronous release . In light of the ubiquitous nature of asynchronous release , it will be critical to determine whether similar mechanisms are at work at synapses within the central nervous system .
Zebrafish ( Danio rerio ) were maintained in accordance with the standards set forth in the International Animal Care and Use Committee . Brian’s wild-type strain and the tb204a mutant line ( Wen et al . , 2013 ) were used for all experiments . The electrophysiology and imaging experiments were performed exclusively on larva between the ages of 72–96 hr post-fertilization ( hpf ) . Methodology for mounting and preparing the fish for paired recordings are detailed in the video publication ( Wen and Brehm , 2010 ) . The intracellular solution contains ( in mM ) : 115 K-gluconate , 15 KCl , 2 MgCl2 , 10 K-HEPES , 4 Mg-ATP , pH 7 . 2 with 0 . 5 mM EGTA or 5 mM EGTA as indicated with each experiment . The extracellular solution contains ( in mM ) : 134 NaCl , 2 . 9 KCl , 1 . 2 MgCl2 , 2 . 1 CaCl2 , 10 glucose , 10 Na-HEPES , pH 7 . 8 . For calcium imaging , either 100 μM Fluo-4 or 100 μM Fluo-5F was loaded into the CaP neuron by means of the recording patch pipette . Live confocal images were acquired using a Yokogawa CSU-10 spinning disc ( Yokogawa , Tokyo , Japan ) with a Stanford Photonics 620 Turbo ICCD camera ( Stanford Photonics , Palo Alto , CA ) . The laser lines used included an Argon 488 nm ( DLS2000 , Dynamic Laser , Salt Lake City , UT ) , 561 nm ( Sapphire 561-50 CW CDRH , Coherent , Santa Clara , CA ) and 640 nm ( Chromalase II diode laser , Blue Sky Research , Milpitas , CA ) . Low power images that included the entire CaP motor neuron utilized the Zeiss Plan-Apochromat 40×/1 . 0 n . a . dip objective , whereas higher resolution of the bouton field was obtained with the 63x equivalent objective . The CaP motor neuron was also co-loaded with 40 μM Alexa Fluor 647 hydrazide ( Invitrogen , Eugene , OR ) to identify the boutons and perform morphological reconstructions . For paired recordings with simultaneous calcium imaging , muscle was filled with 40 μM Alexa Fluor 555 hydrazide ( Molecular Probes , Eugene , OR ) . For each synaptopHluorin or calcium indicator recording , an acquisition plane was selected to contain 10–15 synaptic boutons in the field . Sequential images were acquired continuously at 33 ms intervals during 100 Hz stimulation . The calcium signals for each bouton were baseline subtracted , and the ratio of stimulus induced increase in calcium signal ( green ) to fill signal ( red ) was computed for regions of interest ( ROIs ) . Images were acquired with Piper Control 2 . 5 . 04 ( Stanford Photonics , Palo Alto , CA ) and analyzed with ImageJ ( NIH , Bethesda , MD ) , Microsoft Excel ( Redmond , WA ) and custom scripts in Igor Pro 6 . 3 ( Lake Oswego , OR ) . Unlike the calcium images taken in a single plane , images of the Alexa Fluor 647 filled motor neuron were acquired before and after the end of the experiment using 50–70 1 µm steps ( Mipos 100SG piezo driver; Piezosystems Jena , Jena , Germany ) to create stacks for the purpose of full morphological determination . The dye filled CaP motor neuron was reconstructed using Imaris filament software ( Bitplane , Zurich , Switzerland ) after completion of the calcium imaging . The Tol2 transposon system was used to generate the transgenic fish line expressing synaptopHluorin in the CaP motor neuron . SynaptopHluorin was constructed by fusing super-ecliptic pHluorin ( Miesenbock et al . , 1998 ) in frame to C-terminus of the zebrafish Vamp2 with an 8-amino acid linker . It was cloned into pTol2000 vector with the Huc promoter for primary motor neuron expression . The plasmid DNA was co-injected with in vitro synthesized transposase mRNA into one cell embryos . Founder fish were screened by PCR and fluorescence microscopy for expression in the spinal motor neurons . All data are presented as mean ± SD and statistical comparisons were made using standard t tests . TTX , ω-conotoxin GVIA and α-btx were obtained from Alomone Labs ( Jerusalem , Israel ) . Fluo-4 , Fluo-5F , Alexa Fluor dyes were obtained from Invitrogen ( Eugene , OR ) . Ruthenium red , ryanodine , Mibefradil , Nimodipine , and Ni2+ were obtained from Sigma-Aldrich ( St . Louis , MO ) . Nitrendipine , Nifedipine , Isradipine , SNX 482 , thapsigargin , 2-APB , and Xestospongin-C were obtained from Tocris Bioscience ( Bristol , UK ) . | Neurons communicate with one another at junctions called synapses . The arrival of an electrical signal known as an action potential at the first ( presynaptic ) neuron causes calcium ions to flood into the cell . This in turn causes the neuron to release packages of chemicals called neurotransmitters into the synapse . These activate receptors on the second ( postsynaptic ) neuron , triggering a new action potential that travels down the axon to the next synapse . The ions that trigger the release of the neurotransmitters are thought to enter the neuron through special calcium channels on or near the synapse . A sudden discrete influx of calcium ions causes the neuron to release many packages of transmitter simultaneously . This is called synchronous release . By contrast , when successive action potentials occur in the same neuron , the ions entering through the calcium channels accumulate inside the cell . This is thought to account for a sustained release of the neurotransmitter that continues even in the absence of nerve action potentials . This is called asynchronous release . Wen et al . have now obtained evidence that these two forms of release might be triggered by calcium from different sources . The work was performed using a synapse between nerve and muscle cells in zebrafish: it has been shown that channels called P/Q calcium channels control the release of neurotransmitters at this synapse in zebrafish . Mutant zebrafish with greatly reduced numbers of P/Q channels showed reduced synchronous release , but normal asynchronous release . Blocking the P/Q channels with a specific toxin in normal zebrafish eliminated synchronous release but left asynchronous release intact . Imaging experiments on these toxin-treated zebrafish revealed that a wave of calcium ions that propagated from a distant source coincided with the onset of asynchronous release . This wave of calcium fully accounted for the delayed onset and the persistence of asynchronous release following termination of the action potentials . This study further demonstrates that asynchronous release can be triggered by calcium ions that do not enter through the P/Q calcium channels . Waves of calcium have been described in the nervous system before , but their significance has always been unclear . The work of Wen et al . offers the first possible explanation for the role of these waves , and further experiments are now needed to determine whether this process happens at other types of synapses . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2013 | Synchronous and asynchronous modes of synaptic transmission utilize different calcium sources |
Evolutionary theory predicts that the lack of recombination and chromosomal re-assortment in strictly asexual organisms results in homologous chromosomes irreversibly accumulating mutations and thus evolving independently of each other , a phenomenon termed the Meselson effect . We apply a population genomics approach to examine this effect in an important human pathogen , Trypanosoma brucei gambiense . We determine that T . b . gambiense is evolving strictly asexually and is derived from a single progenitor , which emerged within the last 10 , 000 years . We demonstrate the Meselson effect for the first time at the genome-wide level in any organism and show large regions of loss of heterozygosity , which we hypothesise to be a short-term compensatory mechanism for counteracting deleterious mutations . Our study sheds new light on the genomic and evolutionary consequences of strict asexuality , which this pathogen uses as it exploits a new biological niche , the human population .
Obligate asexual reproduction has been argued to carry considerable negative evolutionary consequences ( Maynard Smith , 1986 ) hence most clonal species undergo some degree of recombination , albeit infrequent , which generates novel genotypes ( Heitman , 2006 ) . Here we investigate the reproductive strategy of a putatively asexual yet successful human pathogen , Trypanosoma brucei gambiense Group 1 , the main causative agent of human African trypanosomiasis , which contrasts with closely-related , sexually reproducing sub-species . T . b . gambiense has been divided into two groups . T . b . gambiense Group 1 , found in West and Central Africa , is the main human-infective sub-species , causing >97% of all human cases of trypanosomiasis ( Simarro et al . , 2010 ) . T . b . gambiense Group 2 was detected in the 1980/90s in Côte d’Ivoire and Burkina Faso but may now be extinct ( Capewell et al . , 2013 ) . A third human infective sub-species , T . b . rhodesiense , is found in East Africa and causes <3% of human cases ( Simarro et al . , 2010 ) . Each of these human infective sub-species appears to have arisen independently from the non-human infective T . brucei and possesses a different mechanism of human infectivity ( Capewell et al . , 2013; Capewell et al . , 2011; Uzureau et al . , 2013; Van Xong et al . , 1998 ) . All sub-species of T . brucei , with the important exception of T . b . gambiense Group 1 , show evidence for mating in natural populations ( Capewell et al . , 2013; Duffy et al . , 2013; Gibson and Stevens , 1999 ) and have the ability to undergo sexual reproduction with Mendelian allelic segregation and independent assortment of unlinked markers ( Cooper et al . , 2008; MacLeod et al . , 2005 ) . In addition , haploid gametes have been observed in T . b . brucei ( Peacock et al . , 2014 ) and although meiosis genes appear to be expressed in T . b . gambiense Group 1 ( Peacock et al . , 2014 ) , no haploid gametes have ever been observed in these parasites ( Peacock et al . , 2014 ) . This is consistent with clonality in all Group 1 populations analysed ( Koffi et al . , 2009; Morrison et al . , 2008; Tibayrenc and Ayala , 2012 ) , however , these studies were based on limited sets of genetic markers , which lack the necessary discriminatory power to distinguish between predominantly clonal evolution , with occasional bouts of genetic exchange , and strictly asexual propagation . Genomic-level analyses of T . brucei diversity to date have concentrated on T . b . brucei and T . b . rhodesiense and for T . b . gambiense Group 1 , include only the genome reference strain ( DAL972 ) ( Goodhead et al . , 2013 ) or two ( Sistrom et al . , 2014 ) field isolates . We hereby present a population-level genomic analysis as a means to determine whether this species is truly asexual , when the switch to asexuality arose and to provide insights into the genomic consequences of asexual evolution , including possible compensating strategies for eliminating deleterious mutations .
The genomes of 75 isolates of T . b . gambiense Group 1 ( Supplementary file 1 ) were sequenced , including multiple samples from geographically separated disease foci within Guinea ( n=37 ) , Côte d’Ivoire ( n=36 ) and Cameroon ( n=2 ) collected over fifty years ( 1952–2004 ) . For comparative purposes , isolates of T . b . rhodesiense ( n=4 ) , T . b . gambiense Group 2 ( n=4 ) and T . b . brucei ( n=2 ) were also sequenced . A total of 230 , 891 single nucleotide polymorphisms ( SNPs ) were identified compared to the haploid consensus assembly of the T . b . brucei reference genome ( Berriman et al . , 2005 ) . These were evenly distributed over the eleven major chromosomes , covering 85% of the genome ( Figure 1—figure supplement 1 ) . T . b . gambiense Group 1 showed a 5–10 fold lower number of SNPs ( 11 , 398 ) and SNP density compared to the other groups ( Figure 1—source data 1 ) , despite an over-representation in terms of the number of samples . Phylogenetic network analysis revealed that T . b . gambiense Group 1 genotypes showed an extremely low level of intra-group diversity ( e . g . the two most distantly related isolates differed only at 435 SNP loci ) and formed a monophyletic group ( Figure 1A ) . The network features reticulation among non-Group 1 parasites indicating the presence of recombinant genotypes; this stands in contrast to Group 1 parasites and is consistent with an absence , or rarity , of recombination in this group . Network analysis of T . b . gambiense Group 1 revealed the population is geographically sub-structured ( Figure 1B ) . A significant deviation from Hardy-Weinberg Equilibrium ( HWE ) was observed at 97 . 4% of SNP loci ( P<10-17 at each locus ) and this was found to be associated with every sampled genotype being heterozygous at these loci ( Figure 1—source data 2 ) . To control for geographical and temporal population sub-structure , isolates from three sub-populations were analysed and HWE deviation and heterozygote excess was confirmed ( Figure 1—source data 2 ) . FIS was calculated for each SNP locus , giving a uni-modal distribution with a median of -1 ( Figure 1—figure supplement 2 and Figure 1—source data 3 ) , as would be predicted for a strictly asexual population . Using a genome-wide panel of SNP loci , strong evidence of linkage disequilibrium ( LD ) was obtained for each chromosome and the whole genome formed a single genetic linkage group ( Figure 1—figure supplement 3 ) . 10 . 7554/eLife . 11473 . 003Figure 1 . Phylogenetic network analysis . SplitsTree phylogenetic networks were constructed using ( A ) each isolate for the collection of T . b . brucei ( Tbb ) , T . b . rhodesiense ( Tbr ) , T . b . gambiense Group 1 ( Tbg1 ) and T . b . gambiense Group 2 ( Tbg2 ) and ( B ) for just T . b . gambiense Group 1 isolates . The number of samples in each group is indicated in parenthesis . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 00310 . 7554/eLife . 11473 . 004Figure 1—source data 1 . Number of SNP loci with respect to different sub-species . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 00410 . 7554/eLife . 11473 . 005Figure 1—source data 2 . Testing Hardy-Weinberg Equilibrium ( HWE ) across the T . b . gambiense Group 1 genome . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 00510 . 7554/eLife . 11473 . 006Figure 1—source data 3 . FIS by sub-population . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 00610 . 7554/eLife . 11473 . 007Figure 1—source data 4 . Number and type of T . b . gambiense Group 1 SNPs . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 00710 . 7554/eLife . 11473 . 008Figure 1—figure supplement 1 . Genome-wide SNP density map for each sub-species . The densities of SNP loci were assessed among isolates of ( a ) T . b . brucei ( n=2 ) , ( b ) T . b . gambiense Group 1 ( n=75 ) , ( c ) T . b . gambiense Group 2 ( n=4 ) and T . b . rhodesiense ( n=4 ) over each of the 11 chromosomes ( horizontal scale ) . The number of SNPs per 10Kb window is plotted in blue , with the scale indicating SNP density between 0 and 150 loci per window ( vertical scale ) . T . b . gambiense Group 1 isolates have a SNP density , which is approximately ten-fold lower than other sub-species , however they are evenly distributed throughout the genome . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 00810 . 7554/eLife . 11473 . 009Figure 1—figure supplement 2 . Weir and Cockerham’s fis . The number of SNP loci exhibiting different levels of fis ( Weir and Cockerham’s fis ( Goudet , 1995 ) ) was plotted , showing a distribution around -1 , indicating strict asexuality . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 00910 . 7554/eLife . 11473 . 010Figure 1—figure supplement 3 . Genome-wide linkage disequilibrium ( LD ) among T . b . gambiense Group 1 parasites . Normalised LD was assessed across all eleven chromosomes . The position of each SNP used in this analysis is illustrated . The black triangle illustrates the single linkage group identified in the analysis , which spans the entire genome . Numbers 1–11 represent the 11 chromosomes . Red ( D'>1 , LOD≥2 ) , Pink ( D'<1 , LOD≥2 ) , Blue ( D'=1 , LOD<2 ) and White ( D'<1 , LOD<2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 010 Inspection of the SNP distribution across the genome of Group 1 isolates identified multiple long tracts of homozygosity , termed Loss of Heterozygosity ( LOH ) ( Figure 2—figure supplement 1 ) . Examination of read depth variation confirmed that this is not due to the loss of part of a homologue and is consistent with mitotic gene conversion; similarly there is no evidence of aneuploidy . In T . b . gambiense Group 1 , LOH has occurred on every chromosome , with many isolates showing the same LOH patterns ( Figure 2—figure supplement 1 ) , strongly suggesting that many of these regions arose early in Group 1 evolution . Chromosome 10 displayed the greatest variation in LOH sites , with a total of eighteen different patterns of variation among Group 1 genomes ( Figure 2 ) . This varied from 2% to 82% of the chromosome length , with LOH occurring predominantly towards one telomere . Distinct LOH patterns were associated with particular phylogenetic lineages , indicating LOH has occurred at various points in Group 1 evolution ( Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 11473 . 011Figure 2 . Loss of heterozygosity on chromosome 10 . Loss of heterozygosity regions ( blue ) spanning Chromosome 10 show 18 different patterns ( A-R ) . The number of isolates possessing each pattern and the percentage of the chromosome affected are indicated . The table ( inset ) shows the extent of LOH ( min and max ) for each chromosome as a percentage of chromosome length . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 01110 . 7554/eLife . 11473 . 012Figure 2—figure supplement 1 . Loss of heterozygosity across the T . b . gambiense Group 1 genome . Every T . b . gambiense Group 1 isolate ( first column ) was analysed for loss of heterozygosity ( LOH-see Materials and methods ) . Blocks of LOH are shown in blue across each of the eleven chromosomes , as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 01210 . 7554/eLife . 11473 . 013Figure 2—figure supplement 2 . Loss of heterozygosity across chromosome 10 . Every T . b . gambiense Group 1 isolate was analysed for loss of heterozygosity ( LOH ) . Blocks of LOH across chromosome 10 ( horizontal scale ) are shown in blue . Eighteen different patterns ( A-R ) are evident and the phylogenetic tree is shown on the left . Patterns of LOH can be seen to cluster in agreement with the tree , clearly showing that a stable LOH profile may be inherited . A selection of inherited blocks of LOH together with the branch on which they emerge are marked 1–4 . In addition , very recent LOH events can be observed , for example the large LOH region in ‘DEOLA’ , which is not observed in its close relative ‘LISA’ , highlighted in yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 013 Taken together , these analyses clearly demonstrate that Group 1 parasites are evolving asexually . A key predicted feature of asexual diploid species is the independent evolution and divergence of alleles on chromosome homologues ( Birky , 1996 ) , often referred to as the ‘Meselson effect’ ( Birky , 1996; Butlin , 2002; Mark Welch and Meselson , 2000 ) . To test whether this phenomenon occurred in T . b . gambiense Group 1 , three regions of the genome were chosen where it was possible to manually phase the genomic data , using LOH events to guide the phasing in non-LOH closely related isolates ( Materials and methods ) . For all three regions , a clear sequence of the accumulating mutations could be inferred ( Figure 3 ) , with each haplotype evolving independently , thus illustrating the Meselson effect ( Judson and Normark , 1996 ) . 10 . 7554/eLife . 11473 . 014Figure 3 . The Meselson effect . An accumulation of mutations on separate haplotypes , the ‘Meselson Effect’ , is illustrated using three regions on chromosome 10 . For each region ( 1 , 2 and 3 ) the two haplotypes are shown for a series of isolates with the accumulating mutations ( filled boxes ) indicated in red or blue for each haplotype . The sequences of accumulating mutations observed in the isolates ‘Lisa’ and ‘B4_F303P’ are shown . The number of mutations arising between each ancestral haplotype pair is indicated with two distinct lineages apparent at each locus , illustrating the independent evolution of each haplotype . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 01410 . 7554/eLife . 11473 . 015Figure 3—source data 1 . Estimated time since the most recent common ancestor . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 015 To test whether haplotypes have evolved independently at the whole-chromosomal level , the sequence dataset for all sub-species was computationally phased and haplotypes inferred for each isolate . Excluding LOH regions , phylogenetic trees for each chromosome were generated , revealing that A and B haplotypes separate into distinctive clusters ( example in Figure 4A ) and for every chromosome this pattern was maintained ( Figure 4—figure supplement 1 ) . Co-phylogenetic analysis revealed congruence between the A and B haplotype trees across different chromosomes ( Figure 4B; Figure 4—figure supplements 2 and 3 ) illustrating that these haplotypes are evolving in parallel . 10 . 7554/eLife . 11473 . 016Figure 4 . Haplotype co-evolution of chromosome 8 . ( A ) Phylogenetic tree of phased haplotype sequences of chromosome 8 shows a complete divergence between A ( blue ) and B ( red ) haplotypes of T . b . gambiense Group 1 ( identical genotypes removed ) . The tree is rooted using a Group 2 isolate ( green ) ; ( B ) Co-phylogenetic analysis reveals 100% consensus between the A and B haplotype trees of a subset of T . b . gambiense Group 1 isolates . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 01610 . 7554/eLife . 11473 . 017Figure 4—source data 1 . Co-phylogenetic analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 01710 . 7554/eLife . 11473 . 018Figure 4—figure supplement 1 . Phylogenetic trees of phased data showing ‘A’ and ‘B’ haplotypes . Maximum likelihood phylogenetic trees of phased haplotype sequences demonstrate divergence between A ( blue ) and B ( red ) haplotypes of T . b . gambiense Group 1 over each of the 11 chromosomes . This represents all SNP loci identified in each sub-species ( n = 230 , 891 ) . The haplotypes of non-Group 1 isolates are shown in green . Divergence of A and B haplotypes of Group 1 isolates can be observed for every chromosome . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 01810 . 7554/eLife . 11473 . 019Figure 4—figure supplement 2 . Co-phylogenetic analysis . For the three chromosomes with more than 1 , 000 SNP loci ( Figure 1—source data 4 ) , subsets of isolates ( discriminated by highly supported nodes , see Materials and methods ) were used to construct haplotype trees . In each case , the A and B haplotype trees show identical topology , illustrating the co-evolution of partner haplotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 01910 . 7554/eLife . 11473 . 020Figure 4—figure supplement 3 . Phylogenetic tree of all T . b . gambiense Group 1 isolates . A maximum likelihood phylogenetic tree was constructed with all T . b . gambiense Group 1 isolates using the panel of SNPs associated with derived alleles ( Figure 1—source data 4 ) . Bootstrap support is shown for each node . The 27 isolates chosen to illustrate co-evolution of partner haplotypes in Figure 4—figure supplement 2 are shown in blue ( Guinea ) and red ( Côte d'Ivoire ) . The clade used for molecular clock calculations is marked with an asterisk . DOI: http://dx . doi . org/10 . 7554/eLife . 11473 . 020 The time of emergence of T . b . gambiense Group 1 was determined . We estimated the genome-wide mutation rate using the number of accumulated mutations ( both genome-wide and on Chromosome 9 ) together with the year of isolation for each isolate , using root-to-tip linear regression ( Drummond et al . , 2003 ) ( Figure 3—source data 1 ) . A rate of 1 . 82 x 10–8 substitutions per site per year was estimated , similar to that of the COII-NDI ‘clock’ locus in T . cruzi ( Lewis et al . , 2011 ) . Given the observed number of accumulated mutations per isolate , the genome size , and our calculated mutation rate , the time since the existence of the most recent common ancestor ( MRCA ) of the Group 1 isolates analysed in the study is estimated to be in excess of one thousand years before present ( Figure 3—source data 1 ) . Similarly , using the mutation rates for two different T . cruzi clock genes ( COII-NDI and GPI ) ( Lewis et al . , 2011 ) , the date of emergence was estimated to be between 750 and 9 , 500 years before present ( Figure 3—source data 1 ) .
The theory of clonality in parasitic protozoan populations has been proposed and debated over the last quarter of a century ( Tibayrenc and Ayala , 2002; Tibayrenc et al . , 1990 ) . To advance our understanding of clonal evolutional , we have undertaken a whole-genome population-level analysis of T . b . gambiense Group 1 focussing on a large number of isolates sampled from two countries , together with a small out-group from a more distant West African country . We provide robust evidence that this important human parasite reproduces exclusively asexually , demonstrating complete genetic linkage across the genome and the absence of allelic segregation . Despite meiosis-specific genes being intact and expressed ( Peacock et al . , 2014 ) , the population genetic data is incompatible with sexual reproduction , self-fertilisation , aneuploidy ( Schurko et al . , 2009 ) , a parasexual cycle ( Ramírez and Llewellyn , 2014; Forche et al . , 2008 ) or the atypical meiosis observed in Rotifers ( Signorovitch et al . , 2015 ) . The barrier to sexual reproduction in T . b . gambiense Group 1 remains unclear . The lack of decay of ‘meiosis-associated’ genes may be explained by a number of non-mutually exclusive hypotheses including the relatively recent emergence of this asexual lineage , such that insufficient time has elapsed to allow decay . Alternatively , these genes may perform additional roles in processes such as DNA repair or VSG-related recombination . Our data indicates the parasite population comprises just two independently evolving haplotypes; this remarkable observation confirms the Meselson effect at a whole-genome level for the first time . Despite being predicted for almost twenty years , empirical evidence for this phenomenon has been elusive . The original report of the Meselson effect focused on Bdelliod rotifers ( Welch , 2000 ) , however this was later shown to instead be due to the entirely different phenomenon of cryptic tetraploidy ( Mark Welch et al . , 2008 ) . Subsequent attempts to demonstrate the Meselson effect have been thwarted , such as in the case of the obligately apomictic crustacean , Daphnia ( Omilian et al . , 2006 ) . The relatively high rate of mitotic recombination in comparison to the mutation rate results in novel heterozygous sites being eliminated by gene conversion ( LOH ) events much faster than they are generated and consequently allelic divergence is not observed ( Omilian et al . , 2006 ) . More recently , in the genome of Timema stick insects , nuclear alleles have been shown to display a higher level of divergence in asexual rather than sexual species ( Schwander et al . , 2011 ) . In that system , as in T . b . gambiense Group 1 , the mitotic recombination rate is sufficiently low so as not to obscure the pattern of accumulating mutations , and this has underpinned our ability to detect and confirm the Meselson effect . The similarity of the genomes studied from different geographical locations , together with a lack of recombination in the evolution of T . b . gambiense Group 1 , suggests this sub-species emerged from a single progenitor . Each clade represents a separate country and these are partitioned by the earliest branches of the phylogenetic tree , implying early radiation and colonisation . The emergence of T . b . gambiense Group 1 within the last 10 , 000 years coincided with an important period in human history when civilisation and livestock farming were developing in West Africa ( Oliver , 1966 ) , but whether asexual reproduction was a prerequisite for human infection or occurred subsequently is uncertain . This successful asexual human pathogen contrasts markedly with the virtually extinct sexual T . b . gambiense Group 2 parasite that occupied similar biological and geographical niches and this may be an example of the dominance of an asexual lineage over its sexual counterpart ( Charlesworth , 1980 ) . Another remarkable feature of the T . b . gambiense Group 1 genome is the extensive loss of heterozygosity across large regions of each chromosome , which likely arose from gene conversion/mitotic recombination . Gene conversion provides a mechanism for removing a proportion of deleterious mutations in asexual eukaryotes , as hypothesised in other systems ( Tucker et al . , 2013 ) . A fitter allele on one haplotype would be positively selected , resulting in long-range tracts of LOH that encompass the loci under selection together with extensive flanking regions . However , it has been predicted that following an LOH event , individuals will experience a slight decline in long-term fitness due to newfound homozygosity featuring pre-existing sub-optimal alleles , leaving a signature equivalent to inbreeding ( Tucker et al . , 2013 ) in the progeny . This process has been described as a more powerful evolutionary force than the accumulation of point mutations ( Omilian et al . , 2006 ) and thus may drive Muller’s Ratchet more quickly than de novo mutations . This suggests that although LOH may effectively counteract the accumulating mutational load in the short-term , it is unclear whether it can prevent this uniquely well-adapted pathogen from ultimately entering an extinction vortex .
A panel of eighty-five DNA samples was collected , representing T . brucei isolates from East and West Africa ( Supplementary file 1 ) , including T . b . brucei ( n=2 ) , T . b . rhodesiense ( n=4 ) , T . b . gambiense Group 1 ( n=75 ) and T . b . gambiense Group 2 ( n=4 ) . The main focus of the study was human-derived T . b . gambiense Group 1 , with the samples collected from sleeping sickness patients in Guinea ( n=37 ) , Côte d'Ivoire ( n=36 ) and Cameroon ( n=2 ) . This included collections from Bonon ( n=14 , collected 2000–2004 ) in the Côte d'Ivoire and Boffa ( n=18 , collected 2002 ) and Dubreka ( n=19 , collected 1998–2002 ) in Guinea . The T . brucei genome is approximately 26 Mb in size and comprises eleven major chromosomes between one and six megabases together with an array of intermediate and mini-chromosomes ( Ogbadoyi et al . , 2000 ) . Illumina paired-end sequencing libraries were prepared from genomic DNA and sequenced by standard procedures on Illumina HiSeq machines , to yield paired sequence reads of 75 bases in length . For each parasite strain , the data yield from the sequencing machines passing the default purity filter was between 7 . 4 million and 40 . 4 million read pairs ( median of 15 . 3 million ) , which corresponds to a nominal genome coverage of between 37 . 9-fold and 207 . 1-fold ( median of 78 . 2-fold ) . For the purpose of calling SNPs , mapping of the paired sequencing reads to the genome reference sequence from GeneDB ( Trypanosoma brucei brucei TREU927 , referred to here as Tb927 ) was carried out with SMALT ( www . sanger . ac . uk/resources/software/smalt/ ) version 0 . 7 . 4 using the following parameters: word length = 13 , skip step = 3 , maximum insert size = 800 , minimum Smith-Waterman score = 60 , and with the exhaustive search option enabled . A median fraction of 79 . 8% of sequencing reads were mapped and a median fraction of 64 . 9% of sequencing reads were classified as 'proper pairs' , i . e . with the two mates of a sequencing read pair mapped within the expected distance and in the correct orientation . Only sequence reads mapped as 'proper pairs' were used for SNP calling , and the first 5 and last 15 nucleotides were clipped from all reads prior to subsequent analysis . Genotypes for every genomic position were determined using SAMtools version 0 . 1 . 19 ( Li et al . , 2009 ) by using the 'samtoolsmpileup' command with minimum baseQ/BAQ ratio of 15 ( -Q ) followed by SAMtools' 'bcftools view' command with options -c and -g enabled . SNP calls were filtered according to the following criteria: a minimum of 6 high-quality base calls ( DP4 ) ; a minimum and maximum coverage depth ( DP ) of 0 . 25 times and 4 times the median , respectively; a minimum quality score ( QUAL ) of 22; a minimum mapping quality ( MQ ) of 22; a minimum second best genotype likelihood value ( PL ) of 0 . 25 times the median; a maximum fraction of conflicting base calls for homozygous genotype calls of 10%; and a minimum percentage of 5% for base calls ( as a fraction of all base calls for a given genotype ) that mapped either to the forward or the reverse strand of the reference sequence . Only loci that passed the quality control criteria for every sample were used for the population analysis . To ensure the SNP-calling parameters used in our analysis did not skew the distribution of variant sites detected ( a ) within individual genotypes and ( b ) across the sample collection , SNP-calling was performed using a range of stringencies . While the total number of SNP loci identified per individual and throughout the population varied depending on the stringency of the filter , the allele frequency spectrum remained constant ( data not shown ) . For the analysis presented in the manuscript , our SNP-calling parameters corresponded with a relatively low stringency filter , which was considered capable of detecting a very high proportion of variant sites while ablating the effects of sporadic read errors . Thus , the filter was designed to eliminate artefactual variants in the first instance , although a relatively low number of variant sites may remained undetected due to the necessity for every sample to pass QC at a given locus . The entire SNP dataset is deposited at the TritrypDB pathogen database , which is freely accessible at http://www . tritrypdb . org/tritrypdb/ . A subset of SNP loci was selected where a high-confidence genotype could be identified for every sample in the dataset ( Figure 1—source data 1 ) . This totalled 230 , 891 bi-allelic markers , which were used as the basis for the population genomic analyses presented in this study . The number of SNP loci was calculated for each sub-species using two methods: ( 1 ) in comparison to the Tb927 reference genome; and ( 2 ) in comparison to other members of that sub-species ( Figure 1—source data 1 ) . A total of 130 , 180 SNP loci were identified among the seventy-five Group 1 isolates in comparison to the reference T . b . brucei genome although only 11 , 398 of these showed polymorphism within Group 1 itself . In order to facilitate different types of analysis , a series of panels of a sub-set of SNP loci were defined ( Figure 1—source data 4 ) . For some analyses , Loss of Heterozygosity ( LOH ) regions of the genome were excluded and therefore a sub-set of SNP loci in non-LOH regions of the genome were identified ( Figure 1—source data 4; n=5201 ) . In addition , SNP loci in non-LOH regions where the minimum allele frequency was greater than 20% were identified , n=3 , 549 ( Figure 1—source data 4 ) , in order to provide sufficient power for testing linkage disequilibrium among the sequenced samples . In order to investigate whether SNP loci where the minimum allele frequency ( MAF ) is low correspond to localised areas where recombination events have occurred , the distribution of these loci was visually compared with the distribution of other SNPs in the non-LOH regions of the genome . Similar to the other SNPs , these SNP loci were evenly distributed over the non-LOH regions of the genome , excluding this possibility ( data not shown ) . Finally , a set of SNP loci was identified over non-LOH regions of the genome , excluding fixed heterozygous loci , which was polymorphic only among T . b . gambiense Group 1 isolates . These correspond to ‘derived’ alleles , which arose since the most recent common ancestor of the Group 1 isolates studied ( Figure 1—source data 4 , ‘Tbg1 derived’ ) . A set of SNPs loci polymorphic both within and outside the Group 1 population was also defined ( Figure 1—source data 4 , ‘Tbg1 ancient’ ) . Phylogenetic networks were constructed using the Split Decomposition method of SplitsTree4 ( Huson and Bryant , 2006 ) : Figure 1A shows the reconstruction using all the isolates sequenced in this study ( n=85 ) and the full panel of SNPs; Figure 1B shows relationships among the Group 1 isolates ( n=75 ) using the SNP panel corresponding to derived alleles in non-LOH areas ( Figure 1—source data 4 ) . The virtually non-reticulated topology of the network presented in Figure 1B supports our finding of strict asexuality in the Group 1 population and it is therefore appropriate to utilise a classical phylogenetic tree approach for the analysis of this sub-species . Maximum Likelihood phylogenetic trees were constructed using RAxML ( Stamatakis , 2014 ) using a generalised time-reversible model of sequence evolution . Confidence in individual branching relationships was assessed using 100 bootstrap pseudo-replicates and trees visualised using FigTree 1 . 4 ( tree . bio . ed . ac . uk ) . The Haploview software package ( Barrett , 2009 ) was used to investigate LD across each chromosome using unphased data and to calculate the normalised measure of allelic association , D' ( Daly et al . , 2001 ) . Linkage blocks were defined using the method of Gabriel et al . ( 2002 ) with a block being created if 95% of informative comparisons between SNP loci were found to possess ‘strong LD’ . Strong LD was defined as D'=1 and LOD score ≥2 . Haploview was also used to calculate the probability that any observed deviation from HWE could be explained by chance using a χ² test . This was performed for SNP loci across the T . b . gambiense Group 1 genome utilising the set of ‘Tbg1 ancient’ SNP loci ( Figure 1—source data 4 ) . Statistically significant loci at P<0 . 001 were additionally tested to determine whether deviation from HWE was associated with a heterozygote excess ( Figure 1—source data 2 ) , by comparing the predicted with the observed heterozygote frequency . Overall , a high proportion of SNP loci ( 97 . 4% ) showed a statistically significant departure from HWE ( P<10–17 ) and strikingly , all of these loci showed an excess of heterozygotes ( Figure 1—source data 2 ) . Along with the entire set of T . b . gambiense Group 1 isolates , a separate HWE analysis was performed on three spatio-temporally defined sub-populations ( Bonon , Boffa and Dubreka ) , thus accounting for any geographical and temporal population sub-structure . A series of further genetic tests was performed using Fstat version 2 . 9 ( Goudet , 1995 ) . FIS was calculated across the T . b . gambiense Group 1 genome , again utilising the set of ‘ancient’ SNP loci ( Figure 1—figure supplement 2 ) . A median figure of -1 was calculated for the entire set of T . b . gambiense Group 1 isolates and for each of the Bonon , Boffa and Dubreka sub-populations , indicating strict asexuality . This was supported by permutation testing ( n iterations = 30 , 000 ) , which indicated that FIS was lower than expected ( Figure 1—source data 3 ) . Similarly , using the Bonon , Boffa and Dubreka sub-populations , Weir & Cockerham’s fis was also calculated using Fstat . This had a uni-modal distribution with a median of -1 ( fis = -1 at 91 . 4% of loci ) , indicating strict asexuality . To assess the distribution of heterozygous sites across the genome , the density of these sites was calculated in 10 kb segments for every isolate . These density figures were used to determine whether each 10 kb segment could be considered a candidate area for long-range LOH . LOH blocks were defined using a custom Perl script to perform Interval Analysis using the following criteria: max number of heterozygous sites allowed per block = 0 , minimum number of contiguous blocks = 6 , maximum gap size in a contiguous block = 2 , max number of heterozygous sites allowed within gap = 2 . LOH block data was converted for viewing in the Integrative Genome Viewer ( IGV ) ( Thorvaldsdottir et al . , 2013 ) . In order to determine whether genomic structural variation could explain observed LOH , we performed a systematic copy number variation ( CNV ) analysis across the genome using CNVnator ( Mills et al . , 2011 ) . This reveals that there was no loss or gain of chromosomal material associated with LOH segments ( data not shown ) . In order to validate the computational phasing and investigate the relationship between haplotypes , three loci were selected where recent LOH events had occurred independently on chromosome 10 in different isolates . Such independent LOH events may be identified by examining patterns of LOH in comparison to the phylogenetic tree ( Figure 2—figure supplement 2 ) . For locus 1 , for example , an LOH region recently arose independently in B7_2 and DEOLA and can be observed over approximately 650 kb of chromosome 10 ( Figure 2—figure supplement 2 ) . Examination of the 506 SNP loci identified in this region indicated that two ancestral haplotypes existed . The nearest neighbours of these isolates ( B7_2 and DEOLA ) on the phylogenetic tree , CP1_2_KIVI and LISA , respectively , did not possess LOH in this region and therefore the sequences of B7_2 and DEOLA could be used as guides to allow CP1_2_KIVI and LISA to be confidently phased . Since the divergence of CP_1_2_KIVI with LISA , at this locus 15 mutations arose in the former isolate and 9 mutations in the latter and this is illustrated for LISA in Figure 3 . Again , because of the existence of only two haplotypes , more distant isolates could also be phased in this manner , which allowed intermediate haplotypes to be inferred and the accumulating sequence of Meselson mutations to be determined . To permit a genome-wide analysis , computational phasing of the diploid genotypic data was performed using the segmented haplotype estimation and imputation tool SHAPEIT2 ( Delaneau et al . , 2012; Delaneau et al . , 2013 ) . The default parameters were used together with an adjusted window size of 0 . 5 Mb and a recombination rate of 0 . 0003 ( MacLeod et al . , 2005 ) . The accuracy of the computational phasing for each isolate was assessed in comparison to a large LOH region on one isolate ( DEOLA at the 650 kb Locus 1 on Chromosome 10 ) . LOH in this region provided a set of ‘gold standard’ phasing information , which was used to check the phasing of all isolates , except the five which shared LOH in this region . A switch error rate ( Lin et al . , 2004 ) of between 4% and 12% was observed ( mean 9 . 3% ) across the 69 isolates , validating the results of the computational phasing . Phased sequence data from all isolates in the collection was used to create a separate Maximum Likelihood phylogenetic tree for each chromosome with RAxML ( Stamatakis , 2014 ) ( Figure 4—figure supplement 1 ) . For T . b . gambiense Group 1 isolates , co-phylogeny of the phased haplotypes ( A vs B ) was then assessed for each chromosome in turn using Jane ( Conow et al . , 2010 ) . Tree topologies were resolved using a Genetic Algorithm for co-phylogeny reconstruction with the default cost model . To assess whether the trees were more similar than would be expected by chance , 1 , 000 simulations were carried out using each of: ( a ) a random tip-mapping method; and ( b ) a random tree method ( beta = -1 ) . For every chromosome , A and B haplotype trees were significantly similar to each other ( Figure 4—source data 1 , P < 0 . 001 ) . In order to illustrate this similarity between A and B haplotype trees , a set of 27 isolates was selected which could be resolved with 100% bootstrap support from a phylogenetic tree constructed using the whole-genome dataset ( Figure 4—figure supplement 3 ) . The three chromosomes with the largest number of SNPs among T . b . gambiense Group 1 ( 8 , 10 and 11 ) were then selected , as these were the most informative . Isolates which could be resolved by the most highly supported nodes in trees representing the A and B haplotypes were subsequently selected and this corresponded to 16 , 18 and 13 isolates for chromosomes 8 , 10 and 11 respectively . In each case , the A and B haplotype sub-trees showed identical topology ( Figure 4—figure supplement 2 ) , illustrating the co-evolution of partner haplotypes . The date of emergence of T . b . gambiense Group 1 was calculated by combining the mutational rate with the estimated number of mutations arising since the most recent common ancestor of our collection of Group 1 isolates . The mutation rate was calculated using two approaches , the first using all seventy-five isolates , the second using a subset of samples representing a discrete lineage isolated in the Côte d'Ivoire ( Figure 3—source data 1 ) . Both methods gave rise to a similar mutation rate of approximately 2 x 10–8 mutations per base per year . Two approaches were used to assess the number of accumulated mutations present in the genome of each isolate . In the first , only the non-LOH portion of the genome was utilised to avoid the risk of mutations being ‘erased’ by way of LOH/gene conversion events . The number of SNP loci where derived alleles were present in the genome was counted . 388 such loci ( +/- 12 ) were identified on both homologues over 8 . 47 Mb of the non-LOH portion of the genome . The second involved focusing on Chromosome 9 , where a large region of LOH is found over much of the chromosome in every Group 1 isolate . Any mutations occurring since the existence of the most recent common ancestor of all the isolates analysed can easily be identified as they manifest themselves as heterozygous loci; this avoids the issue of identifying fixed heterozygous loci , which represent uninformative pre-existing mutations rather than accumulated mutations . Similar results were achieved with both methods , providing us with estimates of the time to most recent common ancestor ( TMRCA ) of approximately 1 , 000 to 1 , 500 years before present ( Figure 3—source data 1 ) . The mutation rates of two loci in T . cruzi , for which a 95% confidence interval was available , were also utilised to date the emergence of Group 1 parasites . These provided a confidence interval of approximately 750 to 9 , 500 years before present ( Figure 3—source data 1 ) . | An organism’s genetic ‘blueprint’ is encoded in DNA packaged within structures called chromosomes . Most organisms have two copies of each chromosome and , through sexual reproduction , the DNA within a pair of chromosomes can recombine randomly in a process that could be likened to shuffling a deck of cards . This process generates genetic diversity and means that any undesirable combinations of genes or mutations can be eliminated from the population by natural selection . While these activities are thought to promote the long-term survival of a species , some organisms appear not to have sex at all . Evolutionary theory predicts that ‘asexual’ organisms should face extinction in the long-term and that a lack of sexual recombination should leave a characteristic genetic ‘signature’ in their DNA . The theory also predicts that pairs of chromosomes will evolve independently , a phenomenon that is termed the ‘Meselson effect’ . However , while it was first predicted almost twenty years ago , evidence for this signature has been elusive . Now , Weir et al . have used the asexual parasite ( Trypanosoma brucei gambiense ) , which causes African sleeping sickness in humans , to search for signs of the Meselson effect . Sequencing the whole genome of a large number of parasites revealed that the population of this parasite arose from a single individual within the last 10 , 000 years . Over this time , mutations have built up and the lack of sexual recombination means that the two chromosomes in each pair have evolved independently of the other . These results provide the first demonstration of the Meselson effect at a genome-wide level in any organism . Weir et al . also uncovered evidence that this parasite uses a mechanism called “gene conversion” to compensate for its lack of sex . This mechanism essentially repairs the inferior , or mutated , copy of a gene on a chromosome by ‘copying and pasting’ the superior copy from the chromosome’s partner . The findings also suggest that gene conversion can only go some way to compensating for a lack of sex . A future challenge will be to investigate how effective this mechanism can be in the long term and to predict whether the parasite will ultimately become extinct . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"short",
"report",
"genetics",
"and",
"genomics"
] | 2016 | Population genomics reveals the origin and asexual evolution of human infective trypanosomes |
Superelongation complexes ( SECs ) are essential for transcription elongation of many human genes , including the integrated HIV-1 genome . At the HIV-1 promoter , the viral Tat protein binds simultaneously to the nascent TAR RNA and the CycT1 subunit of the P-TEFb kinase in a SEC . To understand the preferential recruitment of SECs by Tat and TAR , we determined the crystal structure of a quaternary complex containing Tat , P-TEFb , and the SEC scaffold , AFF4 . Tat and AFF4 fold on the surface of CycT1 and interact directly . Interface mutations in the AFF4 homolog AFF1 reduced Tat–AFF1 affinity in vivo and Tat-dependent transcription from the HIV promoter . AFF4 binding in the presence of Tat partially orders the CycT1 Tat–TAR recognition motif and increases the affinity of Tat-P-TEFb for TAR 30-fold . These studies indicate that AFF4 acts as a two-step filter to increase the selectivity of Tat and TAR for SECs over P-TEFb alone .
Transcription of the HIV genome by RNA polymerase II ( Pol II ) , like the expression of many cellular genes , is largely regulated at the step of transcript elongation . ( Levine , 2011; Lin et al . , 2011; Luo et al . , 2012; Zhou et al . , 2012 ) . Pol II is recruited to the HIV promoter and initiates transcription , which stalls after a 30–50 nucleotide transcript containing the trans-activating response region ( TAR ) is formed . The HIV Tat protein bound to a host super elongation complex ( SEC ) recognizes TAR and releases the paused polymerase ( He et al . , 2010; Sobhian et al . , 2010 ) . It is yet unclear how TAR and Tat specifically recruit SECs in preference to other complexes in the cell that contain SEC subunits . HIV Tat binds simultaneously to TAR and positive elongation factor b ( P-TEFb ) , composed of CDK9 and Cyclin T1 ( CycT1 ) subunits . In turn , P-TEFb and the transcriptional elongation factors ELL2 and ENL/AF9 associate with hydrophobic segments in the approximately 1200-amino-acid AFF1 or AFF4 scaffold , together forming the SEC ( He et al . , 2010; Lin et al . , 2010; Sobhian et al . , 2010 ) . P-TEFb triggers promoter escape by phosphorylating two negative elongation factors ( DSIF and NELF ) , as well as the C-terminal domain ( CTD ) of Pol II ( Ott et al . , 2011; Zhou et al . , 2012; Lu et al . , 2013 ) . In contrast , ELL2 is thought to stimulate Pol II processivity ( Shilatifard et al . , 1997 ) and ENL/AF9 appears to bridge the SEC to RNA polymerase II-associated factor complexes ( He et al . , 2011 ) . Overexpression of an AFF1 fragment that binds to P-TEFb , but not the other components of SECs , has a strong inhibitory effect on HIV transcription , indicating that productive HIV transcription requires a complete SEC ( Lu et al . , 2013a ) . As a key regulator of transcription , P-TEFb itself is regulated through complex formation with other proteins and RNA . Recent studies have shown that the CycT1 subunit tightly associates with AFF1 or AFF4 and that the ternary complex of CDK9 , CycT1 , and AFF1 ( He et al . , 2010; Lin et al . , 2010 ) moves between various active and inactive P-TEFb complexes ( Lu et al . , 2014 ) . These assemblies include the SECs ( He et al . , 2010; Lin et al . , 2010 ) , a complex with Brd4 ( Jang et al . , 2005; Yang et al . , 2005 ) , and the inhibitory 7SK snRNP ( Lu et al . , 2013a; Yang et al . , 2001; Yik et al . , 2003 ) . The existence of various P-TEFb complexes in the cell raises the question of how Tat and TAR discriminate among these functionally diverse assemblies . An initial clue to the origin of the specificity of Tat and TAR for recognizing SECs was provided by the crystal structure of the AFF4-P-TEFb complex ( Schulze-Gahmen et al . , 2013 ) . In this complex , the intrinsically disordered AFF4 fragment was folded on the surface of CycT1 in an orientation adjacent to the Tat binding site . A model of the quaternary Tat-AFF4-P-TEFb complex , based on the superposition of AFF4-P-TEFb and Tat-P-TEFb ( Tahirov et al . , 2010 ) , predicted direct interactions between Tat and AFF4 ( Schulze-Gahmen et al . , 2013 ) . These direct contacts were proposed to account for an 11-fold increase in the affinity of Tat for P-TEFb in the presence of AFF4 . To better understand the structural basis for the critical role of AFF1/4 in HIV transcription , we determined the crystal structure of P-TEFb in complex with Tat and AFF4 . The Tat-AFF4-P-TEFb structure and in vivo reporter assays in HeLa cells confirm the direct Tat–AFF4 interactions . In addition , AFF4 contacts CycT1 residues that are part of the flexible Tat–TAR recognition motif ( TRM ) ( Garber et al . , 1998; Das et al . , 2004 ) , increasing the order of the TRM in the quaternary complex . The TRM wraps around Tat , exposing several basic residues , and contributing CycT1 C261 to coordinate a shared Zn2+ ion . Remarkably , AFF4 fragments containing the CycT1- and Tat-interacting segments increased the affinity of Tat-P-TEFb for TAR by 30-fold . These results support the idea that AFF4 contributes to TAR binding through two distinct and sequential mechanisms . AFF4 interactions with Tat directly favor SEC recruitment and AFF4 interactions that constrain the CycT1 TRM indirectly promote TAR binding .
The 2–73 fragment of the SEC scaffold protein , AFF4 , binds with high affinity to P-TEFb and increases P-TEFb affinity for Tat 11-fold ( Chou et al . , 2013; Schulze-Gahmen et al . , 2013 ) . To define the structural basis for the increased affinity of the AFF4-P-TEFb complex for Tat , we determined the crystal structure of the quaternary complex of P-TEFb with AFF42–73 and Tat1–57 . The Tat 1–57 fragment is a minimal construct with high transcriptional activation ( Garcia et al . , 1988 ) . The structure was determined using X-ray data to 3 . 0-Å resolution ( R/Rfree = 0 . 206/0 . 232; Figure 1A , Table 1 ) with three complexes in the asymmetric unit ( a . u . ) . 10 . 7554/eLife . 02375 . 003Figure 1 . Tat and AFF4 bind adjacent to each other on the CycT1 surface . ( A ) Tat-AFF4-P-TEFb ribbon diagram ( left ) showing interactions between Tat ( red ) and AFF4 ( blue ) bound to the CycT1 ( yellow ) subunit of P-TEFb . AFF4 helix 0 is bound to the CDK9 ( cyan ) subunit , and adenosine ( spheres ) is modeled in the CDK9 ATP binding pocket . The close-up view ( right ) obtained by horizontal and vertical 45° rotations of the left hand figure shows similar Tat–AFF4 ( red/dark red-blue/light blue ) interactions in independent complexes . CycT1 ( yellow/green ) TRM residues adopt different structures in different crystal environments . The upper of the two Zn2+ ions ( gray spheres ) anchors the CycT1 TRM . ( B ) Surface representation of the binding pocket for Tat K28 in the Tat–AFF4 ( red-blue ) interface . The CycT1 TRM ( yellow ribbon ) with the fewest crystal contacts is shown . The TRM interacts with a hybrid interface including AFF4 and Tat . DOI: http://dx . doi . org/10 . 7554/eLife . 02375 . 00310 . 7554/eLife . 02375 . 004Figure 1—figure supplement 1 . Crystal contacts of CycT1 TRM for two representative complexes in the a . u . . ( A ) In the two dyad-related complexes , CDK9 ( gray sticks ) molecules from adjacent complexes make contacts with the C-terminal end of the CycT1 ( yellow ) TRM , as well as AFF4 ( blue ) . ( B ) In the third complex , the neighboring CDK9 ( gray sticks ) interacts with the N-terminus of the CycT1 ( green ) TRM . The C-terminal residues of the TRM are exposed to solvent . DOI: http://dx . doi . org/10 . 7554/eLife . 02375 . 00410 . 7554/eLife . 02375 . 005Figure 1—figure supplement 2 . Surface representation of the binding pocket for Tat M1 and the N-acetyl group . The methionine side chain binds in a pocket formed by Tat ( red ) and CycT1 ( yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02375 . 00510 . 7554/eLife . 02375 . 006Figure 1—figure supplement 3 . Schematic drawing of AFF4 secondary structures . Disordered regions ( gray rectangles ) , α helices ( blue springs ) and the short β strand ( black arrow ) are indicated . Helix H0 in AFF4 is only observed in two out of three molecules in the a . u . DOI: http://dx . doi . org/10 . 7554/eLife . 02375 . 00610 . 7554/eLife . 02375 . 007Table 1 . X-ray data collection and refinement statistics for P-TEFb-Tat-AFF4DOI: http://dx . doi . org/10 . 7554/eLife . 02375 . 007Data collection Space groupP6522 Cell dimensions: a , b , c184 . 91 , 184 . 91 , 360 . 40 Resolution ( Å ) *50 . 0–3 . 0 ( 3 . 05–3 . 0 ) Unique reflections*73 , 424 ( 3589 ) I/σ ( I ) *12 . 8 ( 0 . 9 ) Rmerge ( % ) *22 . 2 ( >100 ) Rmerge ( % ) * , I/sigI≥38 . 4 ( 18 . 9 ) Rpim ( % ) †7 . 6 ( 87 . 9 ) CC1/2 high resolution shell0 . 553 Completeness ( % ) *100 . 0 ( 100 . 0 ) Redundancy*24 . 2 ( 23 . 8 ) Temperature ( K ) 100 Mosaicity ( ° ) 0 . 23–0 . 39Refinement Resolution ( Å ) 49 . 0–3 . 0 No . reflections73 , 297 Rwork/Rfree*0 . 206/0 . 232 ( 0 . 316/0 . 335 ) No . atoms/B-factors ( Å2 ) CDK9 , molecule 1 , 2 , 32560 ( 75 . 4 ) , 2521 ( 90 . 9 ) , 2572 ( 88 . 5 ) Cyclin T1 , molecule 1 , 2 , 32061 ( 79 . 4 ) , 2053 ( 85 . 8 ) , 2058 ( 97 . 8 ) AFF434-66 , molecule 1 , 2 , 3438 ( 85 . 0 ) , 268 ( 115 . 7 ) , 422 ( 92 . 3 ) Tat390 ( 79 . 1 ) , 384 ( 78 . 0 ) , 390 ( 102 . 7 ) Water37 ( 58 . 7 ) R . m . s . deviations Bond lengths ( Å ) 0 . 0035 Bond angles ( ° ) 0 . 811 Ramachandran plot‡ Favored ( % ) 96 . 0 Allowed ( % ) 3 . 36 Disallowed ( % ) 0 . 66*Values in parentheses are for the highest resolution shell . †Rp . i . m . is the precision-indicating merging R factor , which is related to the traditional Rsym but provides a better estimate of data quality ( Weiss and Hilgenfeld , 1997; Weiss et al . , 1998 ) . ‡Values from MOLPROBITY ( Chen et al . , 2009 ) . Two complexes related by a non-crystallographic two-fold rotation axis are nearly identical , while the third complex shows small differences due to different crystal contacts ( Figure 1—figure supplement 1 ) . For example , only the two dyad-related complexes show electron density for AFF4 helix 0 ( residues 4–21 , Figure 1A and Figure 1—figure supplement 3 ) , which packs against αE and αI of the CDK9 subunit in the same complex and AFF4 35–39 from the two-fold-related complex in the a . u . Almost identical interactions between AFF4 helix 0 and CDK9 were observed in one out of three assemblies in the AFF4-P-TEFb structure ( Schulze-Gahmen et al . , 2013 ) , in addition to crystal contacts between helix 0 and a crystallographically related CDK9 subunit . This recurrence of the same helical structure in different crystal environments suggests that AFF4 residues 4–21 prefer a helical conformation . The function of helix 0 , however , remains in doubt because mutational effects on transcription do not match the AFF4 contacts in the interface , and stabilization of helix 0 depends on crystal packing ( Schulze-Gahmen et al . , 2013 ) . We will focus on features shared among all complexes and point out differences when they are relevant for the discussion . Tat binds in an extended conformation to AFF4-P-TEFb , with minor changes from the Tat-P-TEFb structure ( Tahirov et al . , 2010 ) . The backbone of loop residues 27–30 shifts in response to contacts with CycT1 250–261 for two complexes in the a . u . Based on mass spectrometry and electron density , the baculovirus-expressed Tat , like Tat expressed in HEK293T cells ( Jäger et al . , 2012 ) , is N-terminally acetylated ( Figure 1—figure supplement 2 ) . Acetylation of the Tat N-terminus removes a positive charge in the moderately hydrophobic Tat–CycT1 interface around Tat M1 and fills a cavity , probably leading to tighter anchoring of the Tat N-terminus to CycT1 . AFF4 residues 34- to 69-fold on the CycT1 surface , making multiple direct contacts with Tat K28 , F32 , and E2 ( Figure 1 ) and burying Tat M1 . The average size of the Tat–AFF4 interface is 305 Å2 on Tat and 330 Å2 on AFF4 . The interactions between AFF4 and Tat are mostly hydrophobic and van der Waals contacts , but also include hydrogen bonds on each end of the interaction site ( Figure 1A ) . The nexus of the Tat–AFF4 interface , Tat K28 , is partially buried in a hydrophobic pocket formed by the side chains of AFF4 M62 , F65 , and Tat F32 , and the main chain of AFF4 helix 2 , residues 59–63 . The Tat K28 side-chain amino group forms a hydrogen bond with the AFF4 E61 main-chain carbonyl at the pocket edge facing the solvent ( Figure 1A ) . In addition , the Tat E2 side chain is positioned to form a hydrogen bond with the AFF4 D68 main-chain amide . The CycT1 TRM forms another side of the K28 pocket ( Figure 1B ) . Although AFF4 contains similar secondary structural elements observed in the absence of Tat ( Schulze-Gahmen et al . , 2013 ) , coupled shifts occur to avoid collisions with Tat ( Figure 2 ) . Backbone RMS deviations in AFF4 residues 34–66 excluding the variable loop 43–45 range from 1 . 6 Å to 2 . 2 Å between AFF4-P-TEFb complexes with and without Tat . In contrast , RMS deviations for the same AFF4 residues of different complexes in the a . u . range from 0 . 32 Å to 0 . 58 Å . While AFF4 residues 34–40 coincide in the presence and absence of Tat , AFF4 helix 1 ( residues 48–55 ) is shifted along the helix axis . This change positions AFF4 M55 to make contacts with L252 in the CycT1 TRM region ( Figure 2A ) . AFF4 helix 2 ( residues 58–66 ) is shifted away from the bound Tat and closer to the CycT1 surface formed by helices H2′ , H3′ and the H3′–H4′ loop . This movement leads to the formation of additional hydrogen bonds between AFF4 and CycT1 in the Tat-AFF4-P-TEFb complex compared to the AFF4-P-TEFb complex . As a consequence of the shifts in the AFF4 backbone , unfavorable close contacts between AFF4 and Tat are avoided ( Figure 2B ) . 10 . 7554/eLife . 02375 . 008Figure 2 . Structural shifts in the subunits of the Tat-AFF4-P-TEFb complex . ( A ) Superposition of the AFF4-P-TEFb complex ( PDB ID 4IMY , pastel colors ) and Tat-AFF4-P-TEFb ( red , blue , yellow ) on the CycT1 subunit shows coupled shifts of the two AFF4 helices . AFF4 helices 1 and 2 shift away from Tat , thereby avoiding close contacts between Tat and helix 2 . ( B ) Superposition of AFF4-P-TEFb ( PDB ID 4IMY , AFF4 light blue ) , Tat-P-TEFb ( PDB ID 3MI9 , Tat pastel-red ) , and Tat-AFF4-P-TEFb ( Tat red , AFF4 blue , CycT1 green ) on the CycT1 subunit . CycT1 of AFF4-P-TEFb and Tat-P-TEFb is omitted to emphasize the changes in AFF4 . The AFF4 backbone shifts 1–2 Å in the presence of Tat , while the Tat conformation displays only small changes associated with AFF4 binding . Side chains undergo only small conformational changes . DOI: http://dx . doi . org/10 . 7554/eLife . 02375 . 008 In contrast to the Tat-P-TEFb complex , in which the CycT1 TRM is disordered between residues 253 and 260 , this functionally important segment is ordered in two conformations in the Tat-AFF4-P-TEFb complex ( Figures 1 , 3 ) . In all three complexes in the a . u . , P249 and N250 at the beginning of the TRM make multiple contacts with the main chain atoms at the C-terminal end of Tat helix 35–44 . In addition , CycT1 L252 forms hydrophobic interactions with AFF4 M55 in all three complexes . The TRM structures start to diverge at this point . CycT1 residues 253–259 loop over Tat helix 28–33 in two conformations that converge at CycT1 C261 ( Figure 3A , Figure 3—figure supplement 1 ) . In the dyad-related complexes , the CycT1 TRM makes crystal contacts that include W258 , R259 , and A260 , but in the third complex , the crystal contacts are restricted to the side chain of R251 . In all complexes , W256 makes buried contacts with AFF4 . Basic residues such as K253 , R254 , R259 , and the polar N257 , on the other hand , show continuous main-chain electron density but the exposed side chains are disordered . C261 binds the shared Tat Zn2+ ion in all three complexes . The presence of multiple conformations and relatively weak electron density for the TRM loop residues 253–260 indicates that this region is conformationally restrained but still quite flexible after Tat and AFF4 binding . These results suggest that in the presence of Tat , AFF4 partially orders the TRM ( Figure 3B ) . 10 . 7554/eLife . 02375 . 009Figure 3 . CycT1 TRM interacts with Tat and AFF4 . ( A ) Ribbon diagram of two distinct TRM conformations observed in the Tat-AFF4-P-TEFb crystal structure ( red , blue , yellow/dark red , light blue , green ) . Zn2+ ions are shown as gray spheres . ( B ) Surface representation of Tat-AFF4-CycT1 interactions . The three subunits intertwine , thereby stabilizing the TRM conformation in the hybrid interface . DOI: http://dx . doi . org/10 . 7554/eLife . 02375 . 00910 . 7554/eLife . 02375 . 010Figure 3—figure supplement 1 . Representative electron density for the Tat-AFF4-P-TEFb complex . 2Fo-Fc map ( 1 . 0 σ ) for Tat ( red ) and CycT1 TRM ( green ) is shown for a dyad-related complex . Residues of the CycT1 TRM were omitted from the model used for molecular replacement and subsequently built into the omit electron density . DOI: http://dx . doi . org/10 . 7554/eLife . 02375 . 010 The structure of the Tat-AFF4-P-TEFb complex points to AFF4 M62 and F65 as the major Tat-interacting residues . To test the contribution of the Tat–AFF4 interface to Tat-dependent transcription , we measured the effect of alanine substitutions on SEC recruitment and Tat-dependent HIV-1 transcription . These assays were performed with the AFF1 scaffold protein , because Tat has a stronger effect on HIV transcription with AFF1 than with AFF4 ( He et al . , 2010; Lu et al . , 2014 ) . AFF1 mutants V67A and F70A ( corresponding to M62A and F65A in AFF4 ) were ectopically expressed in HeLa cells in the absence or presence of Tat ( C22G ) , a mutation in the Zn2+ ion coordination site required for WT Tat activity ( Garber et al . , 1998 ) . The Tat ( C22G ) mutation increases the dependence on AFF1 for efficient transactivation ( Lu et al . , 2014 ) . Immunoprecipitation of tagged WT AFF1 , as well as the V67A , F70A , and V67A/F70A variants , efficiently co-precipitated CDK9 and CycT1 ( Figure 4A ) . In contrast , Tat ( C22G ) failed to co-precipitate P-TEFb ( Figure 4B , lanes 1 & 2 ) . This lack of binding was rescued by co-expressing WT AFF1 but not the three AFF1 alanine variants , V67A , F70A , and V67A/F70A ( Figure 4B ) . These results suggest that AFF1 V67 and F70 are important for interactions between the scaffold and Tat . In turn , this interface stabilizes the AFF1-CycT1 association , as suggested by the Tat-AFF4-P-TEFb structure . 10 . 7554/eLife . 02375 . 011Figure 4 . AFF1 Tat interaction mutants reduce Tat binding and activation of HIV LTR by AFF1 . ( A ) Nuclear extracts ( NE ) were prepared from HeLa cells expressing the truncated Flag-tagged AFF1 protein ( residues 1–308 ) . Anti-Flag immunoprecipitates ( IP ) from the NE were examined by Western blotting ( WB ) for the indicated proteins . ( B ) Nuclear extracts were prepared from HeLa cells co-expressing Flag-tagged Tat ( C22G ) and haemagglutinin ( HA ) -tagged truncated AFF1 . Anti-Flag IPs were analyzed as in A . ( C ) HeLa-based NH1 cells containing the intergrated HIV-1 LTR-luciferase reporter gene were transfected with the Tat ( C22G ) - and/or AFF1-expressing construct as labeled . Luciferase activities were measured in cell extracts , with the level of activity detected in cells transfected with an empty vector ( − ) set to 1 . The error bars represent mean ± SD from three independent measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 02375 . 011 The effects of changes in the Tat–AFF1 interface on Tat-dependent HIV transcription were measured using a HIV LTR-driven luciferase reporter system in HeLa-derived NH1 cells ( He et al . 2010 ) that can express Tat ( C22G ) . Although Tat ( C22G ) barely activated HIV transcription by itself ( 1 . 9-fold ) , this mutant strongly synergized with WT AFF1 to stimulate transcription to a much higher level ( 89-fold ) . Ectopic expression of WT AFF1 by itself only increased Tat-independent luciferase expression 13-fold ( Figure 4C ) . In contrast , the AFF1 single ( V67A and F70A ) and double ( V67A/F70A ) alanine mutants showed significantly reduced cooperation with Tat ( C22G ) in activating HIV transcription compared to WT AFF1 ( Figure 4C ) . Thus , the Tat–AFF4 interface is critical not only to enhance the binding of Tat to CycT1 , but also to stimulate Tat-dependent HIV transcription . The Tat–TAR recognition motif of CycT1 is essential for high affinity binding of P-TEFb-Tat to TAR ( Garber et al . , 1998 ) . This critical segment ( CycT1 250-264 ) at the C-terminal end of the cyclin domain interacts directly with TAR , as judged by RNA–protein cross-linking studies ( Richter et al . , 2002 ) . Since AFF4 binds close to the TRM in the Tat-AFF4-P-TEFb structure , we investigated the effect of AFF4 on TAR binding . Electrophoretic mobility shift assays ( EMSA ) revealed unexpectedly that AFF4 fragments 32–67 , 2–73 , and 2–98 each increased the affinity of TAR for the Tat-P-TEFb complex by 30-fold ( Figure 5A ) . It is unlikely that AFF4 directly contributes to TAR binding , since the AFF4-P-TEFb complex does not show any binding to TAR by itself . Instead , the Tat-AFF4-P-TEFb structure provides evidence that AFF4 binding in the presence of Tat restricts the conformational freedom of the CycT1 TRM region and positions this region for TAR interaction ( Figure 5B ) . 10 . 7554/eLife . 02375 . 012Figure 5 . SECs stimulate TAR recognition . ( A ) Electrophoretic mobility shift assays with 32P-labeled TAR and increasing concentrations of Tat-P-TEFb , or Tat-P-TEFb + AFF432–67 , Tat-P-TEFb + AFF42–73 , Tat-P-TEFb + AFF42–98 . Control assays ( bottom right ) with P-TEFb and AFF42–73 showed no shifts for TAR . Half of TAR was shifted with 35–40 nM Tat-P-TEFb complex . In the presence of excess AFF4 fragments 32–67 , 2–73 , or 2–98 , 50% of TAR was shifted by 1 . 1 nM Tat-AFF4-P-TEFb complex . ( B ) Calculated electrostatic surface potential of Tat-AFF4-CycT1 centered on the CycT1 TRM . The ribbon diagram ( right ) is in the same orientation as the surface representation ( left ) . This orientation converts into the orientation in Figure 1A by consecutive rotations around y ( 70° ) and z ( −35° ) . CDK9 was omitted from the surface figure ( left ) to focus on the TAR interaction region . Solvent-exposed CycT1 residues K253 , R254 , N257 , W258 , R259 , and Tat R49 , which have no side-chain electron density , were modeled in the most common orientation . The electrostatic potential , calculated using APBS ( Baker et al . , 2001 ) was applied to color the solvent excluded surface of Tat-AFF4-CycT1 in Chimera ( Pettersen et al . , 2004 ) from −5 kbTe−1 ( red ) to +5 kbTe−1 ( blue ) . CycT1 residues were labeled in black , Tat residues in red . The TRM region forms a positively charged patch on the SEC surface close to the disordered Tat ARM , which follows Tat R49 . DOI: http://dx . doi . org/10 . 7554/eLife . 02375 . 01210 . 7554/eLife . 02375 . 013Figure 5—figure supplement 1 . Model of TAR binding to SEC . The positively charged CycT1 TRM is positioned close to the predicted location of the Tat ARM , which binds to bases in the TAR bulge ( U23–U25 ) ( Weeks and Crothers , 1991 ) . Considering that the CycT1 TRM is interacting with the TAR loop region ( C30–A35 ) ( Richter et al . , 2002 ) , we manually placed the solution structure of TAR-arginimide ( PDB ID 1ARJ ) onto the Tat-AFF4-P-TEFb model so that the TAR bulge neighbors the Tat ARM region and the TAR loop contacts the CycT1 TRM . The dimensions of the components match well . The color scheme for the SEC is the same as in previous figures . The TAR phosphate backbone is shown in orange with bases in blue . TAR bases important for binding to Tat or CycT1 are drawn in magenta . DOI: http://dx . doi . org/10 . 7554/eLife . 02375 . 013
Efficient HIV transcription requires the Tat/TAR-mediated recruitment of SECs to the HIV promoter . Transcription remains stalled at the promoter in the absence of host elongation factors needed for Pol II to faithfully reach the distal end of the HIV genome . The X-ray structure of Tat-AFF4-P-TEFb reveals subunit interactions that mediate the preference of Tat for SECs over other P-TEFb complexes . As predicted ( Schulze-Gahmen et al . , 2013 ) , Tat and AFF4 bind adjacent to each other on the CycT1 surface ( Figure 1 ) . The Tat–AFF4 interaction surface is centered around Tat K28 , which is acetylated in vivo to regulate HIV transcription ( Kiernan et al . , 1999 ) . Mutations in AFF1 corresponding to the Tat–AFF4 interface in the crystal structure reduce scaffold binding and transcription stimulation functions . Compared to the AFF4-P-TEFb complex , AFF4 structural segments in the Tat-AFF4-P-TEFb complex undergo unanticipated 1–2 Å backbone shifts to avoid unfavorable close contacts with Tat ( Figure 2 ) . These shifts in AFF4 alter the dimensions of the Tat-binding pocket in AFF4-P-TEFb that may serve as a site for targeting HIV transcription inhibitors ( Schulze-Gahmen et al . , 2013 ) . The Tat-AFF4-P-TEFb complex structure also reveals that the scaffold and Tat combine to partially fold the CycT1 TRM . This interaction is an example of multiple natively disordered protein segments coming together to form a structure that depends on the other protein subunits ( Figure 3B ) . The conformations of the CycT1 TRM are strikingly different in the Tat-AFF4-P-TEFb complex compared to other structures containing P-TEFb . The TRM residues 253–260 are disordered in the Tat-P-TEFb crystal structure ( Tahirov et al . , 2010 ) . In P-TEFb alone , crystal contacts stabilize the TRM in a distinct conformation that partially occludes the AFF4 binding site and precludes contacts between C261 and the Tat Zn2+ ion ( Garber et al . , 1998; Baumli et al . , 2008 ) . The position of the TRM in the Tat-AFF4-P-TEFb structure , including exposed side chains for residues R251 , R254 , W258 , and R259 , is consistent with functional data ( Garber et al . , 1998 ) . For example , alanine mutants of eight out of thirteen residues in the CycT1 segment 250–262 abolished or reduced in vitro binding of Tat-CycT1 to TAR . In addition , UV crosslinking of the CycT1 Tat–TAR complex ( in the absence of AFF4 ) suggests that TAR makes direct contacts with the TRM ( Richter et al . , 2002 ) . The CycT1 TRM is not known to make RNA contacts during host transcription in uninfected cells , but the TRM contacts the loop in HIV TAR ( Wei et al . , 1998; Richter et al . , 2002 ) . To probe the complementarity of TAR with the Tat–CycT1 hybrid surface in the Tat-AFF4-P-TEFb complex structure , we manually docked the TAR-argininamide solution structure ( PDB ID 1ARJ ) ( Puglisi et al . , 1992; Aboul-ela et al . , 1995 ) with Tat-AFF4-P-TEFb . Since , we could not model the side chains of CycT1 K253 , R254 , W258 , and R259 in the electron density , we added these side chains in the most frequently observed conformation . The calculated electrostatic potential surface shows a large positively charged region covering the CycT1 TRM and part of Tat including R49 ( Figure 5B ) . Positioning the TAR loop to contact the TRM enables the RNA bulge ( U23–U25 ) to reach the Tat arginine rich motif ( ARM ) that makes essential contacts ( Figure 5—figure supplement 1 ) ( Weeks and Crothers , 1991 ) . The dimensions of all components fit well and provide a plausible working model for the TAR interaction with Tat-AFF4-P-TEFb . AFF4 not only promotes folding of the TRM , but the scaffold also increases TAR binding in vitro by 30-fold . This enhancement of RNA affinity is likely to be an indirect consequence of ordering the CycT1 TRM , because TAR is too small to make direct contacts with AFF4 residues in the complex . In Hela cells , alanine mutations in AFF1 ( M60A/L61A ) corresponding to the AFF4–TRM interaction site ( M55/L56 ) abolish Tat ( C22G ) -SEC assembly and eliminate the inhibitory activity of AFF1 1-308 in Tat transactivation ( Lu et al . , 2014 ) . Taken together , both the Tat-AFF4 interface and the TRM–AFF4 interaction are essential for WT Tat activity . These results show that AFF4 contributes to the selective recruitment of SECs by Tat and TAR through a two-stage mechanism . First , direct interactions involving AFF4 , Tat , and the TRM increase the binding affinity of Tat for AFF4-P-TEFb 11–fold ( Schulze-Gahmen et al . , 2013 ) . Second , AFF4 binding to Tat-P-TEFb indirectly constrains the TRM conformation , increasing the binding affinity for TAR 30-fold . The enhancements of affinity in these successive steps together lead to a 330-fold increase in Tat/TAR binding by AFF4-P-TEFb over P-TEFb . This preference of Tat and TAR for the SECs , in concert with the Tat-stimulated release of scaffold-P-TEFb complexes from the 7SK snRNP ( Lu et al . , 2014 ) , ensures preferential , simultaneous recruitment of the full complement of elongation factors required for efficient HIV transcription .
P-TEFb and TAT-P-TEFb were expressed in High5 insect cells using recombinant baculovirus infections . We co-expressed human CDK9 1–330 and human cyclin T1 1–264 with and without HIV-1 Tat 1–57 . Baculovirus generation and High5 cell infections were described in detail previously ( Schulze-Gahmen et al . , 2013 ) . AFF4 fragments 2–73 and 2–98 with an N-terminal TEV-protease-cleavable His-tag were expressed in E . coli ( Schulze-Gahmen et al . , 2013 ) . Tat-P-TEFb and AFF42-73 were purified separately following procedures described recently ( Schulze-Gahmen et al . , 2013 ) . Tat-P-TEFb and AFF42–73 were combined at a 1:1 . 4 ( mol/mol ) ratio , concentrated to 0 . 6 ml , and injected onto an analytical Superdex S200 size exclusion column equilibrated with 25 mM Na-HEPES pH 7 . 4 , 0 . 2 M NaCl and 1 mM DTT . The center fractions of the eluted four-protein peak were used for crystallization . A synthetic TAR fragment encompassing nucleotides 18–44 was purchased from IDT ( San Diego , CA , USA ) . The RNA was annealed at 0 . 1 mg/ml in 20 mM Na HEPES pH 7 . 3 , 100 mM KCl , 3 mM MgCl2 . Best results were obtained by heating the RNA at 75°C for 2 min , followed by rapid cooling on ice . The purity of the RNA , analyzed by denaturing and native 10% polyacrylamide gel electrophoresis , was at least 95% . The purified Tat-AFF4-P-TEFb complex was combined with refolded synthetic TAR in a 1 . 1-fold molar excess . MgCl2 was added to a 3 mM final concentration . The protein–RNA complex was concentrated in an Amicon Ultra filter with a 30 kDa cutoff to about 10 mg/ml protein concentration . The presence of TAR was confirmed on silver-stained polyacrylamide gels . Crystals were grown in sitting drops from 1 . 0 µl protein–TAR complex combined with 1 . 0 µl reservoir solution . The drops were equilibrated against 2 . 4 M sodium formate , 10 mM MgCl2 at 18°C . After equilibrating for 24 hr , diluted microseeds from previous crystallization experiments were added with a hair . Seeding produced single hexagonal crystals ( 0 . 15 × 0 . 15 × 0 . 2 mm ) . Crystals were soaked in 2 . 8 M sodium formate , 10 mM MgCl2 , 30% glycerol for cryoprotection and flash frozen in liquid nitrogen . X-ray data were collected at Beamline 8 . 3 . 1 at the Advanced Light Source at the Lawrence Berkeley National Laboratory ( MacDowell et al . , 2004 ) . The reflections were processed using HKL2000 ( Otwinowski and Minor , 1997 ) ( Table 1 ) . The Rmerge of the data is high , requiring additional tests of the symmetry of the crystals . Processing the data in space group P1 yielded an Rmerge of 18 . 6% , and analyzing these data using the program Pointless ( Winn et al . , 2011 ) confirmed the presence of the symmetry operators of space groups P6522 or P6122 . Moreover , no twinning was detected using the programs Pointless and Xtriage ( Adams et al . , 2010 ) . Rmerge for the P6522 data set decreases to 8 . 4% after excluding reflection with I/sigI <3 . In addition , the values for Rp . i . m . ( Weiss and Hilgenfeld , 1997; Weiss et al . , 1998 ) and CC1/2 ( Karplus and Diederichs , 2012 ) , as well as the refinement statistics indicate that the data were processed correctly . These tests support the conclusion that the space group was assigned correctly , and the unusually high value of Rmerge was due to the presence of many weak reflections and the high redundancy of the data set . The structure was determined by molecular replacement with PHENIX ( Adams et al . , 2010 ) using the AFF4-P-TEFb complex ( PDB ID 4IMY ) as the search model . After three molecules were placed in the a . u . , the Tat structure from the P-TEFb-Tat complex ( PDB ID 3MI9 ) was combined with the molecular replacement solution by superimposing the CycT1 molecules of the different complexes . The model was refined with PHENIX . refine ( Adams et al . , 2010 ) , using gradient minimization with weight optimization , maximum likelihood targets , non-crystallographic symmetry constraints , individual B-factors , and TLS parameters . Automatic refinement was alternated with manual rebuilding using Coot ( Emsley and Cowtan , 2004 ) . Although the crystallization experiments were set up with Tat-AFF4-P-TEFb-TAR complex , we did not find electron density for TAR RNA , nor was there room for TAR in the crystal lattice . The high-salt conditions of the crystallization probably dissociated the TAR RNA from the protein complex . In the final model , density was missing for residues 1–7 and 89–96 in CDK9 mol1 and mol2 , and residues 1–7 and 92–95 in CDK9 mol3 . Density was also absent for residues 1–6 and 262–264 in all three CycT1 molecules , and residues 50–57 in all three Tat molecules . For AFF4 mol1 ( mol3 ) density for residues 2 , 22–32 , 70–73 ( 2–3 , 22–33 , 70–73 ) was missing , while AFF4 mol2 was missing density for residues 2–33 and 70–73 . The ATP binding pocket of CDK9 contained extra density although ATP was not included in the crystallization . The density was modeled as adenosine . Least squares fitting of protein structures were performed with Coot ( Emsley and Cowtan , 2004 ) and the program ProFit by Dr A Martin from University College London . Profit uses the McLachlan fitting algorithm ( McLachlan , 1982 ) . Potential hydrogen bonds were identified with the program CONTACT in CCP4 ( Winn et al . , 2011 ) and manually inspected . The assay was performed as described ( He et al . , 2010 ) . Briefly , nuclear extracts prepared from HeLa cells transfected with the indicated expression constructs were incubated with anti-Flag or anti-HA agarose beads ( Sigma , St . Louis , MO ) for 2 hr before extensive washing and elution . The HeLa-based NH1 cell line containing an integrated HIV-1 LTR-luciferase reporter construct ( He et al . , 2010 ) was transfected with the indicated expression constructs . At 48 hr post transfection , total cell lysates were prepared from approximately 106 cells per sample and luciferase activity was measured . Refolded synthetic TAR ( nucleotides 18–44 ) was radioactively labeled with 32P-γ–ATP using T4-polynucleotide kinase . A 10-µl reaction was prepared with 200 nM TAR , 0 . 3 mCi 32P-γ–ATP ( 7000 Ci/mmol , MP Biomedicals , Sohon , OH ) , and 10 units of T4-polynucleotide kinase ( New England BioLabs , Ipswich , MA ) in 70 mM Tris/HCl pH7 . 6 , 10 mM MgCl2 , 2 mM DTT . After incubating at 37°C for 1 hr , 25 µl of annealing buffer ( 20 mM Na HEPES pH 7 . 3 , 100 mM KCl , 3 mM MgCl2 ) were added to the reaction . The mixture was purified twice over Illustra G25 spin columns ( GE Healthcare , Piscataway , NJ ) to remove free nucleotides . The purified labeled TAR was diluted to 10 nM ( 3000–5000 cpm/ µl ) with annealing buffer for storage and use in EMSAs . Binding reactions ( 10 µl ) were carried out in 20 mM Na HEPES pH 7 . 3 , 100 mM KCl , 3 mM MgCl2 , 1 mM DTT , 4% glycerol with 12 units RNasin ( Promega , Madison , WI ) , 10 µg/ml BSA , and 5 µg/ml poly ( I:C ) ( Invivogen , San Diego , CA ) . Each reaction contained 100 pM labeled TAR RNA . Reactions were incubated at 20°C for 30 min and RNA-binding complexes were separated on a pre-run 6% polyacrylamide gel in 0 . 5x TBE ( 100 V , 1 hr at 4°C ) . Gels were dried , exposed to storage phosphor screens , and measured on a Typhoon phosphorimager ( GE Healthcare , Piscataway , NJ ) . | The rate at which many genes are expressed as proteins depends on a process called transcriptional elongation . This process takes place as the region of DNA that defines the gene is transcribed into an RNA molecule , and it is catalyzed by an enzyme called RNA polymerase II . However , this process often stalls shortly after it starts , and another enzyme called a positive transcription elongation factor is needed to restart it . The human immunodeficiency virus ( HIV ) is a retrovirus that hijacks the gene expression machinery inside immune cells in order to replicate itself . To do this as efficiently as possible , the elongation factor needs to restart the transcription process as quickly as possible . To ensure that this happens the virus produces a protein called Tat that binds to the short region of RNA that has already been made . At the same time the Tat protein also combines with other proteins to form a multi-protein machine called the super elongation complex . Other proteins in the super elongation complex include a ‘scaffold’ protein called AFF4 , a positive elongation factor called P-TEFb , and at least two additional transcription factors . Until recently researchers did not know how the Tat protein was able to recruit super elongation complexes to the correct location without recruiting other complexes that contained similar protein subunits . Now Schulze-Gahmen et al . have shed new light on this mystery by working out the crystal structure of the complex formed by the elongation factor P-TEFb when it forms a complex with the Tat protein and a scaffold protein called AFF4 . The results show that direct interactions between the Tat and scaffold proteins help to recruit the super elongation complex to the correct location . The three-way interactions between Tat , AFF4 , and P-TEFb form a binding surface that encourages the complex to bind to the RNA . Overall , Schulze-Gahmen et al . show that the super elongation complex is much more likely to be recognized by the Tat protein and then bind to RNA than just the elongation factor on its own . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | AFF4 binding to Tat-P-TEFb indirectly stimulates TAR recognition of super elongation complexes at the HIV promoter |
Behavior of extinct organisms can be inferred only indirectly , but occasionally rare fossils document particular behaviors directly . Brood care , a remarkable behavior promoting the survival of the next generation , has evolved independently numerous times among animals including insects . However , fossil evidence of such a complex behavior is exceptionally scarce . Here , we report an ensign scale insect ( Hemiptera: Ortheziidae ) , Wathondara kotejai gen . et sp . nov . , from mid-Cretaceous Burmese amber , which preserves eggs within a wax ovisac , and several freshly hatched nymphs . The new fossil is the only Mesozoic record of an adult female scale insect . More importantly , our finding represents the earliest unequivocal direct evidence of brood care in the insect fossil record and demonstrates a remarkably conserved egg-brooding reproductive strategy within scale insects in stasis for nearly 100 million years .
Brood care is an altruistic trait that evolved to enhance the fitness of offspring at a cost to the parents and represents a breakthrough in the adaptation of organisms to their environment ( Tallamy , 1984; Clutton-Brock , 1991; Gilbert and Manica , 2010 ) . Fossil evidence of such an ephemeral behavior is extremely rare , reported mainly in dinosaurs ( Meng et al . , 2004; Varricchio et al . , 2008 ) , ostracods ( Siveter et al . , 2007 , 2014 ) , arachnids ( Engel and Grimaldi , 2014 ) , but rarely in insects . Until now only two putative examples in Mesozoic insects have been described based on fossils ( Peñalver et al . , 2012; Cai et al . , 2014 ) . Although phylogenetic analyses suggest some ancient insects evolved brood care ( e . g . , Korb et al . , 2012 ) , only fossils provide unequivocal direct evidence . In this study , we report on an exceptionally preserved insect from mid-Cretaceous Burmese amber , which represents the earliest unequivocal direct evidence of brood care in the insect fossil record and sheds new light on the early evolution of such behavior .
Order Hemiptera Linnaeus , 1758 . Family Ortheziidae Amyot and Serville , 1843 . Wathondara kotejai gen . et sp . nov . Simon , Szwedo and Xia . The generic name refers to Wathondara—goddess of earth in Buddhist mythology from Southeast Asia . Gender: feminine . The species is named after the late Polish entomologist Jan Koteja in recognition of his significant contribution to the study of both extant and fossil scale insects . BA14011 . The amber piece preserves an adult female with eggs , six first-instar nymphs , and a weevil . It is polished in the form of a flattened ellipsoid cabochon , clear and transparent , with diameter about 11 mm , height about 5 mm , and weight about 0 . 8 g . Specimen is from Kachin Province in northern Myanmar . Burmese amber has been dated biostratigraphically from late Albian to Cenomanian ( about 105 to 95 million years old ) , based on an ammonite and palynology ( Cruickshank and Ko , 2003; Ross et al . , 2010 ) . The U-Pb dating of zircons from the volcaniclastic matrix of the amber gave an age of 98 . 8 ± 0 . 6 million year ( Shi et al . , 2012 ) . Body elongate oval , dorsoventrally flattened ( seems to be natural condition ) . Antennae 8-segmented; first segment straight , elongate , thicker than others , trapezoid in shape; second segment cylindrical distinctly longer than others; antennal segments III–VIII with numerous setae of hair-like and fleshy types , some of them almost as long as apical setae on segment VIII . Apical segment cylindrical with long and stout apical seta and additional shorter subapical seta situated on subapical projection . Legs slender; trochanter fused with femur; tibia and tarsus fused , with numerous spine-like setae . Tarsal claw without denticles; claw digitules hair-like , thin , and short . Amber specimen preserves adult female with about 60 elliptical eggs ( 0 . 3 mm long , 0 . 2 mm wide ) in wax ovisac , and six first-instar nymphs near adult ( Figure 1 ) . Adult body elongate oval , 6 mm long , 2 mm wide ( with ovisac ) . Antenna about 1 . 2 mm long , inserted ventrally at frontal margin , with eight segments ( Figure 2C ) ; first , the widest , trapezoidal; second , the longest , cylindrical; segments III to VII , club-like; segment VIII cylindrical , with subapical projection; length of antennal segments ( in mm ) I—0 . 162; II—0 . 350; III—0 . 130; IV—0 . 130; V—0 . 145; VI—0 . 115; VII—0 . 125; VIII—0 . 220 . Segments I and II covered with scarce hair-like setae , segments III to VII with subapical fleshy setae on external margins and hair-like setae; some setae almost as long as apical setae of VIII segment; segment VIII with subapical seta on projection ( 0 . 075 mm long ) and apical seta ( 0 . 097 mm long ) . Eyes not easily observable , placed on short stalks . Labium apparently 2-segmented . Legs well-developed; tarsal claw small , slightly bent , without denticle . Anal ring visible on dorsum . Spiracles , wax glands , and most body setae not visible . Wax secretion of ortheziid type , with nine pairs of marginal lobes , two triangular frontal lobes , three elongate triangular median lobes , nine submedian pairs , and posterior lobes ( Figure 3 ) . Wax covering made partly translucent due to preservation in amber , completely covering dorsum ( Figure 2A ) . Ovisac well-developed , 3 mm long , 2 . 1 mm wide ( Figure 2D ) . Six associated first-instar nymphs are of similar size , 0 . 3 mm long , 0 . 2 mm wide , with only 6-segmented antennae ( Figure 1D ) . 10 . 7554/eLife . 05447 . 003Figure 1 . Wathondara kotejai gen . et sp . nov . Simon , Szwedo and Xia from mid-Cretaceous Burmese amber . ( A ) Habitus in dorsal view , stacked image with a blue filter . ( B ) Habitus in ventral view , stacked image with a green filter . ( C ) Habitus in ventral view , stacked image with a green filter . Note the weevil under the adult . The numbers 1–6 represent six first-instar nymphs . ( D ) Enlargement of a nymph in ( C ) . Scale bars of ( A , B and C ) represent 1 mm; scale bar of ( D ) represents 0 . 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 05447 . 00310 . 7554/eLife . 05447 . 004Figure 2 . Wathondara kotejai gen . et sp . nov . Simon , Szwedo and Xia from mid-Cretaceous Burmese amber . ( A ) Habitus in dorsal view . The numbers 1–9 indicate nine marginal wax lobes . ( B ) Habitus in ventral view . ( C ) Enlargement of the antenna in ( B ) . ( D ) Enlargement of the ovisac in ( B ) . Scale bars of ( A , B , and D ) represent 1 mm; scale bar of ( C ) represents 0 . 25 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 05447 . 00410 . 7554/eLife . 05447 . 005Figure 3 . Drawing of brooding Wathondara kotejai gen . et sp . nov . Simon , Szwedo and Xia in ventral view . The ovisac and wax covering are made nearly transparent by preservation in amber . DOI: http://dx . doi . org/10 . 7554/eLife . 05447 . 005
Scale insects ( Coccoidea ) , with about 7800 species , are highly diverse , and most of them are obligatory plant parasites often of economic importance ( Ben-Dov et al . , 2014 ) . They exhibit many unusual features of morphology , reproduction , and life history and are thus considered as some of the most evolutionarily fascinating organisms amongst insects ( Wappler and Ben-Dov , 2008; Hodgson and Hardy , 2013 ) . The female life cycle involves two or three actively feeding instars prior to the adult stage , and adult females are wingless , resembling the immature stages ( Gullan and Cook , 2007 ) . In contrast , adult males are delicate , ephemeral insects with simplified wing venation ( Hodgson and Hardy , 2013 ) . Scale insects separated from their sister-group , the aphids , at least by the Middle Permian based on the earliest occurrence of Aphidomorpha ( Szwedo et al . , 2015 ) , with the fossil record probably extending back to the Middle Triassic ( trace fossils in Figure 4 ) . However , their fossil record is over-dominated by males entrapped in fossil resins , and fossil adult females are very scarce—probably because they are commonly sedentary or sessile on host plants ( Koteja , 2000 ) . To our knowledge , the new fossil is the only Mesozoic record of an adult female , the next oldest being from the late Eocene Baltic amber ( Koteja and Żak-Ogaza , 1988 ) . 10 . 7554/eLife . 05447 . 006Figure 4 . The evolution of scale insects . Hypothetical phylogeny based on Hodgson and Hardy ( 2013 ) and Vea and Grimaldi ( 2015 ) ( extinct families omitted ) . Matsucoccidae , Ortheziidae , Margarodidae are commonly considered as the most primitive families ( Vea and Grimaldi , 2015 ) , but their phylogenetic relationships are still unresolved ( e . g . , Gullan and Cook , 2007; Hodgson and Hardy , 2013 ) . Thick lines indicate the known extent of the fossil record . ( 1 ) Undescribed scale marks on plants from the Middle Triassic Dont Formation of Italy ( T Wappler , personal observation , October 2014 ) ; ( 2 ) scale marks on plants from the Late Triassic Molteno Formation of South Africa ( Labandeira , 2006 ) ; ( 3 ) putative , undescribed scale insect from the Late Jurassic ( Grimaldi and Engel , 2005 ) ; Red star represents Wathondara kotejai from mid-Cretaceous Burmese amber . An early diversification of scale insects probably occurred during the end of the Jurassic or earliest Cretaceous ( blue area ) , and later radiations are probably closely related to the rise of angiosperms and ants ( Grimaldi and Engel , 2005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05447 . 006 Wathondara kotejai is unambiguously referable to Ortheziidae , as evidenced by its general habitus with its body covered with wax plates , ensign-like ovisac , stalked eyes , and well-developed legs ( Kozár , 2004 ) . Furthermore , W . kotejai shares two potential synapomorphies with Recent and Tertiary , crown-group Ortheziidae: differentiated apical and subapical setae on the last antennal segment , and trochanter and femur fused ( Vea and Grimaldi , 2012 ) . Two Cretaceous genera have been tentatively attributed to Ortheziidae: Burmorthezia Vea and Grimaldi in mid-Cretaceous Burmese amber is considered as an extinct sister group to the crown-group Ortheziidae ( Vea and Grimaldi , 2012 ) , while Cretorthezia Koteja and Azar in Early Cretaceous Lebanese amber is probably a stem group of scale insects ( Koteja and Azar , 2008; Hodgson and Hardy , 2013 ) or an extinct group within Ortheziidae ( Vea and Grimaldi , 2015 ) . Additionally , a putative female ( Cretorthezia sp . ) from Burmese amber was tentatively identified as an ortheziid , and its systematic position is still uncertain ( Koteja and Azar , 2008; Vea and Grimaldi , 2012 ) . Our new fossil supports the view that crown-group Ortheziidae are present in the mid-Cretaceous . The seventh , eighth , ninth , and posterior wax lobes of W . kotejai are distinctly extended and cover the ovisac dorsally . The thick wax cover not only protects the adult female but also serves to shelter her eggs and first instars . Extant ortheziid females have a band of pores on the ventral side of the abdomen , which secrete a waxy ovisac . The eggs and hatched nymphs are protected within the ovisac ( Figure 3 ) , as in extant ortheziids and monophlebids ( Vogelsang and Szklarzewicz , 2001 ) . In extant species , the young nymphs hatch within this ovisac and remain there for a few days until they have acquired a thin covering of wax secretion ( visible in our specimens as a slight white pubescence on the fossil nymphs ) , then crawl out through a split in the wax at the distal end of the ovisac ( Gullan and Kosztarab , 1997 ) . Extant first instars are mobile and serve as principle agents for dispersion and seeking out suitable feeding sites ( Koteja , 2001 ) . This egg brooding is widely considered to be a primitive form of brood care ( e . g . , Royle , et al . , 2012; Wong et al . , 2013; Siveter et al . , 2014 ) . Some Early Cretaceous cockroaches have been reported with an ootheca attached ( Grimaldi and Engel , 2005 , Figure 7 . 72 ) . However , it is not definitive evidence of egg brooding , because some cockroaches subsequently deposit the ootheca in a suitable crevice . Therefore , W . kotejai provides the earliest unequivocal evidence of brood care in insects . Brood care is considered to have evolved independently in at least seven insect orders ( Wong et al . , 2013 ) . This remarkable behavior takes several forms of which the most common are egg brooding and offspring attendance . Scale insects have evolved a variety of methods to protect their eggs and hatched nymphs from unfavorable abiotic conditions and natural enemies . Some extant species ( e . g . , Diaspididae , some Pseudococcidae ) even possess an ovoviviparous form or pseudoplacental viviparity ( Gullan and Kosztarab , 1997 ) . In addition to Ortheziidae , ovisacs occur in many other families , for example , Monophlebidae , unrelated Coccidae and many Pseudococcidae , in all of which the secretions of a variety of tubular ducts and disc-pores combine to form the ovisac ( Ben-Dov et al . , 2014 ) . These various types of ovisacs have evolved convergently to protect their offspring from wet and dry conditions , honeydew contamination , and natural enemies ( Gullan and Kosztarab , 1997 ) . Our study demonstrates that these significant behavioral and morphological adaptations , associated with considerable maternal investment , were already well established by the mid-Cretaceous . Many extant Ortheziidae females feed on roots and fungal mycelia or mosses and lichens ( Vea and Grimaldi , 2012 ) and ‘run about’ in forest litter with the eggs carried in the ovisac attached to their bodies . This is considered to be the most primitive habit in scale insects ( Gullan and Kosztarab , 1997; Koteja , 2001 ) , and similar brood care behavior also occurs in other early scale insects , for example , Margarodidae and Matsucoccidae ( Koteja , 2001 ) . Therefore , this behavior probably has an early origin and maybe a synapomorphy for scale insects . Flowering plants and ants are thought to be important drivers for radiation of the most diverse advanced group , the neococcoids ( Grimaldi and Engel , 2005 ) . However , both factors are absent in the evolutionary history of basal groups of scale insects ( Figure 4 ) . Brood care , greatly promoting the survival of offspring ( Royle et al . , 2012 ) , could therefore have been an important driver for the early radiation of scale insects which occurred during the end of the Jurassic or earliest Cretaceous ( Figure 4 ) . Despite a great taxonomic diversity of extant insects with brood care ( Wong et al . , 2013 ) , direct evidence of such behavior has been reported only in Cenozoic ambers ( Peñalver et al . , 2012 ) . The new fossil is unique in providing evidence of ovarian and juvenile developmental stages in a fossil insect . More remarkably , it represents the earliest direct evidence of brood care in insects and highlights the long-term stasis of this behavior in archaeococcoids , extending nearly 100 million years .
The electronic edition of this article conforms to the requirements of the amended International Code of Zoological Nomenclature , and hence the new names contained herein are available under that Code from the electronic edition of this article . This published work and the nomenclatural acts it contains have been registered in ZooBank , the online registration system for the ICZN . The ZooBank LSIDs ( Life Science Identifiers ) can be resolved and the associated information viewed through any standard web browser by appending the LSID to the prefix ‘http://zoobank . org/’ . The LSID for this publication is: urn:lsid:zoobank . org:pub: 01114A99-586C-4BAD-9F84-4E3FDBFFD86F . The electronic edition of this work was published in a journal with an ISSN , and has been archived and is available from the following digital repositories: PubMed Central , CLOCKSS , Steinmann Institute at University of Bonn , Natural History Museum ( London ) , University of Gdańsk , and Nanjing Institute of Geology and Palaeontology ( CAS ) . | Many animals care for and protect their offspring to increase their survival and fitness . Insects care for their young using a range of strategies: some dig underground chambers for their young , whilst others carry their brood around on their own bodies . However , it was unclear when these strategies first evolved in insects . Now Wang et al . report that they have discovered the earliest fossil evidence of an insect caring for its young , in the form of a female insect preserved with her brood in a specimen of ancient amber . The amber comes from northern Myanmar , where amber deposits are around 95–105 million years old . The fossilised insect is an adult female scale insect with a cluster of around 60 eggs on her abdomen . Six young scale insect nymphs are also preserved in the same piece of amber . Wang et al . named this newly discovered species Wathondara kotejai , after an earth goddess in South-East Asian Buddhist mythology and the late Polish entomologist Jan Koteja . Most scale insect fossils found to date have been males . Fossilised adult females are scarcer , most likely because female scale insects are wingless and less mobile and therefore less prone to accidental burial . The fossil reported by Wang et al . is therefore a rare find , and it is also sufficiently well preserved to reveal that the female's eggs are contained within a wax-coated egg sac . Today there are many species of scale insects , most of which are parasites of plants and many are economically important pests of trees and shrubs . In living relatives of W . kotejai , females use a similar wax coating to protect themselves and their offspring: young nymphs hatch inside the egg sac and remain there for a few days before emerging into the outside world . This new fossil provides a unique insight into the anatomy and life cycle of a long-extinct insect; it also demonstrates that brood care in insects is an ancient trait that dates back to at least around 100 million years ago at the height of the age of the dinosaurs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"short",
"report",
"evolutionary",
"biology"
] | 2015 | Brood care in a 100-million-year-old scale insect |
How cargoes move within a crowded cell—over long distances and at speeds nearly the same as when moving on unimpeded pathway—has long been mysterious . Through an in vitro force-gliding assay , which involves measuring nanometer displacement and piconewtons of force , we show that multiple mammalian kinesin-1 ( from 2 to 8 ) communicate in a team by inducing tension ( up to 4 pN ) on the cargo . Kinesins adopt two distinct states , with one-third slowing down the microtubule and two-thirds speeding it up . Resisting kinesins tend to come off more rapidly than , and speed up when pulled by driving kinesins , implying an asymmetric tug-of-war . Furthermore , kinesins dynamically interact to overcome roadblocks , occasionally combining their forces . Consequently , multiple kinesins acting as a team may play a significant role in facilitating smooth cargo motion in a dense environment . This is one of few cases in which single molecule behavior can be connected to ensemble behavior of multiple motors .
Kinesin is part of a cytoskeletal motor family that moves cellular cargoes primarily to the cell periphery ( microtubule plus end ) . It is important in key cellular processes like cell division and signaling ( Gross et al . , 2002 ) . It is also implicated in several neurological disorders ( McLaughlin et al . , 2016 ) . Due to advances in single molecule microscopy and force measurement techniques , transport properties by a single kinesin are well understood ( Veigel and Schmidt , 2011 ) . For example , kinesin-1 , the prototypical kinesin , moves 8 . 4 nm per ATP consumed , in a hand-over-hand motion , walking about 100 steps before detaching and traveling at a speed of ~0 . 77 μm/sec in vitro ( Cai et al . , 2007 ) , and equal- or higher speed in vivo ( Block et al . , 1990; Stamer et al . , 2002; Yildiz et al . , 2004 ) . A single kinesin also exerts up to ~6 pN force ( Svoboda and Block , 1994 ) , and importantly , has an asymmetric run-length and velocity with regard to the direction of force on the microtubule ( Figure 1A ) ( Coppin et al . , 2002; Milic et al . , 2014 ) . However , a cell is extremely dense , filled with proteins ( ~300 mg/ml ) , and only 60–80% water volume ( Albe et al . , 1990 ) , resulting in a high viscosity and elastic modulus ( Berret , 2016 ) . Despite this , large cargos move at virtually the same rate as in a water-based in-vitro environment ( Furuta et al . , 2013; Howard et al . , 1989 ) . How is this possible ? We argue that it is due to the action of multiple motors acting on a single cargo . Multiple motor transport is important in cellular trafficking ( Blehm et al . , 2013; Gross et al . , 2002; Hendricks et al . , 2012; Holzbaur and Goldman , 2010 ) . Electron microscopy shows that ~1 to 7 motors are bound to cellular cargoes ( Gross et al . , 2007 ) . Motors are inter-dependent ( Gross et al . , 2002 ) , such that the impairment of one motor type ( e . g . dynein ) causes severe impairment in the other ( e . g . kinesin ) ( Gross et al . , 2002 ) . Run length and stall forces are typically greater for multiple vs single motor , although not necessarily in proportion ( Holzbaur and Goldman , 2010 ) . Theoretical studies predict tension between multiple motors carrying the same cargo ( Arpağ et al . , 2014 ) . This tension , we argue , allows for their ability to efficiently bypass roadblocks , which has yet to be shown experimentally . To understand multiple motor transport , it is crucial to probe the force and motion of each kinesin , plus that of the cargo motion . However , no current single molecule assay can achieve this . For example , atomic force microscopy ( AFM ) allows recording of only single motors walking ( Kodera and Ando , 2018 ) ; current fluorescence and optical trap assays can measure overall forces and positions of multiple motors , but not every single motor ( Derr et al . , 2012; Jamison et al . , 2010 ) . The inability to probe all molecular motors may have led to the differing conclusions on the effect of motor number on run length ( Block et al . , 1990; Derr et al . , 2012; Efremov et al . , 2014; Shubeita et al . , 2008; Vershinin et al . , 2007 ) . Two in vitro assays come close to probing the behavior of all participating kinesins . One uses a programmable DNA origami with up to seven kinesins attached ( Derr et al . , 2012 ) . Another uses immobilized GFP-labeled kinesins and tracks gliding microtubules with quantum dots attached ( Leduc et al . , 2007 ) . These studies accurately count the number of kinesins on cargo , but not their behavior . One pertinent question is whether single-motor asymmetry plays a role in multi-motor transport . The Block group found an asymmetry when studying the run length of single kinesin as a function of force in an optical trap ( Milic et al . , 2014 ) ( Figure 1A ) . This asymmetry is between driving kinesin , which experiences force opposite to its walking direction , and resisting kinesin , which experiences force in the same direction as its movement . They found that the run length is greater for driving kinesin ( 50 nm at −6 pN rising rapidly to 1100 nm at 0 pN ) vs . resisting kinesin ( ~100 nm at 2 pN slowly falling to 10 nm at 20 pN ) ( Figure 1A ) ( Milic et al . , 2014 ) . Similarly , the Vale group saw kinesin velocity to be slower for driving-kinesin than resisting-kinesin ( Figure 1B ) ( Coppin et al . , 2002 ) . Whether this run-length and velocity asymmetry occurs when multiple driving and resisting kinesins are carrying a single cargo was not determined , and whether inter-kinesin interactions cause this , will be described here . To overcome current experimental limitations , a number of authors have done simulations which provided insights into multiple-motor behavior . One simulation showed that the asymmetric property of single kinesin under load leads to an average of one-third of kinesin’s resisting in multi-motor situation ( Arpağ et al . , 2014 ) . In addition , the tension was between 0 to 15 pN between the kinesins carrying the cargo ( Arpağ et al . , 2014 ) . Other simulations showed that , on a single cargo being driven by multiple motors , force-dependent detachment of motors ( Arpağ et al . , 2014 ) , particularly the resisting motors ( Nelson et al . , 2014 ) , is important for cargo speed . However , the accuracy of simulations heavily depended on the particular models used ( Arpağ et al . , 2014; Kunwar and Mogilner , 2010; Kunwar et al . , 2008; Xu et al . , 2013 ) . Another question is how well multiple motors cooperate together . A number of experimental studies found negative cooperativity of kinesin ( Furuta et al . , 2013; Jamison et al . , 2010 ) . Negative cooperativity , in this case , means that cargo transport does not fully benefit from having two motors present due to a certain inter-motor inhibition that reduces the motor-filament binding energies in the system . Motor run length and average detachment forces would therefore not increase as dramatically as expected in cases without this interference ( Jamison et al . , 2010 ) . However , exactly how the on- and off-rates of each kinesin ( e . g . run length and binding duration ) change to give rise to negative cooperativity is unknown . Negative cooperativity does not mean that additive forces of two or more kinesins can never occur . It can happen occasionally , and can be observed in optical trap assays ( Jamison et al . , 2010; Vershinin et al . , 2007 ) . How relevant additive forces are to cellular transport is unclear . It is possible that additive forces help kinesin get unstuck upon roadblock encounter , but this has not been shown . Another possibility is that additive forces can detach kinesin ( s ) from the cargo , which has also not been shown , but may facilitate cargo transport in a crowded cell ( Conway et al . , 2012 ) . It is possible that additive forces happen only transiently , but , nevertheless , may be important to bypass roadblocks . In this work , we have developed an in-vitro assay , which we call a force-gliding assay , details of which are discussed below . It allows direct observation of individual kinesin-1’s motion , velocities and forces , acting as a dynamical team of multiple kinesin motors ( 1- ~ 8 ) , transporting a common cargo ( microtubule ) , whose position and velocity can be measured . We can directly observe the attachment and detachment of individual motors from the microtubule , and find that kinesin exists in two distinct states , one driving the microtubule , the other resisting . We observe an asymmetric run length and velocity response to load of driving-kinesins and resisting-kinesins , indicating a non-zero tension between kinesins carrying the same cargo . This leads to one-third of kinesins always resisting . We further show that multiple kinesins exhibit negative cooperativity through decreasing run-length of individual kinesins as more kinesins participate in transport . Lastly , multiple kinesins can combine forces that help in overcoming the roadblocks . We conclude that an asymmetric tug-of-war , with negative cooperativity and additive forces , defines collective kinesin transport and can potentially help with uninterrupted cargo transport inside the cell . We note that our assay measures dynamic interactions between multiple kinesins and a cargo . The ability to do get dynamic interactions may be crucial because it may be only transient interaction which are necessary to bypass roadblocks . Whether the tension embedded within an in vivo system , and whether other molecular motors such as dynein and myosin have a similar cooperative behavior , remains to be seen .
We developed a modified microtubule gliding assay , called a force-gliding assay , capable of measuring the direction and the magnitude of force exerted by individual kinesin-1 on the microtubule cargo in real-time ( Figure 1C ) . The assay can also be used to estimate the attachment-detachment dynamics of each kinesin . Each kinesin was labeled with a quantum dot ( QD ) at a ratio of 1:1 , and attached by a 1565-base pair dsDNA molecule , which acts like a non-linear spring , to a non-stick polyethylene-glycol ( PEG ) coverslip . Issues of multiple DNA-binding to a QD and non-fluorescent QD have been minimized ( see Appendix 1 ) . A fluorescently-labeled microtubule ( shown moving to the right in Figure 1C ) serves as the cargo and is moved by kinesins at saturating ( 1 mM ) ATP conditions . The points at which kinesins are attached to the glass coverslip through the dsDNA are defined as the ‘equilibrium positions’ . The position of each kinesin could then be monitored with nanometer accuracy via a tracking algorithm similar to Fluorescence Imaging with One Nanometer Accuracy ( FIONA ) ( Yildiz and Selvin , 2005 ) ( see Materials and methods ) . Single particle tracking was possible because the kinesins were placed at a distance greater than the diffraction limit apart , allowing many kinesins to be monitored individually , but simultaneously . The force acting on the kinesins can be estimated from the DNA extensions using the extensible Worm Like Chain ( eWLC ) ( Lee and Thirumalai , 2004 ) . Upon tracking the positions of kinesins during cargo transport , we found that kinesins were in one of two possible states , shown in Figure 1C . One state speeds up the microtubule—called the ‘driving kinesin’ , where the kinesin pushes the microtubule in the direction of the microtubule gliding ( right in Figure 1C ) , causing the kinesin to be displaced in the opposite direction of microtubule gliding ( left of kinesin’s equilibrium position in Figure 1C ) . The other state , called the ‘resisting kinesin’ , slowes the microtubule down , resulting in the kinesin being displaced in the direction of microtubule gliding ( right of kinesin’s equilibrium position in Figure 1C ) . Therefore , we define signs such that the resisting kinesins have positive displacements from their equilibrium position and the driving kinesins have negative displacements . Both fluorescent signals from the QD on the kinesin and the organic fluorophores on the microtubule allowed several minutes-long recording . Using the force-gliding assay , we estimated the displacement of each kinesin with respect to their equilibrium positions , as well as measure the corresponding microtubule velocity in real time . To investigate the dynamics of multiple kinesins in detail , we tracked the motion of each microtubule when it was being moved by multiple kinesins ( see Materials and methods ) . Figure 2 is an example of one of these cases where a microtubule ( green ) is transported by three kinesins , labeled #1 , 2 and 3 ( see Video 1 ) . Figure 2A shows the microscope images at different time points during the microtubule transport . The top image of Figure 2A ( t = 0 s ) shows that three kinesins are at their equilibrium position ( marked by yellow arrows , when they are not transporting any microtubule ) . When the microtubule transport begins at t = 1 . 4 s , kinesins start dynamically fluctuating around their respective equilibrium positions . We observed that individual kinesin while transporting the microtubule will be in either the driving or resisting states , and can switch between them . For example , kinesin three is in the resisting state at t = 5 . 2 s and transitions to the driving state at t = 14 s ( Figure 2A ) . Figure 2B shows the microtubule velocity and kinesin displacement analysis of the same three kinesins . xc shows the kymograph of the microtubule . We calculated the microtubule velocity , plotted in Figure 2B2 , by tracing the microtubule kymographs . We tracked the position of individual kinesins , plotted their displacements and correlated them to the microtubule velocity in Figure 2B3 . The dashed horizontal black lines in the kinesin displacement plots represent the equilibrium position of each kinesin . Dashed blue boxes , expanded at the bottom of Figure 2B4 and B5 , show the resisting and driving displacements above ( positive ) and below ( negative ) the equilibrium lines in detail . Four time points ( same as in Figure 2A ) are depicted by dashed yellow vertical lines . At t = 0 s , all kinesins are at their equilibrium position . At 1 . 4 s , a microtubule appears; kinesin #1 and #3 are still at their equilibrium positions , while kinesin #2 rapidly starts driving the microtubule—therefore it has negative displacement . Due to kinesin #2 assuming a driving role , the microtubule velocity increases to ~850 nm/sec . At 5 . 2 s , kinesin #3 is resisting while kinesin #1 and #2 are driving , maintaining the velocity of microtubule at ~850 nm/sec . At 14 s , kinesin #3 is driving the microtubule while kinesin #2 is resisting the motion , resulting in a ~ 600 nm/sec microtubule velocity . This shows that kinesins dynamically switch their roles , which affects the cargo velocity . The fact that kinesins , while working together , can switch states from driving to resisting and vice versa , can be made quantitative . We measured many such transitions ( N = 685 ) , similar to that shown in Figure 2A and B . The average transition rate of a single kinesin is 4 . 9 transitions/min ( Figure 2C ) . The majority ( 52% ) of these transitions are drive-to-drive transitions , during which kinesin drives , then return to equilibrium while still being attached to or after detaching from the microtubule , and then drives again . The drive-to-resist transitions and resist-to-drive percentage contribute 16% of all transitions each . Together they make up approximately one-third of all transitions . This shows the highly dynamic nature of kinesin working in a team . Rather than one kinesin taking the lead all the time , kinesins are constantly changing roles between driving and resisting . Next , we ask if there is a difference in run length and duration between kinesins in the driving and the resisting states , as observed in the single kinesin case of Milic et al . ( 2014 ) . From the Milic et al . data ( Figure 1A ) , we can say that without tension , that is with F = 0 , the forward and backward run lengths are equal to each other ( ~1 μm ) . Run length asymmetry for multiple motor transport will only be observed if there is non-zero tension between driving and resisting kinesins carrying the same cargo , as shown in Figure 1A for F ≠ 0 . With multiple kinesins , Figure 3A shows that driving kinesin stays attached to microtubule for an average of 3 . 0 ± 0 . 21 s , compared to 2 . 15 ± 0 . 15 s for resisting kinesin . Figure 3B shows that the average run length of driving-kinesin is 2 . 31 ± 0 . 18 µm , compared to 1 . 42 ± 0 . 09 µm for resisting kinesin . Our observation of asymmetry in run length and duration suggests there is tension between driving and resisting kinesins with multiple motors driving a single ( microtubule ) cargo . To quantify the tension when multiple kinesins carry the same cargo , we used a semi-quantitative method to estimate the force ( or tension ) on the kinesin based on the extensible WLC model ( Lee and Thirumalai , 2004 ) . Using a persistence length of 50 nm , a dsDNA contour length of 532 nm and a distance-offset of 20 nm to account for the size of QD , proteins and other attachment agents , we obtained the average force as a function of time on kinesins that is shown in Figure 3C . A total of 1478 driving kinesins and 573 resisting kinesins was used . Before 0 . 8 s , which is the time for dsDNA to stretch near its contour length , assuming a kinesin velocity of 800 nm/s , the tension is on average less than ~0 . 4 pN . After 0 . 8 s , the tension increases and stays between below ~4 pN until kinesin detaches . Therefore , we estimate the tension between kinesins to be between 0 and 4 pN . Because of the uncertainty in DNA extension ( due to uncertainty in distance offset , ~0–20 nm , and equilibrium point determination , ~0–40 nm: see Appendix 2 ) , there is significant uncertainty in the force results—from sub-pN up to tens of pN—depending on the DNA extension ( Sup . Figure 1 ) . We , therefore , sought to verify our force results with the published literature . In particular , Milic et al . , found that the relevant forces are 0 to −4 pN for driving kinesins and from 0 to 4 pN for resisting kinesin . Outside of this range ( F< −4 pN or F > 4 pN ) , the run length difference is small between driving and resisting kinesin . ( Note that Milic et al . refer to the load being carried by the kinesin , where the hindering load corresponds to the driving kinesin and the assisting load to the resisting kinesin: see Figure 1B ) . Consequently , our results of 0–4 pN tension between driving and resisting kinesins are in semi-quantitative agreement with the result of Milic et al . Knowing that run length in multiple kinesin transport is asymmetric , we then asked how this affects the fraction of driving and resisting kinesins . We expected that the longer run length of driving kinesin would result in more kinesins being in the driving state than the resisting state . Indeed , we found that there is a constant fraction—about 2/3—of kinesins which are driving . This means that about 1/3 of the kinesins are resisting , regardless of the total number of kinesins ( from 1 to 8 ) attached to the microtubule ( Figure 3D ) . The fact that about 1/3 of the kinesins are in the resisting mode has significant implications , to be discussed in the Discussion section . Next , we asked if the velocity of the driving and resisting kinesins change differently when pulled in opposite directions , that is is there a velocity asymmetry ? For a single kinesin , Coppin et al . ( 2002 ) found that the answer was yes: directional loads—which , in their case , are imposed by pulling the kinesin forward or backwards on a stationary microtubule—will slow down driving kinesin , and speed up resisting kinesin . There is a force range around zero , from −2 pN to +2 pN , where there is no change , that is there is no asymmetry ( Figure 1B ) . Such an asymmetric effect on the velocity has not been shown for multiple kinesins . In multiple motor case , asymmetry can arise because a kinesin can be in the driving mode or the resisting mode and apply tension ( force ) . We find that for multiple kinesins , the driving kinesins slow down to match the cargo ( microtubule ) rate , and the resisting kinesins speed up to match the cargo ( microtubule ) rate . This speeding up or slowing down can be seen in the plateaus for the kinesin displacement graphs . Two examples of the plateaus are shown as the 2’ region in the resisting kinesin trace ( Figure 2B4 ) and driving kinesin trace ( Figure 2B5 ) . What is happening is because there is asymmetry in the tension direction between the two types of kinesins , there is a concomitant asymmetric change in velocity ( see also Appendix 7 ) . To test the prevalence of velocity asymmetry in multiple kinesin transport , we computed the average velocity of 1478 driving kinesins and 573 resisting kinesins relative to the microtubule , summarized in the result of Figure 3E . We observe slowing down of driving kinesins and speeding up of resisting kinesins , confirming velocity asymmetry in bulk kinesin behavior . On average , the driving kinesins start at a velocity of ~1200 nm/s and resisting kinesins at ~500 nm/s relative to microtubule ( =absolute velocity of the microtubule minus absolute velocity of the kinesin ) ( Figure 1A ) . Driving kinesins slow down and resisting kinesins speed up to ~830 nm/s , which is the average microtubule velocity ( taken from the velocity statistics discussed later in Figure 4C ) . Note that at ~1 s , the driving kinesin velocity intersects with resisting kinesin velocity , which is due to kinesin detachment , as explained in Appendix 3 and Figure 4—figure supplement 4 ) . Overall , we show that velocity asymmetry can be observed in individual driving and resisting traces that are involved in multiple kinesin transport . Such velocity asymmetry leads to resisting and driving kinesins speeding up or slowing down , respectively , to match the velocity to the cargo speed . How well do kinesins work together ? Some studies found evidence for negative cooperativity: both the run length and force increase when more kinesins carry cargo , but not in proportion to the increase in the number of kinesins ( Furuta et al . , 2013; Jamison et al . , 2010 ) . A few studies have shown that when multiple kinesins are on a cargo , only a fraction of them are the primary drivers ( Furuta et al . , 2013; Jamison et al . , 2010 ) . We find this to be true in our assay , with only 2-fold increase in the total number of kinesins driving and resisting the microtubule for every 16-fold increase in the kinesin surface concentration ( Figure 3—figure supplement 1 ) . What causes this negative cooperativity ? We find that it is due to a decrease in the run length and binding duration of driving and resisting kinesins as more kinesins participate in transport . Figure 3F shows that individual kinesin duration and run length decreases by 2-to-3-fold when kinesin surface concentration increases from 0 . 03 to 0 . 46 kinesin/µm2 ( ~16 fold increase in surface concentration ) . ( A decrease in both the run length and duration indicates that the kinesins have approximately the same velocity regardless of kinesin surface concentration , since velocity is displacement ( run length ) over time ( duration ) ) . This shorter duration and run length likely results from the higher tension between kinesins as more kinesins are transporting the cargo . Thus , using a force-gliding assay , we find that negative cooperativity of kinesins can be explained by the shorter run length of each participating kinesin . Motors need to walk in dense cellular environment and overcome myriads of roadblocks to transport the cargo to its destination ( Lakadamyali , 2014 ) . To dissect the mechanism of multiple kinesin-based transport in the presence of roadblocks , we introduced roadblocks in our force gliding assay . We labeled microtubules with commercially available quantum dots ( ~20 nm in size ) using streptavidin-biotin linkage ( Sheung et al . , 2018 ) . These are of a different color ( QD605 ) than the QDs on kinesins ( QD705 ) and could be separately detected . Consequently , we could track the motion of kinesins , microtubule , and roadblocks simultaneously . Past studies found that a single kinesin either detaches immediately or pauses when it encounters a roadblock ( Schmidt et al . , 2012; Schneider et al . , 2015 ) . These are readily observable in the force-gliding assay , as shown in Figure 3—figure supplement 1 , Figure 4—figure supplement 1 , and Figure 4—figure supplement 2 , and Video 2 , Videos 3 and 4 . What is unclear is whether multiple kinesins can help cargo navigate through roadblocks . With the force-gliding assay , we can now observe multiple kinesins rescue a microtubule decorated with a QD as roadblock . The multiple kinesins do this by detaching from the microtubule a kinesin stuck at a roadblock , and then collectively guiding the microtubule forward . Figure 4 shows this . Figure 4A is a series of snapshots of Video 5 . A microtubule decorated with roadblock ( QD605 ) is depicted by a yellow line ( marked with ‘microtubule’ at 12 s ) . The QD605 roadblock gets stuck at kinesin K1 ( making it difficult to individually separate their fluorescence since they overlap ) and eventually interacts with other kinesins ( K2–K6 ) to rescue the microtubule motion . The six kinesin-QDs are shown in yellow-orange spots , enclosed in white circles . At t = 12 s to t = 72 s , the microtubule is fluctuating around K1 , which is stuck at a roadblock on the microtubule . At t = 84 s , K4 catches the microtubule and makes failed attempts to drive the stuck microtubule . In the process , K4 stretches and straightens the microtubule , which becomes aligned with four other kinesins . More kinesins then start to interact with the microtubule and rescue its motion at t = 186 s . Figure 4B and its zoomed inset give a more detailed look . From 0 s to ~ 80 s , the QD605 roadblock on the microtubule was stuck at K1 . From ~ 80 s onwards , K4 attempted to drive the microtubule , but the microtubule did not budge until ~ 130 s , when K5 joined the drive attempt . K5 triggered a repositioning of the microtubule in the off-axis direction ( see Video 5 ) , and after a short lag , at ~ 143 s , the microtubule started moving ( time A ) , driven by K4 with slight resistance from K1 . At time B , the microtubule gets stuck again , because K1 starts to show greater resistance . This happens until time C , when K3 joins K4 in driving , and successfully propelling the microtubule forward . At time D , the microtubule stops again , now because of K4 resisting . This takes place up to time E , when K4 resumes driving . In summary , Figure 4 shows that when a cargo ( microtubule ) is stuck at a roadblock ( here a QD605 attached to the microtubule ) , multiple kinesins , by dynamically interacting with the microtubule , can rescue the motion of the cargo . After scanning through our entire dataset , we found 18 similar examples in which a microtubule is stuck at a kinesin and is rescued by the action of other kinesins . We show one additional example in Sup . Figure 5 . Next , we studied the statistics of kinesin motion in the presence of the roadblocks . Specifically , we varied the amount of roadblocks in the force gliding assay . In Figure 4C , we show the results of placing various concentrations of roadblocks ( 0 , 30 , and 100 nM streptavidin-QD ) onto the biotinylated microtubule . Roadblock amounts of 0 , 30 , and 100 nM corresponded to linear densities ~ 0 , 0 . 75 , and 2 . 5 roadblock-QDs/micron of microtubule length . For simplicity , here we denote roadblock concentration in terms of nM . Upon plotting the velocity histograms of the microtubule , regardless of the number of kinesins ( one through ~ 8 ) , we find that the histograms are well represented by two Gaussian populations: stuck ( ~0 nm/s ) and fast ( 800–950 nm/s ) . As the roadblock concentration increases from 0 to 30 to 100 nM , the proportion of stuck microtubule increases from 7% to 12% to 29% . The average velocity of the fast microtubules also decreases from 926 nm/sec to 857 nm/sec to 806 nm/sec , consistent with previous roadblock studies ( Chaudhary et al . , 2018 ) . Taken together , roadblocks reduce the average cargo velocity and induce pauses in a cargo moved by a team of kinesin . Because we use a gliding assay where we can determine the number of kinesins bound to the microtubule , we can further break down the bulk velocity histograms . We can estimate the variation in microtubule velocity with increasing number of kinesins transporting it ( Figure 4D ) . When there are no roadblocks present ( Figure 4D left column , 0 nM roadblocks ) , we observe that peak of the velocity remains almost constant as the number of kinesins increase . This result agrees with previous studies on multiple motors ( Derr et al . , 2012 ) . On the other hand , the 100 nM roadblock data ( Figure 4D right column ) show a different result . Here , as the number of kinesin increases from 1 to 2–3 to 4–5 , the proportion of stuck microtubules decreases from 35% to 31% to 13% , respectively , and eventually to zero when there are 6–8 kinesin moving the microtubule . The velocities of fast microtubules remain constant at around 805 nm/s regardless of motor number . This shows that roadblock-induced pauses can be reduced , eventually to nearly zero , by having more motors available to drive the cargo . One possible explanation of why having more motors makes the motion smoother is that forces of kinesins add up to induce higher tension on the stuck kinesin , which increases its detachment rate . Our results with optical trap studies in the next section further reinforces this explanation . In support of the hypothesis that more motors help cargos overcome roadblocks and reduce cargo pauses , we observed nine instances when a resisting kinesin is stuck at a roadblock while other kinesins keep driving the microtubule , causing the stuck kinesin to detach from the surface . We present one of such case in Figure 5 . As the resisting kinesin detaches from the coverslip , it starts moving with the microtubule , driven by four other driving kinesins ( Figure 5A and Video 6 ) . Figure 5A shows the images of the kinesin ( yellow arrow ) that detaches from the surface due to the force of four other driving kinesins ( marked by white lines ) . At 64 . 8 s , four driving kinesins are pulling on a microtubule while one resisting kinesin is holding the microtubule back ( Figure 5A ) . The resisting kinesin is then ripped off of the glass coverslip and the kinesin travels with the microtubule , as can be seen at 72 . 0 s and 84 . 6 s . Presumably , the weakest link—in this case , the digoxigenin:anti-digoxigenin antibody bond for the DNA linker , shown in the insert of Figure 1C , is rupturing ( Neuert et al . , 2006 ) . This example shows that four kinesin are applying sufficient force to break the digoxigenin:anti-digoxigenin bond . To test the magnitude of the force and compare it to the force that a single kinesin can exert ( ~6 pN ) , we then took the same 1 . 56 kb dsDNA and its linkages , and stretched it in an optical trap until the digoxigenin:anti-digoxigenin linkage ruptures ( Figure 5B; Figure 5—figure supplement 1; Appendix 4 ) . We pulled the dsDNAs at 10 nm/sec and 100 nm/sec loading rates ( see Figure 5—figure supplement 2 for rationale ) and found half of the tethers ruptured at 30 and > 45 pN , respectively ( Figure 5C; Appendix 5 ) . These values are above the stall force of a single kinesin . Hence , the few ( ~4 ) driving kinesins observed pulling on the detaching kinesin may exert additive forces beyond what a single kinesin can exert . There is , however , an important caveat to this argument: due to the broad survival distribution , there is a small but non-negligible probability for the unbinding to occur at low forces , meaning the resisting kinesin could have released at < 6 pN ( see Appendix 6 ) .
The force-gliding assay allows us to understand the interaction of individual kinesins with one another and with the microtubule . We found that underneath the seemingly smooth transport of a microtubule cargo , like that shown in Figure 2B2 , multiple kinesins attached and detached frequently , switching their states from driving to resisting and vice versa ( Figure 2B3 ) . Thus , we uncover the hidden dynamics ( i . e . the attachment-detachment , and the changing states ) of kinesins previously unseen in simple motor walking or gliding assays . Kinesin has two distinct states during cargo transport and these states have asymmetric run length and velocity response to load , depending on the load ( or tension ) direction . Driving kinesin dwells longer than resisting kinesin on the microtubule , and will slow down under hindering load caused by tension between kinesins . This tension is key in multiple motor transport , and our assay is the first experimental assay to show its presence between motors . Tension allows kinesins to communicate with one another: driving kinesin feels the tug of resisting kinesin and vice versa through this tension . This allows the driving kinesin to slow down and resisting kinesin to speed up , so that both driving and resisting kinesins walk in-sync at the same speed , that is the speed of the cargo . Our assay also allows us to measure the force on individual kinesin motor . In our assay , we show that the tensions between kinesins vary between 0 to 4 pN . Since our assay uses a long and flexible linker ( 1565-base pair DNA ) , we therefore predict that in cells , where shorter and stiffer linkers ( adaptor proteins ) are employed , the tension will likely increase . This is because kinesin can travel further at low force with a long and flexible linker , and since there is a finite chance of dissociating at every step , kinesin can dissociate prematurely before high forces are reached when long and flexible linkers are used . This is confirmed by comparing our result with a simulation by Arpağ et al . ( 2014 ) , where shorter and stiffer linker ( elasticity of 0 . 2 pN/nm after 40 nm stretching ) is used . For a rough comparison , the DNA in our assay has an elasticity of 0 . 00066 pN/nm , 300-fold more elastic , at half the DNA contour length ( i . e . 532/2 = 266 nm ) —though the elasticity decreases at larger DNA extension ( 0 . 19 pN/nm at 516 nm ) . Arpağ et al . found that the tension in their simulation varies between 0 to 10 pN for driving kinesin and 0 to 15 pN for resisting kinesin ( Figure 5—figure supplement 3 ) , larger than the 0 to 4 pN tension measured in our system . Even though the tension in our system is likely lower than in the cell , measurements such as the fraction of kinesin resisting will likely remain the same . In fact , Arpağ et al . predicted that one-third of the driving/resisting kinesins will be resisting ( Figure 5—figure supplement 4 ) ( Arpağ et al . , 2014 ) , similar to the result of our measurement . Inside cells , motors are bound to a lipid cargo , which may have varying fluidity . Grover et al . ( 2016 ) found that gliding velocity of microtubules transported by membrane-bound kinesin decreases with increasing membrane fluidity . This is due to the slippage of motor anchors in the lipid bilayer . Our studies are carried out with DNA tethers attached to a rigid , planar cargo ( the coverslip surface ) , so cargo fluidity is not accounted for . Nevertheless , we predict that membrane fluidity will reduce the tension between the kinesins , though the ratio between driving and resisting kinesins will likely remain the same . Another study on Myosin Va shows that fluid membrane allows vesicle travel at velocities up to twice that of a single motor ( Nelson et al . , 2014 ) . This is due to the biased detachment of resisting motors . In fluid membrane , slippage of motor anchor causes resisting motors to lag behind the driving motor , and the detachment of this resisting , lagging motor will cause the vesicle to spring forward . Without a fluid membrane , resisting motors can detach when it is ahead or behind the driving motor , and the vesicle will spring backward or forward once detached ( no biased detachment ) . Our study found that resisting kinesin motors will detach faster than driving motors , and we predict that , just like Myosin Va , a team of kinesin motors will travel faster on a fluid vesicle due to the biased detachment of resisting motors . The one-third resisting kinesin fraction ( and two-third driving kinesin fraction ) is a consequence of the asymmetric run length to load , since resisting kinesin detaches faster than driving kinesin at the same force . These resisting kinesins were not accounted for in published animations of kinesins working as a team ( Bolinsky et al . , 2006; Condeelis et al . , 2014 ) . What is the significance of a one-third ( ~33% ) resisting kinesin ? In particular , what happens when the fraction of resisting kinesin is 0% ( no resisting kinesin ) or 50% ( equal fraction as driving ) ? We propose that the one-third resisting kinesin may be an optimal strategy to increase cargo run length and reduce tension between kinesins . If there is no resisting kinesin ( 0% ) , any kinesin would have detached immediately as soon as it feels an assisting load . An immediate benefit is the absence of drag due to resisting kinesin . The downside is that there will be less kinesin attached to microtubule . Since more kinesin ( even resisting ones ) can help maintain attachment of cargo to microtubule . Less kinesin means that the cargo run length will be shorter . As a rough estimate , a 0% resisting kinesin strategy will reduce the run length a two-motor system from a two-motor run length of 8 μm ( Vershinin et al . , 2007 ) to a one-motor run length of 1 μm ( Vershinin et al . , 2007 ) , assuming the resisting kinesin will detach at the slightest resisting force . Thus a 0% resisting kinesin strategy will reduce drag , but also reduce run length . If there is an equal number of resisting kinesin as driving kinesin ( 50% resisting ) , the rate of resisting kinesin detaching will be the same as driving kinesin . If we have a kinesin with the resisting kinesin having the same detachment rate as the current driving kinesin , this would mean that the kinesin will stay resisting for a longer time before detaching , thus generating larger drag . The upside is that since the resisting kinesin can stay longer , the cargo run length will also be longer . Thus a 50% resisting kinesin strategy increase cargo run length , but also increases drag . Since 0% resisting kinesin has low drag and short run length , while 50% resisting kinesin has high drag and long run length , we hypothesize that a 33% resisting kinesin strategy is a strategy that balances the drag and run length . It will be interesting to investigate this more thoroughly through future simulation studies . We found that kinesin cooperates negatively in our assay: that is , when there are more available kinesins to bind the microtubule , only a fraction of them are actively driving or resisting at any one time . Past studies infer the net negative cooperativity of kinesin through stall forces of two kinesins ( Jamison et al . , 2012; Jamison et al . , 2010 ) . With force gliding assay , we are able to directly observe that when kinesin surface concentration is increased 16-fold , there is only a 2-fold increase in the total number of kinesins driving and resisting the microtubule ( Figure 3—figure supplement 1 ) . Furthermore , we find that this negative cooperativity arise because the run length and binding duration of driving and resisting kinesins decrease as more kinesins participate in transport ( Figure 3F ) . Even though kinesins cooperate negatively , their forces can still combine additively on occasion ( Jamison et al . , 2010; Vershinin et al . , 2007 ) . In our assay , probing the effect of roadblocks on multiple motor cargo transport , we found that when one kinesin is stuck on a roadblock on a microtubule , other kinesins combine forces to help detach the stuck kinesin ( Figure 4A , B ) . As a result , having more kinesins on the cargo leads to smoother cargo velocity and reduction of stuck cargo events ( Figure 4D ) . We also observed a limited number of cases where kinesin on the coverslip surface is detached and moved with roadblocks due to the combined forces of multiple kinesins ( Figure 5A ) . These cases were more common at high roadblock concentration . We infer from such cases that the combined force was so high that it led to the detachment of the resisting kinesin from the surface . Using an optical trap , we indirectly quantified the force in the system and concluded that kinesins can augment their forces in the presence of roadblocks and , thus , can help in overcoming the roadblocks ( Figure 5B , C ) . Figure 6 is an example of how multiple kinesins might interact with a single cargo within a cell . When one kinesin moves slower than the cargo and becomes resisting ( Figure 6A ) , the assisting forces from the cargo tend to increase this kinesin’s speed or cause it to release rapidly , allowing the cargo to experience minimal drag force . Surprisingly , there appears to be ~ 35% resisting kinesins , causing a continuous tug-of-war among the kinesins , which tends to maintain an appreciable tension between kinesins . Kinesins can also rapidly switch between driving and resisting , leading to a fairly continuous and uninterrupted cargo motion forward . Against roadblocks , force augmentation of multiple kinesins may lead to large forces , causing detachment of resisting kinesin from cargo or microtubule ( Figure 6B ) . By combining the motion and the force of single kinesins , we can connect the single molecule behavior of kinesins with their ensemble behavior . Whether the tension embedded within an in vivo system , and whether other molecular motors such as dynein and myosin have a similar cooperative behavior , remains to be seen .
Truncated kinesin with 888 amino acids ( K888 ) from the mouse kinesin heavy chain ( accession number BC090841 ) with a C-terminal biotin-tag and FLAG epitope , and mouse kinesin light chain ( accession number BC014845 ) were cloned separately into the baculovirus transfer vector pAcSG2 ( BD Biosciences ) for recombinant virus production . Sf9 cells were infected with recombinant viruses , grown , harvested , lysed and purified using a published protocol for K888 homodimer kinesin ( Tjioe et al . , 2018 ) . Briefly , infected cells in growth medium supplemented with 0 . 2 mg/ml biotin were harvested after 72 hr and lysed by sonication in lysis buffer ( 10 mM imidazole , pH 7 . 4 , 0 . 3 M NaCl , 1 mM EGTA , 5 mM MgCl2 , 7% ( w/v ) sucrose , 2 mM DTT , 0 . 5 mM 4- ( 2-aminoethyl ) benzenesulfonyl fluoride , 5 µg/ml leupeptin ) prior to clarifying at 200 , 000 x g for 40 min . The supernatant was applied to a FLAG-affinity column ( Sigma-Aldrich ) and washed with 10 mM imidazole , pH 7 . 4 , 0 . 3 M NaCl , 1 mM EGTA . Specifically-bound protein was eluted in the same buffer containing 0 . 1 mg/ml FLAG peptide . Fractions of interest were combined , concentrated with an Amicon centrifugal filter device ( Millipore ) , dialyzed against 10 mM imidazole , pH 7 . 4 , 0 . 2 M NaCl , 1 mM tris ( 2-carboxyethyl ) phosphine TCEP ) , 55% ( v/v ) glycerol , 1 mM DTT , 1 µg/ml leupeptin , 50 µM MgATP , and flash frozen for storage at −80°C . MiCA purification was performed to obtain one to one binding of biotinylated kinesin with streptavidin-QD 655 or 705 . Briefly , kinesin K888 is mixed with 3x excess QD so that each QD has one or no kinesin bound 95% of the time . This reaction is allowed to incubate for > 10 min on ice in a BSA-taxol buffer ( 1 mM THP ( 71194 , EMD Millipore ) , 20 μM Paclitaxel ( Cytoskeleton , Inc ) and ~30 nM ATP ( Magnesium salt , A9187 , Sigma Aldrich ) in DmB-BSA ( dynein motility buffer ( 30 mM HEPES , 50 mM KAcetate , 2 mM MgAcetate , 1 mM EGTA , pH 7 . 2 ) supplemented with 8 mg/mL BSA ) ) at 220 nM final K888 concentration and 660 nM final QD concentration . Excess QD is then removed through MiCA purification , which uses moderately positive magnetic beads ( i . e . magnetic amine beads coated with PEG-amine to reduce highly positive amine charge ) that bind to short microtubules to form MiCA capture beads . This is done by mixing 5 μL sonicated GMPCPP microtubule ( 1 mg/mL short microtubules prepared from 97% pure tubulin ( HTS03-A , Cytoskeleton , Inc ) , stored at −80°C and thawed right before use ) with 8 μL PEG-amine magnetic beads ( 10 mg/mL , prepared as previously published ) with its buffer removed after a magnetic pull to leave only the pellet . After 5 min incubation in an end-to-end rotator at room temperature , the MiCA capture bead is washed 2x with 8 µL BSA-taxol buffer and reconstituted in 1 µL BSA-taxol buffer to give ~ 1 . 5 µL final bead volume . Next , 6 µL kinesin-QD ( 220 nM kinesin ) is mixed with the 1 . 5 µL MiCA capture bead and 1 . 2 µL AMP-PNP ( 8 mM ) , and the mixture is allowed to incubate for 5 min at room temperature in an end-to-end rotator . The AMP-PNP causes kinesin-QD to bind strongly to MiCA capture beads . The mixture is then washed 3x with 8 µL BSA-taxol buffer and 8 µL elution buffer ( 2 mM ATP in BSA-taxol buffer ) is added . After 5 min incubation in an end-to-end rotator at room temperature , the eluant is extracted , yielding approximately 80 nM kinesin-QD ( assuming 50% purification yield ) . 22 square millimeter coverslips were sonicated in 1M KOH and plasma cleaned , then aminosilanized and reacted with N-hydroxysuccinimide ( NHS ) ester modified polyethylene glycol ( PEG ) that includes 1% biotin-PEG-NHS ( Roy et al . , 2008 ) . The attachment of biotin-PEG to the surface is thus covalent . Double sided tape pieces were sandwiched between a thoroughly washed glass slide and the coverslip to make the imaging channels . 600 nM streptavidin was flowed into the channel and incubated for 5 min . The channel was washed with DMB-BSA buffer ( 30 mM HEPES , 50 mM KAcetate , 2 mM MgAcetate , 1 mM EGTA , 8 mg/ml BSA , pH 7 . 4 ) . 10 nM biotinylated anti-digoxigenin ( Abcam ) was flowed into the chamber and incubated for 5 min followed by a subsequent wash with DMB-BSA buffer to remove excess anti-digoxigenin-biotin . MiCA purified kinesin-QD was mixed with eight times less DNA ( IDT ) to minimize conjugation of multiple DNA molecules to single kinesin-QD . The biotin end of DNA was conjugated with the kinesin-QD and the other end with digoxigenin remained free . Kinesin-QD-DNA was flowed into the chamber and the digoxigenin end of the DNA was conjugated with the Anti-digoxigenin on the surface . The chamber was incubated with excess biotin to saturate all the streptavidin binding sites in the chamber and subsequently washed with DMB-BSA . The number of kinesins on the surface were optimized such that they were sufficiently away from each other and could be tracked individually . Finally , the imaging buffer containing the polymerized microtubules , saturating ATP and deoxygenating agents ( pyranose oxidase + glucose ) was flowed in the imaging chamber and movies were acquired . Five sets of experiments were collected at 0 . 03 , 0 . 06 , 0 . 11 , 0 . 23 , and 0 . 46 kinesin/ μm2 . For each set , four to five movies ( technical replicates ) were imaged , each at 0 . 2 s exposure time for 1500 frames . For doing the roadblock experiments , biotinylated-microtubules were incubated with equal volume of streptavidin-QD605 ( Thermo Fisher Scientific ) solution of varying concentration ( 0 nM , 30 nM , 100 nM QD605 ) . Roadblock incubated microtubules were used in the imaging buffer for doing the roadblock experiments . Three sets of experiments were collected at no roadblock , 30 nM roadblock , and 100 nM roadblock ( QD ) concentration . For each set , four to five movies ( technical replicates ) were imaged , each at 0 . 2 s exposure time for 1500 frames . Double-stranded DNA was synthesized through PCR amplification of a 1 . 565-kbp segment of the pBR322 plasmid ( New England Biolabs ) , using forward and reverse primers conjugated with a 5’ biotin and a 5’ digoxigenin , respectively ( Integrated DNA Technologies ) and a high-fidelity master mix ( New England Biolabs ) . The PCR product was purified with a PCR cleanup kit ( QIAGEN ) . For optical trapping experiments , 2 or 2 . 4 µL of 0 . 05 nM dsDNA were incubated for an hour at room temperature with 5 µL of 0 . 2% w/v streptavidin-coated beads ( Spherotech ) . Beads were diluted in approximately 300 µL of buffer ( 100 mM Tris , 20 mM NaCl , 3 mM MgCl2 , pH 7 . 6 ) for delivery to the optical traps through bead channels in a custom flow chamber ( Whitley et al . , 2017 ) . In the trapping channel of the flow chamber , dual-trap optical tweezers were used to trap a DNA-coated streptavidin bead in one trap , and a bead ( Spherotech ) coated with digoxigenin ( Roche Diagnostics ) in the other . The beads were repeatedly brought together until a DNA tether formed . Once a dsDNA tether was formed , a force-extension curve was collected by moving one trap away from the other at a constant rate ( 10 nm/s or 100 nm/s ) over a pre-set distance , then returning at the same rate to the initial position . Most tethers ruptured during the force ramp . Rupture is expected to occur primarily at the linkage betwen digoxigenin and anti-digoxigenin , as rupture forces previously reported for this linkage ( under different buffer conditions ) have been lower than for the biotin-streptavidin linkage ( Merkel et al . , 1999; Neuert et al . , 2006 ) . Each resulting force-extension curve was fitted to the extensible worm-like chain model ( Camunas-Soler et al . , 2016; Wang et al . , 1997 ) to verify that only one molecule was present and that it behaved correctly ( Figure 5B ) . The maximum forces experienced by the single dsDNA tethers were determined and plotted as a survival distribution ( Figure 5C ) . The optical trapping experiments were conducted in a microfluidic flow chamber , in a channel containing trapping buffer consisting of 76% DmB-BSA ( 30 mM HEPES , 5 mM MgSO4 , 1 mM EGTA , pH 7 . 0 and 8 mg/ml BSA ) , 10 µM biotin , 100 µM ATP , 100 µM THP , 2 µM Paclitaxel , and an oxygen scavenging system ( Landry et al . , 2009; Swoboda et al . , 2012 ) ( final concentrations in buffer: 32 mg/mL glucose , 0 . 58 mg/mL catalase ( from Aspergillus niger: Millipore Sigma , formerly EMD Millipore , 219261-100KU , 5668 U/mg ) , 1 . 16 mg/mL pyranose oxidase ( from Coriolus sp . : Sigma P4234-250UN , 12 . 2 U/mg ) , 400 µM Tris-HCl and 2 mM NaCl ) . Total Internal Reflection Fluorescence Microscopy ( TIRFM ) was performed with an inverted light microscope ( Olympus IX71 ) equipped with two EMCCD cameras ( iXon DU-897E ) , a TwinCam ( Cairn Research ) to split two colors into two separate cameras , a 100x magnification oil immersion objective ( Olympus UPlanSApo , NA 1 . 40 ) , and a green laser ( 10 mW power , Coherent OBIS 532 nm attenuated with a neutral density filter with optical density of 1 . 0 . The excitation light was reflected with a 556 long-pass dichroic ( T556lpxr-UF3 UltraFlat , Chroma ) and cleaned up with 532 nm long-pass filter ( BLP01-532R-25 , Semrock ) . Fluorescence from QD and microtubule were split with a 685 nm long-pass filter ( T685lpxr-UF3 , UltraFlat , Chroma ) in TwinCam . QD655 , QD705 and a combined QD625 and HyLite 488 Microtubule emission were filtered using a 655/40 nm , 710/40 , and 600/80 nm ( BrightLine , Semrock ) band-pass filter , respectively . Images were recorded with 0 . 2 s exposure time for all experiments , except for the experiment shown in Figure 2 , where 0 . 1 s exposure time is used . An EM-gain between 10 and 300 was used , adjusted to maximize the signal collected without saturating the camera . No additional magnification was used for all experiments , except one shown in Figure 2 , where 1 . 5x additional magnification is used . The pixel size for each image is thus 16 , 000 nm ( the actual camera pixel dimension ) /100 x objective magnification = 160 nm for most images , and 16 , 000 nm / 150 x total magnification = 106 . 7 nm for those with 1 . 5x additional magnification . Fluorescent images obtained from the two channels of TwinCam were mapped onto each other using a transform file obtained from a set of nanohole images as previously described ( Tjioe et al . , 2018 ) . The 512 × 512 pixels of combined image were visualized in Fiji ( plugin-rich package of ImageJ ) and gliding instances of every microtubule were cropped and saved . Point locations of all kinesin-QD were detected with TrackMate ( Tinevez et al . , 2017 ) , a plugin within Fiji , using a Laplacian of Gaussian ( LoG ) detector , with estimated blob diameter of 4 pixels ( 160 nm/pixel ) , threshold of 50 , and sub-pixel localization turned on . Simple LAP ( Linear Assignment Problem ) algorithm within TrackMate was used to track all detected spots , with maximum distance for frame-to-frame linking of 4 pixels , maximum distance for track segment gap closing of 4 pixels , and maximum frame gap of 20 frames . All spots detected and tracked were then saved as a csv file for subsequent analysis in Matlab . See Video 7 for detailed tutorial . In Matlab , kinesin-QD locations from TrackMate were imported , along with cropped images of microtubule and kinesin-QD . The Matlab code , FFGTraceGenerator . m , along with other necessary codes , are provided in Supplementary Material . Kinesin-QDs exhibiting driving and resisting were manually picked , and their on-axis displacements parallel to the microtubule axis were calculated after manual tracing of microtubule backbone using aggregated images from defined time-points ( see Video 8 from time 6:48 to 7:58 ) . Microtubule bending is accounted for in the analysis . Variation in fluorescent intensity along a microtubule allows a microtubule kymograph to be generated . Edges in the kymograph were detected using the ‘edge’ command in Matlab with the ‘canny’ detection method . Manual clean-up and patching of the edges were then done to make sure microtubule movements were captured for every frame . Next , all kymograph edges were converted into velocity and averaged to obtain the microtubule velocity over time . Microtubule displacement over time was then calculated from the velocity . See Video 8 for detailed tutorial . Kinesin-QD on and off-axis displacements along a microtubule were plotted and their equilibrium positions were manually identified . Drive and resist instances were then picked with the following criteria: 1 ) there must be at least two points with displacements more than 100 nm or larger than two standard deviations from the noise at equilibrium , and 2 ) traces with more than 5 s of missing data points are removed . All drive and resist instances were then saved , containing information such as the duration and kinesin-QD displacement over time . Microtubule length over time was then obtained by manually identifying the microtubule backbone at select frames . Once all the drive and resist instances were identified for every cropped image , we compiled statistics including: the average kinesin drive-to-drive and drive-to-resist transitions; duration , run length , and force histograms; average kinesin velocity relative to microtubule over time; bulk microtubule velocity; and microtubule velocity for specific number of kinesins attached . Force was calculated from the kinesin-QD-DNA displacement by fitting an extensible Worm-Like-Chain ( WLC ) model with double stranded DNA contour length of 532 nm and persistence length of 50 nm . A distance offset of 20 nm was subtracted from the kinesin-QD-DNA displacement to account for the size of QD , proteins , and PEG and to arrive at the DNA extension length . | The inside of a cell is a crowded space , full of proteins and other molecules . Yet , the molecular motors that transport some of those molecules within the cell move at the same speed as they would in pure water – about one micrometer per second . How the molecular motors could achieve such speeds in crowded cells was unclear . Nevertheless , Tjioe et al . suspected that the answer might be related to how multiple motors work together . Molecular motors move by walking along filaments inside the cell and pulling their cargo from one location to another . Other molecules that bind to the filaments should , in theory , act like “roadblocks” and impede the movement of the cargo . Tjioe et al . studied a motor protein called kinesin , which walks on filaments called microtubules . But instead of looking at these motors moving along microtubules inside a cell , Tjioe et al . used a simpler system where the cell was eliminated , and all parts were purified . Specifically , Tjioe et al . tethered purified motors to a piece of glass and then observed them under an extremely accurate microscope as they moved free-floating , fluorescently labelled microtubules . The microtubules , in this scenario , were acting like cargoes , where many kinesins could bind . Each kinesin motor also had a small chemical tag that could emit light . By following the movement of the lights , it was possible to calculate what each kinesin was doing and how the cargo moved . When more than one kinesin molecule was acting , the tension and speed of one kinesin affected the movement of the others . In any group of kinesins , about two-thirds of kinesin pulled the cargo , and unexpectedly , about one-third tended to resist and slow the cargo . These latter kinesins were moved along with the group without actually driving the cargo . These resisting kinesins did come off more rapidly than the driving kinesins , meaning the cargo should be able to quickly bypass roadblocks . This would help to keep the whole group travelling in the right direction at a steady pace . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2019 | Multiple kinesins induce tension for smooth cargo transport |
In the striatum , signaling via G protein-coupled neurotransmitter receptors is essential for motor control . Critical to this process is the effector enzyme adenylyl cyclase type 5 ( AC5 ) that produces second messenger cAMP upon receptor-mediated activation by G protein Golf . However , the molecular organization of the Golf-AC5 signaling axis is not well understood . In this study , we report that in the striatum AC5 exists in a stable pre-coupled complex with subunits of Golf heterotrimer . We use genetic mouse models with disruption in individual components of the complex to reveal hierarchical order of interactions required for AC5-Golf stability . We further identify that the assembly of AC5-Golf complex is mediated by PhLP1 chaperone that plays central role in neurotransmitter receptor coupling to cAMP production motor learning . These findings provide evidence for the existence of stable G protein-effector signaling complexes and identify a new component essential for their assembly .
Neurotransmitters elicit their effects by activating receptors on the surface of neurons . G protein-coupled receptors ( GPCRs ) form the largest group of the receptors responsible for the actions of the majority of neurotransmitters and play a critical role in virtually all neuronal functions ( Gainetdinov et al . , 2004; Wettschureck and Offermanns , 2005 ) . In a classical model , upon binding to neurotransmitter , GPCRs undergo conformational changes activating heterotrimeric G proteins by promoting GTP binding to Gα subunits and triggering the release of the Gβγ subunits . When dissociated , both Gα and Gβγ subunits modulate the activities of downstream effector molecules that are directly responsible for generating cellular responses ( Gilman , 1987; Neer , 1995 ) . However , an emerging alternative model suggests that G protein heterotrimers exist in more stable complexes that rearrange rather than dissociate upon activation and may further form higher order signaling complexes with receptors and effectors ( Bunemann et al . , 2003; Dupre et al . , 2009; Hepler , 2014; Lambert , 2008 ) . The assembly of this macromolecular complex may be tightly regulated . Indeed , several chaperone proteins have been described to be required for the biogenesis of the G protein subunits and their complexes ( Dupre et al . , 2009; Papasergi et al . , 2015; Willardson and Tracy , 2012 ) . One of the central and best studied G protein effectors is adenylyl cyclase ( AC ) , an enzyme that catalyzes the synthesis of the second messenger cyclic adenosine monophosphate ( cAMP ) ( Taussig and Gilman , 1995 ) . Numerous isoforms of AC are differentially modulated by both Gβγ and various Gα-GTP subunits and play critical roles in a variety of fundamental neuronal processes ( Sadana and Dessauer , 2009; Sunahara et al . , 1996 ) . GPCR signaling to AC performs a particularly important function in the striatum , the input structure of the basal ganglia circuit essential for initiating and maintaining movement , mood control , and reward valuation ( Graybiel , 2000; Kreitzer and Malenka , 2008 ) . Imbalance in cAMP homeostasis in this region has been associated with drug addiction , bipolar disorder , schizophrenia and a variety of movement disorders ( Bonito-Oliva et al . , 2011; Girault , 2012; Nestler and Aghajanian , 1997; Wilson and Brandon , 2015 ) . Striatal neurons receive diverse inputs that converge on AC5 , the major AC isoform in the region , accounting for ~ 80% of cAMP generation ( Lee et al . , 2002 ) . Coupling of key neurotransmitter receptors , such as dopamine D1 ( D1R ) and adenosine A2A ( A2AR ) to increase cAMP production in these neurons , is mediated to a large extent by a unique heterotrimer composed of the stimulatory α subunit Gαolf complexed with Gβ2 and Gγ7 subunits ( Herve , 2011 ) . Indeed , deletion of Gαolf , Gγ7 or AC5 in mice severely diminishes cAMP production in response to D1R and A2AR activation , which is paralleled by muted behavioral responses to psychostimulants and antipsychotics that act on these receptors as well as by profound motor deficits ( Corvol et al . , 2007; Iwamoto et al . , 2003; Lee et al . , 2002; Sasaki et al . , 2013; Schwindinger et al . , 2010 ) . Recently , mutations in Gαolf and AC5 have been shown to cause primary dystonia in humans ( Carapito et al . , 2014; Fuchs et al . , 2012; Kumar et al . , 2014 ) , further supporting the key contribution of Golf-AC5 signaling axis to pathophysiology of movement disorders . These observations argue for the lack of functional compensation from other AC isoforms and heterotrimeric G proteins in transducing the signal . However , the functional significance of striatal-specific composition of these specific signaling elements is not well understood . Functionally , Gαolf is similar to another stimulatory G protein Gαs , but Gαolf has lower efficiency of both receptor coupling and AC5 stimulation when tested in in vitro biochemical assays ( Chan et al . , 2011; Jones et al . , 1990 ) . AC5 displays dual regulation by G protein subunits in which Gβγ acts to facilitate its activation by Gαs ( Gao et al . , 2007 ) . In vitro biochemical studies show that AC5 is capable of binding Gαs and Gβγ simultaneously , suggesting that it can scaffold the stimulatory G protein heterotrimers ( Sadana et al . , 2009 ) . In fact , growing evidence suggests that in vivo Golf-AC5 may exist in a signaling complex ( Herve , 2011 ) . In mice , AC5 elimination leads to a reduction in expression levels of Gαolf in the striatum ( Iwamoto et al . , 2004 ) . Similarly , knockout of Gγ7 reduces the expression of Gαolf and Gβ2 ( Sasaki et al . , 2013; Schwindinger et al . , 2010 ) . However , interactions involving elements of the complex in the striatum , their reciprocal relationship , mechanisms of complex assembly as well as implications for the cAMP signaling and behavior are not understood . Here we report that in the striatum , AC5 forms a stable macromolecular complex with heterotrimeric Golf proteins and this pre-assembly is essential for the stability of AC5 and its ability to produce cAMP . We identified that the Gβ chaperone phosducin-like protein 1 ( PhLP1 ) plays a key role in the assembly of this signaling complex in striatal neurons . Elimination of PhLP1 in striatal neurons affects assembly and stability of the complex and causes selective impairment in sensorimotor behavior and motor skill learning , preferentially affecting signaling in the striatopallidal medium spiny neurons .
To begin testing the idea that AC5 may exist in a stable complex with the Golf heterotrimer in vivo , we first analyzed their binding by co-immunoprecipitaiton assays . AC5 was effectively and specifically pulled down together with both Gαolf and , to a lesser extent , Gβ2 from mouse striatal lysates , indicating that in striatal neurons Golf subunits form stable complexes with AC5 in their ground state in the absence of receptor stimulation ( Figure 1A ) . Next , we analyzed the relationship between AC5 and Golf by examining the co-dependence of their expression using mouse knockout models . Elimination of AC5 in mice ( Adcy5-/- ) dramatically reduced the expression of Gαolf but had no appreciable effect on Gβ2 expression , suggesting that AC5 contributes to the stability of Gαolf but not the stability of the Gβ2γ7 complex ( Figure 1B ) . We further used Adcy5-/- striatal tissues in the immunoprecipitation experiments and found that antibodies against AC5 failed to pull-down Golf subunits from striatal lysates lacking AC5 , confirming the specificity of Golf association with AC5 ( Figure 1C ) . In contrast , disruption of Gβ2γ7 by eliminating Gγ7 in Gng7-/- knockout led to a marked down-regulation of both AC5 and Gαolf expression ( Figure 1D ) . The consequences of Gαolf disruption were evaluated in heterozygous mice ( Gnal+/- ) with severely diminished Gαolf expression because complete ablation of Gαolf leads to increased perinatal lethality ( Belluscio et al . , 1998 ) . In these animals , the levels of both AC5 and Gβ2 were substantially reduced ( Figure 1E ) . Therefore , in mouse striatum co-dependence of the complex components appears to be unidirectional: destabilization of either the Gα or Gβγ subunits of the Golf heterotrimer compromised AC5 stability , while only Gα but not Gβγ was affected by the loss of AC5 . 10 . 7554/eLife . 10451 . 003Figure 1 . Co-dependence of heterotrimeric Golf subunits and AC5 in complex formation and expression . ( A ) Immunoprecipitation of AC5 complexes from striatal lysates of wild type mice . Specific anti-AC5 antibodies but not non-immune IgGs pull down Gαolfβ2γ7 subunits from native striatal tissues . ( B ) Significant reduction of Gαolf expression in striatum tissue from mice lacking AC5 . Total striatal lysates were analyzed by immunoblotting using indicated antibodies and quantified by densitometry . ***p<0 . 001 , Student’s t-test , n = 3 mice . ( C ) Immunoprecipitation of AC5 from wild type ( +/+ ) and Adcy5-/- tissues confirms specificity of AC5-Golf binding . Same anti-AC5 antibodies were used in both immunoprecipitaiton experiments and samples were processed in parallel . ( D ) Significant reduction of AC5 and Gαolf expression in striatal tissues from mice lacking Gγ7 . Total striatal lysates from wild type and Gng7-/- mice were analyzed by immunoblotting with indicated antibodies and quantified by densitometry . *p<0 . 05 , **p<0 . 01 , Student’s t-test , n = 3 mice . ( E ) Significant reduction of AC5 and Gβ2 expression in striatal tissues from mice with reduced expression of Gαolf . Total striatal lysates from wild type and Gnal+/- mice were analyzed by immunoblotting with indicated antibodies and quantified by densitometry . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , Student’s t-test , n = 3 mice . ( F ) Mutual stabilization of AC5 and Gαolf upon co-transfection in HEK293 cells . Equal amounts of cDNAs were transfected into cells as described in the Methods section and protein expression was assessed by immunoblotting with specific antibodies ( G ) Lack of co-stabilization between AC5 and Gβ2γ7 subunits in co-transfected HEK293 cells . Indicated constructs were transfected into cells and changes in protein expression were monitored by immunoblotting with specific antibodies . ( H ) Mutual stabilization of Gαolf subunits with Gβ2γ7 subunits in co-transfected HEK293 cells . The experiment was conducted as described for panel G . DOI: http://dx . doi . org/10 . 7554/eLife . 10451 . 003 We further examined the relationship between Golf subunits and AC5 in transfected HEK293 cells . Consistent with the in vivo data , co-expression with Gαolf dramatically increased levels of AC5 , and vice versa , the introduction of AC5 enhanced the expression of Gαolf ( Figure 1F ) . In contrast , no significant changes in AC5 or Gβ2γ7 levels were observed upon their co-expression ( Figure 1G ) . We further observed co-dependence between Gαolf and Gβ2γ7 subunits in which co-expression with Gαolf increased the levels of Gβ2γ7 and co-expression of Gβ2γ7 enhanced levels Gαolf ( Figure 1H ) . Collectively , these data strongly support the existence of a complex between AC5 and the Gαolfβ2γ7 heterotrimer and indicate that the association between components of this complex is required for its stability . In search of factors that may regulate the assembly of Golf-AC5 complex , our attention was drawn to phosducin like protein 1 ( PhLP1 ) , which has been described to serve as a co-factor for the assembly of signaling complexes containing β subunits of heterotrimeric G proteins ( Willardson and Tracy , 2012 ) . In HEK293 cells , co-transfection with PhLP1 significantly increased the expression of all three Golf subunits ( Figure 2A ) . In contrast , a dominant negative construct of PhLP1 ( ΔNT-PhLP1 ) , lacking the first 75 amino acids critical for its interaction with Gβ ( Lukov et al . , 2005 ) had no effect on the expression Golf subunits ( Figure 2A ) . Consistent with the described impact of PhLP1 on Gβγ complexes ( Lukov et al . , 2005 ) , co-expression with full-length PhLP1 resulted in a significant increase in the formation of the Gβ2γ7 complexes as evidenced by the fluorescence complementation assay with the split-Venus system ( Figure 2B ) . In contrast , ΔNT-PhLP1 exerted the opposite effect and inhibited Gβ2γ7 assembly ( Figure 2B ) . To examine the functional effects of PhLP1 in a context of the entire Golf heterotrimer , we tested the ability of Gβ2γ7 and Gαolf subunits to undergo a cycle of dissociation/re-association in response to changes in the D1R activity . We used a cell-based Bioluminescence Energy Transfer ( BRET ) assay to monitor the release of the Gβ2γ7 dimer in the presence of Gαolf ( Figure 2C ) . In this assay , stimulation of D1R by dopamine results in the activation of Gαolf , releasing Venus-tagged Gβ2γ7 to interact with the NanoLuc-tagged reporter , an event detected by changes in the BRET signal . Conversely , application of an antagonist SCH39166 facilitates complex re-association and quenching of the BRET response ( Figure 2D ) . Reversing the sequence of ligand addition abolishes the response indicating that changes in the BRET signal are associated with the activation of the D1R receptors ( Figure 2D ) . Furthermore , omitting D1R or Golf from the transfection also dramatically suppressed the signal indicating that the signal is specifically driven by the D1R-mediated activation of the Golf ( Figure 2E ) . Using this assay we found that PhLP1 coexpression significantly increased dopamine-induced activation of Gαolf-Gβ2γ7 heterotrimer ( Figure 2F , G ) . In contrast , expression of the dominant negative ΔNT-PhLP1 significantly attenuated the BRET response ( Figure 2F , G ) . These observations suggest that PhLP1 facilitates functional coupling of GαolfGβ2γ7 to D1 receptors , likely owning to its ability to promote the assembly of Gβ2γ7 complexes . 10 . 7554/eLife . 10451 . 004Figure 2 . PhLP1 facilitates functional assembly of Gαolfβ2γ7 complex . ( A ) Left , Full length of PhLP1 but not its N terminally truncated mutant ΔNT-PhLP1 increases the expression level of Gαolfβ2γ7 subunits upon overexpression in HEK293 cells . Right , quantification of immunoblot data from 3 independent experiments . Data were normalized to the individual protein expression in the control group without PhLP1 transfection . Data were analyzed by One-Way ANOVA ( Gαolf F[2 , 9] = 8 . 731 , p = 0 . 008; Gβ2 F[2 , 9] = 6 . 688 , p = 0 . 017; Gγ7 F[2 , 9] = 11 . 107 , p = 0 . 004 ) . *p<0 . 01 compared to the control group post hoc Tukey’s test . ( B ) Full length PhLP1 facilitates , while ΔNT-PhLP1 inhibits Gβ2γ7 assembly . Venus fluorescence intensity was used as a readout of Gβ2γ7 complex assembly in a complementation experiment in transfected HEK293 cells . Data were analyzed by One-Way ANOVA ( F[2 , 15] = 2719 . 521 , p<0 . 001 ) . **p<0 . 01 , ***p<0 . 001 compared with the control group , post hoc Tukey’s test . ( C ) Schematic diagram of BRET sensor strategy for examining the dissociation and reassociation of Gαolf and Gβ2γ7 subunits upon D1Rs activation and inactivation . ( D ) Representative BRET response traces . Cells transfected with D1R and Golf were stimulated by 100 μM dopamine followed by 100 μM SCH39166 ( black ) or by 100 μM SCH39166 followed by 100 μM dopamine ( red ) . First and second ligands were applied at 5 and 40 s , respectively . ( E ) Control experiments examining the requirement of both Golf and D1R to transduce the signal . Cells were transfected with the three different conditions , Golf only , D1R only , or D1R plus Golf . Each bar represents the mean of 6 replicates . ( F ) Representative BRET signal traces in response to D1 receptor activation with dopamine ( 100 μM ) . ( G ) Comparison of maximal BRET ratios . Data were analyzed by One-Way ANOVA ( F[2 , 9] = 706 . 655 , p<0 . 001 ) . **p<0 . 01 compared with the control group , post hoc Tukey’s test . DOI: http://dx . doi . org/10 . 7554/eLife . 10451 . 004 The observations that AC5 expression is sensitive to changes in the levels of the Golf and that PhLP1 increases the expression of Golf subunits suggested that PhLP1 might modulate AC5 activity . We tested this possibility by measuring the effect of PhLP1 co-expression on AC5 expression . PhLP1 co-expression resulted in a significant increase ( ~1 . 7 fold ) in AC5 expression ( Figure 3A ) . This effect was paralleled by elevation of the basal cAMP levels ( ~4 fold ) in the cells expressing AC5 ( Figure 3B ) . In addition , AC5-containing cells showed enhanced cAMP generation in response to both the direct AC activator forskolin ( ~4 fold increase ) and the β2-adrenergic agonist isoproterenol ( ~1 . 5 fold increase ) when PhLP1 was overexpressed ( Figure 3B ) . These effects were AC5-dependent as no stimulatory effect on cAMP generation was observed in the absence of AC5 ( Figure 3C ) . Interestingly , while the increase in GPCR-driven cAMP production ( likely through Gαs ) in the presence of PhLP1 closely matched the increase in the total AC5 levels , the effect of PhLP1 on both basal and forskolin-induced cAMP production was much larger . These observations suggest that PhLP1 may exert functional effects on cAMP production that are independent from increasing AC expression . 10 . 7554/eLife . 10451 . 005Figure 3 . PhLP1 augments expression and activity of AC5 . ( A ) Co-expression of PhLP1 significantly increases AC5 expression in HEK293 cells . **p<0 . 01 Students’ t-test , n = 3 . ( B ) PhLP1 enhances cAMP levels under basal condition , or in response to stimulation with forskolin ( FSK , 1 μM , 5 min ) or isoproterenol ( ISO , 1 μM , 5 min ) in AC5 expressing HEK293 cells . **p<0 . 01 Student’s t-test , n = 3 . ( C ) PhLP1 has no effect on cAMP generation in HEK293 cells without AC5 overexpression . n = 3 ( D ) Effects of purified recombinant full length PhLP1 and ΔNT-PhLP1 on adenylyl cyclase activity in striatal membranes . Assays were performed in the absence of GTP . Striatal membranes were pre-incubated with 0 . 5 μM of purified proteins at 4°C for 20 min and then subjected to adenylyl cyclase activity assay . Data were analyzed by One-Way ANOVA ( F[2 , 9] = 35 . 477 , p<0 . 001 ) . **p<0 . 01 , post hoc Tukey’s test . n = 3 ( E ) Purified recombinant PhLP1 enhances forskolin-stimulated adenylyl cyclase activity in striatal membranes . Membranes were pre-incubated with 0 . 5 μM of purified proteins at 4°C for 20 min . Membranes were then stimulated with 1 μM forskolin or 80 nM Gαs-GTPγS . Data were analyzed by One-Way ANOVA ( F ( 4 , 15 ) = 52 . 291 , p<0 . 001 ) . ***p<0 . 001 , post hoc Tukey’s test . n = 3 ( F ) Effect of purified recombinant PhLP1 on the dose response of forskolin-mediated activation of adenylyl cyclase in striatal membranes . The EC50 for forskolin was 32 . 5 ± 2 . 7 μM in the control reaction and 8 . 9 ± 1 . 6 μM in PhLP1-treated reaction . **p<0 . 01 Student’s t-test , n = 3 . ( G ) Effect of purified recombinant PhLP1 on the dose response of Gαs-GTPγS-mediated activation of adenylyl cyclase in striatal membranes . The EC50 for Gαs-GTPγS was 16 . 2 ± 4 . 7 nM in the control reaction and 22 . 3 ± 5 . 7 nM in PhLP1-treated reaction ( n = 3 reactions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10451 . 005 To gain mechanistic insight into the nature of this regulation in a more physiological setting , we performed adenylyl cyclase activity assays using membrane isolated from mouse striatum . Remarkably , purified recombinant full-length PhLP1 significantly enhanced basal AC activity ( Figure 3D ) . This effect was not observed with ΔNT-PhLP1 , suggesting that the effect likely involves an ability of PhLP1 to bind Gβγ . The reaction buffer in these assays did not contain GTP , therefore the effect of PhLP1 was unlikely to result from the classical activation of Gα . Given that in the striatal membranes AC5 forms a complex with GαolfGβ2γ7 we hypothesized that PhLP1 may cause activation by scavenging Gβγ and releasing Gαolf , making it available for the activation of AC5 , as even GDP-bound free Gα subunits are capable of regulating AC ( Sunahara et al . , 1997 ) . To test this notion , we compared the ability of forskolin and Gαs-GTP to regulate AC activity in the striatal membranes in the absence or presence of PhLP1 . Gαs is known to synergize with forskolin increasing its ability to regulate cyclase activity by promoting binding of forskolin to a high affinity site ( Dessauer et al . , 1997 ) . Consistent with our model , PhLP1 significantly enhanced forskolin-stimulated AC activity in striatal membranes and produced an approximately three-fold reduction in EC50 for forskolin ( Figure 3E and F ) . In contrast , PhLP1 failed to exert an effect on AC activated by adding exogenous Gαs-GTPγS ( Figure 3E and G ) supporting the idea that PhLP1 activates AC5 through a Gα-dependent mechanism . In this case , added Gαs-GTPγS likely out competed endogenous Gαolf-GDP activating AC5 after being released by PhLP1 . Thus , PhLP1 appears to regulate AC5 activity by a dual mechanism . On the one hand it enhances the expression of the AC5 by facilitating its complex formation with Golf . On the other , it may act in promoting the receptor-independent release of Gαolf that stimulates AC5 activity . We next sought to examine the effect of PhLP1 on striatal signaling and physiology in vivo . In agreement with previous reports ( Schroder and Lohse , 2000 ) , we found PhLP1 protein to be abundantly expressed in the mouse striatum by immunoblotting ( Figure 4A ) . To study the role of PhLP1 in the striatum , we ablated its expression selectively in the striatum by crossing the conditional PhLP1 mouse strain ( Pdclflx/flx ) with the striatal driver mouse line Rgs9-Cre , generating a conditional , striatal-specific elimination of PhLP1 ( Pdcl cKO ) mouse ( Figure 4B ) . In this line recombination likely occurs postnatally , as expression of the Rgs9 gene is induced around P3 to P6 ( Anderson et al . , 2007 ) . We began our analysis by assessing possible anatomical changes because previous studies indicated that elimination of PhLP1 may lead to neuronal degeneration ( Lai et al . , 2013 ) . Overall , striatal morphology of Pdcl cKO mice looked normal with no signs of degeneration at least until 3–4 months of age ( Figure 4C ) . Morphometric analysis revealed a decrease in the total volume of striatal tissue in Pdcl cKO mice ( Figure 4D ) . Counting the number of neurons ( diameter > 5 μm ) in the striatum tissue using Nissl staining revealed a significantly greater number of neurons in the Pdcl cKO mice . A greater number of cells together with a smaller volume that they occupy indicate that the sizes of individual striatal neurons are likely smaller . These changes are consistent with retarded maturation of neurons , a process controlled by the cAMP signaling ( Fujioka et al . , 2004; Nakagawa et al . , 2002 ) . We next analyzed projections of striatal medium spiny neurons to the target regions Globus Pallidus ( GPe ) and Substantia Nigra ( SNr ) , revealed by immunostaining for enkephalin and substance P , respectively . Quantification of fluorescence intensity revealed no difference in the intensities of the signals for these markers , which were found in appropriate target areas ( Figure 4E ) . In summary , these data indicate that loss of PhLP1 in the striatum does not lead to neuronal degeneration , but rather promotes neuronal survival while inhibiting their growth . 10 . 7554/eLife . 10451 . 006Figure 4 . Elimination of PhLP1 does not impact survival and connectivity of striatal neurons . ( A ) PhLP1 expression in different brain regions from adult mice as determined by immunoblot analysis . ( B ) Generation of PhLP1 conditional knockout out . Pdclflx/flx mice were crossed with Rgs9-Cre mice to generate striatal specific PhLP1 conditional knockout ( Pdcl cKO ) mice ( C ) Representative images of Nissl-stained coronal brain sections from adult control and Pdcl cKO mice . Stm , striatum . ( Scale bar , 1 mm ) . ( D ) Left: Striatal volume of adult Pdcl cKO mice was reduced by 30% as compared with control mice ( n = 4 mice each ) . Error bars represent SEM . Student’s t test: **p<0 . 01 . Right: Counts of striatal neurons in Pdcl cKO and control mice obtained from Nissl-stained sections . The striatal neuron counts were increased by 43% compared with control mice ( n = 4 mice each ) . Error bars represent SEM . Student’s t test: *p<0 . 05 . ( E ) Left: Representative images of anti-enkephalin and anti-substance-P stained brain sections . GPe , external globus pallidus . SNr , substantia nigra pars reticulate . ( Scale bar , 2 mm . ) Right: Quantification of immunofluorescent signal for enkephalin and substance-P . Mean intensity of signals from both antibodies showed no significant difference between control and Pdcl cKO mice . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 10451 . 006 Our studies in vitro and in heterologous expression system indicate a role for PhLP1 in the assembly of the GαolfGβ2γ7 complex . Previous in vivo studies also demonstrated that PhLP1 is required for the assembly of Gβ1 with Gαt1 and Gβ3 with Gαt2 as well as Gβ5 complexes with RGS9-1 ( Lai et al . , 2013 , Tracy et al . , 2015 ) . Therefore , we proceeded to investigate the consequences of PhLP1 ablation on the expression of various subunits of heterotrimeric G proteins , RGS proteins and AC5 in the striatum ( Figure 5A ) . Immunoblotting shows that the levels of PhLP1 protein were reduced by ~60% in Pdcl cKO striatum . Consistent with the results in transfected cells , we found that the levels of Gαolf , Gβ2 and AC5 were severely reduced in Pdcl cKO as well . Deletion of PhLP1 also had a detrimental effect on the expression of Gβ5 and RGS9-2 , as may have been expected from the studies on the rod and cone photoreceptors ( Lai et al . , 2013; Tracy et al . , 2015 ) . Interestingly , the effect was clearly selective as deletion of PhLP1 did not affect the expression of Gβ1 and Gα subunits possibly associated with it: Gαo , Gαi , Gαq ( Figure 5B ) . Furthermore , the levels of another Gβ5 associated protein , RGS7 were also unaffected . Analysis of the mRNA levels for corresponding down-regulated proteins showed no changes in the transcript levels , suggesting that PhLP1 likely contributes to protein stability rather than affects the expression through a transcriptional mechanism ( Figure 5C ) . Therefore , it appears that PhLP1 selectively affects biosynthesis and/or assembly of the AC5 signaling complex that in addition to GαolfGβ2γ7 also contains RGS9-2/Gβ5 ( Xie et al . , 2012 ) . 10 . 7554/eLife . 10451 . 007Figure 5 . Elimination of PhLP1 in striatal neurons significantly impairs expression and function of AC5-Golf complex . ( A ) Representative immunoblot data of AC5 , different G protein subunits and RGS proteins in stratal tissues from adult control or Pdcl cKO mice . ( B ) Quantification of protein levels . Data were normalized to the percentage of protein level in control mice . **p<0 . 01 , ***p<0 . 001 , Student’s t-test , n = 3 mice . ( C ) mRNA quantification in striatum tissue from control and Pdcl cKO mice . Data were normalized to the percentage of mRNA level in control mice . ( D ) , Basal cAMP level was reduced in the striatum tissue of Pdcl cKO mice . ***p<0 . 001 , Students’ t-test , n = 3 mice . ( E ) Striatal membrane adenylyl cyclase activity is reduced in Pdcl cKO mice . Adenylyl cyclase activity was measured either under basal conditions ( without stimulation ) , or in the presence of forskolin ( FSK , 1 μM ) , D1R specific agonist SKF38391 ( SKF , 10 μM ) or A2AR agonist CGS21680 ( CGS , 10 μM ) . *p<0 . 05 , **p<0 . 01 , Student’s t-test , n = 3 reactions . DOI: http://dx . doi . org/10 . 7554/eLife . 10451 . 007 In agreement with the changes in protein levels , we found marked deficits in cAMP signaling . Basal cAMP levels were substantially lower in the striatum tissues from Pdcl cKO mice as compared to their control littermates ( Figure 5D ) . Adenylyl cyclase activity measured with membranes isolated from the striatum was also lower upon the deletion of PhLP1 under both basal and forskolin-stimulated conditions ( Figure 5E ) . Finally , the efficiency of D1R and A2AR coupling to cAMP production was lower in Pdcl cKO striatal membrane ( Figure 5E ) . Together , these data indicate that in the striatum PhLP1 is necessary for high-level expression of the AC5 complex components and its functional activity . To better understand the relevance of the observed molecular changes to normal physiology and pathology , we analyzed the behavioral consequences of eliminating PhLP1 in the striatum . We started by assessing the performance of Pdcl cKO mice in a battery of striatum-dependent tasks comparing their behavior to control littermates . In the open field test , Pdcl cKO mice exhibited normal habituation to a novel environment ( Figure 6A ) and had unaltered levels of basal locomotor activity as evidenced by both total distance traveled during the task ( Figure 6B ) and average locomotion velocity ( Figure 6C ) . The Pdcl cKO mice also had normal thigmotaxis , stereotypy , pre-pulse inhibition and showed no signs of abnormal involuntary movements or clasping behavior typically associated with gross striatal dysfunction ( data not shown ) . 10 . 7554/eLife . 10451 . 008Figure 6 . Behavioral consequences of PhLP1 elimination in striatal neurons . ( A ) Pdcl cKO mice display normal basal locomotion and habituation to a novel environment . ( B ) Total distance traveled in 2 hr in the open field chamber . ( C ) Average velocity in the open field chamber . ( D ) Pdcl cKO mice exhibit severe deficits in motor learning behavior in rotarod test . There were statistically significant differences between genotype as determined by Two-way ANOVA ( F[1 , 153] = 63 . 518 , p<0 . 001 ) . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , post hoc Tukey’s test , n = 9 and 10 for control mice and Pdcl cKO mice , respectively . ( E ) Increased grip strength of the forelimbs in Pdcl cKO mice . *p<0 . 05 , Student’s t-test , n = 8 mice for each genotype . ( F ) Increased acoustic startle response of Pdcl cKO as measured by Vmax in response to 120 dB white noise bursts . ( G ) Normal locomotor response to D1R agonist SKF 38 , 393 ( SKF , 50 mg/kg , i . p . ) in Pdcl cKOcompared to control mice . Mice were injected with vehicle or SKF 38 , 393 ( SKF , 50 mg/kg , i . p . ) and immediately put in open field chambers . The locomotion was recorded for 1 hr . Data were analyzed by Two-way ANOVA ( treatment F[1 , 28]=96 . 068 , p<0 . 001 , genotype F[1 , 28] = 2 . 679 , p = 0 . 113 ) . *p<0 . 01 post hoc Tukey’s test compared to the vehicle control of the same genotype . n=8 mice per each genotype . ( H ) Blunted response to A2AR antagonist SCH58261 ( SCH , 3 mg/kg , i . p . ) treatment . Data were analyzed by Two-way ANOVA ( treatment F ( 1 , 28 ) = 42 . 819 , p<0 . 001 , genotype F ( 1 , 28 ) = 20 . 181 , p<0 . 001 ) . *p<0 . 01 post hoc Tukey’s test compared to the vehicle control of the same genotype , #p<0 . 05 post hoc Tukey’s test in comparison between genotypes in response to SCH 58261 . n = 8 mice per each genotype . ( I ) Reduced catalepsy in response to D2R antagonist haloperidol ( 2 mg/kg , i . p . ) in Pdcl cKO mice . Catalepsy was measured in the bar test 1 hr after haloperidol ( 2 mg/kg , i . p . ) administration . *p<0 . 05 , Student’s t-test . n = 8 mice per each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 10451 . 008 We next tested animal motor behavior . In the accelerating rotarod task , mice of both genotypes that were naïve to the procedure performed equally during initial trial , indicating their normal motor coordination ( Figure 6D ) . However , the behavior of Pdcl cKO mice was dramatically different when multiple trials were conducted . While the performance of control littermates improved with each subsequent trial , Pdcl cKO mice performed at the same level during all trial sessions ( Figure 6D ) . These data indicate that knockout of Pdcl in the striatum completely abolishes motor learning in mice . Pdcl cKO mice also showed enhanced forelimb grips strength ( Figure 6E ) and augmented response to acoustic startle ( Figure 6F ) , suggesting further deficits in processing sensorimotor stimuli . To dissect the influence of PhLP1 on receptor-mediated neurotransmission , we performed pharmacological studies at the behavioral level . Injection of mice with the D1R selective agonist SKF38393 that activates striatal neurons of the direct pathway caused equal psychomotor activation in Pdcl cKO mice and their wild-type littermates ( Figure 6G ) . In contrast , administration of A2AR antagonist SCH58261 , that inhibits striatal neurons of the indirect pathway , had no effect on the behavior of Pdcl cKO mice while significantly increasing locomotor activity of wild-type littermates ( Figure 6H ) . Conversely , the D2R antagonist haloperidol , that activates striatopallidal ( indirect ) pathway , induced catalepsy to a significantly greater extent in control littermates than in Pdcl cKO mice ( Figure 6I ) . In summary , the behavioral data suggest normal baseline behavior of Pdcl cKO mice with selective deficits in the striatal-mediated motor learning and sensorimotor coordination , preferentially affecting the function of the striatopallidal pathway .
In this study we demonstrate that AC5 , the key cAMP-producing enzyme in the striatum , forms stable complexes with its regulatory G protein species Gαolfβ2γ7 in vivo . We find that this interaction occurs at the basal state and can be detected without agonist application , which is typically required to promote dissociation of heterotrimers into Gα-GTP and Gβγ subunits making them competent for binding to effectors . We further demonstrate that binding to Gαolfβ2γ7 complex is required for the proteolytic stabilization of AC5 that results in its high expression level in the striatum . In the course of this study , we further confirmed previously reported interdependence among subunits of the Golf heterotrimer in which interactions between Gαolf and Gβ2γ7 are required for mutual proteolytic stabilization of the complex ( Iwamoto et al . , 2004; Schwindinger et al . , 2003; Schwindinger et al . , 2010 ) . These observations are similar to noted interdependence of Gαt1β1γ1 and Gαoβ3γ13 subunits observed in rod photoreceptors ( Lai et al . , 2013 ) and ON-bipolar cells ( Dhingra et al . , 2012 ) of the retina , respectively , and likely reflect a general principle in setting subunit stoichiometry in G protein heterotrimeric complexes . While there are several examples for the association of both Gα and Gβγ subunits with effector molecule , e . g . PLCβ , GIRK channels , AC isoforms ( Kovoor and Lester , 2002; Lyon et al . , 2014; Sunahara et al . , 1996 ) , to the best of our knowledge this study is the first to document a case in which stable association with all subunits of G protein heterotrimer is required for the stability of an effector . Taken together with biochemical studies showing that AC5 is capable of scaffolding inactive G protein heterotrimers ( Sadana et al . , 2009 ) , our results suggest that in striatal neurons heterotrimeric Golf-AC5 is assembled in a pre-coupled ‘signalosome’ in which subunits rearrange rather than physically dissociate upon GPCR activation . We report that the chaperone protein PhLP1 plays a critical role in the assembly of the Golf-AC5 complex in striatal neurons . Previous studies have demonstrated the role of PhLP1 in the assembly of the complexes involving Gβ subunits of heterotrimeric G proteins . It is thought to function as a co-chaperone with the cytosolic chaperonin complex ( CCT ) , assisting in retrieval of the Gβ subunits emerging from the CCT and presenting them for the association with the Gg subunits ( Willardson and Tracy , 2012 ) . In addition to assisting folding of conventional Gβγ complexes , PhLP1 chaperones the formation of the structurally similar complexes involving the atypical Gβ5 subunits and the Gγ-like domains in RGS proteins ( Howlett et al . , 2009 ) . Consistent with this model , deletion of PhLP1 in photoreceptors disrupts the formation of Gβ1γ1 ( in rods ) , Gβ3γ8 ( in cones ) and Gβ5L/RGS9-1 complexes ( in both rods and cones ) , dramatically decreasing their expression ( Lai et al . , 2013; Tracy et al . , 2015 ) . Similarly , overexpression of the dominant negative mutant of PhLP1 deficient in Gβ binding in rods down-regulates the expression of the Gβ1γ1 and Gβ5L/RGS9-1 complexes ( Posokhova et al . , 2011 ) . Interestingly , loss of PhLP1 function also results in a dramatic decrease in the expression of Gαt1 and Gαt2 subunits in rods and cones , respectively ( Lai et al . , 2013; Tracy et al . , 2015 ) . Given that stability of these Gα subunits depends on their complex formation with Gbg ( Kolesnikov et al . , 2011; Lobanova et al . , 2008; Nikonov et al . , 2013 ) , the role of PhLP1 in stabilizing Gαt1 and Gαt2 is likely indirect and stems from its effects on Gβγ complex assembly . Similar to the situation in photoreceptors , we observe that knockout of PhLP1 in striatal neurons results in marked down-regulation in the expression of all subunits of heterotrimeric Golf complex . Direct side-by-side comparison of the impact on protein expression produced by elimination of individual subunits of the Golf-AC5 complex affords a unique opportunity to analyze the reciprocal relationship between complex components . Our results suggest a hierarchical relationship in the assembly of the complex . The stability of Gαolf requires its association with both Gβ2γ7 and AC5 , but the stability of Gβ2γ7 depends only on the association with Gαolf and not AC5 . In turn , stability of AC5 is dependent on both Gβ2γ7 and Gαolf . Given that PhLP1 directly binds to Gbg subunits ( Savage et al . , 2000b; Thibault et al . , 1997 ) and has been shown to be required for folding Gβ2γ7 in transfected cells ( Howlett et al . , 2009 ) , we propose the following model for the role in PhLP1 the assembly of the Golf-AC5 complex . PhLP1 assists the assembly of the Gβ2γ7 complex , increasing its expression , which in turn upregulates Gαolf . Higher levels of Gαolfβ2γ7 promote stabilization of the AC5 by forming pre-coupled complexes with it . Thus , PhLP1 triggers a chain of events resulting in the stabilization of the entire AC5-Golf complex ensuring its high expression level and setting the stoichiometry ( Figure 7 ) . 10 . 7554/eLife . 10451 . 009Figure 7 . Schematic illustration of PhLP1 involvement in regulating Golf-AC5 complex assembly and signaling . PhLP1 promotes biogenesis of Gβ2γ7 and assembly of its complex with Golf . The trimeric Gαolfβ2γ7 forms stable complexes with AC5 contributing to its proteolytic stability . In addition , PhLP1 regulates cAMP production by influencing Golf arrangement on AC5 . DOI: http://dx . doi . org/10 . 7554/eLife . 10451 . 009 It is interesting that PhLP1 loss in striatal neurons affected G protein complexes in a selective fashion . Studies in transfected cells indicate that PhLP1 participates in folding and stabilization of virtually all Gβγ combinations including complexes of Gβ5 with RGS9 and RGS7 ( Howlett et al . , 2009 ) . Yet , we find that in the striatum , PhLP1 elimination has no effect on the expression of many G protein subunits , including Gβ1 , a subunit clearly impacted by PhLP1 loss in rod photoreceptors ( Lai et al . , 2013 ) . We propose that this apparent selectivity may be explained by higher susceptibility of G proteins with high expression levels to the destabilizing effects associated with PhLP1 loss than those expressed at moderate to low levels . In this scenario , PhLP1 action may be the rate-limiting factor required for achieving high expression of G proteins with abundant mRNA expression in a particular neuronal population , e . g . Gαt1β1γ1 , Gαt2β3γ8 and Gαolfβ2γ7 overexpressed in rods , cones and striatal neurons , respectively . However , at this point , we also cannot rule out an alternative explanation that cellular heterogeneity in the striatum contributes to selectivity of the effects . Although Rgs9-Cre driver line that we used to delete PhLP1 is active in 95% of neurons in the striatum , it might have no effect on other cells such as glia that might contain a higher abundance of G protein subunits that we find to be unregulated by PhLP1 . It is interesting to consider the results of this study in light of the mechanisms linking GPCR signaling in striatal neurons to cAMP production and behavior . The Golf heterotrimer is known to be essential for coupling both D1 and A2A receptors in direct and indirect striatal neurons to stimulation of cAMP production ( Herve , 2011 ) . Knockout of Gnal or Gng7 that encode Gαolf or Gγ7 respectively , markedly reduces the ability of D1R and A2AR agonists to increase cAMP , and this biochemical observation is paralleled by muted behavioral responses of mice to psychostimulants that activate these receptors ( Corvol et al . , 2001; Schwindinger et al . , 2003; Zhuang et al . , 2000 ) . Acute psychomotor responses to drugs appear to be particularly sensitive to loss of Golf , while adaptive behaviors to repeated drug exposure are preserved ( Corvol et al . , 2007; Schwindinger et al . , 2010 ) . Interestingly , levels of Golf complex dictate particular signaling outcomes upon receptor stimulation . For example , while Gαolf is critical for the ability of D1R to stimulate both cAMP production and ERK phosphorylation , Gαolf haploinsufficiency leads to selective deficits in cAMP signaling without detrimentally affecting coupling to ERK ( Corvol et al . , 2007 ) . This signaling dichotomy becomes even more pronounced at the level of AC5 . Although , activation of either D1R and A2A clearly results in the increase in cAMP levels ( Corvol et al . , 2001; Iwamoto et al . , 2003; Lee et al . , 2002 ) and AC5 mediates nearly 80% of this effect ( Lee et al . , 2002 ) , elimination of AC5 in mice does not diminish behavioral responses of mice to the administration of A1R and D1R agonists ( Iwamoto et al . , 2003; Lee et al . , 2002 ) . At the same time , mice lacking AC5 have abolished responses to D2 antagonists , suggesting greater impact of signaling via AC5 in indirect pathway neurons ( Lee et al . , 2002 ) . Balance of neurotransmitter signaling between direct and indirect pathway striatal neurons is thought to set a tightly orchestrated coordination of movements ( Graybiel , 2005; Nelson and Kreitzer , 2014 ) . In agreement with this idea , mice lacking AC5 or Gγ7 show profound deficits in motor learning and coordination ( Iwamoto et al . , 2003; Kheirbek et al . , 2009; Sasaki et al . , 2013 ) . Furthermore , mutations in genes encoding AC5 and Gαolf in humans cause dystonia , a disorder characterized by involuntary movements ( Carapito et al . , 2014; Fuchs et al . , 2012; Kumar et al . , 2014 ) . Thus , it is likely that Golf-AC5 axis is involved in more subtle coordination of signaling in striatal neurons that sometimes might not be evident from measuring gross motor responses to pharmacological treatment with receptor ligands . Our findings with a mouse model lacking PhLP1 in the striatum agree well with the expectations based on the analysis of mice with disruptions in AC5 and Golf components and also provide additional insight that helps further clarify integration of neurotransmitter signaling in the striatum . Our Pdcl cKO model displays complete lack of motor learning , deficits in psychostimulatory effects of A2AR antagonism and diminished neuroleptic responses to D2R antagonism . As with studies on Adcy5-/- mice ( Herve , 2011; Lee et al . , 2002 ) , perhaps the most surprising result of our studies is the intact responses of mice lacking PhLP1 to D1R agonism . This observation is particularly striking given the profound down-regulation of AC5 , Gαolf and Gβ2γ7 in these mice paralleled by equal deficits in coupling of both D1R and A2AR to cAMP production . We think that this selectivity is likely related to downregulation of all signaling components rather than their complete loss . It is possible that striatonigral and striatopallidal neurons have differential sensitivity to changes in the efficiency of the Golf-AC5 coupling , creating signaling imbalance when the expression of signaling components is diminished . This effect could be either exacerbated or compensated by the loss of RGS9-2 that also controls AC5 activity ( Xie et al . , 2012 ) and signaling downstream from D2 receptors ( Cabrera-Vera et al . , 2004; Celver et al . , 2010; Rahman et al . , 2003 ) , further contributing to signaling imbalance . Alternatively , preservation of D1R mediated behavioral responses in Pdcl cKO mice may also indicate that signaling pathways other than cAMP initiated by D1 receptors ( Nishi et al . , 2011 ) play compensatory role . In any event , our study demonstrates the importance of proper Golf-AC5 complex expression and assembly for the balance of the neurotransmitter signaling in striatal neurons and introduces PhLP1 as a critical regulator of the process and an interesting molecular player to consider in the pathology of dystonia and possibly other movement disorders .
Pdcl flx/flxmice ( Lai et al . , 2013 ) were crossed with Rgs9-Cre mice ( Dang et al . , 2006 ) to generate striatal specific PhLP1 conditional knockout ( Pdcl cKO ) mice . Pdclflx/flx Cre ( - ) control littermates derived from heterozygous breeding pairs were used for all experiments . Mice were housed in groups on a 12 hr light–dark cycle with food and water available ad libitum . Males and females ( 2–5 months ) were used for all experiments . All procedures were approved by the Institutional Animal Care and Use Committee ( IACUC ) at The Scripps Research Institute . Generation and characterization of Gnal+/- ( Belluscio et al . , 1998 ) , Gng7-/- ( Sasaki et al . , 2013 ) ) and Adcy5-/- ( Lee et al . , 2002 ) mice have been described previously . Full length of PhLP1 and N-terminal ( 1–75 a . a . ) truncation of PhLP1 ( ΔNT-PhLP1 ) were cloned into pcDNA3 . 1 vector as previously ( Lukov et al . , 2005 ) . Venus155-239-Gβ2 and Venus1-155-Gγ7 constructs were generated by replacing Gβ1 in Venus155-239-Gβ1 construct with Gβ2 and Gγ2 in Venus1-155-Gγ2 construct with Gγ7 ( Hollins et al . , 2009 ) . Construction of masGRK3ct-Nluc ( Posokhova et al . , 2013 ) and Flag tagged AC5 ( Xie et al . , 2012 ) was reported previously . Gαolf and D1R cDNAs were purchased from Missouri S&T cDNA Resource Center . Flag-tagged Ric-8B in pcDNA3 . 1 ( Von Dannecker et al . , 2006 ) were gifts from Dr . Bettina Malnic . Hybridoma cell lines expressing mouse monoclonal AC5 antibody against human AC5 peptide CGNQVSKEMKRMGFEDPKDKN were commercially generated by Genscript . Hybridomas were cultured in DMEM/F-12 supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin . Synthetic peptide ( CGNQVSKEMKRMGFEDPKDKN ) was covalently immobilized to beaded agarose using SulfoLink Immobilization Kit ( Pierce Biotechnology ) . Antibodies were purified from collected hybridoma culturing medium by affinity chromatography using immobilized antigen peptide . Other antibodies used were against: c-myc ( Genescript ) , Gβ1 ( Lee et al . , 2004 ) and Gαolf ( Corvol et al . , 2001 ) . PhLP1 ( Lai et al . , 2013 ) ; β-actin ( AC-15 ) ( Sigma-Aldrich , St . Louis , MO ) ; Gβ2 , Gαq , Gγ7 and Gαo ( K-20 ) ( Santa Cruz Biotechnology , Dallas , TX ) ; Gαi1/2 ( Affinity BioReagents , Golden , CO ) ; GFP ( clones 7 . 1 and 13 . 1; Roche Applied Science ) ; Gβ5 and RGS7 and RGS9-2 ( Martemyanov et al . , 2005 ) . For immunoblot analysis or immunoprecipitation , striatal tissue ( ~15 mg ) or transfected HEK293T/17 cells were homogenized by sonication in lysis buffer ( 1xPBS , 150 mM NaCl , 0 . 5% dodecyl nonaoxyethylene ether ( C12E9 ) containing complete protease inhibitor cocktail ( Roche ) and phosphatase inhibitor cocktail 1 and 2 ( Sigma ) . The homogenate was centrifuged at 16 , 000×g for 15 min . For immunoprecipitation , the supernatant was incubated with 3 μg of antibody as indicated and 10 μL of protein G beads for 1 hr at 4°C . Beads were washed 3 times with lysis buffer . Protein sample was eluted in SDS sample buffer containing 4M urea , incubated at 42°C for 15 min , resolved by SDS-PAGE , transferred onto PVDF membrane and subjected to immunoblot analysis . All experiments involving cultured cells were performed in HEK293/17 cell line obtained from ATCC ( Manassas , VA ) . The company certifies authenticity of the cell line , and guarantees it to be free of contaminants and pathogens . The cells were maintained in standard DMEM medium , and frozen in aliquots from the stock received from the ATCC . The cells were grown for no more than 20 passages . The laboratory has tested tissue culture facility and found it to be free of mycoplasma contamination . Recombinant Gαs were expressed in BL21 ( DE3 ) E . coli strain and purified by affinity chromatography on HisTALON column ( Clontech , Mountain View , CA ) as described previously ( Lee et al . , 1994 ) . Gαs was activated by incubation with 20 μM GTPγS in the assay buffer containing 20 mM Tris-HCl pH 7 . 8 , 10 mM MgCl2 , 1 mM EDTA , 1 mM dithiothreitol for 30 min at 30°C . The unbound GTPγS was then removed by Zeba spin desalting column ( Life Technologies , Carlsbad , CA ) . His-tagged full-length PhLP1 as well as the N-terminal ( 1–75 a . a . ) truncation of PhLP1 were purified from E . coli as previously described ( Savage et al . , 2000a ) . The purity of the recombinant proteins was assessed by Coomassie staining following gel separation and was found to be at least 80% . For membrane preparation , striatal tissues were homogenized in buffer containing 250 mM sucrose , 20 mM Hepes pH 8 . 0 , 1 mM EDTA , 2 mM MgCl2 , 1 mM DTT and proteinase inhibitors . The homogenate was centrifuged at 2000 g to remove nuclei , followed by centrifugation at 25 , 000 rpm in Beckman SW28 . 1 rotor for 35 min in 23/43% sucrose gradient to isolate the membrane fraction . The plasma membranes were carefully collected from the layer at the 23/43% sucrose interface . The protein concentrations in plasma membrane preparations were then determined by Pierce 660nm Protein Assay Reagent ( Thermo Fisher Scientific , Waltham , MA ) . Striatal tissues were homogenized in 0 . 1 N HCl ( 20 μL per mg tissue ) . Lysates were centrifuged at 600 g for 10 min . Supernatants were collected , diluted 50-fold and cAMP concentrations were quantified using a cAMP enzyme immunoassay kit ( cAMP Direct EIA ) following the acetylated version protocol ( Enzo Life Sciences , Farmingdale , NY ) . The activity of adenylyl cyclase in striatal membrane preparations ( 1 μg protein/reaction ) was determined as described previously ( Xie et al . , 2012 ) . Briefly , 1 μg of striatal membrane was treated with vehicle ( basal ) , or indicated stimulator for 10 min at 30ºC in adenylyl cyclase assay buffer ( 50 mM HEPES pH 8 . 0 , 0 . 6 mM EDTA , 100 μg/mL BSA , 100 μM 3-isobutyl-1-methylxanthine ( IBMX ) , 3 mM phosphoenolpyruvate potassium , 10 μg/mL pyruvate kinase , 5 mM MgCl2 and 100 μM adenosine triphosphate ( ATP ) . Reactions were stopped by adding an equal volume of 0 . 2 N HCl . For dose response curves of Gαs-GTPγS or forskolin ( FSK ) stimulated adenylyl cyclase activity experiments , striatal membranes were pre-incubated with 0 . 5 μM purified PhLP1 for 20 min on ice and then stimulated with increasing doses of Gαs-GTPγS or FSK as indicated . The resulting cAMP in the sample was determined by cAMP Direct EIA kit . Total RNA from striatal tissues was extracted and quantified as previously ( Orlandi et al . , 2015 ) . Briefly , striatal tissues were homogenized in TRIZOL reagent ( Invitrogen , ) according to the manufacturer's instructions . cDNA was generated from 0 . 5 µg of total RNA using qScript cDNA SuperMix ( Quanta Biosciences , Gaithersburg , MD ) according to the manufacturer's instructions . To analyze the RNA expression pattern of the target genes , the 7900HT Fast Real-Time PCR System ( Applied Biosystems ) was used with the Taqman gene expression master kit . Three biological replicates and four technical replicates for each sample were used . 10 ng of each sample were used in each real-time PCR ( TaqMan Gene Expression Assay ID probes: Pdcl: Mm01327170_m1; Adcy5: Mm00674122_m1; Gnal: Mm01258217_m1; Gnb2: Mm00515865_g1; Gng7: Mm00515876_m1; Rgs9: Mm01250425_m1; Applied Biosystems ) . The expression ratio of the target genes was calculated using 2−ΔΔCT method ( Livak and Schmittgen , 2001 ) with18S ribosomal RNA ( ID: Mm03928990_g1 ) as reference . Data are shown as mean ± S . E . M . Agonist-dependent cellular measurements of bioluminescence resonance energy transfer ( BRET ) between masGRK3ct-Nluc and Venus-Gβ2γ7 were performed to visualize the action of G protein signaling in living cells as previously described with slight modification ( Kumar et al . , 2014 ) . Briefly , dopamine D1 receptor , Gαolf , Venus156-239-Gβ2 , Venus1-155-Gγ7 , Flag-Ric-8B and masGRK3ct-Nluc constructs together with PhLP1 or ΔNT-PhLP1 were transfected into HEK293T/17 cells at a 1:6:1:1:1:1:1 ratio using lipofectamine LTX transfection reagent ( Invitrogen ) . 7 . 5 µg total DNA was delivered per 3 . 5 × 106 cells in a 6-cm-dish . 16–24 hr post transfection , cells were stimulated with 100 µM dopamine followed by treatment with 100 µM SCH39166 . The BRET signal was determined by calculating the ratio of the light emitted by the Venus-Gβ2γ7 ( 535 nm ) over the light emitted by the masGRK3ct-Nluc ( 475 nm ) . The average baseline value recorded prior to agonist stimulation was subtracted from BRET signal values , and the resulting difference ( ΔBRET ) was obtained . Mice were anesthetized by Avertin ( tribromoethanol ) and perfused transcardially with phosphate buffered saline ( PBS ) and 4% paraformaldehyde ( PFA ) . Brains were collected and postfixed in 4% PFA overnight , and stored in 30% sucrose solution for cryoprotection for 3–5 days . Brains were sectioned into 50 µm slices by a sliding microtome ( SM2000R , Leica ) . After sectioning , sample slices were stored in PBS at 4°C or in antifreeze solution at -20°C for long-term storage . Immunofluorescent staining was performed as described previously ( Gharami et al . , 2008 ) . Primary antibodies against enkephalin and substance-P ( Immunostar Inc . ; 1:1 , 1000 ) were used for detecting the projection from dopamine receptor D1 or D2 medium spiny neurons . Fluorescent dye-conjugated secondary antibodies ( Alexa Fluor 594 , Jackson ImmunoResearch Inc . ) were applied during staining . The staining was performed on two brain sections including either external globus pallidus ( GPe ) or substantia nigra pars reticulata ( SNr ) . After acquiring images by using a 4X objective ( CFI Plan Apo , Nikon ) on Nikon Ti microscope , the Nikon NIS-Elements software was used to measure mean intensity in designated area . Then values of control and mutant groups , with backgrounds subtracted , were evaluated by Student’s unpaired t-test . For neuronal cell counting in striatum , we first performed Nissl-staining in six coronal sections with 300 µm interval . Stereo Investigator software ( MicroBrightField Inc . ) was employed to measure the volume of striatum and evaluate the neuronal number . The number of neurons was estimated using a fractionator sampling method described previously ( Baydyuk et al . , 2011 ) . The analysis was performed blinded to the genotypes . Locomotor activities were evaluated in automated video tracking ANY-maze open field chambers ( Stoelting , Wood Dale , IL ) under illuminated conditions . Mice were habituated to the testing room for 1 hr before the test on each day . On the first day , naïve mice were placed in the novel chambers without injection , and allowed to explore the chambers for 2 hr . Horizontal activity was measured in terms of the total distance traveled or distance traveled in 10-min intervals . Thigmotaxis ( wall-hugging ) for each subject was determined by dividing the distance traveled in the 10-cm-wide perimeter of the chamber by the total distance traveled during the 2-hr session . For pharmacological studies , mice were injected with vehicle ( 10 mL/kg , i . p . ) , D1 receptor agonist SKF 38 , 393 ( 50 mg/kg , i . p . ) , or A2a receptor antagonist SCH 58 , 261 ( 3 mg/kg , i . p . ) , and then immediately placed in the open-field chambers . Activity was monitored for 1 hr . Dose response of the same drug treatments were done on the same mice starting from the low dose first with 2-day intervals between different doses . SKF 38 , 393 was dissolved in saline , and SCH 58 , 261 was dissolved in saline solution containing 10% DMSO and 10% Kolliphor EL ( Sigma Aldrich , St . Louis , MO ) . Accelerating rotarod performance was tested using a five-station rotarod treadmill ( IITC Rotarod , IITC Life Science , Woodland Hills , CA ) . Mice were habituated in the test room for an hour before the testing . Three trials were performed per day over 3 days for a total of nine trials for each animal . After placing a mouse on the rod , it was accelerated from 8 to 22 r . p . m . in 2 min . The endurance of mice on the rotarod was measured by the time to fall to the floor of the apparatus , or to turn around one full revolution while hanging onto the drum . Catalepsy was measured in the bar test . Briefly , one hour after haloperidol ( 2 mg/kg , i . p . ) administration , mice forepaws were gently placed over a horizontal bar fixed at a height of 5 cm above the working surface . The length of time during which the animal retained this position until the removal of one of its forepaws was recorded with the cutoff time of 180 s . Grip strength of mice was assessed using Grip-Strength Meter ( Ugo Basile , Italy ) . The mouse was placed over a base plate , in front of a grasping bar . The bar was fitted to a force transducer connected to the Peak Amplifier . When pulled by the tail , the mouse instinctively grasped to the bar until the pulling force overcomes their grip strength . After the mouse lost its grip on the grasping bar , a peak preamplifier automatically stored the peak pull force ( in grams ) achieved by the forelimbs . Grip strength of each mouse was averaged from 5 consecutive trials . Acoustic startle response tests were performed in acoustic apparatuses with mouse enclosures ( San Diego Instruments , San Diego , CA ) in sound-attenuating cubicles . The amplitude of each startle response was measured using a piezoelectric movement–sensitive platform . Acoustic stimuli and steady background noise ( 70 dB ) were delivered through a loudspeaker . Briefly , mice were put in the mouse enclosures and acclimated for 2 min . Six trials of 40-ms 120 dB white noise bursts were then presented with a variable intertribal interval . Vmax values recorded by Acoustic Startle Program ( San Diego Instruments , San Diego , CA ) were averaged from six startle trials for each mouse . | In the brain , cells called neurons communicate with one another using chemicals called neurotransmitters , which bind to receptors on the surface of cells . In most cases , the binding of neurotransmitter causes a protein known as a G protein to interact with the receptor . G proteins are made up of three subunits: α , β and γ . After binding to the receptor , the α subunit separates from the β/γ subunits – which remain together – and both components then act as signals to activate specific targets in the cell . There are many different α , β and γ subunits , which participate in various signaling pathways in different parts of the brain . In the striatum – a region involved in controlling movement – a particular combination of α , β and γ subunits called Golf is responsible for activating an enzyme called adenylyl cyclase type 5 ( AC5 ) . Traditionally , G protein signalling has been thought to occur in stages so that the binding of neurotransmitter to receptors on striatal neurons would lead to the dissociation of Golf into the α and βγ subunits . Then the α and βγ subunits are thought to bind to and activate AC5 . However , Xie et al . now show that all three subunits of Golf are already found in a stable group ( or complex ) with AC5 in striatal neurons . Neurotransmitter binding to the receptors causes the entire Golf-AC5complex to rearrange and this process activates AC5 . A particular chaperone protein regulates the assembly of the G protein-AC5complex . Mice that lack the gene that encodes this chaperone in striatal neurons struggle to learn how to balance on a rotating rod . In humans , mutations in the genes that encode Golf and AC5 cause dystonia , which is a disorder characterised by involuntary movements . Given the evidence linking this chaperone protein to the regulation of Golf-AC5 signaling , future experiments should investigate whether it might also contribute to dystonia . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2015 | Stable G protein-effector complexes in striatal neurons: mechanism of assembly and role in neurotransmitter signaling |
The dynamics of different ionic currents shape the bursting activity of neurons and networks that control motor output . Despite being ubiquitous in all animal cells , the contribution of the Na+/K+ pump current to such bursting activity has not been well studied . We used monensin , a Na+/H+ antiporter , to examine the role of the pump on the bursting activity of oscillator heart interneurons in leeches . When we stimulated the pump with monensin , the period of these neurons decreased significantly , an effect that was prevented or reversed when the h-current was blocked by Cs+ . The decreased period could also occur if the pump was inhibited with strophanthidin or K+-free saline . Our monensin results were reproduced in model , which explains the pump’s contributions to bursting activity based on Na+ dynamics . Our results indicate that a dynamically oscillating pump current that interacts with the h-current can regulate the bursting activity of neurons and networks .
Rhythmic behaviors such as walking , breathing , and running are controlled by central pattern generators , networks of neurons that produce rhythmic activity without sensory input ( Marder and Calabrese , 1996; Marder and Bucher , 2001 ) . The rhythmic bursting activity of each constituent neuron within a central pattern generator is shaped by the dynamics of various ionic currents that are intrinsic to each neuron ( Harris-Warrick , 2002 ) . Many of these neurons share subsets of ionic currents with similar functional properties that give rise to bursting activity . For example , many central pattern generator neurons have a persistent Na+ current or a low-threshold Ca2+ current that supports bursting ( Opdyke and Calabrese , 1994; Butera et al . , 1999; Del Negro et al . , 2002; Rybak et al . , 2004 ) , a hyperpolarization-activated inward current that provides recovery from inhibition to initiate bursting ( Angstadt and Calabrese , 1989; Golowasch and Marder , 1992 ) , and a transient K+ current that impedes initiations of action potentials and bursts ( Simon et al . , 1992 ) . Moreover , modulation of these ionic currents can alter the timing and intensity of these neurons’ bursting activity ( e . g . , Tobin and Calabrese , 2005; Koizumi and Smith , 2008 ) . Although the Na+/K+ pump is ubiquitous in all animal cells , its role in regulating the bursting activity of neurons in general has not been widely considered . The pump is a transmembrane protein that maintains the intracellular concentrations of Na+ and K+ by exchanging three intracellular Na+ ions for two extracellular K+ ions with each cycle of ATP ( adenosine triphosphate ) hydrolysis ( Thomas , 1972a; De Weer and Geduldig , 1973 ) . Because the exchange of Na+ and K+ ions is unequal , the pump is electrogenic as it generates an outward current ( Glitsch , 2001 ) . In addition to maintaining internal concentrations of Na+ and K+ , the pump contributes a voltage drop to the resting membrane potential ( Hodgkin and Keynes , 1955; Carpenter and Alving , 1968; Smith et al . , 1968; Baylor and Nicholls , 1969 ) and is able to generate a slow afterhyperpolarization after a train of action potentials when its activity is enhanced by increased intracellular Na+ ( Gage and Hubbard , 1966; Nakajima and Takahashi , 1966; Rang and Ritchie , 1968; Baylor and Nicolls , 1969; Sokolove and Cooke , 1971; Bolton , 1973; Mat Jais et al . , 1986; Gordon et al . , 1990; Catarsi and Brunelli , 1991; Lombardo et al . , 2004; Pulver and Griffith , 2010; Gulledge et al . , 2013 ) . In the context of motor patterns , the pump appears to play an important role in regulating bursting activity ( Ballerini et al . , 1997; Tobin and Calabrese , 2005; Zhang and Sillar , 2012 ) . For example , Zhang and Sillar ( 2012 ) found that blocking the pump with ouabain abolished the slow afterhyperpolarization of spinal cord central pattern generator neurons in Xenopus laevis tadpoles , resulting in longer swimming episodes . In a separate study , Tobin and Calabrese ( 2005 ) observed that inhibition of the pump with the neuropeptide , myomodulin , or with ouabain speeds up the bursting activity of oscillator heart interneurons in the leech heartbeat central pattern generator ( Tobin and Calabrese , 2005 ) . These studies show that the pump can serve as a target for modulating the bursting activity of neurons and networks that program motor output . Many studies have explored the function of the pump by inhibiting its activity; fewer have investigated the pump’s function by stimulating its activity . For example , Zhang et al . ( 2015 ) recently found that stimulating the pump activity of central pattern generator neurons in Xenopus enhances the ultraslow hyperpolarization , which suppresses excitability of the entire motor network . Nevertheless , the effects of stimulating pump on the ongoing activity of rhythm generating neurons have yet to be explored . In the present study , we had two principal goals . First , we wanted to reveal experimentally the mechanisms that underlie the effects of a stimulated pump on the bursting activity of central pattern generator neurons , especially with respect to the temporal or burst characteristics of these neuron’s bursting activity such as period or duty cycle . Second , we wanted to develop a mathematical model that could capture our experimental results and help identify mechanisms . For our analysis , we used leech oscillator heart interneurons , which participate in half-center oscillators to pace the heartbeat central pattern generator . We examined the influence of the pump on bursting based on changes in the burst characteristics of these oscillator heart interneurons . We used monensin , a Na+/H+ antiporter , to increase Na+ concentrations to stimulate the pump ( Hill and Licis , 1982 ) . Our results show that monensin enhances the outward pump current , which hyperpolarizes the membrane potential of oscillator heart interneurons . We also found that stimulation of pump activity by monensin speeds up the bursting activity of oscillator heart interneurons . Blocking the h-current of these neurons with Cs+ while stimulating the pump with monensin failed to speed up bursting . Our biophysical model captured these experimental results by simulating the interaction between the pump current and the h-current to control the interburst interval and thus the period . Taken together , our study leads us to conclude that in the presence of the h-current , the electrogenic activity of the pump can play a significant role in the dynamics of bursting activity in the leech heartbeat central pattern generator and likely in other rhythmically bursting neuronal networks .
To delineate the role of the Na+/K+ pump in central pattern generator neurons , we used oscillator heart interneurons that pace bursting activity in the leech heartbeat central pattern generator . There are two pairs of these oscillator heart interneurons in each animal , with each pair located in the third and fourth segmental ganglia of the ventral nerve cord . Both neurons in each ganglion form mutual inhibitory synaptic connections , thereby constituting a half-center oscillator . We used only individual isolated ganglia in our experiments to determine the contribution of the pump to ongoing rhythmic bursting in oscillator heart interneurons . The pump current is proportional to the pump rate , which is itself proportional to intracellular Na+ concentrations ( Thomas , 1972a ) . Thus , the pump can be stimulated by increasing the intracellular loading of Na+ from an electrode filled with a Na+-based solution , which has been demonstrated in other neurons such as mechanoreceptors in leeches ( Jansen and Nicholls , 1973; Catarsi and Brunelli , 1991; Lombardo et al , 2004 ) and snails ( Kerkut and Thomas , 1965 ) . To determine if the intracellular loading of Na+ hyperpolarizes the membrane potential of the oscillator heart interneurons , we recorded one oscillator heart interneuron with an extracellular electrode and impaled the contralateral oscillator heart interneuron with an intracellular electrode , filled with the standard 2M KAcetate and 20 mM KCl solution , in normal saline . Upon impalement , we measured the base potential , defined as midway between an undershoot trough and the next threshold ( first peak of the third derivative ) of the intracellularly recorded neuron in the first 10 min . Within 3 min of impalement , the base potential stabilized to −40 . 8 ± 1 . 6 mV ( n = 5 ) , a voltage that is consistent with previous observations ( Olsen and Calabrese , 1996 ) , and both neurons exhibit their usual rhythmic bursts of action potentials for 20 min or more ( Figure 1A1 ) . In another group of preparations , we substituted the KAcetate and KCl with equimolar concentrations of NaAcetate and NaCl and recorded the activity of the oscillator heart interneurons in normal saline . When an oscillator heart interneuron was impaled with a Na+-filled electrode , its rhythmic activity decreased rapidly over time and its base potential was noticeably more hyperpolarized than the K+-loaded neurons ( Figure 1A2 ) . We compared the average base potential of both groups of preparations for the first ten minutes and found that the base potential of the Na+-loaded neurons was significantly more hyperpolarized than the base potential of the K+-loaded neurons ( Figure 1A3 , n = 5 , split-plot ANOVA , F1 , 8 = 1847 . 7 , p=0 . 006 ) . Despite being hyperpolarized , the Na+-loaded neurons were very responsive to brief depolarizing pulses , indicating that they were still healthy . In summary , these results show that the intracellular leakage of Na+ does hyperpolarize the membrane potential of oscillator heart interneurons . 10 . 7554/eLife . 19322 . 003Figure 1 . Hyperpolarization of the oscillator heart interneurons and suppression of their spiking activity by intracellular leakage of Na+ from an electrode and by monensin . ( A1 ) An extracellular ( blue ) trace of one oscillator heart interneuron and an intracellular ( vermilion ) trace of a contralateral oscillator heart interneuron that was impaled with a K+-filled intracellular electrode . ( A2 ) Impalement of an oscillator heart interneuron with a Na+-filled electrode gradually suppressed its spiking activity and hyperpolarized the neuron . There was no change in the bursting activity of the extracellularly recorded neurons in the ( A1 ) K+ and ( A2 ) Na+ recordings . ( A3 ) During the first ten minutes , the average base potential of Na+-loaded neurons ( closed circles ) was significantly more hyperpolarized than the base potential of K+-loaded neurons ( open circles ) . Such differences persisted well into the 15th and 20th minute . The data are represented as mean ± SEM , with the asterisk ( * ) representing significant differences between the K+ and Na+ base potentials ( split-plot ANOVA , F1 , 8 = 1847 . 7 , p=0 . 006 ) . ( B1 ) Extracellular ( blue ) and intracellular ( vermilion ) traces from a pair of oscillator heart interneurons that were initially bathed in control saline and showed normal alternating bursting . ( B2 ) When the oscillator heart interneurons were bathed in Ca2+-free saline with 2 mM Cs+ and 1 . 8 mM Mn2+ , they produced a more tonic firing pattern that was interspersed with synchronized oscillations . ( B3 ) When the oscillator heart interneurons were subsequently treated with 10 µM monensin in the same Ca2+-free saline , the spiking activity of both oscillator heart interneurons were suppressed and the membrane potential of the intracellularly recoded neuron gradually hyperpolarized . DOI: http://dx . doi . org/10 . 7554/eLife . 19322 . 003 To determine the contributions of the Na+/K+ pump under multiple experimental treatments , we used monensin , an antibiotic that functions as a Na+-H+ antiporter in cell membranes ( Lichtshtein et al . , 1979; Hill and Licis , 1982; Zhang et al . , 2015 ) . Monensin has been found to increase intracellular Na+ concentrations , which stimulates the pump ( Lichtshtein et al . , 1979; Hill and Licis , 1982; Zhang et al . , 2015 ) and hyperpolarizes the membrane potential of various cell types from other systems ( Lichtshtein et al . , 1979; Tsuchida and Otomo , 1990; Satoh and Tsuchida , 1999; Doebler , 2000; Wang et al . , 2012 ) . To determine if monensin also hyperpolarizes the membrane potential of oscillator heart interneurons from the leech heartbeat system , we performed dual intracellular and extracellular recordings from a pair of oscillator heart interneurons and measured the base potential of the intracellularly recorded neuron before and after external application of monensin . In normal saline , both oscillator heart interneurons fired their usual alternating bursts of action potentials and the base potential of the intracellularly recorded neuron was −41 . 5 ± 1 . 5 mV ( n = 5 ) . We then applied Ca2+-free saline , which contained 2 mM Cs+ to block the h-current ( Angstadt and Calabrese , 1989 , 1991 ) and 1 . 8 mM Mn2+ to block all Ca2+ currents as well as synaptic transmission ( Angstadt and Calabrese , 1991 ) . Once the Ca2+-free saline took effect , the oscillator heart interneurons no longer burst regularly but instead fired tonically , indicating that the neurons were synaptically isolated and incapable of normal bursting ( Figure 1B2 ) . Such tonic firing was interspersed by brief Na+-based synchronized oscillations ( Figure 1B2 ) , a characteristic of Mn2+ exposure in leech neurons ( Angstadt and Friesen , 1991 ) . We measured the base potential of these neurons after these synchronized oscillations had appeared , which was 2–7 min after applying the Ca2+-free saline . We then added 10 µM monensin to the Ca2+-free saline , which abolished the synchronized oscillations and eventually suppressed spiking activity ( Figure 1B3 ) . We chose the concentration of 10 µM for monensin because we wanted to produce the maximum effect within the shortest amount of time ( see details below on the effects of lower concentrations of monensin on bursting activity ) . Within 7–10 min of adding monensin , the base potential under the Ca2+-free monensin saline became significantly more hyperpolarized than the base potential under Ca2+-free saline ( Figure 1B3 , −43 ± 1 . 8 mV for pre-monensin vs . −54 . 4 ± 3 . 4 mV for monensin , n = 5 , paired t-test , p=0 . 02 ) . Moreover , the suppression of spiking activity in the extracellularly recorded neurons is consistent with the hyperpolarization of the intracellularly recorded neurons ( Figure 1B3 ) . Thus , consistent with our hypothesis and with previous studies ( e . g . , Doebler , 2000; Lichtshtein et al , 1979; Tsuchida and Otomo , 1990; Satoh and Tsuchida , 1999; Wang et al . , 2012 ) , monensin hyperpolarizes the membrane potential of oscillator heart interneurons , which suppresses their spiking activity . To determine if the hyperpolarized membrane potential by monensin was due to the outward current generated by a stimulated Na+/K+ pump , we voltage-clamped one of the oscillator heart interneurons from an isolated ganglion and looked for shifts in the membrane current brought about by monensin as well as by monensin plus strophanthidin ( Figure 2A1 ) . If monensin does indeed stimulate the pump , we should see an outward shift in the membrane current . We first bathed the oscillator heart interneurons in the same Ca2+-free saline to block the h-current and all Ca2+ currents , thereby also suppressing synaptic transmission . The neurons were then voltage-clamped at −45 mV for five minutes and later treated with 10 µM monensin for ten minutes ( Figure 2A1 ) . The holding potential of −45 mV was chosen because it is near the observed base potential of oscillator heart interneurons when bathed in Ca2+-free saline ( Figure 1B2 ) . Moreover , we wanted to reduce the escape spiking of these neurons ( Norris et al . , 2007 ) . Before adding monensin , the average holding current at −45 mV was −26 . 4 ± 16 . 9 pA ( Figure 2A2 , n = 5 ) , which was measured at the fifth minute of voltage-clamping the neurons in Ca2+-free saline . Monensin suppressed all escape spiking and induced an outward current which moved the holding current to 62 . 4 ± 29 . 6 pA ( Figure 2A1-2 , n = 5 , Tukey’s test , p=0 . 009 ) , measured at the tenth minute after monensin application . This monensin-induced outward current was apparently blocked , however , by the application of 100 µM strophanthidin , resulting in a large inward holding current ( Figure 2A1-2 , −180 . 0 ± 33 . 3 pA , n = 5 , Tukey’s test , p=0 . 0002 ) and a resumption of escape spiking by the neurons ( Figure 2A2 ) . The difference between the average outward current generated by monensin and the average inward current observed upon blocking the pump with strophanthidin , suggests that the pump is able to generate a maximum outward current of 242 . 4 ± 16 . 0 pA , assuming a saturating effect of 10 µM monensin on the pump . Subtracting the outward current generated by monensin from this maximum pump current , suggests that the resting pump current is about 180 . 0 ± 33 . 3 pA ( or 74 . 0 ± 11 . 2% of the maximum pump current ) . Thus , the maximum pump current appears to have a limited range of 202 to 289 pA , with about three-quarters of this current being generated under resting conditions , leaving another quarter of the current available for enhancement by 10 µM monensin . 10 . 7554/eLife . 19322 . 004Figure 2 . Monensin stimulates the outward Na+/K+ pump current . ( A1 ) Membrane current trace from an oscillator heart interneuron with its membrane potential ( Vm ) voltage-clamped at −45 mV ( see inset ) in Ca2+-free saline with 1 . 8 mM Mn2+ plus 2 mM Cs . Changes in the neuron’s membrane current ( Im ) were observed under three experimental treatments: pre-monensin saline for five minutes , 10 µM monensin for 10 min , and 10 µM monensin plus 100 µM strophanthidin ( SPTD ) for another five minutes . ( A2 ) A scatterplot of membrane currents from five preparations , with each green dashed line representing a mean for each of the three experimental treatments . Monensin induced a significant outward current relative to pre-monensin saline . Monensin plus strophanthidin induced a significant inward current relative to pre-monensin or monensin saline . The asterisks ( * ) represent significance from the pre-monensin saline whereas the hashtag ( # ) represents significance from the monensin saline ( Tukey’s test , p<0 . 05 for all tests ) . ( B1 ) Membrane currents from the same oscillator heart interneuron that was voltage-clamped at −45 mV in the same Ca2+-free saline before and after treatment with 10 µM monensin . Both currents were generated by a slow ramp-clamp protocol , with each voltage ramp running from −45 mV to −80 mV and back ( see inset ) . Compared to the membrane current under pre-monensin saline ( blue trace ) , the membrane current under monensin saline ( vermilion trace ) shifted outwards across the entire range of negative voltage-ramp values . ( B2 ) Current-voltage relations under pre-monensin ( blue line ) and monensin ( vermilion line ) treatments from the same oscillator heart interneuron were generated based on the ( B1 ) voltage-ramp traces . DOI: http://dx . doi . org/10 . 7554/eLife . 19322 . 004 To determine if the outward current produced by monensin is due to an enhanced pump current and not by changes in membrane conductance , we voltage-clamped another set of oscillator heart interneurons in the same Ca2+-free saline for 4–6 min , before generating a series of three voltage-ramps using a slow ramp-clamp protocol , with each ramp running from −45 mV to −80 mV and back ( See inset in Figure 2B1 ) . We then added 10 µM monensin to the Ca2+-free saline for 6–9 min before generating another series of voltage ramps ( Figure 2B1 ) . In four of the five preparations , we saw outward shifts in the membrane current across the entire range of negative voltage-ramp values ( Figure 2B1 ) . These outward shifts occurred without any significant change in conductance ( Figure 2B2 , 15 . 3 ± 1 . 6 nS for pre-monensin saline vs . 13 . 0 ± 1 . 3 nS for monensin saline , n = 5 ) , which is consistent with our hypothesis that the outward current brought induced by monensin is driven by stimulation of the pump and not by changes in conductance . Since we have established that monensin stimulates the Na+/K+ pump , which hyperpolarizes the membrane potential , we predicted that treating the oscillator heart interneurons with monensin would slow down their bursting activity as indicated by an increased period ( for the definition of period , see Figure 3A and Materials and methods ) . To test this hypothesis , we again performed dual intracellular and extracellular recordings from a pair of oscillator heart interneurons in normal saline . Contrary to what we expected , when we applied 10 µM monensin for 15 min , we found that the average period of both oscillator heart interneurons actually decreased ( Figure 3 , 9 . 3 ± 1 . 0 s for control saline vs . 3 . 9 ± 0 . 2 s for monensin saline , n = 5 , paired t-test , p=0 . 007 ) , which were detectable within two minutes of applying monensin . We measured the period after 5 . 1–10 . 5 min of adding monensin because shortening of the period stabilized during this time period ( see example in Figure 3—figure supplement 1 ) . In four of the five preparations , the activities of both neurons initially transitioned from bursting to tonic-like firing and back to bursting again ( Figure 3B ) . The disorganized tonic-like firing pattern did not appear in the extracellular recordings ( Figure 4 ) , leaving us to conclude that this pattern was due to a nonspecific leak current introduced by the intracellular electrode ( Cymbalyuk et al . , 2002 ) . We also measured the base potential before and after application of monensin but did not find any significant changes . Thus , monensin actually speeds up the bursting activity of oscillator heart interneurons without significantly affecting their base potential . 10 . 7554/eLife . 19322 . 005Figure 3 . Stimulation of the Na+-K+ pump with monensin speeds up the bursting activity of oscillator heart interneurons as half-center oscillators . ( A ) Extracellular ( blue ) and intracellular ( vermilion ) traces from a pair of oscillator heart interneurons functioning as a half-center oscillator . Burst characteristics such as the period , burst duration , and interburst interval can be measured from each neuron’s bursting activity . The period was measured from the middle action potential ( diamond symbol ) of one burst to the middle action potential of the next burst . ( B ) Extracellular ( blue ) and intracellular ( vermilion ) traces from a pair of oscillator heart interneurons . In control saline , both neurons function as a half-center oscillator by firing alternating bursts of action potentials . Adding 10 µM monensin to the saline resulted in initial tonic firing by both neurons followed by alternating bursts of action potentials with a reduced period . DOI: http://dx . doi . org/10 . 7554/eLife . 19322 . 00510 . 7554/eLife . 19322 . 006Figure 3—figure supplement 1 . The cycle-to-cycle effects of monensin on the period of oscillator heart interneurons . ( A ) Initial application of monensin rapidly shortens the period towards a stable minimum value . The concentration of 10 µM monensin ( vermilion line ) shortens the period more rapidly than the lower concentration of 1 µM ( blue line ) . The amount of time a period needs to reach its minimum value at the 200th cycle can be measured by summing up all the periods leading up to that 200th cycle . ( B ) A scatterplot of the amount of time that has passed before the period has reached its value at the 200th cycle in both 1 µM and 10 µM monensin treatments . ( C ) A scatterplot of the period at the 200th cycle in both 1 µM and 10 µM monensin treatments . The dashed green lines in the scatter plots represent means whereas the asterisk ( * ) represents significance from control ( unpaired t-test , p=0 . 003 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19322 . 00610 . 7554/eLife . 19322 . 007Figure 4 . Effects of stimulating the Na+-K+ pump with monensin while blocking the h-current on the burst characteristics of oscillator heart interneurons as half-center oscillators . Extracellular traces of bursting activity by left ( blue ) and right ( vermilion ) oscillator heart interneurons [HN ( L ) and HN ( R ) neurons] initially bathed in ( A1 ) control ( normal ) saline and subsequently treated with ( A2 ) saline that contained 10 µM monensin followed by another treatment with ( A3 ) saline that contained 10 µM monensin plus 2 mM Cs+ . Extracellular traces of another pair of oscillator heart interneurons first treated with ( A4 ) saline that contained 2 mM Cs+ followed by another treatment with ( A5 ) saline that contained 2 mM Cs+ plus 10 µM monensin . Scatterplots of the ( B1 ) period , ( B2 ) burst duration , ( B3 ) interburst interval , ( B4 ) duty cycle , and ( B5 ) intraburst spike frequency that were measured from the extracellular traces of five preparations under ( A1-3 ) three experimental treatments . Monensin decreased significantly the ( B1 ) period , ( B2 ) burst duration , and ( B3 ) interburst interval relative to control . The ( B4 ) duty cycle and ( B5 ) intraburst spike frequency were unchanged . Monensin plus Cs+ increased significantly ( B1 ) the period , ( B3 ) interburst interval , and ( B5 ) intraburst spike frequency relative to control . Because the ( B2 ) burst duration under monensin plus Cs+ remained unchanged relative to monensin alone , the ( B4 ) duty cycle decreased significantly under the monensin plus Cs+ saline relative to either control or monensin saline . The dashed green lines represent means whereas asterisks ( * ) and hashtags ( # ) represent significance from control and monensin , respectively ( Tukey’s test , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19322 . 00710 . 7554/eLife . 19322 . 008Figure 4—figure supplement 1 . Stimulating the pump with 10 µM monensin in pharmacologically isolated heart interneurons requires h-current to shorten the interburst interval but not to shorten the burst duration . Extracellular traces from left ( blue ) and right ( vermilion ) oscillator heart interneurons [HN ( L ) and HN ( R ) neurons] that were pharmacological isolated as bursters by being treated with ( A1 ) saline that contained 500 µM bicuculline ( Bic ) . ( A2 ) The isolated oscillator heart interneurons were then treated with saline that contained 500 µM bicuculline plus 10 µM monensin . Corresponding scatter plots of ( A3 ) burst duration and ( A4 ) interburst interval . Extracellular traces from another pair of isolated oscillator heart interneurons that were treated with ( B1 ) saline that contained 500 µM bicuculline plus 2 mM Cs+ saline followed by followed by another treatment with ( B2 ) saline that contained 500 µM bicuculline , 2 mM Cs+ , plus 10 µM monensin saline . Corresponding scatter plots of ( B3 ) burst duration and ( B4 ) interburst interval . The dashed green lines represent means whereas asterisks ( * ) and hashtags ( # ) represent significance from control and bicuculline , respectively ( Tukey’s test , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19322 . 008 Because it was difficult to sustain a healthy intracellular recording from an oscillator heart interneuron for more than 40 min , we were not able to determine if the effects of monensin are reversible in our intracellularly-recorded cells beyond 15–20 min of washout with normal saline . To determine if the effects of monensin could be reversed with longer washouts , we recorded extracellularly from both oscillator heart interneurons in normal saline to overcome the difficulties of obtaining long-term recordings and to prevent disorganized firing patterns from being introduced by intracellular electrodes . We treated the oscillator heart interneurons with 10 µM monensin for 10 min , followed by a washout with normal saline for six hours . There was variability across preparations with respect to when the period started to increase , which occurred between 20 . 2 to 76 . 9 min . In preparations that were exposed to monensin for 20–50 min , we found that the effects of 10 µM monensin are not reversible . To determine if the decreased period observed in our combined intracellular and extracellular experiments with monensin ( Figure 3B ) was due to the 10 µM concentration that we used , we performed another set of dual extracellular recordings and treated the oscillator heart interneurons with 1 µM monensin for 30 min followed by the treatment of 10 µM monensin for another 15 min . When we increased the concentration of monensin from 1 to 10 µM , only the burst duration decreased significantly , but only by just 0 . 7 ± 0 . 1 s ( 2 . 4 ± 0 . 1 s for 1 µM monensin vs . 1 . 7 ± 0 . 1 s for 10 µM monensin saline , n = 5 , Tukey’s test , p=0 . 044 ) . Thus , increasing the concentration of monensin in the saline from 1 to 10 µM does not appear to affect the period of neurons that have already been treated with monensin . To determine if the effects of 1 µM monensin occurs more slowly than 10 µM monensin , we performed another set of extracellular recordings with 10 µM monensin and compared the amount of time needed for the period to reach its minimum value at the 200th cycle in both 1 µM and 10 µM monensin ( Figure 3—figure supplement 1A ) . It took significantly longer for the period to reach its 200th cycle value in neurons treated with 1 µM monensin relative to neurons treated with 10 µM monensin ( Figure 3—figure supplement 1B , 18 . 5 ± 0 . 7 min for 1 µM monensin vs . 14 . 5 ± 0 . 6 min for 10 µM monensin saline , n = 5 , unpaired t-test , p=0 . 003 ) . Moreover , the period at the 200th cycle for neurons treated with 10 µM monensin was significantly lower than the period of neurons treated with 1 µM monensin ( Figure 3—figure supplement 1C , 3 . 9 ± 0 . 1 s for 1 µM monensin vs . 2 . 9 ± 0 . 1 s for 10 µM monensin saline , n = 5 , unpaired t-test , p=0 . 00003 ) . Thus , in subsequent experiments , we opted to use 10 µM monensin because its maximal effects can be observed sooner . Moreover , the 10 µM concentration is widely used ( e . g . , Lichtshtein et al . , 1979; Tsuchida and Otomo , 1990; Busciglio et al . , 1993; Itoh et al . , 2000; Lamy and Chatton , 2011; Wang et al . , 2012; Zhang et al . , 2015 ) , allowing us to compare our results to those from other studies . We next sought to determine if stimulation of the Na+/K+ pump by monensin might exert its effect on the period through activation of the h-current . Previous studies in other systems have suggested that the h-current might counteract the hyperpolarizing effects of the pump current ( e . g . , Robert and Jirounek , 1998; Soleng et al . , 2003; Baginskas et al . , 2009 ) . Moreover , Hill et al . ( 2001 ) developed a canonical model of the oscillator heart interneurons and showed that increasing the h-current conductance decreases the period , which was later supported by dynamic clamp experiments ( Sorensen et al . , 2004; Olypher et al . , 2006 ) . Using a hybrid half-center oscillator , Sorensen et al . ( 2004 ) found that increasing the h-current conductance of living or silicon neurons with dynamic clamp decreases the period and interburst interval , without affecting the burst duration . Angstadt and Calabrese ( 1989 ) found that the h-current in leech oscillator heart interneurons can be blocked by 2–4 mM external Cs+ . Thus , if we were to block the h-current with Cs+ while stimulating the pump with monensin , we should observe an increase in the period in oscillator heart interneurons . To determine if the pump current interacts with the h-current to regulate the period and other burst characteristics of oscillator heart interneurons , we performed dual extracellular recordings as previously described and added 10 µM monensin to the saline for 15 min followed by the application of 10 µM monensin plus 2 mM Cs+ to block the h-current for another 15 min ( Figure 4A1-3 ) . We then compared five burst characteristics ( see Figure 3A for definitions of period , burst duration , and interburst interval and Materials and methods for definitions of duty cycle and intraburst spike frequency ) across all three experimental treatments ( Figure 4B1-5 ) . Similar to the previous intracellular and extracellular recordings , we found that monensin decreased the period significantly relative to control saline ( Figure 4A1-2 and B1 , 8 . 2 ± 0 . 7 s for control saline vs . 2 . 7 ± 0 . 1 s for monensin saline , n = 5 , Tukey’s test , p=0 . 0002 ) . Monensin also significantly decreased the burst duration ( Figure 4A1-2 and B2 , 4 . 5 ± 0 . 3 s for control saline vs . 1 . 4 ± 0 . 04 s for monensin saline , n = 5 , Tukey’s test , p=0 . 0002 ) and the interburst interval ( Figure 4A1-2 and B3 , 3 . 7 ± 0 . 3 s for control saline vs . 1 . 2 ± 0 . 1 s for monensin saline , n = 5 , Tukey’s test , p=0 . 0007 ) , resulting in no significant change in the duty cycle ( Figure 4B4 ) . Monensin did not significantly affect the intraburst spike frequency ( Figure 4B5 ) . Taken together , stimulation of the pump with monensin speeds up the bursting activity of half-center oscillators without affecting their duty cycle . When we added Cs+ to the monensin saline , we found that the period increased significantly under monensin plus Cs+ saline relative to the monensin saline ( Figure 4A2-3 and B1 , 2 . 7 ± 0 . 1 s for monensin saline vs . 5 . 5 ± 0 . 2 s for monensin plus Cs+ saline , n = 5 , Tukey’s test , p=0 . 004 ) . The burst duration remained ( Figure 4A2-3 and B2 ) the same whereas the interburst interval ( Figure 4A2-3 and B3 , 1 . 2 ± 0 . 1 s for monensin vs . 4 . 6 ± 0 . 3 s for monensin plus Cs+ , n = 5 , Tukey’s test , p=0 . 0002 ) increased significantly relative to monensin alone . As a result , the duty cycle decreased significantly under the monensin plus Cs+ saline relative to monensin saline ( Figure 4B4 , 53 . 3 ± 0 . 7% for monensin saline vs . 16 . 0 ± 1 . 8% for monensin plus Cs+ saline , n = 5 , Tukey’s test , p=0 . 0002 ) . Finally , the intraburst spike frequency under Cs+ plus monensin saline increased significantly relative to monensin alone ( Figure 4B5 , 15 . 7 ± 0 . 8 Hz for monensin saline vs . 39 . 6 ± 3 . 1 Hz for monensin plus Cs+ saline , n = 5 , Tukey’s test , p=0 . 0002 ) . Thus , stimulating the pump with monensin decreases the burst duration and the interburst interval . However , its effects on the interburst interval is inhibited when the h-current is blocked . We wished to determine if stimulating the pump with monensin can still exert its effect on the burst duration when the h-current has already been blocked , thereby preventing it from affecting the interburst interval and period . We first blocked the h-current with Cs+ ( Figure 4A4 ) and then stimulated the pump with 10 µM monensin ( Figure 4A5 ) . We found that the burst duration decreased significantly relative to Cs+ saline ( 4 . 5 ± 0 . 7 s for Cs+ saline vs . 2 . 6 ± 0 . 4 s for Cs+ plus monensin saline , n = 5 , paired t-test , p=0 . 04 ) . There were no significant changes in the interburst interval ( 4 . 8 ± 0 . 4 s for Cs+ saline vs . 4 . 4 ± 0 . 3 s for Cs+ plus monensin saline , n = 5 ) and period ( 9 . 2 ± 1 . 1 s for Cs+ saline vs . 7 . 0 ± 0 . 5 s for Cs+ plus monensin saline , n = 5 ) relative to Cs+ saline . Thus , when the h-current is blocked , stimulating the pump decreases the burst duration but not the interburst interval . This treatment sequence of Cs+ and monensin did not confound our previous results ( Figure 4A1-3 ) , confirming the critical role of the h-current in shortening the interburst interval when the pump is stimulated by monensin . Similar results were also observed when we stimulated the pump of oscillator heart interneurons that have been isolated with 500 µM bicuculline , which blocks the synaptic transmission between the two oscillator heart interneurons ( Figure 4—figure supplement 1 ) ( Schmidt and Calabrese , 1992; Cymbalyuk et al . , 2002 ) . Although these oscillator heart interneurons form half-center oscillators , they can function as endogenous bursters when recorded extracellularly in bicuculline saline ( Cymbalyuk et al . , 2002 ) . The one notable exception was that stimulating the pump with monensin in isolated oscillator heart interneurons does not significantly decrease the period ( 3 . 9 ± 0 . 1 s for bicuculline saline vs . 3 . 3 ± 0 . 1 s for bicuculline plus monensin saline , n = 5 ) , which can be explained in part by a floor effect; period had already been significantly reduced in bicuculline saline ( 8 . 8 ± 0 . 9 s for control saline vs . 3 . 9 ± 0 . 1 s for bicuculline saline , n = 5 , Tukey’s test , p<0 . 001 ) , to a point where adding monensin could not have a significant effect on shortening period ( Figure 4—figure supplement 1 ) . When the h-current was blocked , the period increased significantly relative to bicuculline saline ( 3 . 5 ± 0 . 1 s for bicuculline saline vs . 7 . 1 ± 0 . 9 s for bicuculline plus Cs+ saline , n = 5 , Tukey’s test , p=0 . 003 ) . Likewise , the interburst interval increased significantly as well relative to bicuculline saline ( 1 . 3 ± 0 . 05 s for bicuculline saline vs . 5 . 8 ± 1 . 0 s for bicuculline plus Cs+ saline , n = 5 , Tukey’s test , p=0 . 0004 ) . The burst duration , however , decreased significantly relative to bicuculline saline ( 2 . 2 ± 0 . 1 s for bicuculline saline vs . 1 . 3 ± 0 . 1 s for bicuculline plus Cs+ saline , n = 5 , Tukey’s test , p=0 . 047 ) . Adding monensin to the bicuculline plus Cs+ saline did not significantly change any of the burst characteristics relative to bicuculline plus Cs+ saline ( Figure 4—figure supplement 1B2-4 ) . Thus , h-current must be present for monensin to decrease period of isolated heart interneurons ( bicuculline ) . Taken together , our results suggest a scenario in which the pump contributes to the normal cycling of the heart interneuron half-center oscillator by providing hyperpolarizing current that shortens the burst duration , and also the interburst interval through activation of the h-current . Stimulating the pump shortens the period by enhancing both of these processes . When the h-current is blocked , we can separate the effects of the pump on interburst interval and on burst duration . Our results are consistent with the findings of Hill et al . ( 2001 ) and Sorensen et al . ( 2004 ) whereby the h-current regulates the interburst interval , which in turn affects the period . In our previous experiments , we found that stimulating the Na+/K+ pump with monensin hyperpolarizes the membrane potential , which activates the h-current to shorten the period . However , Tobin and Calabrese ( 2005 ) found that inhibiting the pump with ouabain or with the neuropeptide myomodulin also shortened the period of these same oscillator heart interneurons just before suppressing spiking activity . We were able to reproduce their results when we performed dual extracellular recordings with a group of oscillator heart interneurons treated with strophanthidin . Within 1 min of adding 100 µM strophanthidin to the normal saline , the period decreased significantly under strophanthidin saline relative to control saline ( Figure 5A1 and A3 , 7 . 8 ± 0 . 2 s for control saline vs 5 . 8 ± 0 . 7 s for strophanthidin saline , n =5 , paired t-test , p=0 . 0467 ) . The duty cycle , however , increased significantly under strophanthidin saline relative to control saline ( Figure 5A1 and A4 , 57 . 7 ± 1 . 7% for control saline vs . 91 . 5 ± 2 . 8% for strophanthidin saline , n=5 , paired t-test , p=0 . 0006 ) . The significant increase in duty cycle was due to the significant decrease in interburst interval under strophanthidin saline relative to control saline ( 3 . 3 ± 0 . 1 s for control vs 0 . 6 ± 0 . 3 s for strophanthidin , n = 5 , paired t-test , p=0 . 001 ) . The burst duration was unchanged . We also observed the suppression of spiking activity by strophanthidin for 2–4 min before washing it out with control saline for at least 12–15 min until spiking activity resumed ( Figure 5A2 ) . Thus , strophanthidin initially speeds up the bursting activity on oscillator heart interneurons followed by a reversible suppression of their spiking activity . 10 . 7554/eLife . 19322 . 009Figure 5 . Inhibition of the Na+-K+ pump by applying strophanthidin speeds up the bursting activity of oscillator heart interneurons . Extracellular traces of bursting activity by left ( blue ) and right ( vermilion ) oscillator heart interneurons [HN ( L ) and HN ( R ) neurons] functioning as a half-center oscillator in ( A1 ) in saline that contained 100 µM strophanthidin . ( A2 ) Strophanthidin transiently decreased the period but eventually suppressed spiking activity , which was reversible once the neurons were again bathed in normal saline . Corresponding scatterplots of the ( A3 ) period and ( A4 ) duty cycle were measured from the extracellular traces of five preparations under ( A1-2 ) two experimental treatments . ( B1 ) Extracellular and intracellular traces of activity in control saline and strophanthidin saline . ( B2 ) A corresponding scatter plot of the base potential in control saline and strophanthidin saline . ( B3 ) Intracellular traces of current and voltage from an oscillator heart interneuron in normal saline . The neuron was injected with 0 . 6 nA for 4 s , which hyperpolarized its membrane potential below −60 mV . ( B4 ) In strophanthidin saline , the same neuron was injected with −1 nA for 4 s to hyperpolarize its membrane potential below −60 mV . The dashed green lines in the scatter plots represent means whereas the asterisks ( * ) represent significance from control ( paired t-test , p<0 . 05 for all tests ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19322 . 009 It has been well-established that inhibition of pump activity by glycosides such as ouabain or strophanthidin can further depolarize a cell ( Kerkut and Thomas , 1965; Carpenter and Alving , 1968; Smith et al . , 1968; Baylor and Nicholls , 1969; Sokolove and Cooke , 1971; Thomas , 1972b; Jansen and Nicholls , 1973; De Weer and Geduldig , 1973; Mat Jais et al . , 1986; Doebler , 2000; Glitsch , 2001; Pulver and Griffith , 2010; Wang et al . , 2012 ) , which is consistent with the inward current produced by strophanthidin in our voltage-clamp experiments . Such depolarization could explain the observed decrease in the period in our experiments ( Figure 5A ) and by Tobin and Calabrese ( 2005 ) . To determine if the decreased period resulted from a membrane potential that has been depolarized by an inhibited pump , we performed dual intracellular and extracellular recordings to measure the base potential of oscillator heart interneurons before and after external application of 100 µM strophanthidin ( Figure 5B1 ) . We found that strophanthidin gradually depolarized the base potential of oscillator heart interneurons ( data not shown ) , which reached an asymptote after 7 min . Compared to normal saline , the base potential under strophanthidin , which was averaged across 7–8 min , was significantly more depolarized ( Figure 5B2 , −42 . 3 ± 1 . 4 mV for control saline vs −32 . 2 ± 1 . 1 mV for strophanthidin saline , n =5 , paired t-test , p=0 . 003 ) . Subtracting the base potential under strophanthidin saline from the base potential under control saline , the resting pump contributed 10 . 1 ± 1 . 6 mV to the membrane potential of oscillator heart interneurons . Thus , inhibiting the pump with strophanthidin does indeed depolarize the membrane potential of oscillator heart interneurons , which decreases the period . To determine if the suppression of spiking activity was due to a depolarization block , we sought to reactivate spiking by injecting a short pulse ( 4 s ) of negative current to hyperpolarize their membrane potential . Under normal saline , a negative current of −0 . 6 to −0 . 8 nA was sufficient to hyperpolarize the cells below −60 mV ( Figure 5B3 ) . In strophanthidin saline , however , more negative current ( −1 . 0 to −2 . 5 nA ) was needed to hyperpolarize the cells beyond −60 mV ( Figure 5B4 ) . Hence , the input resistance was significantly lower in strophanthidin saline relative to control saline ( 114 . 1 ± 4 . 5 MΩ for control saline vs . 27 . 7 ± 6 . 5 MΩ for strophanthidin saline , n=5 , paired t-test , p=0 . 0002 ) . These short hyperpolarizing pulses were sufficient to elicit one or more spikes in all five preparations , leading us to conclude that the suppression of spiking activity in strophanthidin saline was due to a depolarization block , consistent with previous studies ( e . g . , Carpenter and Alving , 1968; Johnson et al . , 1992 ) . Taken together , the decreased period that resulted from inhibiting the pump with strophanthidin was mediated by depolarization , a different mechanism from the one initiated by monensin , which activated the h-current . Another well-established method to inhibit the Na+/K+ pump is to remove external K+ from saline ( Garrahan and Glynn , 1967; Carpenter and Alving 1968; Maetz , 1969; Sokolove and Cooke , 1971; Thomas , 1972a; Gadsby 1980; Bühler et al . , 1991; Lafaire and Schwarz 1986; Catarsi and Brunelli , 1991; Wang et al . , 2012 ) . Because the pump couples the outward exchange of Na+ with the inward exchange of K+ , the removal of external K+ inhibits pump activity . To determine if the effects of strophanthidin on the bursting activity of half-center oscillator could be similarly reproduced with the removal of external K+ , we treated another group of oscillator heart interneurons with K+-free saline . Starting with the normal bursting activity of in control ( 4 mM K+ ) saline ( Figure 6A1 ) , the period decreased significantly under K+-free saline relative to control saline ( Figure 6A1-2 and B1-2 , 9 . 8 ± 0 . 9 s for control saline vs . 6 . 1 ± 0 . 2 s for K+-free saline , n = 5 , paired t-test , p=0 . 01 ) . This effect was observed within 3–7 min of applying the K+-free saline ( Figure 6B1 ) . Similarly , the burst duration ( Figure 6A1-2 , 5 . 4 ± 0 . 5 s for control saline vs . 3 . 5 ± 0 . 1 s for K+-free saline , n=5 , paired t-test , p=0 . 01 ) and interburst interval ( Figure 6A1-2 , 4 . 4 ± 0 . 4 s for control saline vs . 2 . 6 ± 0 . 1 s for K+-free saline , n =5 , paired t-test , p=0 . 01 ) decreased significantly under K+-free saline relative to control saline , resulting in no significant differences in duty cycle in both treatments . The intraburst spike frequency increased significantly under K+-free saline relative to control saline ( 9 . 7 ± 1 . 0 Hz for control saline vs . 13 . 6 ± 1 . 1 Hz for K+-free saline , n=5 , paired t-test , p=0 . 01 ) . Like strophanthidin , long-term exposure ( 16–31 min ) to K+-free saline also suppressed spiking activity ( data not shown ) , which was reversible as both heart interneurons resumed their usual bursting activity after 1–5 min of washing out the K+-free saline with normal saline ( data not shown ) . Thus , the suppressive effects of K+-free saline are also reversible . 10 . 7554/eLife . 19322 . 010Figure 6 . Effects of various concentrations of external K+ on the bursting activity of oscillator heart interneurons . Extracellular traces of bursting activity by left ( blue ) and right ( vermilion ) oscillator heart interneurons [HN ( L ) and HN ( R ) neurons] treated with ( A1 ) control ( 4 mM K+ ) saline , ( A2 ) 0 mM , ( A3 ) 0 . 1 mM , ( A4 ) 0 . 4 mM , and ( A4 ) 2 mM . ( B1 ) A period vs cycle graph of one preparation showing the shortening of the period in K+-free saline over time . Summing up all the periods leading up to the 40th cycle reveals that it takes 6 . 6 min for the effects of K+-free saline to stabilize . Scatterplots of corresponding periods for ( B2 ) 0 mM , ( B3 ) 0 . 1 mM , ( B3 ) 0 . 4 mM , and ( B5 ) 2mM K+ were measured from the extracellular traces of the four groups treated with ( A2-5 ) lower concentrations of external K+ . The dashed green lines represent means whereas the asterisks ( * ) represent significance from control ( paired t-test , p<0 . 05 for all tests ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19322 . 010 To exclude the possibility that the decreased period under K+-free saline was mainly due to a change in the equilibrium potential for K+ , we performed additional extracellular experiments with varying low concentrations ( 0 . 1 , 0 . 4 , and 2 mM ) of external K+ ( Figure 6A3-5 ) . Unlike the period in K+-free saline , the period did not decrease significantly when the oscillator heart interneurons were exposed to any of the three low concentrations of external K+ ( Figure 6B3-5 ) . In fact , the period either remained the same ( 0 . 1 and 0 . 4 mM K+ saline , Figure 6A3-4 and B3-4 ) or increased significantly ( 2 mM K+ saline , Figure 6A5 and B5 , 11 . 4 ± 0 . 5 s for control saline vs . 19 . 6 ± 2 . 4 s for K+-free saline , n=5 , Tukey’s test , p=0 . 005 ) . Only in K+-free saline , did we observe a significantly decreased period . Thus , it is the inhibition of the pump and not the change in the equilibrium potential for K+ that is primarily responsible for the initial increase in bursting activity . To model the bursting activity of the heart interneuron half-center oscillator and to explain mechanistically the results of our monensin experiments , we modified the biophysical model by Hill et al . ( 2001 ) by introducing the Na+/K+ pump current and intracellular Na+ dynamics ( See Equations 4 and 5 in Materials and methods ) . The monensin-mediated diffusion of Na+ across the cell membrane was modeled as a non-electrogenic process controlled by a monensin rate constant ( M ) , which contributed to intracellular Na+ dynamics . To determine the extent to which the bursting activity of our model captures the bursting activity observed in our experiments , we compared the modeled data to the experimental data based on four burst characteristics ( period , burst duration , interburst interval , and duty cycle ) that mimicked our three experimental treatments ( Figure 4A1-3 , Table 1 ) . The first experimental treatment was a half-center oscillator ( control saline , Figures 4A1 and 7A1 ) , which was modeled by including expressions for all voltage-gated and synaptic currents . The second experimental treatment was a half-center oscillator with the pump stimulated by monensin ( monensin saline , Figures 4A2 and 7A2 ) , which was simulated by setting the monensin rate constant to 2 . 2 × 10−3s−1 ( these values have been rounded; see the Appendix for the exact values used in our simulations ) . Finally , the third experimental treatment was a half-center oscillator that has a pump stimulated by monensin and a blocked h-current ( monensin plus Cs+ saline , Figures 4A3 and 7A2 ) , which was simulated by reducing the maximal conductance of the h-current ( g¯h = 0 . 1 nS ) and by setting the monensin rate constant to 1 . 9 × 10−4 s−1 . 10 . 7554/eLife . 19322 . 011Figure 7 . A biophysical model of oscillator heart interneurons that mimics three experimental treatments with monensin . ( A1 ) Sample traces of simulated activity by oscillator heart interneurons functioning as a half center oscillator in normal saline , which were observed when parameters g¯h = 4 . 9 nS and M = 0 s−1 . ( A2 ) Simulated activity of oscillator heart interneurons with a Na+/K+ pump stimulated by monensin ( monensin saline ) , observed when g¯h = 4 . 9 nS and M = 2 . 2 × 10−3 s−1 . ( A3 ) Simulated activity of a half-center oscillator with blocked h-current and pump stimulated by monensin ( monensin plus Cs+ saline ) , observed when g¯h = 0 . 1 nS and M = 1 . 9 × 10−4 s−1 . Sample traces representing membrane potentials ( Vm ) of both left ( L , blue ) and right ( R , vermilion ) oscillator heart interneurons as well as h-current ( Ih , yellow ) , pump current ( Ipump , reddish purple ) , and intracellular Na+ concentration [Na]i ( black ) belonging to the right oscillator heart interneuron . ( B1 ) A scatterplot depicting incremental shortening of the period as the monensin rate constant increases towards 2 . 2 × 10−3 s−1 in a model of a half-center oscillator with the h-current present . Spiking activity was suppressed at rate constant values larger than 2 . 2 × 10−3 s−1 . ( B2 ) In a simulation of a half-center oscillator , the average pump current over the entire burst cycle was fixed at 155 . 5 pA , which resulted in a longer period , burst duration , and interburst interval . Intracellular Na+ concentration [Na]i appeared to increase and decrease more slowly relative to a ( A1 ) normal half-center oscillator with a dynamic pump current . ( B3 ) Fixing the average pump current over the entire burst cycle to 144 . 4 pA ( a value lower than 155 . 5 pA ) produced irregular bouts of bursting . DOI: http://dx . doi . org/10 . 7554/eLife . 19322 . 01110 . 7554/eLife . 19322 . 012Table 1 . Modeled burst characteristics ( period , burst duration , interburst interval , and duty cycle ) were compared to experimental ranges under three parameter regimes that mimicked our experimental treatments . The control treatment was a normal half-center oscillator . The monensin treatment was a half-center oscillator with a pump stimulated by monensin ( M = 2 . 2 × 10−3 s−1 ) . The monensin plus Cs+ treatment was a half-center oscillator with blocked h-current and a pump stimulated by monensin ( M = 1 . 9 × 10−4 s−1 ) . Asterisks denote out-of-range values . DOI: http://dx . doi . org/10 . 7554/eLife . 19322 . 012Comparison of modelled burst characteristics to experimental ranges under four parameter regimes . TreatmentsPeriod ( s ) Burst duration ( s ) Interburst interval ( s ) Duty cycle ( % ) ModelExperimental rangesModelExperimental rangesModelExperimental rangesModelExperimental rangesControl8 . 04 . 3-12 . 33 . 72 . 6-7 . 34 . 21 . 5-6 . 047 . 046 . 8-64 . 7Monensin4 . 4*2 . 5-4 . 11 . 61 . 3-2 . 32 . 8*1 . 1-1 . 835 . 5*44 . 6-59 . 8Monensin , Cs+6 . 65 . 9-8 . 62 . 01 . 7-4 . 14 . 53 . 3-5 . 530 . 926 . 6-48 . 0 The monensin rate constants in the above two modeled experimental treatments were chosen to provide the best fit to our experimental data . Both rate constants can be found close to the boundaries of their respective models ( e . g . , 2 . 2 × 10−3 s−1 for a half-center oscillator in Figure 7B1 ) and differed from each other by an order of magnitude ( 10−3 s−1 vs . 10−3 s−1 ) because the models themselves had different sensitivities to the monensin rate constant , with the half-center oscillator with a blocked h-current being more sensitive . Increasing the monensin rate constant towards its boundary value generally decreases the period , and spiking activity was suppressed when the rate constants were at values greater than its boundary value . To determine if our simulations captured the trends observed in our experiments , we compared the burst characteristics of our model to the ones from our experiments . We used the control data of all extracellular experiments involving monensin . When we simulated the experimental treatments in our model , all the burst characteristics were within range except for the period , duty cycle , and interburst interval from our model of a half-center oscillator with a stimulated pump ( Table 1 ) . Thus , nine of the twelve measures ( four burst characteristics multiplied by three experimental treatments ) were within range of the experimentally observed characteristics . In cases where the burst characteristics were out of range , the simulations still qualitatively captured the trends observed in our experiments whereby stimulation of the pump by monensin speeds up the bursting activity of oscillator heart interneurons when the h-current is present . In our experiments , stimulating the Na+/K+ pump with monensin will only decrease the period of half-center oscillators if the h-current was present ( Figure 4A1-2 and B1 ) , revealing an interaction between the pump current and the h-current to regulate bursting . To explain this interaction further using our model , we analyzed changes in the pump current and the h-current during each burst and interburst interval . We computed average values for the pump current over a burst and over an interburst interval . We only computed the average value for the h-current over an interburst interval because the h-current activates only when hyperpolarized but deactivates when depolarized . Average values for both currents were obtained by computing the integral of each current over a closed interval and then dividing that integral by the duration of the closed interval . When averaging these currents , we only considered the segment of the interburst interval from the time after the last action potential in a burst in which the membrane potential crossed −50 mV to the time before the first action potential of the next burst whereby the membrane potential crossed −50 mV again . This segment was chosen to avoid having substantial parts of each current originating from the end of a plateau of a previous burst and the beginning of a plateau of the next burst . When we set the monensin rate constant to 2 . 2 × 10−3 s−1 to simulate the monensin treatment in our half-center oscillator model , each burst duration decreased from 3 . 7 s to 1 . 6 s ( Table 1 , Figure 7A1-2 ) , which we attributed to the average pump current during each burst , which increased from 183 . 6 pA to 221 . 6 pA . The increased pump current itself can be attributed to the average increase in intracellular Na+ concentration during each burst from 14 . 4 to 14 . 6 mM . Similarly , adding monensin also decreased each interburst interval from 4 . 2 s to 2 . 8 s ( Table 1 , Figure 7A1-2 ) but increased the average intracellular concentration during each interburst interval from 14 . 1 to 14 . 4 mM , which in turn increased the average pump current during each interburst interval from 108 . 3 pA to 166 . 9 pA . As a result , the average membrane potential during each interburst interval hyperpolarized from −60 . 6 mV to −62 . 2 mV , which increased the average magnitude of the h-current during each interburst interval from −69 . 9 pA to −82 . 6 pA . When we reduced the monensin rate constant to 1 . 9 × 10−4 s−1 and the maximal conductance of the h-current to 0 . 1 nS , each burst duration increased from 1 . 6 s to 2 . 0 s and each interburst interval increased from 2 . 8 s to 4 . 5 s ( Table 1 , Figure 7A2-3 ) . Blocking the h-current in our monensin simulation also decreased the average pump current during each burst from 221 . 6 pA to 153 . 0 pA and during each interburst interval from 166 . 9 pA to 92 . 1 pA . The decreased pump current was due to the average decrease in intracellular Na+ concentration from 14 . 6 mM to 14 . 3 mM during each burst and from 14 . 4 mM to 14 . 0 mM during each interburst interval . Taken together , the application of monensin in our simulations decreased the burst duration by increasing the average pump current during each burst . Moreover , monensin also increased the average pump current during each interburst interval , which hyperpolarized the membrane potential and activated the h-current during each interburst interval , which in turn decreases the interburst interval itself . Our simulations also confirmed our experimental results whereby stimulating the pump while blocking the h-current in half-center oscillators shortens the burst duration but increases the interburst interval , which has the effect of decreasing the duty cycle . These simulations confirmed our experimental results that a stimulated pump interacts with the h-current to decrease the period that results from the combined decrease of the burst duration and interburst interval . To determine the dynamic contributions of the Na+/K+ pump current and intracellular Na+ concentrations to the bursting activity of our modeled half-center oscillator , we examined the changes of the pump current and intracellular Na+ concentration during specific time points of each burst and interburst interval . The pump current at the end of each interburst interval had a minimum ( or baseline ) value of about 98 . 0 pA ( Figure 7A1 ) . During the beginning of each burst , intracellular Na+ concentration accumulated with each action potential ( Figure 7A1 ) . As the intracellular Na+ concentration increased , the pump current also increased . Over the course of a burst , the spike frequency decreased and the outward Na+ flux generated by the pump began to balance—at a maximum pump current of 203 . 6 pA —and exceeded the inward Na+ flux produced by the voltage-gated Na+ currents ( Figure 7A1 ) . At the termination of each burst , the intracellular Na+ concentration and pump current relaxed to a baseline of about 14 mM and 100 pA , respectively , over the duration of an interburst interval ( Figure 7A1 ) . The outward pump current was active over the entire cycle of bursting activity , but it was stronger during each burst . These effects are much like the ones we would expect from a leak current with a very negative equilibrium potential: an outward current that seems to track the membrane potential exerting a hyperpolarizing drag on both the burst duration and the interburst interval . The pump current actually tracks intracellular Na+ dynamics , which is delayed ( low-pass filtered ) with respect to the membrane potential and this delay might exert sizable effects in some cases . To determine if a dynamic pump current is essential to the normal bursting activity of oscillator heart interneurons , we performed simulations whereby the average pump current was fixed at a constant value of 155 . 5 pA over the entire burst cycle ( Figure 7B2 ) . Compared to the dynamic pump current , the value of this constant pump current was lower during each burst but higher during each interburst interval . This constant value was chosen because lower values of the pump current ( e . g . , 144 . 4 pA ) produced irregular bouts of bursting ( Figure 7B3 ) . When the pump current was fixed to 155 . 5 pA , the period , burst duration and interburst interval were 19 . 9 s , 9 . 6 s , and 10 . 2 s , respectively ( Figure 7B2 ) . The durations of these burst characteristics were longer than those observed in simulations with a dynamic pump current ( Figure 7A1 and Table 1 ) . The relatively lower constant pump current during each burst increased the burst duration . Because the period in the half center oscillator is determined in part by the burst duration , the period itself became longer relative to simulations with a dynamic pump . Taken together , we concluded that in a half-center oscillator , a dynamic pump current that changes from an interburst interval to the next burst is critical for the formation of burst patterns with normal burst characteristics . Such a conclusion is also consistent with the observed non-spiking activity that result from inhibiting the pump with strophanthidin or K+-free saline in our experiments .
In the present study , we examined the role of the Na+/K+ pump in regulating the bursting activity of oscillator heart interneurons , which pace the leech heartbeat central pattern generator . In doing so , we performed experiments that allowed us to determine the effects of a stimulated pump on the burst characteristics of these oscillator heart interneurons ( Figures 3–4 ) . Because the rate of pump activity is dependent on intracellular Na+ concentrations ( Thomas , 1972a; Mogul et al . , 1990 ) , we were able to stimulate the pump of the oscillator heart interneurons by increasing the intracellular loading of Na+ with an electrode ( Figure 1A2 ) or by applying monensin ( Figures 1B3 and 2A2 ) , a Na+/H+ antiporter ( Hill and Licis , 1982 ) . Monensin has a hyperpolarizing effect when applied to excitable cells such as neurons and muscle cells ( Lichtshtein et al . , 1979; Tsuchida and Otomo 1990 , Satoh and Tsuchida , 1999; Doebler , 2000; Wang et al . , 2012 ) . Our stimulation of the pump with monensin did indeed hyperpolarize the membrane potential of the oscillator heart interneurons when the h-current and Ca2+ currents were blocked ( Figure 1B3 ) . When we voltage-clamped oscillator heart interneurons under the same conditions , we observed that monensin produced a long-lasting outward current , indicating that the outward pump current was enhanced by monensin ( Figure 2A1-2 ) . This outward current is consistent with the results of Tsuchida and Otomo ( 1990 ) , who also showed monensin enhancing the outward current generated by Purkinje fibers that were isolated from the hearts of dogs and goats . Moreover , the monensin-enhanced outward pump current can be blocked by strophanthidin , which is consistent with previous studies that found glycosides such as ouabain and strophanthidin blocked the hyperpolarizing effects of monensin ( Lichtshtein et al . , 1979; Doebler , 2000; Wang et al . , 2012 ) . Since we did not see any corresponding change in the membrane conductance that was assessed from the linear slope of each current-voltage curve ( Figure 2B2 ) , it is unlikely that the observed outward shift in the membrane current is due to other voltage-gated or leak currents that are intrinsic to these oscillator heart interneurons . Our voltage-clamp experiments were performed in Ca2+-free saline that contained Mn2+ , which would have prevented the activation of Ca2+-activated K+ current . Even if the Ca2+-activated K+ current was present , it is a fast transient current in these neurons , which could not sustain a long-term outward current ( Simon et al . , 1992 ) . The combined use of strophanthidin and monensin in our voltage-clamp experiments allowed us to determine the fixed range ( 202–289 pA ) of the maximum pump current by taking the difference between the outward current produced by monensin and the inward current produced by monensin plus strophanthidin . At rest , the pump current varies between 82 pA to 283 pA , a range that is consistent with outward currents attributed to the pump at rest in other voltage-clamp experiments ( e . g . , Sakai et al . , 1996; Mogul et al . , 1990 ) . Because the maximum pump current is usually not generated under resting conditions , there is a certain amount of pump current left available that can be stimulated by monensin . These results indicate that the action of monensin in oscillator heart interneurons is consistent with its well-established role of facilitating an electroneutral exchange of intracellular H+ for extracellular Na+ , thereby increasing intracellular Na+ , which in turn stimulates the pump to generate more outward current that hyperpolarizes the membrane potential ( Lichtshtein et al . 1979 ) . A possible side effect of the continuous extrusion of H+ by monensin is a modest change in intracellular pH ( Smallridge et al . , 1992; Itoh et al . , 2000 ) . For example , at the same concentration of 10 µM used in our experiments , monensin has been found to increase the intracellular pH of astroglial cells by 0 . 07 ( from 7 . 26 to 7 . 33 ) ( Itoh et al , 2000 ) . Intracellular alkalization has been reported to prolong the bursting activity of ventral white cells in snails , increase the number of spikelets in hippocampal neurons of rodents , decrease the viability of endothelial cells , and enhance glycolysis in neurons and glioma cells ( Gillette , 1983; Valiante et al . , 1995; Cutaia et al . , 2000 ) . In our experiments , we did not observe any prolongation of bursting activity ( Figure 4B1 ) or an increase in intraburst spike frequency ( Figure 4B5 ) . Moreover , the monensin-treated oscillator heart interneurons do not appear to be adversely affected by any potential intracellular alkalization as they were able to burst continuously for well over thirty minutes to an hour ( data not shown ) . Nevertheless , if monensin were to enhance glycolytic functions , thereby increasing ATP production , it would only serve to enhance pump activity ( Lombardo et al . , 2004 ) , which would be consistent with the use of monensin in this study . Thus , even if modest intracellular pH changes by monensin occurred , we do not expect these changes to confound our results or to affect our interpretations of them . A notable effect of stimulating the pump of oscillator heart interneurons with monensin was the significant decrease in their period ( Figures 4A1 and B1 ) . This acceleration was unexpected because a stimulated or reactivated Na+/K+ pump has been consistently demonstrated in oscillator heart interneurons ( Figures 1–2 ) and in other tissues and systems ( e . g . , De Weer and Geduldig , 1973; Eisner and Lederer , 1980; Johnson et al . , 1992 ) to cause hyperpolarization . The paradox of stimulating the pump to speed up bursting activity can be explained if we consider other intrinsic ionic currents that are also present and are likely to be activated or deinactivated by a membrane potential hyperpolarized by an outward pump current . The h-current is an obvious candidate as it is an inward current with a voltage-threshold near -50 mV and becomes fully activated at −70 to −80 mV ( Angstadt and Calabrese , 1989 ) . The threshold of the h-current is well within the range of membrane potential produced by monensin . The role of the h-current in controlling the rate of depolarization after hyperpolarization has been demonstrated experimentally in neurons and cardiac pacemakers that oscillate continuously ( Soltesz et al . , 1991; DiFrancesco 1993; Bal and McCormick , 1997; Zhang et al . , 2003 ) . Moreover , based on the biophysical model of oscillator heart interneurons by Hill et al . ( 2001 ) , increasing the h-current from its canonical value produces a substantial decrease in period . Their findings were later confirmed by Sorensen et al . ( 2004 ) , who found that increasing the h-current conductance of living or silicon neurons with dynamic clamp decreases the period and interburst interval , without affecting the burst duration . Thus , activation of the h-current can explain the non-intuitive effects of monensin in shortening the period of central pattern generator neurons . The possible interaction between the pump current and the h-current has been suggested by previous studies on slow afterhyperpolarizations ( e . g . , Robert and Jirounek , 1998; Soleng et al . , 2003; Baginskas et al . , 2009 ) . For example , when Robert and Jirounek ( 1998 ) recorded the slow afterhyperpolarization ( described by the authors as a slow posttetanic hyperpolarization ) , which occurs after a burst of compound action potentials , in the rabbit vagus nerve , they found that blocking the h-current or the inward rectifier K+ with Cs+ and Ba2+ , respectively , would increase the amplitude of the slow afterhyperpolarization . Because the slow afterhyperpolarization has been documented to be a product of the pump current , Robert and Jirounek ( 1998 ) hypothesized that the h-current and the inward rectifier K+ current may serve to counterbalance the effects of the pump current under normal conditions . Such a counterbalancing role by the h-current could explain the previously mentioned paradoxical shortening of the period when we stimulated the pump in the heart interneurons . We were able to confirm this role of the h-current in the shortening of the period by monensin when we blocked the h-current with Cs+ , which prevented the period from shortening ( Figure 4B1 ) . Similar results were also observed when the oscillator heart interneurons functioned as individual bursters when pharmacologically isolated with bicuculline ( Figure 4—figure supplement 1 ) . Thus , our experimental results support our hypothesis that stimulation of the Na+/K+ pump in oscillator heart interneurons activates the h-current , which in turns speeds up rhythmic bursting in half-center oscillators and isolated interneurons . When we blocked the h-current with Cs+ while stimulating the pump with monensin in our extracellular experiments , we found that the interburst interval increased significantly , consistent with the established role of the h-current in regulating the interburst interval of oscillator heart interneurons ( Hill et al . , 2001; Sorensen et al . , 2004 ) . The burst duration , however , remained the same when the h-current was blocked . Thus , the h-current regulates the period by changing the interburst interval , with a shorter period resulting from a decreased interburst interval . Because the burst duration and interburst interval decreased proportionally in monensin saline relative to control saline ( Figure 4B2-3 ) , the duty cycle in monensin saline remained similar to the one under control saline ( Figure 4B4 ) . It was only when the h-current was blocked with Cs+ that we observed a decrease in the duty cycle that resulted an increased interburst interval ( Figure 4B3 ) . Conversely , Zhang et al . ( 2003 ) found that overexpression of the h-current in pyloric dilator neurons of the stomatogastric ganglion increased the duty cycle of these neurons . Thus , the pump current regulates the burst duration whereas the h-current regulates the interburst interval and changes to one or both of these two burst characteristics affect the period and duty cycle of oscillator heart interneurons . In summary , we posit a sequence of events by which stimulation of the pump current with monensin activates the h-current to speed up the bursting activity in a half-center oscillator . First , monensin increases the intracellular loading of Na+ as a result of its electroneutral exchange of H+ for Na+ across the cell membrane . Second , the increased intracellular loading of Na+ stimulates the pump to generate more outward current . Third , the enhanced outward current hyperpolarizes the membrane potential of the neuron , which activates the h-current . Finally , activation of the h-current depolarizes the membrane potential , which promotes faster recovery from inhibition and speeds up the bursting activity of half-center oscillators . One seeming paradox of manipulating the Na+/K+ pump in oscillator heart interneurons current is the increase in bursting activity , regardless of whether the pump has been stimulated or inhibited . Tobin and Calabrese ( 2005 ) first observed that inhibiting the pump with the neuropeptide myomodulin or with the glycoside ouabain shortened the period of oscillator heart interneurons . In contrast to monensin , myomodulin produced an inward shift in the ramp current of oscillator heart interneurons when the h-current and Ca2+ currents were blocked ( Tobin and Calabrese , 2005 ) . When ouabain was present , myomodulin could not produce further inwards shifts in the ramp current , indicating that myomodulin inhibits the pump like ouabain . The results from our strophanthidin experiments are consistent with their observations as we also observed a decrease in the period followed by the suppression of spiking activity in oscillator heart interneurons treated with strophanthidin . Furthermore , we confirmed that the suppression of spiking activity was due to a depolarization block , consistent with previous studies ( e . g . , Rhoades and Gross , 1994; Krey et al . , 2010 ) . Unlike ouabain , the effects of strophanthidin are reversible in leech neurons ( Baylor and Nicholls , 1969 ) , and spiking resumed once the neurons were washed out with normal saline . We were able to reproduce the results of our strophanthidin experiments using K+-free saline , whereby the period initially decreases followed by the suppression of spiking . In contrast , the periods of neurons bathed in lower concentrations of external K+ either increased or remained the same . Thus , it appears that the decreased period in K+-free saline was due to the depolarized membrane potential that resulted from an inhibited pump , whereas the increased or unchanged period observed in low K+ ( 0 . 1–2 mM ) saline was due to a more negative equilibrium potential for K+ . Such a conclusion is consistent with the findings of Carpenter and Alving ( 1968 ) , who found that the resting membrane potential of R2 neurons in Aplysia californica was more depolarized when bathed in K+-free seawater at 20°C . But when they lowered the external concentration of K+ to just 1 mM , the R2 neurons hyperpolarized instead . In summary , the paradox of increased bursting activity following the inhibition or stimulation of the Na+/K+ pump in oscillator heart interneurons can be explained by two different mechanisms initiated by pump stimulation and by pump inhibition . Pump stimulation hyperpolarizes the membrane potential , which activates the h-current to shorten the period whereas pump inhibition depolarizes the membrane potential , increasing the likelihood of spiking activity , thereby also shortening the period . Thus , the activation of these two separate pathways ultimately result in the same decrease in period . It is also interesting to note that myomodulin not only inhibits the pump current but increases the h-current thus ensuring that there is ample h-current during each burst cycle despite the depolarization caused by reduction of the pump current ( Tobin and Calabrese , 2005 ) . Our biophysical model , which includes the Na+/K+ pump and intracellular Na+ dynamics , captured and explained the changes in each of the four burst characteristics ( period , burst duration , interburst interval , and duty cycle ) within each of the three experimental treatments . Our model also confirmed our experimental results showing that increasing intracellular Na+ concentrations enhances the hyperpolarizing pump current , which interacts with the h-current to shorten the interburst interval ( Figures 7A2 ) . Moreover , the model suggests that when a half-center oscillator model is treated with monensin , the pump current contributes to burst termination thereby influencing burst duration ( Figure 7A2 ) . Overall , our model quantitatively reproduced nine of the twelve measures ( four burst characteristics multiplied by three experimental treatments ) . Although the three modeled burst characteristics did not fall within the experimentally observed range , the model still faithfully reproduced the observed trends of the experimental data . Our model also shows that a dynamic pump current is essential to normal bursting activity . When we set the average pump current to be constant over the entire burst cycle , the period of the oscillator heart interneuron simulations became longer , as each burst duration and interburst interval increases . Thus , our model reveals that importance of a pump current that is dynamic and not static for there to be normal bursting activity in the oscillator heart interneurons . Furthermore , stimulating the pump current with monensin offsets the amount pump current being generated , thereby influencing the entire burst cycle . The ability of our model to account for multiple experimental treatments is notable because models that do so are generally expected to better predict dynamics under a wide range of other treatment conditions . These other treatment conditions may include modulation or removal of subsets of currents that elicit characteristic oscillatory behaviors such as seizure-like activities or slow subthreshold oscillations ( Grillner et al . , 1995; Cymbalyuk and Calabrese , 2000 , 2001; Krishnan and Bazhenov , 2011; Jasinski et al . , 2013; Barnett et al . , 2013; Bicanski et al . , 2013; Doloc-Mihu and Calabrese , 2014; Krishnan et al . , 2015 ) . Moreover , the greater sensitivity of the model without h-current to the monensin rate constant suggests that the h-current plays a protective role to prevent disruption of functional activity in oscillator heart interneurons , consistent with previous observations ( Cymbalyuk et al . , 2002; Marin et al . , 2013 ) . The action of the Na+/K+ pump current in bursting neurons and networks is very similar to the action of a leak current . The leak current is predominantly an outward current and if its reversal potential is sufficiently hyperpolarized , it is an outward current over the entire burst cycle . Cymbalyuk et al . ( 2002 ) found that the conductance and reversal potential of the leak current controlled properties of bursting activity in isolated oscillator heart interneurons . For example , when the leak conductance increases , the duty cycle and burst duration decrease ( Cymbalyuk et al . , 2002 ) . The pump current is always an outward current that increases , along with the leak current , during each burst . In simulations of a half-center oscillator , over the course of the burst this outward current creates a drag that promotes burst termination as the underlying inward currents wane . An increase in an outward current , such as a leak current or a pump current , or increase of a constant outward current , could lead to earlier burst termination . In contrast , the interburst interval of a half-center oscillator is mainly determined by the interaction of the h-current with the pump current but also by the burst duration of the inhibiting cell . Adding monensin in our model slightly increases the intracellular Na+ concentration throughout the burst cycle . By increasing the baseline intracellular Na+ concentration and the resulting pump current in both phases of each burst cycle , we achieved an effect similar to introducing an outward leak current . The time course of activation of the pump current differentiates it from a leak current . Because the leak current is an instantaneous function of membrane voltage , its waveform tracks the membrane potential exactly . The pump current activates incrementally as action potentials contribute to the intracellular Na+ concentration . As the pump current gradually builds , it can aid in the termination of each burst . After each burst , the intracellular Na+ concentration takes time to decay to its baseline between bursts , resulting in a strong outward pump current being present immediately after each burst , which hyperpolarizes the membrane potential and activates the h-current during each interburst interval . The Na+/K+ pump is principally credited with maintaining the intracellular concentrations of Na+ and K+ ( Ritchie , 1971; Clausen , 2005 ) . Although the pump generates an outward current , its contribution to the resting membrane potential is often too small to be detectable ( Thomas , 1972a ) . Only recently has the potential role of the pump current in regulating motor activity been explored , and there is growing experimental evidence that the pump current can contribute to the motor activity ( e . g . , Ballerini et al . , 1997; Tobin and Calabrese , 2005; Krey et al . , 2010; Pulver and Griffith , 2010; Zhang and Sillar , 2012; Zhang et al . , 2015 ) . In Drosophila larval motor neurons ( Pulver and Griffith , 2010 ) and in spinal central pattern generator neurons of Xenopus tadpoles ( Zhang and Sillar , 2012 ) , the pump current generates slow afterhyperpolarizations , which are long-lasting and are thought to be involved in short-term motor memory through the integration of spikes . The contribution of the pump current to motor activity may also manifest itself through complex dynamical interactions with other ionic currents ( Pulver and Griffith , 2010; Zhang et al . , 2015 ) . Pulver and Griffith ( 2010 ) observed that the pump-mediated slow hyperpolarization in Drosophila larval motor neurons releases the A-current from inactivation , thereby modifying the neuron’s response to the next depolarizing input by delaying the initiation of the spike . This delay could be abolished by ouabain or K+-free saline , as shown by Zhang et al . ( 2015 ) using Xenopus . Unlike the A-current in Drosophila or Xenopus , the A-current in leech heart interneurons has short activation and inactivation kinetics , in the range of milliseconds , and is therefore unlikely to sustain an effect on the period , which lasts for seconds . Instead , our results point to an interaction between the pump current and the h-current to regulate ongoing bursting activity of central pattern generator neurons . Such regulation has strong behavioral implications as changes in burst characteristics underlie the production of various rhythmic behaviors ( e . g . , walking vs . running or breathing fast vs . breathing slow ) ( Hooper , 1998 ) . Nevertheless , for the pump current to regulate bursting activity , it has to be dynamic and stimulating the pump currents offsets the amount of pump current generated . Thus , the dynamics of the pump current , along with the dynamics of the h-current , can help shape the bursting activity of neurons and networks that control motor output .
Medicinal leeches , Hirudo spp . Linneaus 1758 , weighing between 1–1 . 5 g each , were obtained from Niagara Leeches ( Cheyenne , WY ) and Leeches USA ( Westbury , NY ) and were maintained in artificial pond water [0 . 05% ( w/v ) Instant Ocean sea salt ( Spectrum Brands Inc . , Madison , WI ) diluted in reverse osmosis water] at 16°C . Prior to dissection , each animal was anesthetized in a bed of crushed ice and later immersed in a dissecting dish filled with cold leech saline , which contained ( in mM ) 115 NaCl , 4 KCl , 1 . 7 CaCl2 , 10 D-glucose , and 10 HEPES; pH adjusted to 7 . 4 with 1 M with NaOH . The animal was then pinned dorsal side up and an incision was made through its dorsal body wall to expose its internal organs . Individual ganglia at segments 3 and 4 were removed and pinned , ventral side up , in individual sylgard-coated 35 × 10 mm petri dishes with pins made from 0 . 05 mm tungsten wire ( California Fine Wire Company , Grover Beach , CA ) . Each sylgard-coated dish held 1 . 5 mL of solution and served as a recording chamber in all experiments . Once isolated , the ventral side of each ganglion was then desheathed and continuously superfused with saline at a flow rate of 3 mL/min at room temperature . Two procedures were used to record the activity of both left and right oscillator heart interneurons from isolated ganglia . In the first procedure , both heart interneurons were recorded simultaneously using two extracellular electrodes . These electrodes were fabricated from microfilament-containing thin-wall capillary glass ( o . d . 1 mm , i . d . 0 . 75 mm; A-M Systems , Sequim , WA ) pulled on a Sutter P-97 puller ( Sutter Instrument Company , Novato , CA ) to have a tip diameter of ~15–20 µm . Each electrode was positioned over the cell body of a neuron and a gentle suction was applied until the cell body was inside the electrode . The extracellular signals were amplified using a differential AC Amplifier Model 1700 ( A-M Systems , Sequim , WA ) and a Brownlee Precision Model 410 amplifier ( AutoMate Scientific , Berkeley , California ) . Oscillator heart interneurons were identified based on the pattern of their bursting activities and the position of their cell bodies within the ganglion . In the second procedure , combined intracellular and extracellular recording techniques were used to monitor the activity of both heart interneurons ( Tobin and Calabrese , 2005 ) . Sharp intracellular electrodes were fabricated from the same glass as the extracellular electrodes and had a resistance of 20–29 MΩ when filled with 2 M C2H3KO2 ( KAcetate ) and 20 mM KCl . Air bubbles were removed from the tip of each electrode using negative pressure ( Kueh and Jellies , 2012 ) . In experiments designed to introduce intracellular leakage of Na+ from an intracellular electrode , the intracellular solution was substituted with 2 M C2H3NaO2 ( NaAcetate ) and 20 mM NaCl , which gave the intracellular electrodes a resistance of 30–40 MΩ . All intracellular traces were acquired at a sample rate of ≥4 KHz using an Axoclamp 2A amplifier ( Molecular Devices , Sunnyvale , CA ) in discontinuous current clamp mode or in discontinuous single electrode voltage-clamp mode . During the voltage-clamp experiments , the neuron was voltage-clamped at a holding potential of −45 mV , with a gain set to 0 . 8 nA/mV , the time constant to 20 ms , and the anti-alias filter to 5 µs . The output bandwidth for intracellular signals was set to 0 . 3 Khz . For intracellular traces to be accepted into analysis , the input resistance of a neuron that was hyperpolarized with a −0 . 1 nA pulse had to be ≥60 MΩ and the bath potential at the end of an experiment had to be within ± 5 mV . Both intracellular and extracellular signals were digitized using an Axon Digidata 1440A digitizer and recorded with the Clampex 10 . 4 software ( Molecular Devices , Sunnyvale , CA ) . The principal approach used to stimulate the Na+/K+ pump was to treat the oscillator heart interneurons with 10 µM monensin . Monensin sodium salt ( Sigma-Aldrich , St . Louis , MO ) was initially diluted in 100% ethanol to prepare a 50 mM stock solution , which was then stored at −20°C . The stock solution was further diluted to 10 µM in saline . To inhibit the Na+/K+ pump , we would either use a saline that contained 100 µM strophanthidin ( Sigma-Aldrich , St . Louis , MO ) and that contained 0 mM of K+ . Saline solutions with lower concentrations ( 0 . 1 , 0 . 4 , and 2 mM ) of K+ were also used . To control for the effects of ethanol in monensin saline ( 0 . 0193% ethanol ) or strophanthidin saline ( 0 . 217% ethanol ) , the control saline was supplemented with the same concentration of ethanol . To block Ca2+ and synaptic currents , we replaced the Ca2+ in the saline with 1 . 8 mM Mn2+ . To block the h-current , we added 2 mM Cs+ to the saline solution ( Angstadt and Calabrese , 1989 ) . To record oscillator heart interneurons as isolated bursters , the ganglia were bathed in normal saline containing 500 µM bicuculline methiodide to block synaptic transmission ( Sigma-Aldrich ) ( Schmidt and Calabrese , 1992; Cymbalyuk et al . , 2002 ) . All solutions were superfused at a rate of 3 mL/min and it took approximately 20 s to replace the entire bathing solution . To analyze the burst characteristics of the oscillator heart interneurons , we used custom MATLAB scripts to identify and measure individual bursts based on the methods of Masino and Calabrese ( 2002 ) . A burst was defined as a group of five or more action potentials , with the burst duration defined as the time between the first and last action potential of each burst . An action potential was detected whenever a voltage change exceeds an inner threshold ( approximately 50% of the largest spike amplitude ) but not an outer threshold . A refractory period of 20 ms was imposed immediately after the first detection of an action potential to prevent repeated detection of the same action potential . To discriminate between bursts , an interburst interval of 800 ms between each consecutive burst had to elapse before detection of the first action potential from the next burst could occur . Whenever appropriate , these settings were adjusted to ensure that the sample recordings were analyzable by our MATLAB scripts . We required a minimum of ten consecutive bursts that could be detected for a sample recording to be accepted into analysis . Based on these detected bursts , we were able to analyze five burst characteristics: period , burst duration , interburst interval , duty cycle , and intraburst spike frequency across different experimental treatments ( e . g . , control vs . monensin ) . Period was defined as the time from the middle action potential of one burst to the middle action potential of the next burst ( Kristan and Calabrese , 1976 ) . The middle action potential was chosen because it has more physiological relevance ( Wenning et al . , 2014 ) and there is less variability in measuring the period based on the middle action potential rather than the first action potential . The burst duration was defined as the time from the first action potential to the last action potential of a burst , and the interburst interval was calculated by subtracting the burst duration from the period . The duty cycle was defined as follows: ( 1 ) D=BDT×100% In Equation 1 , D is the duty cycle , BD is the burst duration , and T is the period . Because each preparation contributed ≥10 bursts for each experimental treatment , an average was taken for each of the five burst characteristics , resulting in each preparation contributing only one value for each experimental treatment . Moreover , because we did not find any statistically significant differences between the two oscillator heart interneurons of a pair with respect to their burst characteristics , comparisons of burst characteristics in different saline solutions were based on average values taken from both neurons in each ganglion . In addition to the burst characteristics , we measured the membrane current , membrane conductance , and base potential in experiments involving intracellular recordings . The base potential of an intracellularly recorded oscillator heart interneuron was calculated as follows: ( 2 ) tBP=tUP+ ( tT − tUP ) 2 In Equation 2 , tBP is the time point of the base potential , which was defined as midway between the time point of an undershoot phase ( tUP ) and the time point of the next threshold ( tT ) ( Olsen and Calabrese , 1996 ) . The undershoot phase was identified based on the time point of the undershoot trough of an action potential . The threshold was identified by taking the third derivative of an action potential and identifying the time point of the first positive peak of that third derivative ( Henze and Buzsáki , 2001 ) . In cases where spiking activity had been suppressed for a minute or more , the base potential would be the resting membrane potential at any given time point during that period of suppressed spiking activity . Measurements of specific time points , membrane potentials , and currents were performed using Clampfit 10 ( Molecular Devices , Sunnyvale , CA ) and LabChart Reader 8 ( ADInstruments , Colorado Springs , CO ) . The experimental data from within-group experiments ( e . g . , periods from the same preparations in two or more saline solutions ) were analyzed using a paired t-test or a repeated-measures ANOVA , depending on the number of experimental treatments being compared . In between-groups experiments ( e . g . , 1 vs 10 µM monensin ) , an unpaired t-test or an ANOVA was used instead . Finally , in the Na+ and K+ experiment , which has time as the within-group factor and electrode cation ( Na+ and K+ ) as the between-group factor , a split-plot ANOVA was used to analyze changes in base potential from both groups over time . Whenever an ANOVA revealed a statistical significance , a Tukey’s range test or a Holm-Šídák test was used to identify specific sample means that were significantly different . Statistical significance was defined as p<0 . 05 for all tests . All statistical analyses were done using SigmaPlot 12 ( Systat Software , Inc . , San Jose , CA ) and IBM SPSS statistics 12 ( IBM Corp . , Armonk , NY ) . All experimental data are represented as mean ± SEM and were plotted using Microsoft Excel 2016 ( Microsoft Corp , Redmond , WA ) . All figures were prepared using Adobe Illustrator CS5 ( Adobe Systems Inc . , San Jose , CA ) . We developed a single-compartment model of a heart interneuron using Hodgkin-Huxley style equations . Our model has a leak ( Ileak ) current and a Na+/K+ pump ( Ipump ) current , with the leak current having Na+ ( I ( leak , Na ) and K+ ( I ( leak , K ) components . The model also has eight voltage-gated currents: a fast Na+ current ( INa ) , a persistent Na+ current ( IP ) , a low-threshold rapidly inactivating Ca2+ current ( ICaF ) , a low-threshold slowly inactivating Ca2+ current ( ICaS ) , a hyperpolarization-activated inward current ( Ih ) , a delayed rectifier-like K+ current ( IK1 ) , a persistent K+ current ( IK2 ) , and a fast transient K+ current ( IKA ) . The model of a single heart interneuron can be converted into a half-center oscillator by including a spike-mediated synaptic current ( ISynS ) and a graded synaptic current ( ISynG ) , as follows: ( 3 ) CdVdt=− ( INa+IP+IK1+IK2+IKA+Ih , Na+Ih , K+ICaF+ICaS+ILeak , Na+ILeak , K+IPump+ISynS+ISynG ) where C is the membrane capacitance ( in nF ) , V is the membrane potential ( in V ) , t is time ( in s ) . Our half-center oscillator model in Equation 3 differs from the original Hill et al . ( 2001 ) model because it includes a Na+/K+ pump current and it describes changes in intracellular Na+ concentrations that occur as a result of the Na+ fluxes carried by ionic currents , Na+/K+ pumps , and monensin-facilitated diffusion: ( 4 ) d[Na]idt=M ( [Na]o−[Na]i ) −IP+INa+Ih , Na+ Ileak , Na+3 IpumpvF In Equation 4 , [Na]i is the changing intracellular Na+ concentration , [Na]o is the extracellular Na+ concentration that was kept constant , v is the volume ( ~6 . 7 pL ) of the intracellular Na+ reservoir , F is Faraday’s constant , and M is the exchange rate ( in 1/s ) of Na+ and H+ by monensin , which is based on Fick’s Law of diffusion . Because the Na+/K+ pump exchanges two K+ ions for three Na+ ions , the contribution of the pump current to intracellular Na+ concentrations was multiplied by a factor of 3 . The Na+/K+ pump current has a sigmoidal dependence on intracellular Na+ concentrations , which is expressed as follows: ( 5 ) Ipump=Ipumpmax1+exp ( [Na]ih−[Na]i[Na]is ) where Ipumpmax is the maximum Na+/K+ pump current , [Na]ih is the intracellular Na+ concentration for the half-activation of the Na+/K+ pump , and [Na]is the sensitivity of the Na+/K+ pump to [Na]i . The h-current and the leak current have Na+ and K+ components . In the case of the h-current , we computed the Na+ component ( Ih , Na ) using the equilibrium potential of Na+ and we computed the K+ component ( Ih , K ) using the equilibrium potential of K+: ( 6 ) Ih , Na=37g¯hmh2 ( Vm−ENa ) ( 7 ) Ih , K=47g¯hmh2 ( Vm−EK ) In both equations , g¯his the maximum conductance , mh is the activation variable , Vm is the membrane potential , and Eion is the equilibrium potential at 20°C . Unlike the equilibrium potential of K+ , the equilibrium potential of Na+ was computed at each time step as a function of a constant extracellular Na+ concentration and a changing intracellular Na+ concentration: ( 8 ) ENa=0 . 02526ln ( [Na]o[Na]i ) We also computed two components ( Ileak , Naand I ( leak , K ) of the leak current: ( 9 ) Ileak , Na=g¯leak , Na ( Vm−ENa ) ( 10 ) Ileak , K=g¯leak , K ( Vm−EK ) We fixed the ratio of Na+ to K+ conducted by the leak current and used the equilibrium potentials of K+ and Na+ from Hill et al . ( 2001 ) to compute this ratio . The equations for g¯leak , Na and g¯leak , K are: ( 11 ) g¯leak , Na=g¯leak ( Eleak , Ref−EK ) ( ENa , Ref−EK ) ( 12 ) g¯leak , K=g¯leak ( Eleak , Ref − ENa , Ref ) ( EK−ENa , Ref ) In Equation 12 , ENa , Ref was fixed at 0 . 045 V . The Na+ and K+ components of the leak conductance , g¯leak , Na and g¯leak , K , were computed ahead of time and were fixed for the duration of each simulation . We computed the average value function I of the Na+/K+ pump current and the h-current over a bounded integral as follows: ( 13 ) I=1t2−t1∫t1t2Idt Both average currents were computed over each burst duration and over each interburst interval . When the average currents were computed over a burst duration , t1 represented the time of the first action potential in a burst whereas t2 represented the time of the last action potential in the same burst . When the average currents were computed during an interburst interval , in which the membrane potential was below −50 mV , t1 represented the time at which the membrane potential crossed −50 mV after the last action potential in the preceding burst whereas t2 represented the time at which the membrane potential crossed −50 mV immediately before the first action potential of the next burst . We used the trapezoid method for numerical integration ( the trapz function in MATLAB ) . A complete list of Hodgkin-Huxley style equations that describe ionic and synaptic currents can be found in the Appendix . We computed numerical solutions to these ordinary differential equations using the 8–9 order Prince-Dormand method from the GNU Scientific Library ( www . gnu . org/software/gsl ) . All variables were computed with an absolute tolerance of 1e−9s , a relative tolerance of 1e−10s , and a maximum time step of 1e−3s . We analyzed the burst characteristics of current and voltage trajectories from our model with custom-made routines written in the C programming language . The four burst characteristics that we measured were the period , burst duration , interburst interval , and duty cycle . The burst characteristics from our model were measured in the same way as the burst characteristics from our experimental data ( see Experimental data analysis ) . | In animals , cells called neurons relay information around the body in the form of electrical signals . An enzyme called the sodium and potassium pump is found in the membrane that surrounds neurons . It uses energy to pump sodium ions out of the neuron and potassium ions in the opposite direction . This helps to maintain different concentrations of these ions across the membrane , which is critical for the electrical activity of neurons and also generates an electrical current in the process . The size of the current is influenced by how many sodium ions have leaked back into the neuron due to the neuron’s electrical activity . Neurons control many rhythmic processes in animals including breathing and heartbeats . However , it was not clear how the current produced by the sodium and potassium pump contributes to the rhythms in neural activity that drive these processes . To address this question , Kueh et al . investigated the effect of drugs that alter the activity of the pump in neurons that control heartbeat in leeches . The experiments show that stimulating the pump by altering the amount of sodium ions that leak into the neuron dramatically sped up the rhythmic activity of these neurons . This effect depended completely on the presence of a channel protein – called an h-channel – that was activated with a delay by the altered pump current and allowed sodium and potassium ions to cross the membrane , counteracting the pump current . Inhibiting the pump also sped up the rhythm of neural activity , but this effect did not depend on the h-channel . Kueh et al . developed a computer model that indicated that the time course of the pump current following the sodium ion leak and the slow activation of the h-channel were important for these effects . Previous studies have shown that a particular signal molecule modulates the activity of both the pump and the h-channel in neurons . Therefore , a future challenge is to find out how the pump and the h-channel interact while their activities change in response to the signal molecule . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"neuroscience"
] | 2016 | Na+/K+ pump interacts with the h-current to control bursting activity in central pattern generator neurons of leeches |
Although arthropods are important viral vectors , the biodiversity of arthropod viruses , as well as the role that arthropods have played in viral origins and evolution , is unclear . Through RNA sequencing of 70 arthropod species we discovered 112 novel viruses that appear to be ancestral to much of the documented genetic diversity of negative-sense RNA viruses , a number of which are also present as endogenous genomic copies . With this greatly enriched diversity we revealed that arthropods contain viruses that fall basal to major virus groups , including the vertebrate-specific arenaviruses , filoviruses , hantaviruses , influenza viruses , lyssaviruses , and paramyxoviruses . We similarly documented a remarkable diversity of genome structures in arthropod viruses , including a putative circular form , that sheds new light on the evolution of genome organization . Hence , arthropods are a major reservoir of viral genetic diversity and have likely been central to viral evolution .
Negative-sense RNA viruses are important pathogens that cause a variety of diseases in humans including influenza , hemorrhagic fever , encephalitis , and rabies . Taxonomically , those negative-sense RNA viruses described to date comprise at least eight virus families and four unassigned genera or species ( King et al . , 2012 ) . Although they share ( i ) a homologous RNA-dependent RNA polymerase ( RdRp ) , ( ii ) inverted complementary genome ends , and ( iii ) an encapsidated negative-sense RNA genome , these viruses display substantial diversity in terms of virion morphology and genome organization ( King et al . , 2012 ) . One key aspect of genome organization is the number of distinct segments , which is also central to virus classification . Among negative-sense RNA viruses , the number of segments varies from one ( order Mononegavirales; unsegmented ) to two ( family Arenaviridae ) , three ( Bunyaviridae ) , three-to-four ( Ophioviridae ) , and six-to-eight ( Orthomyxoviridae ) and is further complicated by differences in the number , structure , and arrangement of the encoded genes . Despite their diversity and importance in infectious disease , the origins and evolutionary history of the negative-sense RNA viruses are largely obscure . Arthropods harbor a diverse range of RNA viruses , which are often divergent from those that infect vertebrates ( Marklewitz et al . , 2011 , 2013; Cook et al . , 2013; Ballinger et al . , 2014; Qin et al . , 2014; Tokarz et al . , 2014a , 2014b ) . However , those arthropod viruses sampled to date are generally those that have a relationship with vertebrates or are known to be agents of disease ( Junglen and Drosten , 2013 ) . To determine the extent of viral diversity harbored by arthropods , as well as their evolutionary history , we performed a systematic survey of negative-sense RNA viruses using RNA sequencing ( RNA-seq ) on a wide range of arthropods .
We focused our study of virus biodiversity and evolution on 70 potential host species from four arthropod classes: Insecta , Arachnida , Chilopoda , and Malacostraca ( Table 1 and Figure 1 ) . From these samples , 16 separate cDNA libraries were constructed and sequenced , resulting in a total of 147 . 4 Gb of 100-base pair-end reads ( Table 1 ) . Blastx comparisons against protein sequences of negative-sense RNA virus revealed 108 distinct types of complete or nearly complete large ( L ) proteins ( or polymerase protein 1 ( PB1 ) in the case of orthomyxoviruses ) that encode the relatively conserved RdRp ( Tables 2–4 ) . Four additional types of previously undescribed RdRp sequence ( >1000 amino acids ) were identified from the Transcriptome Shotgun Assembly ( TSA ) database . Together , these proteins exhibited an enormous diversity in terms of sequence variation and structure . Most notably , this data set of RdRp sequences is distinct from both previously described sequences and from each other , with the most divergent showing as little as 15 . 8% amino acid sequence identity to its closest relatives ( Tables 2–4 ) . Overall , these data provide evidence for at least 16 potentially new families and genera of negative-sense RNA viruses , defined as whose RdRp sequences shared less than 25% amino acid identity with existing taxa . 10 . 7554/eLife . 05378 . 003Table 1 . Host and geographic information and data output for each pool of arthropod samplesDOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 003PoolNo of unitsOrderSpeciesLocationsData generated ( bases ) Mosquitoes—Hubei24DipteraAedes sp , Armigeres subalbatus , Anopheles sinensis , Culex quinquefasciatus , Culex tritaeniorhynchusHubei26 , 606 , 799 , 000Mosquitoes—Zhejiang26DipteraAedes albopictus , Armigeres subalbatus , Anopheles paraliae , Anopheles sinensis , Culex pipiens , Culex sp , Culex tritaeniorhynchusZhejiang7 , 233 , 954 , 480True flies24DipteraAtherigona orientalis , Chrysomya megacephala , Lucilia sericata , Musca domestica , Sarcophaga dux , S . peregrina , S . spHubei6 , 574 , 954 , 320Horseflies24DipteraUnidentified Tabanidae ( 5 species ) Hubei8 , 721 , 642 , 060Cockroaches24BlattodeaBlattella germanicaHubei6 , 182 , 028 , 000Water striders12HemipteraUnidentified Gerridae ( 2 species ) Hubei3 , 154 , 714 , 200Insects mix 16Diptera , Coleoptera , Lepidoptera , NeuropteraAbraxas tenuisuffusa , Hermetia illucens , unidentified Chrysopidae , unidentified Coleoptera , Psychoda alternata , unidentified Diptera , unidentified StratiomyidaeZhejiang7 , 745 , 172 , 660Insects mix 24Diptera , HemipteraUnidentified Hippoboscidae ( 2 species ) , Cimex hemipterusHubei5 , 916 , 431 , 520Insects mix 3 ( insect near water ) 10Odonata , Hemiptera , Hymenoptera , IsopodaPseudothemis zonata , unidentified Nepidae ( 2 species ) , Camponotus japonicus , Diplonychus sp , Asellus spHubei11 , 973 , 368 , 200Insects mix 4 ( insect in the mountain ) 12Diptera , Orthoptera , Odonata , Hymenoptera , HemipteraPsychoda alternata , Velarifictorus micado , Crocothemis servilia , unidentified Phoridae , unidentified Lampyridae , Aphelinus sp , Hyalopterus pruni , Aulacorthum magnoliaHubei6 , 882 , 491 , 800Ticks16IxodidaDermacentor marginatus , Dermacentor sp , Haemaphysalis doenitzi , H . longicornis , H . sp , H . formosensis , Hyalomma asiaticum , Rhipicephalus microplus , Argas miniatusHubei , Zhejiang , Beijing , Xinjiang24 , 708 , 479 , 580Ticks Hyalomma asiaticum1IxodidaHyalomma asiaticumXinjiang2 , 006 , 000 , 100Spiders32AraneaeNeoscona nautica , Parasteatoda tepidariorum , Plexippus setipes , Pirata sp , unidentified AraneaeHubei11 , 361 , 912 , 300Shrimps48DecapodaExopalaemon carinicauda , Metapenaeus sp , Solenocera crassicornis , Penaeus monodon , Litopenaeus vannameiZhejiang5 , 365 , 359 , 900Crabs and barnacles35Decapoda , ScalpelliformesCapitulum mitella , Charybdis hellerii , C . japonica , Uca arcuataZhejiang5 , 833 , 269 , 360Millipedes12PolydesmidaUnidentified Polydesmidae ( 2 species ) Hubei , Beijing7 , 176 , 702 , 40010 . 7554/eLife . 05378 . 004Figure 1 . Host component of each pool used in the RNA-seq library construction and sequencing . The taxonomic units in the tree correspond to the unit samples used in the RNA extraction . Species or genus information is marked to the left of the tree . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 00410 . 7554/eLife . 05378 . 005Table 2 . Mononegavirales-related RdRp sequences discovered in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 005Virus nameLength of RdRpClassificationPoolAbundancePutative arthropod hostClosest relative ( aa identity ) Bole Tick Virus 32155ChuvirusTicks202 . 35Hyalomma asiaticumMidway virus ( 17 . 1% ) Changping Tick Virus 22156ChuvirusTicks185 . 73Dermacentor spMidway virus ( 17 . 6% ) Changping Tick Virus 32209ChuvirusTicks41 . 80Dermacentor spMidway virus ( 16 . 5% ) Lishi Spider Virus 12180ChuvirusSpiders5 . 82Parasteatoda tepidariorumMidway virus ( 16 . 9% ) Shayang Fly Virus 12459ChuvirusTrue flies8 . 99Atherigona orientalisMaize mosaic virus ( 16 . 8% ) Shuangao Fly Virus 12097ChuvirusInsect mix 123 . 63Unidentified DipteraLettuce big-vein associated virus ( 16 . 3% ) Shuangao Insect Virus 52291ChuvirusInsect mix 1209 . 31Unidentified Diptera , Abraxas tenuisuffusa , unidentified ChrysopidaePotato yellow dwarf virus ( 16 . 3% ) Shuangao Lacewing Virus2145ChuvirusInsect mix 144 . 48Unidentified ChrysopidaePotato yellow dwarf virus ( 16 . 8% ) Tacheng Tick Virus 42101ChuvirusTicks137 . 22Argas miniatusMidway virus ( 17 . 5% ) Tacheng Tick Virus 52201ChuvirusTicks276 . 32Dermacentor marginatusMidway virus ( 16 . 8% ) Wenzhou Crab Virus 22208ChuvirusCrabs and barnacles4054 . 25Charybdis japonica , Charybdis lucifera , Charybdis helleriiMidway virus ( 15 . 8% ) Wenzhou Crab Virus 32077ChuvirusCrabs and barnacles169 . 21Charybdis japonicaMidway virus ( 16 . 3% ) Wuchang Cockroach Virus 32203ChuvirusCockroaches440 . 14Blattella germanicaMidway virus ( 16 . 3% ) Wuhan Louse Fly Virus 62182ChuvirusInsect mix 24 . 12Unidentified HippoboscidaeMidway virus ( 16 . 4% ) Wuhan Louse Fly Virus 72174ChuvirusInsect mix 299 . 83Unidentified HippoboscidaeMidway virus ( 17 . 2% ) Wuhan Mosquito Virus 82159ChuvirusMosquito Hubei300 . 33Culex tritaeniorhynchus , C . quinquefasciatus , Anopheles sinensis , Armigeres subalbatusMidway virus ( 16 . 7% ) Wuhan Tick Virus 22189ChuvirusTicks154 . 46Rhipicephalus microplusMidway virus ( 16 . 7% ) Culex tritaeniorhynchus rhabdovirus2142Culex tritaeniorhynchus rhabdovirusMosquito Hubei3517 . 32Culex tritaeniorhynchus , C . quinquefasciatus , Anopheles sinensis , Armigeres subalbatus , Aedes spIsfahan virus ( 38 . 5% ) Wuhan Insect Virus 42105CytorhabdovirusInsect mix 494 . 92Hyalopterus pruni OR Aphelinus spLettuce necrotic yellows virus ( 40 . 6% ) Wuhan Insect Virus 52098CytorhabdovirusInsect mix 4622 . 97Hyalopterus pruni OR Aphelinus spPersimmon virus A ( 47 . 9% ) Wuhan Insect Virus 62079CytorhabdovirusInsect mix 4991 . 99Hyalopterus pruni OR Aphelinus spPersimmon virus A ( 45 . 2 ) Wuhan Louse Fly Virus 52123Kolente virus likeInsect mix 298 . 92Unidentified HippoboscidaeKolente virus ( 54 . 5% ) Yongjia Tick Virus 22113Nishimuro virus likeTicks13 . 14Haemaphysalis hystricisNishimuro virus ( 54 . 2% ) Shayang Fly Virus 22170Sigmavirus likeTrue flies36 . 83Musca domestica , Chrysomya megacephalaIsfahan virus ( 44 . 1% ) Wuhan Fly Virus 22134Sigmavirus likeTrue flies18 . 37Musca domestica , Sarcophaga spVesicular stomatitis Indiana virus ( 43 . 4% ) Wuhan House Fly Virus 12098Sigmavirus likeTrue flies31 . 04Musca domesticaIsfahan virus ( 42 . 8% ) Wuhan Louse Fly Virus 102146Sigmavirus likeInsect mix 2235 . 94Unidentified HippoboscidaeDrosophila melanogaster sigmavirus ( 51 . 2% ) Wuhan Louse Fly Virus 82145Sigmavirus likeInsect mix 2292 . 11Unidentified HippoboscidaeDrosophila melanogaster sigmavirus ( 50 . 6% ) Wuhan Louse Fly Virus 92145Sigmavirus likeInsect mix 269 . 37Unidentified HippoboscidaeDrosophila melanogaster sigmavirus ( 51 . 4% ) Bole Tick Virus 22171Unclassified dimarhabdovirus 1Ticks38 . 19Hyalomma asiaticumIsfahan virus ( 38 . 1% ) Huangpi Tick Virus 32193Unclassified dimarhabdovirus 1Ticks15 . 81Haemaphysalis doenitziEel virus European X ( 40% ) Tacheng Tick Virus 32182Unclassified dimarhabdovirus 1Ticks96 . 30Dermacentor marginatusEel virus European X ( 39 . 8% ) Taishun Tick Virus2226Unclassified dimarhabdovirus 1Ticks24 . 56Haemaphysalis hystricisVesicular stomatitis Indiana virus ( 36 . 6% ) Wuhan Tick Virus 12191Unclassified dimarhabdovirus 1Ticks119 . 92Rhipicephalus microplusEel virus European X ( 38 . 3% ) Wuhan Insect Virus 72120Unclassified dimarhabdovirus 2Insect mix 4241 . 7Hyalopterus pruni OR Aphelinus spIsfahan virus ( 42 . 6% ) Lishi Spider Virus 22201Unclassified mononegavirus 1Spiders5 . 57Unidentified AraneaeMaize fine streak virus ( 19 . 6% ) Sanxia Water Strider Virus 42108Unclassified mononegavirus 1Water striders4767 . 82Unidentified GerridaeOrchid fleck virus ( 20 . 5% ) Tacheng Tick Virus 62068Unclassified mononegavirus 1Ticks17 . 92Argas miniatusMaize mosaic virus ( 20 . 6% ) Shuangao Fly Virus 21966Unclassified mononegavirus 2Insect mix 125 . 94Psychoda alternataMidway virus ( 21 . 3% ) Xincheng Mosquito Virus2026Unclassified mononegavirus 2Mosquito Hubei400 . 12Anopheles sinensisMidway virus ( 19 . 2% ) Wenzhou Crab Virus 11807Unclassified mononegavirus 3Crabs and barnacles382 . 29Capitulum mitella , Charybdis japonica , Charybdis luciferaMidway virus ( 22 . 2% ) Tacheng Tick Virus 72215Unclassified rhabdovirus 1Ticks35 . 86Argas miniatusOrchid fleck virus ( 24 . 5% ) Jingshan Fly Virus 21970Unclassified rhabdovirus 2True flies4 . 43Sarcophaga spMaize fine streak virus ( 23 . 4% ) Sanxia Water Strider Virus 52264Unclassified rhabdovirus 2Water striders4373 . 68Unidentified GerridaeNorthern cereal mosaic virus ( 22 . 6% ) Shayang Fly Virus 32231Unclassified rhabdovirus 2True flies27 . 73Chrysomya megacephala , Atherigona orientalisMaize fine streak virus ( 22 . 6% ) Shuangao Bedbug Virus 22207Unclassified rhabdovirus 2Insect mix 216 . 29Cimex hemipterusMaize fine streak virus ( 22 . 5% ) Shuangao Insect Virus 62088Unclassified rhabdovirus 2Insect mix 114 . 37Unidentified Diptera , Abraxas tenuisuffusaPotato yellow dwarf virus ( 21 . 2% ) Wuhan Ant Virus2118Unclassified rhabdovirus 2Insect mix 3169 . 79Camponotus japonicusLettuce necrotic yellows virus ( 21 . 4% ) Wuhan Fly Virus 32230Unclassified rhabdovirus 2True flies6 . 00Musca domestica , Sarcophaga spMaize fine streak virus ( 21 . 9% ) Wuhan House Fly Virus 22233Unclassified rhabdovirus 2True flies221 . 04Musca domesticaNorthern cereal mosaic virus ( 23 . 4% ) Wuhan Mosquito Virus 92260Unclassified rhabdovirus 2Mosquito Hubei56 . 19Culex tritaeniorhynchus , C . quinquefasciatus , Aedes spPersimmon virus A ( 23 . 2% ) Wuhan Louse Fly Virus 112110Vesiculovirus likeInsect mix 26 . 11Unidentified HippoboscidaeVesicular stomatitis Indiana virus ( 52 . 9% ) 10 . 7554/eLife . 05378 . 006Table 3 . Bunya-arenaviridae-related RdRp sequences discovered in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 006Virus nameLength of RdRpClassificationPoolAbundancePutative arthropod hostClosest relative ( aa identity ) Huangpi Tick Virus 13914Nairovirus likeTicks11 . 32Haemaphysalis doenitziHazara virus ( 39 . 5% ) Tacheng Tick Virus 13962Nairovirus likeTicks88 . 91Dermacentor marginatusHazara virus ( 39 . 6% ) Wenzhou Tick Virus3967Nairovirus likeTicks44 . 30Haemaphysalis hystricisCrimean-Congo hemorrhagic fever virus ( 39 . 1% ) Shayang Spider Virus 14403Nairovirus likeSpiders90 . 95Neoscona nautica , Parasteatoda tepidariorum , Plexippus setipesCrimean-Congo hemorrhagic fever virus ( 26 . 2% ) Xinzhou Spider Virus4037Nairovirus likeSpiders3 . 79Neoscona nautica , Parasteatoda tepidariorumErve virus ( 22 . 9% ) Sanxia Water Strider Virus 13936Nairovirus likeWater striders26 , 483 . 38Unidentified GerridaeHazara virus ( 23 . 4% ) Wuhan Louse Fly Virus 12250OrthobunyavirusInsect mix 267 . 06Unidentified HippoboscoideaLa Crosse virus ( 57 . 8% ) Shuangao Insect Virus 12335Orthobunyavirus likeInsect mix 17 . 97Unidentified Chrysopidae , Psychoda alternataKhurdun virus ( 29 . 1% ) Wuchang Cockroach Virus 12125Phasmavirus likeCockroaches11 , 283 . 22Blattella germanicaKigluaik phantom virus ( 35 . 9% ) GAQJ010071891554Phasmavirus likeDatabaseN/AOstrinia furnacalisKigluaik phantom virus ( 35 . 9% ) Shuangao Insect Virus 21765Phasmavirus likeInsect mix 136 . 32Abraxas tenuisuffusa , unidentified DipteraKigluaik phantom virus ( 31 . 9% ) Wuhan Mosquito Virus 12095Phasmavirus likeMosquito Hubei , Mosquito Zhejiang3523 . 08Culex tritaeniorhynchus , Anopheles sinensis , Culex quinquefasciatusKigluaik phantom virus ( 39 . 5% ) Wuhan Mosquito Virus 22111Phasmavirus likeMosquito Hubei , Mosquito Zhejiang39 . 66Culex tritaeniorhynchus , Anopheles sinensis , Culex quinquefasciatus , Aedes spKigluaik phantom virus ( 39 . 6% ) Huangpi Tick Virus 22121PhlebovirusN/AN/AHaemaphysalis spUukuniemi virus ( 49 . 3% ) Bole Tick Virus 12148PhlebovirusTicks67 . 86Hyalomma asiaticumUukuniemi virus ( 37 . 9% ) Changping Tick Virus 12194PhlebovirusTicks335 . 25Dermacentor spUukuniemi virus ( 39 . 7% ) Dabieshan Tick Virus2148PhlebovirusTicks250 . 62Haemaphysalis longicornisUukuniemi virus ( 39 . 2% ) Lihan Tick Virus2151PhlebovirusTicks60 . 40Rhipicephalus microplusUukuniemi virus ( 38 . 6% ) Tacheng Tick Virus 22189PhlebovirusTicks132 . 59Dermacentor marginatusUukuniemi virus ( 39 . 0% ) Yongjia Tick Virus 12138PhlebovirusTicks119 . 49Haemaphysalis hystricisUukuniemi virus ( 40 . 5% ) GAIX010000592151Phlebovirus likeDatabaseN/APararge aegeriaCumuto virus ( 24 . 1% ) GAKZ010482601583Phlebovirus likeDatabaseN/AProcotyla fluviatilisCumuto virus ( 22 . 8% ) GAQJ010086812261Phlebovirus likeDatabaseN/AOstrinia furnacalisGouleako virus ( 22 . 0% ) Shuangao Insect Virus 32050Phlebovirus likeInsect mix 1339 . 41Unidentified Chrysopidae , unidentified DipteraCumuto virus ( 23 . 7% ) Wuhan Louse Fly Virus 22327Phlebovirus likeInsect mix 23 . 57Unidentified HippoboscoideaUukuniemi virus ( 25 . 2% ) Wuhan Insect Virus 12099Phlebovirus likeInsect mix 3178 . 53Asellus sp , unidentified Nepidae , Camponotus japonicusCumuto virus ( 24 . 8% ) Huangshi Humpbacked Fly Virus2009Phlebovirus likeInsect mix 413 . 13Unidentified PhoridaeCumuto virus ( 18 . 1% ) Yichang Insect Virus2100Phlebovirus likeInsect mix 471 . 50Aulacorthum magnoliaeGouleako virus ( 45 . 3% ) Wuhan Millipede Virus 11854Phlebovirus likeMillipedes and insect mix 3825 . 66Unidentified PolydesmidaeCumuto virus ( 25 . 3% ) Qingnian Mosquito Virus2243Phlebovirus likeMosquito Hubei17 . 09Culex quinquefasciatusRazdan virus ( 21 . 0% ) Wutai Mosquito Virus2185Phlebovirus likeMosquito Hubei70 . 72Culex quinquefasciatusRice stripe virus ( 26 . 4% ) Xinzhou Mosquito Virus2022Phlebovirus likeMosquito Hubei98 . 95Anopheles sinensisCumuto virus ( 24 . 7% ) Zhee Mosquito Virus2443Phlebovirus likeMosquito Hubei , Mosquito Zhejiang308 . 98Anopheles sinensis , Armigeres subalbatusCumuto virus ( 22 . 6% ) Wenzhou Shrimp Virus 12051Phlebovirus likeShrimps5859 . 37Penaeus monodonUukuniemi virus ( 32 . 2% ) Wuhan Spider Virus2251Phlebovirus likeSpiders17 . 71Neoscona nautica , Parasteatoda tepidariorum , Plexippus setipesUukuniemi virus ( 21 . 7% ) Wuhan Fly Virus 12192Phlebovirus likeTrue flies68 . 58Atherigona orientalis , Chrysomya megacephala , Sarcophaga sp , Musca domesticaGrand Arbaud virus ( 27 . 8% ) Wuhan Horsefly Virus3117Tenuivirus likeHorseflies13 . 50Unidentified TabanidaeUukuniemi virus ( 28 . 2% ) Jiangxia Mosquito Virus 11889Unclassified segmented virus 1Mosquito Hubei11 . 55Culex tritaeniorhynchusGouleako virus ( 16 . 7% ) Shuangao Bedbug Virus 12015Unclassified segmented virus 2Insect mix 212 . 71Cimex hemipterusMurrumbidgee virus ( 16 . 3% ) Jiangxia Mosquito Virus 21860Unclassified segmented virus 2Mosquito Hubei2 . 81Culex tritaeniorhynchusHantavirus ( 18 . 9% ) Shuangao Mosquito Virus1996Unclassified segmented virus 2Mosquito Zhejiang11 . 67Armigeres subalbatusHantavirus ( 18 . 7% ) Wenzhou Shrimp Virus 22241Unclassified segmented virus 3Shrimps3824 . 55Penaeus monodon , Exopalaemon carinicaudaLa Crosse virus ( 19 . 0% ) Shayang Spider Virus 22165Unclassified segmented virus 4Spiders12 . 75Neoscona nautica , Pirata sp , Parasteatoda tepidariorum , unidentified AraneaeAkabane virus ( 16 . 6% ) Wuhan Insect Virus 22377Unclassified segmented virus 5Insect mix 4223 . 06Hyalopterus pruni OR Aphelinus spKigluaik phantom virus ( 19 . 2% ) Sanxia Water Strider Virus 22349Unclassified segmented virus 5Water striders707 . 09Unidentified GerridaeKigluaik phantom virus ( 19 . 8% ) Wuhan Millipede Virus 23709Unclassified segmented virus 6Millipedes1513 . 41Unidentified PolydesmidaeDugbe virus ( 17 . 2% ) Wuhan Insect Virus 32231Unclassified segmented virus 7Insect mix 33 . 50Asellus spHerbert virus ( 17 . 2% ) 10 . 7554/eLife . 05378 . 007Table 4 . Orthomyxoviridae-related RdRp sequences discovered in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 007Virus nameLength of RdRpClassificationPoolAbundancePutative arthropod hostClosest relative ( aa identity ) Jingshan Fly Virus 1795QuaranjavirusTrue flies21 . 93Atherigona orientalis , Chrysomya megacephala , Sarcophaga sp , Musca domesticaJohnston Atoll virus ( 36 . 9% ) Jiujie Fly Virus653QuaranjavirusHorseflies10 . 30Unidentified TabanidaeJohnston Atoll virus ( 39 . 7% ) Sanxia Water Strider Virus 3789QuaranjavirusWater striders1101 . 03Unidentified GerridaeJohnston Atoll virus ( 36 . 7% ) Shayang Spider Virus 3768QuaranjavirusSpiders1 . 95Neoscona nauticaJohnston Atoll virus ( 38 . 5% ) Shuangao Insect Virus 4793QuaranjavirusInsect mix 159 . 90Unidentified Diptera , unidentified StratiomyidaeJohnston Atoll virus ( 36 . 9% ) Wuhan Louse Fly Virus 3784QuaranjavirusInsect mix 2500 . 77Unidentified HippoboscoideaJohnston Atoll virus ( 37 . 7% ) Wuhan Louse Fly Virus 4783QuaranjavirusInsect mix 296 . 80Unidentified HippoboscoideaJohnston Atoll virus ( 38 . 2% ) Wuhan Mosquito Virus 3801QuaranjavirusMosquito Hubei40 . 07Culex tritaeniorhynchus , Culex quinquefasciatus , Armigeres subalbatusJohnston Atoll virus ( 35 . 6% ) Wuhan Mosquito Virus 4792QuaranjavirusMosquito Hubei86 . 21Culex tritaeniorhynchus , Culex quinquefasciatus , Armigeres subalbatusJohnston Atoll virus ( 34 . 8% ) Wuhan Mosquito Virus 5806QuaranjavirusMosquito Hubei75 . 05Culex tritaeniorhynchus , Culex quinquefasciatus , Armigeres subalbatusJohnston Atoll virus ( 35 . 5% ) Wuhan Mosquito Virus 6800QuaranjavirusMosquito Hubei56 . 30Culex quinquefasciatusJohnston Atoll virus ( 34 . 2% ) Wuhan Mosquito Virus 7779QuaranjavirusMosquito Hubei20 . 74Anopheles sinensis , Culex quinquefasciatusJohnston Atoll virus ( 34 . 1% ) Wuhan Mothfly Virus710QuaranjavirusInsect mix 414 . 47Psychoda alternataJohnston Atoll virus ( 39 . 7% ) Wuchang Cockroach Virus 2671Unclassified orthomyxovirus 1Cockroaches4 . 01Blattella germanicaInfluenza C virus ( 27 . 0% ) Next , we measured the abundance of these sequences as the number transcripts per million ( TPM ) within each library after the removal of rRNA reads . The abundance of viral transcripts calculated in this manner exhibited substantial variation ( Figure 2 , Tables 2–4 ) : while the least abundant L segment ( Shayang Spider Virus 3 ) contributed to less than 0 . 001% to the total non-ribosomal RNA content , the most abundant ( Sanxia Water Strider Virus 1 ) was at a frequency of 21 . 2% , and up to 43 . 9% if we include the matching M and S segments of the virus . The remaining viral RdRp sequences fell within a range ( 10–1000 TPM ) that matched the abundance level of highly expressed host mitochondrial genes ( Figure 2 ) . 10 . 7554/eLife . 05378 . 008Figure 2 . Abundance level ( transcripts per million—TPM ) of the RdRp genes from the negative-sense RNA viruses detected in this study . Abundance is calculated after the removal of ribosomal RNA reads . As a comparison , we show the abundance of the two well characterized ( positive-sense ) RNA viruses: Japanese encephalitis virus and Gill-associated virus found in the Mosquito-Hubei and Shrimp libraries , respectively , as well as the range of abundance of host mitochondrial COI genes in these same multi-host libraries . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 008 With this highly diverse set of RdRp sequences in hand we re-examined the evolution of all available negative-sense RNA viruses by phylogenetic analysis ( Figure 3; Figure 3—figure supplement 3 ) . These data greatly expand the documented diversity of four viral families/orders—the Arenaviridae , Bunyaviridae , Orthomyxoviridae , and Mononegavirales—as well as of three floating genera—Tenuivirus , Emaravirus , and Varicosavirus ( King et al . , 2012 ) . Most of the newly described arthropod viruses fell basal to the known genetic diversity in these taxa: their diversity either engulfed that of previously described viruses , as in the case of phlebovirus , nairovirus , and dimarhabdovirus , or appeared as novel lineages sandwiched between existing genera or families , and hence filling in a number of phylogenetic ‘gaps’ ( Figure 3; Figure 3—figure supplement 3 ) . One important example was a large monophyletic group of newly discovered viruses that fell between the major groups of segmented and unsegmented viruses ( Figure 4 ) ; we name this putative new virus family the ‘Chuviridae’ reflecting the geographic location in China where most of this family were identified ( ‘Chu’ is a historical term referring to large area of China encompassing the middle and lower reaches of the Yangzi River ) . Also of note was that some of the previously defined families no longer appear as monophyletic . For example , although classified as distinct families , the family Arenaviridae fell within the genetic diversity of the family Bunyaviridae and as a sister group to viruses of the genus Nairovirus . Furthermore , the floating genus Tenuivirus was nested within the Phlebovirus-like clade , and another floating genus , Emaravirus , formed a monophyletic group with the Orthobunyavirus and Tospovirus genera ( Figure 3C; Figure 3—figure supplement 2 ) . Hence , there are important inconsistencies between the current virus classification scheme and the underlying evolutionary history of the RdRp revealed here . 10 . 7554/eLife . 05378 . 009Figure 3 . Evolutionary history of negative-sense RNA viruses based on RdRp . This is initially displayed in an unrooted maximum likelihood ( ML ) tree including all major groups of negative-sense RNA viruses ( A ) . Separate and more detailed ML phylogenies are then shown for the Orthomyxoviridae-like ( B ) , Bunya-Arenaviridae-like ( C ) , and Mononegavirales-like viruses ( D ) . In all the phylogenies , the RdRp sequences described here from arthropods are either shaded purple or marked with solid gray circles . The names of previously defined genera/families are labeled to the right of the phylogenies . Based on their host types , the branches are shaded red ( vertebrate-specific ) , yellow ( vertebrate and arthropod ) , green ( plant and arthropod ) , blue ( non-arthropod invertebrates ) , or black ( arthropod only ) . For clarity , statistical supports ( i . e . , approximate likelihood-ratio test ( aLRT ) with Shimodaira–Hasegawa-like procedure/posterior probabilities ) are shown for key internal nodes only . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 00910 . 7554/eLife . 05378 . 010Figure 3—figure supplement 1 . A fully labeled ML phylogeny for Orthomyxoviridae-like viruses . The phylogeny is reconstructed using RdRp alignments . Statistical support from the approximate likelihood-ratio test ( aLRT ) is shown on each node of the tree . The names of the viruses discovered in this study are shown in red . The names of reference sequences , which contain both the GenBank accession number and the virus species name , are shown in black . The names of previously defined genera/families are shown to the right of the phylogenies . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 01010 . 7554/eLife . 05378 . 011Figure 3—figure supplement 2 . A fully labeled ML phylogeny for Bunya-Arenaviridae-like viruses . The phylogeny is reconstructed using RdRp alignments . Statistical support from the aLRT is shown on each node of the tree . The names of the viruses discovered in this study are shown in red . The names of reference sequences , which contain both the GenBank accession number and the virus species name , are shown in black . The names of previously defined genera/families are shown to the right of the phylogenies . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 01110 . 7554/eLife . 05378 . 012Figure 3—figure supplement 3 . A fully labeled ML phylogeny for Mononegavirales-like viruses . The phylogeny is reconstructed using RdRp alignments . Statistical support from the aLRT is shown on each node of the tree . The names of the viruses discovered in this study are shown in red . The names of reference sequences , which contain both the GenBank accession number and the virus species name , are shown in black . The names of previously defined genera/families are shown to the right of the phylogenies . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 01210 . 7554/eLife . 05378 . 013Figure 4 . The unrooted ML phylogeny based on RdRp showing the topological position of segmented viruses within the genetic diversity of negative-sense RNA viruses . The segmented viruses are labeled with segment numbers and shaded red . The unsegmented viruses are shaded green . The Chuviridae , which exhibit a wide variety of genome organizations , are shaded cyan . Three major types of putative chuvirus genomes ( circular , circular and segmented , and linear ) are shown in the right panel and are annotated with predicted ORFs: putative RdRp genes are shaded blue , putative glycoprotein genes are shaded orange , and the remaining ORFs are shaded gray . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 013 A key result of this study is that much of the genetic diversity of negative-sense RNA viruses in vertebrates and plants now appears to be contained within viruses that utilize arthropods as hosts or vectors . Indeed , it is striking that all vertebrate-specific segmented and unsegmented viruses ( arenavirus , bornavirus , filovirus , hantavirus , influenza viruses , lyssavirus , and paramyxovirus ) fall within the genetic diversity of arthropod-associated viruses ( Figures 3 , 5 ) . Also nested with arthropod-associated diversity were plant viruses ( emaravirus , tospovirus , tenuiviruses , nucleorhabdovirus , cytorhabdovirus , and varicosavirus ) ( Figures 3 , 5 ) . Surprisingly , our phylogeny similarly placed two non-arthropod invertebrate viruses , found in nematodes ( Heterodera glycines ) and flatworms ( Procotyla fluviatilis ) , within arthropod-associated diversity ( Figure 3C , Figure 3—figure supplement 2 ) , indicating that the role of non-arthropod invertebrates should be explored further . Finally , it was striking that although individual arthropod species can harbor a rich diversity of RNA viruses , many viruses seemed to be associated with different arthropod species that share the same ecological niche ( Tables 2–4 ) . Interestingly , host species in the same niche had similar viral contents that were generally incongruent with the host phylogeny ( Figure 6 ) . Such a pattern is indicative of frequent cross-species and occasional cross-genus virus transmission in the context of ecological and geographic proximity . 10 . 7554/eLife . 05378 . 014Figure 5 . The unrooted ML phylogeny of negative-sense RNA viruses ( RdRp ) with the common names of the principle arthropod hosts analyzed in this study indicated . Vertebrate-specific viruses are shaded red , those infecting both vertebrates and arthropods ( or with unknown vectors ) are shaded yellow , those infecting both plants and arthropods are shaded green , those infecting non-arthropod invertebrates are shaded blue , and the remainder ( arthropod only ) are shaded black . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 01410 . 7554/eLife . 05378 . 015Figure 6 . Phylogenetic congruence between viruses ( M segments ) and hosts . The comparisons include ( A ) Wuhan Horsefly Virus , ( B ) Wuhan Fly Virus 1 , ( C ) Wuhan Mosquito Virus 2 , and ( D ) Wuhan Mosquito Virus 1 . Different host species/genera are distinguished with different colors , which are then mapped onto virus phylogeny to assess the phylogenetic congruence . ML phylogenetic trees were inferred in all cases . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 015 The diversity of genome structures in these virus data was also striking . This can easily be documented with respect to the evolution of genome segmentation . The number of genome segments in negative-sense RNA viruses varies from one to eight . Our phylogenetic analysis revealed no particular trend for this number to increase or decrease through evolutionary time ( Figure 4 ) . Hence , genome segmentation ( i . e . , genomes with >1 segment ) has clearly evolved on multiple occasions within the negative-sense RNA viruses ( Figure 4 ) , such that it is a relatively flexible genetic trait . Although most segmented viruses were distantly related to those with a single segment ( Figure 4 ) , close phylogenetic ties were seen in other cases supporting the relatively recent evolution of multiple segments , with the plant-infecting varicosavirus ( two segments ) and orchid fleck virus ( bipartite ) serving as informative examples . In this context , it is notable that the newly discovered chuviruses fell ‘between’ the phylogenetic diversity of segmented and the unsegmented viruses . Although monophyletic , the chuviruses display a wide variety of genome organizations including unsegmented , bi-segmented , and a circular form , each of which appeared multiple times in the phylogeny ( Figures 4 , 7 ) . The circular genomic form , which was confirmed by ‘around-the-genome’ RT-PCR and by the mapping of sequencing reads to the genome ( Figure 7C ) , is a unique feature of the Chuviridae and can be distinguished from a pseudo-circular structure seen in some other negative-sense RNA viruses including the family Bunyaviridae and the family Orthomyxoviridae . Furthermore , this circular genomic form was also present in both segments of the segmented chuviruses ( Figure 7B ) . In addition , the chuviruses displayed a diverse number and arrangement of predicted open reading frames that were markedly different from the genomic arrangement seen in the order Mononegavirales even though these viruses are relatively closely related ( Figures 4 , 7 ) . In particular , the chuviruses had unique and variable orders of genes: the linear chuvirus genomes began with the glycoprotein ( G ) gene , followed by the nucleoprotein ( N ) gene , and then the polymerase ( L ) gene , whereas the majority of circular chuviruses were most likely arranged in the order L- ( G ) -N ( i . e . , if displayed in a linear form ) as the only low coverage point throughout the genome lay between the 5′ end of N gene and the 3′ end of L gene ( Figure 7B ) . In addition , the genome organizations of the chuviruses were far more concise than those of the order Mononegavirales , with ORFs encoding only 2–3 major ( >20 kDa ) proteins ( Figure 7 ) , and hence showing more similarity to segmented viruses in this respect . 10 . 7554/eLife . 05378 . 016Figure 7 . The differing genome organizations in the Chuviridae . ( A ) ML trees of three main putative proteins conserved among the chuviruses . Viruses with circular genomes ( Type I ) are shaded blue , while those with segmented genomes ( Type II ) are shaded red . ( B ) Structures of all complete chuvirus genomes . Circular genomes are indicated with the arrow ( blue ) situated at the 3′ end , and the genome is drawn in a linear form for ease of comparison only , being broken at the region of variable sequence ( refer to the ‘Materials and methods’ ) . ( C ) An example showing mapping of sequencing reads to the circular chuvirus genome . The template for mapping contains two genomes connected head-to-tail . The two boxes magnify the genomic region containing abundant sequence variation . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 016 Although our phylogenetic analysis focused on the relatively conserved RdRp , in the case of segmented viruses we searched for other putative viral proteins from the assembled contigs . Accordingly , we were able to find the segments encoding matching structural proteins ( mainly glycoproteins and nucleoproteins ) for many of the viral RdRp sequences ( Figure 8 ) , although extensive sequence divergence prevented this in some cases . Surprisingly , M segments were apparently absent in a group of tick phleboviruses whose RdRps and nucleoproteins showed relatively high sequence similarity to Uukuniemi virus ( genus Phlebovirus; Table 3 and Figure 8 ) . Genomes with missing glycoprotein genes were also found in the chuviruses ( Changping Tick Viruses 3 and 5 , Wuhan Louse Viruses 6 and 7 , Figure 7 ) and the unsegmented dimarhabdovirus ( Taishun Tick Virus , Wuhan Tick Virus 1 , Tacheng Tick Virus 6 , Figure 9 ) . Although it is possible that the glycoprotein gene may have been replaced with a highly divergent or even non-homologous sequence , we failed to find any candidate G proteins within the no-Blastx-hit set of sequences under the following criteria: ( i ) structural resemblance to G proteins , ( ii ) similar level of abundance to the corresponding RdRp and nucleoprotein genes , and ( iii ) comparable phylogenies or levels of divergence ( among related viruses ) to those of RdRps and nucleoproteins . The cause and biological significance of these seemingly ‘incomplete’ virus genomes require further study . Finally , it was also of interest that a virus with four segments was discovered in the horsefly pool . Although the predicted proteins of all four segments showed sequence homology to their counterparts in Tenuivirus ( Falk and Tsai , 1998 ) , this virus lacked the ambisense coding strategy of tenuiviruses ( Figure 10 ) . While the capability of this virus to infect plants is unknown , it is possible that it represents a transitional form between plant-infecting and arthropod-specific viruses . 10 . 7554/eLife . 05378 . 017Figure 8 . Genome structures of segmented negative-sense RNA viruses . Predicted viral proteins homologous to known viral proteins are shown and colored according to their putative functions . The numbers below each ORF box give the predicted molecular mass . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 01710 . 7554/eLife . 05378 . 018Figure 9 . Genome structures of unsegmented negative-sense RNA viruses . Predicted ORFs encoding viral proteins with >10 kDa molecular mass are shown and colored according to their putative functions . The numbers below each ORF box give the predicted molecular mass . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 01810 . 7554/eLife . 05378 . 019Figure 10 . Comparison of the genome structure of a potential tenui-like virus from horsefly with a prototype tenuivirus ( Rice grassy stunt virus ) genome . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 019 As well as novel exogenous RNA viruses , our metagenomic analysis also revealed a large number of potential EVEs ( Katzourakis and Gifford , 2010 ) in more than 40 arthropod species; these resembled complete or partial genes of the major proteins—the nucleoprotein , glycoprotein , and RdRp—but without fully intact genomes ( Table 5 ) . As expected given their endogenous status , most of these sequences have disrupted reading frames and many are found within transposon elements , suggesting that transposons have been central to their integration . Interestingly , in some cases , such as the putative glycoprotein gene of chuviruses , the homologous EVEs from within a family ( Culicidae ) or even an order ( Hymenoptera ) form monophyletic groups ( Figure 11 ) . However , they are unlikely to be orthologous because they do not share homologous integration sites in the host genome as determined by an analysis of flanking sequences , which in turn limited the applicability of molecular-clock based dating techniques . Furthermore , phylogenetic analyses of those EVEs shared among different host species revealed extremely complex tree topologies which do not exhibit simple matches to the host phylogeny at both the species and genera levels ( Figure 11B–C ) . In sum , these results suggest that EVEs are relative commonplace in arthropod genomes and have been often generated by multiple and independent integration events . 10 . 7554/eLife . 05378 . 020Table 5 . Summary of Endogenous Virus Elements ( EVEs ) determined hereDOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 020Host classificationHost nameVirus classificationGene ( s ) presentChelicerataIxodes scapularisChuvirusG , NDimarhabdovirusRdRp , NNairovirus likeNPhlebovirusRdRp , NQuaranjavirusRdRpTetranychus urticaeDimarhabdovirusNCrustaceaDaphnia pulexPhlebovirus likeRdRpEurytemora affinisChuvirusGDimarhabdovirusRdRp , NHyalella aztecaChuvirusG , NUnclassified mononegavirus 3RdRp , NLepeophtheirus salmonisPhlebovirus likeN , GInsecta: ColeopteraDendroctonus ponderosaeChuvirusGPhasmavirusG , NTribolium castaneumChuvirusGInsecta: DipteraAedes aegyptiChuvirusRdRpDimarhabdovirusRdRp , NPhasmavirusGPhlebovirus likeNQuaranjavirusRdRpAnopheles spp . ChuvirusGDimarhabdovirusRdRp , NPhasmavirusG , NPhlebovirus likeNQuaranjavirusRdRpCulex quinquefasciatusChuvirusG , NDimarhabdovirusNDrosophila spp . DimarhabdovirusRdRp , NPhasmavirusNUnclassified rhabdovirus 2RdRp , NInsecta: IsopteraZootermopsis nevadensisChuvirusNInsecta: HemipteraAcyrthosiphon pisumChuvirusG , NDimarhabdovirusNPhlebovirus likeNQuaranjavirusRdRpUnclassified mononegavirus 1RdRp , NRhodnius prolixusChuvirusGPhasmavirusGInsecta: HymenopteraAtta cephalotesUnclassified mononegavirus 2RdRpAcromyrmex echinatiorChuvirusGUnclassified mononegavirus 2RdRpCamponotus floridanusChuvirusGUnclassified mononegavirus 1NUnclassified mononegavirus 3RdRpUnclassified rhabdovirus 2RdRpHarpegnathos saltatorChuvirusGLinepithema humileChuvirusGNasonia spp . ChuvirusGPogonomyrmex barbatusChuvirusGSolenopsis invictaChuvirusGUnclassified mononegavirus 1NUnclassified mononegavirus 3RdRp , NInsecta: LepidopteraBombyx moriChuvirusRdRp , GQuaranjavirusRdRpUnclassified rhabdovirus 2RdRpMelitaea cinxiaDimarhabdovirusNQuaranjavirusRdRpPlutella xylostellaDimarhabdovirusN , GSpodoptera frugiperdaPhlebovirus likeGMyriapodaStrigamia maritimaChuvirusNPhlebovirus likeG10 . 7554/eLife . 05378 . 021Figure 11 . ML phylogeny of EVEs . The phylogeny is based on the glycoprotein of chuviruses in the context of exogenous members of this family ( A ) , with subtrees magnified for ( B ) the Culicidae clade and ( C ) the Hymenoptera clade . The EVEs used in the phylogeny covered the complete or near complete length of the glycoprotein gene and are shown in red and labeled according to host taxonomy in the overall tree . For clarity , monophyletic groups are collapsed based on the host taxonomy . Only bootstrap values >70% are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 021
Our study suggests that arthropods are major reservoir hosts for many , if not all , of the negative-sense RNA viruses in vertebrates and plants , and hence have likely played a major role in their evolution . This is further supported by the high abundance of viral RNA in the arthropod transcriptome , as well as by the high frequencies of endogenous copies of these viruses in the arthropod genome , greatly expanding the known biodiversity of these genomic ‘fossils’ ( Katzourakis and Gifford , 2010; Cui and Holmes , 2012 ) . The often basal position of the arthropod viruses in our phylogenetic trees is also compatible with the idea that the negative-sense RNA viruses found in vertebrates and plants ultimately have their ancestry in arthropods , although this will only be confirmed with a far wider sample of virus biodiversity . The rich genetic and phylogenetic diversity of arthropod RNA viruses may in part reflect the enormous species number and diversity of arthropods , and that they sometimes live in large and very dense populations that provide abundant hosts to fuel virus transmission . Furthermore , arthropods are involved in almost all ecological guilds and actively interact with other eukaryotes , including animals , plants , and fungi , such that it is possible that they serve as both sources and sinks for viruses present in the environment . In addition , not only were diverse viruses present , but they were often highly abundant . For example , in the pool containing 12 individuals ( representing two species ) from the Gerridae ( Water striders ) collected at the same site , we identified at least five negative-sense RNA viruses whose TPM values are well above 100 , and where the viral RNA collectively made up more than 50% of the host total RNA ( rRNA excluded ) . Determining why arthropods are able to carry such a large viral diversity and at such frequencies clearly merits further investigation . The viruses discovered here also exhibited a huge variation in level of abundance . It is possible that this variation is in part due to the stage or severity of infection in individual viruses and may be significantly influenced by the process of pooling , since most of our libraries contain an uneven mixture of different host species or even genera . In addition , it is possible that some low abundance viruses may in fact be derived from other eukaryotic organisms present in the host sampled , such as undigested food or prey , gut micro flora , and parasites . Nevertheless , since the majority of the low abundance viruses appear in the same groups as the highly abundant ones in our phylogenetic analyses , these viruses are most likely associated with arthropods . Viral infections in vertebrates and plants can be divided into two main categories: ( i ) arthropod-dependent infections , in which there is spill-over to non-arthropods but where continued virus transmission still requires arthropods , and ( ii ) arthropod-independent infections , in which the virus has shifted its host range to circulate among vertebrates exclusively ( Figure 12 ) . The first category of infections is often associated with major vector-borne diseases ( Zhang et al . , 2011 , 2012 ) . Given the biodiversity of arthropod viruses documented here , it seems likely that arthropod-independent viruses were ultimately derived from arthropod-dependent infections , with subsequent adaptation to vertebrate-only transmission ( Figure 12 ) . 10 . 7554/eLife . 05378 . 022Figure 12 . Transmission of negative-sense RNA viruses in arthropods and non-arthropods . Three types of transmission cycle are shown: ( i ) those between arthropods and plants are shaded green; ( ii ) those between arthropods and vertebrates are shaded yellow; and ( iii ) those that are vertebrate-only are shaded red . Viruses associated with each transmission type are also indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 05378 . 022 One of the most notable discoveries was that of a novel family , the Chuviridae . The identification of this diverse virus family provides a new perspective on the evolutionary origins of segmented and unsegmented viruses . In particular , the chuviruses occupy a phylogenetic position that is in some sense ‘intermediate’ between the segmented and unsegmented negative-sense RNA viruses and display genomic features of both . Indeed , our phylogenetic analysis reveals that genome segmentation has evolved multiple times within the diversity of chuviruses ( Figure 7 ) , such that this trait appears to be more flexible than previously anticipated . In addition , the majority of the chuviruses possess circular genomes . To date , the only known circular RNA virus is ( hepatitis ) deltavirus , although this potentially originated from the human genome ( Salehi-Ashtiani et al . , 2006 ) and requires hepatitis B virus for successful replication . As such , the chuviruses may represent the first report of autonomously replicating circular RNA viruses , which opens up an important line of future research . Our results also provide insights into the evolution of genome segmentation . Within the bunya-arena-like viruses ( Figures 3C , 4 ) , the three-segment structure is the most common , with the viral polymerase , nucleoprotein , and surface glycoproteins present on different segments . Notably , our phylogenetic analysis seemingly revealed independent occurrences of both increasing ( Tenuivirus and Emaravirus ) and decreasing ( Arenavirus ) segment numbers from the three-segment form ( Figure 4 ) . Independent changes of genome segmentation numbers are also observed in the mononegavirales-like viruses ( Figure 4 ) and , more frequently , in the chuviruses ( Figure 7A ) . Consequently , the number of genome segments appears to be a relatively flexible trait at a broad evolutionary scale , although the functional relevance of these changes remains unclear . While the segmented viruses ( bunya-arenaviruses , orthomyxoviruses , and ophioviruses ) appear to be distinct from the largely unsegmented mononegavirales-like viruses in our phylogenetic analysis , this may be an artifact of under-sampling , especially given that only a tiny fraction of eukaryotes have been sampled to date . With a wider sample of eukaryotic viruses it will be possible to more accurately map changes in segment number onto phylogenetic trees and in so doing come to a more complete understanding of the patterns and determinants of the evolution of genome segmentation . In sum , our results highlight the remarkable diversity of arthropod viruses . Because arthropods interact with a wide range of organisms including vertebrate animals and plants , they can be seen as the direct or indirect source of many clinically or economically important viruses . The viral genetic and phenotypic diversity documented in arthropods here therefore provides a new perspective on fundamental questions of virus origins , diversity , host range , genome evolution , and disease emergence .
Between 2011 and 2013 we collected 70 species of arthropods from various locations in China ( Table 1 ) . Among these , ticks were either directly picked from wild and domestic animals or captured using a tick drag-flag method; mosquitoes were trapped by light-traps; common flies were captured by fly paper; horseflies were picked from infested cattle; bed bugs and cockroaches were trapped indoors; louse flies were plucked from the skin of bats; millipedes were picked up from the ground; spiders were collected from their webs; water striders were captured using hand nets from river surfaces; and crabs and shrimps were bought ( alive ) from local fisherman . In addition , three pools of mixed insect samples ( Table 1 ) were collected from a rural area adjacent to rice fields ( Insect Mix 1 ) , from a lakeside ( Insect Mix 3 ) , and from a mountainous area near Wuhan ( Insect Mix 4 ) . After brief species identification by experienced field biologists , these samples were immediately stored in liquid nitrogen and were later put on dry ice for shipment to our laboratory . The specimens were first grouped into several units ( Table 1 ) . Depending on the size of specimens , one unit could include from 1 to 20 individual arthropods belonging to the same species and sampling location . These units were first washed with phosphate-buffered saline ( PBS ) three times before homogenized with the Mixer mill MM400 ( Restsch , Germany ) . The resultant homogenates were then subjected to RNA extraction using TRIzol LS reagent ( Invitrogen , Carlsbad , CA ) . After obtaining the aqueous phase containing total RNA , we performed purification steps from the E . Z . N . A Total RNA Kit ( OMEGA , Portugal ) according to the manufacturer's instructions . The concentration and quality of final extractions were examined using a ND-1000 UV spectrophotometer ( Nanodrop , Wilmington , DE ) . Based on host types and/or geographic locations , these extractions were further merged into 16 pools for RNA-seq library construction and sequencing ( Table 1 ) . To verify the field species identification , we took a proportion of the homogenates from each specimen or specimen pool for genomic DNA extraction using E . Z . N . A . DNA/RNA Isolation Kit ( OMEGA ) . Two genes were used for host identification: the partial 18S rRNA gene ( ∼1100 nt ) which was amplified using primer pairs 18S#1 ( 5′-CTGGTGCCAGCGAGCCGCGGYAA-3′ ) and 18S#2RC ( 5′-TCCGTCAATTYCTTTAAGTT-3′ ) and partial COI gene ( ∼680 nt ) using primer pairs LCO1490 ( 5′-GGTCAACAAATCATAAAGATATTGG-3′ ) and HCO2198 ( 5′-TAAACTTCAGGGTGACCAAAAAATCA-3′ ) . PCRs were performed as described previously ( Folmer et al . , 1994; Machida and Knowlton , 2012 ) . For taxonomic determination , the resulting sequences were compared against the nt database as well as with all COI barcode records on the Barcode of Life Data Systems ( BOLD ) . Total RNA was subjected to a slightly modified RNA-seq library preparation protocol from that provided by Illumina . Briefly , following DNase I digestion , total RNA was subjected to an rRNA removal step using Ribo-Zero Magnetic Gold Kit ( Epicentre , Madison , WI ) . The remaining RNA was then fragmented , reverse-transcribed , ends repaired , dA-tailed , adaptor ligated , purified , and quantified with Agilent 2100 Bioanalyzer and ABI StepOnePlus Real-Time PCR System . Pair-end ( 90 bp or 100 bp ) sequencing of the RNA library was performed on the HiSeq 2000 platform ( Illumina , San diego , CA ) . All library preparation and sequencing steps were performed by BGI Tech ( Shenzhen , China ) . The resulting sequencing reads were quality trimmed and assembled de novo using the Trinity program ( Grabherr et al . , 2011 ) . All sequence reads generated in this study were uploaded onto NCBI Sequence Read Achieve ( SRA ) database under the BioProject accession SRP051790 . The assembled contigs were translated and compared ( using Blastx ) to reference protein sequences of all negative-sense RNA viruses . Sequences yielding e-values larger than 1e−5 were retained and compared to the entire nr database to exclude non-viral sequences . The resulting viral sequences were merged by identifying unassembled overlaps between neighboring contigs or within a scaffold using the SeqMan program implemented in the Lasergene software package v7 . 1 ( DNAstar , Madison , WI ) . To prevent missing highly divergent viruses , the newly found viral sequences were included in the reference protein sequences for a second round of Blastx . For each potential viral sequence , we first used nested RT-PCR to examine which unit contained the target sequence , utilizing primers designed based on the deep-sequencing results . In the case of segmented viruses this information was also used to determine whether and which of the segments recovered from the pool belonged to the same virus . We next designed overlapping primers to verify the sequence obtained from the deep sequencing and assembly processes . Based on the verified sequences , we determined the sequencing depth and coverage by mapping reads to target sequences using bowtie2 ( Langmead and Salzberg , 2012 ) . All virus genome sequences generated in this study have been deposited in the GenBank database under accession numbers KM817593–KM817764 . Before quantification , we first removed the rRNA reads from the data sets to prevent any bias due to the unequal efficiency of rRNA removal steps during library preparation . To achieve this , we blasted the Trinity assembly results against the SILVER rRNA database ( Quast et al . , 2013 ) and then used the resulting rRNA contigs as a template for mapping using BOWTIE2 ( Langmead and Salzberg , 2012 ) . The remaining reads from each library were then mapped on to the assembled transcripts and analyzed with RSEM ( Li et al . , 2010 ) , using the run_RSEM_align_n_estimate . pl scripts implemented in the Trinity program ( Grabherr et al . , 2011 ) . The relative abundance of each transcript is presented as transcripts per million ( TPM ) which corrects for the total number of reads as well as for transcript length ( Li et al . , 2010 ) . Some of the sequences obtained were substantially shorter than expected . To obtain longer sequences , we used a Genome walking kit ( TaKaRa , Japan ) . Briefly , three gene-specific primers close to the end of the known sequence were designed . RNA from positive samples was used as input for reverse transcription primed by random primer N6 . TAIL-PCR ( thermal asymmetric interlaced PCR ) was performed according to the manufacturer's protocol . The cDNA was used as a template for PCR with specific primers and the manufacturer-supplied degenerate primers . After three rounds of amplification , the products were analyzed on 1 . 0% agarose gels , and single fragments were recovered from the gels and purified using an agarose gel DNA extraction kit ( TaKaRa ) . The purified products were then ligated into pMD19-T vector ( TaKaRa ) which contains the gene for ampicillin resistance . The vector was transformed into DH5α cells , which were spread on agar plates and incubated overnight at 37°C . A total of 10 clones were randomly selected and sequenced using M13 primers on ABI 3730 genetic analyzer ( Applied Biosystems , Carlsbad , CA ) . The extreme 5′ sequences were recovered by performing a 5′-Full RACE kit with TAP ( TaKaRa ) according to the manufacturer's protocol . Briefly , two gene-specific primers close to the end of the known sequence were designed . The 5′ end of RNA was ligated to the 5′RACE adaptor ( without 5′ end dephosphorylating and decapping ) and then reverse-transcribed using random 9 mers . The resulting cDNA was used as a template for nested PCR with 5′ RACE primers provided by the kit and gene-specific reverse primers . The PCR products were separated on an agarose gel , cloned into pMD19-T cloning vector , and subsequently sequenced . The extreme 3′ sequences were recovered by performing a 3′-full RACE Core Set with PrimeScript RTase ( TaKaRa ) according to the manufacturer's protocols . Because the RNA template lacks a polyadenylated tail , a Poly ( A ) Tailing Kit ( Applied Biosystems ) was used to add this to the RNAs prior to first-strand 3′-cDNA synthesis . 20 μl of the Poly ( A ) -tailing reaction mixture was prepared according to the manufacturer's instructions and was incubated at 37°C for 1 hr before reverse transcription using PrimeScript Reverse Transcriptase . The cDNA was then amplified by nested PCR using the 3′ RACE primers provided by the kit and gene-specific reverse primers . The PCR products were separated on agarose gels , cloned into pMD19-T cloning vector , and subsequently sequenced . The 5′ and 3′ ends of the genome fragment were also determined by RNA circularization . RT-PCR amplification was performed across the ligated termini and the resulting PCR products were subsequently cloned and sequenced . Potential viral proteins identified from this study were aligned with their corresponding homologs of reference negative-sense RNA viruses using MAFFT version 7 and employing the E-INS-i algorithm ( Katoh and Standley , 2013 ) . The sequence alignment was limited to conserved domains , with ambiguously aligned regions removed using TrimAl ( Capella-Gutierrez et al . , 2009 ) . The final alignment lengths were 224 amino acids ( aa ) , 412aa , 727aa , and 364aa for data sets of overall , bunya-arena-like , mononega-like , and orthomyxo-like data sets , respectively . Phylogenetic trees were inferred using the maximum likelihood method ( ML ) implemented in PhyML version 3 . 0 ( Guindon and Gascuel , 2003 ) , with the WAG + Γ amino acid substitution model and a Subtree Pruning and Regrafting ( SPR ) topology searching algorithm . Phylogenetic trees were also inferred using a Bayesian method implemented in MrBayes version 3 . 2 . 2 ( Ronquist and Huelsenbeck , 2003 ) , with the same substitution model as used in ML tree inference . In the MrBayes analyses , we used two simultaneous runs of Markov chain Monte Carlo sampling , and the runs were terminated upon convergence ( standard deviation of the split frequencies <0 . 01 ) . The phylogeny was subsequently summarized from both runs with an initial 10% of trees discarded as burn-in . For each of the putative viral protein sequences , we used TMHMM v2 . 0 ( http://www . cbs . dtu . dk/services/TMHMM/ ) to predict the transmembrane domains , SignalP v4 . 0 ( http://www . cbs . dtu . dk/services/SignalP/ ) to determine signal sequences , and NetNGlyc v1 . 0 ( http://www . cbs . dtu . dk/services/NetNGlyc/ ) to identify N-linked glycosylation sites . For some of the highly divergent viruses belonging to the Mononegavirales and the Chuviridae , a protein was regarded as a potential glycoprotein if it contained ( i ) a N-terminal signal domain , ( ii ) a C-terminal transmembrane domain , and ( iii ) glycosylation sites in cytoplasmic domains . Endogenous copies of the exogenous negative-sense RNA viruses newly described here were detected using the tBlastn algorithm against arthropod genomes available in the Reference Genomic Sequences Database ( refseq_genomic ) and Whole Genome Shotgun Database ( WGS ) in GenBank , and using viral amino acid sequences as queries . The threshold for match was set to 1e−05 for the e-value and 50 amino acids for matched length . The query process was reversed for each potential endogenous virus to determine their corresponding phylogenetic group . Orthologous insertion events were determined by examining flanking gene sequences . Sequence alignment and phylogenetic analyses were carried out as described above . Within the Chuviridae , Wuhan Louse Fly Virus 6 and 7 , Wenzhou Crab Virus 2 , Lishi Spider Virus 1 , and Wuchang Cockroach Virus 3 possessed bi-segmented genomes . Both segments were discovered using Blastx against pools of predicted proteins from unsegmented chuvirus or mononegavirales sequences . To determine that these sequences were indeed from separate segments , we performed all combinations of head-to-tail RT-PCR which allowed us to ascertain whether the sequence fragments came from a single genome . Furthermore , checking sequencing depth can help to eliminate the possibility of separate contigs being generated due to inadequate sequencing coverage . To prove that a pair of segments belonged to the same virus , we checked: ( i ) sequencing depth for both segments , ( ii ) the presence of conserved regulatory sequences at non-coding regions of the genome , ( iii ) whether there is match for PCR-positive units , and ( iv ) the phylogenetic positions of the different viral proteins ( Figure 7A ) . The circular genome organization within the Chuviridae was identified after we found that their genome sequences were ‘over assembled’ ( i . e . , generating contigs that contained more than one genome connected head-to-tail ) . This circular genomic form was also observed in both segments of the segmented chuviruses ( Figure 7B ) . In addition , RT-PCR and sequencing over the entire genome did not reveal any break-points . As a control , the same protocol failed to connect the genome termini within the Mononegavirales , suggesting the circular genomic form is unique to the chuviruses . To further validate that these genomes are circular , we mapped the high-throughput sequencing reads to these assembled genomes . The coverage and depth were adequate throughout the genome with the exception of one location upstream to the 3′ end of the ORF encoding RdRp ( Figure 7C ) . This genomic location had only 0–20 X coverage depending on the virus , although all RT-PCRs were successful across this location . Interestingly , sequencing of the cloned PCR products revealed extensive sequence variation ( i . e . , insertions and deletions ) ( Figure 7C ) , which is the likely cause of the low sequence coverage in this location . Collectively , these data provide strong evidence for circular genomes in the chuviruses , although this does not exclude the potential presence of linear genomic forms . | Many illnesses , including influenza , hemorrhagic fever , and rabies , are caused by a group of viruses called negative-sense RNA viruses . The genetic information—or genome—of these viruses is encoded in strands of RNA that must be copied before they can be translated into the proteins needed to build new viruses . It is currently known that there are at least eight different families of these viruses , which have a wide range of shapes and sizes and arrange their RNA in different ways . Insects , spiders , and other arthropods carry many different RNA viruses . Many of these viruses have not previously been studied , and those that have been studied so far are mainly those that cause diseases in humans and other vertebrates . Researchers therefore only know a limited amount about the diversity of the negative-sense RNA viruses that arthropods harbor and how these viruses evolved . Studying how viruses evolve helps scientists to understand what makes some viruses deadly and others harmless and can also help develop treatments or vaccines for the diseases caused by the viruses . Li , Shi , Tian , Lin , Kang et al . collected 70 species of insects , spiders , centipedes , and other arthropods in China and sequenced all the negative-sense RNA viruses in the creatures . This revealed an enormous number of negative-sense RNA viruses , including 112 new viruses . Many of the newly discovered arthropod viruses appear to be the ancestors of disease-causing viruses , including influenza viruses and the filoviruses—the group that includes the Ebola virus . Indeed , it appears that arthropods host many—if not all—of the negative-sense RNA viruses that cause disease in vertebrates and plants . While documenting the new RNA viruses and how they are related to each other , Li et al . found many different genome structures . Some genomes were segmented , which may play an important role in evolution as segments can be easily swapped to create new genetic combinations . Non-segmented and circular genomes were also found . This genetic diversity suggests that arthropods are likely to have played a key role in the evolution of new viruses by acting as a site where many different viruses can interact and exchange genetic information . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2015 | Unprecedented genomic diversity of RNA viruses in arthropods reveals the ancestry of negative-sense RNA viruses |
Mechanisms of muscle atrophy are complex and their understanding might help finding therapeutic solutions for pathologies such as amyotrophic lateral sclerosis ( ALS ) . We meta-analyzed transcriptomic experiments of muscles of ALS patients and mouse models , uncovering a p53 deregulation as common denominator . We then characterized the induction of several p53 family members ( p53 , p63 , p73 ) and a correlation between the levels of p53 family target genes and the severity of muscle atrophy in ALS patients and mice . In particular , we observed increased p63 protein levels in the fibers of atrophic muscles via denervation-dependent and -independent mechanisms . At a functional level , we demonstrated that TAp63 and p53 transactivate the promoter and increased the expression of Trim63 ( MuRF1 ) , an effector of muscle atrophy . Altogether , these results suggest a novel function for p63 as a contributor to muscular atrophic processes via the regulation of multiple genes , including the muscle atrophy gene Trim63 .
Muscle atrophy is associated with aging , cancer , AIDS and neurodegenerative diseases such as amyotrophic lateral sclerosis ( ALS ) ( von Haehling et al . , 2010 ) . Although muscle atrophy is not necessarily the primary target of the pathology , it is often an important cause of lethality . For example , atrophy and dysfunction of respiratory muscles lead to death in ALS patients ( Rothstein , 2009 ) . As muscle atrophy is associated with complex pathologies , the exact mechanisms inducing muscle atrophy are varied and still debated . Typically , ALS has been considered a neurodegenerative pathology specifically causing alteration in motor neurons , but more recent findings indicate that the etiology of the pathology is more complex . Indeed , a number of additional cell types , such as astrocytes ( Yamanaka et al . , 2008 ) , microglia ( Boillée et al . , 2006 ) and muscle cells ( Wong and Martin , 2010 ) , have been described to be directly affected by the pathology and therefore to participate in the muscle atrophy . Around 20% of all inherited ALS cases can be linked to mutations in the gene encoding SOD1 . Cellular events ( Pansarasa et al . , 2014 ) that have been shown to be triggered by these different mutations include aggregation of SOD1 proteins in the cytoplasm ( Hart , 2006 ) , increase in oxidative stress ( Barber and Shaw , 2010 ) and subsequent DNA damage ( Aguirre et al . , 2005 ) , endoplasmic reticulum ( ER ) stress ( Nishitoh et al . , 2008 ) or alterations of mitochondrial function ( Manfredi and Xu , 2005 ) . In addition , novel mutated genes ( FUS , TARDBP… ) have been linked to ALS with differences in the pathophysiological outcomes ( Chen et al . , 2013 ) . These differences might be linked to the different impacts of the mutated proteins at the molecular level . Indeed , protein aggregates or other alterations induced by SOD1 mutants have been characterized in muscle cells , while other mutated proteins linked to ALS seem to not directly affect muscles ( Pansarasa et al . , 2014 ) . To date , the exact molecular mechanisms driving muscle catabolism in the symptomatic phase of ALS remain poorly understood . It is also only during the symptomatic phase that ALS pathology can be diagnosed . The absence of pre-symptomatic markers highlights the need for understanding the muscle catabolic processes for therapeutic purposes . Several observations indicate that the p53 family members ( p53 , p63 , p73 ) play an important role in muscle physiopathology and might therefore represent actors of the muscle atrophy ( Schwarzkopf et al . , 2006; Mazzaro et al . , 1999; Cam et al . , 2006; Fontemaggi et al . , 2001; Martin et al . , 2011; Rouleau et al . , 2011; Belloni et al . , 2006; Su et al . , 2012 ) . The p53 family of transcription factors is a central regulator of cellular processes such as apoptosis , cell cycle arrest , metabolism or cellular differentiation through the regulation of several target genes ( CDKN1A , BAX , GADD45A , MDM2 and others ) ( Arrowsmith , 1999; Menendez et al . , 2009 ) . All three members encode TA and △N isoforms that vary in their N-terminus due to alternate promoter usage where TA has a canonical transactivation domain . ∆N isoforms lack such a domain and can serve as dominant negatives versus the TA isoforms in some cases , although they are also capable of transactivating certain genes ( De Laurenzi et al . , 1998; Casciano et al . , 1999; Murray-Zmijewski et al . , 2006 ) . Through their cellular activities , p53 proteins are involved in a broad variety of physiological functions that include tumor suppression and organ development ( Arrowsmith , 1999 ) . For example , p53 plays a role in the response against tumor-inducing events such as DNA damage , oncogene activation , and a variety of additional cellular stresses ( hypoxia , reactive oxygen species ( ROS ) , or alteration of energy metabolism ) ( Marcel et al . , 2011; Rufini et al . , 2013; Gonfloni et al . , 2014 ) . In addition , several studies have highlighted the involvement of the p53 family members in neurodegenerative diseases . p53 as well as p63 and p73 have been shown to regulate neuronal apoptosis and their activation has been observed in various neurodegenerative diseases , such as Alzheimer , Parkinson and Angelman syndromes ( Jiang et al . , 1998; de la Monte et al . , 1997; Seidl et al . , 1999; Bui et al . , 2009; Benosman et al . , 2007; Benosman et al . , 2011 ) . We have previously reported an induction of p53 in degenerating spinal cord motor neurons in an ALS mouse model expressing mutated Cu/Zn superoxide dismutase 1 ( SOD1[G86R] ) ( González de Aguilar et al . , 2000 ) . In muscles , p53 is activated during myogenic differentiation , participates with MyoD to induce myogenesis , and mediates doxorubicin-induced muscle atrophy via its target gene pw1 ( Schwarzkopf et al . , 2006; Mazzaro et al . , 1999 ) . Nonetheless , p53 expression is not essential for muscle development ( Donehower et al . , 1992 ) or regeneration ( White et al . , 2002 ) , which could be explained by compensatory mechanisms involving p63 and p73 . Indeed , more recent studies have shown that p63 and p73 are also involved in myoblast differentiation ( Cam et al . , 2006; Fontemaggi et al . , 2001; Martin et al . , 2011; Rouleau et al . , 2011 ) and ΔNp73 appears to protect muscle cells against stresses ( Belloni et al . , 2006 ) . Finally , a study showed that p63 is important for the regulation of muscle cell metabolism via the regulation of Sirtuins and AMPK ( Su et al . , 2012 ) . In this study , we investigated the regulation and the role of the transcription factors of the p53 family in muscular atrophy during ALS based on a meta-analysis we performed with 4 microarray experiments obtained with biopsies of muscles from ALS patients or with muscles from ALS mouse models .
To identify the molecular mechanisms involved in muscle atrophy during ALS we performed a meta-analysis using four independent microarray experiments deposited at the Array Express database ( EMBL-EBI ) . Two experiments contained gene expression data for the muscle of ALS patients and control individuals ( E-MEXP-3260; E-GEOD-41414 , [Pradat et al . , 2012; Bernardini et al . , 2013] ) . One experiment contained gene expression data for muscles of SOD1 ( G86R ) mice that represents an ALS model in which the onset of the pathology is at 105 days of age ( E-TABM-195 [Gonzalez de Aguilar et al . , 2008] ) . The last experiment contained gene expression data for muscles of SOD1 ( G93A ) mice in which onset of the pathology occurs at 14 weeks of age ( E-GEOD-16361 , [Capitanio et al . , 2012] ) . Beside the better pathophysiological relevance , data obtained from biopsies of ALS patients also provided a better representation of the diversity of the genetic anomalies observed in patients . In addition , patients were at various stage of the pathology , hence establishing a representative scale of muscle alterations . The panel of datasets we chose also included two different mouse models of ALS , allowing us to pinpoint common and specific deregulations . Importantly , the SOD1 mouse models are well characterized for their muscular phenotype alterations . In particular , it has already been established that SOD1 mutants present altered functions in muscles , in contrast to other mutated proteins linked to ALS ( TARDBP , FUS etc ) ( Pansarasa et al . , 2014 ) . After standard normalization and statistical analyses , each experiment was independently subjected to gene ontology , signaling pathway , transcription factor , and miRNA analyses . Fold induction between control individuals and ALS individuals was set to twofold change and rawp value inferior to 0 . 05 . We decided to focus on transcription factor deregulations . The bioinformatic analyses we performed pinpointed to only 7 transcription factors whose activity , indicated by coherent changes in expression of their target genes , was potentially deregulated in at least two out of four experiments ( Figure 1A ) . The activity of one transcription factor , NfKB , appeared deregulated only in experiments done with the mouse models . Deregulation of STAT1 activity was identified in three experiments . Interestingly , the activity of only three transcription factors , MyoD , Myogenin and p53 , was identified to be commonly deregulated in all four experiments that included biopsies from patients and animal models . MyoD and Myogenin are muscle specific transcription factors involved in muscle cell differentiation ( Zanou and Gailly , 2013 ) . P53 was the transcription factor with the highest number of deregulated genes ( 51 genes ) . Notably the p53 target genes CDKN1A , GADD45A and PMAIP1 , among others , were found induced in all four experiments . 10 . 7554/eLife . 10528 . 003Figure 1 . Microarray meta-analysis highlights links between the deregulation of p53 family related genes and ALS . ( A ) Representation of the number of deregulated target genes of the indicated transcription factors . Data were obtained using the indicated datasets from the Array Express database ( EMBL-EBI ) and quantification was carried out from AltAnalyze software analyses on transcription factor databanks ( complete data in Supplementary file 2A , B , C , D . ( B ) mRNA levels from nine ALS patient deltoid muscles as by DNA microarray were correlated with the intensity of muscle injury . Expression data were generated using a murine gene profiling database deposited at ebi . ac . uk/arrayexpress ( accession number E-MEXP-3260 ) . In the corresponding study , muscle injury was estimated according to a composite score combining manual testing of strength of shoulder abductors and the degree of myofiber atrophy . This score ranges from 6 ( normal strength and very low level of atrophy ) to 1 ( total paralysis and high level of atrophy ) . Each point represents an individual . Correlation coefficients ( r ) and p-values were determined by Spearman correlation test . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 00310 . 7554/eLife . 10528 . 004Figure 1—figure supplement 1 . Regulation of p53-family related genes in skeletal muscle of SOD1 ( G86R ) and denervated mice . Data were generated using a gene expression database deposited at ebi . ac . uk/arrayexpress ( accession number E-TABM-195 ) . Gastrocnemius muscle samples from male SOD1 ( G86R ) mice with no symptoms ( Healthy , at 75 days of age ) , altered hind limb extension reflexes ( Preparalysis , at 90 days of age ) , and at the onset of hind limb paralysis ( symptomatic mice , at about 105 days of age ) were analyzed by DNA microarray . Denervated muscles were obtained from wild-type mice after 7 days of sciatic nerve axotomy . Non-transgenic male littermates served as controls . 3–4 animals were pooled per group , and each condition was done in duplicate . Values are expressed as means of normalized expression levels . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 00410 . 7554/eLife . 10528 . 005Figure 1—figure supplement 2 . mRNA levels from control and ALS patient deltoid muscles as by DNA microarray were correlated with the intensity of muscle injury . Expression data were generated using a murine gene profiling database deposited at ebi . ac . uk/arrayexpress ( accession number E-MEXP-3260 ) . In the corresponding study , muscle injury was estimated according to a composite score combining manual testing of strength of shoulder abductors and the degree of myofiber atrophy . This score ranges from 6 ( normal strength and very low level of atrophy ) to 1 ( total paralysis and high level of atrophy ) . Each point represents an individual . In this experiment , patients with high level of atrophy ( L , score 1–3 ) and low degree of atrophy ( E , score 4–6 ) were grouped . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 00510 . 7554/eLife . 10528 . 006Table 1 . Fold induction of p53-related genes in the ALS model SOD1 ( G86R ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 006Gene nameFunction90 d . 105 d . p53-family target genesCdkn1a ( p21 ) Cell cycle arrest413Gadd45aCell cycle arrest5 , 621Peg3Apoptosis inducing37PerpCell cycle arrest412Pmaip1Apoptosis effector512BaxApoptosis effector38SivaApoptosis inducing35Zmat3Growth regulation1 , 61 , 1Eda2RNF . Kb/JNK pathway3 , 49 , 4TigarGlucose metabolism0 , 750 , 2Sens1ROS homeostasis-16 , 3Sens2ROS homeostasis1 , 271 , 46Sco2Glucose metabolism1 , 180 , 91Ddit3 ( Chop ) ER stress1 , 140 , 35Bip ( Grp78 ) ER stress1 , 251 , 08Xbp1ER stress22 , 51p53-family regulatorsMlf1Cell cycle arrest/differentiation0 , 90 , 2Myf6Differentiation47Mdm2p53 degradation46Txn1Oxidative stress response46Id2Inhibition of differentiation23 , 1p53-family membersP5343TAp63412∆Np630 , 50 , 3TAp7323∆Np730 , 90 , 8Denervation/atrophy markersChrna1 ( ACh Receptor alpha ) Neuromuscular junction4 , 212 , 4 As one of the experiments using biopsies of ALS patients included a scale ( von Haehling et al . , 2010; Rothstein , 2009; Yamanaka et al . , 2008; Boillée et al . , 2006; Wong and Martin , 2010; Pansarasa et al . , 2014 ) of muscle alteration , we analyzed whether the expression of some of these genes might correlate with the severity of the pathology . We found that CDKNA1 , GADD45A and PMAIP1 expression correlated with the degree of the pathology of the muscle from ALS patients ( Figure 1B ) . Besides the bioinformatic analysis on the deregulation of transcription factors , the signaling pathway analyses also indicated alterations in the p53 pathway characterized by deregulations in upstream regulators of p53 , such as MDM2 and thioredoxin , and a p53 family member , P63 ( Table 1 , Figure 1B , Figure 1—figure supplement 1 ) . In particular , the expression of P63 correlated with the severity of the pathology in muscles biopsies from ALS patients ( Figure 1B ) . In order to validate the bioinformatic analyses we performed RT-qPCR experiments with RNA from muscle biopsies of an independent group of ALS patients . We confirmed that CDKN1A , GADD45A and PMAIP1 were induced in the muscle biopsies of ALS patients ( Figure 2A , B , C ) . Similarly , we analyzed the expression of these genes using muscle samples of independent groups of SOD1 ( G86R ) mice . Groups analyzed at 60 days and 75 days of age correspond to the asymptomatic stage , while 90 day-old groups correspond to an early or pre-symptomatic stage associated with established gene deregulations ( von Grabowiecki et al . , 2015 ) . Finally , the symptomatic stage group ( beginning after 105 days ) is characterized by the onset of paralysis and marked muscle atrophy ( Figure 2—figure supplement 1 Upregulation of the p53 target genes Gadd45a , Cdkn1a , Bax , Pmaip1 and Perp was observed at 90 days and further increased at 105 days in SOD1 ( G86R ) mice ( Figure 2D , E , F , Figure 2—figure supplement 2 ) . In addition to these genes , we also analyzed by RT-qPCR the expression of additional p53 target genes and regulators of p53 by RT-qPCR ( Table 1 ) . In particular , the expression of p53 target genes involved in apoptosis ( Pmaip1 , Peg3 and Siva ) was also induced . 10 . 7554/eLife . 10528 . 007Figure 2 . p53-family target gene expression in muscles from ALS patients and in an ALS mouse model correlates with disease intensity . ( A–C ) RNA from muscle biopsies of control and ALS patients ( n = 8 , Neuromuscular Unit [BioBank of Skeletal Muscle , Nerve Tissue , DNA and cell lines] ) was extracted and analyzed by RT-qPCR . Absolute levels are normalized against the average of the control group . ( D–F ) p53 family target genes mRNA levels were assayed in SOD1 ( G86R ) mouse gastrocnemius muscle by RT-qPCR . Graphs are means of fold induction versus 60 days-old WT and of matching age ( 60 , 75 , 90 , 105-days-old , n = 6 ) and experimental condition ( wild-type , or SOD1 ( G86R ) ) . *p<0 . 01 compared to control , as calculated by a one-way ANOVA test followed by a Tukey post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 00710 . 7554/eLife . 10528 . 008Figure 2—figure supplement 1 . Gastrocnemius muscles from wild-type or symptomatic SOD1 ( G86R ) ( 105 days ) mice were dissected and weighted . Graph represents the weight ( n = 5 ) . *p<0 . 01 compared to control , as calculated by a one-way ANOVA test followed by a Tukey post-test . NS: non denervated , S: denerveted as assessed by acetylcholine alpha receptor ( AchRα ) expression . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 008 p53 proteins have recently been linked to energy metabolism and endoplasmic reticulum ( ER ) stress pathway activation ( Su et al . , 2012; Ramadan et al . , 2005; Zhu and Prives , 2009 ) . Analysis of the expression of p53 family target genes implicated in several metabolic pathways ( Tigar , sestrins , Sco2 , Sirtuin1 or Prkaa1 ) ( Su et al . , 2012; Vousden and Ryan , 2009 ) or ER stress ( Chop , Bip , or Xbp1 ) ( Stavridi and Halazonetis , 2004 ) , did not reveal coherent regulation in respect to disease progression ( Table 1 ) . For example , the expressions of Sesn2 and Tigar ( Vousden and Ryan , 2009 ) were regulated in opposite directions during the progression of the disease . Therefore , our data suggest that the correlation between ALS progression and p53 function might mostly be due to cell growth arrest and cell death regulation . We also confirmed by RT-qPCR an upregulation of several upstream regulators of the p53 family , including Mdm2 , Myf6 , Mlf1 , and Txn ( Table 1 ) ( Arrowsmith , 1999 ) . Taken together , our results suggest that a p53-like pathway is activated in ALS muscles both in patients and the murine SOD1 ALS-models . As we observed in the muscle biopsies of ALS patients a correlation between p63 expression and the severity of the pathology , we investigated the expression levels of p53 family members in the muscles of SOD1 ( G86R ) mice . Our analysis revealed an increased expression of TA isoforms of Trp63 in SOD1 ( G86R ) ( Figure 3 ) . Strikingly , the mRNA levels of TA isoforms of Trp63 were strongly induced towards the end of the disease ( 105 day ) , while the mRNA levels for ΔN isoforms of Trp63 were downregulated during the same time period . A similar tendency was observed for p53 , TA and ∆N isoforms of p73 , albeit at a lower magnitude . The expression of TA isoforms of Trp63 correlated with acetylcholine receptor alpha ( Chrna1 ) expression , a molecular marker indicating the severity of muscular denervation . In addition , we analyzed the expression of two documented effectors of muscular atrophy , namely Fbxo32 ( Atrogin-1 ) and Trim63 ( MuRF1 ) . These proteins are E3 ubiquitin ligases that target muscular proteins for degradation during muscular atrophy or remodeling ( Murton et al . , 2008 ) . Importantly , the deregulation of Trp63 expression also correlated with the upregulation of these two markers . This is in accordance with our data from ALS patient muscle biopsies , whereby the expression of P63 also correlated with the degree of muscle pathology ( Figure 1E ) . 10 . 7554/eLife . 10528 . 009Figure 3 . Expression of p53-family members in SOD1 ( G86R ) muscles . p53 family members , Chrna1 ( Acetylcholine receptor subunit alpha ) or muscle atrophy effectors Trim63 ( MuRF1 ) and Fbxo32 ( Atrogin1 ) mRNA levels were assayed in SOD1 ( G86R ) mouse gastrocnemius muscle by RT-qPCR . Bars are means of fold induction versus ‘WT 60 days-old’ and of matching age ( 60 , 75 , 90 , 105 days-old , n = 6 ) and experimental condition ( WT or SOD1 ( G86R ) ) . *p<0 . 01 compared to control , as calculated by a one-way ANOVA test followed by a Tukey post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 009 Based on the observed deregulation of P63 expression in ALS patients and the stronger upregulation of TAp63 in SOD1 ( G86R ) mice , we further analyzed p63 protein levels . Immunoblotting with a TAp63 isoforms specific antibody revealed a striking accumulation of p63 proteins in muscles of SOD1 ( G86R ) mice that correlated with the progression of the disease ( Figure 4A ) . When probing with a ΔNp63 specific antibody , however , we did not observe any specific band . The use of a p63 antibody directed against all p63 isoforms confirmed an upregulation of p63 in muscles of SOD1 ( G86R ) mice ( Figure 4—figure supplement 1 ) . Immunohistochemistry with the same antibody also revealed markedly increased immunoreactivity in the nuclei of muscle fibers of SOD1 ( G86R ) ( Figure 4B , Figure 4—figure supplement 3 ) . In contrast , there was no significant increase in p73 staining ( Figure 4—figure supplement 2 ) . In this case , the apparent higher number of p73 positive nuclei appeared to be due to the atrophy of the muscle fibers , increasing the density of cells/nuclei . Similar experiments to detect expression of p53 did not yield a specific staining . However , we observed by western blot some slight increase in p53 protein levels in protein extract of muscle from SOD1 ( G86R ) mice ( Figure 4—figure supplement 1 ) . Taken together , our data indicated a complex regulation of p53 family members during muscular atrophy , highlighted by significant increase of TAp63 messenger and protein expression levels in the skeletal muscles during ALS . 10 . 7554/eLife . 10528 . 010Figure 4 . p63 protein expression in SOD1 ( G86R ) muscle . ( A ) Proteins from muscles were immuno-precipitated with a p63 antibody and then separated on a 10% SDS PAGE gel . Western blot experiment was performed using an antibody against TAp63 . Each experimental point is a pool of proteins from 6 animals . Graph represents quantification of the blot using ImageJ image analyzer software indicated a %/WT 60 day-old animals . ( B ) Gastrocnemius muscles from wild-type or symptomatic SOD1 ( G86R ) ( 105 days ) mice were cryodissected and probed for total p63 protein . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 01010 . 7554/eLife . 10528 . 011Figure 4—figure supplement 1 . p53 and p63 protein expression in muscles of SOD1 ( G86R ) mice . ( A ) Proteins from muscles were immuno-precipitated with a p63 antibody and then separated on a 10% SDS PAGE gel . Western blot experiment was performed using an antibody against p63 total . Shows pools of proteins from 3 animals at 105d . TBP was used as loading control . ( B ) Proteins ( 40 µg ) from muscles were separated on 10% SDS PAGE gel . Western blot probing was performed with p53 antibody ( IC12 , 1/2000 , Cell Signaling , Danvers , MA ) and True Blot ( Rockland Immunochemicals , Pottstown , PA ) secondary antibody avoiding Ig heavy chain recognition . Tubilin was used as loading control . Graph below shows% of induction relative to the mean of p53 expression level in WT animals normalised with tubulin . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 01110 . 7554/eLife . 10528 . 012Figure 4—figure supplement 2 . Gastrocnemius muscles from wild-type or symptomatic SOD1 ( G86R ) ( 105 days ) mice were cryodissected and probed for total p73 protein . Graph represents the number of fibers per surface unit as indicated ( n = 5 ) . *p<0 . 01 compared to control , as calculated by a one-way ANOVA test followed by a Tukey post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 01210 . 7554/eLife . 10528 . 013Figure 4—figure supplement 3 . Gastrocnemius muscles from wild-type or symptomatic SOD1 ( G86R ) ( 105 days ) mice were cryodissected and probed for total p63 protein and nuclei ( Hoechst ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 013 We then further investigated the possible cause of the deregulation of Trp63 expression . Several studies showed that the ALS etiology is complex and multifactorial , involving different cell types and molecular mechanisms . One established cause of muscular atrophy is motor neuron degeneration that leads to muscle denervation . However , it has also been shown that SOD1 mutants can also directly cause alteration in muscle cells such as SOD1 protein aggregates and mitochondrial abnormalities ( Pansarasa et al . , 2014 ) . To verify the first hypothesis , we induced denervation in 80 day-old wild type and SOD1 ( G86R ) mice by sciatic nerve crush , and gastrocnemius muscles were analyzed 7 days later . Our results showed that denervation upregulated TAp63 mRNA levels five- to sixfold in wild-type mice ( Figure 5A ) . Concomitantly , ΔNp63 levels were downregulated 0 . 4-fold ( Figure 5B ) . In SOD1 ( G86R ) mice , nerve crush further accentuated changes in mRNA levels for TA and ∆N isoforms of Trp63 . In addition , the TAp63 target genes Cdkn1a and Gadd45a were found strongly induced after nerve crush ( Figure 5C , D ) . These results show that nerve injury leading to alteration of the motor axis seems to be sufficient to activate a TAp63 response . 10 . 7554/eLife . 10528 . 014Figure 5 . Expression of p63 and p53-family target genes following sciatic nerve crush , SOD1 expression of induction of stress ( A–D ) WT and SOD1 ( G86R ) mice ( 80 days of age ) were anesthetized and the sciatic nerve crushed . Sham-operated contra limbs served as control ( Ct ) . After 7 days , expression of TA isoforms of Trp63 ( A , TAp63 ) , ∆N isoforms of Trp63 ( B , ∆Np63 ) , Gadd45a ( C ) and Cdkn1a ( D ) was assayed by RT-qPCR ( n = 6 ) . Values were normalized to the value of sham-operated WT muscles/animals . Bars represent means ( relative induction versus Ct ) with standard deviation ( n = 3 ) . *p<0 . 01 as calculated by a one-way ANOVA test followed by a Tukey post-test . ( E ) C2C12 myoblasts were transfected with expression vectors for SOD1 variants ( WT or G86R ) . mRNA from SOD1 transfected cells were analyzed by RT-qPCR for p63 and p63 target gene expression . Bars represent means ( relative induction versus Ct ) with standard deviation ( n = 3 ) . *p<0 . 01 as calculated by a one-way ANOVA test followed by a Tukey post-test . ( F ) Proteins were extracted from C2C12 myoblasts treated with compounds: FCCP , Tunicamycin ( Tun ) , Etoposide ( Eto ) , menadione ( Men ) . Western blot analysis revealed TAp63 expression . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 01410 . 7554/eLife . 10528 . 015Figure 5—figure supplement 1 . Regulation of p63 and Mdm2 expression by SOD1 ( G86R ) . ( A ) Protein were extracted from C2C12 myoblasts expressing WT or SOD1 ( G86R ) after 5-days puromycin selection . Western blot analysis revealed TAp63 , Bax or SOD1 expression . Actin was used as loading control . ( B , C ) C2C12 myoblasts were transfected with expression vectors for SOD1 variants ( WT or G86R ) or TAp63γ ( 2 concentrations , 1 , 2 ) and luciferase reporter genes containing deletions of the promoter of the ΔN isoforms of P63 ( -1584-+32 or -46/+32 ) or Mdm2 promoter . Bars represent means ( relative induction versus Ct ) with standard deviation ( n = 3 ) . Results are standardized with the 'minimal' promoter reporter gene -46/+32-luc . Ct = cells transfected with an empty vector . Bars represent means ( relative induction versus Ct ) with standard deviation ( n = 3 ) . *p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 01510 . 7554/eLife . 10528 . 016Figure 5—figure supplement 2 . Functional interaction between members of the p53 family and ER or mitochondrial stress . ( A ) Quantification: Proteins were extracted from C2C12 myoblasts treated with compounds: FCCP , Tunicamycin ( Tun ) , Etoposide ( Eto ) , menadione ( Men ) . Western blot analysis revealed TAp63 expression . Bars correspond to means with SD ( n = 3 ) . *p<0 . 01 . ( B ) C2C12 myoblasts were transfected with expression vectors encoding transcription factors involved in the ER or mitochondrial stress pathway ( CHOP , ATF6 , ATF4 , XBP1s ) . RNA levels for TA isoforms of Trp63 , TA isoforms of P73 and P53 were followed by RT-qPCR . Bars represent means ( relative induction versus Ct ) with standard deviation ( n = 3 ) . *p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 016 Although it remains a challenge to reproduce in vitro the long-term development of ALS , we tried to assess the effect of the mutated SOD1 on muscle cells via an overexpression of SOD1 ( G86R ) in the mouse myoblast cell line C2C12 . Several target genes of the p53-family ( Bax , Cdkn1a , Gadd45a ) were induced upon overexpression of SOD1 ( G86R ) ( Figure 5E ) . Similarly TAp63 expression was increased at the mRNA level and the protein level ( Figure 5E and Figure 5—figure supplement 1 ) . In contrast , the mRNA levels as well as the promoter activity of △N isoforms of P63 were downregulated ( Figure 5E , Figure 5—figure supplement 1 ) ( Romano et al . , 2006 ) . However , we were not able to confirm this result on △Np63 at the protein level . We also tested for a possible cross-regulation of △Np63 expression by the increased expression of TAp63 observed in ALS . We observed that TAp63 represses the promoter activity of the △N isoforms of P63 , while it expectedly induces the Mdm2 promoter ( Figure 5—figure supplement 1 ) . These results demonstrated that expression of SOD1 ( G86R ) was sufficient to trigger a p53-like response similar to our in vivo observations in atrophic muscle tissues . We then investigated whether TAp63 could be induced by different stresses related to the cellular damages caused by SOD1 mutants . We used pharmacological inductors for oxidative stress ( menadione ) ( Barber and Shaw , 2010 ) , DNA damage ( etoposide ) ( Aguirre et al . , 2005 ) , mitochondrial deregulation ( FCCP ) ( Manfredi and Xu , 2005 ) and ER Stress ( tunicamycin ) ( Hart , 2006; Nishitoh et al . , 2008 ) . Treated cells revealed an increase of TAp63 upon the four stresses ( Figure 5F , Figure 5—figure supplement 2 ) . Mitochondrial and ER stress triggers specific signaling pathways that involve a complex network of transcription factors such as ATF4 , ATF6 , XBP1 and CHOP ( Senft and Ronai , 2015 ) . Interestingly , overexpression of ATF4 and ATF6 induces the RNA level for TAp63 and TAp73 respectively , but not p53 ( Figure 5—figure supplement 2 ) . This result indicated that upregulation of TAp63 expression might be involved in the muscle cell response to diverse stresses including stresses related to SOD1 mutants . As the expression profile of the TA isoforms of Trp63 correlated with the expression of the muscle atrophy effectors Fbxo32 and Trim63 ( Figure 3 ) , we hypothesized that TAp63 could regulate them directly . Bioinformatic analyses revealed the presence of several putative p63-binding sites in the promoter of Trim63 ( Figure 6A ) . Therefore , we tested whether TAp63 could regulate Trim63 expression . Indeed , TAp63 overexpression in C2C12 cells strongly induced Trim63 mRNA levels ( Figure 6B ) . Fbxo32 expression level was much less affected ( data not shown ) . Note that under this condition p53 or p73 had less effect on Trim63 expression ( Figure 6B ) . Under the same experimental conditions , other p63 target genes , Cdkn1a , Pmaip1 , Casp1 and Prkaa1 were less induced ( Figure 6C ) . 10 . 7554/eLife . 10528 . 017Figure 6 . Effects of p53-family expression on Trim63 and p53-family target genes . ( A ) Schematic representation of the Trim63 promoter indicating the location of putative p53/p63 binding sites . ( B , C ) C2C12 myoblasts were transfected ( inserted panel: western blot ) with various p53-family members ( TAp63γ , ΔNp63γ , p53 , TAp73β , ΔNp73β ) . Total C2C12 RNA was subjected to RT-qPCR after 10 hr or 24 hr of transfection and Trim63 ( B ) or p63 target ( C , Cdkn1a , Pmaip1 , Casp1 , Prkaa1 ) expressions are shown relative to control-transfected cells . Bars are means of fold induction versus the control ( Ct ) with SD ( n=3 ) . *p<0 . 01 as calculated by a one-way ANOVA test followed by a Tukey post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 017 To further characterize the regulation of Trim63 by TAp63 , we used luciferease reporter constructs containing progressive deletions of the Trim63 promoter . We found that p53 family members induced Trim63 promoter reporters that contained at least the fragment -500 bp to -1000 bp ( Figure 7A and B ) ( Waddell et al . , 2008 ) . Interestingly , that fragment contains potential p63 binding sites with high probability scores , such as RE1/2 ( -660/-690 bp ) . We then assessed the capacity of p63 to bind the Trim63 promoter on binding sites that have high probability scores . Chromatin immunoprecipitation experiments ( ChIP ) covering RE1/2 and RE4 binding sites showed that TAp63 proteins bound preferentially onto RE1/2 ( Figure 7C ) . Similarly , ChIP experiments indicated that p73 and p53 bound to RE1/2 ( Figure 7—figure supplement 1 ) . However , p73 seemed also to bind RE4 . 10 . 7554/eLife . 10528 . 018Figure 7 . Regulation of Trim63 promoter by p63 . ( A , B ) Trim63 promoter reporter constructs were co-transfected with pCDNA3 ( Ct ) or TAp63 into C2C12 cells and luciferase activity was assessed 16 hr later . pGL3 was used as a negative control . Bars correspond to means with SD ( n = 3 ) . *p<0 . 01 as calculated by a one-way ANOVA test followed by a Tukey post-test . ( C ) Chromatin immunoprecipitation assay was performed on the Trim63 promoter using RT-qPCR on RE1/2 and RE4 ( see Figure 6A ) . Bars correspond to means with SD ( n = 3 ) . *p<0 . 01 as calculated by a one-way ANOVA test followed by a Tukey post-test . ( D ) Trim63 mRNA levels were assayed in C2C12 cells by RT-qPCR after TAp63 silencing by siRNA for 36 hr and after treatment with FCCP for 12 hr . Bars correspond to means with SD ( n=3 ) . *p<0 . 01 as calculated by a one-way ANOVA test followed by a Tukey post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 01810 . 7554/eLife . 10528 . 019Figure 7—figure supplement 1 . Regulation of Trim63 by p53 and p73 proteins . ( A ) mRNA levels of Trim63 in C2C12 cells following transfection with siRNA control and siRNA directed against p73 , p53 and a mix of siRNA against P53 , and the TA isoforms of Trp63 and P73 ( siMIX ) . Bars represent means ( relative induction versus Ct ) with standard deviation ( n = 3 ) . *p<0 . 01 . ( B ) mRNA level for TA isoforms of Trp63 , TA isoforms of P73 and P53 in C2C12 cells following transfection with siRNA control and siRNA directed against p63 , p73 , and p53 . Bars represent means ( relative induction versus Ct ) with standard deviation ( n = 3 ) . *p<0 . 01 . ( C , D ) Chromatin immunoprecipitation ( ChIP ) assay was performed on the Trim63 promoter using RT-qPCR on RE1/2 and RE4 . p53 immunoprecipitation ( C ) was performed using p53 antibody IC12 ( Cell Signalling ) , p73 immunoprecipitation was performed using p73 antibody IMG-259a ( Imgenex ) . Bars correspond to means with SD ( n = 3 ) . *p<0 . 01 as calculated by a one-way ANOVA test followed by a Tukey post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 01910 . 7554/eLife . 10528 . 020Figure 7—figure supplement 2 . Impact of p63 on C2C12 cell survival . ( A , B ) C2C12 were transfected with a GFP expression vector and either TAp63γ or ΔNp63γ expression vectors . After 24 hr , cells were left untreated ( Ct ) or treated with FCCP ( 1 µM ) or menadione ( 1 µM ) for 24 hr . Cells were stained with Hoechst and examined with a fluorescence microscope ( B ) . Above , GFP-positive control cells ( untreated ) . Below , dead GFP-positive cell treated with FCCP . C2C12 cells were grown on coverslips coated with poly-ornithine in 24-wells plates . Cells were co-transfected with the indicated expression vectors ( 200 ng/well ) and a GFP-expression vector ( 50 ng/well ) as previously described ( Broadley and Hartl , 2008 ) . Cells were cultured for 18 hr with the indicated agents . Cells were subsequently washed with PBS and fixed with 4% paraformaldehyde for 15 min . After two washes , cells were incubated for 10 min with the Hoechst 33 , 342 staining agent ( 1 µg/ml , Sigma , Germany ) . GFP positive cells were then observed with an epi-fluorescent microscope ( Zeiss , Germany ) to assess the nucleus morphology . ( C ) C2C12 cells were transfected either with the ∆Np63γ expression vector or siRNA directed against the TA isoforms of Trp63 . Cell survival was evaluated using MTT assay after 48 hr of treatment with the indicated drugs at 1 µM . *p<0 . 01 compared to control , as calculated by a one-way ANOVA test followed by a Tukey post-test . DOI: http://dx . doi . org/10 . 7554/eLife . 10528 . 020 To assess the physiological importance of TAp63 in Trim63 expression we used TAp63-specific silencing RNA ( siRNA ) . Transfection in C2C12 cells of TAp63siRNA diminished the expression of TAp63 at the protein and mRNA levels ( Figure 7—figure supplement 1 ) . TAp63 silencing or overexpression of ∆Np63 had a partial protective effect on C2C12 ( Figure 7—figure supplement 2 ) . Importantly , silencing of TAp63 reduced Trim63 mRNA levels in both basal state and following stress induced by FCCP ( Figure 7D ) . SiRNA against p53 also diminished Trim63 RNA level , while siRNA against p73 had not significant effect ( Figure 7—figure supplement 2 ) . The combination of siRNA against TA isoforms of Trp63 , TA isoforms of P73 and P53 diminished further Trim63 RNA level up to ~50% , but did not abolish it . Taken together , these results indicate a complex regulation of the Trim63 promoter , in which the direct binding of p63 and p53 correlates with the modification of gene expression in C2C12 muscular cells .
Our results demonstrate that there is a complex p53-like response developed by the atrophic muscle during ALS progression . This assertion is first based on the bioinformatic signalling pathway analyses of 4 independent microarray experiments performed on muscle biopsies of ALS patients as well as two different mouse models of ALS . These analyses pointed out a deregulation of p53 as one of the only three transcription factors deregulated in all four experiments , and common between mouse and human patient samples . Moreover , detailed expression profile analyses of several p53 target genes ( Cdkn1a , Gadd45a , Pmaip1 ) or the p53 family member , P63 , showed that their expression correlated significantly with the severity of the pathology in humans . The signalling pathway analyses were confirmed with groups of individuals independent of those used for the microarrays and by additional experimental methods . RT-qPCR confirmed the induction of multiple target genes of the p53 family . In addition , expression analysis of p53 family members indicated that in ALS the P63 gene seems more likely to play a regulatory role as the TAp63 isoforms are strongly upregulated and localized in the nuclei of the fibers during the ALS pathology ( Figures 1 , 3 , 4 , 5 , figure supplement 2 , 3 , 4 , 7 ) . Our observation that the deregulation of p63 and p63 target genes occurs in muscle of ALS patients that have not been selected for a particular genetic alteration indicates that these deregulations are likely to be a common feature in ALS , independently of whether it is SOD1 that is mutated or another gene . Additional experiments using other mouse models for TDP43 or FUS might confirm that . It was previously reported that p63 participates in muscle cell differentiation and metabolism , and contributes to cardiac muscle development ( Cam et al . , 2006; Martin et al . , 2011; Rouleau et al . , 2011; Su et al . , 2012; Osada et al . , 1998 ) . Now , by combining biopsies from ALS patients and an animal model for ALS , the present study provides the first solid evidence that p63 might also participate in muscular atrophy . Although our results indicate that TAp63 is strongly induced in muscle atrophy during ALS , we cannot exclude the possible contribution of p53 and p73 proteins due to the fact that their mRNA expression is upregulated , although to a much weaker extent than TAp63 ( Figure 3—figure supplement 4 ) . In addition , we detected p53 and p73 protein expression in muscles tissues . Several studies support this possibility by showing that p53 and p73 play a role in muscle cell differentiation , cachexia and survival ( Schwarzkopf et al . , 2006; Cam et al . , 2006 , Soddu et al . , 1996; Tamir and Bengal , 1998; Porrello et al . , 2000; Weintraub et al . , 1991 ) . However , genetic inactivation of p53 does not affect ALS progression , muscle development or muscle regenerative capacity ( Donehower et al . , 1992; White et al . , 2002; Kuntz et al . , 2000; Prudlo et al . , 2000 ) . Nevertheless , our results suggest that the absence of p53 could be compensated by p63 or even p73 . The slight increased in p53 protein levels observed in protein extracts of muscle from SOD1 ( G86R ) mice might be caused by the production of ROS that stabilized p53 through post-translational modifications as previously described in other stresses ( Vurusaner et al . , 2012 ) . In addition , the slight increase in p53 RNA level we observed might also contribute . The causes of p63 regulation during ALS muscle atrophy seemed complex and reflect the debated etiology of the pathology ( Yamanaka et al . , 2008; Boillée et al . , 2006; Wong and Martin , 2010; Chen et al . , 2013 ) . For instance , we showed that P63 deregulation could have an intercellular origin represented by the loss of interaction between the muscle and the nerve cells , as provoked in the nerve crush experiment ( Figure 5 ) . Hence , the degeneration of the motor neurons that is characteristic of ALS would be sufficient to explain the increased expression of TAp63 and its target genes in ALS . However , we also observed that P63 deregulation can have an intrinsic origin resulting from expression of SOD1 ( G86R ) in muscle cells causing intracellular stresses ( Figure 5 ) . Indeed , ALS-associated SOD1 mutations have been shown to induce SOD1 protein aggregates and mitochondrial dysfunction in muscle cells ( Pansarasa et al . , 2014 ) . Interestingly , we observed that both activation of a protein aggregate stress response pathway or mitochondrial dysfunction could induce a TAp63 response . This finding is supported by a previous report showing that tp63 is an effector of the ER stress pathway in zebrafish allowing the regulation of the pro-apoptotic gene bbc3 ( puma ) ( Pyati et al . , 2011 ) . Protein aggregates , mitochondrial stress and oxidative stress triggered selective complex stress pathways named ER stress ( or UPR , unfolded response ) or mitochondrial stressed pathways that utilize several common transcription factors as effectors , such as ATF4 , ATF6 , CHOP and XBP1 ( Senft and Ronai , 2015; Broadley and Hartl , 2008; Lee , 2015; Michel et al . , 2015 ) . Therefore , we investigated whether these effectors could drive the expression of p53 family protein . We showed that some of these transcription factors , notably ATF4 and ATF6 , were able to induce the RNA levels of TA isoforms of Trp63 and P73 in C2C12 cells ( Figure 5—figure supplement 2 ) . However , bioinformatic analyses did not reveal potential canonical binding sites for these transcription factors neither in promoters of TA isoforms of P63 nor P73 , suggesting that the regulation might occur through indirect mechanisms that remain to be identified . Based on the literature , the p53 family could mediate different cellular outcomes in muscles and therefore on muscle pathology . p53/p63/p73 proteins have been linked to cell death , differentiation , metabolism , ER stress induction and ROS defence , which have all been reported during ALS ( Hart , 2006; Barber and Shaw , 2010; Aguirre et al . , 2005; Nishitoh et al . , 2008; Manfredi and Xu , 2005 ) . Our study revealed that the majority of the p53/p63/p73 target genes upregulated during ALS in the atrophic muscles are connected to cell death ( Gadd45a , Peg3 , Perp , Pmaip1 , Bax , Siva , Eda2r , Wig1/Pag608 ) ( Figure 1 and Table 1 ) . Genes connected to other functions , such as ER stress ( chop , bip , xbp1 , scotin ) or energy metabolism ( Tigar , Sesn1 , Sesn2 , Sco2 ) seem to be less consistently regulated , as some are upregulated ( Sesn1 , Sesn2 , Xbp1 ) , while others are downregulated ( Sco2 , Tigar , Chop , Bip , see Table 1 ) . Therefore , it seems more likely that p53 family members , notably TAp63 , function in ALS is connected to muscular atrophy via control of muscle cell survival and catabolism . This hypothesis is further supported by three of our results . First , the p53 family members TAp63 , p53 and TAp73 induce the muscle atrophy effector gene Trim63 ( MuRF1 ) , most likely via a direct binding to the Trim63 promoter ( Figure 7 , Figure 7—figure supplement 1 ) . Second , overexpression of TAp63 induces cell death in C2C12 myoblasts ( Figure 7—figure supplement 1 ) . Third , overexpression of △Np63 protects myoblastic cells against stresses ( Figure 7—figure supplement 2 ) . Although these results were obtained in a myoblastic cell line , they are consistent with numerous other studies describing the ability of p63 to control cell death in various pathophysiological conditions . To establish the exact pathophysiological importance of TAp63 upregulation in ALS represent a difficult challenge . Indeed , we already observed that in vitro the silencing of TAp63 with siRNA does not entirely abolish the expression of Trim63 ( Figure 7 ) and does not significantly reduce cell death induced by stresses ( Figure 7 , Figure 7—figure supplement 2 ) . This has several reasons . The first is that the expression of Trim63 involves several transcription factors , such as FOXO1 and the glucocorticoid receptors that certainly participate in the regulation of Trim63 during ALS . Indeed , the coordinated silencing of p53 , TAp63 and TAp73 did not completely abolish Trim63 RNA levels ( Figure 7—figure supplement 1 ) , supporting the involvement of other types of transcription factors . The second reason is intrinsic to the p53 family . Indeed , we already know that p53 , p63 and p73 have some redundant functions and target genes . We have also already established that p53 and p73 are expressed in muscles during the pathology ( Figure 4—figure supplements 1 , 2 ) and also bind to the Trim63 promoter ( Figure 7—figure supplement 1 ) . Therefore in absence of TAp63 , p53 and/or TAp73 might replace it in some conditions . For example , we observed an upregulation of TAp73 in C2C12 cells when TAp63 is silenced . This compensatory mechanism might therefore also explain why TAp63 siRNA do not protect C2C12 cells from death , in contrast to the expression of the △Np63 isoform that could inhibit p53 , TAp73 and TAp63 function altogether ( Figure 7—figure supplement 2 ) . ALS patients are currently diagnosed at a stage where denervation and muscular alterations already are established and , because of the lack of curative treatment , lead to death within 2 to 5 years . The results presented here suggest that p53 family members , via the regulation of selected target genes such as Eda2r , Peg3 but also , as we show , via Trim63 , might contribute to muscle catabolism in these patients . It remains to be established whether this signalling pathway is uniquely critical for muscular atrophy during ALS or whether it is common to other muscular atrophies occurring in pathologies such as cachexia , diabetes , and others .
SOD1 ( G86R ) mice were genotyped as described in ( Ripps et al . , 1995 ) . For surgery , 80-day-old FVB mice were anesthetized and both sciatic nerves were exposed at mid thigh level and crushing was performed ( or not – CT ) with a forceps during 20 s ~5 mm proximal to the trifurcation . Control animals used in the experiments were wild-type littermates . Randomization was performed based on body weight . Time course for animal pathology was performed based on a previous study on denervation and muscle atrophy ( von Grabowiecki et al . , 2015 ) . Animal experiments were performed following the European guidelines and protocols validated by the local ethical committee . C2C12 cells were obtained from ATCC ( ATCC CRL-1772 ) and grown in DMEM ( Dulbecco's modified Eagle's medium; Life Technology , Carlsbad , CA ) with 10% fetal bovine serum ( Life Technology ) at 37°C in a humidified atmosphere and 5% CO2 . Mycoplasma contamination has been tested negatively using PlasmoTest ( Invivogene , San Diego , CA ) . Differentiation of C2C12 cells was performed using 2% horse serum at 90% cell confluence . TRIzol ( Invitrogen , Carlsbad , CA ) was used to extract RNA . One µg of RNA was used for reverse transcription ( iScript cDNA kit , Bio-Rad , France ) and qPCR was carried out ( iQ SYBR Green , Bio-Rad ) ( Supplementary file 1 ) . Expression levels were normalized using either 18S , TBP or RPB1 as previously described ( Vidimar et al . , 2012 ) . Cells or tissue were lysed with LB ( 125 mM Tris-HCl pH 6 . 7 , NaCl 150 mM , NP40 0 . 5% , 10% glycerol ) . Proteins were denatured and deposited directly ( 75 μg of proteins ) onto a SDS-PAGE gel , or they were precipitated ( 2 mg of proteins ) with a p63 antibody and G Sepharose beads before separation . Western blotting was performed using antibodies raised against p53 ( rabbit anti-p53 , FL-393 , Santa Cruz Biotechnology , Dallas , TX ) , p63 ( mouse anti-p63 , 4A4 , Santa Cruz Biotechnology; p63 , Abcam , France ) or TAp63 ( Biolegend , CA ) . Secondary antibodies ( anti-rabbit , anti-mouse: Sigma , France ) were incubated at 1:1000 . Loading was controlled with actin ( rabbit anti-β-actin , Sigma , 1:4000 ) or TBP ( anti-TBP 1:1000 , Santa Cruz Biotechnology ) ( Antoine et al . , 1996 ) . Cells were transfected by polyethylenimmine ( PEI ) -based or JetPrim ( Polyplus , Strasbourg , France ) as previously described ( Gaiddon et al . , 1999 ) . For luciferase assays , cells were seeded in 24-well plates , and transfected with the indicated expression vectors ( 200 ng ) and reporter constructs ( 250 ng ) ( Sohm et al . , 1999 ) . Luciferase activity was measured in each well 24 hr later and results were normalized with a CMV-driven reporter gene ( Benosman et al . , 2011 ) . The -1584 ΔNp63 luc and -46 ΔNp63 luc constructs were previously described ( Romano et al . , 2006 ) . The Trim63 luc constructs were previously described ( Waddell et al . , 2008 ) . SiRNA tranfection was performed using 30 nM of siRNA and with RNAiMAX protocol as described by the provider ( Life Technology ) . TAp63 siRNA sequences were covering the sequence: GAA CUU UGU GGA UGA ACC UCC GAA . ChIP assays were performed using the standard protocol from the Magna ChIP G kit ( Millipore ) . C2C12 lysates were sonicated 12 times at 10% power . For each 1 million cells , 1 µg of antibody was used . p63 was immunoprecipitated with a mouse antibody raised against total p63 ( 4A4 , Santa Cruz Biotechnology ) . Mouse-anti-RAB11A was used as negative control ( Santa Cruz Biotechnology ) . ECL files from microarray experiments ( E-MXP-3260; E-GEOD-41414; E-TABM-195; E-GEOD-16361 ) were obtained form the Array Express database ( EMBL-EBI ) . Each experiment was first analyzed individually using AltAnalysis software ( Emig et al . , 2010 ) . Deregulated gene were identified based on two fold change expression and t-test p-value <0 . 05 . Deregulated genes were then analyzed by GO-Elite with Prune Ontology term using Z-score ( cutoff 1 . 96 , p-value 0 . 05 ) and Fisher's Exact Test for ORA ( 2000 permutations ) for over-representation in selected biological processes in several resources: Gene Ontology , MPhenoOntology , Disease Ontology , GOSlim , PathwayCommons , KEGG , Transcription Factor Targets , miRNA Targets , Domains , BioMarkers , RVista Transcription Sites , DrugBank , BioGrid . Mouse gastrocnemius muscles were sampled , submersed in freezing medium ( Tissue-Tek O . C . T compound , Sakura , Japan ) and immediately frozen in a nitrogen-cooled isopentane bath . Muscles were sliced in transversal axis at 14 µm in a cryostat ( Leica CM3050S , Leica , France ) and placed on slides covered with 0 . 5% gelatine . The samples were then dried for 20 min on a hot plate and fixed in 4% paraformaldehyde for 10 min . After a 5 min wash with PBS , the samples were permeabilized with 3% Triton X-100 in PBS for 10 min , washed with TBS , incubated in 100 mM glycine in TBS for 20 min and finally washed again in PBS . The samples were incubated with mouse antibody raised against p63 ( p63 clone 4A4 , Santa Cruz Biotechnology ) at 1:100 with 0 . 1% Triton X-100 in PBS ( Triton buffer ) overnight at room temperature . They were then washed three times with Triton buffer for 10 min and incubated with cyanine 3-coupled goat anti-mouse antibody ( Jackson ImmunoResearch , West Grove , PA ) at 1:1000 , as well as with 1 µg/ml Hoechst 33 , 342 ( Sigma , France ) , in Triton buffer at room temperature for 1 hr . After washing three times with Triton buffer , the slides were covered with mounting medium ( Aqua-Poly/Mount , Polysciences , Warrington , PA ) on glass slips and observed by confocal microscopy ( Zeiss , Germany ) . Antibody specificity was verified with slides probed with only the secondary antibody . | Many conditions , including cancer and AIDS , lead to a person’s muscles wasting away . Often the muscles are not necessarily the prime targets of these diseases . However , in the case of a neurodegenerative disease called amyotrophic laterals sclerosis ( or ALS ) , it is the muscle degeneration that ultimately leads to the death of patients within a few years . There is currently no treatment for ALS . This is partly because the mechanisms behind the disease are poorly understood . However , proteins belonging to the so-called p53 family have been implicated as possibly being involved in the degenerative processes in muscles . The p53 family comprises three proteins called p53 , p63 and p73 . These proteins bind to DNA to switch genes on or off , and target genes involved in a range of cellular processes such as DNA repair and cell death . von Grabowiecki et al . have now identified p63 as a protein that contributes to muscle wasting in ALS . The production of p63 in degenerated muscles increases as the disease progresses , both in patients and in a mouse model of the condition . Further analysis then showed that the protein p63 regulates an enzyme that actually breaks down the building blocks of muscles , which directly leads to the muscle wasting . Following on from this work , the next step is to verify if the p53 family are also involved in muscle wasting that arises from other diseases . This will enable researchers to look for similarities that could highlight new opportunities to treat conditions that cause muscle wasting . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2016 | Transcriptional activator TAp63 is upregulated in muscular atrophy during ALS and induces the pro-atrophic ubiquitin ligase Trim63 |
Polo-like kinase 1 ( PLK1 ) is a key cell cycle regulator implicated in the development of various cancers , including prostate cancer . However , the functions of PLK1 beyond cell cycle regulation remain poorly characterized . Here , we report that PLK1 overexpression in prostate epithelial cells triggers oncogenic transformation . It also results in dramatic transcriptional reprogramming of the cells , leading to epithelial-to-mesenchymal transition ( EMT ) and stimulation of cell migration and invasion . Consistently , PLK1 downregulation in metastatic prostate cancer cells enhances epithelial characteristics and inhibits cell motility . The signaling mechanisms underlying the observed cellular effects of PLK1 involve direct PLK1-dependent phosphorylation of CRAF with subsequent stimulation of the MEK1/2-ERK1/2-Fra1-ZEB1/2 signaling pathway . Our findings highlight novel non-canonical functions of PLK1 as a key regulator of EMT and cell motility in normal prostate epithelium and prostate cancer . This study also uncovers a previously unanticipated role of PLK1 as a potent activator of MAPK signaling .
Mammalian polo-like kinase 1 ( PLK1 ) is a serine/threonine kinase that plays key roles in the regulation of the cell cycle ( Barr et al . , 2004; Llamazares et al . , 1991 ) . It contains a conserved N-terminal kinase catalytic domain and a C-terminal polo-box domain ( PBD ) that is involved in substrate binding . PLK1 mediates almost every stage of cell division , including mitotic entry , centrosome maturation , bipolar spindle formation , sister chromatid segregation , mitotic exit , and cytokinesis execution ( Barr et al . , 2004 ) . In addition to its canonical role in mitosis and cytokinesis , recent studies suggest that PLK1 may have other important functions such as regulation of microtubule dynamics , DNA replication , chromosome dynamics , p53 activity , and recovery from the G2 DNA damage checkpoint ( Liu et al . , 2010; Song et al . , 2011 ) . PLK1 is overexpressed in a variety of human tumors and its expression level often correlates with increased cellular proliferation , enhanced metastatic potential , and poor prognosis in cancer patients ( Cholewa et al . , 2013; Takai et al . , 2005 ) . PLK1 is frequently ( >50% ) overexpressed in prostate cancer ( PCa ) , and PLK1 overexpression is linked to higher tumor grade ( Weichert et al . , 2004 ) , suggesting that PLK1 may play a pivotal role in PCa etiology . Constitutive expression of Plk1 in NIH/3T3 cells causes oncogenic foci formation and these transformed cells are tumorigenic in nude mice ( Smith et al . , 1997 ) . In contrast , depleting PLK1 in U2OS cells abrogates anchorage-independent growth ( Eckerdt et al . , 2005 ) . These results highlight PLK1 as a possible driver of oncogenic transformation , although it remains unclear if PLK1 itself is sufficient to induce tumor development . It has been suggested that PLK1 controls cancer development through multiple mechanisms that include canonical regulation of mitosis and cytokinesis , as well as modulation of DNA replication and cell survival ( Deeraksa et al . , 2013; Luo and Liu , 2012 ) . Importantly , previous studies reported that increased PLK1 expression levels positively correlate with the invasiveness of colorectal , breast , and thyroid tumors ( Han et al . , 2012; Rizki et al . , 2007; Zhang et al . , 2012 ) . These data imply a possible role for PLK1 in tumor invasion and metastasis; however , direct evidence supporting this hypothesis and mechanisms of the proinvasive activity of PLK1 during PCa progression are lacking . In this study , we investigated the roles of PLK1 in regulating the motility of prostate epithelial cells and PCa cells . Our data highlight PLK1 as a crucial positive regulator of different modes of cell migration . This pro-migratory activity of PLK is mediated by induction of the epithelial-to-mesenchymal transition ( EMT ) via activation of the CRAF/MEK/ERK/Fra1/ZEB1/2 signaling cascade .
It has been reported that PLK1 is frequently overexpressed in human PCa ( Weichert et al . , 2004 ) . To examine the expression level and activity status of PLK1 in a panel of PCa cell lines , we performed immunobloting analysis using antibodies that recognize total PLK1 or its active form , phosphorylated at Tyrosine 210 ( pT210 ) . Both the protein abundance and activity of PLK1 were elevated in PCa cell lines when compared to RWPE-1 cells ( immortalized normal prostate epithelial cells; Figure 1A ) , which is consistent with the PLK1 expression profile in PCa tissue specimens reported by another group ( Weichert et al . , 2004 ) . Moreover , PLK1 was differentially expressed and/or activated in PCa cells ( higher in the metastatic PCa cell lines [DU145 , C4-2B and PC3] and lower in the non-metastatic cell lines [LNCaP and LAPC4]; Figure 1A ) . 10 . 7554/eLife . 10734 . 003Figure 1 . Ectopic expression of PLK1 in RWPE-1 cells promotes cell motility . ( A ) Cell lysates were prepared from the indicated PCa cell lines and subjected to Western blots in order to detect the level and activity of PLK1 protein using anti−PLK1 and anti−PLK1 ( pT210 ) antibodies , respectively . β-actin was used as a loading control . ( B ) RWPE-1 cells were infected with lentivirus encoding Flag-PLK1 ( PLK1 ) or empty vector ( EV ) . The protein levels of PLK1 , AR , PSA , and β-actin were determined by Western blot . C4-2B cells with high endogenous PLK1 expression were included for comparison . ( C ) Control RWPE-1 and RWPE-1–PLK1 cells were subjected to a wound healing assay . The figure shows representative images as well as calculated percentage of wound closure during 48 hr of cell migration . Scale bar , 500 µm . ( D ) In vitro Matrigel invasion assay . The figure shows representative images of invaded cells and quantification of the relative number of cells that invaded over 48 hr . The data are presented as the mean ± s . e . m . *p<0 . 01 , two-tailed Student’s t-test . Scale bar , 100 µm . ( E–G ) Time-lapse video microscopy motility experiments to monitor random migration of control and PLK1-overexpressing RWPE-1 cells . The trajectories of individual cells of different experimental groups ( E ) , track distance ( F ) , and velocity of cell migration ( G ) are shown . Horizontal bars in F-G represent median and interquartile range . Each dot represents a single-cell measurement . Thirty cells per experimental group were measured . *p<0 . 01 , two-tailed Mann-Whitney rank sum tests . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 00310 . 7554/eLife . 10734 . 004Figure 1—figure supplement 1 . Androgen receptor ( AR ) mRNA is expressed in RWPE-1 cells . Total RNA from control RWPE-1 ( EV ) and RWPE-1–PLK1 ( PLK1 ) cells was subjected to real-time RT-PCR . AR mRNA levels were normalized to glyceraldehyde phosphate dehydrogenase ( GAPDH ) and are expressed as relative expression values . The data are presented as the mean ± s . e . m . *p<0 . 05 , two-tailed Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 00410 . 7554/eLife . 10734 . 005Figure 1—figure supplement 2 . Ectopic expression of PLK1 in RWPE-1 cells induces cellular transformation and tumorigenicity . ( A ) The effect of PLK1 upregulation in RWPE-1 cells on anchorage-independent growth in soft agar . RWPE-1–PLK1 ( PLK1 ) cells and vector control ( EV ) cells were plated in soft agar and grown for 21 days . The data are presented as the mean ± s . e . m . *p<0 . 01 , two-tailed Student’s t-test . ( B , C ) Control RWPE-1 or PLK1-overexpressing cells ( 1×106 ) were injected subcutaneously into the flanks of NSG mice . Seven mice were used per group . Tumor growth was observed for 5 weeks . ( B ) Images of primary tumors found in NSG mice subcutaneously engrafted with RWPE-1–PLK1 cells . The table below summarizes the tumor incidence in the 2 groups . ( C ) Representative images of H & E staining and PSA IHC of primary tumor and lung metastasis . Arrows point to lung micrometastases . Scale bar = 500 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 00510 . 7554/eLife . 10734 . 006Figure 1—figure supplement 3 . Ectopic expression of PLK1 in PrEC cells promotes cell migration and invasion . ( A ) Modulated PLK1 levels in PrEC cells . PrEC cells were infected with lentivirus encoding Flag-PLK1 ( PLK1 ) or control vector ( EV ) . PLK1 protein levels were determined by Western blot . ( B ) A migration assay was performed and a quantitative analysis is shown . ( C ) The in vitro invasiveness of those cells was measured using Matrigel-coated Transwell chambers . The data are presented as the mean ± s . e . m . *p<0 . 01 , two-tailed Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 006 In order to define the potential oncogenic role of PLK1 upregulation in PCa , we overexpressed PLK1 in RWPE-1 cells . RWPE-1 cells are derived from normal human prostate epithelial cells and are immortalized with human papillomavirus 18 E7 proteins ( Bello et al . , 1997 ) . In contrast to E6-immortalized cells , RWPE-1 cells express p53 and have a functional p53-dependent checkpoint ( Bello et al . , 1997; Roh et al . , 2008 ) . Furthermore , they express luminal cytokeratins and do not grow in soft agar or form tumors in nude mice . We detected low levels of androgen receptor ( AR ) expression in RWPE-1 cells by quantitative real-time RT-PCR and immunoblotting ( Figure 1—figure supplement 1 , Figure 1B ) . Therefore , RWPE-1 cells provide an excellent model for studying normal prostate epithelial functions , prostate epithelial transformation , and different stages of prostate carcinogenesis ( Bello et al . , 1997; Roh et al . , 2008 ) . RWPE-1 cells were infected with lentivirus containing human PLK1 cDNA . To avoid clonal variation , stable cell lines were established from a mixed population of multiple clones . Cells transduced with empty lentivirus served as a control . RWPE-1–PLK1 cells were designed to express PLK1 at a level comparable to asynchronous metastatic C4-2B PCa cells ( Figure 1B ) in order to determine the effect of cell cycle-independent PLK1 overexpression on PCa development . We examined the ability of PLK1-overexpressing cells to grow in soft agar , a property that frequently correlates with cell tumorigenicity . Under these conditions , control RWPE-1 cells formed few , if any , colonies in soft agar , whereas RWPE-1–PLK1 cells showed robust colony formation after 3 weeks of growth ( Figure 1—figure supplement 2A ) . This indicates that PLK1 overexpression is sufficient for cellular transformation of RWPE-1 cells . To assess in vivo tumorigenicity of PLK1-overexpressing prostate epithelial cells , control RWPE-1 or RWPE-1–PLK1 cells were injected subcutaneously into the flanks of NOD/SCID/γcnull ( NSG ) mice ( Fu et al . , 2003 ) . Six weeks after implantation , 100% ( 7 out of 7 ) of mice injected with RWPE-1–PLK1 cells developed primary tumors with an average size of 2 , 356 ± 589 ( SEM ) mm ( Barr et al . , 2004 ) ( Figure 1—figure supplement 2B ) . In contrast , no tumors developed in mice injected with control RWPE-1 cells ( Figure 1—figure supplement 2B ) . Interestingly , lung micrometastases were found in 5 of the 7 mice injected with RWPE-1–PLK1 cells . The micrometastases stained positive for PSA ( Figure 1—figure supplement 2C ) , indicating that they were of human origin and were derived from RWPE-1–PLK1 cells ( Figure 1B ) ( Bello et al . , 1997 ) . This data suggests that PLK1 overexpression not only leads to oncogenic transformation of prostate epithelial cells , but may also drive PCa metastasis . Based on the results of our tumor xenograft experiments implicating PLK1 in PCa metastasis , we next investigated the role of PLK1 in regulating prostate epithelial cell motility in vitro . Two classical assays of cell motility were used: one examined the collective cell migration during closure of planar epithelial wounds and the other analyzed the invasion of individual cells into Matrigel . Figure 1C , D demonstrates that PLK1 overexpression significantly accelerated both wound closure and Matrigel invasion of RWPE-1 cells . Importantly , PLK1 overexpression also promoted migration and invasion of PrEC ( human primary prostate epithelial cells ) ( Rudolph et al . , 2009 ) ( Figure 1—figure supplement 3 ) , thereby indicating that this pro-migratory effect is not a peculiar feature of the RWPE-1 cell line . Since PLK1 is a crucial regulator of the cell cycle , it may promote epithelial cell motility indirectly by stimulating cell proliferation . To rule out this indirect effect , we compared the random movement of individual RWPE-1–PLK1 cells to that of control RWPE-1 cells using time-lapse microscopy . Both migratory distance and velocity were significantly higher in RWPE-1-PLK1 cells compared to control RWPE-1 cells ( Figure 1E−G , Videos 1 , 2 ) . These data indicate that PLK1 overexpression stimulates cell motility in a proliferation-independent fashion . 10 . 7554/eLife . 10734 . 007Video 1 . Video showing the random migration of control RWPE-1 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 00710 . 7554/eLife . 10734 . 008Video 2 . Video showing the random migration of RWPE-1−PLK1 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 008 PLK1 overexpression in RWPE-1 cells causes the cells to change shape from an orthogonal epithelial cell morphology to a spindle-shaped fibroblast-like morphology ( Figure 2A ) , reminiscent of cells having undergone EMT . EMT is believed to play key roles in tumor progression and metastasis by enabling a switch from the stationary epithelial-like cell phenotype to the motile mesenchymal phenotype . On a molecular level , EMT is characterized by the decreased expression of epithelial markers and the increased expression of mesenchymal markers . In order to investigate whether PLK1 overexpression triggers EMT in RWPE-1 cell , we examined the expression of most of the characteristic epithelial and mesenchymal markers . Remarkably , relative to control RWPE-1 cells , RWPE-1–PLK1 cells downregulated epithelial markers ( E-cadherin and cytokeratin 19 ) and upregulated mesenchymal markers ( N-cadherin , vimentin , fibronectin , and SM22 ) , at both the mRNA and protein levels ( Figure 2B , C ) . The switch from epithelial to mesenchymal markers did not depend on the stages of the cell cycle ( Figure 2—figure supplement 1 ) . In agreement with the results obtained in RWPE-1 cells , PLK1 overexpression also induced an EMT-like phenotype in PrEC cells ( Figure 2—figure supplement 2 ) . Furthermore , we compared induction of EMT in cells expressing wild-type ( WT ) , constitutively active ( TD ) , or kinase-defective ( KM ) PLK1 . Constitutively active PLK1 induced the most robust reprogramming of gene expression in RWPE-1 cells , whereas expression of kinase inactive PLK1 failed to induce EMT ( Figure 2D ) . These results suggest that a PLK1-mediated phosphorylation event contributes to the induction of EMT in prostate epithelial cells . Importantly , PLK1 overexpression in RWPE-1 cells disrupted localization of E-cadherin , β-catenin , and junctional adhesion molecule ( JAM ) -A at the areas of cell-cell contacts , thereby indicating profound disassembly of adherens and tight junctions ( Figure 2E , arrows ) . This was accompanied by dramatic reorganization of the actomyosin cytoskeleton manifested by redistribution of non-muscle myosin IIB from the perijunctional F-actin bundles into basal stress fibers ( Figure 2E , arrowheads ) . Taken together , these data demonstrate induction of EMT in PLK1-overexpressing prostate epithelial cells , which is likely to promote a pro-motile phenotype by disrupting intercellular adhesions and creating isolated mesenchymal cells with invasive properties ( Moreno-Bueno et al . , 2008 ) . 10 . 7554/eLife . 10734 . 009Figure 2 . Overexpressing PLK1 in RWPE-1 cells induces EMT . ( A ) Representative phase-contrast images of control RWPE-1 and RWPE-1–PLK1 cells . Scale bar = 50 μm . ( B , C ) mRNA and protein expression of different EMT markers in RWPE-1–PLK1 cells ( PLK1 ) and vector control cells ( EV ) were examined by real-time RT-PCR ( B ) and Western blot ( C ) , respectively . The data are presented as the mean ± s . e . m . *p<0 . 01 , two-tailed Student’s t-test . ( D ) RWPE-1 cells were infected with lentivirus expressing wild-type PLK1 ( WT ) , constitutively active T210D ( TD ) , or kinase-dead K82M ( KM ) PLK1 mutants . Expression of EMT markers was examined by immunoblotting . ( E ) The architecture of adherens junctions ( E-cadherin and β-catenin labeling ) , tight junctions ( JAM-A ) , and the actomyosin cytoskeleton ( myosin IIB ) in control and PLK1-overexpressing cells was evaluated by immunolabeling and confocal microscopy . Arrows indicate disrupted adherens junctions and tight junctions , whereas arrowheads point to basal stress fibers in PLK1-overexpressing cells . Scale bar , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 00910 . 7554/eLife . 10734 . 010Figure 2—figure supplement 1 . The effect of PLK1 overexpression on EMT is independent of its cell cycle function . RWPE-1–PLK1 ( PLK1 ) cells and vector control ( EV ) cells were left either unsynchronized ( Asyn ) or synchronized at G1 , S , and M phases . The expression of EMT markers in these cells was examined by Western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 01010 . 7554/eLife . 10734 . 011Figure 2—figure supplement 2 . Overexpression of PLK1 in PrEC cells induces EMT . ( A ) Representative phase-contrast images of control PrEC ( EV ) and PrEC-PLK1 ( PLK1 ) cells . Scale bar = 50 μm . ( B , C ) EMT marker mRNA and protein expression in PrEC cells ( PLK1 ) and vector control cells ( EV ) was examined by real-time RT-PCR ( B ) and Western blot ( C ) , respectively . The data are presented as the mean ± s . e . m . *p<0 . 01 , two-tailed Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 011 Given the dramatic acceleration of cell motility in PLK1-overexpressing prostate epithelial cells , we sought to investigate whether downregulation of endogenous PLK1 could attenuate the migration of PCa cells . To exclude the effects of severe PLK1 depletion on growth inhibition and cell death , stable cell lines with partial PLK1 knockdown using a lentivirus encoding shRNA that targets the 3’-UTR of PLK1 were established using 2 known metastatic PCa cell lines ( DU145 and C4-2B ) . Downregulation of PLK1 expression in these cells was confirmed by immunoblotting ( Figure 3A , Figure 3—figure supplement 2A ) . As shown in Figure 3—figure supplement 1 , the partial PLK1 knockdown did not affect cell cycle progression and did not trigger cell apoptosis . Importantly , PLK1 downregulation promoted an epithelial phenotype in PCa cells by decreasing the expression of mesenchymal markers and increasing E-cadherin expression , inducing morphological alterations to a more epithelial-like appearance , and triggering the assembly of intercellular junctions ( Figure 3A−C ) . Furthermore , downregulation of PLK1 expression resulted in significant attenuation of PCa cell motility based on both wound closure and Matrigel invasion assays ( Figure 3D , E ) . All the observed changes induced by PLK1 knockdown were reversed by re-expression of WT PLK1 , but not by kinase-defective KM PLK1 ( Figure 3A−E ) . 10 . 7554/eLife . 10734 . 012Figure 3 . Downregulation of PLK1 reverses EMT and inhibits the motility of metastatic PCa cells . ( A ) Two metastatic PCa cell lines ( DU145 and C4-2B ) were infected with lentiviral shRNA constructs that target either the 3’-UTR of endogenous PLK1 ( shPLK1#1 ) or serve as a control ( shCTL ) . Wild-type ( WT ) or kinase-defective ( KM ) PLK1 were then re-expressed in PLK1 knockdown cells . The expression of PLK1 protein and EMT markers was determined by immunoblotting . In the PLK1 immunoblot , the lower bands show endogenous PLK1 expression and the upper bands show exogenous PLK1 expression . ( B ) Representative phase-contrast images of control , PLK1 knockdown cells , and PLK1 knockdown cells with re-expression of WT or KM PLK1 . Scale bar = 50 μm . ( C ) The architecture of adherens junctions ( E-cadherin ) and tight junctions ( JAM-A ) in control cells and cells with PLK1 manipulation was evaluated by immunolabeling and confocal microscopy . Scale bar , 20 µm . ( D , E ) The effects of PLK1 knockdown on planar migration and invasion of PCa cells were measured by the wound healing assay ( D ) and Matrigel invasion assay ( E ) , respectively . The data are presented as the mean ± s . e . m . *p<0 . 01 , two-tailed Student’s t-test . Scale bar , 500 µm ( D ) , and 100 µm ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 01210 . 7554/eLife . 10734 . 013Figure 3—figure supplement 1 . Partial knockdown of PLK1 in PCa cells has minimal effect on cell cycle progression and cell death . ( A ) The effect of PLK1 downregulation on cell cycle distributions was analyzed by flow cytometric analysis . ( B ) The effect of PLK1 downregulation on apoptosis was determined by staining with annexin V and PI solution , followed by flow cytometric analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 01310 . 7554/eLife . 10734 . 014Figure 3—figure supplement 2 . Downregulation of PLK1 results in EMT reversion and reduction of migration of metastatic PCa cells . ( A ) Three metastatic PCa cell lines ( PC3 , C4-2B , and DU145 ) were infected with lentiviral shRNA constructs that target either the 3’-UTR of endogenous PLK1 ( shPLK1#2 ) or serve as a control ( shCTL ) . The expression of PLK1 protein and EMT markers was determined by immunoblotting . ( B ) Representative phase-contrast images of control and PLK1 knockdown cells . Scale bar = 50 μm . ( C ) The architecture of adherens junctions ( E-cadherin ) and tight junctions ( JAM-A ) in control and PLK1 downregulated cells was evaluated by immunolabeling and confocal microscopy . Scale bar , 20 µm . ( D , E ) The effects of PLK1 knockdown on planar migration and invasion of PCa cells were measured by wound healing assay ( D ) and Matrigel invasion assay ( E ) , respectively . The data are presented as the mean ± s . e . m . *p<0 . 01 , two-tailed Student’s t-test . Scale bar , 500 µm ( D ) , and 100 µm ( E ) . ( F ) The effect of PLK1 downregulation on cell proliferation was determined by MTS assay as described in 'Materials and methods' . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 01410 . 7554/eLife . 10734 . 015Figure 3—figure supplement 3 . Validation of the role of PLK1 in induction of EMT in the ARCaP cell culture model . ( A ) The level and activity of PLK1 protein in ARCaPE and ARCaPM cells were examined by Western blotting analysis using anti-PLK1 and anti-PLK1 ( pT210 ) antibodies , respectively . β-actin was used as a loading control . ( B ) ARCaPE cells were transiently transfected with a PLK1 expression construct . The expression of PLK1 protein was determined by immunoblotting . The effects of PLK1 overexpression on EMT marker expression , morphological changes , assembly of intracellular junctions , and invasiveness were determined as described in Figures 1–2 . ( C ) The endogenous PLK1 in ARCaPM cells was downregulated by transfecting with 2 different siRNAs targeting endogenous PLK1 ( si#1 and si#1 ) . The expression of PLK1 protein was determined by immunoblotting . The effects of PLK1 downregulation in ARCaPM cells on EMT marker expression , morphology , intracellular junctions , and invasiveness were determined as described in Figures 1–2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 01510 . 7554/eLife . 10734 . 016Figure 3—figure supplement 4 . PLK1 contributes to physiological EMT events . RWPE-1 and ARCaPE cells were treated with a combination of TGF-β1 ( 4 ng/mL ) and EGF ( 50 ng/mL ) in the presence and absence of BI2536 ( 1 nM ) for 5 days . The EMT marker expression and morphological alteration were determined as described in Figures 1–2 . The phosphorylation of FoxM1 ( a PLK1 substrate ) at S724 was used as a readout for the PLK1 activity . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 016 To exclude the possibility of off-target effects of individual shRNAs , another shRNA ( shPLK1#2 ) targeting a different site within the 3’-UTR of PLK1 was used to verify our observations ( Figure 3—figure supplement 2 ) . Similarly , upon partially knocking down endogenous PLK1 with yet another shRNA in 3 metastatic PCa cell lines ( PC3 , C4-2B , and DU145 ) , the effects of PLK1 on EMT induction and pro-motile phenotype were reversed ( Figure 3—figure supplement 2 ) . Furthermore , we also used an androgen-refractory cancer of the prostate ( ARCaP ) model to verify the role of PLK1 in induction of EMT . ARCaP cells were derived from the ascites fluid of an 83-year-old Caucasian man diagnosed with metastatic carcinoma of the prostate ( Zhau et al . , 1996 ) . Epithelium-like ARCaPE cells and mesenchymal-like ARCaPM cells are sublines of ARCaP cells that were isolated by single-cell dilution cloning ( Xu et al . , 2006 ) . Interestingly , PLK1 is differentially expressed and activated in those 2 cell lines ( higher in highly metastatic ARCaPM cells and lower in less metastatic ARCaPE cells; Figure 3—figure supplement 3 ) , further suggesting a role for PLK1 in PCa metastasis . Upregulation of PLK1 in ARCaPE cells led to the induction of EMT , whereas downregulation of PLK1 in ARCaPM cells promoted the reversion of EMT manifested by biochemical changes as well as altered morphology , assembly of intracellular junctions , and motility ( Figure 3—figure supplement 3 ) . Taken together , our overexpression and knockdown experiments revealed a novel function of PLK1 as a critical regulator of prostate epithelial cell motility and EMT . To determine whether PLK1 is involved in physiologically-relevant EMT , RWPE-1 and ARCaPE cells were treated with a combination of EMT-inducing growth factors ( Zhau et al . , 2008 ) . Following treatment with transforming growth factor β1 ( TGF-β1 ) and epidermal growth factor ( EGF ) , cells underwent an EMT-like process as demonstrated by acquisition of spindle-like cell morphology and switched expression of epithelial and mesenchymal markers ( Figure 3—figure supplement 4 ) . Although growth-factor−induced EMT was not accompanied by increased expression or activity of PLK1 , this process was partially blocked by a low dose of a potent pharmacological inhibitor of PLK1: BI2536 ( Steegmaier et al . , 2007 ) ( Figure 3—figure supplement 4 ) . The phosphorylation of FoxM1 ( a PLK1 substrate ) at S724 was used as a readout for PLK1 activity ( Figure 3—figure supplement 4 ) . These data suggest that PLK1 could be important for EMT-induced by physiological and pathophysiological stimuli . We next sought to elucidate the signaling pathways downstream of oncogenic PLK1 that are responsible for the EMT induction and increased motility of prostate epithelial cells . Expressional reprogramming of epithelial cells undergoing EMT is known to be mediated by several transcriptional regulators , most notably Snail1 , Snail2 ( Slug ) , ZEB1 , ZEB2 , E47 , and Twist ( Sánchez-Tilló et al . , 2012 ) . Therefore , we examined which of these transcriptional regulators are induced by PLK1 overexpression in prostate epithelial cells . Quantitative real-time PCR analysis demonstrated marked and selective upregulation of ZEB1 and ZEB2 in RWPE-1–PLK1 cells , whereas expression of Snail1 , Snail2 , E47 , and Twist was not significantly changed ( Figure 4A ) . Furthermore , increased protein expression of ZEB1 and ZEB2 was observed by immunoblotting in PLK1-overexpressing RWPE-1 cells ( Figure 4B ) and PrEC cells ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 10734 . 017Figure 4 . Both ZEB1 and ZEB2 play a causal role in PLK1-induced EMT and increased motility of prostate epithelial cells . ( A ) Expression of different EMT-inducing transcription factors was examined in control and PLK1-overexpressing RWPE-1 cells by quantitative real-time RT-PCR . mRNA expression of genes of interest was normalized by the level of glyceraldehyde phosphate dehydrogenase ( GAPDH ) mRNA and is presented as relative expression . The data are presented as the mean ± s . e . m . *p<0 . 05 , two-tailed Student’s t-test . ( B ) The levels of ZEB1 and ZEB2 proteins in control and PLK1-overexpressing cells were examined by immunoblotting . ( C−F ) The effects of either individual shRNA-mediated downregulation of ZEB1 and ZEB2 , or their dual knockdown in PLK1-induced EMT ( C ) ; cell-cell junctional disassembly ( D ) ; planar cell migration ( E ) ; and cell invasion ( F ) were determined as described in Figures 1–2 . N . S . : no significant difference . Scale bar , 20 µm ( D ) , 500 µm ( E ) , and 100 µm ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 01710 . 7554/eLife . 10734 . 018Figure 4—figure supplement 1 . Elevated PLK1 levels result in activation of ERK , and upregulation of ZEB1 , ZEB2 , and Fra1 in PrEC cells . Protein levels of ZEB1 and ZEB2 , Fra1 , and activated and total ERK1/2 in PrEC ( EV ) and PrEC-PLK1 ( PLK1 ) cells were determined by Western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 018 In order to determine if ZEB1 and/or ZEB2 have a causal role in PLK1-dependent induction of EMT , we depleted these transcriptional regulators individually or in combination , in RWPE-1–PLK1 cells ( Figure 4C ) . Downregulation of ZEB2 increased the expression of E-cadherin and cytokeratin 19 , while it decreased the levels of N-cadherin , fibronectin , and vimentin; conversely , depletion of ZEB1 was less effective and primarily decreased vimentin expression ( Figure 4C ) . Interestingly , dual knockdown of ZEB1 and ZEB2 completely reversed expressional reprogramming of EMT markers in RWPE-1–PLK1 cells ( Figure 4C ) . Depletion of ZEB1 and ZEB2 proteins not only eliminated the biochemical signature of the EMT , but also reversed the functional effects of PLK1 overexpression . For example , either individual or co-knockdown of ZEB1 and ZEB2 restored assembly of adherens and tight junctions in RWPE-1–PLK1 cells ( Figure 4D ) . Moreover , downregulation of ZEB1 or ZEB2 significantly decreased planar migration and Matrigel invasion of RWPE-1–PLK1 cells ( Figure 4E , F ) . Remarkably , dual knockdown of ZEB1 and ZEB2 completely abolished PLK1-induced motility of RWPE-1–PLK1 cells ( Figure 4E , F ) . These findings indicate that ZEB1 and ZEB2 play causal roles in PLK1-induced EMT and motility in prostate epithelial cells . Next , we investigated the signaling pathways that are responsible for the induction of ZEB1 and ZEB2 in PLK1-overexpressing prostate epithelial cells . Aberrant activation of the mitogen-activated protein kinase/extracellular signal-regulated kinase ( MAPK/ERK ) pathway has been shown to contribute to tumor invasion and progression , and has recently been linked to the EMT process ( Reddy et al . , 2003; Shin et al . , 2010 ) . Therefore , we examined the effect of PLK1 overexpression on the activation status of key molecular constituents of the ERK signaling cascade . PLK1 overexpression in RWPE-1 cells resulted in increased phosphorylation of ERK1/2 and its upstream kinase MEK1/2 ( Figure 5A ) , thereby indicating activation of MEK1/2-ERK1/2 signaling . Consistently , we also detected enhanced phosphorylation of ERK1/2 in PLK1-overexpressing PrEC cells ( Figure 4—figure supplement 1 ) . In order to determine the potential causal role of the MEK/ERK pathway in PLK1-induced EMT and accelerated motility of prostate epithelial cells , we blocked this pathway with RO5126766 , a highly selective dual inhibitor of MEK and its upstream kinase , RAF ( Martinez-Garcia et al . , 2012 ) . Exposure of RWPE-1–PLK1 cells to RO5126766 abrogated ERK activation , reduced expression of ZEB1 and ZEB2 , and completely reversed the biochemical manifestation of EMT ( Figure 5B ) . Moreover , inhibition of MEK/ERK signaling restored assembly of intercellular junctions ( Figure 5C ) and inhibited the increased wound closure and Matrigel invasion of RWPE-1–PLK1 cells ( Figure 5D , E ) . These results strongly support the role of the MEK/ERK pathway in PLK1-induced EMT and cell motility . 10 . 7554/eLife . 10734 . 019Figure 5 . ERK1/2 activation is essential for PLK1-induced EMT and increased motility of prostate epithelial cells . ( A ) The effect of PLK1 overexpression on activation ( phosphorylation ) of ERK1/2 and MEK1/2 was determined by immunoblotting . ( B−E ) Control ( EV ) and PLK1-overexpressing ( PLK1 ) RWPE-1 cells were treated for 24 hr with either vehicle ( DMSO ) or RO5126766 ( 10 mM ) . The effects of MEK inhibition on PLK1-dependent induction of EMT ( B ) , junctional disassembly ( C ) , accelerated planar cell migration ( D ) , and Matrigel invasion ( E ) were determined as described in Figures 1 and 2 . The data are presented as the mean ± s . e . m . *p<0 . 05 , two-tailed Student’s t-test . NS: no significant difference . Scale bar , 20 µm ( C ) , 500 µm ( D ) , and 100 µm ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 019 What is the molecular link between ERK activation and upregulation of EMT inducers such as ZEB1 and ZEB2 ? Recent studies have uncovered Fra-1 as an essential player in the ERK/ZEB signaling pathway , which induces EMT and cell migration ( Shin et al . , 2010 ) . Fra1 belongs to the Fos gene family , whose protein products can dimerize with protein of the JUN family , thereby forming the transcription factor complex AP-1 . Ectopic expression of Fra1 in epithelioid cells resulted in morphologic changes that resemble fibroblastoid conversion , and increased motility and invasiveness ( Kustikova et al . , 1998 ) . Fra1 has been implicated as a potent regulator of anti-apoptosis , cell motility , and invasion in a variety of tumor cell types ( Milde-Langosch , 2005; Young and Colburn , 2006 ) . Based on these data , we sought to elucidate if Fra1 is essential for PLK1-induced EMT and accelerated cell motility in PCa . Overexpression of PLK1 in RWPE-1 cells ( Figure 6A , B ) and PrEC cells ( Figure 4—figure supplement 1 ) resulted in increased Fra1 expression at both the mRNA and protein levels compared to control cells . Moreover , knocking down Fra1 in RPWE-1–PLK1 cells significantly decreased ZEB1/2 expression and reversed the biochemical signature of the EMT ( Figure 6C ) . Importantly , Fra1 depletion reversed the increased wound healing and Matrigel invasiveness and restored assembly of intercellular junctions in PLK1-overexpressing RWPE-1 cells ( Figure 6D–F ) . These data highlight Fra1 as a crucial molecular effector connecting ERK activation and induction of ZEB1/2 expression in the PLK1-induced signaling pathway leading to EMT and increased motility of PCa cells . 10 . 7554/eLife . 10734 . 020Figure 6 . Fra1 is a critical mediator of PLK1-induced EMT and accelerated motility of prostate epithelial cells . ( A , B ) mRNA and protein expression of Fra1 were examined in control ( EV ) and PLK1-overexpressing ( PLK1 ) RWPE-1 cells by real-time RT-PCR ( A ) and immunoblotting ( B ) , respectively . The data are presented as the mean ± s . e . m . *p<0 . 05 , two-tailed Student’s t-test . ( C–F ) The effects of Fra1 depletion on PLK1-induced EMT ( C ) , planar cell migration ( D ) , Matrigel invasion ( E ) , and epithelial junctional disassembly ( F ) were determined as described in Figures 1 and 2 . NS: no significant difference . Scale bar , 500 µm ( D ) , 100 µm ( E ) , and 20 µm ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 020 We next investigated the upstream events that could potentially mediate PLK1-dependent activation of MEK/ERK signaling . The RAF family of serine/threonine protein kinases consists of 3 isoforms , ARAF , BRAF , and CRAF , all of which serve as upstream kinases that evoke a serine/threonine phosphorylation cascade through sequential phosphorylation of MEK1/2 , ERK1/2 , and further downstream effectors to elicit a variety of cellular responses ( Kolch , 2000 ) . CRAF is ubiquitously expressed in mammalian cells , whereas ARAF and BRAF display more tissue-specific expression ( Hagemann and Rapp , 1999 ) . Activation of CRAF has served as a framework for the other 2 isoforms . CRAF is the cellular proto-oncogene homologue of v-RAF , a retroviral oncogene , and is a central component of the MAP kinase cascade ( Heidecker et al . , 1990; Morrison and Cutler , 1997 ) . Oncogenic CRAF causes EMT and invasion ( Hou et al . , 2014; Lan et al . , 2004 ) . To determine the involvement of CRAF in PLK1-induced EMT and accelerated motility of prostate epithelial cells , we determined if CRAF is activated in RWPE-1–PLK1 cells . It has been reported that phosphorylation of CRAF at S338 and S339 is essential for its activation and that the first essential role of CRAF kinase activity is to autophosphorylate S621 ( Diaz et al . , 1997; Noble et al . , 2008 ) . S621 phosphorylation plays a critical role in preventing CRAF proteasome-mediated degradation ( Noble et al . , 2008 ) . Immunoblotting analysis with antibodies specifically recognizing those 2 different phosphorylated sites ( S338 and S339 ) demonstrated a significant increase in CRAF phosphorylation following PLK1 overexpression ( Figure 7A ) . Furthermore , CRAF autophosphorylation of S621 was also significantly elevated upon PLK1 overexpression ( Figure 7A ) . Downregulation of CRAF expression by a specific shRNA caused several prominent biochemical alterations in PLK1-overexpressing RWPE-1 cells . These alterations include inhibition of ERK signaling manifested by decreased MEK1/2 and ERK1/2 phosphorylation; decreased expression of Fra1 , ZEB1 , and ZEB2; and complete reversal of EMT-like expressional reprogramming of prostate epithelial cells ( Figure 7B ) . Moreover , CRAF depletion restored assembly of intercellular junctions and suppressed wound healing and Matrigel invasion of PLK1-overexpressing RWPE-1 cells ( Figure 7C–E ) . These results demonstrate that CRAF plays an essential role in PLK1-induced EMT and cell motility . 10 . 7554/eLife . 10734 . 021Figure 7 . CRAF plays an essential role in PLK1-induced EMT and accelerated motility of prostate epithelial cells . ( A ) The effects of PLK1 overexpression on the levels of total and phosphorylated CRAF were determined by immunoblotting . p/t: indicates densitometric intensity ratio of phosphorylated to total CRAF . ( B−E ) The effect of shRNA-mediated CRAF knockdown on PLK1-induced EMT ( B ) , epithelial junctional disassembly ( C ) , planar cell migration ( D ) , and Matrigel invasion ( E ) were determined as described in Figures 1–2 . N . S . : no significant difference . Scale bar , 20 µm ( C ) , 500 µm ( D ) , and 100 µm ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 021 We next investigated the molecular mechanism by which PLK1 activates CRAF . The sequence surrounding S339 ( residue 337–340: DSSY ) on CRAF resembles the PLK1 consensus phosphorylation sequence D/E-X-S/T-ψ ( X denotes any amino acid and ψ denotes a hydrophobic amino acid ) ( Nakajima et al . , 2003 ) . We hypothesized that CRAF is a substrate of PLK1 , and that PLK1-mediated phosphorylation of CRAF leads to its activation . To test these hypotheses , we first examined whether PLK1 interacts with CRAF . Endogenous PLK1-CRAF complex was detected in HeLa cells by immunoprecipitation , and PLK1-CRAF interaction appeared to be enhanced during the M phase of the cell cycle when PLK1 activity is maximal ( Figure 8A ) . A GST pull-down assay was performed and showed CRAF association with GST-PBD , but not with GST or the GST-PBD mutant ( Figure 8B ) , thereby suggesting that CRAF is a potential substrate for PLK1 . 10 . 7554/eLife . 10734 . 022Figure 8 . PLK1-mediated phosphorylation of CRAF leads to CRAF activation and stabilization . ( A ) Co-immunoprecipitation of endogenous PLK1 and CRAF from HeLa cell lysates obtained under control conditions or following nocodazole treatment to arrest cells at M phase . ( B ) GST pull-down assay was performed using GST-tagged wild-type ( WT ) or mutant ( MUT ) PLK1 PBD ( upper panel ) . Equal loading of kinase substrates is indicated by Coomassie Blue staining ( lower panel ) . ( C ) Bacterially expressed CRAF was subjected to in vitro kinase assays with constitutively active ( TD ) or kinase-defective ( KM ) PLK1 mutants purified from insect cells . The phosphorylation of CRAF was observed by immunoblotting with the indicated antibodies . Ponceau S staining of the blot was used to indicate equal loading of the assays . ( D ) RWPE-1 cells were infected with lentivirus expressing empty vector ( EV ) , wild-type PLK1 ( WT ) , constitutively active T210D ( TD ) , or kinase-dead K82M ( KM ) mutants . Total cell lysates were subject to immunoprecipitation with CRAF antibody and then analyzed by immunoblotting . p/t: indicates densitometric intensity ratio of phosphorylated to total CRAF . ( E ) Immunoblotting analysis of ERK1/2 activation ( phosphorylation ) in RWPE-1 cells expressing EV or PLK1 WT , TD , or KM mutants . ( F ) C4-2B cells were infected with lentiviral shRNA constructs that target either the 3’-UTR of endogenous PLK1 ( shPLK1#1 ) or serve as a control . Wild-type ( WT ) or kinase-defective ( KM ) PLK1 were then re-expressed in PLK1 knockdown cells . The levels of phosphorylated and total CRAF , MEK and ERK , Fral , and ZEB1/2 were determined by Western blotting analysis . ( G ) C4-2B cells were arrested in M phase by nocodazole ( 0 . 1 mg/mL ) treatment for 16 hr , and then subject to BI 2536 treatment for 30 min . The levels of phosphorylated and total CRAF were determined by Western blotting analysis . ( H ) Flag-tagged CRAF WT- , CRAF-A/A- , and CRAF-D/E-overexpressing RWPE-1 cells were subjected to immunoprecipitation with anti-Flag antibody followed by immunoblotting with the indicated antibodies . ( I ) Control- ( EV ) , PLK1 WT- , PLK1 TD- and PLK1 KM-overexpressing RWPE-1 cells were treated with cycloheximide ( CHX: 20 μg/ml ) for the indicated times , and the CRAF protein levels were monitored by immunoblotting ( left panel ) . Quantification of the endogenous CRAF protein levels relative to β-actin expression is shown in the right panel . The data are presented as the mean ± s . e . m . ( J ) PLK1 WT- , and PLK1 KM-overexpressing RWPE-1 cells were treated with cycloheximide ( CHX: 20 μg/ml ) along with vehicle ( DMSO ) , MG132 ( 2 . 5μM ) , or chloroquine ( CLQ , 50 mM ) for the indicated times , and level of CRAF protein was examined by immunoblotting . ( K ) Flag-tagged CRAF WT- , CRAF-S621A- , CRAF-S621D- , CRAF-A/A- , and CRAF-D/E-overexpressing RWPE-1 cells were treated with cycloheximide ( CHX: 20 μg/ml ) for the indicated time and the level of CRAF protein was determined by immunoblotting . ( L ) A proposed diagram of a novel signaling cascade that mediates PLK1-dependent induction of EMT and cell motility . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 02210 . 7554/eLife . 10734 . 023Figure 8—figure supplement 1 . Downregulation of PLK1 led to dramatic reduction of CRAF phosphorylation at S338 , S339 , and S621 that was rescued by WT PLK1 , but not KM PLK1 . DU145 cells were infected with lentiviral shRNA constructs that target either the 3’-UTR of endogenous PLK1 ( shPLK1#2 ) or serve as a control . WT or KM PLK1 was then re-expressed in PLK1 knockdown cells . The levels of phosphorylated and total CRAF , MEK and ERK , Fral , and ZEB1/2 were determined by Western blotting analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 10734 . 023 An in vitro kinase assay demonstrated that PLK1 directly phosphorylated CRAF at S338 and S339 , but not at S621 ( Figure 8C ) . Phosphorylation of CRAF at S338 , S339 , and S621 was significantly elevated in cells expressing WT or TD PLK1 , but not KM PLK1 ( Figure 8D ) . Cells expressing TD PLK1 displayed higher levels of phosphorylated CRAF compared to those with WT PLK1 ( Figure 8D ) . These data suggest that the enzymatic activity of PLK1 is both directly and indirectly responsible for those phosphorylation events . Consistently , phosphorylated ERK was significantly increased in cells expressing WT or TD PLK1 , but not KM PLK1 ( Figure 8E ) , suggesting that a PLK1-induced phosphorylation event triggers the activation of ERK signaling . On the other hand , shRNA-mediated downregulation of PLK1 led to dramatic reduction of CRAF phosphorylation at S338 , S339 , and S621 , which was rescued by WT PLK1 , but not KM PLK1 ( Figure 8F and Figure 8—figure supplement 1 ) . Furthermore , inhibition of ERK signaling , manifested by decreased MEK1/2 and ERK1/2 phosphorylation along with reduced expression of Fra1 , ZEB1 , and ZEB2 , was observed upon PLK1 knock down , and these effects were reversed by re-expressing WT , but not KM PLK1 ( Figure 8F and Figure 8—figure supplement 1 ) . To further confirm that PLK1 directly phosphorylates CRAF at S338 and S339 , cells were synchronized in M phase by nocodazole treatment followed by a transient inhibition of PLK1 activity with 2 different concentrations of BI 2536 for 30 min . Significant reduction in phosphorylation of CRAF at S338 and S339 was immediately observed upon treatment with the higher dose of the PLK1 inhibitor ( Figure 8G ) , thereby suggesting that PLK1 indeed directly phosphorylates CRAF at S338 and S339 . Furthermore , when we blocked the PLK1 phosphorylation sites on CRAF by mutating S338 and S339 to alanines ( CRAF-A/A ) , S621 phosphorylation was significantly diminished as compared to WT CRAF or the phospho-mimetic mutant ( CRAF-D/E ) ( Figure 8H ) , suggesting that PLK1-induced phosphorylation of CRAF at S338 and S339 contributes to the CRAF autophosphorylation of S621 . Taken together , these findings demonstrate that PLK1 is responsible for phosphorylation-mediated activation of CRAF , which leads to CRAF autophosphorylation . Elevated CRAF protein levels , but unchanged CRAF mRNA expression , were observed in RWPE-1–PLK1 cells ( Figure 7A , data not shown ) . In order to elucidate the mechanisms of this selective upregulation of CRAF protein , we asked whether PLK1 can stabilize CRAF . Cells were treated with cycloheximide to inhibit de novo protein synthesis and the expression level of CRAF was compared at different time points ( 0 , 2 , 4 , 6 , 8 , and 15 hr ) in control cells and cells overexpressing different forms of PLK1 ( WT , TD , and KM ) . Immunoblotting analysis demonstrated significant degradation of CRAF in control cells and cells expressing an inactive PLK1-KM mutation ( Figure 8I ) . CRAF degradation was inhibited by MG132 ( a proteasome inhibitor ) , but not by chloroquine ( a lysosomal inhibitor ) ( Figure 8J ) . In contrast , no significant CRAF degradation was observed in cells overexpressing either WT PLK1 or its constitutively active TD mutant ( Figure 8I ) . Given the previous finding that CRAF autophosphorylation of S621 is critical for preventing its degradation ( Noble et al . , 2008 ) and our observations that PLK1 phosphorylation of CRAF leads to its activation and subsequent autophosphorylation at S621 ( Figure 8A–H ) , we hypothesized that PLK1-induced phosphorylation of CRAF contributes to CRAF stabilization by promoting its autophosphorylation . To corroborate the previous findings from another group ( Noble et al . , 2008 ) , we mutated S621 of CRAF to alanine ( CRAF-S621A ) or aspartate ( CRAF-S621D ) and examined their stability . Consistently , CRAF-S621A displayed a much shorter half-life when compared to WT CRAF or CRAF-S621D ( Figure 8K ) . Importantly , the CRAF-A/A mutant ( lacking PLK1-mediated phosphorylation ) was quickly degraded over a 15-h time course ( Figure 8K ) , which is in agreement with our observation that the CRAF-A/A mutant harbors a much lower level of S621 phosphorylation ( Figure 8H ) . In sharp contrast , the CRAF WT and the CRAF-D/E mutant showed minimal degradation during that period of time ( Figure 8K ) . These results demonstrate that PLK1 increases the stability of CRAF protein by preventing proteasome degradation .
PLK1 is an established mitotic kinase that is involved in multiple steps of mitotic progression . The present study revealed unanticipated non-canonical functions of PLK1 as a potent inducer of EMT and a stimulator of the motile phenotype of prostate epithelial cells . These functions of PLK1 are mediated by the CRAF-MEK1/2-ERK1/2-Fra1-ZEB1/2 signaling pathway and are independent of effects on cell cycle regulation . Our results are consistent with previous studies that suggested pro-migratory activity of PLK1 in colorectal , breast , thyroid cancer , and melanoma ( Han et al . , 2012; Kneisel et al . , 2002; Rizki et al . , 2007; Zhang et al . , 2012 ) . However , our study is the first to demonstrate the causal role of PLK1 in regulating cancer cell motility . Indeed , by using a combination of gain- and loss-of-function approaches , we established that PLK1 is necessary and sufficient to promote both planar migration and invasion of prostate epithelial cells ( Figures 1 and 2 ) . Furthermore , our time-lapse microscopy experiments clearly demonstrated that PLK1 directly regulates the velocity of epithelial cell migration , independently of its effects on other cellular processes ( Figures 1D–F ) . PLK1 controls prostate epithelial cell motility by mechanisms requiring transcriptional reprogramming and dramatic alterations of cell phenotype . Indeed , overexpression of PLK1 induced EMT-like alterations in prostate epithelial cells ( Figure 2 ) , whereas PLK1 knockdown restored epithelial features of invasive prostate cancer cell lines ( Figure 3 ) . This PLK1-dependent regulation of cell phenotype has not been previously reported in any cell type , and it is likely to underlie the observed effects of PLK1 on epithelial cell motility . There are several potential mechanisms by which the induction of EMT can promote cell motility . One mechanism involves disassembly of epithelial junctions that weaken intercellular adhesions , thereby allowing cell dissemination ( Godde et al . , 2010; Le Bras et al . , 2012 ) . Another mechanism involves rearrangement of the actomyosin cytoskeleton from epithelia-specific perijunctional bundles to basal stress fibers that are characteristic of mesenchymal cells . This rearrangement enhances cell-matrix adhesion and enables more efficient cell migration ( Martin et al . , 2014; Yilmaz and Christofori , 2009 ) . We observed both junctional disassembly and cytoskeletal rearrangements in prostate epithelial cells undergoing PLK1-mediated EMT ( Figure 2 ) . Our finding that PLK1 overexpression potently induced EMT in prostate epithelium is important given recent evidence highlighting EMT as an emerging mechanism of PCa progression , metastasis , and therapeutic resistance . For instance , several EMT markers , such as vimentin and N-cadherin , are commonly expressed in circulating tumor cells from patients with relapsed metastatic PCa ( Armstrong et al . , 2011 ) . A 'cadherin switch' with increased N-cadherin and reduced E-cadherin expression has been associated with relapse after PCa surgery and development of metastatic disease ( Gravdal et al . , 2007; Umbas et al . , 1992 ) . Work from Reiter’s group demonstrates a central role for N-cadherin in PCa metastasis ( Tanaka et al . , 2010 ) . Finally , the levels of ZEB proteins positively correlate with Gleason grading and PCa metastasis ( Graham et al . , 2008 ) . Through a series of biochemical analyses , we delineated the molecular mechanism underlying PLK1-mediated cell motility and EMT . Previous studies demonstrated that CRAF-MEK1/2-ERK1/2 signaling plays a crucial role in the regulation of EMT as well as cell migration and invasion in several cancers ( Birchmeier et al . , 1993; Hay and Zuk , 1995; Schoenenberger et al . , 1991 ) . The CRAF-MEK1/2-ERK1/2 pathway is regulated by Ras as well as various kinases including Src , PKC , and PAK ( King et al . , 1998 ) . In this study , we discovered a novel regulatory pathway for MAPK signaling . We demonstrated that CRAF is a physiological substrate of PLK1 ( Figure 8 ) . CRAF consists of a N-terminal regulatory domain and a C-terminal catalytic domain . The 'N-region' , located at the N-terminal of the kinase domain , contains the critical activating phosphorylation sites , S338 and S339 ( Diaz et al . , 1997; Edin and Juliano , 2005 ) . S338 can be phosphorylated by the p21 activated kinase ( PAKs ) ( King et al . , 1998; Chaudhary et al . , 2000; Sun et al . , 2000 ) and other as yet unidentified kinases ( Chiloeches et al . , 2001 ) . Our data support a dual-mechanism model of PLK1-mediated regulation of CRAF signaling ( Figure 8L ) : PLK1 directly interacts with and phosphorylates CRAF at S338 and S339 , which leads to CRAF activation . The activated CRAF undergoes autophosphorylation of S621 , thereby preventing proteasome-mediated degradation of CRAF , which generates a positive-feedback loop , leading to a further increase in CRAF level and activity . This activation event triggers the activation of downstream MEK1/2-ERK1/2 signaling in prostate epithelial cells overexpressing PLK1 ( Figure 8L ) . Consistently , Mielgo et al . reported that PLK1 associates with CRAF and that this interaction does not require CRAF kinase activity ( Mielgo et al . , 2011 ) . Interestingly , they also found that CRAF indirectly promotes PLK1 activation ( Mielgo et al . , 2011 ) . This adds an additional layer of complexity to the PLK1-CRAF interplay to further activate CRAF . In addition , PLK1 may differentially trigger distinct signaling pathways under different physiological conditions . For instance , PLK1 can directly activate MEK/ERK signaling through phosphorylation of the MEK activating site in airway smooth muscle cells ( Jiang and Tang , 2015 ) , although that is not the case for PCa ( Figure 8 ) . We showed that the ERK1/2-Fra1-ZEB1/2 pathway is aberrantly activated in PLK1-overexpressing prostate epithelial cells and that blocking the ERK/Fra1/ZEB1/2 pathway by shRNA or pharmacological inhibition reverses EMT and decreases cell motility in these cells ( Figures 4–6 ) . These results indicate that PLK1 promotes EMT and motility through the CRAF-MEK1/2-ERK1/2-Fra1-ZEB1/2 pathway in prostate epithelial cells . These findings highlight PLK1 as a critical molecular rheostat that controls phenotypic plasticity of normal prostate epithelial cells as well as PCa progression and dissemination . Identification of the CRAF-MEK1/2-ERK1/2 signaling cascade as an essential downstream effector of PLK1 in prostate epithelial cells corresponds with the known roles of this cascade in the pathogenesis of various types of cancer , including PCa ( McCubrey et al . , 2007; Roberts and Der , 2007 ) . Given the fact that the frequency of Ras gene mutations in prostate tumors is low ( Carter et al . , 1990 ) , PLK1-induced activation of CRAF provides mechanistic insights into the aberrant activation of CRAF-MEK1/2-ERK1/2 signaling frequently detected in PCa . Future experiments will further establish PLK1-mediated regulation of CRAF-MEK1/2-ERK1/2 signaling in patients with PCa . Interestingly , it was reported that ERK2/Fra1/ZEB1/2 signaling is responsible for induction of EMT in RasG12V-transformed MCF10A cells ( Shin et al . , 2010 ) , suggesting that ERK/Fra1/ZEB signaling could be a common pathway to induce EMT in mammalian epithelial cells . PLK1 deregulation has been linked with the initiation and progression of many human cancers , including PCa ( Cholewa et al . , 2013; Takai et al . , 2005; Weichert et akl . , 2004 ) . The current dogma in the field implies that PLK1 controls cancer development through multiple mechanisms , including canonical regulation of mitosis and cytokinesis as well as modulation of DNA replication and cell survival ( Deeraksa et al . , 2013; Luo and Liu , 2012 ) . Whether and how PLK1 drives PCa metastasis in vivo remain to be elucidated . Our xenograft studies showed that PLK1 overexpression in human prostate epithelial cells leads to cellular transformation in vitro and promotes tumor formation in NSG mice , which suggests that PLK1 has a tumor-promoting role in the prostate . Strikingly , NSG mice engrafted with RWPE-1–PLK1 cells developed lung micrometastases at a high frequency ( Figure 1—figure supplement 2 ) , which provides in vivo evidence that PLK1 plays a role in PCa invasion and metastasis . Our study establishes the role of PLK1 in the induction of EMT and stimulation of cell motility , which adds a novel and previously unanticipated role for PLK1 during PCa development . Furthermore , our observation that partial downregulation of PLK1 in metastatic PCa cells has no effect on cell cycle progression and cell viability suggests that a low level of PLK1 is sufficient to maintain cell viability and regulate the cell cycle . In contrast , upregulation of PLK1 promotes its non-canonical functions , including stimulation of EMT and acceleration of cell motility . This could explain why normal cells develop a complex system to tightly control the level and activity of PLK1 throughout the cell cycle , and how increased PLK1 promotes tumorigenesis . EMT is an important mechanism of tumor progression and metastasis ( Kalluri and Weinberg , 2009; Yang and Weinberg , 2008 ) . Loss of cell-cell contacts and reorganization of the intracellular cytoskeleton during EMT results in increased cell migration and invasion ( Moreno-Bueno et al . , 2008 ) . This allows cells to invade the surrounding stroma and vasculature , which leads to tumor dissemination and metastases ( Hugo et al . , 2007 ) . However , the mechanisms that drive EMT in PCa remain elusive ( Gravdal et al . , 2007; Howard et al . , 2008 ) . This study provides strong evidence that PLK1 is a key regulator of EMT in prostate epithelial cells and in PCa , which represents an underlying mechanism for PLK1-driven PCa progression and spreading . EMT also enables cancer cells to avoid apoptosis , anoikis , and oncogene addiction ( Jordan et al . , 2011; Tiwari et al . , 2012 ) . An emerging mechanism of EMT is reprogramming epithelial cells into cancer stem cells that have been strongly associated with resistance to chemotherapy and disease recurrence ( Chang et al . , 2013 ) . Whether PLK1-induced EMT contributes to these oncogenic processes will need further investigation . Interestingly , the low level of endogenous PLK1 observed in RWPE-1 cells appears to be sufficient to mediate growth-factor–induced EMT ( Figure 3—figure supplement 4 ) . This suggests that in tumor cells that do not overexpress PLK1 , the low level of PLK1 that is required for cell division could also mediate induction of EMT under certain circumstances , such as in the tumor microenvironment . This finding would have important implications in targeting PLK1 for anticancer treatments . For instance , targeting PLK1 might be applied to a wide variety of cancers , regardless of their PLK1 expression levels , to prevent induction of EMT and tumor metastasis . Including PLK1 inhibitors in current standard therapeutic regimens could expand the scope of clinical efficacy that currently available drugs have established . In conclusion , we demonstrate for the first time that PLK1 drives planar cell migration and matrix invasion of prostate epithelial cells and PCa by mechanisms involving induction of EMT . This previously unanticipated pro-migratory activity of PLK1 is driven by direct phosphorylation and activation of CRAF , which results in enhanced signaling through the MEK1/2-ERK1/2-Fra1-ZEB1/2 pathway . The insights gained from this study will fundamentally advance our understanding of the oncogenic functions of PLK1 and the molecular basis of PCa development and metastatic dissemination , which have the potential to facilitate optimization of treatment regimens targeting PLK1 signaling to significantly enhance anticancer efficacy . Furthermore , our novel findings may be generalizable to many other cancers since preliminary clinical evidence suggests that PLK1 is involved in the development of several other human cancers .
The cell lines used were as follows: RWPE-1 ( prostate epithelial cells ) ; LNCaP , LAPC4 , PC3 , C4-2B , and DU145 ( prostate cancer cells ) ; 293T ( human embryonic kidney cells ) ; HeLa ( human cervical adenocarcinoma cells , obtained from ATCC , Manassas , VA ) ; PrEC ( human primary prostate epithelial cells , obtained from Lonza , Basel , Switzerland ) ; and ARCaPE and ARCaPM ( Novicure Biotechnology , Birmingham , AL ) . All cell lines were tested and found to be free of mycoplasma and were cultured for no more than 10 passages according to the manufacturer’s recommendations . All cell types were checked for proper morphology prior to every experiment and consistently monitored for changes in cell replication that might suggest Mycoplasma contamination . Cell synchronization was performed as described previously ( Yuan et al . , 2014 ) . The inhibitors used were as follows: RO5126766 ( Active Biochem , Maplewood , NJ ) , nocodazole ( Sigma-Aldrich , St . Louis , MO ) , BI2536 ( Selleck Chemicals , Houston , TX ) , cycloheximide ( Sigma-Aldrich ) , MG132 ( Sigma-Aldrich ) , chloroquine ( Sigma-Aldrich ) , recombinant human TGF-β1 ( Thermo Fisher Scientific , Grand Island , NY ) , and recombinant human EGF ( BD Biosciences ) . Baculovirus encoding human WT PLK1 and kinase inactive ( K82M , KM ) mutants were generous gifts from Dr . R . L . Erikson ( Harvard University , Boston , MA ) . Flag-PLK1 WT , the constitutively active form ( TD ) and KM were subcloned into the pLVX-AcGFP-N1 lentiviral vector ( Clontech , Mountain View , CA ) . Human full-length PLK1 WT and KM mutant in the pRc/CMV vector were a gift from Dr . E . A . Nigg ( Max Planck Institute of Biochemistry ) . pGEX-4T-1-PLK1-PBD WT and MUT ( mutant , H538A/K540M ) were kindly provided by Dr . MB Yaffe ( Massachusetts Institute of Technology , Cambridge , MA ) . pcDNA 3 . 0-Flag-CRAF was a kind gift from Dr . KC Yeung ( University of Toledo , Toledo , OH ) . The fragment of CRAF spanning 275-648aa was cloned into pET-28a ( + ) . MISSION lentiviral shRNAs for PLK1 ( #1: TRCN0000121072; Clone ID: NM_005030 . 3-1893s1c1 and #2: TRCN0000121323; Clone ID: NM_005030 . 3-1073s1c1 ) , ZEB1 ( TRCN0000017565; Clone ID: NM_030751 . 2-572s1c1 ) , ZEB2 ( TRCN0000013530; Clone ID: NM_014795 . 2-2202s1c1 ) , Fra1 ( TRCN0000019539; Clone ID: NM_005438 . 2-780s1c1 ) , and CRAF ( TRCN0000001068 , Clone ID: NM_002880 . x-1529s1c1 ) were from Sigma-Aldrich . siRNA duplexes against human PLK1 ( access no . J-003290-09-0005 and J-003290-10-0005 ) , and negative control siRNA ( accession no . D-001810-10-05 ) were purchased from Dharmacon Research . The cell line overexpressing PLK1 was produced by a third-generation lentiviral system ( pLVX-AcGFP-N1 , pMDLg/pRRE , pRSV-Rev , and pMD2 . G ) . The 2 shRNAs were produced by a second-generation lentiviral system ( pLKO . 1 , psPAX2 , and pMD2 . G ) . Briefly , a lentiviral vector containing the shRNA or a transgene along with packing vector and envelope vector were cotransfected into 293T cells using Lipofectamine 2000 ( Thermo Fisher Scientific ) according to the manufacturer’s instructions . The supernatants containing virus particles were collected , filtered , and concentrated . Suspended virus was applied to target cells with 8 μg/mL polybrene . The cells were selected in the presence of puromycin ( 2 μg/mL for prostate cancer cell lines; 0 . 5 μg/mL for RWPE-1 and PrEC cells ) 72 hr after infection . Transient transfections of ARCaPE cells with a control or a PLK1 expression vector were performed using Lipofectamine 2000 per the manufacturer’s protocol . For siRNA depletion experiments , ARCaPM cells were transfected with these siRNA duplexes using DharmaFECT1 siRNA transfection reagent following the manufacturer’s instructions . The antibodies used in this study were: anti-PLK1 ( clone 36–298 , Thermo Fisher Scientific , 1:1000 ) , anti−phospho-PLK1 ( T210 ) ( no . 5472 , Cell Signaling Technology , Danvers , MA , 1:500 ) , anti−Flag ( F3165 , M2 , Sigma-Aldrich , 1:5000 ) , anti–β-actin ( SAB4200248 , Sigma-Aldrich , 1:5000 ) , anti–E-cadherin ( no . 610181 , BD Biosciences , Franklin Lakes , NJ , 1:2000 ) , anti–N-cadherin ( ab12221 , Abcam , Cambridge , MA , 1:1000 ) , anti−Vimentin ( V6389 , Sigma-Aldrich , 1:1000 ) , anti−Cytokeratin 19 ( GTX27754 , GeneTex , Irvine , CA , 1:1000 ) , anti−SM22α ( ab10135 , Abcam , 1:1000 ) , anti−Fibronectin ( F3648 , Sigma-Aldrich , 1:2000 ) , anti−ERK ( no . 4695 , Cell Signaling Technology , 1:500 ) , anti–phospho-ERK ( Thr202/Tyr204 ) ( no . 4370 , Cell Signaling Technology , 1:500 ) , anti−MEK1/2 ( no . 9122S , Cell Signaling Technology , 1:500 ) , anti–phospho-MEK1/2 ( Ser217/221 ) ( 9121S , Cell Signaling Technology , 1:500 ) , anti−CRAF ( sc-133 , Santa Cruz Biotechnology , Dallas , TX , 1:500 ) , anti–phospho-CRAF ( S338 ) ( 9427 , Cell Signaling Technology , 1:500 ) , anti–phospho-CRAF ( S339 ) ( Bs-5652R , Bioss , Wobum , MA , 1:250 ) , anti−ZEB1 ( no . 5825 , ProSci , 1:250 ) , anti−ZEB2 ( AP12086b , ABGENT , San Diego , CA , 1:250 ) , anti−Fra1 ( sc-28310 , Santa Cruz Biotechnology , 1:500 ) , anti−Snail ( no . 4719 , Cell Signaling Technology , 1:500 ) , and anti−PSA ( sc-7316 , Santa Cruz Biotechnology , 1:1000 ) . Cells were lysed in RIPA buffer ( 50 mM Tris-HCl [pH 7 . 5] , 150 mM NaCl , 0 . 1% SDS , 1% Nonidet P-40 , 1% Na-deoxycholate , and 1 mM EDTA ) containing protease and phosphatase inhibitors for 20 min at 4°C . Cell lysates were centrifuged for 10 min at 4°C . For immunoprecipitation or co-immunoprecipitation experiments , cleared lysates ( 1–2 mg protein ) were immunoprecipitated with the indicated polyclonal antibodies and protein A-Sepharose for 1 hr . Proteins were separated on a 7 . 5% SDS-PAGE gel , transferred to Immobilon P membranes , and immunoblotted with the indicated antibodies . In order to label AJ/TJ and cytoskeletal proteins , cells were fixed with 100% methanol for 20 min at -20°C . Fixed cell monolayers were washed with HBSS , blocked with 1% bovine serum albumin in HBSS ( blocking buffer ) for 60 min at room temperature , and incubated for another 60 min with primary antibodies diluted in blocking buffer . Cells were then washed , incubated for 60 min with Alexa-Fluor−conjugated secondary antibodies , rinsed with blocking buffer , and mounted on slides with ProLong Gold Antifade Reagent ( Thermo Fisher Scientific ) . Fluorescently-labelled cell monolayers were examined using a Zeiss LSM700 laser scanning confocal microscope ( Zeiss Microimaging Inc . , Thornwood , NY ) . The Alexa Fluor 488 and 555 signals were imaged sequentially in frame-interlaced mode to eliminate cross-talk between channels . The images were processed using Zen 2011 software and Adobe Photoshop . Images shown are representative of at least 3 experiments , with multiple images taken per slide . Antibodies used for immunofluorescence labeling: anti−E-cadherin ( no 610181 , BD Bioscience , 1:300 ) , anti−ZO_1 ( no 40–2200 , Thermo Fisher Scientific , 1:300 ) , anti−non-muscle myosin IIB ( no . 3404 , Cell Signaling Technology , 1:400 ) , anti–β-catenin ( C2206 , Sigma-Aldrich , 1:600 ) , and anti–JAM-A ( clone J10 . 4 , provided by Dr . Charles Parkos , University of Michigan , 1:200 ) . Total RNA was isolated using TRIzol ( Thermo Fisher Scientific ) and reverse transcribed into cDNA using SuperScript III First-Strand Synthesis System ( Life Technologies ) according to the manufacturer’s instructions . Real-time qPCR was performed using an ABI 7900HT thermal cycler and the FastStart Universal SYBR Green Master ( Roche , Basel , Switzerland ) . The data were normalized to the amount of GAPDH transcript . The primer sequences were as follows: E-cadherin: 5'-ACAGCCCCGCCTTATGATT-3' ( forward ) and 5'-TCGGAACCGCTTCCTTCA-3' ( reverse ) ; Cytokeratin 19: 5'-GGTCATGGCCGAGCAGAA-3' ( forward ) and 5'-TTCAGTCCGGCTGGTGAAC-3' ( reverse ) ; N-cadherin: 5'-TGGGAATCCGACGAATGG-3' ( forward ) and 5'-GCAGATCGGACCGGATACTG-3' ( reverse ) ; Fibronectin: 5'-CATGAGACTGGTGGTTACATGTTAGA-3' ( forward ) and 5'-GCATGATCAAAACACTTCTCAGCTA-3' ( reverse ) ; Vimentin: 5'-AATGACCGCTTCGCCAACT-3' ( forward ) and 5'- ATCTTATTCTGCTGCTCCAGGAA-3' ( reverse ) ; SM22α: 5'-GGCATGAGCCGCGAAGT-3' ( forward ) and 5'-TCCTCCAGCTCCTCGTCATACT-3' ( reverse ) ; SNAIL: 5'-ACCCCAATCGGAAGCCTAAC-3' ( forward ) and 5'-GCTGGAAGGTAAACTCTGGATTAGA-3' ( reverse ) ; SLUG: 5'- CAGCTACCCAATGGCCTCTCT-3' ( forward ) and 5'-GGACTCACTCGCCCCAAAG-3' ( reverse ) ; ZEB1: 5'-CAAATGTGGAAAGCGCTTCTC-3' ( forward ) and 5'-GTAGGAGTAGCGATGATTCATGTGTT-3' ( reverse ) ; ZEB2: 5'-CGCATTTCCCCCTGCTACT-3' ( forward ) and 5'-TGGTCGTAGCCCAGGAATACTG-3' ( reverse ) ; E47 ( TCF3 ) : 5'-GCGGAACCTGAATCCCAAA-3' ( forward ) and 5'-CACACCTGACACCTTTTCCTCTT-3' ( reverse ) ; Fra1: 5'-GCCGCCCTGTACCTTGTATC-3'; ( forward ) and 5'-CAGTGCCTCAGGTTCAAGCA-3' ( reverse ) ; TWIST: 5'-GGAGTCCGCAGTCTTACGAG-3' ( forward ) and 5'-TCTGGAGGACCTGGTAGAGG-3' ( reverse ) ; AR: 5'-TCACAGCCTGTTGAACTCTTC-3' ( forward ) and 5'-ACCTACTTCCCTTACCCCGCCT -3' ( reverse ) ; GAPDH: 5'-GAAATCCCATCACCATCTTCCA-3' ( forward ) and 5'-CCAGCATCGCCCCACTT-3' ( reverse ) . Cells were plated on collagen-I–coated slides . Once cells grew to confluence , a wound was introduced by scratching the confluent monolayer with a pipette tip . The images were acquired at 0 , 24 , and 48 hr post-wounding . The relative surface area travelled by the leading edge was calculated using TSscratch software ( Gebäck et al . , 2009 ) . The figure shows representative images of 3 independent experiments performed in triplicate . The transwell invasion assay was performed using Matrigel Invasion Chambers ( BD Bioscience ) as we previously described ( Fu et al . , 2003 ) . RWPE-1 cells ( 2×105 cells/well ) , PrEC cells ( 2×105 cells/well ) , or prostate cancer cell lines ( PC3 , DU125 , and C4-2B; 5×104 cells/well ) were plated into culture inserts coated with Matrigel . Cells were sparsely plated on a collagen-coated , 35-mm culture dish and were imaged for 9 hr at 3-min time intervals . For all time-lapse recordings , the culture dish was placed in a microincubator to maintain proper environmental conditions ( 37°C , pH7 . 4 ) . All images were acquired using a Zeiss Cell Observer Spinning Disc confocal microscope and analyzed using NIH ImageJ software . Velocity measurements and tracking diagrams were made using the manual tracking plugin for NIH ImageJ software and Adobe Photoshop . Thirty random cells per experiment were analyzed . The experiments were repeated 3 times . Glutathione-sepharose resin-coupled GST-PLK1 PBD ( WT and MUT ) or the GST control was incubated with cell lysates for 1 hr at 4°C . The resins were washed 4 times using NETN buffer ( 150 mM NaCl , 1 mM EDTA , 20 mM Tris pH 8 , 0 . 5% NP-40 ) . Resin-bound complexes were eluted by boiling , separated by SDS-PAGE , and analyzed by Western blotting . Cells were collected , washed with PBS and fixed with ice cold 70% ethanol for at least 1 hr . Cells were then washed twice in PBS and treated for 30 min at 37°C with RNase A at 5 μg/mL and PI at 50 μg/mL , and analyzed on a FACScan flow cytometer ( Becton Dickinson ) . The percentage of cells in different cell cycle phases was calculated using ModFit LT for Mac ( BD Biosciences ) . Upon genetic modification of PLK1 expression ( downregulation of endogenous PLK1 and re-expression of exogenous PLK1 ) in DU145 and C4-2B cells , the entire cell population was subjected to double staining for FITC-annexin V and PI using a FITC-annexin V apoptosis detection kit ( BD Biosciences ) and analyzed by flow cytometry for apoptotic events according to the manufacturer’s instructions . Cells were plated in 96-well tissue culture plates at a density of 1 x 103cells per well . The cell viabilities were assessed by means of a CellTiter 96 Aqueous One solution cell proliferationassay ( Promega ) accordingto the manufacturer's instructions . Experiments were performed in triplicate . Human PLK TD and KM were expressed in the baculovirus/insect cell system as we described previously ( Fu et al . , 2008 ) . His-CRAF 275–648 aa fusion protein was expressed in the Escherichia coli BL21 strain . Kinase and substrates ( CRAF proteins ) were incubated in kinase buffer ( 20 mM Hepes , pH 7 . 4 , 150 mM KCl , 10 nM MgCl2 , 1 mM EGTA , 0 . 5 mM dithiothreitol , 5 mM NaF , 100 μM ATP , ) for 30 min at 30°C . Reactions were stopped by the addition of SDS sample buffer . Then samples were heated for 5 min to 95°C before analysis by SDS-PAGE and Western blot analysis with specific antibodies . Assays of colony formation in soft agar were performed using standard methods as we described previously ( Fu et al . , 2003 ) . Cells were plated ( 1×104 cells/well ) onto the previously prepared under layers . All experiments involving animals were performed with approval from the Virginia Commonwealth University Institutional Animal Care and Use Committee ( IACUC; protocol #: AD20282 ) . Control RWPE-1 or PLK1-overexpressing cells ( 1×106 ) were mixed with an equal volume of BD Matrigel Basement Membrane Matrix and injected subcutaneously into the flanks of NOD/SCID/γcnull mice ( 6-week-old male mice , from The Jackson Laboratory , stock no . 005557 ) . Seven mice were used per group . Mice were housed under specific-pathogen-free conditions . Tumor growth was observed for 5 weeks . Tumors were measured 3 times weekly with a caliper and their volumes calculated using the following formula: π × [length in millimeters] × [width in millimeters]/6 . At the end of the experiment , mice were euthanized and the tumors and lungs were removed and preserved for pathological examination . Formalin-fixed sections were stained with hematoxylin and eosin . Immunochemistry for PSA on paraffin-embedded tissue sections ( 5–6 μm ) was done using a catalyzed amplification system ( Dako , Carpinteria , CA ) as we described previously ( Fu et al . , 2009 ) . The statistical tests used are indicated in the figure legends . All statistical tests were performed using GraphPad Prism version 6 . 02 for Windows ( GraphPad Software ) . Data are expressed as mean ± s . e . m . , and p<0 . 05 was considered statistically significant . A two-tailed Student’s t-test was used to compare differences between treated groups and their paired controls . To compare the track distance and velocity between 2 groups , a two-tailed Mann-Whitney rank sum test was performed . Groups of 7 mice were used for the tumor xenograft experiments . No statistical method was used to pre-determine sample size . The experiments were not randomized . The investigators were blinded to allocation during the xenograft tumor model experiments . For other experiments , the investigators were not blinded to allocation during experiments and outcome assessment . | Living cells grow and divide via a series of events called the cell cycle . If this process is disturbed in animals , it can lead to cancer . In the later stages of tumor development , cancer cells frequently change their structure and behavior in a process called the epithelial-to-mesenchymal transition ( EMT ) , which enables them to migrate and form new tumors around the body . A protein called Polo-like kinase 1 ( PLK1 ) plays a central role in the cell cycle and has been implicated in the development of various cancers , including prostate cancer . Recent evidence suggests that PLK1 also has other roles in cells , but it is not clear how much they contribute to the development of cancer . Wu et al . studied PLK1 in human cells and mice and showed that manipulating healthy prostate epithelial cells to produce more PLK1 caused the cells to go through the EMT and increased their ability to migrate . In other experiments , the levels of PLK1 in prostate cancer cells were deliberately lowered , which caused the cells to change to become more like epithelial cells and become less mobile . Wu et al . also investigated how PLK1 promotes the EMT and cell migration . These experiments showed that PLK1 activates a protein that controls an important chain of signaling events called the ERK/MAPK pathway , which is essential for cell growth and migration . Wu et al . ’s findings uncover a new role for PLK1 in promoting the spread of cancer cells around the body . A future challenge is to find out how PLK1 is regulated in people with prostate cancer and whether the EMT is involved in promoting other processes in cancer cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2016 | Polo-like kinase 1 induces epithelial-to-mesenchymal transition and promotes epithelial cell motility by activating CRAF/ERK signaling |
Located in the midbrain , the inferior colliculus ( IC ) is the hub of the central auditory system . Although the IC plays important roles in speech processing , sound localization , and other auditory computations , the organization of the IC microcircuitry remains largely unknown . Using a multifaceted approach in mice , we have identified vasoactive intestinal peptide ( VIP ) neurons as a novel class of IC principal neurons . VIP neurons are glutamatergic stellate cells with sustained firing patterns . Their extensive axons project to long-range targets including the auditory thalamus , auditory brainstem , superior colliculus , and periaqueductal gray . Using optogenetic circuit mapping , we found that VIP neurons integrate input from the contralateral IC and the dorsal cochlear nucleus . The dorsal cochlear nucleus also drove feedforward inhibition to VIP neurons , indicating that inhibitory circuits within the IC shape the temporal integration of ascending inputs . Thus , VIP neurons are well-positioned to influence auditory computations in a number of brain regions .
The inferior colliculus ( IC ) is the hub of the central auditory pathway . Nearly all ascending output from the lower auditory brainstem and a large descending projection from the auditory cortex converge in the IC ( Adams , 1979; Glendenning and Masterton , 1983; Oliver , 1987; Oliver , 1984; Winer et al . , 1998 ) . In turn , the IC provides the main auditory input to the thalamocortical system ( Calford and Aitkin , 1983 ) . Neurons in the IC exhibit selective responses to the spectral and temporal content of sounds and perform computations important for sound localization and the identification of speech and other communication sounds ( Felix et al . , 2018; Winer and Schreiner , 2005 ) . Despite these critical functions , we have limited knowledge about the organization and function of neural circuits in the IC . This is because probing neural circuits requires the ability to identify and manipulate specific classes of neurons , but IC neurons have proven difficult to delineate into distinct classes . Anatomical studies have shown that IC neurons have disc-shaped or stellate morphologies ( Malmierca et al . , 1993; Meininger et al . , 1986; Oliver and Morest , 1984 ) . Disc-shaped neurons maintain their dendritic arbors within isofrequency lamina and make up the majority of neurons in the tonotopically organized central nucleus of the IC ( ICc ) . Stellate neurons in the ICc extend their dendritic arbors across laminae and are therefore thought to integrate information across sound frequencies ( Oliver et al . , 1991 ) . Both disc-shaped and stellate cells can be glutamatergic or GABAergic , an indication that each of these morphological groups consists of at least two neuron types ( Oliver et al . , 1994 ) . Based on soma size and extracellular markers , IC GABAergic neurons have been divided into four classes ( Beebe et al . , 2016 ) . Among these , ‘large GABAergic’ neurons are the one consistently identified neuron type in the IC ( Geis and Borst , 2013; Ito et al . , 2015; Ito et al . , 2009; Ito and Oliver , 2012 ) . However , there are currently no known molecular markers specific for large GABAergic neurons ( Schofield and Beebe , 2019 ) . Defining IC neuron types based on physiology has also proven difficult . IC neurons exhibit diverse responses to tones , but a comprehensive study showed that these responses form a continuum and cannot be used on their own to define functionally significant groups of neurons ( Palmer et al . , 2013 ) . In addition , disc-shaped neurons could not be divided into distinct groups by matching their morphology with their in vivo physiology ( Wallace et al . , 2012 ) . Similarly , GABAergic and glutamatergic IC neurons exhibit overlapping and equally diverse responses to sounds ( Ono et al . , 2017 ) . In vitro recordings have shown that IC neurons exhibit diverse firing patterns , but these firing patterns do not correlate with neuronal morphology or neurotransmitter phenotype ( Ono et al . , 2005; Peruzzi et al . , 2000; Reetz and Ehret , 1999; Sivaramakrishnan and Oliver , 2001 ) . In many brain regions , a multidimensional analysis that includes molecular markers has proven key to identifying neuron classes ( Ascoli et al . , 2008; Tremblay et al . , 2016; Zeng and Sanes , 2017 ) . Here , by combining molecular , morphological , and physiological analyses , we identify vasoactive intestinal peptide ( VIP ) neurons as a novel class of IC principal neurons . Our results show that VIP neurons are glutamatergic stellate neurons and represent approximately 20% of the stellate neurons in the ICc . VIP neurons are labeled in the VIP-IRES-Cre mouse line and are present in the major subdivisions of the IC , with a higher prevalence in caudal regions of the ICc , where most ascending input originates from monaural brainstem nuclei . Using viral tract tracing , we found that VIP neurons project to multiple auditory and non-auditory areas , demonstrating that a single neuron class can participate in most of the major projection pathways out of the IC . Using Channelrhodopsin-assisted circuit mapping ( CRACM ) , we found that VIP neurons integrate input from the contralateral IC and the auditory brainstem . Input from the auditory brainstem also drove local , feedforward inhibition onto VIP neurons . Thus , our data reveal a novel circuit motif that may control the temporal summation of ascending input to the IC . Together , these results represent a critical step toward determining how defined neural circuits in the IC support sound processing .
By crossing VIP-IRES-Cre mice with Ai14 reporter mice , we obtained mice in which VIP+ neurons expressed the fluorescent protein tdTomato . Distribution of VIP+ neurons throughout the brain matched the description provided by Taniguchi et al . ( 2011 ) . There were very few or no VIP-expressing neurons in auditory centers outside of the IC ( including cochlear nucleus , superior olivary complex , nuclei of the lateral lemniscus , nucleus of the brachium of the IC and medial geniculate nucleus ) , with the exception of auditory cortex , where sparse neurons matching descriptions of VIP-expressing interneurons were labeled . The present report focuses on the IC , which contained many VIP-expressing neurons . Figure 1 shows the distribution of VIP+ neurons ( magenta ) in transverse sections through the IC . Labeled neurons were present throughout most of the rostro-caudal extent of the IC , including in the ICc , ICd , and IClc , but were most numerous in the caudal regions . Labeled neurons were rare or absent in the IC rostral pole and intercollicular tegmentum . Previous studies have shown that IC neurons are either glutamatergic or GABAergic ( Merchán et al . , 2005; Oliver et al . , 1994 ) . To investigate the neurotransmitter content of VIP neurons , we performed immunohistochemical staining against GAD67 , a marker for GABAergic neurons , in brain slices from three VIP-IRES-Cre x Ai14 animals , aged P58 ( Figure 2A ) . We then counted tdTomato+ neurons and GAD67-labeled cell bodies in one caudal and one rostral IC slice per animal . Because there were no regional differences in GAD67 staining among VIP neurons , neurons located in the ICc , ICd , and IClc were combined for this analysis . Across 793 tdTomato+ neurons , only 10 neurons co-labeled with GAD67 ( 1 . 3% , Table 1 ) . We suspect the 10 tdTomato+ neurons that stained for GAD67 were non-specifically labeled or represent rare cases of tdTomato expression in non-VIP neurons . These data suggest that VIP neurons are a subgroup of glutamatergic neurons in the ICc , ICd , and IClc . To determine the percentage of neurons in the ICc and IC shell ( ICd plus IClc ) that are VIP neurons , we performed immunostaining with anti-NeuN , a neuron-selective antibody previously shown to label most or all neurons in the IC ( Beebe et al . , 2016; Foster et al . , 2014; Mellott et al . , 2014 ) ( Figure 2B ) , and anti-bNOS , a marker that differentiates the ICc from the IC shell regions ( Coote and Rees , 2008 ) . Coronal IC sections from two VIP-IRES-Cres x Ai14 mice were stained with anti-NeuN and anti-bNOS . Three sections per mouse were analyzed: one caudal , one middle , and one rostral . To ensure unbiased counting of neurons , we applied the optical fractionator method , a design-based stereology approach ( see Materials and methods; West et al . , 1991 ) . Accordingly , we used systematic random sampling to collect confocal image stacks with a 63x objective at evenly spaced intervals from each IC section . Each image stack was inspected to determine the boundaries of the slice , and guard zones were set at the top and bottom of the slice to delineate a central , 15 µm-thick section of the slice for subsequent analysis . Within this 15 µm-thick region , we separately marked NeuN+ neurons and tdTomato+ neurons , then overlaid the NeuN and tdTomato images to determine the number of double-marked cells . We also collected tile scan images of each IC section analyzed using a 20x objective and used bNOS staining to determine the border separating the ICc from the IC shell . The 63x z-stacks were aligned to these tile scans , and counted neurons were assigned to the ICc or IC shell . The results of the stereological analysis revealed that VIP neurons represented a larger portion of neurons in the ICc ( 3 . 5 ± 1 . 0% ) than in the IC shell ( 1 . 5 ± 0 . 2%; two-tailed t-test , t ( 79 ) = 2 . 86 , p = 0 . 005; Table 2 ) . In addition , the prevalence of VIP neurons was highest in caudal regions of the ICc and the IC shell and tended to decrease in a caudal to rostral gradient . This trend was significant when comparing the caudal ICc to the rostral ICc ( one-way ANOVA , F ( 2 , 37 ) = 7 . 27 , p = 0 . 002 , Tukey’s post hoc , p=0 . 001 ) and the caudal IC shell to the middle IC shell ( one-way ANOVA , F ( 2 , 38 ) = 3 . 88 , p = 0 . 03 , Tukey’s post hoc , p = 0 . 03 ) . These results suggest that VIP neurons are not evenly distributed along the rostral-caudal extent of the isofrequency lamina of the ICc and , according to the functional domain hypothesis , may be more likely to receive input from ascending afferents that preferentially target the caudal ICc ( Cant and Benson , 2006; Loftus et al . , 2010; Oliver et al . , 1997 ) . The functional implications of this anatomical arrangement will be addressed in more detail in the Discussion . Overall , a combined count of VIP neurons from the ICc and IC shell showed that VIP neurons represent 2 . 3 ± 0 . 3% of the total population of neurons in the mouse IC ( 208 of 9304 neurons , n = 69 systematic random samples ) . Next , we investigated the firing pattern and intrinsic physiology of VIP neurons by targeting whole cell patch clamp recordings to tdTomato+ neurons in brain slices from VIP-IRES-Cre x Ai14 mice . For the majority of neurons , neuronal location relative to the IC subdivisions was determined post hoc , during retrieval of neuronal morphology ( see below and Materials and methods ) . Recordings made from the ICc , ICd , and IClc were lumped together for this experiment because there were no clear differences in VIP neuron physiology across these subdivisions of the IC . VIP neurons had a resting membrane potential of −69 . 5 mV ± 4 . 4 mV ( n = 216 , corrected for liquid junction potential ) . In response to a current step protocol with hyperpolarizing and depolarizing current injections , VIP neurons showed minimal to no voltage sag to hyperpolarizing current steps and a sustained firing pattern of action potentials to depolarizing current steps ( Figure 3A1 , A2 ) . Neurons were classified as sustained if their spike frequency adaptation ratio ( SFA ) was less than 2 ( Peruzzi et al . , 2000 ) . The SFA ratio was calculated by dividing the last interspike interval by the first for a depolarizing current step that elicited ~10 spikes . 90 . 3% ( 214 of 237 ) of patched VIP neurons exhibited a sustained firing pattern , 8 . 4% ( 20 of 237 ) showed an adapting firing pattern ( SFA ratio >= 2 ) , and 1 . 3% ( 3 of 237 ) of VIP neurons had a transient firing pattern ( firing stopped before the end of the current step ) . To compare the physiology of VIP neurons to that of the general population of neurons in the IC , we patched neurons in IC slices of C57BL/6J mice in a random , non-targeted approach as a control . Neurons patched with the non-targeted ( NT ) approach showed a higher diversity in firing patterns , with a higher proportion of transient neurons ( 21 . 8% , 12 out of 55 ) and adapting neurons ( 16 . 4% , 9 out of 55 ) when compared to VIP neurons . Sustained firing neurons were the most prevalent group in NT recordings ( 58 . 2% , 32 out of 55 ) . Additionally , 3 . 6% ( 2 of 55 ) of randomly patched IC neurons fired only one spike at the onset of the depolarizing current step . This firing pattern was never observed in VIP neurons . The intrinsic physiology of VIP neurons also differed significantly from the general population of IC neurons ( see Figure 3B ) . VIP neurons on average had a higher peak input resistance ( Rpk ) than non-targeted neurons ( mean ± SD: VIP 242 . 1 ± 139 . 4 MΩ vs NT 191 . 1 ± 161 . 4 MΩ , p = 0 . 0003 , Wilcoxon rank sum test ) , a higher steady-state input resistance ( Rss ) ( mean ± SD: VIP 239 . 7 ± 170 . 7 MΩ vs NT 153 . 4 ± 153 . 2 , p = 0 . 0001 , Wilcoxon rank sum test ) , a slower membrane time constant ( mean ± SD: VIP 15 . 0 ± 8 . 8 ms vs NT 9 . 7 ± 7 . 6 MΩ , p = 6 . 8*10−7 , Wilcoxon rank sum test ) , lower rheobase values ( mean ± SD: VIP 67 . 8 ± 96 . 2 pA vs NT 120 . 0 ± 100 . 9 pA , p = 0 . 013 , Wilcoxon rank sum test ) , and a much less pronounced voltage sag ( mean ± SD: VIP 0 . 87 ± 0 . 16 vs NT 0 . 75 ± 0 . 21 , p = 0 . 0003 , Wilcoxon rank sum test ) . Most striking , the SFA ratio of VIP neurons was tightly clustered at 1 . 47 ± 1 . 62 , whereas SFA of NT neurons showed a significantly higher value and spread ( 3 . 62 ± 5 . 43 , mean ± SD , p = 2 . 07*10−7 ) . Although statistically different from the general neuronal population in the IC and showing a tightly clustered SFA ratio , the intrinsic physiology of VIP neurons still showed some level of variability . In the lower auditory brainstem , it has been found that the intrinsic physiology of some neurons varies along the tonotopic axis ( Baumann et al . , 2013; Hassfurth et al . , 2009 ) . We therefore hypothesized that the intrinsic physiology of VIP neurons in the ICc varied along the tonotopic axis of the ICc . During patch clamp experiments , VIP neurons were passively filled with biocytin via the internal solution . Slices were fixed and stained post hoc with a streptavidin-Alexa Fluor conjugate ( see Materials and methods ) . We then used confocal imaging to map the location of the recorded neurons relative to a two-dimensional ( medial-lateral and dorsal-ventral ) coordinate system superimposed on the left IC ( n = 61 neurons; Figure 4A ) . Correlations between intrinsic physiology and location in the ICc were tested by fitting a plane to scatter plots of intrinsic parameters versus medial-lateral and dorsal-ventral coordinates ( Figure 4B ) . Because the ICd and IClc are not tonotopically organized , only neurons located in the ICc were included in this analysis . We found that variability in the intrinsic physiology of VIP neurons was at least partially correlated to their location within the coronal plane of the ICc . This was particularly true for the voltage sag ratios , which measure hyperpolarization-activated cation current ( Ih ) . Approximately one quarter of the variability in sag ratios was explained by location in the ICc ( sag ratio at −91 mV: R = 0 . 536 , R2adj = 0 . 262 , p = 1 . 24×10−05 , n = 60 , Figure 4B4; sag ratio at −111 mV: R = 0 . 516 , R2adj = 0 . 233 , p = 0 . 0002 , n = 47 , Figure 4B5 ) . A significant but smaller portion of the variability in membrane time constant was explained by location in the ICc ( τ: R = 0 . 343 , R2adj = 0 . 088 , p = 0 . 007 , n = 61; Figure 4B3 ) . There was also a significant relationship between the steady-state input resistance of VIP neurons and location in the ICc and a trend toward a relationship between peak input resistance and location ( Rss: R = 0 . 328 , R2adj = 0 . 076 , p = 0 . 011 , n = 60; Rpk: R = 0 . 227 , R2adj = 0 . 018 , p = 0 . 084 , n = 60; Figures 4B1 and 2 ) . The tonotopic axis of the ICc runs along a dorsolateral ( low frequency ) to ventromedial ( high frequency ) axis ( Malmierca et al . , 2008; Portfors et al . , 2011; Stiebler and Ehret , 1985; Willott and Urban , 1978 ) . For each of the above intrinsic parameters , values tended to be lower , indicating faster membrane properties , at more dorsolateral locations and higher , indicating slower membrane properties , at more ventromedial locations . Combined , these results suggest that variability in the intrinsic physiology of VIP neurons is at least in part due to their localization along the tonotopic axis of the ICc and that the membrane properties of VIP neurons tend to be faster in lower frequency regions of the ICc . The streptavidin staining of biocytin-filled VIP neurons allowed for a detailed analysis of morphology . In total , we recovered the morphology of 55% of patched VIP neurons ( n = 100 of 183 ) . Nearly all ( 81/86 = 94 . 2% ) VIP neurons had spiny dendrites ( Figure 5A , B insets ) . This contrasts sharply with the 28% ( 12 of 43 neurons ) of neurons that had spiny dendrites in non-targeted recordings from C57BL/6J mice ( Figure 5C ) , suggesting that VIP neurons represent a subset of IC cells . Nonetheless , dendritic spines can be present on a variety of cell types , including stellate and disc-shaped cells ( Herrera and Correa , 1988; Paloff et al . , 1992; Willard and Ryugo , 1983 ) . On average , VIP neurons had five primary dendrites ( mean ± SD: 4 . 77 ± 1 . 38 ) that spread out in all directions from the soma , consistent with a stellate morphology . This is unsurprising in the ICd and IClc , where stellate morphology predominates , but warranted further analysis in the ICc , where stellate cells can have oriented dendritic trees but are outnumbered by the more highly oriented disc-shaped cells ( Malmierca et al . , 1993; Oliver and Morest , 1984; Willard and Ryugo , 1983 ) . There were no obvious differences between the morphology of VIP neurons in the ICc versus the ICd or IClc , but this question warrants more detailed analysis in a future study . Figure 5D shows the variability in the morphology of VIP neurons located in the ICc . Neurons are displayed as they would appear in a coronal slice of the left IC viewed from a caudal perspective . Oliver and colleagues distinguished disc-shaped from stellate neurons by calculating the length to width ratio of the dendritic arbor: neurons with a ratio <3 were stellate and those with a ratio >= 3 were disc-shaped ( Oliver et al . , 1991 ) . Applying this classification to our sample , 93% of VIP neurons in the ICc ( 39 of 42 neurons ) had a length to width ratio <3 , therefore being classified as stellate ( Figure 5H ) . Only three VIP neurons from the ICc had a length to width ratio >3 . These results demonstrate that the dendritic arbors of VIP neurons are not as highly oriented as disc-shaped neurons , again consistent with the hypothesis that VIP neurons are a class of stellate neurons . Although less oriented than disc-shaped cells , the VIP dendrites tended to show some orientation that could influence the range of frequencies that converge on the cell . To measure the orientation of ICc VIP neurons in relation to the isofrequency laminae , which in mouse run in a ~45° angle through the ICc ( Stiebler and Ehret , 1985 ) , we identified the longest and second longest axis of each neuron through principal component analysis and plotted the orientation of these axes on a standardized model of the IC ( Figure 5E ) . No preferred orientation was apparent ( Figure 5E , combined ) . Only 17% of VIP neurons ( 7 of 42 ) had their longest axis oriented within ± 15° of the 45° laminar plane , indicating that the dendritic arbors of most VIP neurons ( 83% , 35 of 42 ) may be positioned to cross one or more isofrequency laminae in the ICc ( Figure 5F ) . To quantify this , we calculated the length the dendritic arbor extended perpendicular to a 45° laminar plane . The dendritic arbors of 83% of ICc VIP neurons ( 35 of 42 ) spread more than 100 µm perpendicular to the laminar plane , and more than 36% ( 15 of 42 ) spread more than 200 µm across the laminar plane ( Figure 5G ) . Previous work in rats has shown that the isofrequency laminae have a center-to-center distance that ranges from 90 to 150 µm , while neurons contained within a lamina had a thickness ranging from 30 to 70 µm ( Malmierca et al . , 1993 ) . If we assume that laminar dimensions in mouse are no thicker than those in rats , our results indicate that the dendritic fields of VIP neurons usually extend beyond at least one isofrequency lamina , consistent with the conclusion that VIP neurons in the ICc are a class of stellate neurons . Previous estimates indicate that approximately 15 – 20% of ICc neurons are stellate ( Oliver et al . , 1991; Oliver and Morest , 1984 ) . Since our stereological analysis showed that VIP neurons represent 3 . 5% of ICc neurons , our results indicate that VIP neurons account for ~18 – 23% of ICc stellate neurons . Injections of an AAV encoding a Cre-dependent eGFP construct , AAV1 . CAG . FLEX . eGFP . WPRE . bGH , led to eGFP expression in VIP+ IC cells . Figure 6A shows a representative deposit site , with eGFP-positive neurons ( yellow ) located among a population of VIP+ cells ( magenta , labeled by cross-breeding the VIP-IRES-Cre mice with Ai14 reporter mice ) . Neurons that expressed eGFP routinely co-expressed tdTomato , confirming VIP expression by those neurons ( Figure 6B ) . Many tdTomato+ neurons did not express the eGFP , despite their intermingling with many virally-labeled neurons . eGFP-labeled axons were prominent within the injected IC , where the labeled boutons were located in the neuropil or in close apposition to IC somas , suggesting extensive contributions to local circuits ( Figure 6C ) . In addition , eGFP-labeled axons were present in several fiber tracts carrying projections from the IC , including the brachium of the IC , the commissure of the IC and the lateral lemniscus . Labeled axons and boutons were found in numerous auditory nuclei , including the contralateral IC , medial geniculate body and superior olivary complex ( Figure 6D–F ) . Additional targets included the nucleus of the brachium of the IC , the periaqueductal gray , and the superior colliculus ( not shown; details of termination patterns and terminal axon morphology will be described in a subsequent report ) . These data indicate that VIP+ IC neurons contribute to ascending , commissural and descending pathways from the IC . In addition to axonal projection patterns , the sources of synaptic input to a neuron class are an important predictor of neuronal function . Anatomical studies have shown that the IC receives ascending , descending , and commissural input , but , with the exception of large GABAergic neurons ( Ito et al . , 2015; Ito and Oliver , 2014 ) , it has rarely been possible to identify the sources and physiology of synaptic input to a specific class of neurons in the IC . This is largely because axons from multiple sources overlap in the IC , making it difficult to use electrical stimulation to selectively activate axons from specific presynaptic sources . In addition , electrical stimulation of commissural projections cannot differentiate between axons originating in the ipsilateral and contralateral IC . This is because electrical stimulation evokes both orthodromic and antidromic spikes , and , since most IC neurons have local axon collaterals ( Oliver et al . , 1991 ) , antidromic spikes can lead to synaptic release from the local collaterals of neurons ipsilateral to the recording site . Furthermore , the commissure contains axons from sources other than IC cells , including cells in the auditory cortex , sagulum , nuclei of the lateral lemniscus and superior paraolivary nucleus ( reviewed by Saldaña and Merchán , 2005 ) ; electrical stimulation in the commissure could also activate these axons . To overcome these obstacles , we turned to Channelrhodopsin-assisted circuit mapping ( CRACM ) ( Petreanu et al . , 2007 ) . With CRACM , it is possible to selectively activate a single population of presynaptic neurons by anatomically and/or molecularly restricting the expression of an optogenetic protein . In our experiments , we used stereotaxic coordinates and intracranial virus injections to anatomically restrict Chronos expression . Epi-fluorescence imaging confirmed that Chronos-GFP expression was always clearly limited to the right IC ( Figure 7B ) or the right DCN ( Figure 8A ) . Numerically , the contralateral IC provides the largest single source of input to the IC ( Moore , 1988 ) . We therefore first used CRACM to test whether VIP neurons in the ICc receive commissural input . Using stereotaxic , intracranial virus injections with AAV1 . Syn . Chronos-GFP . WPRE . bGH , we drove expression of GFP-tagged Chronos , a fast Channelrhodopsin variant ( Klapoetke et al . , 2014 ) , in the right IC . Note that we did not attempt to limit Chronos expression to a particular subdivision of the right IC . Visual inspection of Chronos-GFP fluorescence suggested that most of the somata labeled in the right IC were located in the ICc , but labeled somata were often also present in the ICd and occasionally the IClc . We then targeted recordings to VIP neurons in the contralateral ( left ) ICc ( Figure 7A ) . In each experiment , we used GFP fluorescence to verify transfection of the right IC ( Figure 7B ) and restricted our recordings to VIP neurons in areas of the left ICc where GFP-labeled axons were visible . Because commissural projections are a mixture of glutamatergic and GABAergic projections ( González-Hernández et al . , 1996; Hernández et al . , 2006; Nakamoto et al . , 2013; Saint Marie , 1996 ) , we used pharmacology to isolate EPSPs and IPSPs . Indeed , without adding receptor antagonists to the bath , postsynaptic potentials often were mixtures of IPSPs and EPSPs ( data not shown ) . In the presence of AMPA and NMDA receptor antagonists ( 10 µM NBQX and 50 µM D-AP5 , bath application ) , 2 – 5 ms flashes of blue light elicited IPSPs in 6 out of 12 neurons tested ( Figure 7C , left ) . IPSPs were completely abolished by the GABAA receptor antagonist gabazine ( 5 µM , Figure 7C , right; n = 6; amplitude , p = 0 . 006; rise time , p = 0 . 003; halfwidth , p = 0 . 001; membrane time constant , p = 0 . 003 , paired t-test ) . On average ( n = 6 ) , commissural IPSPs in ICc VIP neurons were small ( −1 . 53 mV ± 0 . 96 mV ) and had moderate 10 – 90% rise times ( 7 . 8 ms ± 2 . 1 ms ) , halfwidths ( 15 . 1 ms ± 6 . 8 ms ) and decay time constants ( 32 . 4 ms ± 17 . 0 ms ) ( Figure 7E ) . To investigate excitatory commissural inputs to ICc VIP neurons , recordings were carried out in the presence of GABAA and glycine receptor antagonists ( 5 µm gabazine and 1 µM strychnine , bath application , Figure 7D ) . We found that 2 – 5 ms flashes of blue light elicited EPSPs in 11 out of 27 neurons ( Figure 7D , left ) . On average ( n = 6 ) , commissural EPSPs in VIP neurons were small ( 1 . 52 mV ± 1 . 08 mV ) and had moderate 10 – 90% rise times ( 8 . 3 ms ± 4 . 3 ms ) , halfwidths ( 19 . 6 ms ± 7 . 6 ms ) and decay time constants ( 43 . 5 ms ± 16 . 8 ms ) ( Figure 7F ) . Adding the NMDA receptor antagonist D-AP5 to the bath significantly reduced the halfwidth of EPSPs ( 14 . 3 ms ± 4 . 7 ms , p = 0 . 006 ) and revealed a trend toward reducing the rise time ( 6 . 3 ms ± 1 . 6 ms , p = 0 . 09 ) and decay time constant ( 30 . 6 ms ± 7 . 3 ms , p = 0 . 06 ) of EPSPs ( ANOVA for repeated measurements with Tukey post-hoc test ) . The remainder of the EPSP was completely blocked by the AMPA receptor antagonist NBQX ( Figure 7F ) . These results suggest that VIP neurons in the ICc receive excitatory commissural input and express both AMPA and NMDA receptors at excitatory commissural synapses . Interestingly , commissural input activated NMDA receptors even though there was 1 mM Mg2+ in the bath and the somatic membrane potential was at or near the resting membrane potential throughout the recording . This may indicate that commissural synapses are located on the distal dendrites and/or the dendritic spines of VIP neurons . Combined , the commissural CRACM experiments show that VIP neurons in the ICc receive excitatory and inhibitory synaptic input from the contralateral IC . Surprisingly , although GABAergic neurons make up <= 20% of commissural projections ( Hernández et al . , 2006; Nakamoto et al . , 2013 ) , we found a higher connection probability for inhibitory commissural projections ( 50%; 6 out of 12 recordings performed in the presence of NBQX and D-AP5 ) , than for excitatory connections ( 41%; 11 out of 27 recordings performed in the presence of gabazine and strychnine ) . Due to the kinetics of optogenetic proteins and the possibility of activating optogenetic proteins in synaptic terminals , optogenetically-evoked PSPs may have slower kinetics than PSPs evoked by electrical stimulation ( Jackman et al . , 2014; Zhang and Oertner , 2007 ) . To test whether the kinetics of Chronos-evoked commissural PSPs were similar to those of electrically-evoked PSPs , we electrically stimulated the IC commissure while recording PSPs from VIP neurons in the contralateral IC . As indicated above , this raises the complication that electrically-evoked EPSPs were probably elicited both from the desired contralateral projections and from antidromic stimulation of ipsilateral axons . Compared to optogenetically-evoked EPSPs , electrically-evoked EPSPs ( n = 6 ) had similar amplitudes and rise times , but had smaller halfwidths and trended toward faster decay time constants ( amplitude: 2 . 4 mV ± 1 . 7 mV , p = 0 . 37; rise time: 6 . 28 ms ± 6 . 3 ms , p = 0 . 40; halfwidth: 10 . 3 ms ± 4 . 8 ms , p = 0 . 008*; decay time constant 28 . 4 ms ± 10 . 0 ms , p = 0 . 016 , two tailed t-test , *critical p value with Bonferroni correction for multiple comparisons = 0 . 0125 ) . Electrically-evoked IPSPs ( n = 6 ) showed no significant difference in amplitude , halfwidth or decay time constant , but trended toward faster rise times ( amplitude: −1 . 7 mV ± 1 . 3 mV , p = 0 . 78; rise time: 4 . 2 ms ± 2 . 3 ms , p = 0 . 025; halfwidth: 8 . 8 ms ± 7 . 3 ms , p = 0 . 16; decay time constant 36 . 8 ms ± 17 . 3 ms , p = 0 . 67 , two tailed t-test , critical p value with Bonferroni correction for multiple comparisons = 0 . 0125 ) . It should be noted that to position the stimulus electrode in the IC commissure contralateral to the recording site , these recordings had to be biased to VIP neurons in more rostral parts of the IC where the commissure was clearly visible , a bias that was not present in the CRACM recordings . Also , we cannot exclude that antidromic spikes may have influenced PSP analysis . Both may account for some of the observed differences in PSP kinetics . Overall , however , these data suggest that the kinetics of Chronos-evoked PSPs were generally similar to those of electrically-evoked PSPs . The DCN provides one of the major sources of excitatory input to the IC ( Adams , 1979; Brunso-Bechtold et al . , 1981; Oliver , 1984; Osen , 1972; Ryugo et al . , 1981 ) . A previous study has shown that DCN afferents synapse onto glutamatergic and GABAergic neurons in the IC ( Chen et al . , 2018 ) , but it is not known which specific classes of IC neurons receive input from the DCN . In addition , the physiology of DCN afferent synapses remains unknown . To test whether VIP neurons receive synaptic input from DCN principal neurons , we injected the right DCN with the AAV1 . Syn . Chronos-GFP . WPRE . bGH viral vector and recorded from VIP neurons in the contralateral ( left ) ICc ( Figure 8A , left ) . To confirm selective transfection of the DCN , we sliced the brainstem of every animal used and determined whether GFP expression was present and limited to the right DCN . If there was no transfection or if there was considerable expression of GFP in the auditory nerve or VCN , no recordings were performed . In most cases , GFP expression was limited to the DCN ( Figure 8A , right ) and GFP-labeled axons were present in the left ICc . To block spontaneous IPSPs , we performed DCN CRACM experiments with GABAergic and glycinergic blockers in the bath ( 5 µm gabazine and 1 µM strychnine , Figure 8B ) . We found that 2–5 ms pulses of blue light elicited EPSPs in 19 of 25 neurons tested , confirming that ICc VIP neurons receive synaptic input from the DCN . Light-evoked EPSPs had moderate amplitudes ( 2 . 85 mV ± 2 . 98 mV ) and relatively slow rise times ( 4 . 2 ms ± 1 . 3 ms ) , halfwidths ( 20 . 6 ms ± 14 . 4 ms ) and decay time constants ( 22 . 0 ms ± 6 . 7 ms ) ( n = 6 cells , Figure 8B , left and 8C ) . Because EPSP kinetics were relatively slow , we hypothesized that DCN synapses activate NMDA receptors on VIP neurons . Interestingly , D-AP5 had no significant effect on any of the measured properties ( amplitude: 2 . 80 mV ± 2 . 54 mV , p = 0 . 96 , rise time: 4 . 4 ms ± 1 . 7 ms , p = 0 . 95 , halfwidth: 15 . 4 ms ± 6 . 1 ms , p = 0 . 37 , decay time constant: 29 . 0 ms ± 7 . 1 ms , p = 0 . 16 , two tailed t-test , critical p value with Bonferroni correction for multiple comparisons = 0 . 0125 ) ( Figure 8B , middle , and 8C ) . In contrast to commissural inputs where D-AP5 had a significant effect on EPSP kinetics , this suggests that DCN inputs to ICc VIP neurons do not activate NMDA receptors under resting conditions . Subsequent addition of NBQX completely abolished DCN-evoked EPSPs ( Figure 8B , right ) , confirming that DCN synapses activate AMPA receptors on ICc VIP neurons . Since fusiform cells in the DCN often fire at rates exceeding 100 Hz ( Davis et al . , 1996; Ma and Brenowitz , 2012; Nelken and Young , 1994; Spirou and Young , 1991; Young and Brownell , 1976 ) , the slow kinetics of DCN-evoked EPSPs suggests that these EPSPs undergo temporal summation in VIP neurons . To test whether Chronos-elicited synaptic release from DCN terminals was driven by action potentials as opposed to direct activation of DCN terminals by light , we repeated the DCN CRACM experiments in the presence of 1 µM tetrodotoxin ( TTX ) . After TTX wash-in , EPSPs elicited by blue light flashes were completely and reversibly abolished in 3 of 3 tested VIP neurons ( data not shown ) . This suggests that synaptic release in CRACM experiments was driven by action potentials , or was at least dependent on sodium channel activation in synaptic terminals . To determine whether PSPs evoked by DCN inputs to VIP neurons differed from those evoked in non-VIP neurons , we recorded from non-fluorescent neurons in the ICc of VIP-IRES-Cre x Ai14 mice in which the DCN was transfected with Chronos . Recordings were targeted to neurons in regions of the ICc where Chronos-GFP-labeled axons were clearly visible . We found that 8 out of 12 non-VIP neurons received input from the DCN . EPSPs evoked by DCN inputs to non-VIP neurons ( n = 8 ) had similar amplitudes , rise time , halfwidth and decay times ( amplitude: 1 . 7 mV ± 0 . 6 mV , p = 0 . 34; rise time: 4 . 7 ms ± 1 . 6 ms , p = 0 . 52; halfwidth: 16 . 5 ms ± 9 . 6 ms , p = 0 . 77; decay time constant 25 . 9 ms ± 17 . 8 ms , p = 0 . 72; two tailed t-test , critical p value with Bonferroni correction for multiple comparisons = 0 . 0125 ) compared to those recorded from VIP neurons . Although the means were not statistically different , DCN-evoked EPSPs in VIP neurons showed less variability in halfwidth and decay time than EPSPs in non-VIP neurons ( Figure 8F ) . These results suggest that ICc neurons located near DCN afferents have a high probability of receiving input from the DCN , consistent with predictions made from anatomical studies ( Oliver , 1984; Oliver et al . , 1997 ) . Together , these results identify VIP neurons in the ICc as a distinct postsynaptic target of DCN afferents . Given the number of conditions that must be met for a long-range CRACM experiment to succeed , our observation that the connection probability for DCN to VIP projections was 76% ( 19 of 25 neurons ) suggests that most VIP neurons in the ICc receive input from the DCN . We next repeated the DCN-CRACM experiments without GABAA and glycine receptor antagonists in the bath . Under these conditions , we observed that an IPSP was elicited 2–3 ms after the onset of the light-evoked EPSP ( Figure 8D ) . This IPSP could vary in strength between recorded VIP neurons . In some instances , the IPSP slightly altered the halfwidth and decay time constant of the EPSP . In other cases , the IPSP strongly limited the EPSP duration and generated a hyperpolarization after the EPSP ( Figure 8D , left ) . Washing in gabazine restored the EPSP to values comparable to the EPSPs recorded in the presence of inhibitory receptor antagonists ( compare Figure 8B , C ) . Washing out gabazine restored the IPSP and limited EPSP halfwidth again ( Figure 8D , right ) . Across five ICc VIP neurons , the IPSP significantly shortened the halfwidth of the elicited EPSP ( p = 0 . 048 , paired t-test , Figure 8E ) . Halfwidth reductions ranged from 22% to 73% , with a median reduction of 36% . Because DCN to IC projections are glutamatergic ( Ito and Oliver , 2010; Oliver , 1984 ) , and the periolivary nuclei and nuclei of the lateral lemniscus , the sources of ascending GABAergic input to the IC ( González-Hernández et al . , 1996 ) , were not present in the brain slice , this inhibition must be due to DCN afferents activating a local feedforward inhibitory circuit within the IC . The latency to the IPSP onset also supports the theory of a disynaptic , local inhibitory circuit , as the IPSP always succeeded the initial EPSP . Thus , these results indicate that ascending input from the DCN activates a feedforward inhibitory circuit within the IC and that this circuit regulates the duration of DCN-evoked excitation in ICc VIP neurons .
It has long been argued that the classification of neurons requires a multifaceted analysis of morphological and physiological features ( Tyner , 1975 ) . More recent efforts have emphasized the importance of combining these features with molecular markers ( Ascoli et al . , 2008; Tremblay et al . , 2016; Zeng and Sanes , 2017 ) . This combination has proven to be particularly effective for unambiguously classifying neuron types . In the cerebral cortex , a multifaceted classification scheme including molecular markers has enabled investigations into how specific classes of interneurons shape circuit computations , sensory processing , and animal behavior ( Cichon et al . , 2017; Kato et al . , 2017; Kato et al . , 2015; Kuchibhotla et al . , 2017; Lee et al . , 2013; Milstein et al . , 2015; Pfeffer et al . , 2013; Pi et al . , 2013 ) . Similar approaches have succeeded in the amygdala , hypothalamus , basal ganglia , and other brain regions where it was previously difficult to identify neuron types ( Campbell et al . , 2017; Capogna , 2014; Wallace et al . , 2017 ) . In the auditory midbrain , previous efforts to identify neuron classes relied on in vitro physiology , in vivo physiology , morphology , neurochemical markers , or some combination of these ( Beebe et al . , 2016; Fujimoto et al . , 2017; Malmierca et al . , 1993; Oliver and Morest , 1984; Ono et al . , 2005; Palmer et al . , 2013; Peruzzi et al . , 2000; Ramachandran et al . , 1999; Schofield and Beebe , 2019; Sivaramakrishnan and Oliver , 2001 ) . Despite these and other attempts , the only neuron class that has been consistently identified in the IC are the large GABAergic neurons ( Beebe et al . , 2016; Geis and Borst , 2013; Ito et al . , 2015; Ito et al . , 2009; Ito and Oliver , 2014 ) . However , there are currently no known molecular markers for large GABAergic neurons , making it difficult to test the role of these neurons in auditory computations ( Schofield and Beebe , 2019 ) . Using a multifaceted approach , we identified VIP neurons as a novel class of IC neurons . VIP neurons share a common set of molecular , neurochemical , morphological , and physiological features . VIP neurons are labeled in the VIP-IRES-Cre mouse line and are glutamatergic . Ninety-three percent of the ICc VIP neurons in our data set had a stellate morphology and 94% of all IC VIP neurons had dendritic spines . Similarly , 90% of VIP neurons had sustained firing patterns . Although the input resistance , membrane time constant , and expression of Ih varied within the population of VIP neurons , for VIP neurons in the ICc , we found that a portion of this variability reflected their location along the tonotopic axis of the ICc . Tonotopic variation in the intrinsic physiology of VIP neurons was consistent with VIP neurons having faster membrane properties and therefore better temporal coding potential in lower frequency regions of the ICc . This parallels an in vivo study that found a gradient in the temporal coding capacity of cat ICc neurons that varied from better temporal coding in low frequency regions to worse temporal coding in high frequency regions ( Rodríguez et al . , 2010 ) . It is important to note that it would not be possible to identify VIP neurons based on their morphology or physiology alone . VIP neurons are not the only stellate neurons in the IC , nor are they the only neurons with sustained firing patterns or dendritic spines . These results provide insight to why it has been difficult to classify neuron types in the IC . We propose that a multidimensional approach incorporating molecular markers will be essential to identifying the remaining neuron classes in the IC . We found that VIP neurons project not only to MGB and contralateral IC , the most common recipients of IC projections , but also to the nucleus of the brachium of the IC , superior colliculus , periaqueductal gray , superior olivary complex , and ipsilateral IC ( Figure 6 ) . The number of extrinsic targets reached by VIP axons was a surprise given the relatively small population of VIP neurons . Do individual VIP neurons project to multiple targets ? Previous retrograde labeling studies suggest that some patterns of collateral projection are more common than others for IC cells . IC cells that project to the contralateral thalamus appear to quite commonly have a collateral projection to the ipsilateral thalamus ( Mellott et al . , 2019 ) . In contrast , very few IC neurons project to the thalamus and the cochlear nucleus ( Coomes and Schofield , 2004; Hashikawa and Kawamura , 1983; Okoyama et al . , 2006 ) , or to the left and right cochlear nuclei ( Schofield , 2001 ) . Whether IC commissural cells can have collateral projections to the thalamus has been supported ( González-Hernández et al . , 1996 ) or denied ( Okoyama et al . , 2006 ) . Because retrograde tracing studies underestimate collateral projections ( Schofield et al . , 2007 ) , such studies may have missed VIP neurons with collateral projections . Alternatively , individual VIP neurons might project to one or a few targets . It would then be possible to subdivide VIP neurons based on their axonal projections . This would parallel the cerebral cortex , where the major classes of interneurons contain subclasses that often differ in their axonal targeting ( Tremblay et al . , 2016 ) . If such is the case , then the unifying feature of VIP neurons might be that they perform similar computational roles within circuits , even when the circuits themselves are involved in different functions . In any event , the axonal projection patterns of individual VIP neurons , to extrinsic targets and within the IC , will be important features for further characterizing VIP subclasses and their functional roles . Our results show that VIP neurons receive input from at least four sources: principal neurons in the DCN , local inhibitory neurons , and excitatory and inhibitory neurons in the contralateral IC ( Figure 8G ) . In multiple instances , we observed that VIP neurons received input from the DCN and a local inhibitory neuron or a combination of excitatory and inhibitory commissural input . Given that optogenetic circuit mapping experiments underestimate connection probabilities ( not all synapses are transfected by the virus ) , these results suggest that many individual VIP neurons integrate input from ascending , local , and commissural sources . This is consistent with previous studies showing that individual IC neurons can integrate input from numerous sources ( Ito et al . , 2015; Ito and Oliver , 2014 ) . Excitatory postsynaptic responses in IC neurons often involve activation of NMDA receptors ( Ma et al . , 2002; Wu et al . , 2004 ) . Under in vivo conditions , the activation of NMDA receptors can influence how IC neurons respond to tones ( Sanchez et al . , 2007 ) . In VIP neurons , we found that excitatory commissural synapses activated AMPA and NMDA receptors , while synaptic input from DCN afferents activated only AMPA receptors . The activation of NMDA receptors occurred even though our ACSF contained 1 mM Mg2+ and neurons were at their resting membrane potential . Interestingly , previous studies have shown that NMDA receptors in some IC neurons can be activated under similar conditions , even when AMPA are receptors blocked ( Ma et al . , 2002; Sivaramakrishnan and Oliver , 2006 ) . The activation of NMDA receptors in our recordings may indicate that excitatory commissural synapses tend to be located on the distal dendrites or dendritic spines of VIP neurons , where the local membrane potential might be sufficiently depolarized by activation of AMPA receptors to remove Mg2+ block of NMDA receptors . A distal location would be consistent with the proposed modulatory role for commissural inputs ( Orton et al . , 2016; Orton and Rees , 2014 ) . Similarly , the lack of NMDA receptor activation by DCN afferents might indicate that DCN synapses are located on proximal dendrites or possibly on the soma itself , or that these synapses lack NMDA receptors . These synaptic arrangements may have important implications for auditory coding and synaptic plasticity mechanisms in VIP neurons . In many brain regions , feedforward inhibitory circuits control the time window for temporal integration of synaptic input ( Gabernet et al . , 2005; Pouille and Scanziani , 2001; Roberts et al . , 2013; Stokes and Isaacson , 2010 ) . In the ICc , it was recently shown that GABAergic neurons provide local inhibitory input mainly to neurons in the same isofrequency lamina ( Sturm et al . , 2014 ) . However , the conditions that recruit local inhibition have remained unclear . Our data provide direct evidence that activation of DCN afferents can elicit both direct excitatory input and disynaptic feedforward inhibition to VIP neurons . Feedforward inhibition can dramatically reduce EPSP halfwidth , suggesting that local feedforward inhibition regulates the temporal summation of synaptic inputs . In addition , while DCN afferents elicited modest EPSPs in VIP neurons , DCN input presumably drove spiking in the GABAergic neurons that provided feedforward inhibition . It will be important for future studies to identify this population of GABAergic neurons and the extent of their influence on auditory computations in VIP and other IC neurons . Ascending projections from most auditory brainstem nuclei restrict their synapses to subregions of the ICc , dividing the ICc into functional zones that may be specialized for processing specific classes of auditory cues ( Brunso-Bechtold et al . , 1981; Cant , 2013; Cant and Benson , 2006; Loftus et al . , 2010; Oliver , 2005; Oliver et al . , 1997 ) . We found that VIP neurons in the ICc were distributed with a significant bias toward the caudal ICc ( Table 2 ) and a tendency to be enriched in medial and dorsal regions of the ICc ( Figure 1 ) . Intriguingly , this distribution overlaps with the monaural domain of the ICc , an area that predominantly receives input from monaural nuclei , including the contralateral DCN and AVCN and the ipsilateral ventral nucleus of the lateral lemniscus ( VNLL ) ( Cant and Benson , 2006; Loftus et al . , 2010 ) . In addition , a recent study in mice showed that excitatory ICc neurons that receive input from the DCN are unlikely to receive input from other ascending auditory sources but likely to receive input from the ipsilateral auditory cortex ( Chen et al . , 2018 ) . These results lead us to hypothesize that VIP neurons are specialized for processing monaural cues , and in particular , monaural cues from the DCN . In future studies , it will be important to determine whether ascending input to VIP neurons is predominantly from the DCN , from a combination of monaural brainstem nuclei , or from both monaural and binaural sources . It will also be important to assess whether VIP neurons receive descending input from the auditory cortex and how commissural and cortical inputs shape sound processing in VIP neurons . An important question is whether IC VIP neurons use VIP signaling to modulate the activity of their postsynaptic targets . In cerebral cortex , 99 . 1% of neurons labeled in the VIP-IRES-Cre mouse immunostained with an antibody against VIP , suggesting that Cre expression in the VIP-IRES-Cre mouse is well correlated with the production of VIP ( Prönneke et al . , 2015 ) . There is therefore a strong likelihood that VIP neurons in the IC produce VIP . VIP acts by binding to one of three G-protein coupled receptors , VPAC1 , VPAC2 , and PAC1 ( Dickson and Finlayson , 2009; Vaudry et al . , 2009 ) . Anatomical studies suggest that all three VIP receptors are expressed in the MGB and SOC ( Joo et al . , 2004; Sheward et al . , 1995 ) , both of which are postsynaptic targets of VIP neurons ( Figure 6 ) . Activation of VIP receptors can have significant effects on neuronal excitability . In somatosensory thalamus , activation of VIP receptors increased the activation of HCN channels , depolarizing thalamocortical neurons and inducing them to switch from burst firing to tonic firing ( Lee and Cox , 2003; Sun et al . , 2003 ) . These data suggest that IC VIP neurons may potently modulate the activity of MGB neurons and other postsynaptic target neurons through VIP signaling . Thus , VIP neurons in the IC may exert an outsize influence on their long-range postsynaptic targets .
All experiments were approved by the University of Michigan Institutional Animal Care and Use Committee and were in accordance with NIH guidelines for the care and use of laboratory animals . Animals were kept on a 12 hour day/night cycle with ad libitum access to food and water . VIP-IRES-Cre mice ( Viptm1 ( cre ) Zjh/J , Jackson Laboratory , stock # 010908 ) ( Taniguchi et al . , 2011 ) were crossed with Ai14 reporter mice ( B6 . Cg-Gt ( ROSA ) 26Sortm14 ( CAG-tdTomato ) Hze/J , Jackson Laboratory , stock # 007914 ) ( Madisen et al . , 2010 ) to yield F1 offspring that expressed the fluorescent protein tdTomato in VIP neurons . For control experiments , C57BL/6J mice ( Jackson Laboratory , stock # 000664 ) were used . Because mice on the C57BL/6 background undergo age-related hearing loss , experiments were restricted to an age range where hearing loss should be minimal or not present ( Zheng et al . , 1999 ) . This age range was P70 or less for all mice except for three C57BL/6J mice used for electrophysiology experiments that were aged P89 , P90 , and P113 . Mice were deeply anesthetized and perfused transcardially with 0 . 1 M phosphate-buffered saline ( PBS ) , pH 7 . 4 , for 1 min and then with a 10% buffered formalin solution ( Millipore Sigma , cat# HT501128 ) for 10 min . Brains were collected and post-fixed in the same fixative for 2 hr and cryoprotected overnight at 4°C in 0 . 1 M PBS containing 20% sucrose . Brains were cut into 40 μm sections on a vibratome or freezing microtome . Sections were rinsed in 0 . 1 M PBS , and then treated with 10% normal donkey serum ( Jackson ImmunoResearch Laboratories , West Grove , PA ) and 0 . 3% Triton X-100 for 2 hr . Slides were incubated overnight at 4 °C in mouse anti-GAD67 ( 1:1000; Millipore Sigma , cat# MAB5406 ) , rabbit anti-NeuN ( 1:500; Millipore Sigma , cat# ABN78 ) , or mouse anti-bNOS ( 1:1000; Millipore Sigma , cat# N2280 ) . The next day , sections were rinsed in 0 . 1 M PBS and incubated in Alexa Fluor 488-tagged donkey anti-mouse IgG or donkey anti-rabbit IgG ( 1:500 , Thermo Fisher , cat# A-21202 and A-21206 ) for 1 hr at room temperature . Sections were then mounted on gelatin-subbed slides ( SouthernBiotech , cat# SLD01-BX ) and coverslipped using Fluoromount-G ( SouthernBiotech , cat# 0100–01 ) . Images were collected using a 1 . 30 NA 40x oil-immersion objective or a 1 . 40 NA 63x oil-immersion objective on a Leica TCS SP8 laser scanning confocal microscope . The mouse monoclonal anti-GAD67 antibody ( Millipore Sigma , cat# MAB5406 ) was raised against the 67 kDA isoform of glutamic acid decarboxylase ( GAD ) . The manufacturer reports that Western blot analysis shows no cross-reactivity with the 65 kDa isoform of GAD ( GAD65 ) . This antibody has been previously used to identify GABAergic cells in the IC ( Beebe et al . , 2016; Ito et al . , 2009; Mellott et al . , 2014 ) . The mouse monoclonal anti-nitric oxide synthase-brain ( bNOS ) ( Sigma , cat# N2280 ) was raised against the IgG1 isotype from the NOS-B1 hybridoma . The manufacturer reports that anti-bNOS reacts specifically with nitric oxide synthase ( NOS ) , derived from brain ( bNOS , 150 – 160 kDa ) . This antibody has been previously used , in guinea pig and mouse , to delineate the borders of the IC ( Coote and Rees , 2008; Keesom et al . , 2018 ) . To perform NeuN staining , we used a rabbit polyclonal antibody ( Millipore Sigma , cat# ABN78 ) . The manufacturer reports that anti-NeuN specifically recognizes the DNA-binding , neuron-specific protein NeuN , which is present in most central and peripheral neuronal cell types of all vertebrates tested . Previous studies reported the use of this antibody to label neurons in the IC ( Beebe et al . , 2016; Foster et al . , 2014; Mellott et al . , 2014 ) . Images from representative sections of the IC ( n = 3 animals , two sections per animal , one caudal and one middle ) were collected at 2 µm depth intervals with a 1 . 30 NA 40x oil-immersion objective on a Leica TCS SP8 laser scanning confocal microscope . Images were analyzed using Fiji software ( Rueden et al . , 2017; Schindelin et al . , 2012 ) . Consistent with previous studies , we found that the anti-GAD67 antibody did not penetrate the entire depth of the tissue sections ( Beebe et al . , 2016; Mellott et al . , 2014 ) . We therefore restricted our analysis to the top 10 – 12 µm of each section , where the antibody was fully penetrant . Within this region , we manually marked every GAD67+ cell body and every tdTomato+ cell body in the left IC . The green ( GAD67 ) and red ( tdTomato ) color channels were analyzed separately , so that labeling in one channel did not influence analysis of the other channel . After cells were marked , the GAD67 and tdTomato color channels were merged , and every instance where a cell body contained markers for both GAD67 and tdTomato was counted . The number of double-labeled cells was compared to the total number of tdTomato+ neurons to determine the percentage of tdTomato+ neurons that were GAD67+ . A design-based stereology approach was used to estimate the numbers of NeuN+ and tdTomato+ neurons in anti-NeuN stained sections ( Schmitz and Hof , 2005 ) . To collect systematic random samples , a virtual 370 µm x 370 µm grid was overlaid on the IC section . The starting coordinates for the grid were set using the Mersenne Twister random number generator in Igor Pro 7 or 8 ( WaveMetrics Inc ) . Images were then collected at coordinates determined by the upper-left intersection of each grid-square that fell over the left IC . Each image consisted of a 184 µm x 184 µm Z-stack collected at 1 µm depth intervals with a 1 . 40 NA 63x oil immersion objective on a Leica TCS SP8 confocal microscope . Six to sixteen images were collected per slice . Three slices were analyzed per mouse , with slices from each mouse evenly distributed along the rostral-caudal axis of the IC . Images were imported to Neurolucida 360 ( MBF Bioscience ) , where neurons were counted using the optical fractionator approach ( West et al . , 1991 ) . In this approach , we determined the image planes corresponding to the top , center , and bottom of the slice in each image stack . Top and bottom regions of each slice ( ≥2 µm thick ) were treated as guard zones and discarded from subsequent analysis . Removal of guard zones left a 15 µm-thick region at the center of the slice for subsequent analysis . Neurons within this region were counted by making a single mark at the top of each cell . Cells crossing the right and top borders of the image stack were counted , whereas those crossing the left and bottom borders were not . The green ( NeuN ) and red ( tdTomato ) color channels were analyzed separately , so that labeling in one channel did not affect analysis of the other . Next , the color channels were merged and cells with both NeuN and tdTomato markers were counted . In every instance , tdTomato+ cells were also NeuN+ ( 208/208 cells ) . The total number of double-labeled ( tdTomato+/NeuN+ ) cells was then compared to the total number of NeuN+ cells . Following transcardial perfusion as described previously , brains from three VIP-IRES-Cre x Ai14 mice were frozen and sectioned on a sliding microtome . Brains were cut into 40 µm sections , one each in the transverse , sagittal , and horizontal planes , and sections were collected in three series . The distribution of tdTomato-expressing ( VIP+ ) cells in one series from each case was analyzed using a Neurolucida system ( MBF Bioscience , Williston , VT ) attached to a Zeiss Axioimager . Z1 fluorescence microscope . Major IC subdivisions , including the central nucleus ( ICc ) , dorsal cortex , ( ICd ) and lateral cortex ( IClc ) , were identified by comparing bNOS and GAD67 immunostains with previous studies of mouse IC ( Dillingham et al . , 2017; Meininger et al . , 1986; Ono et al . , 2016; Willard and Ryugo , 1983 ) . Neurolucida Explorer was used to export drawings to Adobe Illustrator for figure preparation . Mice of either sex were used , aged postnatal day ( P ) 21 to P70 for VIP-IRES-Cre x Ai14 crosses and P21 to P113 for C57BL/6J animals . Mice were deeply anesthetized with isoflurane , decapitated , and the brain was dissected quickly in ~34 °C artificial cerebrospinal fluid ( ACSF ) containing ( in mM ) : 125 NaCl , 12 . 5 Glucose , 25 NaHCO3 , 3 KCl , 1 . 25 NaH2PO4 , 1 . 5 CaCl2 , 1 MgSO4 , bubbled to a pH of 7 . 4 with 5% CO2 in 95% O2 . Chemicals were obtained from Fisher Scientific or Millipore Sigma unless stated otherwise . Coronal or parasagittal slices ( 200 – 250 µm ) containing the IC were cut in ~34 °C ACSF with a vibrating microtome ( VT1200S , Leica Biosystems ) and incubated at 34 °C for 30 min in a holding chamber filled with ACSF and bubbled with 5% CO2 in 95% O2 . After incubation , slices were stored at room temperature until used for recordings . To make recordings , slices were placed in a recording chamber under a fixed stage upright microscope ( BX51WI , Olympus Life Sciences ) and were constantly perfused with 34 °C ACSF at ~2 ml/min . All recordings were conducted near physiological temperature ( 34 °C ) . IC neurons were patched under visual control using infrared Dodt gradient contrast and epifluorescence imaging . Recordings were performed with a BVC-700A patch clamp amplifier ( Dagan Corporation ) . Data were low pass filtered at 10 kHz , sampled at 50 kHz with a National Instruments PCIe-6343 data acquisition board , and acquired using custom written algorithms in Igor Pro . For every recording , series resistance and pipette capacitance were corrected using the bridge balance circuitry of the BVC-700A . Recordings with a series resistance above 25 MΩ were discarded . All membrane potentials have been corrected for a liquid junction potential of 11 mV . Electrodes were pulled from borosilicate glass ( outer diameter 1 . 5 mm , inner diameter 0 . 86 mm , Sutter Instrument ) to a resistance of 3 . 5 – 4 . 5 MΩ using a P-1000 microelectrode puller ( Sutter Instrument ) . The electrode internal solution contained ( in mM ) : 115 K-gluconate , 7 . 73 KCl , 0 . 5 EGTA , 10 HEPES , 10 Na2 phosphocreatine , 4 MgATP , 0 . 3 NaGTP , supplemented with 0 . 1% biocytin ( w/v ) , pH adjusted to 7 . 3 with KOH and osmolality to 290 mmol/kg with sucrose . Input resistance was determined by delivering a series of 100 ms hyperpolarizing current steps incremented to elicit hyperpolarization ranging from just below the resting membrane potential to < −110 mV . For each response , the amplitudes of the peak ( most negative value ) and steady-state ( average of last 10 ms of response ) hyperpolarization were measured relative to the resting potential . Voltage versus current plots were prepared , and input resistance was determined from the slopes of lines fit to the peak ( Rpk ) and steady-state ( Rss ) data for current steps that achieved a peak hyperpolarization between 0 and −15 mV relative to rest . Membrane time constant was determined by delivering 50 current steps at an amplitude that hyperpolarized the membrane potential by 1–3 mV . Current step duration was set to ensure that the membrane potential achieved a steady-state value before the end of the current step . An exponential function was then fit to onset of each response and the median time constant determined . For electrical stimulation of commissural inputs , a glass electrode filled with ACSF was placed into the commissure near the midline , connected to a DS3 constant current , isolated stimulator ( Digitimer Ltd . ) . PSPs were elicited with 100 µs electric shocks , ranging from 45 µA to 320 µA . To isolate or manipulate synaptic events , the following pharmacological agents were used , all diluted in standard ACSF: 5 µM SR95531 ( gabazine , GABAA receptor antagonist , Hello Bio ) , 1 µM strychnine hydrochloride ( glycine receptor antagonist , Millipore Sigma ) , 50 µM D-AP5 ( NMDA receptor antagonist , Hello Bio ) , 10 µM NBQX disodium salt ( AMPA receptor antagonist , Hello Bio ) , 1 µM TTX ( voltage dependent sodium channel blocker , Hello Bio ) . Data analysis was performed using custom written algorithms in Igor Pro or MATLAB ( Mathworks ) . Statistical tests were performed in R Studio ( R Studio , Boston ) for R 3 . 5 . 1 ( The R Project for Statistical Computing , The R Foundation ) . After recordings , the electrode was removed slowly to allow the cell membrane to reseal , and the slice was fixed in 4% paraformaldehyde ( PFA ) in 0 . 1 M phosphate buffer ( PB , pH 7 . 4 ) for 12 – 24 hr . Slices were then washed in 0 . 1 M PB and stored in 0 . 1 M PB for up to three weeks . Recorded neurons were stained using fluorescent biocytin-streptavidin histochemistry . In brief , slices were washed in 0 . 1 M PB three times for 10 min ( 3 × 10 min in PB ) , permeabilized in 0 . 2% Triton X-100 in 0 . 1 M PB for 2 hr , washed 3 × 10 min in PB , and stained at 4 °C for 48 hr with streptavidin-Alexa Fluor 488 or 647 , diluted 1:1000 in 0 . 1 M PB . Slices were then washed 3 × 10 min in PB and mounted on Superfrost Plus microscope slides in anti-fade media ( Fluoromount-G ) . Z-stack images of streptavidin-Alexa Fluor labeled cells were obtained with a Leica TCS SP8 laser scanning confocal microscope using a 1 . 40 NA 63x oil-immersion objective . Three-dimensional reconstructions of neuronal morphology and quantitative analyses of soma and dendrite shape were performed on image stacks imported into Neurolucida 360 ( MBF Bioscience ) . To facilitate comparisons of neuronal morphology , all reconstructed neurons are displayed as if they were in the left IC as viewed from a caudal perspective . Reconstructions of neurons that were located in the right IC were flipped along the dorsal-ventral axis so that they appear as if they were in the left IC . Following biocytin-streptavidin histochemistry , tile scan images of the entire IC were collected using a 20x objective on a Leica TCS SP8 confocal microscope . These images were then used to determine medial-lateral and dorsal-ventral coordinates of recorded neurons . The medial-lateral coordinate was measured as the distance of the soma from the medial axis ( midline ) of the IC slice ( x axis ) . The dorsal-ventral coordinate was measured as the distance of the soma from the dorsal-most edge of the IC slice ( y axis ) . Neurons in the right IC were combined with those from the left IC by multiplying the medial-lateral coordinate of neurons from the right IC by −1 . Neurons were assigned to IC subdivisions , including the central nucleus ( ICc ) , dorsal cortex , ( ICd ) and lateral cortex ( IClc ) , by comparing neuron location in 20x tile scans to IC subdivision borders determined from standard series of IC sections immunostained for bNOS or GAD67 ( Choy Buentello et al . , 2015; Coote and Rees , 2008 ) . Neuron coordinates were then compared to physiological parameters obtained during whole cell recordings . To test for correlations , data were fit with a plane using the Levenberg-Marquardt least squares method in Igor Pro . Fit quality was assessed with Pearson’s correlation coefficient and the adjusted R2 . Fit significance ( p value ) was calculated based on the chi-squared statistic from the fit and the chi-squared cumulative distribution function . Neuron morphology was reconstructed as described above . To determine the ‘length’ and ‘width’ axes of the dendritic arbors , the set of coordinates describing the morphology of the dendritic arbor of each neuron was exported from Neurolucida 360 ( MBF Bioscience ) . Coordinates were imported to Igor Pro , where principal components analysis ( PCA ) was performed on either the x and y coordinates ( 2D PCA ) or the x , y , and z coordinates ( 3D PCA ) . The orientation of the length and width axes was then derived from the first and second principal directions of the resulting eigenvector matrices . 2D PCA was used to determine the orientation of neurons within the coronal plane . 3D PCA was used to determine the axes to use for measuring dendritic arbor length/width ratios . For this , the spread of the dendritic arbor along the first and second principal directions was determined by rotating each morphology coordinate set according to its eigenvector matrix , then calculating the range from the minimum to maximum coordinates along the x ( length , first principal direction ) and y ( width , second principal direction ) axes . Intracranial virus injections were performed on mice age P21 – P35 using standard aseptic techniques . Throughout the procedure , mice were anesthetized with isoflurane and their body temperature maintained with a homeothermic heating pad . An injection of the analgesic carprofen ( 5 mg/kg , CarproJect , Henry Schein Animal Health ) was delivered subcutaneously . The scalp was shaved and a rostro-caudal incision was made along the midline to expose the skull . Injection sites were mapped using stereotaxic coordinates relative to the lambda suture . A single craniotomy was performed using a micromotor drill ( K . 1050 , Foredom Electric Co . ) with a 0 . 5 mm burr ( Fine Science Tools ) . Viral constructs were injected with a NanoJect III nanoliter injector ( Drummond Scientific Company ) connected to a MP-285 micromanipulator ( Sutter Instruments ) . Glass injection pipettes were prepared by pulling capillary glass ( Drummond Scientific Company ) with a P-1000 microelectrode puller ( Sutter Instrument ) . The injector tip was cut to an opening of ~20 µm and beveled at 30° with a BV-10 pipette beveller ( Sutter Instrument ) . Injectors were back-filled with mineral oil and then front-filled with virus . AAV1 . Syn . Chronos-GFP . WPRE . bGH ( University of Pennsylvania Vector Core , Lot# CS1027L , 2 . 986e13 genome copies ( GC ) /ml ) was used for CRACM experiments . AAV1 . CAG . FLEX . eGFP . WPRE . bGH ( Allen Institute 854 , University of Pennsylvania Vector Core , Lot# CS0922 , 4 . 65e13 GC/ml ) was used for axonal tract tracing . For CRACM experiments , the IC was injected via two penetrations . Virus deposits were made at 250 µm intervals along the dorsal-ventral axis , resulting in four deposits in penetration one and three deposits in penetration 2 . At each depth , 20 nl of virus was deposited , resulting in seven virus deposits and a total load of 140 nl virus per injected IC . DCN injections were limited to two deposits of 20 nl virus . For axonal tracing studies , viral load was reduced to 40 nl total ( 20 nl per site ) to achieve sparser labeling of neurons . Injections were made at the coordinates shown in Table 3 . After injections were completed , the scalp was sutured with Ethilon 6–0 ( 0 . 7 metric ) nylon sutures ( Ethicon USA LLC ) , and the wound was treated with 0 . 5 – 1 ml 2% Lidocaine hydrochloride jelly ( Akorn Inc ) . Once mice were ambulatory , they were returned to the vivarium where they were monitored daily until sutures fell out and the wound was completely healed . After allowing 3 – 4 weeks for Chronos expression , animals were used for in vitro slice electrophysiology experiments as described above , with the exception that after decapitation all steps were performed in red light and recordings were conducted in darkness or red light to limit Chronos activation . For standard CRACM experiments , recordings were targeted to VIP neurons . In additional control experiments , recordings were targeted to non-fluorescent IC neurons in VIP-IRES-Cre x Ai14 crosses . During whole cell recordings , Chronos was activated by brief pulses of 470 nm light emitted by a blue LED coupled to the epi-fluorescence port of the microscope and delivered to the brain slice through a 0 . 80 NA 40x water immersion objective with a field number of 26 . 5 mm . Accordingly , the blue light spot had an area of 0 . 345 mm2 . In all CRACM experiments , the soma of the recorded neuron was present in the field of view . Blue light flashes were 2 to 5 ms long , illuminated the entire field of a 0 . 80 NA 40x objective , and yielded optical power densities that ranged from 6 to 48 mW/mm² . Optical power was set using a minimal stimulation protocol . In general , the shortest stimulus duration that elicited a PSP was chosen , combined with 120% of the optical power that was determined as the threshold to elicit PSPs . Yet , when using a 5 ms pulse with maximum optical power , EPSP kinetics did not change significantly ( 8 of 8 recorded neurons , data not shown ) compared to the minimal stimulation paradigm . This suggests that light pulses elicited action potentials with consistent durations , regardless of optical power . Recording sweeps with light flashes were repeated 20 to 50 times in 0 . 5 – 1 s intervals to obtain average PSP values . During experiments to investigate receptor contribution to PSPs , drugs were washed in for at least 10 min before recording under the new condition . For each receptor antagonist , 7 – 8 washout experiments were conducted . In each case , drug effects reversed after washout ( data not shown ) . The right IC of VIP-IRES-Cre x Ai14 mice was injected with AAV1 . CAG . FLEX . eGFP . WPRE . bGH and transcardially perfused 3 – 4 weeks later , as described above . Brains were frozen and sectioned at 40 µm on a sliding microtome . For some brains , sections were collected serially , and for others , sections were collected in three series . The brains were examined for eGFP-labeled axons and boutons , which were interpreted as VIP+ projections originating in the IC . Some sections were counterstained with a fluorescent Nissl stain ( Neurotrace 640/660 , ThermoFisher , cat# N21483 ) . Injection sites comprised a collection of eGFP-labeled cell bodies . Cases were included for analysis only if the eGFP labeled cell bodies were restricted to the IC . Images were collected on a Zeiss AxioImager . Z2 microscope . High magnification images were collected as z-stacks using a 1 . 40 NA 63X oil-immersion objective and structured illumination ( Zeiss Apotome 2 ) for optical sectioning at 0 . 5 µm intervals . Images shown are maximum projections of collected stacks . Adobe Photoshop was used to colorize images , to globally adjust levels , and to add scale bars . | Our brains help us make sense of our surroundings . Our sense organs , for example the ears , receive signals from the environment , which are passed on to specialized parts of the brain , called nuclei . There , neurons process the information and send impulses to other areas in the brain , which eventually help us decide how to respond . A nucleus called the inferior colliculus is one of the major hearing centres in the mammalian brain . In humans , it is vital for recognizing speech and pinpointing the location of sounds . It contains several different types of neurons , and links with many other areas of the brain . While the inferior colliculus is crucial for our sense of hearing , little is known about the specific properties of the neurons within it . In particular , these cells are difficult to divide into well-defined groups . Even less is understood about the precise nature of the connections between these neurons , which likely underpin the computational power of the inferior colliculus . Goyer et al . therefore set out to identify a specific class of neurons in this region and map the circuits they formed . Experiments were conducted on genetically modified mice whose neurons were only ‘glowing’ if they had a gene called VIP switched on . Detailed examination of the shape of the ‘VIP cells’ , as well as their chemical and electrical properties , confirmed that they were indeed a distinct class of neurons . Another set of experiments relied on a method that uses light to control the activity of brain cells . This showed that VIP cells received signals both from other neurons in the inferior colliculus , and from another hearing center in the brain . In turn , VIP cells sent signals over long distances to many other parts of the brain that handle sound signals . This suggests that VIP neurons have a wide-ranging influence on our brains’ ability to process sound . The work by Goyer et al . has , for the first time , reliably identified specific circuits in a brain region essential for our sense of hearing . By knowing more about how the brain’s hearing centers are connected to each other , it may become possible to understand their roles in hearing loss . In this effort , the inferior colliculus may become a target for treatments for patients with hearing difficulties . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2019 | A novel class of inferior colliculus principal neurons labeled in vasoactive intestinal peptide-Cre mice |
Many chronic diseases are associated with fibrotic deposition of Collagen and other matrix proteins . Little is known about the factors that determine preferential onset of fibrosis in particular tissues . Here we show that plasma membrane ( PM ) overgrowth causes pericellular Collagen accumulation in Drosophila adipocytes . We found that loss of Dynamin and other endocytic components causes pericellular trapping of outgoing Collagen IV due to dramatic cortex expansion when endocytic removal of PM is prevented . Deposits also form in the absence of negative Toll immune regulator Cactus , excess PM being caused in this case by increased secretion . Finally , we show that trimeric Collagen accumulation , downstream of Toll or endocytic defects , activates a tissue damage response . Our work indicates that traffic imbalances and PM topology may contribute to fibrosis . It also places fibrotic deposits both downstream and upstream of immune signaling , consistent with the chronic character of fibrotic diseases .
Basement membranes are polymers of extracellular matrix ( ECM ) proteins that underlie epithelia and surround organs in all animals ( Yurchenco , 2011; Kelley et al . , 2014 ) . Their main constituent is Collagen IV , a helical trimer consisting of three α chains , capable of forming polymeric networks that interact with other ECM proteins . The fruitfly Drosophila melanogaster has emerged in recent years as an excellent model to study production of Collagen and other ECM proteins thanks to evolutionary conservation , powerful genetic tools and little genetic redundancy ( Denef et al . , 2008; Martinek et al . , 2008; Bunt et al . , 2010; Haigo and Bilder , 2011; Drechsler et al . , 2013; Lerner et al . , 2013; Na et al . , 2013; Devergne et al . , 2014; McCall et al . , 2014; Xiao et al . , 2014; Zhang et al . , 2014 ) . Two Collagen IV chains exist in Drosophila , encoded by Collagen at 25C ( Cg25C , α1 chain ) and viking ( vkg , α2 chain ) . Apart from Collagen IV , BMs include Laminin , Nidogen and Perlecan , which are conserved from flies to humans as well ( Hynes , 2012 ) . In the larva , the main source of Collagen IV is the adipocytes of the fat body ( Pastor-Pareja and Xu , 2011 ) , known for their role in lipid storage and metabolic regulation , but also as an active secretory tissue which produces serum proteins and clotting factors normally present in the hemolymph ( insect blood ) . The fat body is also a key effector of innate immunity , known to produce and secrete to the hemolymph large amounts of antimicrobial peptides in response to infections ( Lemaitre and Hoffmann , 2007 ) . Secreted proteins like Collagen reach the extracellular space through a controlled series of membrane traffic events ensuring fusion of secretory vesicles with the plasma membrane ( PM ) . Besides its role in cargo transport , membrane trafficking is a well-recognized driver of changes in cell shape and PM amount during morphogenesis ( Lecuit and Pilot , 2003 ) . Examples of morphogenetic traffic in Drosophila include contraction of the amnioserosa during dorsal closure ( Mateus et al . , 2011 ) and widening of the lumen of tracheae ( Tsarouhas et al . , 2007 ) . The best studied example of traffic-driven morphogenesis is perhaps blastoderm cellularization in the early Drosophila embryo . During blastoderm cellularization , fast directed PM growth results from membrane contributions from the secretory pathway ( Lecuit and Wieschaus , 2000 ) , endocytic membrane recycling ( Pelissier et al . , 2003; Sokac and Wieschaus , 2008; Fabrowski et al . , 2013 ) and microvillar PM elaborations ( Figard et al . , 2013 ) . While these examples highlight the potential of membrane traffic to elicit drastic changes in cell shape in the context of morphogenetic events , a role of in maintaining stable cortical morphology has not been addressed in detail and little is known on how cells normally regulate PM amount . Furthermore , the consequences for cell physiology of changes in this fundamental property are also unknown . Handling of Collagen entails several challenges to secreting cells . Because of its large size , secretory transport of Collagen molecules requires carriers larger than regular COPII vesicles ( Saito et al . , 2009 ) . Also , Collagen molecules undergo postranslational modification along the secretory pathway by numerous Collagen-modifying enzymes such as glycosidases , and lysyl- and prolyl-hydroxylases , required for trimer formation ( Myllyharju and Kivirikko , 2004 ) . Prolyl-hydroxylation in particular is essential for trimer formation , mediated in Drosophila by the prolyl-4-hydroxylase PH4αEFB ( Pastor-Pareja and Xu , 2011 ) . Unlike fibrilar Collagen I , which flies lack , Collagen IV is secreted in functional form and does not require extracellular cleavage of the N- and C-terminal propeptides ( Khoshnoodi et al . , 2008 ) . Therefore , and given its ability to form supramolecular assemblies , it is not known how Collagen IV avoids aggregation inside secreting cells or at their PM . Finally , because Collagen is resistant to most proteases , disposing of aggregates when they occur is highly problematic . Such is the case of fibrotic diseases , characterized by aberrant and excessive deposition of ECM due to persistent immune stimulation following diverse types of injuries ( Wynn and Ramalingam , 2012 ) . In a variety of tissues such as the liver , skin , kidney or fat tissue , fibrosis disrupts tissue organization and increases matrix stiffness , affecting normal cell and tissue physiology ( Hoffman et al . , 2011; Friedman et al . , 2013 ) . In this work , we conducted a screening for genes involved in Collagen IV secretion by the fat body adipocytes in Drosophila . We found that loss of shibire and other endocytic genes leads to excess PM growth , which traps Collagen IV in the pericellular space . Pericellular Collagen IV trapping also results upon loss of loss of negative immune regulator cactus , which causes PM overgrowth through a Toll-dependent secretory burst . The accumulation of Collagen IV in the secreting adipocytes has two main consequences: Collagen IV deficit in the BMs of destination tissues and an immune response against adipocytes due to the abnormal ECM accumulation .
To gain new insights into Collagen biogenesis , we conducted a screening for genes affecting production of Collagen IV by fat body adipocytes , its main source in the Drosophila larva ( Figure 1A ) . We used BM-40-SPARC-GAL4 ( Venken et al . , 2011 ) to drive expression in adipocytes of the RNAi transgenes in the TRiP collection ( 8459 lines targeting 6200 genes ) ( Ni et al . , 2008 , 2011 ) and analyzed the localization of Vkg-GFP , a functional GFP-trap fusion to the Collagen IV chain Vkg ( Morin et al . , 2001 ) . While a majority of hits produced intracellular Collagen IV accumulation ( full results to be published later ) , a distinct phenotypical category consisted of 60 genes causing accumulation at or near the PM ( Supplementary file 1 ) . Among the strongest hits in this category were two different RNAi transgenes targeting shibire ( shi ) , encoding fly Dynamin , a GTPase involved in excision of endocytic vesicles ( Ferguson and De Camilli , 2012 ) . shibire knock-down ( shii ) , as well as expression of dominant negative Dynamin ( ShiK44A ) ( Moline et al . , 1999 ) , caused Vkg accumulation in adipocytes ( Figure 1B ) . In validation of the phenotype , antibody staining confirmed reduced Dynamin expression in shii cells , whereas the staining increased after Shi . K44A overexpression ( Figure 1C ) , attesting to the sensitivity of the antibody . Since Collagen IV is a heterotrimer combining the α2 chain Vkg with Cg25C α1 chains , we performed a staining with an anti-Cg25C antibody we generated for this study ( see Figure 1—figure supplement 1 ) . This staining revealed that Cg25C , same as Vkg , accumulates in shii adipocytes ( Figure 1D ) , thus confirming Collagen IV accumulation in these cells . The accumulation of Collagen IV occurs at the cell periphery , under a basement membrane surrounding the tissue in the wild type and which still forms in shii adipocytes ( Figure 1D; see also Figure 3B later ) . Further validating the requirement of shibire in normal Collagen IV distribution , shi1 and shi2 thermosensitive mutations ( Kim and Wu , 1990 ) also caused Vkg accumulation in adipocytes when larvae grew at restrictive temperature ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 07187 . 003Figure 1 . Endocytic defects cause Collagen accumulation in Drosophila adipocytes . ( A ) Schematic depiction of Collagen IV production , secretion and incorporation into basement membranes . ( B ) shibire knock-down ( BM-40-SPARC>shii ) and dominant negative shibireK44A ( BM-40-SPARC>shiDN ) cause Vkg-GFP accumulation in third instar larva adipocytes ( marked with RFP ) . ( C ) Confocal images of third instar larva adipocytes stained with anti-Dynamin antibody . Staining is absent upon shi knock-down and increased by shiK44A expression . Nuclei stained with DAPI . ( D ) Localization of Vkg-GFP and Cg25C ( anti-Cg25C staining ) in wild type and BM-40-SPARC>shii adipocytes . Collagen IV accumulates in the periphery of shii adipocytes . ( E ) Vkg accumulation in BM-40-SPARC>Rab5i and >Rab5DN adipocytes . ( F ) Vkg accumulation in BM-40-SPARC>Chci and >ChcDN adipocytes . ( G ) Presence of Vkg-GFP is reduced in discs from BM-40-SPARC>shii , >shiDN , >Rab5i and >Rab5DN larvae . Vkg-GFP decrease quantified in graph . n ≥ 5 for each genotype . Differences with wild type are in all cases significative ( Mann–Whitney tests , **p < 0 . 01 ) . ( H ) Elongation of the ventral nerve cord ( VNC ) in BM-40-SPARC>shiDN and >Rab5DN . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 00310 . 7554/eLife . 07187 . 004Figure 1—figure supplement 1 . ( A ) Confocal images of adipocytes from shi1 and shi2 thermosensitive mutants . Shifting larvae to restrictive temperature for 3 hr causes mild pericellular accumulation of Collagen IV ( Vkg-GFP in green ) . myr-RFP membrane marker in red . ( B ) Western blots of hemolymph probed with an anti-Cg25C antibody ( 1:5000 ) . Hemolymph was collected by turning 10 larvae inside-out inside 100 μl of PBS . 10 μl of 2-Mercaptoethanol-reduced sample ( equivalent to the blood of 1 larva ) were loaded per genotype . We bled wild type larvae ( w1118 ) and larvae where vkg or Cg25C were knocked down in adipocytes ( Cg>vkgi+tub-GAL80ts and Cg>Cg25Ci+tub-GAL80ts respectively ) . For vkg and Cg25C knock-down , and in order to circumvent embryonic/L1 lethality , temporary inhibition of GAL4-driven knock-down was achieved with thermosensitive GAL4 inhibitor tub-GAL80ts ( larvae were grown at 18°C to prevent knock-down , transferred to 30°C to initiate knock-down in L1/L2 stage and bled 3 days later in L3 stage ) . Note that knock-down of Vkg increases Cg25C signal , expected as monomeric Cg25C cannot be incorporated into BMs in the absence of Viking ( Pastor-Pareja and Xu , 2011 ) . ( C ) Pericellular Vkg accumulation in adipocytes from BM-40-SPARC>Hrsi , >RN-trei , >AP-2αi and >AP-2μi larvae . ( D ) Western blots of hemolymph extracted from wild type ( w1118 ) , BM-40-SPARC>shii , >cacti and >Tl10B larvae probed with an anti-GFP antibody ( 1:5000 ) . The amount of blood loaded in each well is equivalent to 1 larva . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 00410 . 7554/eLife . 07187 . 005Figure 1—figure supplement 2 . UAS-Dcr2 expression ( BM-40-SPARC-Gal4>UAS-Dcr2 ) does not affect Collagen IV localization ( Vkg-GFP ) in adipocytes compared to Vkg-GFP control larvae ( + ) and larvae expressing GAL4 but not Dcr2 ( BM-40-SPARC-Gal4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 005 In addition to Dynamin , a number of hits showing a similar peripheral Collagen IV accumulation phenotype in the screening turned out to encode components of the endocytic machinery as well , such as Rab5 ( Figure 1E ) , Clathrin heavy chain ( Figure 1F ) , the Rab5-GAP RN-Tre , the AP-2 components AP-2α and AP-2μ and Hrs ( Figure 1—figure supplement 1 ) , suggesting that defective endocytosis was indeed the cause for the phenotype . Concurrent with the accumulation of Collagen IV in the adipocytes producing it , we found that the amount of Collagen IV present in destination BMs ( Figure 1G ) and in the hemolymph ( Figure 1—figure supplement 1 ) was reduced in these conditions . In agreement with this , we observed elongation of the ventral nerve cord ( Figure 1H ) , a deformation associated with Collagen IV reduction ( Pastor-Pareja and Xu , 2011 ) . We therefore conclude from these data that loss of shibire and other endocytic genes causes accumulation of Collagen IV in the adipocytes that normally secrete it and set out next to ascertain the mechanism by which this accumulation occurs . Accumulations of Collagen IV in shii adipocytes were found in close proximity to the PM , as revealed by the membrane marker myr-mRFP ( myristoylation domain of Src fused to mRFP; Figure 2A ) and by phalloidin staining of F-actin , normally enriched in the cell cortex ( Figure 2B ) . To determine whether Collagen IV accumulations were intracellular or extracellular we stained shii adipocytes with the lipophilic , cell-impermeable dye FM4-64 ( Bolte et al . , 2004 ) to label membrane directly in contact with the extracellular space ( PM ) . Short incubation with FM4-64 showed that the accumulations were surrounded by PM ( Figure 2C ) and , thus , likely extracellular . Labeling of membrane around the accumulations was equally observed when shii adipocytes were fixed and stained with fluorescently-labeled fixable dextrans of molecular weights 70 , 000 ( Figure 2D ) and 3000 ( Figure 2—figure supplement 1 ) . To confirm that the Collagen IV accumulations in shii adipocytes were extracellular , we performed antibody stainings without permeabilizing the cells ( no detergent in washing or blocking solutions ) . As a control , we stained Tango1i adipocytes , known to retain Collagen IV intracellularly ( Pastor-Pareja and Xu , 2011 ) , and found that intracellularly retained Collagen IV could not be stained without permeabilization ( Figure 2E ) . In contrast to this lack of staining , accumulations of Collagen IV in shii adipocytes were still labeled ( Figure 2E ) , indicating that they were indeed extracellular . Altogether , these data show that Collagen IV accumulations in shii adipocytes are pericellular accumulations located outside of the PM ( see also electron micrographs in Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 07187 . 006Figure 2 . Collagen accumulation in endocytosis-defective cells is pericellular and autonomous . ( A ) Vkg-GFP accumulation in a shii adipocyte ( BM-40-SPARC>shii ) expressing membrane marker myr-RFP . ( B ) shii adipocyte stained with phalloidin ( F-actin ) . ( C ) shii adipocyte stained with cell-impermeable membrane dye FM4-64 , labelling plasma membrane ( PM ) around accumulations . ( D ) shii adipocyte stained with fixable Texas-Red-coupled Dextran ( 70 , 000 MW ) , labelling PM around accumulations . ( E ) Antibody stainings of wild type and shii adipocytes performed without permeabilization ( no detergent ) in order to detect extracellular Collagen IV . In contrast to the accumulations in shii adipocytes , intracellular accumulations of Collagen IV in Tango1i adipocytes cannot be stained in the absence of permeabilization and are shown as a control . ( F ) Mosaic fat body ( act-GAL4>shii flip-out clones , marked with RFP ) showing Vkg-GFP accumulation in shii cells . Accumulation is suppressed by a GFP-targeting dsRNA ( iGFPi ) . ( G ) Pericellular accumulation in mosaic shii fat body expressing Cg25C-RFP . Clones marked with GFP . ( H ) Localization of the endocytic marker Transferrin Receptor ( Cg>TfR-GFP ) in wild type adipocytes . ( I ) In shii adipocytes , TfR concentrates in PM pockets containing Collagen IV ( anti-Cg25C ) . ( J ) In Rab11i adipocytes , TfR localizes to intracellular vesicles that completely fill the cytoplasm . No TfR is detected at the PM . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 00610 . 7554/eLife . 07187 . 007Figure 2—figure supplement 1 . ( A ) Confocal images showing the PM of shii and Tl10B adipocytes stained with fixable Texas-Red-coupled Dextrans ( 3000 and 70 , 000 MW ) , labelling PM around Collagen IV ( Vkg-GFP ) accumulations . ( B ) Electron micrographs of the PM in BM-40-SPARC>shii adipocytes showing instances of connection ( arrows ) between the pericellular accumulations ( asterisks ) and the extracellular space showing that these pockets are not isolated cisternae , but part of a very intricate PM . ( C ) Confocal images of wing discs ( posterior ventral hinge ) showing localization of Cg25C-GFP and Cg25C-RFP in the basement membrane after expression in the fat body controlled by Cg-GAL4 ( Cg>Cg25C-GFP and Cg>Cg25C-RFP respectively ) . Images of wild type ( w1118 ) discs are shown as controls to exclude auto-fluorescence . Nuclei stained with DAPI ( blue ) . ( D ) Images of live larvae expressing Cg25C-RFP and Cg25C-GFP in the fat body ( Cg>Cg25C-GFP and Cg>Cg25C-RFP ) . Knock-down of PH4αEFB , required for Collagen IV trimerization , causes tagged Cg25C to accumulate in the blood ( note strong fluorescent signal filling the body cavity ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 007 In light of the accumulation of Collagen IV , we first hypothesized that Collagen was internalized from the hemolymph ( blood ) by wild type adipocytes , which could explain pericellular accumulation when endocytosis was prevented . Contradicting this hypothesis , however , we found in mosaic experiments that accumulation of Vkg-GFP was suppressed when Vkg-GFP expression was knocked-down in the same cell ( Figure 2F ) . This result demonstrates that Collagen IV accumulated in a Dynamin-deficient adipocyte originates autonomously in that same cell . Supporting the autonomous origin of accumulated Collagen IV , mosaic expression of Cg25C-RFP ( see Figure 2—figure supplement 1 ) in shii adipocytes caused autonomous Collagen IV accumulation as well ( Figure 2G ) . These experiments indicate that Collagen IV pericellularly accumulating in a shii adipocyte is not Collagen that failed to be endocytosed . Contrary to that , our data show that pericellularly accumulated Collagen IV , instead of ingressing into the cell , is outgoing Collagen secreted by that same cell and becoming extracellularly trapped in the cell cortex . In order to ascertain how endocytic defects cause pericellular trapping of outgoing Collagen , we proceeded to characterize the locus of its accumulation . The endocytic marker TfR ( Transferrin Receptor , hTfR-GFP [Henthorn et al . , 2011] ) localized uniformly in the adipocyte PM and some intracellular vesicles ( Figure 2H ) . Confirming that TfR is internalized and recycled back to the PM , knock-down of Rab11 , required for recycling endosome trafficking ( Maxfield and McGraw , 2004 ) , caused TfR vesicles to fill the cytoplasm while no TfR was detected at the PM ( Figure 2J ) . In Dynamin-deficient cells , in contrast , TfR concentrated heavily in deep PM pockets containing Collagen IV at their center ( Figure 2I ) , suggesting that Collagen is pericellularly accumulated in PM pockets formed as a consequence of failed endocytosis . To characterize in further detail the topology of the PM at the site of Collagen accumulation , we analyzed confocal and electron micrographs . Wild type adipocyte PM , which is flat during the first and second larval instars , becomes somewhat convoluted in the third instar ( Figure 3—figure supplement 1 ) , displaying multiple invaginations which often encircle surface lipid droplets ( Diaconeasa et al . , 2013 ) . Compared to this , examination of endocytosis-deficient adipocytes revealed a striking increase in PM amount with respect to the wild type ( Figure 3A , B ) . Quantification in both confocal and electron micrographs showed an increase in the depth of PM ingression into the cytoplasm and higher PM sinuosity in endocytosis-defective cells ( Figure 3C; Figure 3—figure supplement 2 ) . The opposite phenotype , PM flattening , along with a vesicle-filled cytoplasm , resulted from Rab11 knock down . These results indicate that both endocytosis and membrane recycling are critical to maintain normal PM amount and cortical morphology , suggesting that Collagen IV is trapped in an abnormally expanded cell cortex when endocytic removal of membrane from the PM is prevented . 10 . 7554/eLife . 07187 . 008Figure 3 . Pericellular Collagen trapping is due to PM overgrowth . ( A ) Confocal sections of adipocyte PM ( myr-RFP marker ) . PM expansion is observed in BM-40-SPARC>shii , >Rab5i and >Chci adipocytes , whereas PM flattening occurs in >Rab11i adipocytes along with accumulation of intracellular vesicles . ( B ) Electron micrographs of adipocyte PM from control , BM-40-SPARC>shii , >Rab5i , >Chci and >Rab11i larvae . Internal cell volume indicated through yellow transparency . Asterisks mark pericellular deposits . ( C ) Quantification of PM depth and sinuosity ( see Figure 3—figure supplement 2 ) . Depth measurements obtained from confocal ( n = 12 ) and electron ( n = 10 ) micrographs . PM sinuosity is the ratio between the length of PM between two points on that membrane and the linear distance separating them ( n ≥ 15 ) . Differences with controls were significative as indicated ( Mann–Whitney tests , **p < 0 . 01 , ***p < 0 . 001 ) . ( D ) Adipocytes from wild type larvae grown on lipid-depleted food , BM-40-SPARC>shii larvae and BM-40-SPARC>shii larvae grown on lipid-depleted food . Vkg-GFP accumulation and PM excess are both suppressed by lipid-depletion . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 00810 . 7554/eLife . 07187 . 009Figure 3—figure supplement 1 . ( A ) Confocal images of fat body dissected from first , second and third instar larvae . myr-RFP membrane marker in white . PM convolution is apparent in third instar larvae . ( B ) Confocal ( BODIPY staining ) and electron micrographs showing surface lipid droplets ( arrows ) surrounded by PM in larva 3 adipocytes . ( C ) Electron micrographs of the PM of adipocytes from control w1118 , BM-40-SPARC>shii , >Chci and >Rab5i third instar fat body . Intracellular volume indicated with a transparent yellow layer in right panels . Pericellular deposits of extracellular material marked by asterisks . ( D ) Electron micrographs of the PM of adipocytes from larvae grown in lipid-depleted medium . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 00910 . 7554/eLife . 07187 . 010Figure 3—figure supplement 2 . Schematic explanation of PM sinuosity and depth measurements performed in electron micrographs . See ‘Materials and methods’ section . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 010 To confirm that an excessive amount of PM can act as a barrier to Collagen release after secretion , we aimed at artificially decreasing PM amount in adipocytes . To that end , we cultured flies in medium depleted of lipids through chloroform extraction ( Palm et al . , 2012 ) . In larvae thus grown , we found that the amount of adipocyte PM was greatly reduced ( Figure 3C , D; Figure 3—figure supplement 1 ) . Furthermore , lipid-depleted food suppressed both PM excess and pericellular Collagen accumulation in Dynamin-deficient adipocytes ( Figure 3C , D ) , indicating that PM overgrowth was indeed responsible for pericellular Collagen trapping . These data , in all , show that PM excess resulting from lack of endocytic membrane removal leads to a hyperconvoluted PM , which in turn causes pericellular trapping of Collagen IV . We asked next whether other proteins besides Collagen IV were pericellularly trapped due to PM overgrowth . Apart from Collagen IV , the main components of basement membranes are Perlecan , Laminin and Nidogen ( Yurchenco , 2011 ) . Whereas evidence exists of significant Laminin and Nidogen production outside the fat body ( Urbano et al . , 2009; Zhu et al . , 2012 ) , production of Perlecan has not been studied . Through iYFPi ( in vivo YFP interference , Figure 4A ) , we knocked down expression of Perlecan-YFP ( trolCPTI-002049 [Rees et al . , 2011] ) , a YFP-trap insertion predicted to label all Perlecan isoforms and found that Perlecan present in imaginal discs originated entirely in the fat body ( Figure 4B ) , same as Collagen IV . Also similar to Collagen IV , Trol-YFP was pericellularly accumulated in endocytosis-defective adipocytes ( Figure 5A , B ) . This accumulation of Trol ( terribly reduced optic lobes ) depended on Collagen IV , as it was suppressed by Collagen IV knock-down ( Figure 5C ) . Through antibody staining , we confirmed Perlecan accumulation ( Figure 5—figure supplement 1 ) and additionally observed accumulation of Nidogen , again in a Collagen-dependent manner ( Figure 5D ) , but not Laminin ( anti-LanB1 staining , not shown ) . 10 . 7554/eLife . 07187 . 011Figure 4 . Perlecan , like Collagen IV , originates in the fat body . ( A ) Schematic representation of the in vivo YFP interference strategy ( iYFPi ) to knock-down expression of YFP-trapped Perlecan ( Trol-YFP ) and ascertain its tissue of origin . Expression of a short hairpin RNA targets the YFP sequence in the YFP-trapped mRNA for degradation through RNAi . ( B ) Localization of Perlecan ( Trol-YFP trap ) in wing discs from trolCPTI-002049/Y flies . iYFPi in the fat body ( BM-40-SPARC>iYFPi ) eliminates expression of Trol-YFP in the wing disc and produces tissue hyperconstriction , a previously described trol loss-of-function phenotype ( Pastor-Pareja and Xu , 2011 ) . Phalloidin staining of F-actin in red to reveal disc deformation . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 01110 . 7554/eLife . 07187 . 012Figure 5 . Pericellular deposits in adipocytes are fibrotic . ( A ) Localization in wild type and BM-40-SPARC>shii adipocytes of Trol-YFP ( trolCPTI-002049; see Figure 4 ) . Perlecan accumulates in >shii adipocytes . ( B ) Perlecan accumulation in BM-40-SPARC>shiDN and >Rab5i adipocytes . ( C ) Perlecan accumulation in r4>shii adipocytes is suppressed by Collagen IV knock down . ( D ) Pericellular Nidogen accumulation ( anti-Ndg staining ) in r4>shii adipocytes is suppressed by Collagen IV knock down . ( E ) Pericellular Vkg accumulation in BM-40-SPARC>shii adipocytes is suppressed by knocking down prolyl-hydroxylase PH4αEFB . ( F ) BM-40-SPARC>shii adipocytes do not accumulate secretion marker secr-GFP . Intracellular secr-GFP retention in BM-40-SPARC>Rab1i is shown as a control . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 01210 . 7554/eLife . 07187 . 013Figure 5—figure supplement 1 . ( A ) PM of fat body adipocytes stained with anti-Trol antibody ( magenta ) . Pericellular accumulation of Perlecan ( Trol ) in r4>shii adipocytes is suppressed by additionally knocking down expression of vkg and Cg25C Collagen IV chains . ( B ) Localization of 26-29-protease ( 26-29-pCA06735 GFP-trap ) and Ferritin 1HCH ( Fer1HCHG188 GFP-trap ) in adipocytes from wild type L3 larvae , Cg>shii L3 larvae and Cg>sec23i L2 larvae . Note intracellular accumulation upon sec23 knock-down , which confirms that 26-29-p and Fer1HCH are indeed adipocyte-secreted . ( C ) Western blots of hemolymph extracted from wild type ( w1118 ) , Fer1HCHG188 , Drs-GFP , BM-40-SPARC>ecr-GFP and 26-29-pCA06735 larvae probed with anti-GFP antibody ( 1:5000 ) . The amount of blood loaded in each well is equivalent to 2 . 5 larvae . Whereas Fer1HCH-GFP ( expected molecular weight 50 kDa ) , Drs-GFP ( 34 kDa ) and secr-GFP ( 27 kDa ) are detected in the hemolymph as clear single bands , 26-29-p ( 87 kDa ) seems to be processed . ( D ) Dorsal view of a live larva expressing secr-GFP in adipocytes ( BM-40-SPARC>secr-GFP ) showing accumulation of GFP in pericardial cells , a hemolymph filtering nephrocyte-like cell type . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 013 The fact that the aggregates contained basement membrane components Collagen IV , Perlecan and Nidogen suggested that they were in essence fibrotic ECM deposits . Indeed , Vkg accumulation was suppressed by knock down of the Prolyl-4-hydroxylase PH4αEFB ( Figure 5E ) , in the absence of which Collagen IV is still secreted to the blood as non-functional monomeric chains ( Pastor-Pareja and Xu , 2011 ) . Additionally supporting the fibrotic nature of the deposits , the secretion marker secr-GFP ( GFP coupled to a signal peptide [Pfeiffer et al . , 2002] ) did not accumulate in the expanded cortex of shii adipocytes ( Figure 5F ) . We also examined two other non-ECM proteins secreted by the fat body , 26-29-protease and Ferritin 1HCH , and none of them accumulated pericellularly upon shi knock down ( Figure 5—figure supplement 1 ) . These results demonstrate that not all proteins secreted by the fat body are pericellularly trapped as a result of PM overgrowth . On the contrary , our data support the notion that the aggregates consist of ECM proteins like trimeric Collagen IV , Perlecan and Nidogen , and thus are fibrotic deposits . Furthermore , our results suggest that Collagen IV is the main protein in this deposits , as accumulation of Perlecan and Nidogen depended on the presence of Collagen IV . Having established that extracellular deposits caused by PM overgrowth were not general protein aggregates but rather fibrotic ECM aggregates , we decided to investigate their wider biological effects . To do that , we first turned our attention to cactus , another PM accumulation hit in our screening ( Figure 6A ) . cactus ( cact ) encodes a negative regulator of the Toll signaling pathway , a key mediator of innate immunity ( Roth et al . , 1991 ) . We confirmed activation of Toll signaling in cacti adipocytes ( Dorsal nuclear accumulation ) and further validated the phenotype by observing pericellular Collagen accumulation in cact4 mutants ( Roth et al . , 1991 ) ( Figure 6—figure supplement 1 ) . Expression of the constitutively active mutant receptor Toll10B also produced pericellular Collagen accumulation ( Figure 6B ) , as did infection with Toll-activating Gram+ bacteria Micrococcus luteus ( Figure 6C; Figure 6—figure supplement 1 ) . Examination of cacti and Toll10B adipocytes in confocal and electron micrographs revealed PM overgrowth ( Figure 6D , E ) , similar to shii adipocytes . Also similar to shii adipocytes , PM overgrowth upon Toll activation was suppressed by lipid-depleted food ( graph in Figure 6E ) . Moreover , pericellular Collagen IV deposits were extracellular ( antibody staining without permeabilization , Figure 6—figure supplement 1 ) , disappeared upon PH4αEFB knock-down ( Figure 6F ) and did not contain secr-GFP ( Figure 6G ) , but contained Perlecan ( Figure 6H ) , indicating accumulation of fibrotic material , same as in endocytosis-defective cells . 10 . 7554/eLife . 07187 . 014Figure 6 . Fibrotic deposits and PM overgrowth upon Toll activation . ( A ) Pericellular Vkg deposits ( Vkg-GFP ) and PM overgrowth in BM-40-SPARC>cacti adipocytes . ( B ) Vkg deposits and VNC elongation in BM-40-SPARC>Tl10B larvae . ( C ) Vkg deposits and PM overgrowth in adipocytes 1 day after infection with Micrococcus luteus . ( D ) Electron micrographs of BM-40-SPARC>cacti and >Tl10B adipocytes . The arrow marks connection of the deposits to the extracellular space . ( E ) Measurements of PM depth and sinuosity in adipocytes of indicated genotypes . Depth measurements obtained from confocal ( n ≥ 7 ) and electron micrographs ( n ≥ 10 ) . Sinuosity measured in electron micrographs ( n ≥ 15 ) . Differences with wild type or appropriate control as indicated were significative in all cases ( Mann–Whitney tests , **p < 0 . 01 , **p < 0 . 001 ) . ( F ) Pericellular Vkg accumulation in BM-40-SPARC>cacti and >Tl10B adipocytes is suppressed by knocking down prolyl-hydroxylase PH4αEFB . ( G ) Secretion marker secr-GFP does not accumulate in BM-40-SPARC>cacti or >Tl10B adipocytes . ( H ) Pericellular Perlecan deposits ( Trol-YFP ) in BM-40-SPARC>cacti and >Tl10B adipocytes . ( I ) Induction of antimicrobial peptide Drosomycin ( Drs-GFP ) fills adipocyte cytoplasm in BM-40-SPARC>cacti , >Tl10B and Micrococcus luteus-infected larvae . ( J ) Rab1 knock-down causes intracellular Drosomycin retention and suppresses PM overgrowth in BM-40-SPARC>Tl10B adipocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 01410 . 7554/eLife . 07187 . 015Figure 6—figure supplement 1 . ( A ) Nuclear accumulation of the Toll downstream transcription factor Dorsal ( anti-Dorsal staining ) in BM-40-SPARC>cacti adipocytes . ( B ) Pericellular Cg25C accumulation in adipocytes of mutant cact4 over cact-uncovering deficiency Df ( 2L ) r10 . ( C ) Electron micrographs of the PM of BM-40-SPARC>cacti and >Tl10B adipocytes . Asterisks mark pericellular deposits . Arrows mark visible connections of the deposits to the extracellular space . ( D ) Antibody stainings of cacti and Tl10B adipocytes performed without permeabilization ( no detergent ) in order to detect extracellular Collagen IV . ( E ) Knock-down of Dif decreases the expression of Toll target gene Drosomycin ( Drs-GFP ) activated by cacti and Tl10B ( BM-40-SPARC>cacti and >Tl10B adipocytes ) . ( F ) Localization of endocytic marker TfR-GFP in wild type , Cg>shii and Cg>cacti adipocytes . Intracellular TfR vesicles are seen in cacti adipocytes , same as wild type . ( G ) Drosomycin-containing vesicles ( Drs-GFP ) in the cytoplasm of BM-40-SPARC>Tl10B adipocytes . ( H ) Pericellular retention of Collagen IV ( Vkg-GFP ) at the PM of adipocytes from larvae dissected 1 or 2 days after Micrococcus luteus infection . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 015 Despite these similarities with endocytosis-defective cells , we could not find evidence of endocytosis/recycling defects when we examined TfR-GFP localization in cacti and Toll10B adipocytes ( Figure 6—figure supplement 1 ) , suggesting that PM overgrowth upon Toll activation did not stem from reduced endocytosis . Toll activation , on the other hand , is known to potently induce production of secreted antimicrobial peptides like Drosomycin , Defensin and Metchnikowin ( Lemaitre et al . , 1996 ) , which led as to consider increased secretion of these Toll targets as an alternative explanation to PM overgrowth upon Toll activation . Consistent with this , knock-down of Dif , a transcription factor involved in expression of Toll target genes in the adult ( Rutschmann et al . , 2000 ) and larval fat body ( see Figure 6—figure supplement 1 ) , suppressed PM overgrowth ( graph in Figure 6E ) and pericellular Vkg-GFP trapping ( not shown ) , suggesting that the transcriptional response to Toll activation contributed to Collagen IV accumulation in cacti and Toll10B adipocytes . Also consistent with a burst in secretory activity causing the PM overgrowth , the cytoplasm of cacti , Toll10B and M . luteus-infected adipocytes was filled with Drosomycin vesicles or granules ( Figure 6I; Figure 6—figure supplement 1 ) . Furthermore , knock-down of Rab1 , required for ER-to-Golgi transport in the secretory pathway , caused not just abundant intracellular Drs-GFP retention in Toll10B adipocytes , but also suppression of PM excess ( Figure 6J ) , implicating membrane input from the secretory pathway in PM overgrowth . In all , our analysis indicates that PM overgrowth and fibrotic deposits upon Toll activation result from increased secretory activity . Our experiments indicated that fibrotic aggregates could form downstream of Toll immune signaling . We therefore decided to characterize further the immune effects of these deposits . In our screening , 32 out of 70 hits producing pericellular Collagen IV accumulation showed clear signs of a fat body melanization response , including cact , shi , Rab5 and Chc ( Supplementary file 1 ) . Melanization is an insect immune response characterized by blackening of the affected tissue , usually accompanied by hemocyte ( blood cell ) recruitment ( Minakhina and Steward , 2006 ) . Because BM-40-SPARC-GAL4 and Cg-GAL4 , the strong fat body drivers we had used so far , are also expressed in blood cells , we tested weaker fat body drivers ppl-GAL4 and r4-GAL4 , inactive in blood cells , and found that melanization still occurred ( Figure 7A , B ) , ruling out that interfering with blood cell function caused the response . Furthermore , confirming the involvement of Collagen deposits in the response , both the number of r4>shii larvae displaying melanization and the extent of it decreased when we additionally knocked down Collagen IV or PH4αEFB ( Figure 7C , D ) . Reduction of Collagen IV and PH4αEFB also reduced melanization of r4>Toll10B larvae ( Figure 7D ) and completely rescued their pupal lethality . These results show that Collagen deposits either trigger or significantly contribute to fat body melanization in these conditions . 10 . 7554/eLife . 07187 . 016Figure 7 . Immune response to fibrotic deposits . ( A ) Melanotic fat body in a ppl>shii fly . ( B ) Melanized fat body from an r4>shii larva . Hemocytes ( blood cells ) encapsulate the tissue . ( C ) Knock-down of Collagen IV reduces fat body melanization in r4>shii larvae . Cultures maintained at 30°C . ( D ) Percentage of larvae displaying signs of melanization in indicated genotypes . n ≥ 30 per genotype . Differences with r4>shii and >Tl10B controls were significative ( χ2 tests , ***p < 0 . 001 ) . Cultures maintained at 30°C . ( E ) Induction of c-Jun N-terminal kinase ( JNK ) downstream puckered ( puc-GFP enhancer trap ) in BM-40-SPARC>shii , >Tl10B and >cacti adipocytes . ( F ) Induction of Matrix Metallo-Protease 1 ( anti-Mmp1 staining ) in BM-40-SPARC>shii , >Tl10B and >cacti adipocytes . ( G ) Expression of JAK/STAT activity reporter 10XSTAT-GFP in BM-40-SPARC>shii , >Tl10B and >cacti adipocytes . ( H ) Expression of JAK/STAT-activating ligands in wild type , BM-40-SPARC>shii and >Tl10B adipocytes assessed by real time RT-PCR . Error bars represent 95% confidence intervals . rp49 expression was used for normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 016 Given that Toll activation is long known to cause melanization ( Gerttula et al . , 1988 ) , we examined Toll activity in shii adipocytes , but found that shibire knock-down did not cause Toll pathway activation , as evidenced by lack of Drs-GFP induction and absence of nuclear Dorsal localization ( not shown ) . In contrast , both shibire knockdown and Toll activation caused activation of the c-Jun N-terminal kinase ( JNK ) and Janus kinase/Signal Transducer and Activator of Transcription ( JAK/STAT ) pathways , known to act together in response to tissue damage in other fly tissues ( Pastor-Pareja et al . , 2008; Buchon et al . , 2009 ) . JNK targets puckered ( puc ) and Matrix metallo-protease 1 ( Mmp1 ) ( Uhlirova and Bohmann , 2006 ) were induced in shii , Toll10B and cacti adipocytes ( Figure 7E–G ) . As for JAK/STAT signaling , we observed induction of the JAK/STAT reporter 10XSTAT-GFP ( Ekas et al . , 2006 ) ( Figure 7G ) and highly upregulated expression of unpaired 3 ( upd3 ) , encoding a JAK/STAT-activating cytokine ( Figure 7H ) . From these data , we conclude that fibrotic deposits caused by endocytic defects or Toll provoke tissue damage in adipocytes and stimulate an innate immune response , as evidenced by tissue melanization , blood cell recruitment , activation of JAK/STAT signaling and JNK ( Figure 8 ) . 10 . 7554/eLife . 07187 . 017Figure 8 . PM overgrowth leads to adipocyte fibrosis . Schematic representation summarizing the genesis of fibrotic deposits caused by PM excess and the ensuing reaction by the immune system . Defective endocytosis or excess secretion induced by Toll activity cause PM overgrowth in fat body adipocytes , which leads to hyperconvoluted PM morphology and pericellular trapping of Collagen IV and other extracellular matrix ( ECM ) proteins in the cell cortex . Fibrotic Collagen IV deposits trigger an immune response , as evidenced by tissue melanization and activation of the JAK/STAT and JNK pathways . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 01710 . 7554/eLife . 07187 . 018Figure 8—figure supplement 1 . Knock-down of BM-40-SPARC ( BM-40-SPARC>BM-40-SPARCi ) causes PM accumulation of Collagen IV without PM overgrowth . DOI: http://dx . doi . org/10 . 7554/eLife . 07187 . 018
In this study , we investigated the release of Collagen IV by the fat body adipocytes in Drosophila larvae . Through a genetic screening , we discovered that reducing endocytosis caused unexpected Collagen IV accumulation in secreting cells ( Figure 1 ) . In investigating the cause for such a paradoxical phenotype , our experiments revealed that Collagen IV was effectively secreted and reached the extracellular space , but it was then trapped in fibrotic deposits at the cell periphery ( Figure 2 ) . Further experiments led us to conclude that endocytic membrane removal and recycling are critical to maintain normal PM amount and that excess of PM when endocytic membrane removal was prevented resulted in a hyperconvoluted cell cortex where Collagen IV became entrapped ( Figure 3 ) . This hyperconvoluted cell cortex is reminiscent of the basal labyrinth found in highly secretory cells , such as enterocytes and renal tubule epithelium . Fibrotic deposits in the absence of negative Toll regulator Cactus were similarly caused by PM overgrowth as well ( Figure 6 ) . In this instance , however , PM overgrowth resulted from increased membrane input from the secretory pathway , consistent with the ability of Toll to induce massive secretion of antimicrobial peptides . Supporting this , knock down of Rab1 , required for ER-to-Golgi transport , was able to suppress Toll-induced PM overgrowth . Therefore , both recycling endosome-derived and ER-derived membrane inputs may contribute to total PM amount while constant endocytic membrane removal prevents PM overgrowth . Sudden surges or shifts in membrane traffic are well known drivers of cell shape changes during development ( Lecuit and Pilot , 2003 ) . However , the extent to which maintenance of a stable cell shape is affected by the balance of membrane outputs and inputs to the PM has been less explored . Our experiments show that endocytosis is critical for preserving cortex morphology in Drosophila adipocytes and that preventing endocytosis can cause dramatic PM overgrowth . A factor that might determine high PM turnover in these cells , and thus a tendency to imbalances , is their elevated secretory activity . Apart from Collagen IV , fat body adipocytes secrete most other proteins present in the hemolymph , including the very abundant larval serum proteins ( Guedes Sde et al . , 2003; Vierstraete et al . , 2003 ) , which imposes a constant input of ER membrane to the PM . A need for control of PM amount , however , does not seem exclusive to adipocytes or highly secretory cells based on reported effects in other cell types . For instance , in Garland cells , a filtering cell type with high endocytic activity , shibire loss causes deep PM tubulation ( Kosaka and Ikeda , 1983 ) , whereas loss of Rab11 flattens the PM ( Satoh et al . , 2005 ) . In neurons , accumulation of membrane cisternae in synaptic boutons was observed upon Clathrin and Dynamin photoinactivation ( Heerssen et al . , 2008; Kasprowicz et al . , 2014 ) , similar to the PM excess we found in adipocytes . Finally , Rab5 and other endocytic genes are required to maintain epithelial polarity ( Lu and Bilder , 2005; Zeigerer et al . , 2012 ) , suggesting that constant endocytosis actively keeps apical and baso-lateral PMs segregated . These examples , together with our findings , point to a requirement for endocytosis and recycling in maintaining stable cell shape wider than currently appreciated . In addition to the finding that membrane traffic controls adipocyte PM amount , our results have implications for the understanding of fibrotic deposits . We arrived at the study of membrane traffic in adipocytes through a Collagen IV accumulation phenotype . Collagen IV , however is not the only protein in these deposits ( Figure 5 ) . Also Perlecan , which we show originates in adipocytes as well ( Figure 4 ) , accumulates pericellularly . Except for Nidogen , a third basement membrane component , no other protein we tested accumulated , including Ferritin and Catalase , two of the most abundant proteins in the blood ( Handke et al . , 2013 ) . Even though a more complete view of the aggregates must await a proteomic analysis , our data suggest that Collagen IV , while not the only protein present in these deposits , it is the most important one , as knock down of Collagen IV suppressed accumulation of Perlecan and Nidogen in shii cells . Because of its abundance and ability to self-interact , Collagen IV seems particularly prone in principle to nucleate aggregates in the conditions of limited diffusion provided by a hyperconvoluted PM . That Collagen IV self–interaction is critical in triggering pericellular aggregation is indeed supported by the fact that knock-down of PH4α-EFB , required for trimer formation , suppresses aggregate formation . In contrast , our data showing that PM surrounding aggregates can be stained equally well with both low ( 3000 ) and high ( 70 , 000 ) molecular weight fixable dextrans ( Figure 2—figure supplement 1 ) suggests that the size of the accumulated protein does not play a major role in the formation of the deposits . This is also supported by the finding that Perlecan , a large 400 kDa protein , does not accumulate by itself , but as a result of Collagen IV accumulation ( Figure 5 ) . Besides endocytic genes and cactus , our screening has found additional hits causing Collagen IV accumulation at the PM ( Supplementary file 1 ) . Loss of BM-40-SPARC , one of these hits , was previously known to cause Collagen IV deposits on adipocytes ( Pastor-Pareja and Xu , 2011 ) . Furthermore , similar to the fibrotic deposits in our study , other basement membrane components accumulate upon loss of BM-40-SPARC ( Shahab et al . , 2015 ) . However , the PM is not overgrown in BM-40-SPARCi adipocytes ( Figure 8—figure supplement 1 ) , suggesting that the way in which aggregates form in BM-40-SPARCi adipocytes is different from the way they arise in endocytosis-defective and Toll-activated adipocytes . We have also tested Toll pathway activation by examining expression of Toll target Drosomycin ( Drs-GFP ) in BM-40-SPARCi and all other PM accumulation hits , but found no Drosomycin-GFP induction in any of them . This suggests that , cactus aside , none of the hits causes Collagen IV deposits through ectopic Toll activation . Further characterization work , therefore , is needed to ascertain the particular mechanisms by which PM accumulation of Collagen IV occurs upon loss of these genes . Such work might offer additional insights for a better understanding of fibrosis . Pericellular trapping of Collagen IV , apart from preventing its incorporation to destination tissues , induces a potent response in adipocytes . Aspects of this response include melanization and activation of the JAK/STAT and JNK pathways ( Figure 7 ) . Concomitant activation of JNK and JAK/STAT signaling makes this response reminiscent of tissue damage responses in imaginal discs and gut tissues ( Pastor-Pareja et al . , 2008; Buchon et al . , 2009 ) . In particular , predominance of upd3 among induced JAK/STAT cytokines is shared with the intestinal response to tissue damage ( Buchon et al . , 2009 ) . It has been suggested that basement membrane disruption could be a signal that intestinal cells sense when flies are fed tissue damage-inducing Dextran Sulfate Sodium ( Amcheslavsky et al . , 2009 ) . Absence of basement membrane , although unable by itself to elicit a response ( Pastor-Pareja et al . , 2008 ) , has been recently linked to melanization as a permissive factor ( Hauling et al . , 2014; Kim and Choe , 2014 ) . In the case of adipocyte fibrosis , however , our data indicate that it is the fibrotic deposits that cause the response rather than absence of a basement membrane , since Collagen IV knock down in shii cells suppressed the response instead of enhancing it . Also in Toll10B adipocytes , Collagen IV reduction partially suppressed melanization , showing that , even though Toll targets include melanization enzymes ( De Gregorio et al . , 2002 ) , Collagen deposits contribute to Toll-induced melanization . An interesting question deserving of further investigation is how the fly immune system can react to these fibrotic deposits . One possibility is that fibrotic deposits are directly detected as altered self by the immune system , for instance through recognition of some specific molecular trait in them . A second possibility is that deposits somehow cause damage in adipocytes and this damage activates the response . In this latter case , the nature of the fibrosis-induced cell damage and the way that damage might be sensed are also worth deeper investigation . Adipocyte fibrosis in humans and mouse models is correlated with obesity and adipose tissue dysfunction ( Sun et al . , 2013 ) . Same as human adipocytes , Drosophila larval adipocytes are surrounded by a basement membrane containing Collagen IV ( Chun , 2012 ) ; also similar , human adipocytes display complex PM topology , with abundant caveolae forming higher-level rosettes ( Fan et al . , 1983 ) . Our study , importantly , provides a model to study fibrosis in this tissue . Fibrosis , affecting adipocytes or otherwise , is thought to result in all cases from inflammatory stimulation of excess ECM deposition . Our results show that accumulation of trimeric Collagen IV can be triggered by Toll immune activation and that deposits in turn further stimulate other arms of the fly immune system different from the Toll-mediated immune response . This raises the possibility that fibrotic accumulations , once initiated , could be self-sustained , with inflammation and fibrosis feeding each other . Interestingly , the response to fibrosis includes induction of Collagen-degrading enzyme Matrix metalloprotease 1 , suggesting that the response is aimed in part at clearing the aggregates . Further research into how innate immunity is activated by these deposits may uncover therapeutically relevant mechanisms by which adipocytes and other tissues react to , and deal with , excessive matrix deposition .
Standard fly husbandry techniques and genetic methodologies , including balancers and dominant genetic markers , were used to assess segregation of mutations and transgenes in the progeny of crosses , construct intermediate fly lines and obtain flies of the required genotypes for each experiment ( Roote and Prokop , 2013 ) . Flies were maintained at 25°C unless otherwise stated . In the initial screening and most experiments afterwards , the GAL4-UAS binary expression system ( Brand and Perrimon , 1993 ) was used to drive expression of UAS transgenes in fat body adipocytes under temporal and spatial control of transgenic GAL4 drivers BM-40-SPARC-GAL4 ( Venken et al . , 2011 ) ( a gift from Hugo Bellen ) , Cg-GAL4 ( Asha et al . , 2003 ) ( BL7011 ) , r4-GAL4 ( Lee and Park , 2004 ) ( a gift from Pierre Leopold ) and ppl-GAL4 ( Colombani et al . , 2003 ) ( a gift from Herve Agaisse ) . In flies bearing both types of transgenes ( a GAL4 driver and a UAS responder ) , expression of the UAS-transgene is induced by the yeast transcription factor GAL4 according to the pattern of expression of GAL4 specific to the driver transgene . The strength of these drivers is BM-40-SPARC-GAL4 > Cg-GAL4 > r4-GAL4 > ppl-GAL4 . While BM-40-SPARC-GAL4 and Cg-GAL4 are expressed in blood cells , r4-GAL4 and ppl-GAL4 are not . UAS-Dcr2 was included in the screening strain for the purpose of enhancing RNAi-mediated knock-down by long dsRNA hairpins ( Dietzl et al . , 2007 ) . Dcr2 expression in adipocytes showed no visible effect by itself on Collagen IV localization ( Figure 1—figure supplement 2 ) . For generation of flip-out clones ( Ito et al . , 1997 ) ( Figure 2F , G ) , vials containing L2 larvae were heat-shocked at 37°C for 7 min . For lipid depletion experiments , lipids were extracted from medium by mixing ingredients with chloroform for 2 days , allowing chloroform to evaporate for at least two more days before food preparation ( Palm et al . , 2012 ) . Genotypes of animals in all experiments are detailed in Supplementary file 2 . Origin of mutants and transgenes used can be found in Supplementary file 3 . The following strains were used: Canton-S w1118 y w; vkgG454/CyO w; vkgG454 UAS-myr-RFP; BM-40-SPARC-GAL4 UAS-Dcr2/SM6a-TM6B w; vkgG454/CyO; BM-40-SPARC-GAL4 UAS-myr . RFP/TM6B w; Cg-GAL4 ( II ) w; Cg-GAL4 UAS-myr-RFP ( II ) w; ppl-GAL4 ( II ) w; ppl-GAL4 UAS-myr . RFP vkgG454/CyO w; ppl-GAL4 UAS-myr . RFP vkgG454/CyO; UAS-Dcr2 w; r4-GAL4 ( III ) w; UAS-myr . RFP vkgG454/CyO; r4-GAL4 y sc v; UAS-shi . RNAiTRiP . JF03133 y sc v; UAS-shi . RNAiTRiP . HMS00154 w; UAS-shi . RNAiNIG . 18102R-1 w; UAS-shi . K44A . 3-7 ( II ) w shi1 shi2 y w hs-Flp1 . 22; vkgG454 act-y+-GAL4 UAS-myr . RFP/CyO y w hs-Flp1 . 22; act-y+-GAL4 UAS-GFP w; UAS-Cg25C . GFP . 2 . 1 ( II ) w; UAS-Cg25C . RFP . 2 . 1 ( II ) w; UAS-hTfR . GFP ( III ) y sc v; UAS-Rab11 . RNAiTRiP . JF02812 w UAS-Rab11 . dsRNA . WIZ y sc v; UAS-Rab5 . RNAiTRiP . HMC03420 y sc v; UAS-Rab5 . RNAiTRiP . HMS00147 w; UAS-Rab5 . S43N ( II ) ( BL42703 ) w; UAS-GFP . Rab5 ( III ) ( BL43336 ) y v sc; UAS-Chc . RNAiTRIP . JF02681 y v sc; UAS-Chc . RNAiTRIP . HMS01222 w; UAS-Chc . DN ( III ) ( BL26874 ) y v sc; UAS-RN-tre . RNAiTRiP . JF03085 ( III ) y sc v; UAS-AP-2α . RNAiTRiP . HMS00653 ( III ) y sc v; UAS-AP-2μ . RNAiTRiP . JF0287 ( III ) y sc v; UAS-Hrs . RNAiTRiP . JF02860 ( III ) y sc v; UAS-Par-1 . RNAiTRiP . GL00253 ( III ) w trolCPTI-002049 y sc v; UAS-EGFP . shRNA . 3 ( II ) ( BL41559 ) y sc v; UAS-EGFP . shRNA . 3 ( III ) ( BL41560 ) y sc v; UAS-EGFP . shRNA . 4 ( III ) ( BL41553 ) y sc v; UAS-EGFP . shRNA ( II ) ( BL35782 ) ( knocks down GFP and YFP efficiently same as three above constructs; unlike those , it causes unspecific cell death in imaginal discs ) w; UAS-vkg . RNAiNIG . 16858R-3 ( III ) w; UAS-vkg . RNAiVDRC . v16986 ( III ) w; UAS-vkg . RNAiVDRC . v106812 ( II ) w; UAS-Cg25C . RNAiVDRC . v28369 ( III ) w; UAS-vkg . RNAiNIG . 16858R-3 UAS- Cg25C . RNAiVDRC . v28369/TM6B w; UAS-PH4αEFB . RNAiVDRC . v2464 ( III ) w; UAS-sec23 . RNAiVDRC . v24552 ( III ) w; 26-29-pCA06735 ( III ) w; Fer1HCHG188 ( III ) y sc v; UAS-cact . RNAiTRiP . GL00627 ( II ) y sc v; UAS-cact . RNAiTRiP . HMS00084 ( III ) cact4/CyO , act-GFP Df ( 2L ) r10 , cn1/CyO w; UAS-Tl10B ( II ) y sc v; UAS-Dif . RNAi TRiP . HM05191 ( III ) y w Drs-GFP . JM804 y sc v; UAS-Rab1 . RNAiTRiP . JF02609 ( III ) w; pucG462/TM6B w; 10XSTAT-GFP ( II ) To obtain UAS-Cg25C-GFP and UAS-Cg25C-RFP lines , the open reading frame ( ORF ) of Cg25C was cloned into vectors pTWG and pTWR ( Drosophila Carnegie Vector collection ) using Gateway recombination ( Life Technologies , Carlsbad , California ) . To do this , we PCR-amplified the Cg25C ORF from cDNA clone RE33133 ( Drosophila Genomics Resource center , Bloomington ) using primers Cg25C-NF: 5′-GGGG ACA AGT TTG TAC AAA AAA GCA GGC TTC ATG TTG CCC TTC TGG AAG CG-3′ and Cg25C-CF: 5′-GGG GAC CAC TTT GTA CAA GAA AGC TGG GTC CGA GGA GTT CTT CAT GCA CA-3′ , thus adding att sites at the 5′ and 3′ termini of the ORF for subsequent Gateway cloning . The product of this reaction was purified by gel extraction with AxyPrep DNA Gel Extraction Kit ( Axygen , Union City , California ) and recombined into vector pDONR221 ( Life Technologies ) with Gateway BP Clonase Enzyme Mix ( Life Technologies ) . The resulting plasmid ( Entry clone ) , was transformed into Escherichia coli Trans5α competent cells , miniprepped and recombined with destination vectors pTWG ( C-terminal GFP ) and pTWR ( C-terminal RFP ) using Gateway LR Clonase Enzyme Mix ( Life Technologies ) . Transgenic lines were obtained through standard P-element transgenesis ( Spradling and Rubin , 1982 ) . For generation of anti-Cg25C antibody , rabbits were immunized with epitope SVKHYNRNEPKFPIDDSY by AbMax Biotechnology ( Beijing , China ) . The following antibodies and dyes were used: rabbit anti-Cg25C ( 1:1 , 000 , this study ) , mouse anti-rat Dynamin ( 1:250 , BD BioScience , Franklin Lakes , New Jersey ) , rabbit anti-Ndg ( Wolfstetter et al . , 2009 ) ( 1:2000 ) , rabbit anti-Pcan ( Friedrich et al . , 2000 ) ( 1:2000 ) , rabbit anti-LanB1 ( 1:500 , Abcam , Cambridge , UK ) , mouse anti-Mmp1 ( Page-McCaw et al . , 2003 ) ( 1:200 , DSHB , University of Iowa ) , mouse anti-Dorsal ( Whalen and Steward , 1993 ) ( 1:100 , DSHB ) , rabbit anti-GFP ( 1:500 , EASYBIO , Seoul , South Korea ) , anti-mouse IgG conjugated to Alexa-488 , Alexa-555 or Alexa-633 ( 1:200 , Life Technologies ) , Texas-Red phalloidin ( 1:100 , Life Technologies ) and BODIPY 493/503 ( 1:500 , Life Technologies ) . Antibody stainings of whole fat body were performed following standard procedures used for imaginal discs . Briefly , larvae were pre-dissected in PBS by turning inside out with fine tip forceps after removing their posterior end . Carcasses with fat body still attached were fixed in PBS containing 4% PFA , washed in PBS ( 3 × 10 min ) , blocked in PBT-BSA ( PBS containing 0 . 1% Triton X-100 detergent , 1% BSA and 250 mM NaCl ) , incubated overnight with primary antibody in PBT-BSA , washed in PBT-BSA ( 3 × 20 min ) , incubated for 2 hr with secondary antibody in PBT-BSA and washed in PBT-BSA ( 3 × 20 min ) and PBS ( 3 × 10 min ) . Fat body tissues were finally dissected on a slide with a drop of DAPI-Vectashield ( Vector Labs , Burlingame , California ) and mounted in this same medium . FM4-64 staining of PM was performed by incubating fat body in ice-cold PBS containing 5 mg/ml fixable FM4-64 ( Life Technologies ) for 1 min before fixing for 20 min in ice-cold PBS containing 4% PFA . Labeling with fixable 3 , 000 and 70 , 000 MW Texas-Red-coupled dextran ( 0 . 1 mg/ml , Life Technologies ) was performed by fixing fat body in 4% PFA for 5 min followed by three quick washes , incubation with dextran in PBS , three quick washes and fixation again for 20 min . Results were the same when the first fixation was omitted . The screening was conducted in a Leica MZ10F stereoscope ( Wetzlar , Germany ) . For confocal microscopy ( Zeiss LSM780 microscope , Oberkochen , Germany ) , tissues were mounted in DAPI-Vectashield ( Vector Labs ) . Vkg-GFP intensity in basement membranes was quantified with ImageJ software as previously described ( Pastor-Pareja and Xu , 2011 ) . PM depth and sinuosity measurements were performed in electron micrographs as indicated in Figure 3—figure supplement 2 . For measurements of both PM depth and sinuosity , images were analyzed with ImageJ using the segmented line tool to calculate distances . Graphs were drawn and statistical analysis performed with GraphPad Prism . For sinuosity measurements , the distance separating the points where the PM intersects the edges of the micrograph was calculated , as well as the length of the PM joining those two points . Islands of cytoplasm with no connection to the cell body were excluded from sinuosity analysis , but not for PM depth measurement . All fat body images in the manuscript correspond to mid-to-late third larval instar fat body dissected prior to the wandering larva stage , except for the first and second larval instar images in Figure 3—figure supplement 1A and sec23i second instar larvae in Figure 5—figure supplement 1B . Fibrotic aggregates can be observed at all times during the third larval instar , including the extremes of early third instar and white prepupa . Ultrathin sections were obtained following standard procedures and imaged in a Hitachi H-7650B microscope ( Tokyo , Japan ) . Briefly , larvae were turned inside-out and fixed in 2 . 5% gluteraldehyde , leaving fat body attached to carcasses to facilitate handling . Once fixed , fat body was postfixed in 1% osmium tetroxide before embedding . Sections were stained in 2% uranyl acetate/lead citrate . Fat body from at least three different specimens per genotype was imaged . Gram+ bacterium M . luteus ( ATCC 4698 ) was cultured in LB medium at 30°C . Larvae were soaked in concentrated culture and pricked with a tungsten needle to allow infection . RNA was isolated using TransZol Up ( Transgen , Beijing , China ) . cDNA was synthesized from 1 μg of RNA with the PrimeScriptTM RT-PCR Kit ( Takara , Kyoto , Japan ) . Analysis was performed in a CFX96 Touch system ( Bio-Rad , Hercules , California ) using SYBR Green Fast kit ( Applied Biosystems , Waltham , Massachusetts ) . rp49 expression was used for normalization . Three experiments per genotype were averaged . A biological replicate was performed with same results . Primers used were: upd-For: 5′ TCCACACGCACAACTACAAGTTC 3′; upd-Rev: 5′ CCAGCGCTTTAGGGCAATC 3′; upd2-For: 5′ AGTGCGGTGAAGCTAAAGACTTG 3′; upd2-Rev: 5′ GCCCGTCCCAGATATGAGAA 3′; upd3-For: 5′ TGCCCCGTCTGAATCTCACT 3′; upd3-Rev: 5′ GTGAAGGCGCCCACGTAA 3′; rp49-For: 5′ GGCCCAAGATCGTGAAGAAG 3′; rp49-Rev: 5′ ATTTGTGCGACAGCTTAGCATATC 3′ . Hemolymph from third instar larvae was collected by turning larvae inside-out with fine-tip forceps while immersed in 100 μl of PBS in a glass depression well . PBS contained 0 . 1 mg/ml of phenylthiourea to avoid melanization and 1 mg/ml of protease inhibitor . 10 μl of 2-Mercaptoethanol-reduced sample were loaded per genotype into a 4–15% Mini-PROTEAN TGX gel ( Bio-Rad ) . The amount of sample loaded in each experiment given in figure legends as larva-equivalents , depending on the number of larvae collected . Precision Plus Protein Dual Color Standards ( Bio-Rad ) was used as a molecular weight marker . Gels were blotted with rabbit anti-GFP antibody ( 1:5000 , EASYBIO ) or rabbit anti-Cg25C ( 1:5000 , this study ) and revealed with anti-rabbit-HRP ( 1:10 , 000 , Abmart ) and an ECL Plus kit ( Pierce , Rockford , Illinois ) . | In animals , so-called ‘basement membranes’ surround organs and help to both anchor certain tissue types together and control which molecules move between them . The basement membrane is made up of various proteins , and a large protein called Collagen IV is the most abundant component . Collagen IV is made inside cells and packaged into bubble-like compartments called vesicles . These vesicles then merge with the cell membrane , which releases the collagen into the space outside the cell . Sometimes , after it has been released from the cell , Collagen IV forms harmful aggregates that the body finds difficult to break down . This condition is known as fibrosis , and can severely damage organs and tissues . Zang , Wan et al . have now studied how fat cells—also known as adipocytes—in the fruit fly Drosophila melanogaster release Collagen IV . This fly is widely used to study collagen production because it is relatively easy to perform genetic investigations on it , and it releases collagen from its cells in the same way as many other species . Unexpectedly , it was observed that proteins that control a process known as endocytosis—which takes substances into the cell—are also involved in releasing Collagen IV from the cell . Zang , Wan et al . found that this is because endocytosis removes part of the cell membrane: if endocytosis is blocked , then the excess cell membrane traps Collagen IV molecules after they have been released , causing aggregates like those seen during fibrosis . However , artificially decreasing the amount of cell membrane restored normal collagen release . Zang , Wan et al . next found that a pathway called Toll , which is important for protecting flies against infections , can also affect collagen release . When a protein that inactivates the Toll pathway is absent , too much cell membrane grows and Collagen IV forms aggregates as well . In both cases , Toll activation or lack of endocytosis , the aggregates trigger a reaction that damages the adipocytes . Understanding this reaction in more detail could help to develop treatments for conditions that produce fibrosis . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"immunology",
"and",
"inflammation"
] | 2015 | Plasma membrane overgrowth causes fibrotic collagen accumulation and immune activation in Drosophila adipocytes |
Sperm are equipped with a unique set of ion channels that orchestrate fertilization . In mouse sperm , the principal K+ current ( IKSper ) is carried by the Slo3 channel , which sets the membrane potential ( Vm ) in a strongly pHi-dependent manner . Here , we show that IKSper in human sperm is activated weakly by pHi and more strongly by Ca2+ . Correspondingly , Vm is strongly regulated by Ca2+ and less so by pHi . We find that inhibitors of Slo3 suppress human IKSper , and we identify the Slo3 protein in the flagellum of human sperm . Moreover , heterologously expressed human Slo3 , but not mouse Slo3 , is activated by Ca2+ rather than by alkaline pHi; current–voltage relations of human Slo3 and human IKSper are similar . We conclude that Slo3 represents the principal K+ channel in human sperm that carries the Ca2+-activated IKSper current . We propose that , in human sperm , the progesterone-evoked Ca2+ influx carried by voltage-gated CatSper channels is limited by Ca2+-controlled hyperpolarization via Slo3 .
Mammalian sperm fulfil several demanding functions during fertilization: sperm track down the oocyte presumably by chemotaxis , rheotaxis , or thermotaxis ( Bahat et al . , 2003; Eisenbach and Giojalas , 2006; Kaupp et al . , 2008; Miki and Clapham , 2013 ) . Moreover , sperm break through the oocyte’s vestments by hyperactivation and acrosomal exocytosis ( Florman et al . , 2008; Ho and Suarez , 2001 ) . Sperm acquire these skills inside the female genital tract during a maturation process called capacitation ( Fraser , 2010 ) . Navigation , capacitation , hyperactivation , and acrosomal exocytosis are controlled by changes in intracellular pH ( pHi ) , membrane voltage ( Vm ) , and intracellular Ca2+ concentration ( [Ca2+]i ) ( Ho and Suarez , 2001; Eisenbach and Giojalas , 2006; Florman et al . , 2008; Kaupp et al . , 2008; Publicover et al . , 2008 ) . These cellular responses are orchestrated by a set of unique ion channels ( Darszon et al . , 2011; Lishko et al . , 2012; Santi et al . , 2013 ) . A picture has emerged that the inventory and control of ion channels in mouse and human sperm are surprisingly different . For example , human but not mouse sperm harbor functional proton Hv1 channels ( Lishko et al . , 2010 ) ; purinergic P2X channels are functional in mouse ( Navarro et al . , 2011 ) , but not in human sperm ( Brenker et al . , 2012 ) ; the sperm-specific CatSper ( cation channel of sperm ) Ca2+ channel is activated by progesterone in humans ( Lishko et al . , 2011; Strünker et al . , 2011 ) , but not in mouse ( Lishko et al . , 2011 ) . These different channel inventories and different mechanisms of channel activation might reflect adaptations to species-specific challenges encountered by sperm in the female genital tract . In mouse sperm , an alkaline-activated K+ current , called IKSper , is a critical determinant of Vm and , thereby , controls other Vm-dependent channels ( Navarro et al . , 2007 ) . Mouse IKSper is carried by the sperm-specific Slo3 channel . Deletion of the Slo3 gene abolishes IKSper ( Santi et al . , 2010; Zeng et al . , 2011 , 2013 ) ; male Slo3−/− mice are infertile due to defects in sperm motility ( Santi et al . , 2010; Zeng et al . , 2011 ) , osmoregulation ( Santi et al . , 2010; Zeng et al . , 2011 ) , and acrosomal exocytosis ( Santi et al . , 2010 ) . In humans , it is unknown whether Slo3 is functionally expressed in sperm and serves a similar key role for fertilization . Here , we examine the properties of human sperm K+ current by patch-clamp recording and also define properties of currents arising from heterologous expression of hSlo3 and its auxiliary subunit hLRRC52 ( Yang et al . , 2011 ) . We find that human IKSper and heterologously expressed human Slo3 currents share similar biophysical properties , pharmacology , and ligand dependence . Furthermore , we identify Slo3 and LRRC52 proteins in human sperm . Remarkably , whereas mouse Slo3 is exclusively controlled by pHi ( Schreiber et al . , 1998; Zhang et al . , 2006a; Yang et al . , 2011; Zeng et al . , 2011 ) , activation of human Slo3 is regulated by [Ca2+]i and also , more weakly , by cytosolic alkalization . These results show that , between mouse and human sperm , signalling pathways controlling the principal K+ channel and , thereby , Vm are also distinctively different .
We recorded currents from human sperm by the patch-clamp technique ( Lishko et al . , 2013 ) . Depolarizing voltage steps from a holding potential of −80 mV evoked outwardly rectifying voltage-gated currents ( Figure 1A , B ) . At pHi 7 . 3 , current amplitudes at −100 mV and 100 mV were −7 . 5 ± 5 pA and 80 ± 15 pA , respectively ( n = 5 ) ( mean ± SD; n = number of experiments ) ( Figure 1F ) . Several controls established that the currents are carried by K+ channels and not by Cl− channels or CatSper ( Zeng et al . , 2013 ) : lowering the extracellular K+ concentration ( [K+]o ) from 150 to 5 mM shifted the reversal potential ( Vrev ) from 9 . 2 ± 1 . 5 mV to −16 . 5 ± 10 mV ( n = 5 ) ( Figure 1B , C ) . At low [K+]o , a decrease of extracellular [Cl−]o did not change Vrev any further ( Figure 1C , Figure 1—figure supplement 1A , B ) , showing that currents are not carried by Cl− channels . Replacing intracellular K+ by Cs+ almost completely abolished outward currents at Vm ≤ 100 mV ( Figure 1F , Figure 1—figure supplement 1C , D ) . However , at Vm ≥ 100 mV , residual Cs+ outward currents persisted . In mouse Slo3−/− sperm , monovalent outward currents persisting at very positive Vm are carried by CatSper ( Zeng et al . , 2013 ) . Monovalent mouse and human CatSper current is suppressed by extracellular Ca2+ ( Kirichok et al . , 2006; Lishko et al . , 2011; Lishko et al . , 2012; Zeng et al . , 2013 ) . Consistent with CatSper channels conducting the residual Cs+ current in human sperm , current amplitudes at 120 mV were progressively suppressed by increasing extracellular Ca2+ ( Figure 1—figure supplement 1E , F ) . 10 . 7554/eLife . 01438 . 003Figure 1 . Voltage- and alkaline-activated K+ currents in human sperm . ( A ) Whole-cell currents before and after application of 10 mM NH4Cl . Traces at 35 mV and 85 mV are depicted in blue and red , respectively . ( B ) Current-voltage relation of recordings from ( A ) and currents recorded in 5 mM [K+]o . ( C ) Mean Vrev of currents at pHi 7 . 3 in different extracellular solutions ( n = 3–5 ) . ( D ) Currents recorded at pHi 6 . 2 . ( E ) Current-voltage relation of recordings from ( D ) . ( F ) Mean currents before and after application of NH4Cl ( 10 mM ) and with Cs+-based intracellular solution ( 180 mM Cs+ ) ( n = 3–6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 00310 . 7554/eLife . 01438 . 004Figure 1—figure supplement 1 . Voltage-gated currents in human sperm are carried by K+ channels . ( A ) Whole-cell currents from human sperm recorded in K+-based intracellular solution at pHi 7 . 3 and in extracellular solutions containing ( in mM ) : 5 K+/150 Cl− or 5 K+/7 Cl− . Current traces at +35 mV and +85 mV are depicted in blue and red , respectively . ( B ) Current-voltage relation of recordings from part ( A ) . ( C ) Whole-cell currents recorded from human sperm in Cs+-based ( left ) and K+-based ( right ) intracellular solutions at pHi 7 . 3 in 5 mM extracellular K+ . Current traces at 85 mV and 135 mV are depicted in red and green , respectively . ( D ) Mean current–voltage relation of recordings as in panel C . ( E ) Whole-cell currents recorded from human sperm in Cs+-based intracellular solutions at pHi 7 . 3 in 5 mM extracellular K+ and Ca2+ as indicated . Current traces at +85 mV and +135 mV are depicted in red and green , respectively . ( F ) Mean current–voltage relation recorded from human sperm in Cs+-based intracellular solutions at pHi 7 . 3 in HS ( 2 mM Ca2+ , 2 mM Mg2+ ) , HS 100 µM Ca2+ ( 2 mM Mg2+ ) , HS 0 Ca2+ ( 2 mM Mg2+ ) , and NaDVF ( 0 Ca2+ , 0 Mg2+ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 004 We conclude that monovalent cation currents at Vm ≤ 100 mV are carried by K+-selective channels; we refer to this current as human IKSper , analogous to the principal K+ current in mouse sperm . At more positive Vm , currents are also carried by CatSper channels . Mouse IKSper is strongly pH dependent ( Navarro et al . , 2007; Santi et al . , 2010; Zeng et al . , 2011; Zhang et al . , 2006a ) . We , therefore , examined the pH sensitivity of IKSper and Vm in human sperm . Decreasing pipette pH from 7 . 3 to 6 . 2 reduced outward currents by 2 . 2 ± 1 . 0-fold ( n = 4 ) ( Figure 1A , D , F ) . Moreover , intracellular alkalization by NH4Cl enhanced outward currents by 1 . 8 ± 0 . 9-fold ( at 100 mV , n = 4 ) at pipette pH 6 . 2 ( Figure 1D–F ) , but not at pipette pH 7 . 3 ( Figure 1A , B , F ) . Thus , human IKSper is less sensitive to pHi than mouse IKSper , which is enhanced about fourfold by increasing pHi from 6 . 5 to 7 . 5 ( Navarro et al . , 2007 ) . In mouse sperm , IKSper sets Vm in a strongly pHi-dependent manner ( Navarro et al . , 2007; Santi et al . , 2010; Zeng et al . , 2011 ) . Therefore , we tested under current-clamp whether , in human sperm , the control of Vm by IKSper is dependent on pHi . At pipette pH 6 . 2 , Vm was −23 ± 5 mV ( n = 4 ) ( Figure 2A , C left ) ; raising pipette pH to 7 . 3 or alkalization by NH4Cl hyperpolarized sperm only slightly to −30 ± 5 mV and −31 ± 4 mV , respectively ( n = 4 ) ( Figure 2A , B , C left ) . At pipette pH 7 . 3 , alkalization by NH4Cl did not further change Vm ( Figure 2B , C left ) . Of note , at pHi 6 . 2 and 7 . 3 , Vm was independent of [Cl−]o and dropped to about 0 mV at high [K+]o ( Figure 2A , B , C left ) . In conclusion , under the recording conditions used here , Vm of human sperm is only modestly pHi sensitive . 10 . 7554/eLife . 01438 . 005Figure 2 . Vm of human sperm is controlled by [Ca2+]i rather than pHi . ( A ) Current-clamp recording from human sperm ( pHi 6 . 2 ) in extracellular solutions containing different [K+] and [Cl−] ( in mM ) . Intracellular alkalization was evoked by superfusion with 10 mM NH4Cl . ( B ) Current-clamp recording at pHi 7 . 3 . ( C ) Left panel: mean Vm under conditions as described in panel A and B; right panel: mean Vm at indicated [Ca2+]i , and at 1 mM [Ca2+]i under conditions described in panel D ( n = 3–4 ) . ( D ) Current-clamp recording at pHi 7 . 3 and 1 mM [Ca2+]i . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 005 Given the modest effect of alkalization on human K+ current , we tested whether Ca2+ controls IKSper in human sperm . To compare current–voltage ( I–V ) relations at low and high [Ca2+]i in the same cell , we studied voltage activation of IKSper before and after rapid photorelease of Ca2+ from caged Ca2+ ( DMNP-EDTA ) ( Figure 3A , B ) , while monitoring [Ca2+]i with the Ca2+ indicator Fluo-4 . Prior to Ca2+ release , currents were similar to those recorded without Ca2+ in the pipette solution ( Figure 3A , B ) . Photorelease of Ca2+ altered IKSper in several ways: currents activated more rapidly; at Vm ≤ 70 mV , amplitudes were enhanced and at Vm ≥ 70 mV , amplitudes saturated or even declined; thereby , the outward rectification was diminished; finally , Vrev was shifted to more negative potentials ( Figure 3A , B ) . 10 . 7554/eLife . 01438 . 006Figure 3 . Ca2+ enhances K+ currents in human sperm . ( A ) Whole-cell currents recorded at pHi 7 . 3 with 5 mM DMNP-EDTA , 4 mM Ca2+ , and 10 µM Fluo-4 in 5 mM extracellular K+ solution before ( control ) and after photorelease of Ca2+; for simplification , only currents evoked by −105 , −55 , −5 ( grey ) , 45 , and 95 mV are depicted . ( B ) Current-voltage relation of recordings from A . ( C ) Whole-cell-currents at pHi 7 . 3 at 0 ( left ) , 40 µM ( middle ) and 1 mM [Ca2+]i ( right ) in extracellular solutions containing 5 mM K+; for simplification , only currents evoked by −105 , −55 , −5 ( grey ) , 45 , and 95 mV are depicted . ( D ) Current-voltage relations of currents recorded at 2 . 5–1000 µM [Ca2+]i; I–V curves were normalized to the amplitude at 115 mV . ( E ) Current-voltage relation of currents recorded at pHi 7 . 3 and 1 mM [Ca2+]i; extracellular solutions contained 150 mM K+ or 5 mM K+ . ( F ) Mean Vrev of currents at pHi 7 . 3 , 1 mM [Ca2+]i , and different [K+]o and [Cl−]o ( in mM ) ( n = 3–4 ) . ( G ) Whole-cell currents recorded at pHi 6 . 2 , 1 mM [Ca2+]i , and 5 mM [K+]o , before and after superfusion with 10 mM NH4Cl . ( H ) Current-voltage relation of recordings from ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 00610 . 7554/eLife . 01438 . 007Figure 3—figure supplement 1 . Ca2+-activated currents in human sperm are carried by K+ channels . ( A ) Whole-cell currents from human sperm recorded in K+-based intracellular solution at pHi 7 . 3 and 1 mM [Ca2+]i in extracellular solutions containing ( in mM ) : 5 K+/150 Cl− or 5 K+/7 Cl− . ( B ) Current-voltage relation of recordings from panel ( A ) . ( C ) Mean currents from human sperm recorded in K+-based intracellular solution at pHi 7 . 3 and different [Ca2+]i in 5 mM extracellular K+ ( n = 3–9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 007 The Ca2+ dependence of IKSper was quantified by recording the I-V relation at 2 . 5-1000 µM Ca2+ in the pipette ( Figure 3C , D ) . In the absence of Ca2+ , the mean current amplitude ( Vm = 3 . 5 mV ) was 3 . 5 ± 3 pA; increasing Ca2+ to 40 µM and 1 mM Ca2+ increased the amplitude to 14 ± 6 pA and 32 ± 17 pA , respectively ( n = 3 ) ( Figure 3C ) . For [Ca2+]i > 10 µM , current amplitudes were increased in a concentration-dependent manner , and the normalized I–V relation and Vrev were shifted to more negative potentials ( Figure 3D , Figure 3—figure supplement 1C ) . At Vm ≿ 100 mV , Ca2+-activated currents levelled off or even declined . Control experiments showed that Ca2+ activated IKSper , and not Ca2+-activated Cl− channels that were identified in human sperm ( Orta et al . , 2012 ) : with 1 mM [Ca2+] in the pipette , a decrease of [K+]o from 150 to 5 mM shifted Vrev from −5 . 1 ± 10 . 3 mV to −52 . 9 ± 7 . 8 mV ( n = 3 ) ( Figure 3E , F ) ; decreasing [Cl−]o did not alter Vrev ( Figure 3F , Figure 3—figure supplement 1A , B ) . Moreover , Ca2+-activated IKSper was enhanced to some extent by alkalization . At pHi 6 . 2 , NH4Cl increased the mean current amplitude from 52 ± 14 pA to 89 ± 8 pA ( 50 mV ) and shifted Vrev from −33 ± 7 mV to −54 ± 14 mV ( n = 3 ) ( Figure 3G , H ) . Thus , the enhancement of IKSper upon alkalization was similar in the presence and absence of intracellular Ca2+ . In current-clamp mode , Vm was −30 ± 5 mV and −35 ± 7 mV at 0 and 1 µM [Ca2+] in the pipette , respectively; 30 µM and ≥ 70 µM [Ca2+]i changed Vm to −43 ± 4 mV and about −60 mV , respectively ( n = 3-4 ) ( Figure 2C , right ) . At high [Ca2+]i , Vm was independent of [Cl−]o , but dropped to about 0 mV at high [K+]o ( Figure 2C right , D ) . In conclusion , under the conditions used here , Vm is set by IKsper that is controlled strongly by Ca2+ and only modestly by pHi . The pH sensitivity of IKSper , although modest , is reminiscent of the IKSper current in mouse sperm carried by Slo3 channels , whereas the Ca2+ sensitivity is reminiscent of the prototypical Ca2+-activated K+ channel Slo1 ( Salkoff et al . , 2006 ) . To identify the ion channel underlying human IKSper , we tested several inhibitors of Slo3 and Slo1 channels at high [Ca2+] in the pipette , that is when Ca2+-activated K+ channels are strongly activated . The non-selective K+ channel inhibitors quinidine and clofilium , previously shown to inhibit mouse IKSper ( Navarro et al . , 2007; Zeng et al . , 2011 ) and heterologous Slo3 ( Tang et al . , 2010 ) , abolished currents ( Figure 4A , B , E ) . The inhibition by clofilium was irreversible ( Figure 4B ) , which is a hallmark of its action on IKSper in mouse sperm ( Navarro et al . , 2007; Zeng et al . , 2011 ) . Moreover , perfusion of sperm with quinidine and clofilium depolarized the cell ( Figure 4F ) . In contrast , tetraethylammonium ( TEA ) and iberiotoxin ( IBTX ) , which block Slo1 but not Slo3 ( Tang et al . , 2010 ) , neither affected IKSper nor Vm of human sperm ( Figure 4C–F ) . Although the Slo3 inhibitors employed are not selective for Slo3 , the action of these drugs together with the ineffective Slo1 inhibitors provides critical evidence that IKSper is carried by Slo3 , but not by Slo1 . 10 . 7554/eLife . 01438 . 008Figure 4 . Ca2+-activated K+ currents in human sperm exhibit hallmarks of Slo3 channels . ( A–D , G ) Current-voltage relation of whole-cell currents from human sperm recorded in K+-based intracellular solution at pHi 7 . 3 and 1 mM [Ca2+]i in 5 mM extracellular K+ before , during , and after application of inhibitors . ( E ) Mean outward currents at 65 mV in the presence of 500 µM quinidine , 10 mM TEA , 50 µM clofilium , or 100 nM IBTX . ( F ) Current-clamp recording from human sperm in intracellular solution ( pHi 7 . 3 ) containing 180 mM K+ and 1 mM Ca2+ . Sperm were bathed in extracellular solution containing 5 mM K+ and 150 mM Cl− . Concentrations of drugs were as in ( A–E , G ) . ( H ) Current trace recorded at pHi 7 . 3 and 70 µM [Ca2+]i at −60 mV in HS ( top ) and K-based HS ( bottom ) Filter: 2 kHz . ( I ) Segments indicated by the red bars in panel ( H ) shown on an extended time scale ( 5 kHz ) , revealing opening events of one ( top ) and two K+ channels ( bottom ) . Red lines correspond to conductance levels of 0 ( c ) , 65 ( o ) , and 130 pS ( o ) . ( J ) Histogram of current amplitudes recorded in 5 mM K+ and 150 mM K+ , at the conditions described in ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 008 Surprisingly , MDL12330A , an inhibitor of CatSper ( Brenker et al . , 2012 ) , also blocked Ca2+-activated IKSper and depolarized the cell under conditions ( extra- and intracellular 1 mM Ca2+ ) where CatSper currents are negligibly small ( Figure 4F , G ) . As a further signature of the K+ channels , we compared single-channel currents of heterologous hSlo3 with those recorded in human sperm . At −60 mV and high intracellular [Ca2+]i , K+ channels openings displayed a single-channel conductance of 60–70 pS ( Figure 4H–J ) , that is similar to that of hSlo3 ( 70 pS; Figure 6—figure supplement 1; Zhang et al . , 2006b ) , but not to that of Slo1 ( 280 pS ) ( Dworetzky et al . , 1994 ) . These results demonstrate that the channel underlying IKSper displays pharmacological and functional properties consistent with hSlo3 . Next , we tested for the presence of Slo1 and Slo3 proteins in human sperm . Using targeted protein mass-spectrometry ( MS ) ( Figure 5A , Supplementary file 1 ) , we identified in purified human sperm 5 and 3 proteotypic peptides corresponding to Slo3 and its auxiliary subunit LRRC52 , respectively . As positive controls , we identified the following proteins known to be expressed in human sperm: CatSper , the Ca2+-ATPase PMCA4 , the proton channel Hv1 , Na+/K+-ATPase α4 , and IZUMO . However , we did not detect Slo1 . Similar results were obtained by shot-gun proteomics ( Wang et al . , 2013 ) , which identifies Slo3 and other known components of human sperm , but not Slo1 . 10 . 7554/eLife . 01438 . 009Figure 5 . Human sperm express Slo3 and its auxiliary subunit LRRC52 . ( A ) Predicted membrane topology of hSlo3 ( yellow ) and hLRRC52 ( red ) polypeptides . Proteotypic peptides identified in human sperm by targeted protein mass-spectrometry are indicated in grey . ( B ) Western blot of total proteins of human sperm , CHO cells ( wt ) , and CHO cells transfected with HA-tagged hSlo3 ( hSlo3 ) . The Western blot was probed with an anti-hSlo3 and anti-HA antibody . The molecular masses ( kDa ) of the protein standard are indicated on the right . ( C ) Human sperm stained with an antibody directed against hSlo3 ( red ) . The DNA in the head was stained with DAPI ( blue ) . Scale bar: 30 µM . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 00910 . 7554/eLife . 01438 . 010Figure 5—figure supplement 1 . Specificity of anti-hSlo3 antibody . CHO cells heterologously expressing human Slo3 that was modified with a C-terminal hemagglutinin ( HA ) tag were stained with an anti-hSlo3 antibody ( green ) and an antibody directed against the HA-tag ( red ) . DAPI ( blue ) was used to label the nucleus . Scale bar: 10 µM . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 010 Moreover , we expressed hSlo3 heterologously with a hemagglutinin ( HA ) -tag in chinese hamster ovary ( CHO ) cells ( Figure 5—figure supplement 1 ) . In Western blots of hSlo3-transfected , but not of wild type cells , both anti-HA and anti-hSlo3 antibodies labelled polypeptides with an apparent molecular weight ( Mw ) of about 120 kDa ( Figure 5B ) . The predicted Mw of Slo3 is 130 kDa . In Western blots of human sperm , the anti-Slo3 antibody labelled polypeptides of ∼125 kDa and ∼80 kDa ( Figure 5B ) ; in both polypeptide bands , we confirmed by MS that the bands recognized by the antibody contained hSlo3 . The 80 kDa polypeptide might be a product of the cleaved Slo3 channel . Post-translational cleavage has been reported also for other ion channels in ciliary structures ( Molday et al . , 1991; Bönigk et al . , 2009 ) . Finally , the anti-hSlo3 antibody stained the flagellum ( Figure 5C ) , consistent with the localization of IKSper in mouse sperm ( Navarro et al . , 2007 ) . Together , MS , Western blot-analysis , and immunocytochemistry show that human sperm express Slo3 . Considering that mouse Slo3 is insensitive to Ca2+ ( Schreiber et al . , 1998 ) , the Ca2+ activation of human IKSper is remarkable . Therefore , we studied human Slo3 co-expressed with hLRRC52 in Xenopus oocytes . First , we investigated the pHi sensitivity of hSlo3 in inside-out patches with step depolarizations between −60 and 260 mV from a holding potential of −140 mV . In the absence of Ca2+ , currents were only modestly activated at pHi 7 ( Figure 6A , upper left ) , but enhanced at pHi 8 ( Figure 6A , upper right ) . Similar to previous observations ( Leonetti et al . , 2012 ) , an increase of pHi from 7 to either pHi 8 or 9 increased currents by 1 . 9 ± 0 . 4-fold and 2 . 2 ± 0 . 5-fold , respectively ( 200 mV; n = 4 ) . 10 . 7554/eLife . 01438 . 011Figure 6 . Activation of heterologous hSlo3 by intracellular Ca2+ . ( A ) Families of hSlo3 + hLRRC52 currents in oocytes at pH 7 and 8 with indicated [Ca2+]i . Current trace at +200 mV is depicted in red . ( B and C ) Current-voltage relations of tail currents determined at −140 mV; amplitudes were normalized to the amplitude evoked by step to 200 mV , 0 [Ca2+]i , and pH 8 . ( D ) Tail current amplitudes ( activated by 200 mV ) as function of [Ca2+]i for pH 7 and pH 8 . Normalization as in ( B ) . ( E ) Current-voltage relation of steady-state hSlo3 + hLRRC52 currents in CHO cells at pHi 7 . 3 and 0 , 70 µM , and 1 mM [Ca2+]i . Currents were normalized to the amplitude evoked at 115 mV . ( F ) Current-voltage relation of hSlo3 + hLRRC52 currents in CHO cells and IKSper recorded from human sperm at 0 and 1 mM [Ca2+]i ( pHi 7 . 2 ) . Normalization as in panel ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 01110 . 7554/eLife . 01438 . 012Figure 6—figure supplement 1 . Ca2+ increases hSlo3 single-channel openings at −60 mV . ( A , C , E ) hSlo3 + hLRRC52 openings at −60 mV . Filter: 2 kHz . ( B , D , F ) Segments highlighted with red bars in panel ( A , C , E ) at faster time base ( 5 kHz filtering ) . ( G ) Total amplitude histogram of a set of current records as in ( B , D , F ) . ( H ) NPo at 60 and 300 µM Ca2+ normalized to NPo at 0 Ca2+ ( n = 7 ) . Red symbols: individual estimates . Black symbols: mean and SD . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 01210 . 7554/eLife . 01438 . 013Figure 6—figure supplement 2 . Currents carried by hSlo3 co-expressed with hLRRC52 in CHO cells . ( A ) Whole-cell hSlo3 + hLRRC52 currents at pHi 7 . 3 . Traces at +35 mV and +85 mV are depicted in blue and red , respectively . ( B ) Current-voltage relation of hSlo3 + hLRRC52 currents recorded in 140 mM and 5 mM [K+]o . ( C ) Currents recorded at pHi 6 . 2 , before and after intracellular alkalization by NH4Cl ( 10 mM ) . ( D ) Current-voltage relation of recordings from panel C . ( E ) Mean current amplitudes at pHi 6 . 2 and pHi 7 . 3 before and after application of NH4Cl ( 10 mM ) ( n = 3 ) . ( F ) Currents recorded at 1 mM [Ca2+]i and 5 mM [K+]o . ( G ) Current-voltage relation of recordings at 1 mM [Ca2+]i in 140 mM and 5 mM [K+]o . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 013 At pHi 7 and pHi 8 , 60 μM Ca2+ enhanced both outward currents and inward tail currents ( Figure 6A , middle traces ) . At 300 μM Ca2+ , inward tail currents were further enhanced but outward currents were reduced at potentials ≥ 200 mV ( Figure 6A , bottom traces ) , indicating a block of outward currents by Ca2+ . Conductance-voltage ( G-V ) curves generated from tail currents illustrate the activation of hSlo3 at pHi 7 and pHi 8 in the presence of 0 , 60 , and 300 μM Ca2+ ( Figure 6B , C ) . At pHi 7 , raising Ca2+ from 0 to 300 μM enhanced hSlo3 conductance by 6 . 6 ± 1 . 7-fold ( 200 mV ) , in contrast to the only 2 . 2-fold increase evoked by raising pHi from 7 to 9 . Thus , Ca2+ activates hSlo3 much more effectively than alkaline pHi . Surprisingly , the amplitudes of Ca2+-activated tail currents , whether at 60 or 300 µM Ca2+ , were similar at pHi 7 and pHi 8 ( compare Figures 6B and 6C ) , suggesting that , at elevated [Ca2+]i , Slo3 is rather insensitive to changes in pHi> 7 . At pHi 8 , we also examined the action of a broader range of [Ca2+]i . For [Ca2+]i ≥ 10 µM , hSlo3 tail currents increased over at least three orders of magnitude of [Ca2+] ( Figure 6D ) . Finally , Slo3 single-channel openings in patches held at −60 mV ( pHi 8 ) were also markedly increased by 60 µM and 300 μM Ca2+ ( Figure 6—figure supplement 1 ) , indicating that Ca2+ enhances Slo3 activation also at physiological Vm . Although the ranges of pHi and Ca2+ concentrations that affect IKSper and hSlo3 expressed in oocytes are similar , compared to IKSper , the I–V relation of hSlo3 currents appears shifted to more positive potentials ( compare Figure 1B , E and 3B with Figure 6B , C ) . This difference might be due to the non-mammalian expression system , the excised-patch conditions , differences in ionic composition of solutions , or a combination of all of them . Therefore , we recorded whole-cell currents evoked by voltage steps in CHO cells that co-expressed hSlo3 and hLRRC52 , using conditions similar to those used for sperm recordings . hSlo3 currents in CHO cells were also modestly enhanced by alkalization ( Figure 6—figure supplement 2 ) , and Ca2+ shifted the I-V relation to more negative Vm ( Figure 6E , Figure 6—figure supplement 2 ) . In the absence and presence of Ca2+ , the I–V relations of hSlo3 and IKSper were similar ( Figure 6F ) . Taken together , the heterologous expression demonstrates that human Slo3 is a Ca2+-activated rather than a strictly alkaline-activated K+ channel . These results strengthen our conclusion that Slo3 underlies the voltage- , Ca2+- , and alkaline-activated IKsper current in human sperm . Although mouse Slo3 is insensitive to intracellular Ca2+ ( Schreiber et al . , 1998 ) , we wondered whether the auxiliary subunit LRRC52 ( Yang et al . , 2011 ) might confer Ca2+ sensitivity on Slo3 channels . Therefore , we co-expressed mSlo3 with mLRRC52 . Raising pHi from 7 to 8 strongly enhanced mSlo3 steady-state ( Figure 7A ) and tail ( Figure 7B ) currents , similar to previous results ( Schreiber et al . , 1998; Zhang et al . , 2006a; Yang et al . , 2011 ) . At pHi 8 , Ca2+ ( 60 or 300 µM ) did not enhance currents ( Figure 7A , B ) . Instead , outward mSlo3 currents were strongly suppressed by Ca2+ ( Figure 7A ) and even tail-current amplitudes were reduced ( Figure 7B ) . A G-V plot derived from tail currents confirms that Ca2+ does not activate mSlo3 , but inhibits mSlo3 following activation by positive potentials ( Figure 7C ) . Currents carried by mSlo3 co-expressed with hLRRC52 were also not activated , but suppressed by Ca2+ ( Figure 7—figure supplement 1 ) . These results exclude the possibility that hLRRC52 confers Ca2+ sensitivity on hSlo3 . Suppression of mSlo3 tail currents by Ca2+ ( Figure 7C , D ) reflects persistent occupancy of the mSlo3 pore by Ca2+ following repolarization . We note that the voltage-dependent suppression of currents by Ca2+ is much more pronounced for mSlo3 compared to hSlo3 . Ca2+ occludes the pore of mSlo3 with about 10-fold higher affinity than that of hSlo3 ( Figure 7—figure supplement 2 ) ; therefore , inhibition of hSlo3 by Ca2+ occurs only at very positive , non-physiological Vm . 10 . 7554/eLife . 01438 . 014Figure 7 . Co-expression of mLRRC52 does not confer Ca2+ sensitivity on mSlo3 . ( A ) Currents were activated with 0 [Ca2+]i pH 7 , 0 [Ca2+]i pH 8 , 60 µM [Ca2+]i pH 8 , and 300 µM [Ca2+]i pH 8 . Red trace corresponds to step to 60 mV . ( B ) Larger gain display of tail currents from ( A ) . ( C ) Normalized current–voltage relations of tail currents determined at −140 mV; currents were normalized to tail current amplitude following the step to 200 mV . ( D ) Normalized tail-current amplitude as a function of [Ca2+]i . Tail currents were normalized to 0 [Ca2+]i at pH 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 01410 . 7554/eLife . 01438 . 015Figure 7—figure supplement 1 . Co-expression of hLRRC52 with mSlo3 does not confer Ca2+ sensitivity on mSlo3 . ( A ) Currents were activated by the indicated voltage-protocol with 0 Ca2+ pH 7 , 0 Ca2+ pH 8 , 60 µM Ca2+ pH 8 , and 300 µM Ca2+ pH 8 , from top to bottom . Red trace corresponds to step to +60 mV . ( B ) Larger gain display of tail currents from panel ( A ) . ( C ) Comparison of currents activated at the indicated voltages at pH 8 with either 0 Ca2+ ( black ) or 60 µM Ca2+ ( red ) . ( D ) Normalized conductances determined from tail currents following each activation voltage for mSlo3 + hLRRC52 channels for the indicated conditions . ( E ) Normalized conductance as a function of Ca2+ determined from tail currents following activation at either +100 or +200 mV at pH 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 01510 . 7554/eLife . 01438 . 016Figure 7—figure supplement 2 . mSlo3 is more sensitive to voltage-dependent block by Ca2+ than hSlo3 . ( A ) Steady-state conductance-voltage ( G-V ) relationships for hSlo3 + hLRRC52 at different [Ca2+]i . ( B ) Steady-state G-V relationships for mSlo3 + mLRRC52 at different [Ca2+]i and pHi . ( C ) Steady-state G-V relationships for mSlo3 + hLRRC52 at different [Ca2+]i and pHi . ( D ) Fractional inhibition of conductance at three voltages is plotted as a function of [Ca2+]i; lines indicate the fit of I ( Ca2+ ) = 1/ ( 1+[Ca2+]/KD ) . For each [Ca2+]i , steady-state conductance reflects both the increase in conductance from Ca2+-dependent inactivation and voltage-dependent block . Values of steady-state conductance from a panel were , therefore , corrected to reflect the measured Ca2+-dependent increase of conductance determined from tail currents . KD values were 128 . 0 ± 77 . 6 µM , 428 . 3 ± 165 . 1 µM , and 5 . 4 ± 2 . 8 mM for +240 , +200 and +160 mV , respectively . ( E ) Fractional inhibition of mSlo3 + mLRRC52 as a function of Ca2+ is plotted along with the best fit curves . KD values were 16 . 9 ± 8 . 2 , 43 . 7 ± 12 . 9 , and 350 . 2 ± 132 . 6 µM , for +240 , +200 , and +160 mV , respectively . ( F ) Fractional inhibition of mSlo3 + hLRRC52 as a function of Ca2+ . KD values were 20 . 8 ? 13 . 9 , 52 . 2 ? 23 . 0 , and 345 . 8 ± 215 . 3 µM for +240 , +200 , and +160 mV , respectively . ( G ) Voltage-dependence of the KD for Ca2+ inhibition is plotted for mSlo3 + mLRRC52 along with the values for hSlo3 + hLRRC52 and mSlo3 + hLRRC52 . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 016 Together , our results show that regulation of mSlo3 and hSlo3 channels by cytosolic ligands is distinctively different , paralleling the differential regulation of IKsper in mouse and human sperm by pHi and Ca2+ . In human sperm , the female sex hormone progesterone directly activates CatSper ( Lishko et al . , 2011; Strünker et al . , 2011; Brenker et al . , 2012; Smith et al . , 2013 ) . Progesterone-evoked Ca2+ influx via CatSper has been implicated in sperm chemotaxis , hyperactivation , and acrosomal exocytosis ( Blackmore et al . , 1990; Publicover et al . , 2007; Publicover et al . , 2008 ) . We examined whether stimulation of human sperm by progesterone enhances Ca2+ levels sufficient to activate Slo3 channels . Sperm were loaded with Ca2+ indicators of different Ca2+ affinity as surrogates for high- to low-affinity Ca2+-binding sites ( Figure 8A , C ) . The progesterone-evoked transient Ca2+ response was faithfully tracked by high- ( KD = 0 . 35 µM ) , moderate- ( KD = 9 . 7 µM ) , and low-affinity Ca2+ indicators ( KD = 90 µM ) ( Figure 8A ) . The Ca2+ ionophore ionomycin evoked a sustained fluorescence increase reflecting the indicator response to near saturating , millimolar [Ca2+]i ( Figure 8A ) . For indicators with a KD value ≤ 2 . 3 µM , the amplitudes of progesterone- and ionomycin-induced Ca2+ signals were similar ( Figure 8A , B ) , suggesting that these high-affinity indicators become saturated with Ca2+ during the progesterone-evoked response . For indicators with KD values ≥ 9 . 7 µM , the amplitude ratio of progesterone-evoked/ionomycin-evoked Ca2+ signals decreased with increasing KD values . However , even for the low-affinity indicator Fluo-5N , which reports [Ca2+]i changes in a concentration range of about 9–900 µM , the amplitude ratio was as large as 0 . 25 . Thus , considering the dynamic range of indicators ( Figure 8C ) , our results indicate that Ca2+ levels reached during a physiological Ca2+ response are sufficient to activate Slo3 . 10 . 7554/eLife . 01438 . 017Figure 8 . Progesterone-evoked Ca2+ responses in human sperm . ( A ) Ca2+ signal evoked by progesterone ( 2 µM ) and ionomycin ( 2 µM ) in sperm loaded with different Ca2+ indicators . ( B ) Relative amplitude of the progesterone- vs ionomycin-induced Ca2+ signal in sperm loaded with indicators of various Ca2+ sensitivity . ( C ) Dynamic range of Ca2+ sensitivity for different indicators assuming 1:1 binding of Ca2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 017
Here , we show that the biophysical and pharmacological properties of human IKSper conform with the properties of hSlo3 , but not with those of hSlo1 or other members of the Slo family . First , hSlo3 and IKSper are modestly sensitive to pHi ≤ 7 . 0 and are more strongly activated by Ca2+ . Second , the I–V relations of hSlo3 and IKSper are similar , in the absence and presence of intracellular Ca2+ . Third , several Slo3 , but not Slo1 inhibitors block IKSper . Fourth , we identify the Slo3 protein and its auxiliary subunit LRRC52 in human sperm . Finally , both human IKSper ( Mannowetz et al . , 2013 ) and heterologously expressed hSlo3 ( Figure 9A , C ) are inhibited by progesterone; progesterone inhibits human IKSper and hSlo3 with constants of half-maximal inhibition ( Ki ) of 7 . 5 µM ( Mannowetz et al . , 2013 ) and 17 . 5 ± 2 µM ( n = 3 ) , respectively . 10 . 7554/eLife . 01438 . 018Figure 9 . Progesterone inhibits hSlo3 but not hSlo1 . ( A ) Whole-cell hSlo3 + hLRRC52 currents recorded in CHO cells at pHi 7 . 3 before and after perfusion with 30 µM progesterone . ( B ) hSlo1 currents recorded in outside-out patches excised from CHO cells at pHi 7 . 3 and 70 µM [Ca2+]i before and after perfusion with 30 µM progesterone . ( C ) Relative amplitude of hSlo1 and hSlo3 + hLRRC52 currents at 80 mV in CHO cells in the presence of progesterone . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 01810 . 7554/eLife . 01438 . 019Figure 9—figure supplement 1 . Lack of homology among Slo3 sequences in the ligand-sensing cytosolic domain . ( A ) Alignments of human Slo1 and Slo3 from various species are shown for the membrane-associated , pore-forming part of the channels , indicating the relatively high extent of conservation through this part of the Slo3 protein . The Slo1 N-terminus is omitted to minimize effects of S0–S1 linker gaps on the alignment . Slo1 numbering starts from amino acids MDAL . Tick marks below each segment of residues counts every 10 residues in human Slo3 . ( B ) Alignments of human Slo1 and Slo3 from various species are shown for the cytosolic gating ring domain beginning with the conserved sequence at the beginning of the first RCK domain . Blue highlights residue differences between human Slo1 and human Slo3 . Yellow highlights differences of Slo3 of various species to human Slo3 . Alignments were generated by Clustal 1 . 2 . 0 and minor adjustments were made based on structural considerations ( Leonetti et al . , 2012 ) . Above the residues , the correspondence of particular amino acid segments to structurally defined α-helical and β-strand segments is shown based on Leonetti et al . ( 2012 ) . In red , residues or segments identified in Slo1 or Slo3 isoforms which are implicated in ligand-sensing or species-specific functional differences are highlighted . Although extensive information is available regarding loci important in ligand-sensing in Slo1 , such information for Slo3 remains lacking . Numbers identify the following: 1 , the sequence of residues termed the Ca2+ bowl ( Schreiber and Salkoff , 1997 ) , for which there is good correspondence of mutations affecting Ca2+-dependent function ( Bao et al . , 2004 ) and coordination of density in a crystal structure ( Yuan et al . , 2012 ) ; 2 , the D367 residue implicated in the role of the RCK1 domain in Ca2+-dependent activation ( Xia et al . , 2002 ) which is clearly distinct from Ca2+-bowl dependent activation ( Zeng et al . , 2005 ) ; 3 , the M513 residue , which also affects Ca2+−dependent activation involving the RCK1 domain ( Bao et al . , 2002 ) , but probably is not involved in ligand coordination; 4 , residues E374 and E399 which have been implicated in low affinity effects of divalent cations , specifically Mg2+ ( Shi et al . , 2002; Xia et al . , 2002; Yang et al . , 2006 ) ; 5 , residue E535 which may also be involved in Ca2+ coordination in RCK1 ( Zhang et al . , 2010 ) ; 6 , residues H365 and H394 , which have been implicated in proton-dependent activation of Slo1 and also influence Ca2+-dependent activation when protonated ( Hou et al . , 2008 ) ; 7 , H417 and segment 368–475 , which influence pH-sensing in mouse Slo3 ( Zhang et al . , 2006 ) ; 8 , segment 495–515 in bovine Slo3 which accounts for part of the different in functional properties between mouse Slo3 and bovine Slo3 ( Santi et al . , 2009 ) . Illustrated sequences and accession numbers include: HsSlo1 ( Homo sapiens ) , NP_001154824 , Gene ID 3778; HsSlo3 ( Homo sapiens ) , NP_001027006 , Gene ID 157855; MmSlo3 ( Mus musculus ) , NP_032458 , Gene ID 16532; RnSlo3 ( Rattus norvegicus ) , XP_006253398 , Gene ID 680912; TcSlo3 ( Tupaia chinensis , Chinese tree shrew ) , XP_006171561 , Gene ID 102493286; CcSlo3 ( Condylura cristata , star-nosed mole ) , XP_004682520 , Gene ID 101620543; CfSlo3 ( Canis lupus familiaris ) , XP_539971 , Gene ID 482856; BtSlo3 ( Bos taurus ) , NP_001156721 , Gene ID 524144; OaSlo3 ( Ovis aries , sheep ) , XP_004021821 , Gene ID 10110209 . DOI: http://dx . doi . org/10 . 7554/eLife . 01438 . 019 While this manuscript was under review , Mannowetz et al . ( 2013 ) reported that the prototypical Ca2+-activated member of the Slo channel family , Slo1 , is the principal K+ channel in human sperm . Several observations strongly argue against this conclusion . First , Slo1 is inhibited rather than activated at alkaline pH ( Avdonin et al . , 2003 ) . Second , specific inhibitors of Slo1 did not inhibit IKSper . Third , human Slo1 is largely insensitive to progesterone ( Figure 9B , C ) . Fourth , we and others ( Wang et al . , 2013 ) are unable to identify the Slo1 protein in human sperm . Fifth , if Slo1 carried the current of about 125–150 pA in human sperm ( recorded at 100 mV in symmetrical K+ and saturating Ca2+ ) ( Figure 3A , C ) , this would correspond to the opening of about 5–6 BK channels . Under such conditions , discrete opening and closing transitions of single BK channels would be readily visible; not only at 100 mV , but even more so at voltages < 100 mV . Thus , recordings of IKSper current in human sperm are consistent with lower conductance openings characteristic of Slo3 , but not Slo1 . Finally , the most obvious discrepancy between the two reports concerns the pharmacology . Mannowetz et al . show that K+ currents in human sperm are abolished by the Slo1 inhibitors IBTX , charybdotoxin ( CTX ) , and paxilline . Although we do not know the reason for this discrepancy , there are differences in experimental conditions . We tested the inhibitors at 1 mM extracellular [Ca2+] to prevent monovalent CatSper currents and at 1 mM intracellular [Ca2+] to strongly activate IKSper . Mannowetz et al . tested the drugs at 100 µM extracellular [Ca2+] and in the absence of intracellular Ca2+ . Under these conditions , sizeable CatSper currents are recorded ( Figure 1—figure supplement 1E , F ) , but activation of Slo1 would be minimal . What might be the function of Slo3 in sperm ? It has been suggested that the IKSper-mediated hyperpolarization reinforces Ca2+ influx via CatSper by increasing the electrical driving force ( Clapham , 2013 ) . Alternatively , we propose that the hyperpolarization serves as a negative feedback that decreases the open probability of CatSper and , thereby , curtails rather than enhances Ca2+ influx . Why did Slo3 switch in human sperm from a strictly pH-sensitive to a pH- and Ca2+-sensitive K+ channel ? In mouse , Slo3 and CatSper are both voltage- and strongly alkaline-activated . In human but not in mouse sperm , CatSper is directly activated by progesterone and prostaglandins ( Lishko et al . , 2011; Strünker et al . , 2011; Brenker et al . , 2012; Smith et al . , 2013 ) . Thus , human CatSper mediates a ligand- rather than an alkaline-activated Ca2+ influx . Moreover , the pH sensitivity of both Slo3 and CatSper is considerably lower in human compared to mouse sperm , suggesting that Ca2+ activation of Slo3 may have evolved in concert with ligand activation of CatSper . This co-evolution might ensure that ligand-evoked Ca2+ influx via CatSper is coupled to the Ca2+-controlled hyperpolarization via Slo3 . Thus , by curtailing Ca2+ influx via CatSper , IKsper may serve a similar role in both mouse and human sperm despite its differential regulation by intracellular ligands . The control of Vm by Ca2+ and pHi and the interplay of CatSper and Slo3 deserve further study in intact , freely moving human sperm with non-invasive and kinetic techniques , for example using voltage-sensitive dyes . Our results suggest that during a progesterone response , global Ca2+ levels can reach concentrations > 10 µM , sufficient to activate Slo3 . CatSper and Slo3 are both located in the principal piece . Moreover , progesterone-induced Ca2+ signals originate in the flagellum and propagate in a tail-to-head direction ( Servin-Vences et al . , 2012 ) . Due to the miniscule flagellar volume ( about 2 . 5 fl ) , opening a few CatSper channels , each conducting several thousand Ca2+ ions per second , would increase local flagellar [Ca2+] to levels that should readily exceed global [Ca2+]i . The potential interplay of Slo3 and CatSper in sperm is reminiscent of the interplay in neurons between Ca2+-activated K+ channels and voltage-gated Ca2+ channels ( Cav ) ( Prakriya and Lingle , 2000 ) . In neurons , Slo1 and Cav channels interact to form local microdomains of Ca2+ signalling near the plasma membrane ( Berkefeld et al . , 2006 ) . In microdomains , [Ca2+] can rise to levels ≿ 100 µM that are readily sensed by Ca2+-activated K+ channels ( Rizzuto and Pozzan , 2006 ) . It needs to be shown whether Slo3 and CatSper are organized in similar microdomains . Channels of the Slo family have been studied as models for allosteric regulation of gating by ligands ( Magleby , 2003; Lingle , 2007 ) . Slo channels and a large number of bacterial channels/transporters harbor a homologous octameric intracellular domain , dubbed the gating ring , that provides a template for ligand regulation of the pore domain ( Lingle , 2007 ) . The gating-ring motif has evolved for regulation of transmembrane ion flux by nucleotides , Ca2+ , H+ , Na+ , Cl− , and probably other cytosolic ligands ( Salkoff et al . , 2006 ) . Among different Slo isoforms , gating rings share structural similarities ( Wu et al . , 2010; Leonetti et al . , 2012; Yuan et al . , 2012 ) . However , the conformational changes that couple binding to gating remain incompletely defined and the location of ligand-binding sites varies markedly ( Salkoff et al . , 2006 ) . Our study adds an interesting twist , demonstrating that , even among Slo3 orthologues , the gating-ring motif has been exploited for regulation by different ligands — primarily Ca2+ in hSlo3 and primarily H+ in mSlo3 . Alignment of human Slo1 sequence with human Slo3 and other mammalian Slo3 sequences illustrates that , although the membrane-associated S0-S6 domain retains considerable identity ( Figure 9—figure supplement 1A ) , there is extensive lack of identity in many segments of the ligand-sensing cytosolic domain ( Figure 9—figure supplement 1B ) . Given that positions of ligand-sensing determinants in regulator of K+ conductance ( RCK ) -containing proteins vary markedly within the RCK domain structures , it is not surprising that simple examination of the Slo3 sequence alignments does not reveal obvious determinants of either pH sensitivity in mSlo3 or Ca2+−sensitivity in hSlo3 . At first glance , the fact that Slo3 orthologues differ in their ligand-dependence seems highly unusual . Yet , it is well-established that proteins essential for fertilization are rapidly evolving; orthologues often display a low degree of sequence similarity ( Swanson and Vacquier , 2002; Cai and Clapham , 2008 ) . For a set of mammalian species ( human , bovine , mouse , dog , and opossum ) , the amino-acid identities ( excluding a non-specific linker in the cytosolic domain ) among Slo1 orthologues is typically > 92 . 6% ( mSlo1 vs hSlo1: 99 . 5% ) . Slo3 orthologues exhibit identities in the range of 57–75% ( mSlo3 vs hSlo3: 70 . 4% ) . Considering that the amino-acid identity between the pH-regulated mSlo3 and Ca2+-regulated mSlo1 is 45 . 7% , the different ligand dependence between hSlo3 and mSlo3 is not so surprising .
IBTX was purchased from Tocris ( Minneapolis , MN , USA ) . DMNP-EDTA and Ca2+ indicators were purchased from Invitrogen ( Carlsbad , CA , USA ) . UPLC grade formic acid , acetonitrile and methanol were purchased from Biosolve ( Valkenswaard , The Netherlands ) . All other reagents were obtained from Sigma-Aldrich ( St . Louis , MO , USA ) . Samples of human semen were obtained from healthy volunteers with their prior consent . Sperm were prepared as described ( Strünker et al . , 2011 ) . In brief , sperm were purified by a ‘swim-up’ procedure in human tubular fluid ( HTF+ ) containing ( in mM ) : 97 . 8 NaCl , 4 . 69 KCl , 0 . 2 MgSO4 , 0 . 37 KH2PO4 , 2 . 04 CaCl2 , 0 . 33 Na-pyruvate , 21 . 4 lactic acid , 2 . 78 glucose , 21 HEPES , and 25 NaHCO3 adjusted between pH 7 . 3-7 . 4 with NaOH . After washing , human serum albumin ( HSA , 3 mg/ml; Irvine Scientfic , USA ) was added to HTF+ ( referring to as HTF++ ) . Sperm were incubated for at least 2 hr in HTF++ at 37°C in a 10% CO2 atmosphere . Under these conditions , sperm undergo capacitation . All recordings were done on capacitated sperm . After purification by a ‘swim-up’ procedure , human sperm were lysed by several ‘freeze/thaw’ cycles and sonification steps in buffer containing ( in mM ) : 10 HEPES pH 7 . 5 , 2 EGTA , 1 DTT , protease inhibitor cocktails ( Roche Applied Science and Sigma , Mannheim , Germany ) , and DNaseI ( AppliChem , Darmstadt , Germany ) . Membranes were sedimented by centrifugation ( 100 , 000×g , 30 min , 4°C ) and membrane proteins were processed by in-gel digestion , in-solution digestion , or a FASP protocol . Peptides were subjected to 1D-ESI-LC-MS/MS using a nanoAcquity UltraPerformance LC System ( Waters , Milford , Massachusetts , USA ) coupled to an LTQ Orbitrap Velos or Elite instrument ( Thermo , Waltham , Massachusetts , USA ) . The resulting tandem MS data were searched using the Sequest algorithms embedded in Proteome Discoverer 1 . 2 ( Thermo ) against a SwissProt/UniProtKB human protein sequence database ( including 56 , 582 entries ) . The mass tolerance for precursor ions was set to ≤10 ppm; the mass tolerance for fragment ions was set to ≤ 1 amu . For search result filtering , the false discovery rate ( FDR ) was set to < 1% and only peptides with search result rank 1 were accepted for identification . For targeted mass spectrometry , the instrument control software used a list of theoretical tryptic peptide masses for the proteins of interest for subsequent CID fragmentation , that is once a mass from the list was detected in the orbitrap full scan , it was preferred over all other co-eluting masses , independent of its signal intensity . Primary antibodies: Slo3 ( Abcam , Cambridge , UK , catalog no . ab104630; for ICC , 1:50 , 000 , for WB , 1:100 ) , HA ( Roche Applied Science , catalog no . 11867431001; for ICC , 1:1 , 000 , for WB , 1:5000 ) . Secondary antibodies: for ICC , rabbit Cy3-conjugated and rat Cy5-conjugated antibodies ( 1:400; Dianova , Hamburg , Germany ) ; for WB , rabbit-HRP and rat-HRP ( 1:5000; Dianova ) . EST clone BC028701 was obtained from Open Biosystems and verified by sequencing . Based on analysis of the NCBI slo3 gene ( kcnu1 , NM_001031836 ) , 62 base pairs ( bp ) present around the S10 region in BC028701 correspond to an intron left over from incomplete mRNA splicing . The intron sequences were removed by site-directed mutagenesis . In addition , there are at least three polymorphic sites present in the EST clone corresponding to amino-acid positions 192 , 739 , and 768 of hSlo3 . The three sites were changed to match those of the human genomic sequence ( NCBI reference sequence: NC_000008 . 10; Chromosome 8 ) . This results in 192W , 739R , and 768W . The full length coding sequence of hSlo3 was subcloned into the oocyte expression vector pXMX ( see details in Tang et al . , 2010 ) . The hSlo3 sequence used here is identical to that used in another study ( Leonetti et al . , 2012 ) . An hLRRC52 ( NM_001005214 . 3 ) clone was generated from two HEK genomic DNA fragments of 622 bp and 320 bp which correspond to hLRRC52 exons 1 and 2 , respectively . The two fragments were amplified via over-lapping PCR and subcloned into the pXMX vector . For the mCherry-tagged hLRRC52 , seven glycines were added as a spacer between the carboxy-terminus of hLRRC52 and the amino-terminus of mCherry; the hLRRC52-mCherry construct was also subcloned into pXMX vector . All cDNA clones were verified by sequencing . For expression in oocytes , cRNA was synthesized by SP6 polymerase after the cDNA template was linearized with the restriction enzyme MluI . For expression in CHO cells , the full length coding sequence of hSlo3 was amplified from human testis cDNA . A perfect Kozak consensus sequence preceding the start codon and a sequence coding for a carboxy-terminal hemagglutinin tag ( HA-tag ) were added . The coding sequence harbored an arginine at the polymorphic site at position 768 . The DNA was subcloned into a pcDNA3 . 1 ( + ) vector ( Invitrogen ) ; the sequence coding for the neomycin resistance gene was replaced by the coding sequence for either citrine or EGFP . The hLRRC52-mCherry construct described above was also subcloned into the pcDNA3 . 1 ( + ) vector for expression in CHO cells . All cDNA clones were verified by sequencing . Human sperm and CHO cells expressing hSlo3 were immobilized on glass coverslips and fixed for 3 min with 4% paraformaldehyde in phosphate buffered saline ( PBS ) containing ( in mM ) : 137 NaCl , 2 . 7 KCl , 6 . 5 Na2HPO4 , 1 . 5 KH2PO4 , pH 7 . 4 . To block nonspecific binding sites , cells were incubated for 20 min with blocking buffer ( 0 . 5% Triton X-100 and 5% chemiblocker ( Merck Millipore , Germany ) in PBS ) . Primary antibodies were diluted in blocking buffer and cells were incubated for 1 hr at room temperature . After washing with PBS , cells were incubated with fluorescent secondary antibodies in blocking buffer containing 0 . 5 µg/µl 4’ , 6-diamidino-2-phenylindole ( DAPI; Invitrogen ) . After washing with PBS , cells were mounted on slides and examined with a confocal microscope ( FV1000; Olympus , Düsseldorf , Germany ) . Wild type CHO cells and cells expressing the human Slo3 channel were lysed in a buffer containing 10 mM Tris/HCl , pH 7 . 5 , 140 mM NaCl , 1 mM EDTA , 1% Igepal CA-630 , 0 . 5% sodium deoxycholate , 0 . 1% SDS , and mammalian protease inhibitor cocktail ( mPIC , Sigma ) and incubated on ice for 30 min . The suspension ( total lysate ) was centrifuged for 5 min at 10 , 000×g ( 4°C ) and the supernatant was used for WB analysis . Human sperm ( 5 × 106 ) were resuspended in 2x SDS sample buffer containing β-mercaptoethanol . All samples were heated for 5 min at 95°C and separated by 4–12% SDS–polyacrylamide gel electrophoresis . For WB analysis , proteins were transferred onto PVDF membranes , probed with antibodies , and analysed using the LAS-3000 System ( Fujifilm ) . We recorded from sperm in the whole-cell configuration as described ( Strünker et al . , 2011 ) . Seals between pipette and sperm were formed either at the cytoplasmic droplet or the neck region in standard extracellular solution ( HS ) containing ( in mM ) : 135 NaCl , 5 KCl , 1 MgSO4 , 2 CaCl2 , 5 glucose , 1 Na-pyruvate , 10 lactic acid , and 20 HEPES adjusted to pH 7 . 4 with NaOH . K-based HS contained ( in mM ) : 135 KCl , 5 NaCl , 1 MgSO4 , 2 CaCl2 , 5 glucose , 1 Na-pyruvate , 10 lactic acid , and 20 HEPES adjusted to pH 7 . 4 with KOH . HS with low [Cl−] contained ( in mM ) : 135 Na-aspartate , 5 KCl , 1 MgSO4 , 2 CaCl2 , 5 glucose , 1 Na-pyruvate , 10 lactic acid , and 20 HEPES adjusted to pH 7 . 4 with NaOH . The following pipette ( 10–15 MΩ ) solutions were used ( in mM ) : Cs-based IS: 130 Cs-asparate , 5 EGTA , 5 CsCl , and 50 HEPES at pH 7 . 3 with CsOH; K-based IS 0 Ca2+: 130 K-aspartate , 10 NaCl , 1 EGTA , and 50 HEPES at pH 7 . 3 with KOH; K-based IS with 100 µM or 1 mM Ca2+: 130 K-aspartate , 7 NaCl , 2 KCl , 100 µM or 1 mM CaCl2 , and 50 HEPES at pH 7 . 3 with KOH; K-based IS with 40 µM Ca2+: 140 K-aspartate , 50 HEPES , 10 NaOH , 5 KCl , 3 NTA , 1 . 3 CaCl2 , pH 7 . 3 with KOH; K-based IS with 10 µM Ca2+: 130 K-asparate , 10 NaOH , 1 KCl , 3 HEDTA , 2 CaCl2 , and 50 HEPES at pH 7 . 3 with KOH . All voltages were corrected for the liquid junction potential of 16 . 5 mV . The osmolarity of intracellular and extracellular solutions was ∼120 mOsm . For functional studies in Xenopus oocytes , RNAs for hSlo3 and hLRRC52 were injected at a ratio of 1:1 by weight . Gigaohm seals were formed while oocytes were bathed in frog Ringer ( in mM ) : 115 NaCl , 2 . 5 KCl , 1 . 8 CaCl2 , and 10 HEPES at pH 7 . 4 . Pipette resistance was typically 0 . 8–2 MΩ after fire-polishing and filling with pipette solution containing ( in mM ) : 140 K-methanesulfonate ( MES ) , 20 KOH , 2 MgCl2 , and 10 HEPES at pH 7 . The standard solution bathing excised inside-out patches was ( in mM ) : 140 K-MES , 20 KOH , 5 mM EGTA , and 10 HEPES with pH adjusted as indicated . For addition of Ca2+ to bathing solutions , in most cases free [Ca2+] was sufficiently high that buffering was not required ( 60 µM , 300 µM , 1 mM , and 10 mM ) . In one test solution , we omitted EGTA so that the estimated free [Ca2+] was defined presumably by the level of contaminant Ca2+ in our water ( estimated to be 10–15 µM based on calibration with a Ca2+ electrode ) . Osmolarity for solutions was approximately 310 mOsm . Solutions were applied directly to patches via an SF-77B fast perfusion stepper system ( Warner Instruments , Hamden , CT , USA ) . K+ currents were recorded from inside-out patches with an Axopatch 200B amplifier ( Molecular Devices , Sunnyvale , CA , USA ) and low-pass filtered at 10 kHz with an integral four-pole Bessel filter . Signals were digitized with a Digidata 1322A data acquisition system ( Molecular Devices ) at 100 kHz . Recordings were controlled by the pClamp 9 . 2 software suite ( Molecular Devices ) . Capacitative component of currents was subtracted using the appropriately scaled trace for a step from −100 mV to 0 mV . All experiments were performed at room temperature ( 21–24°C ) . CHO cells were co-transfected with hSlo3 and hLRRC52-mCherry constructs . Seals between pipette and cell were performed in standard extracellular solution ( ES ) containing ( in mM ) : 140 NaCl , 5 . 4 KCl , 1 MgCl2 , 1 . 8 CaCl2 , 10 D-glucose , and 5 HEPES adjusted to pH 7 . 4 with NaOH . For the recordings in high K+ solution , NaCl was exchanged by KCl and pH was adjusted to 7 . 4 with KOH . To study activation of Slo3 by alkalization , the following pipette ( 4–5 MΩ ) solution was used ( in mM ) : 130 K-aspartate , 10 NaCl , 1 EGTA , 5 HEPES , 15 D-glucose , pH was adjusted either to 6 . 2 or 7 . 3 using KOH . To study the activation of Slo3 by Ca2+ , the following pipette ( 2–3 MΩ ) solutions were used ( in mM ) : divalent-free solution: 130 K-aspartate , 10 NaCl , 1 EGTA , and 20 HEPES adjusted to 7 . 3 using KOH; 70 µM Ca2+ intracellular solution: 130 K-aspartate , 10 NaCl , 0 . 5 CaCl2 , 1 NTA , and 20 HEPES adjusted to pH 7 . 3 using KOH , the final Ca2+concentration was confirmed using the Ca2+ dye Mag-Fura 2; 1 mM Ca2+ intracellular solution: 130 K-aspartate , 10 NaCl , 1 CaCl2 , and 20 HEPES adjusted to pH 7 . 3 using KOH . Series resistance and cell capacitance were compensated to 70–85% . Voltages were corrected for the liquid junction potential . All recordings were performed at 20–22°C . Changes in [Ca2+]i were measured in 384 multi-well plates in a fluorescence plate reader ( Fluostar Omega , BMG Labtech , Ortenberg , Germany ) at 30°C . Sperm were loaded with the respective fluorescent Ca2+ indicator ( 10 μM ) ( Molecular Probes ) in the presence of Pluronic F127 ( 0 . 1% wt/vol ) for 45 min at 37°C . After incubation , excess dye was removed by a centrifugation step ( 700×g , 10 min , RT ) . The sperm pellet was resuspended in HTF++ and equilibrated for 5 min at 37°C . Each well was filled with 50 µl of the sperm suspension ( 1 × 107 sperm/ml ) ; the fluorescence was excited at 480 nm and emission was recorded at 520 nm with bottom optics . Fluorescence was recorded before and after injection of 10 µl ( 1:6 dilution ) of progesterone or ionomycin in HTF++ . The solutions were injected into the wells manually with an electronic multichannel pipette . Statistical analysis and fitting of the data were performed using OriginPro 8 . 1 G SR3 ( OriginLab Corporation , USA ) or Clampfit 10 . 2 ( Molecular Devices ) . All data are given as mean ± standard deviation ( number of experiments ) . | A sperm that has been ejaculated into the female reproductive tract must complete a number of tasks to pass on its genes to the next generation . First it must travel along a meandering route to encounter an egg , before pushing through a jelly-like coating that surrounds the egg and then , finally , fusing with the egg’s surface membrane . In order to complete these steps and fertilise the egg , a sperm must undergo a process called ‘capacitation’ . This process , and a variety of other sperm functions , involves the controlled flux of positive ions into and out of the sperm via specific ion channels that are located in the cell membrane . The properties of the ion channels that allow protons and calcium ions to move into and out of human sperm are well understood , but less is known about the channels that control the movement of potassium ions . In mice , a channel called Slo3 allows potassium ions to flow out of the sperm and makes the membrane voltage of these cells more negative . Also , in mice , this channel is essential for the sperm to function correctly , and for fertilization . However , in humans , it is unclear if the Slo3 channel is present in sperm and if it performs the same role . Now , Brenker et al . have shown that the flow of potassium ions out of human sperm occurs via the Slo3 channel , and that human Slo3 is responsible for setting the membrane voltage of these cells . However , whereas the mouse Slo3 channel is opened in response to a decrease in the concentration of protons within the sperm ( i . e . , an increase of the pH inside the cell ) , human Slo3 is largely controlled by changes in the levels of calcium ions . An increase in the calcium concentration within the cell opens the human Slo3 channel , more than a decrease in the proton concentration does . Altogether , Brenker et al . identify Slo3 as the principal potassium channel in human sperm and reveal more fundamental differences between human sperm and mouse sperm . Thereby , this work further stresses the need to be cautious about using mice as a model of male fertility in humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | The Ca2+-activated K+ current of human sperm is mediated by Slo3 |
Collective migration—the directed , coordinated motion of many self-propelled agents—is a fascinating emergent behavior exhibited by active matter with functional implications for biological systems . However , how migration can persist when a population is confronted with perturbations is poorly understood . Here , we address this gap in knowledge through studies of bacteria that migrate via directed motion , or chemotaxis , in response to a self-generated nutrient gradient . We find that bacterial populations autonomously smooth out large-scale perturbations in their overall morphology , enabling the cells to continue to migrate together . This smoothing process arises from spatial variations in the ability of cells to sense and respond to the local nutrient gradient—revealing a population-scale consequence of the manner in which individual cells transduce external signals . Altogether , our work provides insights to predict , and potentially control , the collective migration and morphology of cellular populations and diverse other forms of active matter .
The flocking of birds , schooling of fish , herding of animals , and procession of human crowds are all familiar examples of collective migration . This phenomenon also manifests at smaller scales , such as in populations of cells and dispersions of synthetic self-propelled particles . In addition to being a fascinating example of emergent behavior , collective migration can be critically important—enabling populations to follow cues that would be undetectable to isolated individuals ( Camley , 2018 ) , escape from harmful conditions and colonize new terrain ( Cremer et al . , 2019 ) , and coexist ( Gude et al . , 2020 ) . Thus , diverse studies have sought to understand the mechanisms by which collective migration can arise . Less well understood , however , is how collective migration persists after a population is confronted with perturbations . These can be external , stemming from heterogeneities in the environment ( Sándor et al . , 2017; Morin et al . , 2016; Wong et al . , 2014; Chepizhko and Peruani , 2013; Chepizhko et al . , 2013; Chepizhko and Peruani , 2015; Toner et al . , 2018; Maitra , 2020 ) , or internal , stemming from differences in the behavior of individuals ( Yllanes et al . , 2017; Bera and Sood , 2020; Alirezaeizanjani et al . , 2020 ) . Mechanisms by which such perturbations can disrupt collective migration are well documented . Indeed , in some cases , perturbations can abolish coordinated motion throughout the population entirely ( Sándor et al . , 2017; Morin et al . , 2016; Yllanes et al . , 2017; Bera and Sood , 2020; Chepizhko and Peruani , 2013; Chepizhko et al . , 2013; Chepizhko and Peruani , 2015; Toner et al . , 2018 ) . In other cases , perturbations couple to the active motion of the population to destabilize its leading edge , producing large-scale disruptions to its morphology ( Wong et al . , 2014; Alert and Trepat , 2020; Alert et al . , 2019; Driscoll et al . , 2016; Doostmohammadi et al . , 2016; Williamson and Salbreux , 2018; Miles et al . , 2019 ) . Indeed , for one of the simplest cases of collective migration—via chemotaxis , the biased motion of cells up a chemical gradient—morphological instabilities can occur due to the disruptive influence of hydrodynamic ( Subramanian et al . , 2011; Lushi et al . , 2012; Lushi et al . , 2018 ) or chemical-mediated ( Ben Amar and Bianca , 2016; Ben Amar , 2016; Funaki et al . , 2006; Brenner et al . , 1998; Mimura and Tsujikawa , 1996; Stark , 2018 ) interactions between cells . By contrast , mechanisms by which migrating populations can withstand perturbations have scarcely been examined . Here , we demonstrate a mechanism by which collectively migrating populations of Escherichia coli autonomously smooth out large-scale perturbations in their overall morphology . We show that chemotaxis in response to a self-generated nutrient gradient provides both the driving force for collective migration and the primary smoothing mechanism for these bacterial populations . Using experiments on 3D-printed populations with defined morphologies , we characterize the dependence of this active smoothing on the wavelength of the perturbation and on the ability of cells to migrate . Furthermore , using continuum simulations , we show that the limited ability of cells to sense and respond to a nutrient gradient causes them to migrate at different velocities at different positions along a front—ultimately driving smoothing of the overall population and enabling continued collective migration . Our work thus reveals how cellular signal transduction enables a population to withstand large-scale perturbations and provides a framework to predict and control chemotactic smoothing for active matter in general .
To experimentally investigate the collective migration of E . coli populations , we confine them within porous media of tunable properties ( Bhattacharjee and Datta , 2019a; Bhattacharjee and Datta , 2019b; Bhattacharjee et al . , 2021 ) , as schematized in Figure 1A and B and detailed in Materials and methods . The media are composed of hydrogel particles that are swollen in a defined rich liquid medium with L-serine as the primary nutrient and chemoattractant . We enclose the particles at prescribed jammed packing fractions in transparent chambers . Because the hydrogel is highly swollen , it is freely permeable to oxygen and nutrient . However , while the particles do not hinder exposure of bacteria to these chemical signals , the cells cannot penetrate the individual particles and are instead forced to swim through the interparticle pores ( Figure 1B ) . Varying the hydrogel particle packing density thus enables us to tune pore size and thereby modulate cellular migration without altering the nutrient field ( Bhattacharjee and Datta , 2019a; Bhattacharjee and Datta , 2019b; Bhattacharjee et al . , 2021 ) . Specifically , we vary the mean pore size ξ between 1 . 2 µm and 2 . 2 µm , causing cellular migration through the pore space to be more and less hindered , respectively , without deforming the solid matrix ( Bhattacharjee and Datta , 2019a ) . Moreover , the packings are transparent , enabling the morphologies of the migrating populations to be tracked in the xy plane using confocal fluorescence microscopy ( Figure 1A ) ; to this end , we use cells that constitutively express green fluorescent protein throughout their cytoplasm . A key feature of the hydrogel packings is that they are yield-stress solids; thus , an injection micronozzle can move along a prescribed path inside each medium by locally rearranging the particles , gently extruding densely packed cells into the interstitial space ( Figure 1A and B ) . The particles then rapidly re-densify around the newly introduced cells , re-forming a jammed solid matrix that supports the cells in place with minimal alteration to the overall pore structure ( Bhattacharjee et al . , 2015; Bhattacharjee et al . , 2016; Bhattacharjee et al . , 2018 ) . This approach is therefore a form of 3D printing that enables the initial morphology of each bacterial population to be defined within the porous medium . The cells subsequently swim through the pores between particles , migrating outward through the pore space . For example , as we showed previously ( Bhattacharjee et al . , 2021 ) , cells of E . coli initially 3D-printed in densely packed straight cylinders collectively migrate radially outward in smooth ( ‘flat’ ) , coherent fronts . These fronts form and propagate via chemotaxis: the cells continually consume surrounding nutrient , generating a local gradient that they in turn bias their motion along ( Adler , 1966; Cremer et al . , 2019; Fu et al . , 2018; Saragosti et al . , 2011; Bai et al . , 2021 ) . As each front of cells migrates , it propagates the local nutrient gradient with it through continued consumption , thereby sustaining collective migration . In the absence of nutrient , migrating fronts do not form at all ( Bhattacharjee et al . , 2021 ) . To test how perturbations in the overall morphology of the population influence its subsequent migration , we 3D-print densely packed E . coli in 1-cm-long cylinders with spatially periodic undulations as perturbations prescribed along the x direction ( Figure 1B ) . Each population is embedded deep within a defined porous medium; an initial population morphology is schematized at time t=0 in Figure 1B , with the undulation wavelength and amplitude denoted by λ and A , respectively . An experimental realization with A ( t=0 ) ≈300 µm , λ≈0 . 8 mm , and ξ=1 . 7 µm is shown in white in Figure 1C , which shows an xy cross section through the midplane of the population . After 3D printing , the outer periphery of the population spreads slowly , hindered by cell-cell collisions in the pore space , as the population establishes a steep gradient of nutrient through consumption ( Bhattacharjee et al . , 2021 ) . Then , this periphery spontaneously organizes into an ∼300 μm-wide front of cells that collectively migrates outward ( yellow in Figure 1C ) . The undulated morphology of this front initially retains that of the initial population . Strikingly , however , the front autonomously smooths out these large-scale undulations as it continues to propagate ( Video 1 ) . We characterize this behavior by tracking the decay of the undulation amplitude , normalized by its initial value A0≡A ( Δt=0 ) , as a function of time elapsed from the initiation of smoothing , Δt ( green circles in Figure 1F ) . The normalized amplitude decays exponentially ( red line in Figure 1F ) , with a characteristic time scale τ≈2 . 5 hr , and the population eventually continues to migrate as a completely flat front ( cyan in Figure 1C ) . We observe similar behavior when the wavelength λ is increased to 3 . 4 mm ( Figure 1D , Video 2 ) or when the pore size ξ is increased to 2 . 2 μm ( Figure 1E , Video 3 ) ; however , the dynamics of front smoothing are altered in both cases . Specifically , increasing the undulation wavelength slows smoothing , increasing τ by a factor of ≈3 to reach τ≈6 . 5 hr ( green squares in Figure 1F ) . Conversely , increasing the pore size—which enables cells to migrate through the pore space more easily—greatly hastens smoothing , decreasing τ by more than a factor of ≈10 to become τ≈0 . 2 hr ( blue circles in Figure 1F ) . This behavior is consistent across multiple experiments with varying λ and ξ , as summarized in Figure 1G . Our experiments thus indicate that the smoothing of collective migration is regulated by both the undulation wavelength and the ease with which cells migrate . To gain further insight into the processes underlying smoothing , we use the classic Keller–Segel model of chemotactic migration ( Lauffenburger , 1991; Keller and Segel , 1971 ) to investigate the dynamics of undulated populations . Variants of this model can successfully capture the key features of chemotactic migration of flat E . coli fronts in bulk liquid ( Keller and Segel , 1971; Fu et al . , 2018 ) and in porous media ( Bhattacharjee et al . , 2021 ) ; we therefore hypothesize that it can also help identify the essential physics of smoothing . To this end , we consider a 2D representation of the population in the xy plane for simplicity , with r→≡ ( x , y ) , and model the evolution of the nutrient concentration c ( r→ , t ) and number density of bacteria b ( r→ , t ) using the coupled equations: ( 1 ) ∂tc=Dc∇2c-bκg ( c ) , ( 2 ) ∂tb=−∇⋅J→b+bγg ( c ) , J→b=−Db∇b+bχ∇f ( c ) . Equation 1 relates changes in c to nutrient diffusion and consumption by the bacteria; Dc is the nutrient diffusion coefficient , κ is the maximal consumption rate per cell , and g ( c ) =c/ ( c+c1/2 ) describes the influence of nutrient availability relative to the characteristic concentration c1/2 through Michaelis–Menten kinetics . Equation 2 relates changes in b to the bacterial flux J→b , which arises from their undirected and directed motion , and net cell proliferation with a maximal rate γ . In the absence of a nutrient gradient , bacteria move in an unbiased random walk ( Berg , 2004 ) ; thus , undirected motion is diffusive over large length and time scales , with an effective diffusion coefficient Db whose value depends on both cellular activity and confinement in the pore space , and is therefore b-dependent as detailed in Materials and methods ( Bhattacharjee and Datta , 2019a; Bhattacharjee and Datta , 2019b ) . In the presence of the local nutrient gradient established through consumption , bacteria perform chemotaxis , biasing this random walk ( Berg , 2004 ) ; the function f ( c ) ≡log[ ( 1+c/c− ) / ( 1+c/c+ ) ] describes the ability of the bacteria to logarithmically sense nutrient with characteristic concentrations c− and c+ ( Cremer et al . , 2019; Fu et al . , 2018 ) , and the chemotactic coefficient χ describes their ability to then bias their motion in response to the sensed nutrient gradient ( Keller and Segel , 1971; Fu et al . , 2018; Cremer et al . , 2019 ) . The chemotactic velocity is thus given by v→ch≡χ∇f ( c ) , where similar to Db , the value of χ depends on both intrinsic cellular properties and pore-scale confinement , and is also b-dependent as detailed in Materials and methods ( Bhattacharjee et al . , 2021 ) . Together , Equations 1 and 2 provide a continuum model of chemotactic migration that has thus far been successfully used to describe unperturbed E . coli populations ( Keller and Segel , 1971; Fu et al . , 2018; Cremer et al . , 2019; Bhattacharjee et al . , 2021 ) . We note that a recently introduced growth-expansion model of chemotactic migration , for which analytical expressions describing chemotactic fronts have been obtained ( Cremer et al . , 2019; Narla et al . , 2021 ) , can be thought of as a limit of our model with bacterial growth taken to be independent of the attractant . An interesting direction for future work would be to study the phenomenon of chemotactic smoothing revealed here in the growth-expansion model , similar to a recent analytical study of small-amplitude perturbations in a simplified version of the model considered here ( Alert and Datta , 2021 ) . Here , to simulate the chemotactic migration of populations with large-amplitude perturbations , we numerically solve Equations 1 and 2 using undulated morphologies as initial conditions for b , similar to those explored in the experiments . The simulations employ values for all parameters based on direct measurements , as detailed in Materials and methods . Although we do not expect perfect quantitative agreement between the experiments and simulations due to their difference in dimensionality and the simplified treatment of cell-cell interactions , the simulated fronts form , collectively migrate , and smooth in a manner that is remarkably similar to the experiments . Three examples are shown in Figure 2C–E ( Videos 4–6 ) , corresponding to the experiments shown in Figure 1C–E ( Videos 1–3 ) . Similar to the experiments , the outer periphery of each population first spreads slowly , then spontaneously organizes into an outward-migrating front that eventually smooths . We again find that the normalized undulation amplitude decays exponentially over time , as shown in Figure 2D . As in the experiments , increasing the undulation wavelength λ slows smoothing; compare Figure 2B to Figure 2A . Also as in the experiments , increasing the pore size ξ , which increases the migration parameters Db and χ , greatly hastens smoothing; compare Figure 2C to Figure 2A . This variation of the smoothing time scale τ obtained from simulations with λ and ξ is summarized in Figure 2E . We observe the same behavior as in the experiments , with the absolute values of τ agreeing to within a factor of ∼3 . This agreement confirms that the continuum Keller–Segel model recapitulates the essential spatio-temporal features of smoothing seen in the experiments . The simulations provide a way to directly assess the relative importance of cellular diffusion , chemotaxis , and cell proliferation to front smoothing . To this end , we perform the same simulation as in Figure 2A , but with each of the corresponding three terms in Equation 2 knocked out , and determine the resulting impact on collective migration . This procedure enables us to determine the factors necessary for smoothing . While diffusion typically causes spatial inhomogeneities to smooth out , we do not expect it to play an appreciable role in the front smoothing observed here: the characteristic time scale over which undulations of wavelength λ≈1 mm diffusively smooth is ∼λ2/Db≈100 to 700 hr , up to three orders of magnitude larger than the smoothing time τ measured in experiments and simulations . We therefore expect that the undirected motion of bacteria is much too slow to contribute to front smoothing . The simulations for λ=0 . 8 mm and ξ=1 . 7 μm confirm this expectation: setting Db=0 yields fronts that still smooth over a time scale τ∼1 hr similar to the full simulations ( Figure 3A ) . Another possible mechanism of front smoothing is differences in bacterial proliferation at different locations along the front periphery—for example , the front would smooth if cells in concave regions were able to proliferate faster than those in convex regions . However , differential proliferation typically destabilizes bacterial communities , as shown previously both experimentally and theoretically ( Fujikawa and Matsushita , 1989; Bonachela et al . , 2011; Nadell et al . , 2010; Farrell et al . , 2013; Trinschek et al . , 2018; Allen and Waclaw , 2019 ) . Furthermore , even if proliferation were to help smooth the overall population , we again expect this hypothetical mechanism to be too slow to appreciably contribute: the shortest time scale over which cells all growing exponentially at a maximal rate γ∼1 hr-1 spread over the length scale A0≈300 μm by growing end-to-end is γ-1log2 ( A0/lcell ) ∼7 hr , where lcell≈2 μm is the cell body length . This time scale is over an order of magnitude larger than the τ measured in experiments and simulations . The simulations again confirm our expectation: setting γ=0 yields fronts that still smooth over a time scale τ∼1 hr similar to the full simulations ( Figure 3B ) . These findings leave chemotaxis as the remaining possible mechanism of front smoothing . The simulations confirm this expectation: setting χ=0 yields a population that slowly spreads via diffusion and proliferation , but that does not form collectively migrating fronts at all ( Figure 3C ) . Therefore , chemotaxis is both necessary and sufficient for the observed front smoothing . How exactly does chemotaxis smooth bacterial fronts ? To address this question , we examine the spatially varying chemotactic velocity v→c=χ∇f ( c ) , which quantifies how rapidly different regions of the population migrate via chemotaxis . To gain intuition for the determinants of v→c , we recast this expression in terms of the nutrient gradient: ( 3 ) v→c=χf ′ ( c ) ⏟Response function ∇c⏟Forcing . As in linear response theory , the chemotactic velocity can be viewed as the bacterial response to the driving force given by the nutrient gradient , ∇c , modulated by the chemotactic response function χf ′ ( c ) . Thus , variations in chemotactic velocity along the leading edge of the front , which specify how the overall front morphology evolves , are determined by the combined effect of variations in the nutrient gradient and the chemotactic response function . We therefore examine each of these modes by which chemotaxis influences front morphology in turn . We first consider the nutrient gradient , which is the typical focus of chemotaxis studies . Our simulations , which numerically solve the coupled system of Equations 1 and 2 , directly yield the spatially varying nutrient field c and therefore ∇c . A snapshot from the representative example of Figure 2A is shown in Figure 4A , with the contours of c=c− and c=c+ indicated by the cyan and magenta lines , respectively . The contours are spaced closer at the convex ‘peaks’ ( e . g . , at y/λ=0 . 5 ) than at the concave ‘valleys’ ( e . g . , at y/λ=0 ) along the leading edge of the front . Thus , the magnitude of the driving force given by ∇c is larger at the peaks . We confirm this expectation by directly quantifying the nutrient gradient along the leading edge , focusing on the component ∂xc in the overall front propagation direction ( x ) for simplicity , as shown by the orange symbols in Figure 4C; as expected , this driving force is stronger at the peaks . This spatial variation in the driving force promotes faster outward chemotactic migration at the peaks than at the valleys , amplifying front undulations—in opposition to our observation that the migrating population self-smooths . Variations in the local nutrient gradient along the leading edge of the front do not contribute to smoothing; rather , they oppose it . We next turn to the chemotactic response function , which characterizes cellular signal transduction . Because χ is a constant for each porous medium ( Bhattacharjee et al . , 2021 ) , spatial variations in the response function are set by variations in f ′ ( c ) . The sensing function f ( c ) is plotted in the upper panel of Figure 4B . It varies linearly as ∼c ( 1/c--1/c+ ) for c≪c- and saturates at log ( c+/c- ) for c≫c+; the characteristic concentrations c− and c+ represent the dissociation constants of the nutrient for the inactive and active conformations of the cell-surface receptors , respectively ( Cremer et al . , 2019; Fu et al . , 2018; Dufour et al . , 2014; Yang et al . , 2015 ) . The response function χf ′ ( c ) therefore decreases strongly as c increases above c+ , which accordingly is often referred to as an upper limit of sensing ( Figure 4B , lower panel ) . That is , because high nutrient concentrations saturate cell-surface receptors , the chemotactic response function decreases with nutrient concentration . Inspection of the nutrient field indicates that nutrient concentrations are larger at the peaks than at the valleys along the leading edge of the front ( Figure 4A ) . Thus , the chemotactic response of cells is weaker at peaks than at valleys , as shown by the points in Figure 4B , yielding slower outward chemotactic migration at peaks than at valleys and thereby reducing the amplitude of front undulations . Variations in the chemotactic response along the leading edge of the front promote smoothing , unlike variations in the nutrient gradient . We therefore hypothesize that the stabilizing effect of the chemotactic response ( Figure 4C , blue ) dominates over the destabilizing influence of the nutrient gradient ( Figure 4C , red ) , leading to smoothing . Computation of the spatially varying chemotactic velocity at the leading edge of the front using Equation 3 , focusing on the x velocity component vc , x≈χf ′∂xc for simplicity , supports this hypothesis: cells at concave regions migrate outward faster than those at convex regions ( Figure 4C , lower panel ) . To further test this hypothesis , we assess the influence of varying c+; we expect that reducing this upper limit weakens chemotactic response not just at the peaks , but also the valleys , thereby slowing smoothing . While tuning solely c+ is challenging in the experiments , this can be readily done in the simulation—yielding slower smoothing , as expected ( Figure 4—figure supplement 1 ) . As a final test of our hypothesis , for each simulation shown in Figure 2 , we determine the difference between the chemotactic velocities of the valleys and peaks , approximated by Δvc , x≈χ[ ( f ′∂xc ) valley− ( f ′∂xc ) peak] , as a function of time Δt . If smoothing is indeed due to variations of the chemotactic velocity along the leading edge , then the smoothing time τ determined by analyzing the decay of the undulation amplitude , A=A0e-Δt/τ ( Figure 2D and E ) , should be similar to the time τ′ at which valleys catch up to peaks , that is , ∫0τ′Δvc , xdΔt≈A0 . To test this expectation , we plot the τ′ values thus obtained for all of our simulations of varying λ and ξ as a function of the corresponding τ , as shown in Figure 4D . We find that τ′ and τ are indeed similar to one another in all cases—confirming that smoothing is determined by spatial variations in chemotactic velocity .
While our study utilizes a specific form of the sensing function f ( c ) established for E . coli ( Cremer et al . , 2019; Fu et al . , 2018 ) , the phenomenon of chemotactic smoothing can manifest more generally . Specifically , our description of smoothing requires that ( i ) convex regions of a population are exposed to more nutrient c than concave regions , and ( ii ) f ( c ) is monotonically increasing and concave , with f ′′ ( c ) <0; when these conditions are satisfied , the chemotactic response is weaker at convex regions than at concave ones , thereby promoting smoothing ( as indicated in Figure 4B ) . The first requirement is frequently satisfied for collective migration in general; for example , in chemotactic migration , nutrient concentration c decreases from the outward boundaries into the population over a length scale given by the interplay between nutrient diffusion and consumption . This first requirement is also satisfied by many other forms of active matter that rely on other modes of sensing to collectively migrate , for which c would generically represent the stimulus being sensed . Documented examples include durotactic cell groups ( Roca-Cusachs et al . , 2013; Sunyer et al . , 2016; Alert and Casademunt , 2019 ) , phoretic active colloids ( Illien et al . , 2017; Liebchen and Löwen , 2018Stark , 2018 ) , and phototactic robots ( Mijalkov et al . , 2016; Palagi and Fischer , 2018 ) —systems for which migration is directed toward regions of larger c , and therefore convex regions are more likely to be exposed to larger c . The second requirement is also satisfied for diverse active matter systems; in the context of chemotaxis , specific examples include other bacteria ( Menolascina et al . , 2017 ) , enzymes ( Jee et al . , 2018; Agudo-Canalejo et al . , 2018; Mohajerani et al . , 2018 ) , aggregating amoeba cells ( Keller and Segel , 1970 ) , and mammalian cell groups during development , immune response , and disease ( Camley , 2018; Iglesias and Devreotes , 2008; Theveneau et al . , 2010; McLennan et al . , 2012; Malet-Engra et al . , 2015; Puliafito et al . , 2015; Tweedy et al . , 2020 ) . This second requirement is again also satisfied for active matter that collectively migrates using other sensing mechanisms , for which sensing has been documented to increase and eventually saturate with the stimulus , be it the stiffness of the underlying surface ( Roca-Cusachs et al . , 2013; Sunyer et al . , 2016; Alert and Casademunt , 2019 ) , temperature ( Illien et al . , 2017; Liebchen and Löwen , 2018 ) , or light intensity ( Mijalkov et al . , 2016; Palagi and Fischer , 2018 ) . Thus , exploring the physics described here in diverse other forms of active matter will be a useful direction for future work . As a final illustration of the necessity of the sensing function f ( c ) to be concave , f ′′ ( c ) <0 , we repeat our analysis but instead consider a strictly linear f ( c ) =c/clin , which does not saturate . We choose clin= ( 1/c--1/c+ ) -1 so that the linear f ( c ) matches our original logarithmic f ( c ) at small c . With this linear sensing function , the chemotactic response is independent of concentration , f ′ ( c ) =1/clin , and the condition of concavity is violated: f ′′ ( c ) =0 . We therefore expect chemotactic smoothing to not occur . Consistent with our expectation , repeating the analysis underlying Figure 4C but for the strictly linear f ( c ) yields fronts for which valleys no longer move faster than peaks . Instead , as shown in Figure 4—figure supplement 2 , the profile of chemotactic velocity is now inverted with respect to that of the bottom panel in Figure 4C . Hence , the front does not smooth . Overall , this sample computation illustrates a way of modifying f ( c ) that abrogates sensing saturation and hence would prevent chemotactic smoothing . The chemotactic smoothing process described here is autonomous , arising without any external intervention . Instead , it is a population-scale consequence of the limitations in cellular signal transduction—motivating future studies of other population-scale effects , beyond smoothing , that may emerge from individual behaviors . Indeed , while studies of chemotaxis typically focus on the role of the external nutrient gradient in driving cellular migration , our work highlights the distinct and pivotal role played by the cellular chemotactic response function in regulating migration and large-scale population morphology more broadly . Our work therefore contributes a new factor to be considered in descriptions of morphogenesis , which thus far have focused on the role of other factors—such as differential forcing by signaling gradients , differential proliferation , intercellular mechanics , substrate interactions , and osmotic stresses ( McLennan et al . , 2012; Fujikawa and Matsushita , 1989; Bonachela et al . , 2011; Nadell et al . , 2010; Farrell et al . , 2013; Trinschek et al . , 2018; Allen and Waclaw , 2019; Beroz et al . , 2018; Fei et al . , 2020; Yan et al . , 2019; Yan et al . , 2017; Copenhagen et al . , 2020; Smith et al . , 2017; Ghosh et al . , 2015; Zhang et al . , 2021 ) —in regulating the overall morphology of cellular communities and active matter in general .
We prepared 3D porous media by dispersing dry granules of crosslinked acrylic acid/alkyl acrylate copolymers ( Carbomer 980 , Ashland ) in liquid EZ Rich , a defined rich medium for E . coli . The components to prepare the EZ Rich were purchased from Teknova Inc , autoclaved prior to use , and were mixed following the manufacturer’s directions; specifically , the liquid medium was an aqueous solution of 10× MOPS Mixture ( M2101 ) , 10× ACGU solution ( M2103 ) , 5× Supplement EZ solution ( M2104 ) , 20% glucose solution ( G0520 ) , 0 . 132 M potassium phosphate dibasic solution ( M2102 ) , and ultrapure Milli-Q water at volume fractions of 10 , 10 , 20 , 1 , 1 , and 58% , respectively . We ensured homogeneous dispersions of swollen hydrogel particles by mixing each dispersion for at least 2 hr at 1600 rpm using magnetic stirring and adjusted the pH to 7 . 4 by adding 10 N NaOH to ensure optimal cell viability . The hydrogel granules swell considerably , resulting in a jammed medium made of ∼5–10 μm diameter swollen hydrogel particles with ∼20% polydispersity and with an individual mesh size of ∼40–100 nm , as we established previously ( Bhattacharjee and Datta , 2019b ) , which enables small molecules ( e . g . , amino acids , glucose , oxygen ) to freely diffuse throughout the medium . Tuning the mass fraction of dispersed hydrogel particles enables the sizes of the pores between particles to be precisely tuned . We measured the smallest local pore dimension by tracking the diffusion of 200-nm-diameter fluorescent tracers through the pore space , as we detailed in a previous paper ( Bhattacharjee et al . , 2021 ) . This previous paper shows the full pore size distributions thereby measured for porous media prepared in an identical manner to those used here; in this paper , we only describe each medium using the mean pore size ξ , for simplicity . Indeed , the measured pore size distributions exhibit exponential decays characterized by the mean value ξ , as reported in Bhattacharjee et al . , 2021 , with pore sizes between 1 and 8 μm in the loosest packings and pores smaller than 4 μm in the tightest packings . Prior to each experiment , we prepared an overnight culture of E . coli W3110 in LB media at 30°C . We then incubated a 1% solution of this culture in fresh LB media for 3 hr until the optical density reached ∼0 . 6 , and then resuspended the cells in liquid EZ Rich to a concentration of 8 . 6×1010 cells/mL . We then used this suspension as the inoculum that was 3D printed into a porous medium using a pulled glass capillary with a ∼100–200 μm-wide opening as an injection nozzle . Each porous medium had a large volume of 4 mL and was confined in a transparent-walled glass-bottom Petri dish 35 mm in diameter and 10 mm in height; in each experiment , the injection nozzle was mounted on a motorized translation stage that traces out a programmed two-dimensional undulating path within the porous medium , at least ∼500–1000 m away from any boundaries , at a constant speed of 1 mm/s . As the injection nozzle moved through the medium , it locally rearranged the hydrogel packing and gently extruded the cell suspension into the interstitial space using a flow-controlled syringe pump at 50 μL/hr , which corresponds to a gentle shear rate of ∼4–36 s-1 at the tip of the injection nozzle . As the nozzle continued to move , the surrounding hydrogel particles rapidly densified around the newly introduced cells , re-forming a jammed solid matrix ( Bhattacharjee et al . , 2018; Bhattacharjee et al . , 2015; Bhattacharjee et al . , 2016 ) that compressed the cellular suspension until the cells are closely packed to an approximate density of 0 . 95×1012 cells/mL . This protocol thus results in a 3D-printed bacterial population having a defined initial amplitude and wavelength . Moreover , as we showed in our previous work Bhattacharjee et al . , 2021 , this process does not appreciably alter the properties of the hydrogel packing and is sufficiently gentle to maintain the viability and motility of the cells . Because the 3D-printed undulated cylinders of densely packed cells are ∼1 cm long , each printing process requires ∼10 s . After 3D printing , the top surface of the porous medium was sealed with a thin layer of 1–2 mL of paraffin oil to minimize evaporation while allowing unimpeded oxygen diffusion . We then commenced imaging within a few minutes after printing . Once an undulated population is 3D printed , it maintains its shape until cells start to move outward through the pore space . The time needed to print each cylinder is two orders of magnitude shorter than the duration between successive 3D confocal image stacks . Moreover , the 3D printing is fast enough to be considered as instantaneous when compared with the speed of bacterial migration . Thus , the imaging is sufficiently fast to capture the front propagation dynamics . To image how the distribution of cells evolves over time , we used a Nikon A1R + inverted laser-scanning confocal microscope maintained at 30 ± 1°C . In each experiment , we acquired vertical stacks of planar fluorescence images separated by 2 . 58 μm along the vertical ( z ) direction , successively every 2–30 min for up to 20 hr . We then produced a maximum intensity projection from each stack at every time frame with the logarithm of fluorescent intensities displayed at every pixel; examples are shown in Figure 1 . Our prior work used high-resolution visualization to obtain magnified views of the bacterial concentration fields at long times for unperturbed flat fronts and verified that the cells are swimming through the pore space as a suspension ( Bhattacharjee et al . , 2021 ) . Here , we instead use lower-resolution visualization to characterize population-scale front dynamics over larger length scales . We note that because our experiments probe fluorescence from GFP-expressing cells , the confocal images only show the actively moving cells near the leading edge of each propagating front because it is exposed to sufficient oxygen for the GFP to properly fold . As these cells move outward , they continually consume nutrient and oxygen—eventually causing the trailing ‘inner’ region of the population to become oxygen-depleted , as shown in our previous work ( Bhattacharjee et al . , 2021 ) . Under these conditions , we conjecture that the GFP expressed by the cells does not properly fold , and the cells lose fluorescence over ∼30 min . Thus , even though some cells remain localized within the inner region , they turn dark and hence seem to disappear from the microscope fluorescence images . We used each maximum intensity projection at each time point to manually measure the time-dependent amplitude ( A ) and radial location of the front ( Rf ) as defined in Figure 1B , identifying the edges of the front as the positions at which the fluorescent signal from cells matches the background noise . As we showed in our previous work ( Bhattacharjee et al . , 2021 ) , due to the initially high cell density in the population , inter-cell collisions limit outward migration of the population; a coherent outward-propagating front only forms after at least ∼1 hr . Here , we do not focus on these initial transient dynamics , but instead examine the long-time smoothing behavior of undulated fronts . We did this by tracking the decay of the time-dependent undulation amplitude over time , as shown in Figure 1F; we identified the time t0 at which smoothing is initiated as the earliest time at which the error associated with an exponential fit to the decay of A ( t ) is minimized . The initial value A0 is then given by A ( t0 ) . To mathematically model the dynamics of bacterial fronts , we use a continuum description of chemotactic migration that we previously showed captures the essential dynamical features of flat fronts ( Bhattacharjee et al . , 2021 ) . This model extends previous work on the classic Keller–Segel model ( Fu et al . , 2018; Saragosti et al . , 2011; Cremer et al . , 2019; Croze et al . , 2011; Keller and Odell , 1975; Keller and Segel , 1971; Keller and Segel , 1970; Odell and Keller , 1976; Lauffenburger , 1991; Seyrich et al . , 2019 ) to the case of dense populations in porous media . In particular , we consider a 2D representation of the population in the xy plane for simplicity and describe the evolution of the nutrient concentration c ( r→ , t ) and number density of bacteria b ( r→ , t ) using the coupled Equations 1 and 2 . While the experimental geometry is three dimensional , in previous work ( Bhattacharjee et al . , 2021 ) , we found that radial and out-of-plane effects do not need to be considered to capture the essential features of bacterial front formation and migration . Thus , for simplicity , we use a 2D representation . In the x direction ( coordinates defined in Figures 2 and 4 ) , no flux boundary conditions are used at the walls of the simulated region for both field variables b and c . In the y direction , no flux boundary conditions are used after one wavelength of the undulation , peak to peak , which comprises a single repeatable unit . The initial cylindrical distribution of cells 3D printed in the experiments has a diameter of ∼100 μm; so , in the x dimension of the numerical simulations , we use a Gaussian with a 100 μm full width at half maximum for the initial bacteria distribution b ( x , t=0 ) , with a peak value that matches the 3D-printed cell density in the experiments , 0 . 95×1012 cells/mL . We vary the center x position of the Gaussian distribution sinusoidally along y to reproduce a given experimental wavelength and amplitude . Experimental wavelengths were measured directly from confocal images and rounded to the nearest 10 μm . The initial condition of nutrient is c=10 mM everywhere , characteristic of the liquid media used in the experiments . The initial nutrient concentration is likely lower within the experimental population initially due to nutrient consumption during the 3D printing process; however , we expect this discrepancy to play a negligible role as nutrient deprivation occurs rapidly in the simulations . As previously detailed ( Bhattacharjee et al . , 2021 ) , while the periphery of a 3D-printed bacterial population forms a propagating front , cells in the inner region remain fixed and eventually lose fluorescence because they are oxygen-limited . Specifically , the fluorescence intensity of this fixed inner population remains constant over an initial duration τdelay=2 hr , and then exponentially decreases with a decay time scale τstarve=29 . 7 min . To facilitate comparison to the experiments , our simulations incorporate this feature to represent the cellular signal , which is the analog of the fluorescence measured in experiments , in Figures 2 and 4 . We do this by multiplying the cellular density obtained by solving Equation 2 by a correction factor that incorporates the history of oxygen depletion . Specifically , wherever c ( r→ ′ , t′ ) drops below a threshold value , for all times t>t′+τdelay , we multiply the cellular density b ( r→ ′ , t ) by e- ( t-t′ ) /τstarve , where t′ is the time at which the position r→ ′ became nutrient-depleted; oxygen and nutrient depletion occur at similar positions and times as detailed in Bhattacharjee et al . , 2021 . To numerically solve the continuum model , we use an Adams–Bashforth–Moulton predictor corrector method ( Seyrich et al . , 2019 ) , where the order of the predictor and corrector are 3 and 2 , respectively . Since the predictor corrector method requires past time points to inform future steps , the starting time points must be found with another method; we choose the Shanks starter of order 6 ( Shanks , 1966 ) . For the first and second derivatives in space , we use finite difference equations with central difference forms in 2D . Time steps of the simulations are 0 . 01 s and spatial resolution is 10 μm . Because the experimental chambers are 3 . 5 cm in diameter , we use a distance of 3 . 5 × 104 μm for the size of the entire simulated system in the x direction with the cells initially situated in the center . Our previous work ( Bhattacharjee et al . , 2021 ) demonstrated that the choice of discretization does not appreciably influence the results in numerical simulations of flat fronts; furthermore , our new results for the simulations performed here ( Figure 4—figure supplement 3 ) indicate that our choice of discretization used is sufficiently finely resolved such that the results in numerical simulations of undulated fronts are not appreciably influenced by discretization . For the analysis shown in Figure 2 , the leading edge is defined as the locus of positions at which b falls below a threshold value equal to 10–4 times the maximum cell density of the initial bacterial distribution , as in Bhattacharjee et al . , 2021 . For the analysis shown in Figure 4 , to more accurately track the leading edge of the front , we define it as the locus of positions at which b falls below a threshold value specific to each condition tested; the threshold is 0 . 003 cells per μm3 for the prototypical case of ξ=1 . 7 μm and λ=0 . 8 mm shown in Figure 4A–C , as well as all simulations for ξ=2 . 2 μm; 0 . 002 cells per μm3 for simulations for ξ=1 . 7 μm and λ=2 . 0 and 3 . 2 mm; and 0 . 001 cells per μm3 for simulations for ξ=1 . 2 μm and λ=0 . 8 mm . We note that the b-dependence of the motility parameters Db and χ does not play an appreciable role in our analysis of smoothing since the definition used for the leading edge of each front is at a fixed , low value of b . One may speculate that smoothing could be avoided or even reversed by lowering the initial nutrient concentration to a value in between c+ and c− , thereby diminishing the difference in chemotactic response between peaks and valleys and allowing the amplifying effects of the nutrient gradient to dominate . However , a simulation performed with a much lower initial nutrient concentration of 10 μM throughout , chosen to be in between c+ and c− , does not even form a traveling front at all over the experimental time scale ( Video 7 ) . This absence of a front is due to the reduction in nutrient consumption as modulated by the Monod function g ( c ) , which results in a drastic reduction in the nutrient gradient that drives front formation and propagation . Thus , despite varying the initial nutrient concentration over three orders of magnitude , the upper limit c+ over an order of magnitude , and the migration parameters Db and χ over an order of magnitude , we have not found conditions under which chemotactic fronts , if they form , do not smooth . Smoothing therefore appears to be robust to large changes in the environmental conditions . | Flocks of birds , schools of fish and herds of animals are all good examples of collective migration , where individuals co-ordinate their behavior to improve survival . This process also happens on a cellular level; for example , when bacteria consume a nutrient in their surroundings , they will collectively move to an area with a higher concentration of food via a process known as chemotaxis . Several studies have examined how disturbing collective migration can cause populations to fall apart . However , little is known about how groups withstand these interferences . To investigate , Bhattacharjee , Amchin , Alert et al . studied bacteria called Escherichia coli as they moved through a gel towards nutrients . The E . coli were injected into the gel using a three-dimensional printer , which deposited the bacteria into a wiggly shape that forces the cells apart , making it harder for them to move as a collective group . However , as the bacteria migrated through the gel , they smoothed out the line and gradually made it straighter so they could continue to travel together over longer distances . Computer simulations revealed that this smoothing process is achieved by differences in how the cells respond to local nutrient levels based on their position . Bacteria towards the front of the group are exposed to more nutrients , causing them to become oversaturated and respond less effectively to the nutrient gradient . As a result , they move more slowly , allowing the cells behind them to eventually catch-up . These findings reveal a general mechanism in which limitations in how individuals sense and respond to an external signal ( in this case local nutrient concentrations ) allows them to continue migrating together . This mechanism may apply to other systems that migrate via chemotaxis , as well as groups whose movement is directed by different external factors , such as temperature and light intensity . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"physics",
"of",
"living",
"systems",
"computational",
"and",
"systems",
"biology"
] | 2022 | Chemotactic smoothing of collective migration |
Although damage to the primary visual cortex ( V1 ) causes hemianopia , many patients retain some residual vision; known as blindsight . We show that blindsight may be facilitated by an intact white-matter pathway between the lateral geniculate nucleus and motion area hMT+ . Visual psychophysics , diffusion-weighted magnetic resonance imaging and fibre tractography were applied in 17 patients with V1 damage acquired during adulthood and 9 age-matched controls . Individuals with V1 damage were subdivided into blindsight positive ( preserved residual vision ) and negative ( no residual vision ) according to psychophysical performance . All blindsight positive individuals showed intact geniculo-hMT+ pathways , while this pathway was significantly impaired or not measurable in blindsight negative individuals . Two white matter pathways previously implicated in blindsight: ( i ) superior colliculus to hMT+ and ( ii ) between hMT+ in each hemisphere were not consistently present in blindsight positive cases . Understanding the visual pathways crucial for residual vision may direct future rehabilitation strategies for hemianopia patients .
Following damage to the primary visual cortex ( V1 ) patients experience homonymous hemianopia , in which vision on one side of the visual field is lost . However , in spite of this cortical blindness , some patients are still able to ascertain information about visual stimulation within the blind area; this is called blindsight . Over the past 30 years , several visual pathways have been proposed to underlie this residual vision , but the relative role of these pathways and the neurobiological bases for blindsight remains unknown ( see Cowey , 2010 for review ) . Diffusion-weighted magnetic resonance imaging ( dMRI ) combined with tractography offers a practical and non-invasive method for estimating large-scale white matter tracts and studying their microstructural properties in living humans ( Johansen-Berg , 2010; Catani et al . , 2012; Jones et al . , 2013 ) . The method provides a unique approach to investigate how white matter properties relate to visual behaviour in blindsight . Using dMRI in a number of individual patients , intact ipsilateral white matter connecting lateral geniculate nucleus ( LGN ) and extrastriate cortex , specifically area hMT+ , has been proposed as a candidate circuit that could support blindsight ( de Gelder et al . , 2008; Bridge et al . , 2010 ) . In agreement with this proposal , the macaque LGN can support residual visual processing after V1 lesion ( Schmid et al . , 2010 ) . Two alternative proposals suggest blindsight results either from visual plasticity , for example to strengthen interhemispheric white matter in humans ( Leh et al . , 2006; Bridge et al . , 2008; Tamietto et al . , 2012 ) or intact connections to hMT+ from the superior colliculus and pulvinar , demonstrated in the macaque ( Warner et al . , 2010 , 2015 ) . The superior colliculus has also been implicated in human residual vision after V1 damage , particularly for indirect blindsight and saccadic localisation ( Mohler and Wurtz , 1977; Leh et al . , 2006; Kato et al . , 2011 ) . To date the necessary circuitry supporting preserved vision after V1 damage in humans has not been identified . The present study investigated visual white matter tracts in the largest group of patients measured to date with chronic unilateral V1 damage sustained in adulthood ( n = 17 , see Supplementary file 1 for clinical and demographic details ) and healthy age-matched controls ( n = 9 ) . The large subject group enabled the division of patients into those demonstrating blindsight , and those who did not . Three pathways were selected ( 1 ) ipsilateral connections between the LGN and hMT+ , ( 2 ) ipsilateral tracts between the superior colliculus and hMT+ , and ( 3 ) interhemispheric tracts between hMT+ bilaterally . We evaluated the ability to identify these tracts in all individuals and characterised their anatomy and white matter properties . The preservation or destruction of the geniculate-hMT+ tract predicted presence or absence of blindsight respectively . More specifically , the geniculate-hMT+ tract was reliably identified in all blindsight positive patients , but was impossible to track or showed considerably impaired white matter microstructure in all blindsight negative individuals . In contrast , the two alternative candidate tracts showed variable predominance in both patient groups and therefore seem unlikely to underlie blindsight function .
Blindsight was determined according to performance on a high salience 2-AFC temporal detection paradigm presented within the blind region of the visual field ( Figure 1A ) . Patients detected the interval in which the target appeared ( Figure 1B ) and were classified as ‘blindsight positive’ if average performance or performance for stimuli of 100% contrast was significantly above chance ( Figure 1C; Ajina et al . , 2015b ) . Based on these criteria , 12 were classified as ‘blindsight positive’ and this relatively sensitive binary measure allowed us to be confident that patients labelled as ‘blindsight negative’ ( n = 5 ) showed no residual visual function . No patients could describe the stimulus in their blind field , although the degree of awareness varied from a complete absence of awareness to an appreciation of motion at times in the minority of cases . 10 . 7554/eLife . 08935 . 003Figure 1 . Psychophysics protocol and results . ( A ) Example Humphrey visual field deficit drawn schematically , with the location of the target stimulus superimposed . Dense visual field loss is shown in black ( <0 . 5% ) and partial loss in grey ( <2% ) . ( B ) Illustration of the 2AFC-temporal detection procedure . Participants fixated on a central cross , with the onset of each 1500ms interval alerted by a low ( interval 1 ) or high pitch ( interval 2 ) tone . The stimulus could appear in either interval , for a period of 500 ms . At the end of the trial , participants were instructed to decide in which interval the stimulus appeared . ( C ) Detection performance with increasing stimulus contrast , shown separately for blindsight positive ( blue ) and blindsight negative ( red ) patients . Individual results are also plotted for each patient . Chance level is 50% . DOI: http://dx . doi . org/10 . 7554/eLife . 08935 . 003 Classification of participants as either ‘blindsight positive’ or ‘blindsight negative’ was further validated using cross-validation . Two different classification algorithms were applied to participants' performance across all contrast levels and they were compared to classification based only on performance at 100% contrast . The first algorithm was k-nearest-neighbours: it classified participants based on the labels ( ‘blindsight positive’ or ‘blindsight negative’ ) assigned to the majority of the 5 participants with behavior most similar to theirs , based only on the performance of these 5 participants at 100% contrast . The second algorithm was a Gaussian mixture model classification: centroids of two Gaussian distributions were used to fit the data without labels ( i . e . with no knowledge of whether any of the participants were classified as ‘blindsight positive’ or ‘blindsight negative’ based on performance at 100% contrast ) . Each participant was then assigned to one of the two classes based on their similarity to the centroid of each of these distributions . Both classification algorithms agree with the distinction based on performance in 100% contrast in all cases . All 12 blindsight positive patients and the 9 age-matched controls were found to have ipsilateral , uncrossed tracts between the LGN and hMT+ . We combined High-Angular Resolution Diffusion-weighted magnetic resonance Imaging ( HARDI; 60 diffusion directions , b-value = 1500 ) and modern probabilistic tractography ( Tournier et al . , 2012; see ‘Materials and methods’ for more details ) to track between different pairs of regions of interest ( ROIs ) with a fixed number of fascicles , or steamlines ( target 10 , 000 , max generated 1 , 000 , 000 ) . We counted the number of fascicles between each of several ROI pairs in each brain . The precise number of fascicles is an arbitrary value , dependent on many interacting tracking parameters and properties of the measured diffusion data ( see Pestilli et al . , 2014 and ‘Discussion’ for more details ) . Here , we used fascicle count as an indirect measure of the difficulty of tracking a white matter pathway . To further standardise the measure , we used an anatomical standardisation method to eliminate outlier fascicles from counts while constraining the number to a conservative lower bound within each individual brain . This was achieved by removing outlier fascicles defined as those more than about 2 . 5 standard deviations away from or longer than each core tract ( 2 . 6 and 2 . 8 SD respectively; see Allen et al . , 2015; Pestilli et al . , 2014; Yeatman et al . , 2012 and ‘Materials and methods’ for details ) . This process generated a core tract-bundle containing a conservative 25% ± 8% of the original number of fascicles in each subject ( see Supplementary file 2 for original numbers ) . The number of fascicles measured was of similar magnitude in control and blindsight positive individuals . All tracts were reliably measured in both hemispheres , including the hemisphere with V1 damage in blindsight positive patients ( see Table 1 ) . As expected ( Jones et al . , 2013; Pestilli et al . , 2014 ) , even after cleaning , there was considerable variation in fascicle numbers between participants and , in some cases , between hemispheres although this variability was similar for controls ( range = 19–653 ) and patients ( range = 17–635 ) . In blindsight negative patients it was possible to track a pathway between the LGN and hMT+ in the damaged hemisphere of 4/5 patients ( we failed to identify the tract in PN4 ) , with a similar number of fascicles to blindsight positive patients ( Figure 2A , Table 1 ) . However , all of these patients showed considerable abnormality in the microstructure of these tracts compared to their intact hemisphere or control participants , highlighting the importance of considering white matter microstructure in patient tractography studies . Figure 2A shows examples of the anatomical trajectory of these identified pathways for participants from the three groups . 10 . 7554/eLife . 08935 . 004Table 1 . Number of cleaned fascicles for the three pathways of interest in patients and control participantsDOI: http://dx . doi . org/10 . 7554/eLife . 08935 . 004SubjectLGN ↔ hMT+Crossing hMT+SC ↔ hMT+Ipsi-lesionalContra-lesionalLeft → rightRight → leftIpsi-lesionalContra-lesionalBlindsight positive patientsPB1751159no1212PB2191966nono18PB3506724241717PB419315nonono15PB593837191414PB612641315no17PB739717nono168PB8873712162017PB963553nono16noPB103229nonononoPB112914791812noPB1219417171315noBlindsight negative patientsPN115719nono158PN2172261315721PN335189nono1916PN4no101nonono19PN515122nonono13ControlsC130833919141418C2619269nono3917C3575981618noC4176114861816C58430581716C65719nono1516C77846nono9noC849818219143514C9653625752319 ( 1 ) Ipsilateral pathway between LGN and hMT+ ( 2 ) Pathway between hMT+ bilaterally via the corpus callosum ( 3 ) Ipsilateral pathway between SC and hMT+ . Results are shown separately for the intact and damaged ‘ipsi-lesion’ hemispheres ( right and left for control participants ) . ‘no’ = zero fascicles survived the cleaning process . 10 . 7554/eLife . 08935 . 005Figure 2 . ( A ) 3-D representations of ipsilateral tracts between the LGN and hMT+ . Examples are shown for blindsight positive patients PB9 and PB8 , blindsight negative patients PN2 and PN3 and control participants C8 and C4 . Dark green tracts are in the ipsilesional damaged hemisphere , light green tracts are in the intact hemisphere and controls . Tracts are overlaid on a 3-D representation of participant's structural T1-weighted images . ( B ) Average FA along the ipsilateral geniculate-hMT+ pathways of blindsight positive patients , blindsight negative patients , and controls . Blindsight positive patients show a slight reduction in anisotropy over the proximal half of the ipsilesional pathway , although the distal half shows no notable difference to the intact hemisphere . Blindsight negative patients show a marked reduction in FA in the damaged hemisphere beyond the 35th node , continuing to the end of the tract . Control participants show similar results for both hemispheres ( right hemisphere blue , left hemisphere red ) , with FA close to 0 . 5 throughout . The control range for this pathway is displayed in yellow on all charts . DOI: http://dx . doi . org/10 . 7554/eLife . 08935 . 005 Fractional anisotropy ( FA ) and mean diffusivity ( MD ) are commonly used diffusion MRI indices , representing tissue microstructure in situations of neuronal damage ( Werring et al . , 2000; Jones et al . , 2013 ) . These measures are quite sensitive to a number of tissue properties , such as axonal ordering , axonal packing density , degree of myelination , membrane permeability , without being very specific to any one of them . This can result in difficulties related to interpretation . FA is derived from the relationship between the amounts of free water anisotropic diffusion in a single ( principal ) direction , relative to all other directions ( Basser , 1995; Jones et al . , 2013 ) . Decreases in FA have been associated with impaired tissue microstructure . MD is a measure of the total mean diffusion magnitude in all directions in a voxel , and its value also reflects a complex relationship with tissue microstructure . In general , white matter tissue damage has been associated with an increase in MD ( Jones et al . , 2013 ) . We used an advanced tract-anatomy informed analysis ( Allen et al . , 2015; Yeatman et al . , 2012; see ‘Materials and methods’ for more details ) and measured FA and MD along the length of each individual tract . We then computed the mean FA and MD measures along each tract using the core portion of the pathway to eliminate artifactual measurements due to potential partial voluming with grey matter and scar tissue . Mean FA and MD for each core tract were averaged for each participant to generate separate measures for the ipsilesional and intact hemispheres . Mean FA , calculated across all blindsight positive patients and collapsed along the whole geniculate-hMT+ pathway , was 0 . 43 ± 0 . 05 ( mean ± s . d . ) in the damaged hemisphere and 0 . 49 ± 0 . 05 in the intact hemisphere , corresponding to a laterality of 13 . 7% ( see Figure 2B and ‘Materials and methods’ for details ) . Laterality , representing the relative difference in diffusivity for equivalent tracts in opposite hemispheres , was slightly more prominent over the early-mid portions of the pathway . In blindsight negative patients ( Figure 2B , middle column ) , the microstructure of ipsilesional tracts was particularly abnormal . Mean FA was 0 . 35 ± 0 . 1 ( mean ± s . d . ) on the ipsilesional side , vs 0 . 47 ± 0 . 03 in the intact hemisphere ( laterality = 34 . 7% ) . In comparison , control participants ( Figure 2B , right column ) show a left-right laterality of 3 . 3% for FA ( range = 0 . 3–0 . 66 . Mean FA = 0 . 51 ± 0 . 03 left , 0 . 49 ± 0 . 03 right hemisphere ) and 1 . 6% for MD ( range = 0 . 56 x 10−3–0 . 91 x 10−3 , Mean MD = 0 . 73 x 10−3 ± 0 . 03 x 10−3 left , 0 . 72 x 10−3 ± 0 . 03 x 10−3 right hemisphere ) . White matter tract MD in patients was consistent with the findings for FA . In blindsight positive cases , mean MD was 0 . 81 x 10−3 ±0 . 07 x 10−3 in the damaged hemisphere , and 0 . 73 x 10−3 ±0 . 05 x 10−3 in the intact hemisphere ( laterality = 9 . 6% ) . Conversely , blindsight negative patients had a mean MD of 1 . 05 x 10−3 ±0 . 22 x 10−3 in the damaged hemisphere , compared to 0 . 77 x 10−3 ± 0 . 05 x 10−3 on the intact side , representing a laterality of 27 . 0% . The differences between blindsight patients and lesion side can be illustrated using a two-way ANOVA of the FA values within the geniculate-hMT+ tract . While there was no significant effect of blindsight status ( positive or negative; F = 2 . 6 , p = 0 . 13 ) , there was a highly significant effect of lesion side ( ipsilateral or contralateral; F = 35 . 7; p < 0 . 00005 ) and interaction , suggesting a differential effect of the lesion in the two patient groups ( F = 5 . 1; p = 0 . 04 ) . This effect was even stronger for MD: significant effect of blindsight status ( F = 9 . 2; p < 0 . 01 ) , significant effect of lesion side ( F = 35 . 7; p < 0 . 00005 ) and interaction ( F = 12 . 4; p < 0 . 005 ) . Since estimates of distinct pathways frequently overlapped ( Figure 3A ) , the slight laterality in patients may , at least in part , be driven by an overlap with degenerated optic radiations supplying damaged V1 . Figure 3A shows how this overlap can occur in the early-mid portions of the geniculate- hMT+ pathway . In the central nervous system , anterograde ( Wallerian ) or retrograde neuronal degeneration can occur following axonal injury . Consequently , the integrity of optic radiation fibres innervating damaged V1 would be abnormal throughout their course ( similar to Danek et al . , 1990 ) . Where overlap with such fibres occurs , the dMRI measurements would not distinguish between separate axonal bundles due to limitations in spatial resolution ( restricted here to 2 mm isotropic voxel size ) . Thus measures of the geniculate-hMT+ tract could become contaminated with overlapping degenerated optic radiation fibres . It may therefore be useful to measure the diffusivity spanning only the distal portion of the geniculate-hMT+ tract , which has branched away from the large geniculate-V1 radiation bundle . This may represent a purer measure of the pathway , removing artefacts due to overlapping tracts . If the pathway to hMT+ were actually damaged , one would still expect this measure to reflect the damage . 10 . 7554/eLife . 08935 . 006Figure 3 . ( A ) Normal ipsilateral tracts between the LGN and hMT+ , and the LGN and V1 demonstrate a proximal region of overlap . Tracts are demonstrated in a control participant , C2 , comparing ipsilateral connections between the LGN and hMT+ ( pink ) and the LGN and V1 ( blue ) . When these pathways are superimposed , there is a significant region of overlap in the proximal portion of these pathways . In cases of V1 damage where there is retrograde degeneration , this overlapping region of the geniculate-hMT+ pathway may become contaminated by degenerated tracts in the V1 pathway . ( B ) Box plots comparing FA and MD in the distal portion of the geniculate-hMT+ pathway , in blindsight positive and negative patients . The ipsilesional hemisphere is shown in purple , and the intact hemisphere in green . Blindsight positive patients show significant overlap in the FA of the distal portion of this pathway in the damaged and sighted hemispheres . There is a slight increase in MD in the damaged hemisphere , however this is not marked and both measures fall within the control range . In comparison , blindsight negative patients show a marked difference in FA and MD for this pathway in the damaged and sighted hemispheres . The ipsilesional measures extend beyond the control range , implying that they are pathological and significantly impaired . Adjacent values are defined as the lowest and highest observations that are still inside the region defined by the following limits: Lower Limit = Q1 − 1 . 5 x IQR , Upper Limit = Q3 + 1 . 5 × IQR . The age-matched control FA and MD range for this pathway are displayed in yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 08935 . 006 Figure 3B shows the microstructure of just the distal portions of the geniculate-hMT+ pathway ( region between 60% and 85% of the total tract length from the LGN ) . Mean FA in blindsight positive patients was 0 . 39 ± 0 . 06 in the damaged hemisphere and 0 . 42 ± 0 . 04 in the intact hemisphere , corresponding to a laterality of only 7 . 7% ( mean MD = 0 . 79 x 10−3 ± 0 . 06 x 10−3 vs 0 . 73 x 10−3 ± 0 . 06 x 10−3 , laterality = 7 . 6% ) . Individual data also confirmed that the FA or MD standard deviation in each blindsight positive patient overlapped with the opposite hemisphere . Overall , this implies a less pronounced impairment to the white matter than suggested by the entire extent of the tract . In blindsight negative patients , mean FA in the distal portion of this tract was 0 . 29 ± 0 . 07 , compared to 0 . 44 ± 0 . 05 in the intact hemisphere ( laterality = 51 . 7% ) , mean MD = 1 . 04 x 10−3 ± 0 . 23 x 10−3 vs 0 . 74 x 10−3 ± 0 . 07 x 10−3 , ( laterality = 28 . 8% ) . Therefore , unlike the blindsight positive group , blindsight negative patients still showed a relatively marked and significant drop in FA and increase in MD throughout this purer geniculate-hMT+ tract ROI when compared to the intact hemisphere . Although the effect of blindsight status on mean FA within this distal portion was not significant ( F = 0 . 7; p = 0 . 42 ) , both the effect of lesion side ( F = 48 . 6; p < 0 . 00001 ) and the interaction ( F = 15 . 6; p < 0 . 0005 ) were highly significant . The effect of blindsight status on mean MD was significant ( F = 7 . 9; p = 0 . 01 ) , as were the effect of lesion side ( F = 31 . 4; p < 0 . 0001 ) and the interaction ( F = 15 . 6; p = 0 . 001 ) . This difference can also be appreciated in brain images by inspecting the tracts in the white matter , and their corresponding FA and MD maps ( Figure 4 ) . Only the blindsight negative patients ( Figure 4 , lower portion ) possess tracts in the damaged hemisphere that appear to traverse a region of white matter displaying very abnormal FA and MD levels . Thus , although fascicles successfully propagated through this region , they passed through regions of profoundly abnormal , damaged tissue ( see also Figure 4—figure supplement 1 for greater detail ) . 10 . 7554/eLife . 08935 . 007Figure 4 . FA and MD maps in blindsight positive and negative patients , demonstrating the spatial relationship with the geniculate-hMT+ pathways . Individual results are shown for two blindsight positive patients PB5 , and PB10 and two blindsight negative patients , PN1 and PN5 . All four patients showed bilateral ipsilateral fascicles between the LGN and hMT+ , including the damaged hemisphere ( column two ) . In the damaged hemisphere of blindsight positive patients the region directly underlying tracts corresponds to relatively intact MD and FA measures , not notably different from the intact hemisphere . However , both blindsight negative patients have tracts in the damaged hemisphere that traverse a region of tissue with markedly abnormal FA and MD values ( columns three and four ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08935 . 00710 . 7554/eLife . 08935 . 008Figure 4—figure supplement 1 . Zoomed in view demonstrating ipsilesional geniculate-hMT+ tracts with the corresponding T1-weighted structural , FA and MD maps . Blindsight positive patients are shown in A and B with blindsight negative patients in C and D . DOI: http://dx . doi . org/10 . 7554/eLife . 08935 . 008 To ensure that any differences in tract microstructure between blindsight positive and negative patients were not driven by differences in grey matter volume in hMT+ , the volume was directly compared both between hemispheres and groups . In the blindsight positive patients the mean grey matter volume was 159 mm3 ± 59 in the ipsilesional hemisphere and 167 mm3 ± 55 in the intact hemisphere . In blindsight negative patients the equivalent numbers were 135 mm3 ± 71 and 169 mm3 ± 83 . There was no significant effect of blindsight status ( F = 0 . 1; p = 0 . 8 ) , lesion side ( F = 1 . 7; p = 0 . 2 ) or interaction ( F = 1 . 0; p = 0 . 3 ) , indicating that differences in grey matter volume within hMT+ are unlikely to have affected the results significantly . So far we have observed a difference in geniculate-hMT+ tract properties between blindsight positive and negative patients , indicating that this pathway might be a candidate for blindsight . Visual motion information could , however , travel via other pathways such as a transcallosal tract connecting left and right hMT+ ( Figure 5A–C ) , or a pathway connecting the superior colliculus and hMT+ ( Figure 5D–F ) . Next we tested whether these alternate tracts could account for blindsight . 10 . 7554/eLife . 08935 . 009Figure 5 . 3-D representations of interhemispheric tracts between hMT+ bilaterally and ipsilateral tracts between SC and hMT+ . ( A–C ) Interhemispheric hMT+ tracts in blindsight positive patient , PB3 , a blindsight negative patient , PN2 and a control participant , C9 . ( D–F ) Ipsilateral collicular-hMT+ tracts in blindsight positive patient , PB8 , a blindsight negative patient , PN3 and a control participant , C2 . Red tracts represent crossing , interhemispheric connections between hMT+ bilaterally . Dark blue tracts are connections between SC and hMT+ in the ipsilesional damaged hemisphere , light blue tracts show the same collicular-hMT+ pathway in the intact hemisphere , and in controls . Tracts are overlaid on a 3-D representation of participant's structural T1-weighted images . DOI: http://dx . doi . org/10 . 7554/eLife . 08935 . 009 Interestingly , in a number of blindsight positive patients it was not possible to identify either a pathway between hMT+ and the superior colliculus , between hMT+ in the two hemispheres , or both . Similarly , intact pathways between these regions were present in blindsight negative cases . Overall both pathways generated approximately 10-fold fewer fascicles than the geniculate-hMT+ pathway ( mean 18 . 9/18 . 7 vs 202 . 8 in controls , and mean 15 . 6/15 . 0 vs 122 . 7 in blindsight positive patients ) . Furthermore , the collicular-hMT+ tracts appeared less consistent in shape and trajectory between individuals . Crossing tracts between hMT+ bilaterally were identified in only 6/12 patients with blindsight and 6/9 controls ( Table 1 , columns 3–4 ) . As expected , pathways always crossed to the opposite hemisphere via the corpus callosum ( Figure 5A–C ) . Where present , tracts also appeared to possess normal FA and MD . In blindsight positive cases , mean FA was 0 . 64 ± 0 . 07 ( mean ± s . d . ) and mean MD 0 . 70 x 10−3 ±0 . 03 x 10−3 , averaged along the entire tract for both directions ( left to right , and right to left ) , and across participants ( Figure 6A ) . These values were similar to controls ( Figure 6C , mean FA = 0 . 64 ± 0 . 07 , mean MD = 0 . 67 x 10−3 ±0 . 05 x 10−3 ) . 10 . 7554/eLife . 08935 . 010Figure 6 . Average fractional anisotropy along the Interhemispheric hMT+ pathway and ipsilateral pathway between SC and hMT+ . ( A ) Blindsight positive patients show a similar FA to controls along the length of interhemispheric hMT+ pathways . ( B ) Blindsight negative patient , PN2 , also shows a similar FA to controls along the length of this pathway . ( C ) Control participants show a normal peak in FA at the centre of the interhemispheric hMT+ pathway , representing the high degree of anisotropy at the corpus callosum . ( D ) Blindsight positive patients show a similar FA in the ipsilesional collicular-hMT+ pathway as the intact hemisphere and controls . ( E ) Blindsight negative patients show a slight drop in mean FA in the distal third of the ipsilesional collicular-hMT+ pathway . ( F ) Control participants show a fairly constant FA along the length of the collicular-hMT+ pathway , around 0 . 4 . The control range for each pathway is displayed in yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 08935 . 010 Only a single blindsight negative patient showed an interhemispheric connection between hMT+ bilaterally ( PN2; Table 1 , Figure 5B ) , and in this case , the tract appeared to be largely intact and remained within the control FA range ( 0 . 28–0 . 91 ) , with mean FA = 0 . 59 and MD = 0 . 73 x 10−3 ( see Figure 6B for FA plots along this path ) . Similarly , the collicular-hMT+ pathway , could not be tracked in all patients with blindsight , and was demonstrable in some blindsight negative individuals . Specifically , these pathways were tracked in the damaged hemisphere of 8/12 blindsight positive patients ( Figure 5D ) . The same proportion of patients had tracts in their intact hemisphere , although not necessarily in the same cases ( Table 1 , columns 5–6 ) . In comparison , this pathway was present in 3/5 blindsight negative patients ( Figure 5E ) . Control participants showed these pathways in all nine cases on the left , and 7/9 on the right ( Figure 5F ) . Of the patients demonstrating this pathway , mean FA was 0 . 40 ± 0 . 05 in the damaged hemisphere of blindsight positive cases , vs 0 . 46 ± 0 . 03 on the intact side , with some regions of overlap along their trajectory ( laterality of 13 . 9% , see Figure 6D for FA plots ) . Mean MD was 0 . 78 x 10−3 ± 0 . 08 x 10−3 vs 0 . 69 x 10−3 ±0 . 04 x 10−3 ( laterality = 10 . 8% ) . In blindsight negative patients ( Figure 6E ) , the pattern of FA was more variable along its trajectory compared to other tract profiles ( i . e . Figures 2B , and 6A–C ) . Collapsed along the pathway , mean FA was 0 . 37 ± 0 . 05 vs 0 . 42 ± 0 . 03 on the intact side ( laterality = 14 . 0% ) , and MD was 0 . 83 x 10−3 x 10−3 ±0 . 13 vs 0 . 77 x 10−3 ±0 . 07 x 10−3 ( laterality = 7 . 5% ) . In fact , this laterality and distal drop in FA was strongly influenced by data from one patient ( PN1 ) . The other two patients ( PN2 and PN3 ) showed a similar microstructure in the distal portion of their collicular tracts in both hemispheres ( t = 1 . 7 , p = 0 . 2 , mean FA = 0 . 32 vs 0 . 37 ) despite a significant laterality in their geniculate-hMT+ pathway ( t = 12 . 2 , p = 0 . 01 , mean FA = 0 . 34 vs 0 . 48 ) . This implies that intact tracts from the superior colliculus can occur in blindsight negative patients . However , it is worth noting that the FA in one of these two patients ( PN2 ) does drop below the control range despite a normal laterality , reaching an FA of 0 . 22 between nodes 69 and 79 ( control range = 0 . 26–0 . 62 , see Figure 6E ) . Statistical comparison of these values is complicated because only 5/12 of the blindsight positive and 3/5 blindsight negative patients had tracts in both hemispheres . This makes the comparison between the ipsi- and contra-lesional hemispheres problematic . However , a comparison of FA and MD in just the ipsilesional hemisphere indicated no significant difference in either measure between the two groups ( FA: t = 0 . 9; p = 0 . 4; d . f . = 9; MD: t = 0 . 8; p = 0 . 45; d . f . = 9 ) . The analyses thus far have addressed group differences by division of patients into blindsight positive and negative groups . However , even within the blindsight positive group , there is considerable variability in performance . Therefore , the percentage of correct responses in the blindsight task was correlated with the measures of mean FA and MD extracted from the three tracts of interest . In the geniculate-hMT+ tract , 16 patients were included as one blindsight negative patient did not have an identifiable tract , and for this tract only the distal portion was used . Figure 7A shows the correlation for FA and MD from this distal portion of the tract across all patients , with blindsight negative indicated by the open symbols and blindsight positive indicated by the filled symbols . Although both correlations were in the predicted direction , positive for FA and negative for MD , neither was significant ( r = 0 . 44; p = 0 . 09 for FA; r = −0 . 48; p = 0 . 06 for MD ) . Since age can be a confounding factor in tract microstructural properties , the partial correlation coefficients , accounting for age were also calculated , but did not differ from the full correlations . Neither the collicular-hMT+ ( r = −0 . 03; p = 0 . 93 for FA; r = −0 . 10; p = 0 . 77 for MD ) nor the interhemispheric hMT+ ( r = −0 . 33; p = 0 . 39 for FA; r = 0 . 26; p = 0 . 50 for MD ) tracts showed any correlation with behaviour . 10 . 7554/eLife . 08935 . 011Figure 7 . Correlation of tract microstructure in the distal region of the geniculate-hMT+ pathway with behavioural performance on the contrast detection task . In both plots the filled symbols represent the values for the blindsight negative patients while the open symbols are from the blindsight positive patients . ( A ) shows the data for the FA values ( r = 0 . 43; p = 0 . 09 ) and ( B ) shows the corresponding values for MD ( r = −0 . 48; p = 0 . 06 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08935 . 011 In addition to performing tractography between pre-defined regions of interest , a useful and unbiased approach to understand why certain patients have blindsight and others do not is to quantify lesion extent and location from the T1-weighted anatomical image . This is particularly valuable in larger patient cohorts given the heterogeneity of damage in such groups . Figure 8 shows the total lesion volume and distribution of damage across all patients . Average lesion volume in the blindsight positive group was 13 , 461 mm3 ± 7101 mm3 s . d . , compared to 36 , 923 mm3 ± 23 , 035 s . d . in the blindsight negative group . On average , blindsight negative patients had lesions approximately 2 . 5 times larger than the blindsight positive group , although Figure 8B shows the overlap of occipital lobe damage between patients . There was a significant association between the extent of occipital lobe damage and the microstructural measures of ipsilesional geniculate-hMT+ pathways across all patients ( FA: r = −0 . 59 , p = 0 . 015; MD: r = 0 . 63 , p = 0 . 01 ) . This reinforces the likelihood that reduced FA and increased MD in blindsight negative individuals reflects an involvement of surrounding white matter pathways in occipital lesions . 10 . 7554/eLife . 08935 . 012Figure 8 . Comparison of lesion size and location in blindsight positive and negative patients . ( A ) Lesion size is given for each patient , and demonstrates a wide range of volumes in both patient groups . ( B ) Lesion location shows the proportion of lobe damage in each patient , within the occipital , temporal , and parietal lobes , as well as the subcortex . Subcortex incorporates the thalamus ( including LGN and pulvinar ) , striatum , and superior colliculi , with an approximate unilateral volume of 50 , 000 mm3 . Only one patient , PN5 , demonstrated some involvement of this region , including the ipsilesional LGN and pulvinar , but not the superior colliculi . DOI: http://dx . doi . org/10 . 7554/eLife . 08935 . 012 There was no clear association between blindsight function and the presence of additional , non-occipital damage . As seen in Figure 8B , the damage in some patients extended to other regions , including the temporal ( PB8 , PN3 , and PN5 ) or parietal lobe ( PN4 and PN5 ) . However , the pattern of such damage was not associated with a particular group . Furthermore , only one participant showed evidence of significant subcortical pathology ( PN5 ) , seen to extend to a region including the ipsilesional LGN and pulvinar , although the superior colliculi appeared intact . At least three patients with blindsight showed complete destruction of calcarine cortex or its underlying white matter ( PB1 , PB4 , PB10 ) . Similarly there were blindsight negative cases with small regions of V1 apparently intact ( PN2 , PN4 ) . The majority of cases with lesions affecting less than 20% of the occipital lobe had some small area of V1 sparing , which usually corresponded to the occipital pole , or the anterior tip of the calcarine sulcus .
The principal finding was that all patients with blindsight function showed intact , undamaged tracts between the LGN and hMT+ in the hemisphere with V1 damage . This is consistent with our recent fMRI report of motion processing after V1 damage ( Ajina et al . , 2015a ) . A similar direct geniculate pathway was identified in all age-matched controls , and is consistent with neuroanatomical investigation in the macaque ( Sincich et al . , 2004 ) . This was not the case in blindsight negative patients , where such geniculo-extrastriate tracts were either absent or demonstrated significant impairment in mean diffusivity and fractional anisotropy . This has two important implications that support geniculate-extrastriate connections in blindsight . ( 1 ) It is possible that intact connections between the LGN and hMT+ are sufficient for blindsight , since no patients with blindsight function demonstrated an absence or impairment in these pathways . ( 2 ) Intact connections between the LGN and hMT+ may also be necessary for blindsight function . This is supported by the results in blindsight negative patients , as none of the patients without blindsight function possessed normal , intact connections in this pathway . However , since the current study only investigated a limited number of potential pathways , it is possible that other , unexplored , pathways could also underlie blindsight function in these patients . Unlike direct geniculate connections , there were examples of patients with blindsight function who had absent or impaired collicular and interhemispheric pathways . Similarly , there were blindsight negative cases with apparently undamaged connections between these regions . These results suggest that neither of these other putative blindsight pathways have a major role in blindsight function , but support the argument that in blindsight positive cases , another pathway must have facilitated visual performance . However , these pathways may still contribute to blindsight in some circumstances . Only the geniculate-hMT+ tract showed a marginal correlation of behavioural performance with MD and FA that was in the correct direction: improved performance correlated positively with FA and negatively with MD . However , neither of these correlations was significant . Although the current study provides the largest participant group reported to date , the considerable variability in values for tract microstructure means that there may not be sufficient power to find differences , particularly for smaller tracts . Here we have selected three pathways with very distinct trajectories to investigate . There are , however , other potential pathways by which blindsight information could be processed , the most prominent of which is the one from the medial portion of the inferior pulvinar to MT identified in multiple primate species ( Maunsell and van Essen , 1983; Warner et al . , 2012 ) . There is some debate as to the relative strength of the connections to hMT+ from LGN or the inferior pulvinar ( Sincich et al . , 2004; Warner et al . , 2010 ) , although there is evidence that the pulvinar connection is stronger during early development ( Warner et al . , 2015 ) . There are several practical reasons why the pulvinar connection with hMT+ has not been quantified in the current study . Firstly , the most commonly described pathway is a di-synaptic pathway from colliculus to hMT+ via the inferior pulvinar ( Lyon et al . , 2010 ) . Thus , this corresponds to the collicular-hMT+ pathway examined here that was both difficult to track , but also showed reduced microstructure in a number of blindsight positive patients . A direct , retinorecipient pathway from the inferior pulvinar has been described in the marmoset ( Warner et al . , 2010 ) , suggesting that it would also be worth considering only a connection between inferior pulvinar and hMT+ . A recent human tractography study considered tracts from both LGN and pulvinar to hMT+ as part of an investigation into the visual pathways in amblyopia ( Allen et al . , 2015 ) . Quantification of the tract microstructure found that they were very similar with almost identical values for both FA and MD . Thus , at the current resolution of 2 mm isotropic voxels , dissecting apart these two tracts may be impossible , due to the proximity of the thalamic structures . In future studies , higher spatial resolution may help to disentangle these two important pathways . Thus , we cannot completely rule out the presence of an intact direct pathway from the retina to hMT+ via the pulvinar . All three of the pathways studied here have been previously investigated in case studies , although they have never been compared in the same patients . Two studies investigated the pathways underlying motion ( Bridge et al . , 2008 ) or affective blindsight ( Tamietto et al . , 2012 ) in blindsight patient GY . The motion study reported a direct ipsilateral connection between LGN and hMT+ in the damaged hemisphere , similar to the results for blindsight positive patients here . However , GY also showed unusual patterns of connectivity that may be indicative of plasticity . These included a cortico-cortical callosal connection between hMT+ bilaterally ( tested here , but absent in 6/12 blindsight positive patients ) and a crossing pathway between LGN in the undamaged hemisphere and ipsilesional hMT+ . In both cases these unusual pathways were largely demonstrable in controls , although GY showed a considerably greater number of fascicles ( Bridge et al . , 2008 ) . The only study to investigate collicular pathways was in patients following hemispherectomy , two of whom had attentional blindsight ( Leh et al . , 2006 ) . Only patients with blindsight showed crossing tracts between the superior colliculus in the damaged hemisphere and regions of the intact hemisphere , as well as strong ipsilateral connections in the damaged hemisphere . These crossing tracts were seen in some control participants , although were arguably less prominent and were therefore also taken as a possible indicator of plasticity . The current study found no evidence to support such plasticity in adult-onset V1 damage and blindsight . Furthermore , additional evidence against a necessary transcallosal connection comes from cases of bilateral cortical damage with significant fMRI hMT+ activity and blindsight ( Bridge et al . , 2010 ) . Where occipital damage is bilateral , the corpus callosum undergoes profound degeneration and is unlikely to provide useful visual information ( de Gelder et al . , 2008 ) . One possible explanation for some of these differences is the age of brain injury onset , since damage acquired in childhood may lead to greater plastic changes ( Anderson et al . , 2011; Tinelli et al . , 2013 ) . GY sustained his brain injury aged 8 years , and the hemispherectomy patients sustained severe structural brain damage at birth or in early childhood , despite undergoing resective surgery later in life . Both studies identified increased interhemispheric connectivity in blindsight , unlike the cases of cortical blindness and patients in the current study , all of whom sustained damage in adulthood . This could be consistent with an increased propensity for plasticity in the corpus callosum , which continues to grow in cross-sectional area until early adulthood ( Keshavan et al . , 2002 ) . The other factor to consider is how blindsight is assessed , and the type of blindsight present . It has been argued in the past that different forms of blindsight may be mediated by distinct anatomical pathways or structures ( Danckert and Rossetti , 2005 ) . For example , collicular processing may be involved in ‘action’ or ‘attention’ blindsight , whilst the LGN is implicated in perceptual characteristics , described as ‘agnosopsia’ ( Zeki and Ffytche , 1998 ) . The definition of blindsight differs considerably between tractography studies , ranging from comparable 2-AFC testing ( Bridge et al . , 2010 ) to navigational tests ( de Gelder et al . , 2008 ) and indirect or ‘attentional’ blindsight ( Leh et al . , 2006 ) . In particular , patients with extensive cortical damage beyond V1 appear to lack any awareness or direct response to blind field stimulation ( Tomaiuolo et al . , 1997; Leh et al . , 2006; de Gelder et al . , 2008 ) . Indirect blindsight assessments may be more sensitive than the 2-AFC tests used here , and may rely on different structures . The only way to tackle this would be to improve consistency amongst experiments , and to include multiple methods of assessing blindsight in future work . A significant concern highlighted from the current study is that it was possible to track robust fascicles in patients traversing regions of extremely impaired FA and MD . Indeed , it may even be the case that fascicles are biased towards narrow regions of white matter running alongside a lesion boundary . Patient PN1 , for example , showed almost ten times more fascicles in the geniculate-hMT+ pathway of his damaged hemisphere compared to his intact side , even though tracts quite clearly passed through a region of abnormal ( damaged ) tissue ( Figure 4 ) . These tracts are unlikely to be functional , as indicated by the negative psychophysical performance . Emphasis on fascicle numbers without considering the underlying microstructure and pathway viability is therefore problematic . Indeed , there are many reasons why fascicle numbers provide unreliable measures of true axonal projections and function ( Jones et al . , 2013; Pestilli et al . , 2014 ) . Even if the ‘true’ fibre count is uniform , the number of reconstructed fascicles may differ due to the length , curvature , and degree of branching present ( Jones and Cercignani , 2010 ) . Such variability was apparent here , as even control participants showed notable differences in fascicle numbers between hemispheres and individuals . Two of the key early papers on blindsight have focused on this measure ( Leh et al . , 2006; Bridge et al . , 2008 ) , interpreting a quantitative difference in fascicles as suggestive of plasticity . Whilst this may be correct , any tractography algorithm with a bias for peri-lesional pathways could contribute to such findings . One of the more controversial uses of MRI diffusion tractography is to comment on the existence or absence of a specific pathway , with false positive connections particularly problematic ( Sherbondy et al . , 2008; Gao et al . , 2013 ) . This is not surprising if one considers that the success or failure of fascicle propagation in tractography algorithms is subject to the same limitations as the fascicle count . In the current study , an important source of variation was the process of ‘cleaning’ to isolate robust and consistent tracts . If a less stringent cut-off had been used , interhemispheric hMT+ connections would be identified in 100% of controls , thus necessitating care in their interpretation . Although the interhemispheric and geniculate pathways in patients would remain unaffected , a less stringent cut-off would suggest collicular tracts were present in all blindsight positive patients . However , when visualized , these pathways containing fewer than 5 fascicles appear largely implausible , reinforcing the need for a cleaning process to improve data reliability and reduce false positives . A novel mechanism to address this in the future may be to estimate the accuracy of an estimated connectome and tract , such as using Linear Fascicle Evaluation ( Pestilli et al . , 2014 ) . In summary this work provides strong evidence to support a direct geniculate connection to extrastriate cortex as being important for blindsight function in adult-onset V1 damage . Although alternate interhemispheric and collicular pathways were also demonstrable in a number of patients , these connections were unable to account for all blindsight cases and were often found to be intact in patients with absent blindsight performance . The results also highlight the importance of considering white matter microstructure when performing tractography in patients , which is applicable to anyone working with clinical diffusion data . Finally , appreciation of the important tracts may help to direct attempts to boost residual function through rehabilitative strategies in hemianopia .
Seventeen patients ( five female ) took part in this study , of which 15 had sustained posterior circulation stroke and two had undergone benign tumour resection , see Supplementary file 1 for details . All patients had sustained unilateral damage to V1 , causing homonymous visual field loss recorded by Humphrey perimetry . Average age at the time of participation was 54 . 9years ± 14 . 4 , average time after pathology onset 45 months ( range 6–252 months ) . Nine healthy participants ( 54 . 9 ± 11 . 7 years old , three female ) served as controls . Written consent was obtained from all participants . Control participants and patients were matched by age and sex at the time of testing . Controls all had normal or corrected-to-normal visual acuity and no history of neurological disease . Ethical approval was provided by the Oxfordshire Research Ethics Committee ( Ref B 08/H0605/156 ) . Testing was performed at the John Radcliffe Hospital , Oxford . Psychophysical testing was conducted outside the MRI scanner , with a 60Hz CRT monitor at a distance of 68 cm . Visual stimuli consisted of a drifting achromatic Gabor patch of 5o or 8o diameter , displayed on a uniform grey background; temporal frequency 10Hz , spatial frequency 1 . 3 cycles/o . Five contrast levels were used: 1% , 5% , 10% , 50% , and 100% , with stimulus location restricted to the scotoma and its corresponding location in the sighted hemifield in patients , a minimum of 3° from fixation ( see Figure 1A for schematic representation of stimulus location ) . Participants were asked to indicate whether a stimulus appeared in the first or second time-interval ( Figure 1B ) . If they saw nothing , they were instructed to guess . Onset of each interval was indicated by a 500ms auditory tone , 300Hz marking onset of the first interval , and 1200Hz for the second . Visual stimuli appeared for 500 ms with jittered onset while the participant fixated on a central black cross . Stimulus contrast was altered parametrically between the five levels at random , with 20 trials per condition . The allocated interval ( first or second ) was also generated at random . Participants additionally performed a run of control testing , with stimuli presented to the equivalent location in their sighted visual field . Fixation was recorded throughout with an Eyelink 1000 eye tracker ( SR Research Limited , Ontario , Canada ) , and any trials in which eye position exceeded 1° from fixation were excluded from analysis . Participants were reminded to maintain fixation , with the investigator observing this in real-time . Anyone making even a small eye movement into their damaged hemifield was given specific instruction not to do so , and it was explained that these data would have to be discarded . The presence or absence of blindsight , or residual visual function was determined for each patient . This was defined as achieving either an average score , or a score for stimuli of 100% contrast that was significantly above chance , using a statistical threshold of p < 0 . 01 and a cumulative binomial distribution . This criterion led to the allocation of 12 patients as ‘blindsight positive’ ( PB1-PB12 ) and five as ‘blindsight negative’ ( PN1-PN5 ) , see Figure 1C and Table 1 for details . Classification of patients into these two groups ( ‘blindsight positive’ and ‘blindsight negative’ ) was therefore further validated using cross-validation with two other cross-validation strategies:K-nearest neighbours: in each iteration one of the participants was held out and was blindly labelled ( as ‘blindsight negative’ or ‘blindsight positive’ ) according to the label previously assigned to the majority of their k-nearest neighbours ( using only performance at 100% contrast ) . Neighbourhood distance between the currently labelled individual and other individuals was measured in terms of their performance in all contrast levels ( k was set to 5 , but other values of k were also tested and results were found to be robust to choice of k ) . A Gaussian mixture model was fit to behavioural performance data across all contrast level . Fitting was ‘blind’ . That is , no class labels ( ‘blindsight positive’ or ‘blindsight negative’ ) were used in fitting the multi-dimensional Gaussian distributions . Each participant was then classified into one of two groups according to their distance from the centroids of the two Gaussian distributions . Both algorithms were implemented in scikit-learn ( Pedregosa et al . , 2011 ) . Accuracy of the classification was evaluated relative to the labels ( ‘blindsight positive’ or ‘blindsight negative’ ) derived from the classification based only on performance at 100% contrast ( also used in Ajina et al . , 2015b ) . Behavioural testing of control participants and the sighted hemisphere of patients was not possible , since the contrast task is too easy , resulting in 100% detection of even the 1% contrast stimulus . | Visual information from our eyes projects to a region at the back of the brain called the primary visual cortex , which is where the information is processed to allow us to see the world around us . If a person suffers a stroke that affects this primary visual cortex , he or she can become blind on one side . However , some people can still detect images within this ‘blind’ area , even if they are not consciously aware of it . This phenomenon is known as ‘blindsight’ , but it remains unclear which pathways and structures in the brain might allow this information to be detected . Ajina et al . have now examined the brains of a large group of patients with damage to the visual cortex . The results for the patients with blindsight were compared to those without , and to a group of sighted control participants . This analysis identified a pathway that seems to underlie blindsight . This pathway ( which runs between an area of the brain called the lateral geniculate nucleus and another called the motion area hMT+ ) was present in all patients with blindsight , but was missing or disrupted in those patients without blindsight . Ajina et al . then examined other pathways that had previously been suggested to support blindsight and revealed that they were unlikely to do so . This is because the suggested connections were not identifiable in all patients with blindsight , and were often intact in those patients without blindsight . So far , this work has addressed the structure of the pathways rather than their activity . Future work will attempt to determine whether it is possible to strengthen such pathways to improve visual ability . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Human blindsight is mediated by an intact geniculo-extrastriate pathway |
Cell surface receptors govern a multitude of signalling pathways in multicellular organisms . In plants , prominent examples are the receptor kinases FLS2 and BRI1 , which activate immunity and steroid-mediated growth , respectively . Intriguingly , despite inducing distinct signalling outputs , both receptors employ common downstream signalling components , which exist in plasma membrane ( PM ) -localised protein complexes . An important question is thus how these receptor complexes maintain signalling specificity . Live-cell imaging revealed that FLS2 and BRI1 form PM nanoclusters . Using single-particle tracking we could discriminate both cluster populations and we observed spatiotemporal separation between immune and growth signalling platforms . This finding was confirmed by visualising FLS2 and BRI1 within distinct PM nanodomains marked by specific remorin proteins and differential co-localisation with the cytoskeleton . Our results thus suggest that signalling specificity between these pathways may be explained by the spatial separation of FLS2 and BRI1 with their associated signalling components within dedicated PM nanodomains .
Multicellular organisms employ cell-surface receptors for surveying the environment and adjusting to changing physiological conditions . In plants , the repertoire of cell surface receptors has been considerably expanded and receptor kinases ( RKs ) form one of the largest protein families with over 600 members in Arabidopsis thaliana ( hereafter , Arabidopsis ) ( Shiu and Bleecker , 2001 ) . The schematic architecture of plant RKs is similar to that of animal receptor tyrosine kinases ( RTKs ) ; comprising an extracellular ligand binding domain , a single transmembrane helix , and an intracellular kinase domain ( Shiu and Bleecker , 2001 ) . Prominent examples of plant RKs are the immune receptor FLAGELLIN SENSING 2 ( FLS2 ) ( Gómez-Gómez and Boller , 2000 ) and the growth receptor BRASSINOSTEROID INSENSITIVE 1 ( BRI1 ) ( Clouse et al . , 1996; Li and Chory , 1997 ) . FLS2 is a pattern recognition receptor ( PRR ) that perceives the pathogen-associated molecular pattern ( PAMP ) flg22 , an immunogenic epitope of bacterial flagellin , to initiate PAMP-triggered immunity ( PTI ) ( Felix et al . , 1999; Zipfel et al . , 2004; Chinchilla et al . , 2006; Boller and Felix , 2009 ) . BRI1 binds brassinosteroids ( BRs ) , a class of phytohormones involved in various aspects of plant growth and development ( Kinoshita et al . , 2005; Kim and Wang , 2010; Singh and Savaldi-Goldstein , 2015 ) . Despite their different biological functions , FLS2- and BRI1-mediated signalling pathways share several similarities , in particular at or close to the plasma membrane ( PM ) . The PM is the cellular compartment , where both receptors localise to ( Robatzek et al . , 2006; Friedrichsen et al . , 2000 ) , where they bind their respective ligands flg22 or BRs ( Gómez-Gómez et al . , 2001; Bauer et al . , 2001; Kinoshita et al . , 2005 ) , and where presumably their main signalling activity is executed ( Smith et al . , 2014; Irani et al . , 2012 ) . Although FLS2 and BRI1 are competent for ligand binding via their extracellular leucine-rich repeat ( LRR ) domains , they rely on SOMATIC EMBRYOGENESIS RECEPTOR-LIKE KINASE ( SERK ) co-receptors for signalling initiation ( Nam and Li , 2002; Li et al . , 2002; Chinchilla et al . , 2007; Heese et al . , 2007; Roux et al . , 2011; Gou et al . , 2012 ) , which are also LRR-RKs ( Aan den Toorn et al . , 2015 ) . Structural and biochemical analysis of FLS2- and BRI1-SERK hetero-oligomers revealed that flg22 and BRs act as ‘molecular glues’ that stabilise or induce receptor complexes ( Sun et al . , 2013; She et al . , 2011; Hothorn et al . , 2011 ) . Ligand binding additionally triggers auto- and trans-phosphorylation events within the receptor complexes ( Schulze et al . , 2010; Wang et al . , 2008 ) and , in the case of BRI1 , also the release of inhibitory mechanisms ( Wang and Chory , 2006; Jaillais et al . , 2011 ) . After gaining their full kinase activities , FLS2 and BRI1 receptor complexes initiate phosphorylation cascades that culminate in flg22- or BR-responsive transcriptional regulation ( Guo et al . , 2013; Li et al . , 2016 ) . The relay of phosphorylation signals from the PM to the nucleus involves receptor-like cytoplasmic kinases ( RLCKs ) that can associate to the PM and that are direct substrates of the ligand-binding receptor complexes ( Lin et al . , 2013; Belkhadir and Jaillais , 2015; Couto and Zipfel , 2016 ) . Similar to the SERK co-receptors , the RLCKs BRASSINOSTEROID SIGNALING KINASE 1 ( BSK1 ) and BOTRYTIS-INDUCED KINASE 1 ( BIK1 ) are common signalling components in both pathways . Whereas BSK1 is a positive regulator for both signalling routes ( Tang et al . , 2008; Shi et al . , 2013 ) , BIK1 is a positive regulator for PTI responses ( Lu et al . , 2010; Zhang et al . , 2010 ) , but a negative regulator for BR signalling ( Lin et al . , 2013 ) . Even though FLS2- and BRI1-mediated signalling pathways have been extensively studied genetically and biochemically , little is known about how FLS2 and BRI1 are organised within the PM and how these two receptors fulfil their sensory activity at the cell periphery . In contrast to the original fluid mosaic model ( Singer and Nicolson , 1972 ) , which considered the PM as a two-dimensional liquid composed of a lipid bilayer that is interspersed by integral or associated proteins , it is nowadays accepted that the PM is a highly structured and dynamic cellular compartment organised at three hierarchic levels ( Kusumi et al . , 2011; Nicolson , 2014 ) . The first level of PM organisation is characterised by the interaction of the lipid bilayer with the underlying cortical cytoskeleton , the second level by protein-lipid interactions with the PM , and the third level of PM organisation is the result of protein-protein interactions that lead to formation of PM-associated or -integral homo- and hetero-oligomers ( Kusumi et al . , 2011 ) , e . g . FLS2- or BRI1-SERK3/BAK1 complexes . In plants , the cell wall has additional influence on the PM organisation and dynamics ( Martinière et al . , 2012 ) . As a consequence , lateral mobility and distribution of lipids and proteins within the PM is highly heterogeneous leading to the formation of dynamic protein clusters and PM sub-compartments with different shapes and sizes ( Jaqaman and Grinstein , 2012; Jarsch et al . , 2014; Ziomkiewicz et al . , 2015 ) . Each compartment or domain provides specific biophysical and biochemical environments for its residents and thus directly influences associated signalling activities ( Kusumi et al . , 2012; Saka et al . , 2014; Garcia-Parajo et al . , 2014; Tapken and Murphy , 2015 ) . With regard to PM signalling , specialised PM areas often referred to as PM nanodomains have attracted particular attention . For example , EPIDERMAL GROWTH FACTOR RECEPTOR ( EGFR ) , a mammalian RTK , forms clusters that co-localise with PM nanodomains , and EGFR cluster formation depends on the integrity of the PM lipid composition ( Gao et al . , 2015 ) . A recent report showed that BRI1 also localises to PM nano- or micro-domains in Arabidopsis roots and that partitioning of BRI1 into different PM microdomains is crucial for BR signalling ( Wang et al . , 2015 ) . Due to the numerous similarities between BRI1 and FLS2 signalling initiation , we were interested in elucidating the localisation patterns of these two LRR-RKs in a comparative fashion within the highly structured PM of plant cells . Here , we investigated the steady-state PM organisation of FLS2 and BRI1 receptors using live-cell imaging , single-particle tracking and quantitative co-localisation analysis in the PM of leaf epidermal cells , a cell file , in which both receptors are similarly expressed . Our results showed that FLS2 and BRI1 are heterogeneously distributed and that both receptors form transient receptor clusters . Although the spatial characteristics of FLS2 and BRI1 receptor clusters were similar , we observed differences in their dynamic behaviour with FLS2 clusters being more stable . Moreover , we detected only a limited overlap between the two receptor populations . This finding was confirmed by visualising FLS2 and BRI1 clusters within distinct remorin-labelled PM nanodomains . We additionally investigated the PM localisation patterns of BSK1 and BIK1 in complex with the two ligand-binding LRR-RKs . Imaging of FLS2 and BRI1 signalling complexes revealed a confined PM localisation and cluster formation . Importantly , BIK1 signalling complexes localised differentially , whereby BRI1-BIK1 , but not FLS2-BIK1 , complexes associated with cortical microtubules . Together , our data suggest that the distinct spatiotemporal localisation of FLS2 and BRI1 within specialised PM nanodomains may contribute to signalling specificity between immune and growth signalling mediated by these receptors .
To obtain an overview of the PM distribution of FLS2 and BRI1 , we investigated two stable transgenic Arabidopsis lines that express the C-terminally GFP-tagged receptors under their native promoters ( Göhre et al . , 2008; Geldner et al . , 2007 ) . Live-cell imaging using confocal laser scanning microscopy ( CLSM ) revealed that FLS2 and BRI1 were heterogeneously distributed within the PM ( Figure 1A and B ) . Both receptors formed dispersed punctate structures with increased fluorescence intensities . This local concentration of FLS2 and BRI1 subpopulations within the PM indicates the formation of receptor clusters , a phenomenon known from mammalian transmembrane receptors or helper proteins , such as EGFR ( Clayton et al . , 2007 ) or LINKER OF ACTIVATED T CELLS ( LAT ) ( Su et al . , 2016 ) . To emphasis our observation of receptor clusters , we processed the presented images using a spot-enhancing filter described by Sage et al . ( 2005 ) and applied the ‘fire’ lookup table ( Figure 1—figure supplement 1 ) . Importantly , we observed similar cluster formation of FLS2 and BRI1 whether upon native expression in Arabidopsis or after transient expression in Nicotiana benthamiana ( N . benthamiana ) ( Figure 1C and D ) . This feature enabled us to perform further co-localisation analysis in this heterologous system . 10 . 7554/eLife . 25114 . 003Figure 1 . FLS2 and BRI1 form receptor clusters within the plasma membrane . ( A , B ) Plasma membrane localisation of FLS2-GFP ( A ) and BRI1-GFP ( B ) in epidermal cells of Arabidopsis seedling cotyledons . ( C , D ) Plasma membrane localisation of FLS2-GFP ( C ) and BRI1-GFP ( D ) after transient expression in epidermal leaf cells of N . benthamiana . ( E ) Quantification of FLS2-GFP and BRI1-GFP plasma membrane receptor cluster diameters in epidermal cells of Arabidopsis cotyledons and after transient expression in N . benthamiana leaves . The coloured data points represent the technical replicates of 3 independent experiments . No statistical differences were observed based on two-tailed heteroscedastic t-tests and a Bonferroni multiple hypothesis correction . ( F ) Quantification of FLS2-GFP and BRI1-GFP plasma membrane receptor cluster densities in epidermal cells of Arabidopsis seedlings and after heterologous expression in N . benthamiana . The coloured data points represent the technical replicates of 3 independent experiments . No statistical differences were observed based on two-tailed heteroscedastic t-tests and a Bonferroni multiple hypothesis correction . The presented images were acquired using confocal laser scanning microscopy ( CLSM ) . Scale bars represent 5 µm . The colour bar represents the colour code for fluorescence intensities . Black dots represent outliers . Red arrowheads indicate endosomal compartments of BRI1-GFP . Endosomal compartments were omitted for quantitative analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 00310 . 7554/eLife . 25114 . 004Figure 1—figure supplement 1 . Illustration of the image processing steps for emphasising receptor cluster formation . ( A ) Raw data of a confocal micrograph showing the fluorescence intensity of FLS2-GFP in grey scale . ( B ) Confocal micrograph of FLS2-GFP fluorescence intensity as shown in ( A ) but using the ‘fire’ lookup table . ( C ) Confocal micrograph of FLS2-GFP fluorescence intensity as shown in ( A ) after applying the LoG3D plugin . ( D ) Confocal micrograph of FLS2-GFP fluorescence intensity as shown in ( C ) but using the ‘fire’ lookup table . To emphasise our observation of receptor clusters , we applied a spot-enhancing filter on the presented images and time series . The presented images were acquired using confocal laser scanning microscopy ( CLSM ) . The scale bar represents 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 00410 . 7554/eLife . 25114 . 005Figure 1—figure supplement 2 . Plasma membrane localisation of FLS2 and BRI1 with respect to the cytoskeleton . ( A I–A IV ) Confocal micrographs of actin filaments visualised using LifeAct-tRFP ( A I ) and FLS2-GFP ( A II ) in epidermal leaf cells of Arabidopsis seedling cotyledons as well as the merged image ( A III ) and a 3D surface plot of the raw data ( A IV ) . ( B I–B IV ) Confocal micrographs of actin filaments visualised using LifeAct-tRFP ( B I ) and FLS2-GFP ( B II ) after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( B III ) and a 3D surface plot of the raw data ( B IV ) . ( C I–C IV ) Confocal micrographs of actin filaments visualised using LifeAct-tRFP ( C I ) and BRI1-GFP ( C II ) in epidermal leaf cells of Arabidopsis seedling cotyledons as well as the merged image ( C III ) and a 3D surface plot of the raw data ( C IV ) . ( D I–D IV ) Confocal micrographs of actin filaments visualised using LifeAct-tRFP ( D I ) and BRI1-GFP ( D II ) after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( D III ) and a 3D surface plot of the raw data ( D IV ) . ( E I–E IV ) Confocal micrographs of TUB5-mCherry ( E I ) and FLS2-GFP ( E II ) in epidermal leaf cells of Arabidopsis seedling cotyledons as well as the merged image ( E III ) and a 3D surface plot of the raw data ( E IV ) . ( F I–F IV ) Confocal micrographs of TUA6-GFP ( F I ) and FLS2-mCherry ( F II ) after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( F III ) and a 3D surface plot of the raw data ( F IV ) . ( G I–G IV ) Confocal micrographs of TUB5-mCherry ( G I ) and BRI1-GFP ( G II ) in epidermal leaf cells of Arabidopsis seedling cotyledons as well as the merged image ( G III ) and a 3D surface plot of the raw data ( G IV ) . ( H I–H IV ) Confocal micrographs of TUA6-GFP ( H I ) and BRI1-mRFP ( H II ) after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( H III ) and a 3D surface plot of the raw data ( H IV ) . Fluorescence signals of cytoskeleton components are shown in magenta , fluorescence signals of FLS2 or BRI1 receptors are shown in green . The abbreviation tRFP stands for TagRFP . Scale bars represent 5 µm . The image series indicate that FLS2 and BRI1 receptor clusters repeatedly localised on top of actin filaments as indicated by white arrows in the respective 3D surface plots . In contrast , FLS2 and , to a minor extend , BRI1 receptors were largely excluded from plasma membrane areas that co-localised with cortical microtubules as indicated by white arrows in the respective 3D surface plots . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 005 Besides residing within the PM , BRI1-GFP additionally localised to mobile endomembrane compartments with high fluorescence intensities ( highlighted by red arrowheads in Figure 1B and D ) . These intracellular compartments most likely resembled trans-Golgi network/early endosomes and late endosomes/multi-vesicular bodies ( Geldner et al . , 2007; Irani et al . , 2012 ) . In contrast to the highly mobile endomembrane compartments , the visualised FLS2- and BRI1-GFP clusters appeared rather immobile and seemed to follow a certain spatial organisation . FLS2 and BRI1 clusters often aligned in pearl chain-like structures and interconnected lines with low fluorescence intensities frequently separated by PM areas containing several receptor clusters , in particular for FLS2-GFP ( Figure 1—figure supplement 2 ) . It is known that the cytoskeleton influences the hierarchic PM organisation ( Plowman et al . , 2005; Kusumi et al . , 2011 ) and therefore we visualised FLS2 and BRI1 in the presence of actin and microtubule markers . As shown in Figure 1—figure supplement 2 , pearl chain-like aligned FLS2 and BRI1 clusters often coincided with the underlying actin cytoskeleton , whereas the interconnected lines with low fluorescence intensities overlapped with cortical microtubule filaments , both in Arabidopsis and N . benthamiana . Next , we quantified the size and density of individual receptor clusters . The mean values ( ± standard deviation ) for FLS2 receptor cluster diameters were 356 ± 49 nm and 387 ± 55 nm in Arabidopsis and N . benthamiana , respectively , compared to 372 ± 38 nm in Arabidopsis and 392 ± 37 nm in N . benthamiana for BRI1 ( Figure 1E and Supplementary file 1 ) . Although there was a slight trend towards increased receptor cluster sizes in N . benthamiana , no statistical difference was observed . On average , we observed 2 . 21 ± 0 . 33 FLS2 clusters in Arabidopsis and 2 . 23 ± 0 . 26 in N . benthamiana per µm2 , whereas the densities of BRI1 clusters were 2 . 02 ± 0 . 27 and 2 . 43 ± 0 . 39 per µm2 in Arabidopsis and N . benthamiana , respectively ( Figure 1F and Supplementary file 1 ) . Thus , based on our confocal micrograph analysis , FLS2 and BRI1 showed similar spatial features . Both LRR-RKs were heterogeneously distributed within the PM and formed receptor clusters of comparable size and density . Only a difference with regard to the exclusion of receptor clusters from PM areas by cortical microtubules was observed , with FLS2 being more strongly affected . In addition to confocal microscopy , we applied variable angle epifluorescence microscopy ( VAEM ) to further characterise the dynamic behaviour of PM-localised FLS2 and BRI1 receptors . VAEM is a technique related to total internal reflection microscopy ( TIRF ) ( Vizcay-Barrena et al . , 2011 ) . In comparison to CLSM , VAEM has the advantages of improved z-resolution and fast image acquisition ( Vizcay-Barrena et al . , 2011 ) . Therefore VAEM is ideal for studying protein localisation and protein dynamics within or close to the PM ( Wan et al . , 2011; Vizcay-Barrena et al . , 2011 ) . As shown in Figure 2A–D , single frame VAEM images resembled our observations made using CLSM ( Figure 1A–D ) . We repeatedly observed dispersed FLS2- and BRI1-GFP clusters in the PM of Arabidopsis or N . benthamiana leaf epidermal cells . However , using VAEM we were able to acquire image time series ( Video 1 and 2 ) . Micrograph analysis using kymograph representations revealed increased stability and reduced mobility for FLS2-GFP compared to BRI1-GFP clusters , both in Arabidopsis and N . benthamiana ( Figure 2E ) . These observations were confirmed using single-particle tracking . BRI1-GFP showed a trend towards increased lateral cluster displacement ( Figure 2F ) and FLS2-GFP showed an increased population of long-lived receptor clusters ( Figure 2G ) . Additionally , we noticed that some FLS2 and BRI1 clusters suddenly appeared or disappeared from the PM . These observations most likely reflect constitutive exo- and endocytosis processes ( Beck et al . , 2012; Martins et al . , 2015; Wang et al . , 2015 ) . 10 . 7554/eLife . 25114 . 006Figure 2 . FLS2 receptor clusters are more stable than BRI1 clusters . ( A , B ) Plasma membrane localisation of FLS2-GFP ( A ) and BRI-GFP ( B ) in epidermal cells of Arabidopsis seedling cotyledons . ( C , D ) Plasma membrane localisation of FLS2-GFP ( C ) and BRI1-GFP ( D ) after transient expression in epidermal leaf cells of N . benthamiana . ( E ) Kymograph analysis of FLS2-GFP and BRI1-GFP plasma membrane receptor clusters in epidermal cells of Arabidopsis seedling cotyledons and after transient expression in N . benthamiana . Kymographs were obtained from VAEM time series with a temporal resolution of 0 . 5 s over 250 frames along the indicated arrows in micrographs ( A ) to ( D ) . ( F ) Quantification of FLS2-GFP and BRI1-GFP receptor cluster displacements in epidermal cells of Arabidopsis seedling cotyledons obtained from VAEM time series with a temporal resolution of 0 . 5 s over 250 frames . The coloured data points represent the technical replicates of 4 independent experiments . No statistical differences were observed based on two-tailed heteroscedastic t-tests and a Bonferroni multiple hypothesis correction . ( G ) Quantification of FLS2-GFP and BRI1-GFP receptor cluster lifetimes in epidermal cells of Arabidopsis seedling cotyledons obtained from VAEM time series with a temporal resolution of 0 . 5 s over 250 frames . The coloured data points represent the technical replicates of 4 independent experiments . The indicated p-values were obtained using a two-tailed heteroscedastic t-test and a Bonferroni multiple hypothesis correction . The presented images were acquired using variable angle epi-fluorescence microscopy ( VAEM ) . Scale bars represent 5 µm . The colour bar represents the colour code for fluorescence intensities . Black dots represent outliers . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 00610 . 7554/eLife . 25114 . 007Figure 2—figure supplement 1 . The stability of FLS2 clusters depends on the PM lipid composition . ( A , B ) VAEM micrograph of FLS2-GFP ( A ) and a kymograph of the corresponding VAEM time series ( B ) after mock treatment . ( C , D ) VAEM micrograph of BRI1-GFP ( C ) and a kymograph of the corresponding VAEM time series ( D ) after mock treatment . ( E , F ) VAEM micrograph of FLS2-GFP ( E ) and a kymograph of the corresponding VAEM time series ( F ) after methyl-β-cyclodextrin ( MβCD ) treatment . ( G , H ) VAEM micrograph of BRI1-GFP ( G ) and a kymograph of the corresponding VAEM time series ( H ) after MβCD treatment . Application of MβCD , which extracts sterols from the PM , destabilised FLS2-GFP and BRI1-GFP clusters as indicated by the kymographs in ( F ) and ( H ) . The VAEM micrograph and time series was acquired from a 5 day old Arabidopsis seedling cotyledon after 20 min incubation in 30 mM MβCD , dissolved in liquid MS medium , or mock solution ( liquid MS medium ) . The white arrows in ( A ) , ( C ) , ( E ) and ( G ) indicate the spatial dimensions of the kymographs . The scale bars represent 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 00710 . 7554/eLife . 25114 . 008Video 1 . Dynamics of FLS2 receptor clusters within the plasma membrane . The presented time series was acquired from an epidermal cell of an Arabidopsis seedling cotyledon expressing FLS2-GFP under its native promoter using variable angle epi-fluorescence microscopy ( VAEM ) . The acquisition time was 0 . 5 s per frame over 250 frames in total . The scale bar represents 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 00810 . 7554/eLife . 25114 . 009Video 2 . Dynamics of BRI1 receptor clusters within the plasma membrane . The presented time series was acquired from an epidermal cell of an Arabidopsis seedling cotyledon expressing BRI1-GFP under its native promoter using variable angle epi-fluorescence microscopy ( VAEM ) . The acquisition time was 0 . 5 s per frame over 250 frames in total . The scale bar represents 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 009 Taken together , we observed that FLS2 and BRI1 formed non-randomly distributed receptor clusters across the PM . These receptor clusters were comparable in size and density; however , showed differences with regard to dynamic features . Using our non-treated conditions , we visualised both inactive FLS2 populations and partially active BRI1 receptors given the presence of endogenous BRs . This experimental setup most likely best reflects the steady-state configuration of both receptors in a natural situation prior to pathogen attack . Still , we were also interested to test how ligand availability influences the dynamic behaviour of FLS2 and BRI1 receptor clusters . First , we deprived seedlings from endogenous BRs by cultivating them for 2 days in liquid medium containing 5 µM brassinazole ( BRZ ) , an inhibitor of brassinosteroid biosynthesis ( Asami et al . , 2000 ) . As shown in Figure 3 , depletion of endogenous BRs resulted in increased BRI1 cluster mobility ( Figure 3A ) . In contrast , the lateral displacement of FLS2 clusters was not affected ( Figure 3A ) . The determined cluster lifetimes for FLS2 and BRI1 showed a similar distribution as under non-treated conditions with a slight trend towards increased stability for BRI1 clusters upon BZR treatment ( Figure 3B ) . Subsequently , we activated BRI1-mediated signalling by exogenous application of 100 nM 24-epi-brassinolide ( BL ) . In analogy to BRZ-treatment , BL had no effect on FLS2 clusters but decreased the lateral displacement of BRI1 receptor clusters within a time frame of 30 min ( Figure 3C–H ) . 10 . 7554/eLife . 25114 . 010Figure 3 . BRs reduce BRI1 cluster displacement within the plasma membrane . ( A ) Quantification of FLS2-GFP and BRI1-GFP receptor cluster displacements in epidermal cells of Arabidopsis seedling cotyledons after 2 days in liquid medium containing 5 µM BRZ . ( B ) Quantification of FLS2-GFP and BRI1-GFP receptor cluster lifetimes in epidermal cells of Arabidopsis seedling cotyledons after 2 days in liquid medium containing 5 µM BRZ . ( C ) Time-dependent quantification of short-range FLS2-GFP and BRI1-GFP receptor cluster displacements in epidermal cells of Arabidopsis seedling cotyledons after BRZ-treatment and subsequent application of 100 nM BL . ( D ) Time-dependent quantification of short FLS2-GFP and BRI1-GFP receptor cluster lifetimes in epidermal cells of Arabidopsis seedling cotyledons after BRZ-treatment and subsequent application of 100 nM BL . ( E ) Time-dependent quantification of medium-range FLS2-GFP and BRI1-GFP receptor cluster displacements in epidermal cells of Arabidopsis seedling cotyledons after BRZ-treatment and subsequent application of 100 nM BL . ( F ) Time-dependent quantification of medium FLS2-GFP and BRI1-GFP receptor cluster lifetimes in epidermal cells of Arabidopsis seedling cotyledons after BRZ-treatment and subsequent application of 100 nM BL . ( G ) Time-dependent quantification of long-range FLS2-GFP and BRI1-GFP receptor cluster displacements in epidermal cells of Arabidopsis seedling cotyledons after BRZ-treatment and subsequent application of 100 nM BL . ( H ) Time-dependent quantification of long FLS2-GFP and BRI1-GFP receptor cluster lifetimes in epidermal cells of Arabidopsis seedling cotyledons after BRZ-treatment and subsequent application of 100 nM BL . The presented data points were obtained from VAEM time series with a temporal resolution of 0 . 5 s over 250 frames . The coloured data points represent the technical replicates of 3 independent experiments . The indicated p-values were obtained using a one-tailed heteroscedastic t-test and a Bonferroni multiple hypothesis correction . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 010 Accordingly , we observed a reduction in lateral cluster displacement for FLS2 receptors after seedlings were exposed to 100 nM flg22 , whereas BRI1 was unaffected ( Figure 4 ) . Again , there was no quantitative effect on the receptor cluster lifetimes of both receptors ( Figure 4 ) . 10 . 7554/eLife . 25114 . 011Figure 4 . Activation of FLS2 results in reduced lateral receptor cluster displacement . ( A ) Time-dependent quantification of short-range FLS2-GFP and BRI1-GFP receptor cluster displacements in epidermal cells of Arabidopsis seedling cotyledons after application of 100 nM flg22 . ( B ) Time-dependent quantification of short FLS2-GFP and BRI1-GFP receptor cluster lifetimes in epidermal cells of Arabidopsis seedling cotyledons after application of 100 nM flg22 . ( C ) Time-dependent quantification of medium-range FLS2-GFP and BRI1-GFP receptor cluster displacements in epidermal cells of Arabidopsis seedling cotyledons after application of 100 nM flg22 . ( D ) Time-dependent quantification of medium FLS2-GFP and BRI1-GFP receptor cluster lifetimes in epidermal cells of Arabidopsis seedling cotyledons after application of 100 nM flg22 . ( E ) Time-dependent quantification of long-range FLS2-GFP and BRI1-GFP receptor cluster displacements in epidermal cells of Arabidopsis seedling cotyledons after application of 100 nM flg22 . ( F ) Time-dependent quantification of long FLS2-GFP and BRI1-GFP receptor cluster lifetimes in epidermal cells of Arabidopsis seedling cotyledons after application of 100 nM flg22 . The presented data points were obtained from VAEM time series with a temporal resolution of 0 . 5 s over 250 frames . The coloured data points represent the technical replicates of 3 independent experiments . The indicated p-values were obtained using a one-tailed heteroscedastic t-test and a Bonferroni multiple hypothesis correction . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 011 Our findings that receptor activation leads to reduced lateral mobility within the PM are contrary to Wang et al . ( 2015 ) , who reported increased BRI1 diffusion after ligand application in Arabidopsis roots , but in line with observations made for the plant receptors FLS2 and LYK3 ( Ali et al . , 2007; Haney et al . , 2011 ) as well as the mammalian cell surface receptor EGFR ( Low-Nam et al . , 2011 ) . Thus , FLS2 and BRI1 exhibited comparable behaviour in response to their respective ligands , nonetheless , the dynamic features of both cluster populations still differed from each other . To address whether FLS2 and BRI1 clusters coincide or are spatially separated within the PM , we performed co-localisation studies . As positive control , we first determined the overlap of two differently tagged FLS2 receptor populations . We co-expressed FLS2-GFP and FLS2-mCherry ( Mbengue et al . , 2016 ) in leaf epidermal cells of N . benthamiana and , as shown in Figure 5A–D , both fluorescently tagged FLS2 populations showed similar PM localisation patterns and also co-localised ( Figure 5C and D ) . Based on quantitative co-localisation analysis , we determined moderate to high Pearson correlation coefficients for FLS2-GFP and FLS2-mCherry fluorescence signals ( Figure 5I and Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 25114 . 012Figure 5 . FLS2 and BRI1 show distinct plasma membrane localisation patterns . ( A–D ) Confocal micrographs of FLS2-GFP ( A ) and FLS2-mCherry ( B ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( C ) and an image inset ( D ) . ( E–H ) Confocal micrographs of BRI1-GFP ( E ) and FLS2-mCherry ( F ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( G ) and an image inset ( H ) . ( I ) Quantitative co-localisation analysis for FLS2-GFP or BRI1-GFP , respectively , with FLS2-mCherry after transient co-expression in epidermal leaf cells of N . benthamiana . The coloured data points represent the technical replicates of 6 independent experiments . The indicated p-values were obtained using a two-tailed heteroscedastic t-test and a Bonferroni multiple hypothesis correction . ( J ) Quantitative co-localisation analysis for FLS2-GFP or BRI1-GFP , respectively , with FLS2-mCherry after transient co-expression in epidermal leaf cells of N . benthamiana and BRZ-treatment . The coloured data points represent the technical replicates of 2 independent experiments . The indicated p-values were obtained using a two-tailed heteroscedastic t-test and a Bonferroni multiple hypothesis correction . The presented images were acquired using confocal laser scanning microscopy ( CLSM ) . Scale bars in ( C ) and ( G ) represent 5 µm , scale bars in ( D ) and ( H ) represent 2 µm . The areas that correspond to the images ( D ) and ( H ) are indicated by the dashed squares in images ( C ) and ( G ) . Red arrowheads indicate endosomal compartments of BRI1-GFP . Endosomal compartments were omitted for quantitative analysis . The colour bar represents the colour code for fluorescence intensities . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 01210 . 7554/eLife . 25114 . 013Figure 5—figure supplement 1 . Control experiments for verifying the specific localisation patterns of FLS2 and BRI1 . ( A ) Quantitative co-localisation analysis for FLS2-GFP or BRI1-GFP , respectively , with FLS2-mCherry after transient co-expression in epidermal leaf cells of N . benthamiana . ( B ) Quantitative co-localisation analysis for FLS2-GFP or BRI1-GFP , respectively , with FLS2-mCherry after transient co-expression in epidermal leaf cells of N . benthamiana and BRZ-treatment . The coloured data points indicate the values of technical replicates; black dots indicate the position of outliers . To assess whether the determined co-localisation values ( original or ori . ) were significant , co-localisation analysis was also carried out after image randomisation . For this purpose , one of the two image channels was rotated by 90 degrees prior to co-localisation analysis ( rotated or rot . ) . In addition , the co-localisation values of a reciprocal experiment are shown . Here , co-localisation of FLS2-GFP or BRI1-GFP was determined with regard to BRI1-mRFP . As emphasised by the represented p-values , we observed a specific co-localisation for FLS2-GFP and FLS2-mCherry . In contrast , the co-localisation of BRI1-GFP with FLS2-mCherry was not different from randomised images , and thus non-specific . Similar observations were made after BRZ-treatment for 2 days to deplete endogenous BRs as well as for reciprocal experiments using BRI1-mRFP as reference . The indicated p-values were obtained using a two-tailed heteroscedastic t-test and a Bonferroni multiple hypothesis correction . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 013 Subsequently , we compared the distributions of BRI1-GFP and FLS2-mCherry receptors ( Figure 5E–H ) . Quantitative image analysis of the obtained image series indicated a strongly reduced co-localisation between BRI1-GFP and FLS2-mCherry when compared to the FLS2-GFP/FLS2-mCherry combination ( Figure 5A–D ) . As shown in Figure 5I , we determined Pearson correlation coefficients of around zero , which represents non-correlated localisation or no co-localisation ( McDonald and Dunn , 2013 ) . Based on our previous findings , which indicated increased lateral BRI1 mobility by depletion of endogenous BRs , we also analysed the co-localisation of both receptors after BRZ-treatment . However , we observed the same co-localisation pattern as under non-treated steady-state conditions ( Figure 5J ) , suggesting a ligand-independent spatial separation of FLS2 and BRI1 receptors . These results were confirmed using image randomisation and by using BRI1-mRFP as reference ( Figure 5—figure supplement 1 and Supplementary file 1 ) . Consequently , our quantitative co-localisation analysis revealed distinct immune and growth receptor clusters within the PM of leaf epidermal cells . To obtain a more dynamic view on the co-localisation or spatial separation between the FLS2 and BRI1 receptor populations , we additionally applied dual-colour VAEM on leaf epidermal cells that co-expressed BRI1-GFP and FLS2-mCherry ( Video 3 and 4 , Figure 6 ) . We hardly observed overlap between the two LRR-RKs as indicated by the kymograph representation in Figure 6H . 10 . 7554/eLife . 25114 . 014Video 3 . Visualization of FLS2 receptor cluster dynamics within the plasma membrane . The presented time series was acquired from an epidermal leaf cell after transient co-expression of FLS2-GFP and FLS2-mCherry in N . benthamiana using variable angle epi-fluorescence microscopy ( VAEM ) . The acquisition time was 0 . 25 s per frame per channel over 200 frames in total . FLS2-GFP fluorescence is shown in green , FLS2-mCherry fluorescence is shown in magenta . The scale bar represents 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 01410 . 7554/eLife . 25114 . 015Video 4 . Simultaneous visualization of FLS2 and BRI1 receptor cluster dynamics within the plasma membrane . The presented time series was acquired from an epidermal leaf cell after transient co-expression of BRI1-GFP and FLS2-mCherry in N . benthamiana using variable angle epi-fluorescence microscopy ( VAEM ) . The acquisition time was 0 . 25 s per frame per channel over 200 frames in total . BRI1-GFP fluorescence is shown in green , FLS2-mCherry fluorescence is shown in magenta . The scale bar represents 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 01510 . 7554/eLife . 25114 . 016Figure 6 . FLS2 and BRI1 clusters are spatiotemporally separated . ( A–C ) Plasma membrane localisation of FLS2-GFP ( A ) and FLS2-mCherry ( B ) after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( C ) . ( D ) Kymograph analysis of VAEM micrograph shown in ( C ) . The spatial dimension of the kymograph is indicated by the white arrow in ( C ) . The acquisition time of a single channel was 0 . 25 s . For each channel 200 frames were collected . ( E–G ) Plasma membrane localisation of BRI1-GFP ( E ) and FLS2-mCherry ( F ) after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( G ) . ( H ) Kymograph analysis of VAEM micrograph shown in ( G ) . The spatial dimension of the kymograph is indicated by the white arrow in ( C ) . The acquisition time of a single channel was 0 . 25 s . For each channel 200 frames were collected . The presented images were acquired using variable angle epi-fluorescence microscopy ( VAEM ) . In the merged images FLS2-GFP or BRI1-GFP signals are shown in green and FLS2-mCherry signals are shown in magenta . The scale bar represents 5 µm . The colour bar represents the colour code for fluorescence intensities . Two independent experiments with similar results were performed . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 016 Collectively , our co-localisation analysis revealed that the vast majority of the two LRR-RKs formed distinct receptor clusters that were spatiotemporally separated . So far , we provided evidence for the formation of distinct FLS2 and BRI1 clusters that were heterogeneously distributed within the PM of leaf epidermal cells . Considering the proposed hierarchic organisation of the PM ( Kusumi et al . , 2011 ) , one could assume that these receptor clusters may reside within PM nanodomains . In fact , this was recently shown for BRI1 in root epidermal cells by co-localising it with the PM nanodomain marker AtFLOT1 ( Wang et al . , 2015 ) . Interestingly , proteomic studies identified two remorin ( REM ) proteins , REM1 . 2 and REM1 . 3 , that were enriched in detergent-resistant membranes upon flg22 application ( Keinath et al . , 2010 ) . REMs form a large plant-specific protein family with characteristic PM nanodomain localisation patterns ( Jarsch et al . , 2014 ) . Moreover , different members of this family have previously been associated with plant-microbe interactions ( Raffaele et al . , 2009; Lefebvre et al . , 2010; Tóth et al . , 2012; Perraki et al . , 2012 , 2014; Bozkurt et al . , 2014 ) and with SERK-dependent processes ( Gui et al . , 2016 ) . Based on the findings of Keinath et al . ( 2010 ) , we investigated the spatial relationship between the two LRR-RKs and REM1 . 2 in leaf epidermal cells . As shown in Figure 7A I–A III and I , we identified a positive correlation for FLS2-GFP and mRFP-REM1 . 2 fluorescence intensities . In contrast , the co-localisation of this PM nanodomain marker with BRI1 was unspecific ( Figure 7B I−B III and I , Figure 7—figure supplement 1 ) . Similar results were obtained using mRFP-REM1 . 3 as PM nanodomain marker ( Figure 7C I–D III and I , Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 25114 . 017Figure 7 . FLS2 and BRI1 co-localize differentially with remorin markers . ( A I–A III ) Confocal micrographs of mRFP-REM1 . 2 ( A I ) and FLS2-GFP ( A II ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( A III ) . ( B I–B III ) Confocal micrographs of mRFP-REM1 . 2 ( B I ) and BRI1-GFP ( B II ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( B III ) . ( C I–C III ) Confocal micrographs of mRFP-REM1 . 3 ( C I ) and FLS2-GFP ( C II ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( C III ) . ( D I–D III ) Confocal micrographs of mRFP-REM1 . 3 ( D I ) and BRI1-GFP ( D II ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( D III ) . ( E I–E III ) Confocal micrographs of mRFP-REM6 . 1 ( E I ) and FLS2-GFP ( E II ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( E III ) . ( F I–F III ) Confocal micrographs of mRFP-REM6 . 1 ( F I ) and BRI1-GFP ( F II ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( F III ) . ( G I–G III ) Confocal micrographs of mRFP-REM6 . 2 ( G I ) and FLS2-GFP ( G II ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( G III ) . ( H I–H III ) Confocal micrographs of mRFP-REM6 . 2 ( H I ) and BRI1-GFP ( H II ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( H III ) . ( I ) Quantitative co-localisation analysis for FLS2-GFP and BRI1-GFP with mRFP-REM1 . 2 , mRFP-REM1 . 3 , mRFP-REM6 . 1 , and mRFP-REM6 . 2 , respectively . The coloured data points represent the technical replicates of 3 independent experiments . The indicated p-values were obtained using a two-tailed heteroscedastic t-test and a Bonferroni multiple hypothesis correction . In the merged images FLS2-GFP or BRI1-GFP signals are shown in green and REM signals are shown in magenta . Scale bars represent 2 µm . The colour bar represents the colour code for fluorescence intensities . Black dots represent outliers . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 01710 . 7554/eLife . 25114 . 018Figure 7—figure supplement 1 . Control experiments for verifying the specific co-localisation of FLS2 and BRI1 with remorin markers . The coloured data points indicate the values of technical replicates; black dots indicate the position of outliers . To assess whether the determined co-localisation values ( orig . ) were significant , co-localisation analysis was also carried out after image randomisation . For this purpose , one of the two image channels was rotated by 90 degrees prior to co-localisation analysis ( rota . ) . As shown by the represented p-values , FLS2-GFP ( FLS2 ) showed to a varying degree specific co-localisation with the four tested mRFP-tagged remorin markers . In contrast , BRI1-GFP ( BRI1 ) only specifically co-localised with mRFP-REM6 . 1 ( REM6 . 1 ) and mRFP-REM6 . 2 ( REM6 . 2 ) . The indicated p-values were obtained using a two-tailed heteroscedastic t-test and a Bonferroni multiple hypothesis correction . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 018 Besides REM1 . 2 and REM1 . 3 , Benshop et al . ( 2007 ) identified REM6 . 1 and REM6 . 2 as flg22-induced phosphoproteins . Therefore , we investigated also the co-localisation of FLS2 and BRI1 with these two REM proteins . The results are shown in Figure 7E I to Figure 7H III . Similar to REM1 . 2 and REM1 . 3 , we revealed a positive correlation between FLS2-GFP and mRFP-REM6 . 1 protein populations . However , REM6 . 1-labelled PM nanodomains also harboured a considerable amount of BRI1 receptors ( Figure 7F III and I , Figure 7—figure supplement 1 ) . For mRFP-REM6 . 2 , we observed unexpectedly the opposite behaviour ( Figure 7G I–H III ) . REM6 . 2-positive PM nanodomains contained a significantly elevated amount of BRI1 compared to FLS2 receptors . The results of quantitative co-localisation analysis are summarised in Figure 7I . Taken together , these findings demonstrated that the heterogeneously distributed FLS2 and BRI1 receptor clusters are indeed residing within PM nanodomains of leaf epidermal cells . The differential co-localisation of the two LRR-RKs with regard to the tested REM marker proteins further emphasised the spatial separation and distinct localisation of FLS2 and BRI1 receptors . Our cell biological study indicated a spatial separation between immune and growth receptors in steady-state conditions . Though , genetically and biochemically there exist apparent connections between FLS2- and BRI1-mediated signalling pathways , although many of these interconnections seem to occur at the transcriptional level ( Albrecht et al . , 2012; Belkhadir et al . , 2012 , 2014; Lozano-Durán et al . , 2013; Fan et al . , 2014; Malinovsky et al . , 2014; Lozano-Durán and Zipfel , 2015; Jiménez-Góngora et al . , 2015 ) . Nevertheless , both receptors depend on additional PM-localised or PM-associated signalling components for relaying the information of ligand binding to the extracellular LRRs domains across the PM and into the cell interior . Therefore we assume that a signalling competent unit contains at least one ligand-binding receptor , one ( or several ) co-receptor ( s ) , and one ( or several ) RLCK ( s ) . Intriguingly , FLS2 and BRI1 employ , at least from a genetic perspective , the same components; SERK co-receptors ( Nam and Li , 2002; Li et al . , 2002; Chinchilla et al . , 2007; Heese et al . , 2007; Roux et al . , 2011; Gou et al . , 2012 ) , and the RLCKs BSK1 and BIK1 ( Tang et al . , 2008; Shi et al . , 2013; Lu et al . , 2010; Zhang et al . , 2010; Lin et al . , 2013 ) . To investigate the spatial organisation of FLS2 and BRI1 signalling units , we made use of bimolecular fluorescence complementation ( BiFC ) . Since epitope-tagging of BAK1/SERK3 ( and potentially other SERKs ) compromises its function in immune signalling ( Ntoukakis et al . , 2011 ) we decided to omit these co-receptors for our study . Instead , we visualised FLS2 and BRI1 in complex with BSK1 or BIK1 using CLSM . As shown in Figure 8 and Figure 9 , in addition to the ligand binding receptors , the two RLCKs also appeared heterogeneously distributed , and BSK1 and BIK1 clusters became evident . 10 . 7554/eLife . 25114 . 019Figure 8 . FLS2 and BRI1 signaling complexes also undergo cluster formation within the plasma membrane . ( A–D ) Confocal micrographs of BSK1-CFP ( A ) , BSK1-nYFP/BRI1-cYFP ( BiFC ) ( B ) , and BRI1-mRFP ( C ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( D ) . ( E–H ) Confocal micrographs of BSK1-CFP ( E ) , BSK1-nYFP/FLS2-cYFP ( BiFC ) ( F ) , and FLS2-mCherry ( G ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( H ) . ( I ) Quantitative co-localisation analysis for the reconstituted YFP fluorescence intensities ( BiFC ) with BSK1-CFP ( BSK1 ) as well as BRI1-mRFP ( BRI1 ) or FLS2-mCherry ( FLS2 ) and the quantified co-localisation between BSK1-CFP with BRI1-mRFP or FLS2-mCherry . The coloured data points represent the technical replicates of 3 independent experiments . No statistical differences were observed based on two-tailed heteroscedastic t-tests and a Bonferroni multiple hypothesis correction . BiFC stands for bimolecular fluorescence complementation and the labelled image panels show YFP fluorescence signals for the respective protein complexes . Yc and Yn indicate the C- and N-terminal fragments of the split YFP fluorophore , respectively . Scale bars represent 5 µm . Black dots represent outliers . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 01910 . 7554/eLife . 25114 . 020Figure 9 . BRI1-BIK1 , but not FLS2-BIK1 , complexes associate with cortical microtubules . ( A–D ) Confocal micrographs of BIK1-CFP ( A ) , BIK1-nYFP/BRI1-cYFP ( BiFC ) ( B ) , and BRI1-mRFP ( C ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( D ) . ( E–H ) Confocal micrographs of BIK1-CFP ( E ) , BIK1-nYFP/FLS2-cYFP ( BiFC ) ( F ) , and FLS2-mCherry ( G ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( H ) . ( I–K ) Confocal micrographs of BIK1-GFP ( I ) and LifeAct-tRFP ( J ) fluorescence intensities after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( K ) . ( L–N ) Confocal micrographs of BIK1-mRFP ( L ) and TUB5-GFP ( M ) fluorescence intensities after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( N ) . ( O ) Quantitative co-localisation analysis for the reconstituted YFP fluorescence intensities ( BiFC ) with BIK1-CFP ( BIK1 ) as well as BRI1-mRFP ( BRI1 ) or FLS2-mCherry ( FLS2 ) and the quantified co-localisation between BIK1-CFP with BRI1-mRFP or FLS2-mCherry . The coloured data points represent the mean values of 3 independent experiments . No statistical differences were observed based on two-tailed heteroscedastic t-tests and a Bonferroni multiple hypothesis correction . BiFC stands for bimolecular fluorescence complementation and the labelled image panels show YFP fluorescence signals for the respective protein complexes . Yc and Yn indicate the C- and N-terminal fragments of the split YFP fluorophore , respectively . LifeAct-tRFP was employed to visualise actin filaments , whereby tRFP stands for TagRFP . Scale bars represent 5 µm . Black dots represent outliers . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 02010 . 7554/eLife . 25114 . 021Figure 9—figure supplement 1 . BIK1 and BRI1-BIK1 complexes associate with cortical microtubules . ( A–D ) Confocal micrographs of BIK1-CFP ( A ) , BIK1-nYFP/BRI1-cYFP ( B ) , and BRI1-mRFP ( C ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( D ) . ( E–H ) Confocal micrographs of BIK1-CFP ( E ) , BIK1-nYFP/BRI1-cYFP ( F ) , and BRI1-mRFP ( G ) plasma membrane localisation after transient co-expression in epidermal leaf cells of N . benthamiana as well as the merged image ( H ) . ( I–K ) Confocal micrographs of BIK1-mRFP ( I ) and TUA6-GFP ( J ) after transient co-expression in epidermal leaf cells of N . benthamiana together with the merged image ( K ) . The illustrated images show the different degrees of cortical microtubule association of BIK1 and BIK1 complexes . BiFC stands for bimolecular fluorescence complementation and the labelled image panels show YFP fluorescence signals for the respective protein complexes . Scale bars represent 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 25114 . 021 In agreement with the localisation patterns of BSK1 and BIK1 alone , also BiFC complexes formed clusters within the PM ( Figure 8B and F , Figure 9B and F ) . Quantitative image analysis and visual image inspection indicated independent cluster populations but similar co-localisation behaviour for BSK1 in complex with FLS2 and BRI1 ( Figure 8I ) . Thus , we did not reveal major differences for FLS2 and BRI1 signalling complexes at this stage . This situation changed dramatically when we focused our attention on BIK1 complexes . Although quantitative image analysis did not indicate differences between FLS2- and BRI1-BIK1 complexes ( Figure 9O ) , localisation patterns were entirely different for immune and growth signalling complexes . As shown in Figure 9 , BRI1-BIK1 , but not FLS2-BIK1 , complexes co-localised with microtubule-associated BIK1 populations ( Figure 9—figure supplement 1 ) . The finding that signalling complexes also localised differentially further underlined our notion of spatial separation for the two LRR-RKs within the PM of plant cells , in addition to the aforementioned differences in PM localisation for FLS2 and BRI1 themselves . In conclusion , using various different imaging approaches we were able to discriminate specific PM localisation patterns for the PRR FLS2 and the hormone receptor BRI1 . We showed that both ligand-binding receptors are heterogeneously distributed due to the formation of transient receptor clusters . We observed that FLS2 and BRI1 clusters undergo different dynamics but that ligand availability reduces lateral cluster mobility in both cases . Moreover , we could visualise receptor clusters for both LRR-RKs within distinct PM nanodomains and differential PM localisation for steady-state FLS2- and BRI1 signalling complexes . Taken together , our findings emphasise a model of signalling pathway-specific pools of downstream components and suggest that spatial separation of immune and growth signalling platforms may contribute to the generation of signalling specificity .
Plants and animals employ cell surface receptors for the perception of various extracellular signals like hormones or PAMPs . In Arabidopsis , the immune receptor FLS2 and the growth receptor BRI1 represent two of the best-studied transmembrane receptors . Genetic and biochemical studies have linked extracellular ligand binding to the generation of specific flg22- or BR-triggered physiological responses . However , little is known about the organisation of these two receptors within the PM , the compartment where both LRR-RKs perceive their ligands and initiate their signalling cascades . Here , we provide evidence that FLS2 and BRI1 are heterogeneously distributed within the PM by formation of distinct PM nanodomain-localised receptor clusters . In addition , we show that the RLCKs BSK1 and BIK1 also form PM clusters , and that FLS2- and BRI1-BIK1 complexes localise differentially . Recently , Somssich et al . ( 2015 ) reported a comparative study that investigated PM-associated signalling principles of plant RKs . Using the flg22- and CLAVATA3 ( CLV3 ) -triggered pathways as examples for plant immune and growth signalling , they addressed the receptor/co-receptor association prior and during signalling initiation . Similar to previous studies ( Chinchilla et al . , 2007; Heese et al . , 2007; Schulze et al . , 2010; Roux et al . , 2011; Sun et al . , 2013 ) , Somssich et al . ( 2015 ) revealed that the initiation of FLS2-mediated signalling strictly relies on ligand-induced complex formation between FLS2 and SERKs . In contrast , CLV-mediated signalling was found to depend on preformed receptor/co-receptor complexes ( Somssich et al . , 2015 ) , in line with findings for the BRI1-BAK1 association in Arabidopsis roots ( Bücherl et al . , 2013 ) . Interestingly , the association between receptors and co-receptors , but also BAK1 and its negative regulator BIR2 ( Halter et al . , 2014 ) , occurs in clusters or subdomains of the PM , further emphasising our observations presented here . Formation of PM receptor clusters is a well-known phenomenon for mammalian receptors like T-cell receptors ( TCRs ) or EGFR ( Dinic et al . , 2015; Gao et al . , 2015; Su et al . , 2016 ) . The RTK EGFR constitutively undergoes clustering and EGFR clusters play important roles in cell signalling ( Clayton et al . , 2007; Gao et al . , 2015; Paviolo et al . , 2015 ) . Using super-resolution microscopy , Gao et al . ( 2015 ) showed that EGFR clusters localise to PM nanodomains and that cluster formation relies on the integrity of the PM lipidome since sterol depletion of the PM using methyl-β-cyclodextrin ( MβCD ) disrupts clustering . Similar observations have recently been made for BRI1 in Arabidopsis roots ( Wang et al . , 2015 ) . Wang et al . ( 2015 ) observed that BRI1 localises to FLOT1- and clathrin heavy chain ( CHC ) -labelled PM nanodomains and that partitioning of BRI1 into PM nanodomains as well as the PM lipidome are crucial for BR responses . These findings are in line with our observations for FLS2 and BRI1 in epidermal leaf cells since MβCD treatment also affected FLS2 clusters ( Figure 2—figure supplement 1 ) . Thus , both LRR-RKs constitutively formed PM receptor clusters that were influenced by the PM lipid-composition and localised to PM nanodomains labelled by distinct REM protein markers . However , the differential co-localisation to specific PM nanodomains allowed us to discriminate FLS2 and BRI1 receptor clusters , thus indicating their spatial separation within the PM . Clustering of PM receptors is thought to provide a mechanism for modulating intermolecular interactions and for fine-tuning signal transduction ( Bray et al . , 1998; Abulrob et al . , 2010 ) . For example , Hsieh et al . , 2010 reported that EGFR clustering enhances the recruitment of downstream signalling components . Interestingly , we not only observed cluster formation for FLS2 and BRI1 , but also for their downstream signalling components BSK1 and BIK1 , as well as for the respective RK-RLCK complexes . We propose that similar mechanisms as described for EGFR may also account for FLS2- and BRI1-mediated signalling , and , furthermore , that clustering of PM proteins is a more general organising principle for plant PM proteins ( Kleine-Vehn et al . , 2011; Wang et al . , 2013; Demir et al . , 2013; Jarsch et al . , 2014 ) . In fact , in animal cells , most lipids and proteins are heterogeneously distributed across the PM due to intermolecular interactions that generate inhomogeneities of varying size and stability ( Jaqaman and Grinstein , 2012; Saka et al . , 2014 ) . Besides molecular interactions among PM constituents , the cortical cytoskeleton influences the formation and/or stability of PM nanodomains in animals and plants ( Plowman et al . , 2005; Chichili and Rodgers , 2009; Jaumouillé et al . , 2014; Dinic et al . , 2013; Szymanski et al . , 2015; Su et al . , 2016 ) . Intriguingly , we repeatedly observed close association between FLS2 and BRI1 clusters with actin filaments . Though , it is currently unclear how or whether actin contributes to the formation and/or stability or FLS2 and BRI1 clusters . It was however shown previously that actin-myosin function is required for flg22-induced endocytosis ( Beck et al . , 2012 ) . In addition to a contribution to PM nanodomain formation , cortical cytoskeleton components also affect lateral mobility of PM proteins in animal cells ( Chichili and Rodgers , 2009; Jaqaman and Grinstein , 2012 ) . However , in plant cells , it seems that the cell wall is mainly responsible for restricting movements of PM proteins within the lipid bilayer ( Martinière et al . , 2012 ) . Similar to the findings of Martinière et al . ( 2012 ) , we observed very limited dynamics of FLS2 and BRI1 clusters within the PM . The study of Jarsch et al . ( 2014 ) , which described various different localisation patterns for the REM protein family in plant PMs , revealed that the clusters of these PM-associated proteins also hardly undergo lateral movements . Since REMs bind to the inner leaflet of PMs ( Konrad et al . , 2014 ) and therefore cannot directly interact with the extracellular cell wall , a restrictive influence of the cortical cytoskeleton on the lateral PM mobility should not be excluded . Our quantification of FLS2 and BRI1 cluster densities and sizes yielded similar values as described by Jarsch et al . ( 2014 ) for REM nanodomains . With around two clusters per µm2 , the detected cluster densities were slightly higher than determined for the REM nanodomain markers , which range from 0 . 1 to 1 per µm2 ( Jarsch et al . , 2014 ) . Although we revealed cluster diameters in the range of approximately 250–500 nm that are again in line with the dimensions of REM PM nanodomains ( Jarsch et al . , 2014 ) , we assume that the actual size of FLS2 and BRI1 clusters is smaller . Similar to our observations , Demir et al . ( 2013 ) reported PM nanodomain sizes of ca . 250 nm for the potato REM1 . 3 when visualized by CLSM . However , subsequent analysis of these protein clusters using the super-resolution imaging method STED ( stimulated emission depletion ) led to more confined dimensions of around 100 nm ( Demir et al . , 2013 ) . Even though the spatial features of FLS2 and BRI1 clusters were comparable and similar to REM proteins , we observed major differences for the dynamics of protein clusters , both among the transmembrane receptors and with regard to REM proteins . In contrast to the high stability of REM nanodomains ( Jarsch et al . , 2014 ) , the lifetimes of FLS2 and BRI1 clusters were much shorter and they exhibited a more dynamic behaviour . Furthermore , comparison between FLS2 and BRI1 populations revealed that the PRR clusters were characterized by increased stability under steady-state conditions . The low lateral mobility of the two LRR-RK cluster populations and the observation of subpopulations with higher displacement values are in line with the recent report of Wang et al . ( 2015 ) . They described two subpopulations for BRI1 with low and high lateral mobility , respectively , in the PM of epidermal root cells . Under steady-state conditions , the majority of BRI1 receptors underwent only short-range movements ( Wang et al . , 2015 ) , similar to our results for epidermal leaf cells . In stark contrast to Wang et al . ( 2015 ) are our observations of reduced cluster mobility for FLS2 and BRI1 receptors in the presence of their respective ligands . However , our observations are in line with a previous report about FLS2 in Arabidopsis protoplasts ( Ali et al . , 2007 ) and with findings for the mammalian PM receptor EGFR ( Low-Nam et al . , 2011 ) . Thus , increased receptor confinement in response to ligand binding may be a more general phenomenon for cell surface receptors . A plausible explanation for reduced lateral mobility is the well-established formation or stabilisation of receptor ( hetero- ) oligomers within the PM for activation of signal transduction as previously reported for FLS2 and BRI1 ( Chinchilla et al . , 2007; Somssich et al . , 2015; Wang et al . , 2008; Bücherl et al . , 2013 ) . The low lateral mobility of PM proteins in general and the additional confinement in response to ligands also suggests a ligand-independent pre-organisation of receptors and/or signalling units ( Martinière et al . , 2012; Abulrob et al . , 2010; Sandor et al . , 2016 ) . Wang et al . ( 2015 ) also showed that BRI1 receptors co-localise with the PM nanodomain marker FLOT1 . Since proteomic studies previously revealed that REM1 . 2 and REM1 . 3 are phosphorylated and enriched in detergent-resistance membranes in a flg22-dependent manner ( Benschop et al . , 2007; Keinath et al . , 2010 ) , we used several REM proteins as references for investigating the localisation of FLS2 and BRI1 to PM nanodomains . The identified differential co-localisation of the two transmembrane receptors with four different REM markers clearly demonstrated that both FLS2 and BRI1 clusters reside within PM nanodomains . Combined with our hypothesis of an interplay between receptor clusters and the cortical cytoskeleton , these findings are in agreement with the three-tiered hierarchical PM organization model proposed by Kusumi et al . ( 2011 ) . Moreover , the visualisation of the two LRR-RKs in different PM nanodomains underlined our results obtained by FLS2-BRI1 co-localisation that indicated a spatial separation for immune and growth receptors in plant PMs in steady-state conditions . The visualisation of FLS2 and BRI1 signalling complexes additionally confirmed these results . In particular , we found that BRI1-BIK1 , but not FLS2-BIK1 , complexes associated with cortical microtubules . Cortical microtubules are also anchor-points for the cellulose synthase complex ( Paredez et al . , 2006; Gutierrez et al . , 2009 ) . Intriguingly , activation of BR signalling has an immediate effect on cell wall morphology ( Elgass et al . , 2009 ) . Thus , a link between microtubule-associated BRI1-BIK1 complexes and BR-induced cell wall changes may exist . Interestingly , BIK1 fulfils opposing roles for BR and immune signal transduction ( Lin et al . , 2013 ) . Whereas BIK1 is a positive regulator of FLS2-mediated signalling by connecting , for example , ligand perception to ROS production ( Kadota et al . , 2014; Li et al . , 2014 ) , BIK1 negatively regulates BRI1-mediated cellular responses ( Lin et al . , 2013 ) . Based on our observations , this differential signalling specificity may be encoded in differential protein complex localisation ( and composition ) at the PM . Only 5–10% of BRI1 receptor molecules are actively involved in BR signal transduction ( van Esse et al . , 2012 ) . In addition , hyper-activation of BRI1 by exogenous application of BRs results in only a partial mobility shift of BRI1 , whereas its chemical or genetic inactivation only affects a subpopulation of BRI1 receptors ( Wang et al . , 2015 ) . Interestingly , the observation that only a subset of FLS2 receptors are present in steady-state complexes with their downstream substrates BIK1 and BSK1 , in what could be defined as pre-formed signalling platforms , correlates with the previous observation that flg22 binding to only a small subset of receptors is required for inducing a saturating response ( Meindl et al . , 2000; Bauer et al . , 2001 ) . Collectively , our results suggest a spatial separation of FLS2 and BRI1 signalling platforms under steady-state conditions . These findings demonstrate the existence of pathway-specific signalling component pools as previously postulated ( Albrecht et al . , 2012; Halter et al . , 2014 ) . The establishment of physically separated signalling units is an economically favourable concept for plant cells , since common signalling components can be employed along different and even antagonistic signal transduction routes . We propose that the spatiotemporal separation of FLS2 and BRI1 signalling platforms in steady-state conditions is a means to generate signalling specificity between the two pathways upon ligand perception , in addition to potential differential phosphorylation of common signalling components ( e . g . SERK3/BAK1 , BIK1 , and BSK1 ) . Whether differential phosphorylation of common signalling components in fact is a consequence of spatially separated signalling complexes will be investigated in future studies , besides the composition and stoichiometry of distinct PM signalling pools .
Arabidopsis seedlings were grown on Murashige and Skoog ( MS ) salt medium containing 1% sucrose and 0 . 8% agar with a 16 hr photoperiod at 22°C . N . benthamiana plants were soil-grown under a photoperiod of 16 hr and at 22°C . We used the previously published Arabidopsis lines Col-0/pFLS2::FLS2-GFP ( Göhre et al . , 2008 ) , Col-0/pBRI1::BRI1-GFP ( Geldner et al . , 2007 ) , and Col-0/p35S::mCherry-TUA5 ( Gutierrez et al . , 2009 ) . The double transgenic lines pBRI1::BRI1-GFP/p35S::mCherry-TUA5 and pFLS2::FLS2-GFP/p35S::mCherry-TUA5 were generated by crossing and imaged in the F1 population . For the generation of the Col-0/p35S::Lifeact-TagRFP line , a previously described plasmid was used ( Tilsner et al . , 2012 ) . Col-0 plants were stably transformed using the floral dip method ( Clough and Bent , 1998 ) . The double transgenic pBRI1::BRI1-GFP/p35S::Lifeact-TagRFP and pFLS2::FLS2-GFP/p35S::Lifeact-TagRFP lines were generated by crossing and imaged in the F1 population . The REM constructs used in this study were described previously ( Jarsch et al . , 2014 ) . For the generation of the bimolecular fluorescence complementation ( BiFC ) constructs , BIK1-CFP , BIK1-mRFP , BRI1-GFP , BRI1-mRFP , and BSK1-CFP BIK1 , BRI1 , and FLS2 were PCR amplified from Arabidopsis Col-0 cDNA and BSK1 was PCR amplified from Arabidopsis Col-0 genomic DNA . Subsequently , the gel purified PCR products were inserted via Gateway TOPO reaction into pENTR-D-TOPO plasmids and the products were verified via sequencing . Expression vectors for BiFC experiments were generated by Gateway LR reactions using the respective pENTR clones and the pAM-35S-GW-YFPc as well as pAM-35S-GW-YFPn ( Lefebvre et al . , 2010 ) , CFP-tagged constructs using pGWB44 ( Nakagawa et al . , 2007 ) , RFP-tagged using pB7RWG2 ( Karimi et al . , 2002 ) , GFP-tagged constructs using pK7FGW2 ( Karimi et al . , 2002 ) . FLS2-mCherry was generated from pCAMBIA2300 pFLS2::FLS2-3xMYC-GFP . Using the SalI restriction enzyme and subsequent ligation , GFP was substituted with mCherry . Agrobacterium tumefaciens GV3101 or GV3103 strains carrying p35S::BIK1-CFP , p35S::BSK1-CFP , pFLS2::FLS2-GFP , p35S::BRI1-GFP , p35S::BIK1-YFPn , p35S::BSK1-YFPn , p35S::BRI1-YFPc , p35S::FLS2-YFPc , p35S::BRI1-mRFP , pFLS2::FLS2-mCherry , p35S::GFP-TUA6 , or p35S::Lifeact-TagRFP were grown overnight in Luria-Bertani ( LB ) medium containing the respective antibiotics . Bacteria were harvested by centrifugation at 3 . 000 g for 5 min and re-suspended in H2O . After a second centrifugation step bacteria were re-suspended in infiltration buffer containing 10 mM MgCl2 , 10 mM MES pH 5 . 6 , and 100 µM acetosyringone . Bacterial suspensions were adjusted to OD600 = 0 . 2 or , for bimolecular fluorescence complementation constructs containing suspensions , to OD600 = 0 . 4 . Subsequently , the respective bacterial suspensions were combined for co-infiltration with a bacterial suspension providing the p19 helper plasmid . The final Agrobacterium tumefaciens suspension mixes were syringe infiltrated into the adaxial side of N . benthamiana leaves . The used N . benthamiana plants were 4–5 weeks old . Cotyledons of Arabidopsis seedlings or leaf discs of N . benthamiana were imaged on a Leica TCS SP5 ( Leica , Germany ) confocal microscope using a 63 × 1 . 2 NA water immersion objective . CFP , GFP , and YFP were excited using the Argon ion laser lines 458 nm , 488 nm , and 514 nm , respectively . TagRFP , mRFP and mCherry were excited using a DPSS laser ( 561 nm ) when imaged in combination with GFP or using a HeNe laser ( 594 nm ) when imaged in combination with YFP . Fluorescence emission was collected within following band width generated by an AOTF: 465–505 nm for CFP , 500–540 nm for GFP , 520–560 nm for YFP , and 590–630 nm ( DPSS excitation ) or 600–640 nm ( HeNe excitation ) for TagRFP/mRFP/mCherry . Cotyledons of Arabidopsis seedlings or leaf discs of N . benthamiana were imaged on a Zeiss Elyra PS1 ( Zeiss , Germany ) microscope using a 100 × 1 . 4 NA oil immersion objective . GFP was excited using a 488 nm solid-state laser diode and mCherry was excited using a 561 nm solid-state laser diode . Fluorescence emission was collected with an EM-CCD camera with bandwidth filters ranging from 495–550 nm for GFP and 575–635 nm for mCherry . For samples expressing GFP only fluorescence was collected with an acquisition time of 500 ms . For GFP/mCherry expressing samples the acquisition times were 250 ms for each channel using the ‘fast-channel’ mode . Throughout image acquisition the microscope was operated in the ‘TIRF’ mode . For brassinazole ( BRZ; TCI , UK ) treatment , 20 mM DMSO stock solution was diluted to a final concentration of 5 µM BRZ using liquid MS medium . Arabidopsis seedlings were transferred 3 days post germination in a 12-well plate containing the 5 µM BRZ solution or mock , respectively , and cultivated for another 2 days prior to imaging . For treatment of Nicotiana benthamiana leaves , BRZ was added to the infiltration suspension in a final concentration of 10 µM and co-infiltrated . For 24-epi-brassinolide ( BL; Sigma-Aldrich , UK ) treatment , 1 mM EtOH stock solution was diluted to a final concentration of 100 nM BL using liquid MS medium . For microscopy cotyledons were directly mounted in 100 nM BL and incubated for the indicated time period . For flg22 treatment , an aqueous 100 µM stock solution of flg22 ( EZ Biolabs , USA ) was diluted to a final concentration of 100 nM flg22 using liquid MS medium . For microscopy , cotyledons were directly mounted in 100 nM flg22 and incubated for the indicated time period . For methyl-β-cyclodextrin ( MβCD; Sigma-Aldrich , UK ) treatment , a 30 mM MβCD solution was prepared in liquid MS . Plants were incubate for the indicated time period in a 12-well plate at room temperature prior to mounting in the same solution on object slides for microscopy . Confocal and VAEM micrographs were analysed and modified using FIJI ( ImageJ 2 . 0 . 0–39/rc-1 . 50b ) . To emphasise cluster formation , the presented images and time series were modified using the ‘LoG3D’ plugin ( Sage et al . , 2005 ) . Time series were additionally neutralised with a saturation of zero prior to contrasting . For the generation of kymographs , we used the plugin ‘Multi Kymograph’ with a line width of one . For single particle analysis we used the plugin TrackMate ( 2 . 7 . 4 ) . After selecting a region of interest ( ROI ) encompassing the cell outline the LoG detector was selected . Based on preliminary particle analysis using FIJI the estimated blob diameter was set to 0 . 3 µm and the threshold was set to zero . To exclude false-positive particles and endomembrane compartments we applied a quality ( Auto setting ) and mean intensity filter . Subsequently , the simple LAP tracker was selected with a maximal linking distance of 0 . 5 µm and without gap-closing . Using the analysis option within the TrackMate dialog window we obtained statistics for particle size , track duration , and particle displacement . Co-localisation analysis was carried out as described previously ( Jarsch et al . , 2014 ) using ImageJ/FIJI . Endomembrane compartments were excluded from the co-localisation analysis by ROI selection . Briefly , acquired confocal images were ‘Mean’ filtered with a radius of 2 pixels . Background was subtracted using the ‘Rolling ball’ method with a radius of 20 pixels . ROIs were manual selected and for quantifying the co-localisation the plugin ‘Intensity Correlation Analysis’ ( Li et al . , 2004 ) was applied . Statistical analysis using Student’s t-tests was carried out in Numbers ( Apple , USA ) . Box plots were generated in R . BIK1 ( AT2G39660 ) , BRI1 ( AT4G39400 ) , BSK1 ( AT4G35230 ) , FLS2 ( AT5G46330 ) , REM1 . 2 ( AT3G61260 ) , REM1 . 3 ( AT2G45820 ) , REM6 . 1 ( AT2G02170 ) , REM6 . 2 ( AT1G30320 ) , TUA5 ( AT5G19780 ) , TUB5 ( AT1G20010 ) . | Unlike most animals , plants cannot move away if their environment changes for the worse . Instead , a plant must sense these changes and respond appropriately , for example by changing how much it grows . Disease-causing microbes in the immediate environment represent another potential threat to plants . To detect these microbes , plant cells have proteins called “pattern recognition receptors” in their surface membranes that sense certain molecules from the microbes ( similar receptors are found in animals too ) . When a receptor protein recognises one such microbial molecule , it becomes activated and forms a complex with other proteins referred to as co-receptors . The protein complex then sends a signal into the cell to trigger an immune response . Plants also use similar receptor proteins to sense their own signalling molecules and regulate their growth and development . These growth-related receptors rely on many of the same co-receptors and signalling components as the immunity-related receptors . This posed the question: how can plant cells use the same proteins to trigger different responses to different signals ? Bücherl et al . have now used high-resolution microscopy and the model plant Arabidopsis thaliana to show that the plant’s immune receptors and growth receptors are found in separate clusters at the plant cell’s surface membrane . These clusters are only a few hundred nanometres wide , and they also contained other signalling components that are needed to quickly relay the signals into the plant cell . Bücherl et al . suggest that , by organizing their receptors into these physically distinct clusters , plant cells can use similar proteins to sense different signals and respond in then different ways . This idea will need to be tested in future studies . Further work is also needed to understand how these clusters of signalling proteins are assembled and inserted at specific locations within the surface membrane of a plant cell . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Methods",
"and",
"materials"
] | [
"plant",
"biology",
"cell",
"biology"
] | 2017 | Plant immune and growth receptors share common signalling components but localise to distinct plasma membrane nanodomains |
Visual cues exert a powerful control over hippocampal place cell activities that encode external spaces . The functional interaction of visual cortical neurons and hippocampal place cells during spatial navigation behavior has yet to be elucidated . Here we show that , like hippocampal place cells , many neurons in the primary visual cortex ( V1 ) of freely moving rats selectively fire at specific locations as animals run repeatedly on a track . The V1 location-specific activity leads hippocampal place cell activity both spatially and temporally . The precise activities of individual V1 neurons fluctuate every time the animal travels through the track , in a correlated fashion with those of hippocampal place cells firing at overlapping locations . The results suggest the existence of visual cortical neurons that are functionally coupled with hippocampal place cells for spatial processing during natural behavior . These visual neurons may also participate in the formation and storage of hippocampal-dependent memories .
Spatial information is presumably encoded by hippocampal ‘place cells’ , which fire predominantly when an animal is at one or a few specific places ( place fields ) of a given space ( O'Keefe and Dostrovsky , 1971; McNaughton et al . , 1983; Wilson and McNaughton , 1993 ) . How these cells establish the location specificity has been a subject under intensive investigation . It is known that path integration of the information generated from self-motion is important for hippocampal place cell activity ( Hafting et al . , 2005; McNaughton et al . , 2006 ) . However , external sensory cues are equally important , if not more . In particular , visual cues play a pivotal role in place cell activities ( Muller , 1996; Knierim , 2002; Colgin et al . , 2008; Ravassard et al . , 2013 ) . For example , changing the proximal or distal visual cues alters firing rates and/or firing locations of place cells in a one- or two-dimensional space ( Lee et al . , 2004a , 2004b; Leutgeb et al . , 2005 ) , and the rotation of a salient cue card on the wall of a symmetric open arena evokes an equivalent rotation of place field locations ( Muller and Kubie , 1987 ) . How visual cue information contributes to place cell activities is unknown . Visual cues are presumably encoded and processed in the visual cortex , which is connected to the hippocampus via a multi-synaptic pathway involving the associational cortices including the entorhinal cortex ( Miller and Vogt , 1984; Vaudano et al . , 1991; Lavenex and Amaral , 2000; Furtak et al . , 2007 ) . Computational models have shown that hippocampal cells can integrate input from cortical cells , in particular those encoding visual cues , to form a conjunctive response that is location-specific ( de Araujo et al . , 2001; Schonfeld and Wiskott , 2015 ) . Alternatively , it is proposed that local visual cues influence place field location through entorhinal ‘border’ cells , whereas distal cues influence place field orientation via entorhinal ‘head direction’ cells ( Knierim and Hamilton , 2011 ) . Obviously , these models and others similar require the functional interaction between those visual cortical neurons encoding visual cues and corresponding hippocampal place cells . However , experimental data for such an interaction are very limited ( Ji and Wilson , 2007 ) , and the nature of this interaction during natural spatial behavior remains unexplored . To probe the visual cortical—hippocampal interaction during spatial processing , we simultaneously recorded cells in the primary visual cortex ( V1 ) and in the CA1 area of the hippocampus , while freely moving rats traversed back and forth on a track . We aimed to test a hypothesis that there exists a specific population of visual cortical neurons that responds to visual features of the track with specific activity patterns and functionally interacts with those hippocampal place cells representing the same track . We found that , like CA1 place cells , a large number of V1 neurons also displayed firing activities that were specific to particular locations on the track , especially around the visual landmarks . The location-specific activities of V1 neurons , on average , led those of CA1 place cells in both the spatial and temporal domains . For those V1 and CA1 cells that displayed spatially overlapping activity , their precise firing characteristics co-fluctuated every time the animal traveled through the track . These findings suggest the existence of specific V1 activities that participate in the encoding of external spaces . Given the critical role of the hippocampus in spatial memories ( Scoville and Milner , 1957; Squire , 1992; Eichenbaum et al . , 1999; Burgess and O’Keefe , 2003 ) , these visual cells may also participate in the formation and storage of the spatial memories encoded by hippocampal place cells .
To test our hypothesis , we asked whether V1 cells responded to particular locations on a given trajectory of the track in a manner similar to CA1 place cells , with the assumption that the V1 response might not be a spatial response per se , but possibly resulted from the visual cues . As expected , CA1 cells exhibited typical place-field firing characteristics on the track ( Figure 2A ) . Many V1 cells also dramatically increased their firing rates at specific locations of a trajectory ( Figure 2B–D ) , as we have shown previously ( Ji and Wilson , 2007 ) . The location-specific increase in firing activity of V1 cells was apparently stable during each lap of the trajectory . The firing rate curves , defined as the lap-averaged firing rate at every position of a trajectory , of these V1 cells displayed a few well-defined peaks , similar in character to the place fields of CA1 cells and to the multi-peak firing of medial entorhinal grid cells on linear tracks ( Hafting et al . , 2008 ) . Henceforth , we refer to the locations corresponding to the rate curve peaks of a cell as its ‘firing fields’ . 10 . 7554/eLife . 08902 . 004Figure 2 . V1 cells fired predominantly at specific locations during track running . ( A–D ) Firing activities of a CA1 place cell ( A ) and 3 V1 cells ( B , C , D in layer L2/3 , L4 , L5/6 respectively ) . For each panel , the top displays the spike raster of a cell within every lap of running on a trajectory , which is linearized and plotted as the x-axis . Each tick represents a spike . The bottom is the firing rate curve averaged across all the laps . Arrows: firing fields . ( E ) Cumulative distributions of spatial information content ( SIc ) values of V1 and CA1 cells with actual and shuffled spiking activities . ( F ) Same as E , but for spatial information rate ( SIr ) . ( G–I ) Cumulative distribution of spatial modulation index ( SMI , G ) , number of fields per trajectory ( H ) , and field length ( I ) for V1 and CA1 cells . ( J ) Same as E , but for spatial stability . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 00410 . 7554/eLife . 08902 . 005Figure 2—figure supplement 1 . Illustration of computing spatial modulation indices ( SMIs ) for two example cells with different firing rates . ( A ) Lap by lap spike raster and spike rate curves of the two cells averaged across all laps . The plots are arranged the same way as in Figure 2A–D . Spatial information content ( SIc ) and spatial information rate ( SIr ) computed from the rate curves are shown . Note that the higher-rate Cell 2 appeared more spatially modulated than Cell 1 , but had a lower SIc . ( B ) An example of shuffled firing rate curve for each of the two cells , generated via circularly shifting the spike trains in A by a random time interval for every lap . Note the comparable peaks between the actual and the shuffled rate curves in the low-rate Cell 1 , but not in Cell 2 . The SIc and SIr computed from the shuffled rate curves were shown . ( C ) Histograms of SIc and SIr values computed from 100 randomly shuffled firing rate curves for the two cells . The mean and standard deviation ( std ) of each histogram are shown . Red arrows mark the actual values . SMI is computed as ( actual value−mean ) /std . It can be seen that the shuffle-generated values differed between SIc and SIr for the same cell and between the two cells , even though the shuffled spike trains by definition contained no spatial information . The shuffled spikes of the low-rate Cell 1 yielded a higher SIc , but a lower SIr , than the shuffled spikes of the high-rate Cell 2 . The examples illustrate the rate–dependence of SIc and SIr . Second , by normalizing the actual SIc/SIr values relative to their shuffle-generated values , SMIs computed from SIc and SIr become equivalent . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 005 We quantified the location-specificity of each individual V1 and CA1 cell active on a trajectory . To compare with the location-specificity arising randomly from chance , we randomly shuffled the cell's spiking activity by circularly shifting the spikes within each lap of the trajectory with a random time interval ( Henriksen et al . , 2010; Igarashi et al . , 2014 ) . First , we computed spatial information content ( SIc ) , a measure of how much information ( in bits per spike ) a cell's spiking activity contained about the animal's location ( Skaggs et al . , 1993 ) . Although the median SIc value of V1 cells ( 0 . 17 [0 . 085 0 . 34] bits/spike , N = 1501 cell × trajectories ) was relatively small , compared with that of CA1 place cells ( 1 . 6 [1 . 1 2 . 2] bits/spike , N = 2909; p < 0 . 0001 , ranksum test ) , it was significantly greater than that of the shuffled V1 cells ( 0 . 061 [0 . 025 0 . 13] bits/spike; p < 0 . 0001; Figure 2E ) . Second , we computed spatial information rate ( SIr ) , which measures spatial information in bits per second . Similarly to SIc , the median SIr of V1 cells ( 0 . 70 [0 . 42 1 . 2] bits/s ) was smaller than that of CA1 cells ( 2 . 4 [1 . 2 4 . 2] bits/s; p < 0 . 0001 ) , but significantly greater than that of the shuffled V1 cells ( 0 . 22 [0 . 14 0 . 37] bits/spike; p < 0 . 0001; Figure 2F ) . Third , using a method modified from previous studies ( Henriksen et al . , 2010; Igarashi et al . , 2014 ) , we derived a normalized spatial modulation index ( SMI ) . The reason for this additional measure was that SIc and SIr are affected by firing rate ( Figure 2—figure supplement 1 ) . Since V1 and CA1 cells had different firing rates , the SIc and SIr values between V1 and CA1 cells were not directly comparable . SMI was defined as the SIc ( or equivalently SIr ) of a cell relative to its chance-level distribution produced by the random shuffling ( Figure 2—figure supplement 1 ) . SMI does not directly quantify the location-specificity of a cell's firing activity , but provides a measure of the degree of location modulation relative to random spike trains with identical firing rate and temporal spiking patterns . SMI is therefore insensitive to firing rate . The chance-level of SMI for any given cell is zero . The median SMIs of both V1 ( 8 . 1 [3 . 4 15 . 4] ) and CA1 cells ( 12 . 0 [3 . 3 23 . 2] ) were much higher than zero ( p < 0 . 0001 , ranksum test; Figure 2G ) . Finally , we defined a cell with SMI >2 . 325 ( 99th percentile of the chance-level ) as a ‘location-responsive’ V1 cell . We found that 81% of trajectory-active V1 cells were location-responsive on a trajectory and for comparison , 90% of the trajectory-active CA1 cells were location-responsive . Next , we attempted to understand why location-responsive V1 cells had lower location-specificity than CA1 place cells . One obvious reason is that V1 cells , unlike CA1 place cells , often contained much baseline firing outside of a concentrated firing field ( Figure 2B–D ) . A second observation is that V1 cells often fired at multiple firing fields on a trajectory ( Figure 2B–D ) , whereas many CA1 cells only exhibited one or two fields on our track ( Figure 2A ) . We quantified the number and average spatial length of firing fields for location-responsive V1 cells . On average , the location-responsive V1 cells displayed nearly twice the number of firing fields as the CA1 place cells ( mean ± SE for V1: 2 . 9 ± 0 . 04 per trajectory , N = 1216 cell × trajectories; CA1: 1 . 5 ± 0 . 02 per trajectory , N = 2618; p < 0 . 0001 , t-test; Figure 2H ) . On the other hand , the average field length of V1 cells was significantly smaller than that of CA1 cells ( V1: 37 . 5 ± 0 . 3 cm , N = 3521 fields; CA1: 47 . 2 ± 0 . 4 cm , N = 3918; p < 0 . 0001; Figure 2I ) . The result shows that the relative low location-specificity in V1 cells was partially due to more firing fields , but not due to larger field sizes . We also quantified the lap-by-lap stability of a V1 cell's location-specific activity as an animal traveled the same trajectory repeatedly . For each cell active on a trajectory , we computed its firing rate curve separately for each individual lap on the trajectory and then computed the correlation between the rate curves of any two laps ( Cheng and Ji , 2013 ) . Spatial stability was the average correlation among all combinations of laps . To understand how the stability compared with the chance level , we also computed the spatial stability of each cell after the random shifting of its rate curves lap by lap . Similarly as in spatial information measures , the median spatial stability of V1 cells ( 0 . 19 [0 . 06 0 . 31] , N = 1501 cell × trajectories ) was lower than that of CA1 cells ( 0 . 51 [0 . 28 0 . 70] , N = 2909; p < 0 . 0001 , ranksum test ) , but significantly greater than the chance level ( 0 . 0014 [-0 . 0074 0 . 0060]; p < 0 . 0001; Figure 2J ) , indicating that V1 cells responded reliably to same locations during each lap of track running . We next asked how V1 and CA1 firing activities interacted . First , we analyzed how the V1 and CA1 firing fields were distributed along the two trajectories on the track with opposite running directions . We found that V1 and CA1 fields were not equally distributed along the trajectories , but tended to increase in number close to the food wells , as shown previously for CA1 cells ( Dupret et al . , 2010 ) . V1 and CA1 field numbers also increased before and after the animals traveled through the corners of the track , referred to as landmarks , and decreased as the animals traveled through the long segment between the two middle corners ( Figure 3A , B ) . The numbers of V1 and CA1 fields along the trajectories fluctuated together in a highly correlated fashion for both trajectories ( Figure 3C ) . We quantified this observation by computing the cross-correlation between the field distribution curves of V1 and CA1 cells on each trajectory . The cross-correlograms for both trajectories had peaks not exactly at position lag 0 , but at a positive position ( 2 cm , p < 0 . 0001 , Pearson's r ) on one running direction and a negative position ( −4 cm , p < 0 . 0001 ) on the opposite direction ( Figure 3D , E ) , meaning that the number of V1 fields fluctuated ahead of the number of CA1 fields consistently on both running directions . Given the mean speed of 35 cm/s of our rats , the 2–4 cm leading distance is equivalent to 60–120 ms of leading time . While the higher numbers of firing fields around the food wells and corners might simply reflect more sensory information available around the landmarks , the significant correlation between the field distribution curves of V1 and CA1 cells with a spatial delay suggests an interaction between these groups of cells , possibly reflecting the propagation of the landmark-related sensory information from V1 to CA1 . 10 . 7554/eLife . 08902 . 014Figure 3 . Firing fields of V1 and CA1 cells appeared to accumulate around the landmarks of the track . ( A , B ) Histograms of the number of firing fields for V1 ( A ) and CA1 ( B ) cells on the two trajectories of the track with opposite running directions . Vertical gray lines: landmark positions ( corners ) of the track . Arrows: running directions . Black lines: smoothed curves of the histograms . Note that the number of fields tended to peak before and after the landmarks for both running directions and for both V1 and CA1 cells . ( C ) The smoothed curves for V1 and CA1 cells in A and B are normalized ( by total number of firing fields ) and re-plotted together for each of the running direction ( arrow ) . Note that the V1 curve ( red ) tended to rise and fall slightly earlier than the CA1 ( blue ) curve on both running directions . ( D , E ) Cross-correlogram between the V1 and CA1 smoothed field curves in C on each of the running directions ( arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 014 The bias of V1 and CA1 fields around the landmarks on both directions prompted us to examine the directionality of V1 and CA1 firing fields . We observed that some V1 cells fired in front of or behind the landmarks on two opposite running directions , which we termed ‘prospective’ or ‘retrospective’ firing as in a previous study ( Battaglia et al . , 2004 ) . An example of V1 cell with prospective firing is shown in Figure 4A , which was quantified by a significant peak at a positive position lag ( 36 cm , p < 0 . 0001 , Pearson's r ) in the cross-correlogram between the cell's firing rate curves on the two trajectories . An example of CA1 cell with retrospective firing is shown in Figure 4B , with a significant peak at a negative position lag ( −18 cm , p < 0 . 0001 ) in the cross-correlogram between its firing rate curves on the two trajectories . We found that among all trajectory-active cells , 9 . 7% of V1 cells ( N = 75 ) and 10 . 4% of CA1 cells ( N = 212 ) were bidirectional , defined as those with a significant peak within [−50 , 50] cm in their rate curve cross-correlograms ( see ‘Materials and methods’ ) . The cross-correlogram peak positions of these bidirectional V1 cells display a bimodal distribution with two groups around 26 cm and −26 cm ( Figure 4C ) , which corresponded to prospective and retrospective firing , respectively . Similarly , the peak positions of CA1 cells also show a bimodal distribution with two groups around 18 cm and −28 cm ( Figure 4D ) . This result suggests that a group of V1 cells responded to the same landmarks on both directions by either ‘looking’ ahead or back to the landmarks . 10 . 7554/eLife . 08902 . 006Figure 4 . Bi-directional firing of V1 and CA1 cells on the C-shaped track . ( A ) Firing activity of a V1 cell ( lap-by-lap spike raster and average firing rate curve; see Figure 2 legend for details ) on two trajectories with opposite running directions ( left ) and the cross-correlogram of the cell's two firing rate curves ( right ) . Vertical Gray lines: land mark positions ( corners ) of the track . Arrows: running directions . Note that the peaks appeared before the animal passed the same landmarks ( ∆ , prospective firing ) on both running directions , resulted in a primary peak ( * ) at a positive position lag in the cross-correlogram . ( B ) Same as A , but for an example of CA1 cell showing consistent firing after a landmark ( o , retrospective firing ) on both directions . ( C , D ) Histograms of the cross-correlogram peak positions of all V1 ( C ) and CA1 ( D ) cells with significant bi-directional firing . Black lines: smoothed curves of the histograms . Note that both the V1 and CA1 distributions appeared to be bi-modal , suggesting prospective or retrospective firing for V1 and CA1 bi-directional cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 006 We then analyzed the pair-wise cross-correlation between V1 location-responsive cells and CA1 place cells in the time domain . We used a normalized spike count cross-correlation ( see ‘Materials and methods’ ) , which is insensitive to the cells' firing rates , to quantify how the spiking activities of two cells were temporally correlated . Figure 5A–C shows an example pair of V1 and CA1 cells with both having well-defined firing fields on the same trajectory , which led to a prominent peak in its normalized cross-correlogram . The peak time quantified the temporal relationship of the two cells . For individual pairs of cells , this peak , especially that with a long peak time , might just passively reflect the fact that they had firing fields on the same trajectory , not necessarily reflecting any functional interaction . However , if a large number of such pairs are collected , the distribution of their peak times may inform the temporal relationship between the V1 and CA1 activities at the population level . Among 22 , 969 pairs of V1 location-responsive cells and CA1 place cells , we obtained 997 highly significantly correlated pairs ( see ‘Materials and methods’ for definition ) with peak correlation times within [−0 . 2 , 0 . 2] s . The distribution of these peak times was significantly biased toward positive values ( 58% positive , 46% negative , p = 0 . 00017 , binomial test; Figure 5D ) . The result indicates that , on average , V1 cells led CA1 cells in the time domain , suggesting the propagation of visual information from V1 to CA1 . We also analyzed whether this direction of interaction was true to those V1-CA1 cell pairs with both showing bi-directional firing . Out of the 12 pairs with both the V1 and CA1 displaying prospective firing and 9 pairs with one displaying prospective firing while the other displaying retrospective firing ( only 1 pair with both displaying retrospective firing ) , 9 ( 75% ) and 6 ( 68% ) , respectively , showed positive peak times ( Figure 5E ) , suggesting that the V1-CA1 interaction of these pairs followed the general trend at the population level . 10 . 7554/eLife . 08902 . 007Figure 5 . Pair-wise cross-correlation between V1 and CA1 cells . ( A , B ) Firing activity ( lap-by-lap spike raster and average firing rate curve; see Figure 2 legend for details ) of a pair of V1 ( A ) and CA1 ( B ) cells on a trajectory of the C-shaped track . ( C ) Normalized cross-correlogram of the two cells in A and B . *: peak time of the cross-correlogram . ( D ) Histogram of the peak times for all highly significantly correlated pairs of V1 and CA1 cells ( see ‘Materials and methods’ ) . Note the bias of peak times toward positive time lags . ( E ) Peak times of those highly correlated V1-CA1 pairs with both displaying prospetive firing ( Pros pairs ) and of those pairs with one displaying prospective while the other displaying retrospective firing ( Mix pairs ) . Each dot is a pair . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 007 As many pairs of V1 and CA1 cells displayed overlapping firing fields on a trajectory ( Figure 6A; see more examples in Figure 6—figure supplement 1 ) , we reasoned that these pairs of V1 and CA1 cells could be specifically interacting for integrating the visual information present at a location to the hippocampal place cell activity encoding the same location . Therefore , we next focused on the possible interaction between V1 and CA1 cells with overlapping firing fields . We observed that the firing rates and firing locations of such V1 and CA1 cells fluctuated from lap to lap within their respective firing fields , and interestingly , they often fluctuated together in a correlated fashion ( Figure 6A , B , Figure 6—figure supplement 1 ) . We quantified this observation by an analysis similar to the ‘noise’ correlation in previous studies on V1 cell correlation in primates ( Zohary et al . , 1994; Ecker et al . , 2010; Hansen et al . , 2012 ) . For a pair of cells with overlapping firing fields , we identified each cell's spikes within its firing field for each individual lap , and computed two quantities of these spikes: the ( within-field ) firing rate and the center of mass ( COM ) of their firing locations . We then obtained the lap-by-lap fluctuations in within-field firing rate ( Δrate - the difference between a lap's firing rate and the averaged firing rate across all laps ) and in COM ( ΔCOM ) for each cell . Finally , we computed the Pearson correlation in Δrate and in ΔCOM between two cells , which measures how closely the two cells varied their precise firing rate or firing location each time the animal traveled through the overlapped firing fields . Indeed , the pair of cells shown in Figure 6A were significantly correlated in both Δrate and ΔCOM ( Figure 6C; see more examples in Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 08902 . 008Figure 6 . Pairs of V1 and CA1 cells with overlapping firing fields displayed correlated lap-by-lap fluctuations in firing rate and firing location . ( A ) Lap-by-lap spike raster and the average rate curves ( see Figure 2 legend for details ) of a pair of V1 and CA1 cells on the same trajectory . Boxes: the overlapping firing fields of the two cells . ( B ) The spikes within the marked laps ( * ) of the two cells in A are expanded and plotted together . Note the correlated lap-by-lap shifting of the V1 and CA1 cells in their spikes . ( C ) The lap-by-lap fluctuations in firing rate ( ΔRate ) and COM ( ΔCOM ) within the firing fields of the two CA1 and V1 cells in A . Each dot is a lap . Solid line: linear regression . R , P: Pearson's correlation between the CA1 and V1 fluctuations and the associated p-value . ( D ) Average correlation in ∆rate and ∆COM for overlapping , non-overlapping , and non-responsive pairs of CA1 and V1 cells ( see text for definitions ) . ( E , F ) Same as C and D , but for modified ∆rate and modified ∆COM after removing the modulations of firing rate and COM by speed and head direction . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 00810 . 7554/eLife . 08902 . 009Figure 6—figure supplement 1 . Two more examples of overlapping V1-CA1 cell pairs with correlated lap-by-lap fluctuations in , each from a different animal . For each example ( A or B ) , plotted on the left are the lap-by-lap spike raster and firing rate curves of the V1 ( red ) and CA1 ( blue ) cells , while plotted on the right are the lap-by-lap fluctuations in rate ( ∆rate ) and COM ( ∆COM ) for the two cells on the left , and in rate ( modified ∆rate ) and COM ( modified ∆COM ) after the modulation by speed and head direction was removed . See the main figure ( Figure 6 ) legend for details . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 00910 . 7554/eLife . 08902 . 010Figure 6—figure supplement 2 . Illustration of overlapping , non-overlapping , and non-responsive V1-CA1 cell pairs . ( A ) An overlapping pair: a location-responsive V1 cell ( top ) and a CA1 place cell ( bottom ) with spatially overlapping firing fields ( with 50–100% overlap ) . ( B ) A non-overlapping pair: a location-responsive V1 cell ( top ) and a CA1 place cell ( bottom ) with spatially non-overlapping firing fields . ( C ) A non-responsive pair: a non-location-responsive V1 cell ( top ) and a CA1 place cell ( bottom ) . Boxed areas: the spatial intervals for probing the co-fluctuation of a given pair . For overlapping pair ( A ) and non-overlapping pairs ( B ) , the spatial intervals were their firing fields . For Non-responsive pairs ( C ) , the spatial interval for the location-responsive cell was its firing field , whereas for the non-location-responsive cell it was the location-responsive cell's firing field shifted with a small random distance that yielded a random overlap of 50–100% with the firing field of the other cell . The same V1 cell and CA1 place cell can be involved in multiple pairs . In this example the same CA1 place cell appeared in 3 pairs . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 01010 . 7554/eLife . 08902 . 011Figure 6—figure supplement 3 . Overlapping V1-V1 and CA1-CA1 cell pairs displayed correlated lap-by-lap fluctuation in firing rate and COM within their firing fields . ( A ) Average correlation in ∆rate and ∆COM for pairs of V1 location-responsive cells with overlapping firing fields ( Overlapping ) , pairs of V1 location-responsive cells with non-overlapping firing fields ( Non-overlapping ) , and pairs made of one location-responsive V1 cell and one non-location-responsive V1 cell ( Non-responsive ) . ( B ) Same as ( A ) , but after the modulation by speed and head direction was removed . ( C , D ) Same as A and B , but for CA1-CA1 cell pairs . Number of V1-V1 pairs: N = 803 overlapping pairs; 953 non-overlapping pairs; 30 non-responsive pairs . Number of CA1-CA1 pairs: N = 1621 overlapping pairs; 11 , 273 non-overlapping pairs; 121 non-responsive pairs . *: p < 0 . 05; **: p < 0 . 01; ***: p < 0 . 001; ****: p < 0 . 0001; t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 01110 . 7554/eLife . 08902 . 012Figure 6—figure supplement 4 . Modulation of V1 and CA1 firing activities by speed and head direction . ( A ) Correlations between lap-by-lap speed and Δrate and between lap-by-lap speed and ΔCOM for example V1 and CA1 cells ( the same as in Figure 6—figure supplement 1 , panel B ) . Solid line: linear regression . R , P: Pearson's correlation and the associated p-value . Overall , speed was significantly correlated ( p < 0 . 05 ) with Δrate in 25% ( N = 107 out of 428 ) of V1 location-responsive cells and 41% ( N = 619 out of 1510 ) of CA1 place cells , and with ΔCOM in 14% ( N = 60 ) of V1 location-responsive cells and 25% ( N = 378 ) of CA1 place cells . ( B ) Same as A , but between head direction fluctuation ( Δhdir ) and Δrate and between Δhdir and ΔCOM for the same V1 and CA1 cells in A . Δhdir was significantly correlated with Δrate in 12% ( N = 51 ) of location-responsive V1 cells and in 21% ( N = 317 ) of CA1 place cells , and with ΔCOM in 23% ( N = 98 ) of location-responsive V1 cells and in 23% ( N = 347 ) of CA1 place cells . ( C ) Distributions of speed-Δrate and speed-ΔCOM correlation values for all CA1 place cells and location-responsive V1 cells . Dashed lines: 0 correlation . The speed-Δrate distribution was skewed to the positive side for both V1 ( 0 . 064 ± 0 . 014; p < 0 . 0001 , t-test compared with 0 ) and CA1 cells ( 0 . 16 ± 0 . 009 , p < 0 . 0001 ) , indicating that at the population level both V1 and CA1 cells increased their firing rates as speed increased . The speed-ΔCOM distribution was slightly but significantly skewed toward a positive mean for V1 ( 0 . 029 ± 0 . 011 , p = 0 . 014 ) , but not for CA1 cells ( 0 . 010 ± 0 . 008 , p = 0 . 18 ) , suggesting that , as speed increased , V1 cells' firing locations tended to move slightly forward along the animal's movement direction . ( D ) Same as ( C ) , but for Δhdir-Δrate and Δhdir-ΔCOM distributions . The Δhdir–Δrate distribution was centered near 0 for both V1 ( −0 . 0052 ± 0 . 011 , p = 0 . 44 ) and CA1 ( 0 . 0057 ± 0 . 0073 , p = 0 . 64 ) cells , indicating no systematic relationship between firing rate and head direction at the population level , even though each individual cell could increase or decrease firing rate as head direction was changed from left to right or vice versa . This result is expected if we assume that the V1 or CA1 cells as a group should not show any preferred head direction , even through individual V1 cells are tuned to particular directions . The Δhdir - ΔCOM distribution was similarly centered near 0 for both V1 ( 0 . 009 ± 0 . 015 , p = 0 . 53 ) and CA1 ( 0 . 003 ± 0 . 008 , p = 0 . 68 ) ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 01210 . 7554/eLife . 08902 . 013Figure 6—figure supplement 5 . V1 location-responsive cells showed much less lap-by-lap backward shift in their firing locations than CA1 place cells . ( A ) Average lap-by-lap changes in COM for V1 location-responsive cells ( red , N = 670 ) and CA1 place cells ( blue , N = 1743 ) . The COM change of a firing field at each lap was computed relative to its stabilized value , which was the average of those values at laps #21–25 . Solid lines: linear regressions between the COM change and lap numbers for the first 10 laps . It can be seen that the COMs of CA1 place fields significantly and systematically shifted backward ( COM decreased with lap number ) along the animal's moving direction ( p < 0 . 0001 , one-way ANOVA; Pearson's R = −0 . 89 , p = 0 . 0006 ) . The COMs of V1 firing fields appeared to shift backward during the first 5 laps or so , but fluctuated forward/backward in later laps . As a result , there was no significant change in COM ( p = 0 . 09 , one-way ANOVA ) within the first 10 laps and no significant correlation between average COM change and lap number ( Pearson's R = 0 . 10 , p = 0 . 79 ) . In addition , the average change in COM of V1 firing fields was significantly less than that of CA1 place fields within the first 10 laps ( p < 0 . 0001 , two-way ANOVA ) . ( B ) Same as A , but for lap-by-lap COM change of V1 and CA1 cells after removing the modulation by speed and head direction . The results are similar . There was a systematic backward shifting of the modified COM for CA1 place fields ( p < 0 . 0001 , one-way ANOVA; Pearson's R = 0 . 86 , p = 0 . 0014 ) , but not so for V1 firing fields ( p = 0 . 13; R = 0 . 23 , p = 0 . 53; comparison between CA1 and V1: p < 0 . 0001 , two-way ANOVA ) . Therefore , the analysis indicates that V1 firing fields showed much less dynamics at the short-term lap-by-lap time scale than CA1 place fields . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 013 In our data , we found 786 pairs with overlapping firing fields ( see ‘Materials and methods’ ) . We referred to these pairs as ‘overlapping pairs’ . We found that 48% and 23% of the overlapping pairs were significantly correlated in Δrate and ΔCOM ( p < 0 . 05 , Pearson's r ) , respectively . As a group , the average correlation of all the overlapping pairs were significantly greater than 0 , the chance-level correlation , for both Δrate ( 0 . 29 ± 0 . 01 , p < 0 . 0001 , t-test; Figure 6D ) and ΔCOM ( 0 . 16 ± 0 . 01 , p < 0 . 0001; Figure 6D ) . This result indicates that there was a precise lap-by-lap co-fluctuation in firing rate and COM between many pairs of V1 cells and CA1 place cells with overlapping firing fields . We also performed the same analysis on two control groups of V1-CA1 cell pairs ( see Figure 6—figure supplement 2 for examples of V1-CA1 pairs ) . The first group consisted of 6681 pairs , each made of a CA1 place cell and a V1 location-responsive cell that both exhibited firing fields on a trajectory , but that their firing fields were non-overlapping . In this case , the Δrate and ΔCOM were computed within their most dominant firing fields on the trajectory . This control group , referred to as ‘non-overlapping pairs’ , allowed us to test whether the CA1-V1 co-fluctuation was spatially confined within the overlapped firing fields . The second group contained 238 pairs , each composed of a CA1 cell that had place field on a trajectory and an active V1 cell that was not location-responsive on the trajectory . In this case , the Δrate and ΔCOM fluctuations of the V1 cell were computed within a spatial interval that was overlapped with the dominant CA1 place field ( see ‘Materials and methods’ ) . We call this group ‘non-responsive’ pairs , which allowed us to test whether the co-fluctuation was specific to the location-responsive V1 and CA1 cells . We found that the overlapping pairs had significantly higher correlation in Δrate than both non-overlapping ( correlation: 0 . 062 ± 0 . 003; p < 0 . 0001 , t-test ) and non-responsive pairs ( −0 . 023 ± 0 . 015; p < 0 . 0001; Figure 6D ) . Similarly , the overlapping pairs had significantly higher correlation in ΔCOM than non-overlapping ( correlation: 0 . 0005 ± 0 . 0028; p < 0 . 0001 ) and non-responsive pairs ( correlation: 0 . 047 ± 0 . 020; p < 0 . 0001; Figure 6D ) . We also computed the Δrate and ΔCOM correlations for CA1-CA1 cell pairs and for V1-V1 cell pairs . The results were similar: Overlapping pairs within each of the two brain areas were significantly correlated in both Δrate and ΔCOM and had significantly higher correlation than non-overlapping pairs and non-responsive pairs ( Figure 6—figure supplement 3 ) . Taken together , the results above demonstrate a specific , precise co-fluctuation in the firing rates and COMs of V1 location-responsive cells and CA1 place cells with overlapping firing fields , suggesting a functional interaction between these cells . Remarkably , this long–range interaction between cells in the distal V1 and CA1 was qualitatively similar to the local interaction within CA1 place cells . Next , we examined whether the co-fluctuation of activity in overlapping V1-CA1 cell pairs could be explained by the lap-by-lap behavioral fluctuations . As expected ( Huxter et al . , 2003; Saleem et al . , 2013 ) , the firing rate and COM were modulated by speed and head direction in many V1 and CA1 cells ( Figure 6—figure supplement 4 ) . We quantified the modulation of both speed and head direction on Δrate/ΔCOM by a multi-variant linear regression and then removed the modulation to obtain the modified lap-by-lap fluctuations in firing rate/COM ( modified Δrate/ΔCOM ) , which were no longer correlated with speed or head direction ( see ‘Materials and methods’ ) . We computed the correlations in the modified Δrate/ΔCOM for overlapping V1-CA1 cell pairs . For the pair in Figure 6A , the correlations in their modified Δrate and ΔCOM remained unchanged ( Figure 6E ) , although there was a modest reduction in other pairs ( Figure 6—figure supplement 1 ) . For the group of overlapping V1-CA1 pairs , the average correlation between the modified Δrate ( 0 . 14 ± 0 . 009 ) remained significantly greater than 0 ( p < 0 . 0001 , t-test ) , and was significantly higher than those of non-overlapping ( 0 . 018 ± 0 . 002 , p < 0 . 0001 ) and non-responsive ( −0 . 019 ± 0 . 014 , p < 0 . 0001 ) V1-CA1 pairs ( Figure 6F ) . Similarly , the average correlation between the modified ΔCOM ( 0 . 13 ± 0 . 009 ) was also significantly greater than 0 ( p < 0 . 0001 ) , and was significantly higher than those of non-overlapping ( −0 . 003 ± 0 . 003 , p < 0 . 0001 ) and non-responsive ( 0 . 029 ± 0 . 018 , p < 0 . 0001 ) V1-CA1 pairs ( Figure 6F ) . From these results , we conclude that , behavioral variations cannot fully account for the correlation in the firing rate and COM of overlapping V1-CA1 cell pairs . The co-fluctuation of V1-CA1 cells suggests that those overlapping pairs of V1-CA1 cells with both displaying bidirectional firing should show the same type of directionality ( prospective or retrospective firing ) , similarly as in previous reports on CA1 cells and entorhinal grid cells ( De Almeida et al . , 2012; Bieri et al . , 2014 ) . Our data yielded 4 such overlapping and 10 non-overlapping pairs of V1 and CA1 cells . We found that all the 4 overlapping pairs showed the same type of directionality , but only 5 of the 10 non-overlapping pairs did so whereas the other 5 showed the opposite directionality . Although the number of such bi-directional overlapping pairs is low ( mainly due to the low percentage of bidirectional V1 and CA1 cells ) , the result is consistent with the correlated fluctuation in the firing rate and firing location between overlapping V1 and CA1 cells . In addition , CA1 cells are known to systematically shift their COMs backward lap by lap ( Mehta et al . , 1997 , 2000 ) , presumably as a result of synaptic plasticity ( Ekstrom et al . , 2001 ) . To understand whether the observed co-fluctuation in COM between V1 and CA1 cells were related to this lap-by-lap plastic change , we examined the lap-by-lap shift in the COMs of V1 cells . Our analysis shows that , although there was a sign of backward shifting in V1 cells during the first 5 laps or so , the shift was not as robust as and much smaller than that of CA1 cells ( Figure 6—figure supplement 5 ) . The result suggests that V1 firing activities are less plastic than those of CA1 cells at this short time scale of laps ( McClelland et al . , 1995 ) , and that the co-fluctuation between V1 and CA1 cells is not primarily driven by rapid plasticity in their firing activities . In our experiments , V1 and CA1 cells were recorded as the animals ran the same track for multiple days . We next analyzed how the activities of V1 location-responsive cells and CA1 place cells changed over the many days' experience of track running . We grouped the recording days to 3 different time points ( T1–T3 ) : T1 included Day 1 and 2 , a novel condition to the animals , in which most of the behavioral changes occurred ( Figure 1B , C ) , T3 included Day 6 to Day 7+ , which we consider to be a familiar condition , and T2 included Days 3–5 , an intermediate condition between novel and familiar . We examined 279 V1 and 628 CA1 cells at T1 , 173 V1 and 519 CA1 cells at T2 , and 212 V1 and 557 CA1 cells at T3 that were active on at least one trajectory . First , we found that there was a significant increase from T1 to T3 in the median overall firing rate of V1 cells ( 50% increase between T1 and T3; p < 0 . 0001 , Kruskal–Wallis test including all data at T1 , T2 and T3 ) , but not in the median firing rate of CA1 cells ( 3% increase between T1 and T3; p = 0 . 079; Figure 7A ) . Second , there was a significant increase in the median SMI for both V1 ( 55% increase between T1 and T3; p < 0 . 0001 ) and CA1 cells ( 39% increase between T1 and T3; p < 0 . 0001; Figure 7B ) , suggesting an experience-dependent increase in the location-specificity of V1 cells at the time scale of days . Third , we examined the correlation in Δrate and ΔCOM between overlapping V1-CA1 cell pairs at different time points ( N = 244 pairs at T1 , 108 pairs at T2 , 174 pairs at T3 ) . For the Δrate correlation , there was no significant change from T1 to T3 ( 2% decrease between T1 and T3; p = 0 . 20 , one-way ANOVA comparing all data at T1 , T2 and T3; Figure 7C ) . This is also true for the modified Δrate correlation after the speed and head direction modulation was removed ( 19% increase between T1 and T3; p = 0 . 13; Figure 7D ) . For the ΔCOM correlation , there was a small , but statistically non-significant decrease from T1 to T3 ( 17% decrease between T1 and T3; p = 0 . 11 , one-way ANOVA; Figure 7C ) . For the modified ΔCOM , the decrease was more prominent , but remained non-significant ( 33% decrease between T1 and T3; p = 0 . 088; Figure 7D ) . These data show that , although there was an indication of slightly decreased co-fluctuation between V1 and CA1 cells with experience , their functional interaction persisted as the animals became familiarized with the track . Taken together , day-to-day track experience was accompanied by an increase in the firing rate and location-specificity of V1 cells , which remained correlated in their lap-by-lap fluctuations with overlapping CA1 cells . 10 . 7554/eLife . 08902 . 015Figure 7 . Experience dependence of V1 and CA1 firing activities on the C-shaped track . ( A , B ) Cumulative distributions of overall firing rate ( A ) and SMI ( B ) of active V1 and CA1 cells on different days ( T1 – T3 , see texts for definition ) . ( C ) The average ( mean ± S . E . ) correlation in the lap-by-lap Δrate ( left ) and ΔCOM ( right ) fluctuations for overlapping V1-CA1 cell pairs on different days . ( D ) Same as C , but for the correlation in modified Δrate and ΔCOM after removing the modulation by speed and head direction . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 015 Finally , we asked whether the firing properties of V1 neurons differed across layers . We determined the laminar locations of the recorded neurons on the final recording day by performing small electrolytic lesions and subsequently examining the Nissl and acetylcholinesterase ( AChE ) staining ( Figure 1D ) . We estimated the locations of those neurons recorded on previous days by relative distances tetrodes had traveled . A neuron's location was assigned to either the superficial layers 2 and 3 ( L2/3 ) , layer 4 ( L4 ) , or the deep layers 5 and 6 ( L5/6 ) . We obtained 478 cells in L2/3 , 120 in L4 , and 83 in L5/6 that were active on a trajectory . A small number of V1 cells ( 95 out of 776 , 12% ) were not assigned to a layer due to failed AChE staining . We first analyzed the overall firing rate and SMI among the cells recorded in different layers ( Figure 8A , B ) on all days . There was a significant difference in firing rate across the layers ( p < 0 . 0001 , Kruskal–Wallis test ) . The post-hoc pair-wise comparisons revealed that L4 cells had significantly higher firing rates ( 10 . 2 [6 . 5 14 . 9] Hz ) than both L2/3 ( 3 . 2 [1 . 7 7 . 0] Hz , p < 0 . 0001 , ranksum test ) and L5/6 ( 2 . 9 [1 . 8 6 . 3] Hz , p < 0 . 0001 ) cells , whereas the latter two were similar ( p = 0 . 67 ) . The analysis of SMI shows that 86% of L2/3 , 78% of L4 , and 70% of L5/6 cells were location-responsive . The results indicate that L4 cells had the highest firing rate , whereas L2/3 cells had the highest percentage of location-responsive cells . Second , we computed the correlations in Δrate and ΔCOM for overlapping pairs of CA1 cells with the cells in each of the three cortical locations ( CA1-L2/3 , CA1-L4 , and CA1-L5/6 ) . We found that L2/3 , L4 and L5/6 cells were similarly correlated in Δrate and ΔCOM with CA1 cells ( data not shown ) . Third , we have shown that V1 cells increased their firing rates and SMIs with track running experience ( T1 to T3 , Figure 7A , B ) . Here we examined whether this was the case for cells in all layers of V1 . We found a significant increase in firing rate only in L2/3 cells ( p = 0 . 00023 , Kruskal–Wallis test comparing all data at T1 , T2 and T3; 39% increase between T1 and T3 ) , but not in L4 ( p = 0 . 60; 2 . 5% increase between T1 and T3 ) or L5/6 cells ( p = 0 . 10; 2 . 3% increase between T1 and T3; Figure 8C ) . However , there was a significant increase in SMI from T1 to T3 for cells in all layers ( L2/3: p < 0 . 0001 , 52% increase between T1 and T3; L4: p < 0 . 0001 , 130% increase between T1 and T3; L5/6: p = 0 . 05 , 54% increase between T1 and T3; Figure 8D ) . 10 . 7554/eLife . 08902 . 016Figure 8 . Layer differences in V1 . ( A , B ) Cumulative distributions of overall firing rate ( A ) and SMI ( B ) for active cells in L2/3 , L4 and L5/6 . ( C , D ) Cumulative distributions of firing rate and SMI for L2/3 , L4 , and L5/6 cells on different days ( T1–T3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08902 . 016
We have analyzed the firing activities of V1 and CA1 cells in freely moving rats as they ran back and forth on a C-shaped track for food reward . We found that a large number of cells in all layers of V1 , like CA1 place cells , fire selectively and reliably at specific locations of the track . These location-responsive V1 cells have multiple , narrow firing fields on a given trajectory of the track . The number of V1 firing fields rises and falls as animals approach and move away from visual landmarks , leading the rise and fall in the number of CA1 place fields . The precise lap-by-lap fluctuations in the firing rate and firing location of V1 cells are correlated with those of CA1 cells with overlapping firing fields . Finally , V1 cells display an experience-dependent change in their location-specificity at the time scale of days . These results reveal that V1 cells have location-specific responses to a particular environment and are functionally coupled with the hippocampal place cells encoding the same environment . Our data supports a functional interaction between V1 location-responsive cells and hippocampal place cells . Here we provide three pieces of evidence . First , the number of V1 firing fields fluctuates as the animal travels through a trajectory and this fluctuation is highly correlated with the change in the number of CA1 firing fields . More interestingly , we found that the V1 fields rise and fall earlier than the CA1 fields and this leading effect is consistent on both running directions . Second , the pair-wise cross-correlation analysis reveals a bias toward positive time lags , suggesting that , again , V1 location-specific activity appears earlier than the CA1 place field activity at the population level . Third , the lap-to-lap fluctuations in the firing rates and firing locations of individual V1 cells within their firing fields are significantly correlated with those of CA1 place cells with overlapping place fields . This is true even after removing the modulation of speed and head direction on firing rate and firing location . The correlated variation is specific to CA1-V1 cell pairs with the overlapping firing fields , suggesting a remarkably precise interaction at a fine temporal and spatial scale . These observed correlations are unlikely driven by non-specific common factors because of the following reasons . First , both the field distribution and pair-wise cross-correlation analyses show that V1 activities lead CA1 activities during trajectory running . Although the lead could be caused by CA1 and V1 cells independently responding to some unknown common factors with a delay , which we cannot rule out , the lead time of 60–120 ms , is consistent with our previous studies on the interaction between V1 and CA1 during slow-wave sleep ( Ji and Wilson , 2007; Haggerty and Ji , 2014 ) , when behavioral factors do not vary and the brain state is relatively uniform . Therefore , the delay likely reflects the propagation of activity from V1 to CA1 . Second , the lap-by-lap co-fluctuation of overlapping pairs is much higher than that of non-responsive pairs . Since both types of pairs share overlapping locations , if common factors were mainly responsible , the non-responsive pairs would show correlated fluctuations comparable to those of overlapping pairs . The leading of V1 activity over CA1 activity suggests a forward interaction , through which V1 location-responsive cells send visual information to CA1 place cells . The location-specific activity in V1 cells does not necessarily mean that V1 cells directly respond to places per se as hippocampal place cells do . Given what we know about the primary function of V1 neurons in behaving rodents ( Yao et al . , 2007; Niell and Stryker , 2010; Xu et al . , 2012 ) , they are likely driven by the visual cues associated with specific locations on the track , supported by our observation that V1 firing fields are more distributed around the landmarks . In particular , prospective firing may occur when V1 neurons ‘see’ landmarks in front of the animal . Retrospective firing may occur as V1 neurons ‘see’ landmarks behind the animal's head , which is possible because the side-facing eyes of rats have a wide visual field and can see visual stimuli behind or above the head ( Wallace et al . , 2013 ) , although other mechanisms such as feedback from the higher cortical areas or the hippocampus may also be responsible . The landmark information represented by the V1 cells can reach the hippocampus via a multi-synaptic pathway , which includes the secondary visual cortex , temporal cortex , perirhinal/postrhinal cortex , and entorhinal cortex ( Miller and Vogt , 1984; Vaudano et al . , 1991; Lavenex and Amaral , 2000; Furtak et al . , 2007 ) . In this way , the V1 location-responsive cells may allow local and distal visual cues to influence CA1 place cell activities ( Muller and Kubie , 1987; Lee et al . , 2004a , 2004b; Leutgeb et al . , 2005 ) . Our finding of location-responsive cells in the sensory area V1 raises the possibility that the distinction between what we call either a spatial or sensory response is not as clear as we thought . On one hand , given the apparent modulation of hippocampal place cells by visual cues ( Muller and Kubie , 1987; Lee et al . , 2004a , 2004b; Leutgeb et al . , 2005 ) , the response of a hippocampal cell to a particular place also contains a ‘sensory’ response to the visual cues associated with that place . In our data , the CA1 place fields are preferentially distributed around the landmarks , just as the V1 firing fields are , suggesting a bias of CA1 place cell activity by sensory input . On the other hand , the response of a V1 cell to visual clues at a particular place effectively establishes a ‘spatial’ response to the place , if the visual cues are stable in space . In this manner , the firing fields of V1 cells may appear to be similar to the CA1 place fields , with the important differences in that V1 cells exhibited less specificity by nature of non-zero baseline firing and multiple firing fields . The difference in V1 and CA1 location-specific activity is just that , because hippocampal place cell activities result from an integration of not just visual information , but also information from other sensory modalities and self-motion , hippocampal place fields are more specific and predictive of the animal's spatial location than the predominantly sensory-driven V1 firing fields . Therefore , it is possible that spatial representation in the brain may involve a gradual transformation along the anatomical pathway connecting the sensory cortex to the hippocampus , which converts a primarily sensory-driven , less reliable spatial response in the sensory cortex to a more precise and robust spatial response in the hippocampus . Hippocampal place cells are believed not just to form a neural representation of space , but encode spatial memory and other type of memories ( Eichenbaum et al . , 1999; Burgess and O'Keefe , 2003; Ferbinteanu et al . , 2006; Knierim at al . , 2006; Moser et al . , 2008 ) . If so , then the gradual transformation of spatial representation also suggests that a spatial memory trace is not only encoded by hippocampal place cells , but also by the cortical cells involved in the transformation , which are presumably distributed in many cortical areas . This idea is an essential component in many modern theories of memory , including index theory ( Teyler and Rudy , 2007 ) , two-stage memory processing theory ( Buzsaki , 1996 ) , and the theory of two complementary memory systems ( McClelland et al . , 1995 ) . All these theories propose the involvement of relevant cortical cells in long-term memory storage . Our previous study shows that activity patterns of V1 and CA1 cells during track running are coordinately replayed later during slow-wave sleep ( Ji and Wilson , 2007 ) , suggesting the involvement of V1 cells in memory consolidation . The current study provides another piece of key evidence that V1 cells and CA1 place cells functionally interact during the stage of memory formation as animals run a novel track for a first week or so . In our data , the lap-by-lap co-fluctuation between V1-CA1 cell pairs with overlapping firing fields is qualitatively similar to that between pairs of CA1-CA1 place cells with overlapping place fields . Since CA1 place cells with overlapping place fields are believed to constitute functional ‘cell assemblies’ that encode the spatial memory of the animal's environment ( Harris et al . , 2003; Dragoi and Buzsaki , 2006 ) , it is possible that the overlapping V1 location-responsive cells and hippocampal place cells are part of the larger ‘cortical-hippocampal assemblies’ that encode and store the same spatial memory trace . Therefore , we propose that the V1 location-specific activities are a substrate for encoding and storing the visual component of long-term spatial memories , and possibly episodic memories .
Fifteen Long-Evans rats , all male , 3–6 months old , were used in the electrophysiological recording experiments . The rats were first food deprived to ≥85% of their ad libitum weight and pre-trained for 5–10 days to run back and forth on a ∼2 m long linear track for milk reward delivered at both ends of the track . The rats were then placed back onto an ad libitum diet and went through a surgery for implanting a tetrode recording device ( see below ) . About 4 weeks after the surgery , animals were once again food deprived and recording began as the animals were placed on a novel C-shaped track ( Figure 1A ) , in a different room from the pre-training . The track , made of metal , was ∼3 m long and ∼6 cm wide with 3 cm high walls . The track was surrounded by a black curtain . No additional visual cues were added to the experimental set up . The only local cues were those inherent to the C-shaped metal track with 4 corners and low walls ( Figure 1A ) . During the recording , rats ran back and forth along 2 trajectories for milk remotely delivered at both ends of the track by syringe and tubing from outside the curtain . The animals were free to move along the track , but rewarded only after they reached one end from the other . The recording started on the very first day the animals experienced the track ( Day 1 ) and continued for 2–14 days , with 20–60 min each day . The experimental protocol was approved by the Institutional Committee on Animal Care at Baylor College of Medicine and followed National Institutes of Health guidelines . For each rat , we implanted a tetrode recording device ( tetrode drive ) , containing 24 independently movable tetrodes made of 4 fine nichrome wires ( diameter 13 μm ) , under isoflurane anesthesia . Twelve tetrodes were implanted into an exposure at coordinates anteroposterior ( AP ) −3 . 8 mm , mediolateral ( ML ) 2 . 5 mm from Bregma and the other 12 at AP -6 . 5 mm and ML 4 . 8 mm , for targeting CA1 and V1 , respectively . The tetrode drive was fixed to the skull with stainless anchoring screws and dental cement . The analgesic ketoprofen ( 5 mg/kg ) was administrated by subcutaneous injection before the animal was allowed to recover from the anesthesia . Tetrode recordings , by using a Digital Lynx system ( Neuralynx , Bozeman MT ) , were performed as previously described ( Ji and Wilson , 2007 , 2008 ) . Briefly , during the ∼4 weeks post surgery , tetrodes were slowly moved to the CA1 and V1 . Once stable single units ( spikes presumably from single neurons ) in CA1 were obtained , recordings started on Day 1 when the animals were introduced to the track and performed the track running task as described above . Once the recordings started on Day 1 , CA1 tetrodes were never moved again on later recording days , but V1 tetrodes were moved no more than 125 µm a day to sample cells in different layers of V1 . Some V1 tetrodes were already moved to deep layers before the recording started , in order to sample deep-layer cells on early days of track running . Signals recorded by each tetrode were filtered with a pass band of 600 Hz–9 kHz . Spikes were identified by a pre-set threshold of 50–70 µV and digitized at a sampling rate of 32 kHz . Two color diodes ( red , green ) were mounted over the animal's head to track the animals' position and head direction . Positions were sampled at 33 Hz with a resolution approximately 0 . 2 cm . Rat position was analyzed offline , x and y position was linearized , and animal speed was calculated as the linearized spatial distance between each 2 time points . Animals were euthanized using pentobarbital overdose ( ≥200 mg/kg ) after the tetrode recording . For each tetrode , an electric current ( 30 µA ) was passed for ∼15 s to create a small lesion at the tip of the tetrode . Brains were fixed in 10% formalin for at least 24 hr and then sectioned at 50 µm thickness . The sections were alternatively stained with either 0 . 2% Cresyl violet or 1% sodium sulfide nonahydrate ( Na2S9H2O ) to detect acetylcholineesterase ( AChE ) activity following the established protocol ( see Paxinos and Watson , 2007 for details in procedure and the agents used ) . The lesion sites were verified from the lesion marks in the stained sections ( Figure 1D ) , according to the standard rat brain atlas ( Paxinos and Watson , 2007 ) . At the implantation coordinates used here , area V1 is flanked laterally and medially by the secondary visual cortex V2 . V1 was delineated from surrounding cortical areas by a granulized layer 4 that exhibited denser staining in the Cresyl violet and the presence of a double band of darker AChE stain restricted to V1 in layers 4 and 5 . The presence of two dark AChE bands and the darker layer 4 Cresyl staining were also used to locate the laminar location of the lesion sites within V1 ( Zilles et al . , 1984 ) . Because CA1 tetrodes were never moved during the recording days , the locations of all CA1 cells were identified by the lesion sites . The locations of V1 cells recorded on the final recording day were also the same as the lesion sites in V1 . The locations of V1 cells recorded on previous days were estimated , based on the distances tetrodes traveled to the final locations . In two rats , the AChE staining procedure did not work . V1 neurons in these animals were not assigned to a layer . Single units were sorted using custom software ( xclust , M . Wilson at MIT , available at GitHub repository: https://github . com/wilsonlab/mwsoft64/tree/master/src/xclust ) on all data recorded across all recoding days . Since some tetrodes ware not moved from one day to another ( but could have drifted slightly overnight ) , it is difficult to know whether different or same cells were recorded across days from these tetrodes . Therefore , certain cells might be repeatedly sampled across days , but were included in the analysis because we would like to examine how V1 and CA1 activities changed across days ( Figures 7 and 8 ) . A total of 852 V1 and 3627 CA1 cells were recorded . For CA1 cells , only putative pyramidal cells ( firing rate <5 Hz ) that were active on at least one trajectory ( firing rate ≥0 . 5 Hz ) were included in the analysis ( Ji and Wilson , 2007 , 2008 ) . For V1 cells , all active cells ( firing rate >0 . 5 Hz ) were included . We did not exclude high-rate V1 cells because we did not find reliable indicators that high-rate V1 cells were a distinct cluster from low-rate V1 cells in our data . This criterion yielded a total of 776 V1 cells and 2033 CA1 cells across all the recording days , which were the basis for all of the analyses . Only the activities of these cells during active track running were analyzed . The activities at the food wells ( the final ∼10 cm at each end of the track ) or during the stopping periods on the track ( with running speed <6 cm/s for at least 0 . 5 s ) were excluded from the analysis . This exclusion is due to the fact that our purpose was to study the location responses of CA1 and V1 cells , but it is known that during stopping behavior CA1 cells fire in a different mode from active running and their activities become non-location-specific ( Foster and Wilson , 2006; Diba and Buzsaki , 2007; Cheng and Frank , 2008; Karlsson and Frank , 2009 ) . Results were expressed as median [25–75%] range values , mean ± S . E . values , or as otherwise specified . Accordingly , ranksum and Kuskal-Wallis tests were used for statistical comparisons with median values , t-test and ANOVA for comparisons with mean values , or as specified otherwise . p-values of the statistical tests were reported as exact values unless <0 . 0001 . | The brain is able to create and maintain a map of our surroundings as we go about our daily activities . It is thought that some of this spatial information is first processed in a region of the brain called the visual cortex , the information is then relayed to a region called the hippocampus , where the map is reliably stored . However , researchers do not fully understand how the brain transfers spatial information between these two regions . To explore what happens to the information , Haggerty and Ji recorded electrical signals from the brains of rats that were being trained to run back and forth along a C-shaped track . As they ran on the track , the rats gradually developed spatial maps of there surroundings . Electrodes were used to record brain signals from both the visual cortex and the hippocampus as this development took place . Similar to previous studies , analysis of the recordings showed that a specific population of neurons in the hippocampus , called CA1 neurons , produced an electrical signal whenever the rat ran past specific locations on the track , such as corners . Building on this Haggerty and Ji showed a population of neurons in the visual cortex called V1 neurons also produced location-specific electrical signals at the same time . Moreover , the electrical signals of these two populations of neurons fluctuate together in a coordinated fashion . The results of Haggerty and Ji support the idea that there exists a specific population of visual cortical neurons that communicate with hippocampal neurons in the development of a particular spatial map . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Activities of visual cortical and hippocampal neurons co-fluctuate in freely moving rats during spatial behavior |
Optimal decision-making requires balancing fast but error-prone and more accurate but slower decisions through adjustments of decision thresholds . Here , we demonstrate two distinct correlates of such speed-accuracy adjustments by recording subthalamic nucleus ( STN ) activity and electroencephalography in 11 Parkinson’s disease patients during a perceptual decision-making task; STN low-frequency oscillatory ( LFO ) activity ( 2–8 Hz ) , coupled to activity at prefrontal electrode Fz , and STN beta activity ( 13–30 Hz ) coupled to electrodes C3/C4 close to motor cortex . These two correlates differed not only in their cortical topography and spectral characteristics but also in the relative timing of recruitment and in their precise relationship with decision thresholds . Increases of STN LFO power preceding the response predicted increased thresholds only after accuracy instructions , while cue-induced reductions of STN beta power decreased thresholds irrespective of instructions . These findings indicate that distinct neural mechanisms determine whether a decision will be made in haste or with caution .
Fast decisions come at the cost of reduced accuracy . This elementary aspect of decision-making , often referred to as speed-accuracy trade-off , has been studied for over a century ( for a review see Heitz , 2014 ) and can be observed in a multitude of tasks and across various species including rats , non-human primates and humans ( Bogacz et al . , 2010; Forstmann et al . , 2010 , 2008; Hanks et al . , 2014; Heitz and Schall , 2012; Ivanoff et al . , 2008; Ratcliff and McKoon , 2008 , 1998; Reinagel , 2013; Schouten and Bekker , 1967; Thura and Cisek , 2016; van Veen et al . , 2008; Wickelgren , 1977 ) . For example , when approaching prey , the hunter must plan the advance in order to avoid mistakes . However , if deliberation takes too long , the prey might escape and the attempt would be to no avail . Therefore , it is crucial for intelligent agents to balance opposing demands of speed and accuracy . Mathematically , speed-accuracy adjustments can be implemented by elevating or lowering the ‘decision threshold’ , that is , a criterion which defines when the continuous accumulation of evidence should be terminated and the option with the strongest evidence chosen ( Bogacz et al . , 2006; Ratcliff and McKoon , 2008 ) . In neurobiological models of decision-making , such modulations of decision thresholds subserving speed-accuracy trade-offs have been assigned to the basal ganglia ( BG ) ( Bogacz et al . , 2010; Frank , 2006; Lo and Wang , 2006 ) , which are connected to a wide range of decision- and movement-related cortical areas in a closed-loop fashion ( Alexander et al . , 1986; Lambert et al . , 2012 ) . Within these loops , the BG exert tonic inhibition over cortical areas , which can be decreased through activation of a facilitatory , direct pathway connecting the striatum with BG output areas or increased by activation of two net inhibitory circuits passing through the subthalamic nucleus ( STN ) , which incorporate the indirect and hyperdirect pathways ( Alexander et al . , 1986; Bogacz et al . , 2010; Kravitz et al . , 2010 ) . Increased activity of cortical neurons computing decision-related signals during speed emphasis , as predicted by this model , has been observed as firing rate changes in non-human primates ( Hanks et al . , 2014; Heitz and Schall , 2012; Thura and Cisek , 2016 ) and indicated by functional magnetic resonance imaging ( fMRI ) studies in humans ( Forstmann et al . , 2008; Ivanoff et al . , 2008; van Veen et al . , 2008 ) . At the BG level , fMRI studies have inferred increased activity in striatum during speed emphasis ( Forstmann et al . , 2008; van Veen et al . , 2008 ) . An involvement of the STN has been suggested by studies showing that treatment with deep brain stimulation ( DBS ) of the STN affects patients’ ability to switch between fast and accurate decision-making ( Green et al . , 2013; Pote et al . , 2016 ) . However , these behavioral studies are non-informative regarding the neurophysiological correlates of speed-accuracy adjustments . To address this , we exploited the very high spatial and temporal resolution of local field potential ( LFP ) recordings directly from the STN while Parkinson’s disease patients who had undergone STN DBS surgery performed a perceptual decision-making task . Simultaneous recordings of electroencephalography ( EEG ) at electrodes C3 and C4 over or close to the motor cortex , and electrode Fz over the prefrontal cortex as well as computational modeling of participants’ latent decision-making parameters allowed us to relate modulations of STN activity and cortico-STN connectivity to adaptations of decision threshold during speed-accuracy adjustments .
Eleven patients with Parkinson’s disease performed a moving dots task , in which they had to decide whether a cloud of moving dots appeared to move to the left or to the right . Task difficulty was manipulated by changing the percentage of dots moving coherently in one direction ( 8% or 50% ) . Before the onset of the moving dots , subjects were either instructed to respond as quickly or as accurately as possible ( Figure 1A ) . While trials with low and high coherence were pseudorandomly interspersed , speed vs . accuracy instructions alternated in blocks of 20 trials ( Figure 1B ) resulting in a 2 ( low vs . high coherence ) * 2 ( speed vs . accuracy instructions ) design ( Figure 1C ) . An overview of the behavioral data is given in Figure 1D–E . Analysis of reaction times ( RT ) showed that participants were significantly faster during high compared to low coherence trials ( 974 ± 302 ms vs . 1478 ± 298 ms , main effect of coherence: F ( 1 , 10 ) = 83 . 284 , p<0 . 001 ) and when they were instructed to emphasize speed over accuracy ( 1146 ± 322 ms vs . 1306 ms ±259 ms , main effect of instruction: F ( 1 , 10 ) = 18 . 172 , p=0 . 002 ) . In addition , there was a significant interaction of instruction*coherence ( F ( 1 , 10 ) = 9 . 744 , p=0 . 011 ) , since speed instructions led to a stronger decrease in RT during low coherence compared to high coherence ( 180 ± 137 ms vs . 77 ± 113 ms decrease in RT , t ( 10 ) = 2 . 662 , p=0 . 024 ) . This difference was , however , not significant when considering % change in RT rather than the absolute difference ( t ( 10 ) = 1 . 414 , p=0 . 188 ) . Accuracy rates were significantly lower during low vs . high coherence trials ( 73 . 9% ± 9% vs . 95 . 7% ± 5 . 6% , main effect of coherence: F ( 1 , 10 ) = 165 . 107 , p<0 . 001 ) , while there was no significant effect of instruction ( 83 . 9% ± 8% during speed vs . 85 . 7% ± 7% during accuracy instructions , F ( 1 , 10 ) = 1 . 374 , p=0 . 268 ) nor an interaction of instruction*coherence ( F ( 1 , 10 ) = 1 . 152 , p=0 . 308 ) . To assess whether participants’ behavior was , nevertheless , in line with our a-priori hypothesis that subjects would have decreased decision thresholds during speed instructions , we analyzed the RT distribution of error trials in the speed and accuracy conditions . Previous studies have shown that during conditions with high thresholds , errors are primarily observed during slow responses ( Ratcliff and Rouder , 1998; see Materials and methods for more details ) . In line with this , we found that participants made significantly more errors during slow compared to fast trials after accuracy instructions ( 16 . 6% ± 10 . 4% decrease in accuracy rates during slow responses , t ( 10 ) = −4 . 731 , Pcorrected = 0 . 002 ) , but not during speed instructions ( 6% ± 17 . 3% , t ( 10 ) = −1 . 641 , Pcorrected = 0 . 264 ) . Thus , even though there was no overall effect of speed instructions on accuracy rates , which is compatible with previous studies showing little effects on trials with very high or very low coherence ( Hanks et al . , 2014; Ratcliff and McKoon , 2008 ) , our behavioral data were in line with the hypothesis that participants would decrease their decision thresholds during speed emphasis . To directly test this , we then applied the drift diffusion model ( DDM ) , which allows computation of the latent decision-making parameters underlying participants’ behavior . 10 . 7554/eLife . 21481 . 003Figure 1 . Paradigm and behavioral data . ( A–C ) Overview of experimental task and the 2*2 study design . ( D ) Histograms of RT distributions for correct ( blue ) and error ( red ) trials for all four conditions . ( E ) Second level comparison of reaction times ( left ) , accuracy rates ( middle ) and change in accuracy rates in slow vs . fast trials ( right ) . HC , high coherence; LC , low coherence; ** significant at p<0 . 01; ns , not significant . Error bars indicate standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 21481 . 00310 . 7554/eLife . 21481 . 004Figure 1—figure supplement 1 . Behavioral data of healthy participants . ( A ) Histograms of RT distributions for correct ( blue ) and error ( red ) trials for all four conditions . ( B ) Group-averaged reaction times and accuracy rates . HC , high coherence; LC , low coherence . Error bars indicate standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 21481 . 004 DDM includes three main parameters of interest . First , the drift rate v reflects the rate of sensory evidence accumulation . Second , the decision threshold a determines the amount of sensory evidence that needs to be accumulated before the choice is executed . During speed-accuracy adjustments , when speed is required the decision threshold is thought to be decreased requiring less evidence before responding ( Ratcliff and McKoon , 2008 ) . Of note , an increased baseline level is mathematically equivalent to decreased decision thresholds ( Figure 2A ) . Third , the non-decision time t reflects the portion of RT which is not directly related to the decision process , such as afferent delay , sensory processing and motor execution . In the current study , we assumed that drift rates were related to the coherence of the moving dots ( low vs . high coherence ) and thresholds were related to differences in task instructions ( speed vs . accuracy ) . Furthermore , the non-decision time was allowed to vary between coherence and instruction conditions . We fitted this simple model to the data using a hierarchical Bayesian estimation of DDM parameters ( HDDM ) and computed the posterior distribution of model parameters for statistical inference considering posterior probabilities ≥95% significant ( Wiecki et al . , 2013 ) . The model fitted the data well , as indicated by accurate predictions of the observed RT distributions in all four conditions ( Figure 2B ) . As expected , trials with low coherence had significantly lower drift rates than trials with high coherence ( 100% posterior probability ) and decision thresholds after speed instructions were significantly lower than after accuracy instructions ( >99% posterior probability , see Figure 2C ) . As a control analysis , we also assessed whether modulations of drift rates were related to changes in performance during speed vs . accuracy instructions , for example , due to increased attention to the stimuli . However , there was no effect of instruction on drift rates ( 67% posterior probability ) . Similarly , the non-decision time was neither modulated by instructions nor coherence ( both 75% posterior probability ) . Thus , the HDDM analysis confirmed our a-priori hypotheses that changes in coherence of the moving dots would selectively alter drift rates , while speed vs . accuracy instructions would be related to adaptations of decision thresholds . 10 . 7554/eLife . 21481 . 005Figure 2 . Drift diffusion modeling . ( A ) Schematic illustration of DDM . t is the non-decision time , v the drift rate and a the decision threshold . The upper boundary in the upper row indicates the threshold for the correct response , while the upper boundary in the lower row reflects the threshold for the incorrect response . Emphasizing speed over accuracy is thought to decrease the distance from the starting point of evidence accumulation to the decision threshold , which can be achieved by increasing the baseline ( left column ) or decreasing the boundary ( right column ) . Both mechanisms are mathematically equivalent and cannot be distinguished in the DDM framework . ( B ) Quantile probability plots showing the observed ( x ) and predicted ( ellipses ) RT against their cumulative probabilities ( 10 , 30 , 50 , 70 and 90 percentiles ) . The widths of the ellipses represent uncertainty ( standard deviation of the posterior predictive distribution ) . Blue symbols are used for correct , red symbols for incorrect trials ( incorrect trials are only shown for low coherence trials ) . ( C ) Posterior probability densities for changes in decision thresholds by instruction and changes in drift rates by coherence levels . Both effects were highly significant . DOI: http://dx . doi . org/10 . 7554/eLife . 21481 . 00510 . 7554/eLife . 21481 . 006Figure 2—figure supplement 1 . Drift diffusion modeling in healthy participants . ( A ) Quantile probability plots showing the observed ( x ) and predicted ( ellipses ) RT against their cumulative probabilities ( 10 , 30 , 50 , 70 and 90 percentiles ) . The widths of the ellipses represent uncertainty ( standard deviation of the posterior predictive distribution ) . Blue symbols are used for correct , red symbols for incorrect trials ( incorrect trials are only shown for low coherence trials ) . ( B ) Posterior probability densities for changes in decision thresholds by instruction and changes in drift rates by coherence levels . DOI: http://dx . doi . org/10 . 7554/eLife . 21481 . 006 To confirm that the observed behavior in Parkinson’s disease patients resembled ‘physiological’ task performance , we additionally conducted the same task in 18 healthy age-matched participants ( age of healthy participants: range 28–75 y , mean age 61 ± 16 y; age of Parkinson’s disease patients: range 31–75 y , mean age 57 ± 12 y; difference between groups: t ( 27 ) = −0 . 675 , p=0 . 505 ) . In these healthy participants , RT were significantly faster in high compared to low coherence trials ( 652 ± 124 ms vs . 1238 ± 334 ms , main effect of coherence: F ( 1 , 17 ) = 65 . 218 , p<0 . 001 ) and after speed compared to accuracy instructions ( 879 ± 194 ms vs . 1011 ± 219 ms , main effect of instruction: F ( 1 , 17 ) = 57 . 436 , p<0 . 001 , see Figure 1—figure supplement 1 ) . There was also an interaction instruction*coherence ( F ( 1 , 17 ) = 15 . 803 , p=0 . 001 ) , since RT decreased more strongly after speed instructions in low compared to high coherence trials ( 183 ± 121 ms vs . 81 ± 47 ms decrease in RT , t ( 17 ) = 3 . 924 , p=0 . 001 ) , which , however , did not remain significant when considering % change in RT ( t ( 17 ) = 1 . 548 , p=0 . 140 ) . Importantly , neither the effect of coherence nor the effect of instruction on RT differed between patients and healthy controls when directly comparing the groups ( effect of coherence t ( 27 ) = −0 . 793 , p=0 . 435; effect of instruction t ( 27 ) = 0 . 809 , p=0 . 425 ) . Accuracy rates were lower in low compared to high coherence trials ( 98 . 9% ± 2 . 5% vs . 81 . 8% ± 6 . 6% , main effect of coherence: F ( 1 , 17 ) = 355 . 647 , p<0 . 001 ) , while there was no significant effect of instruction ( 89 . 9% ± 4 . 1% after speed vs . 90 . 9% ± 4 . 5% after accuracy instructions , main effect of instruction: F ( 1 , 17 ) = 2 . 193 , p=0 . 157 ) , nor an interaction instruction*coherence ( F ( 1 , 17 ) = 0 . 599 , p=0 . 450 ) . Again there were no differences in the effect of coherence or instruction between patients and healthy participants ( effect of coherence t ( 27 ) = −0 . 460 , p=0 . 649; effect of instruction t ( 27 ) = 0 . 418 , p=0 . 679 ) . We also tested whether there were differences in task-related changes in the latent decision-making parameters using HDDM ( Figure 2—figure supplement 1 ) . As in patients , we found that low coherence trials had significantly lower drift rates compared to high coherence trials ( 100% posterior probability ) . Speed instructions significantly reduced thresholds compared to accuracy instructions ( >99% posterior probability ) , but had no significant effects on drift rates ( 79% posterior probability ) or non-decision times ( 76% posterior probability ) . When comparing groups , we found that changes in coherence affected drift rates more strongly in healthy participants compared to patients ( 99% posterior probability ) . Importantly , however , there were no differences between groups regarding adjustments of decision thresholds ( 51% posterior probability ) , that is , patients and healthy participants changed their thresholds to a similar extent between speed and accuracy instructions . Together , these findings show that task effects on performance and decision threshold adjustments were comparable between patients and healthy participants . In the next step , we aimed to elucidate the neural correlates of these effects by analyzing task-related changes in STN activity obtained from invasive LFP recordings in Parkinson’s disease patients . Aligning STN power to the onset of the motor responses showed the well-known decrease in 13–30 Hz beta power around the motor response ( from ~250 ms before until 250 ms after the response ) , which was followed by an increase in beta power from ~500 ms until 1000 ms after the response ( Figure 3A ) . LFO power of 2–8 Hz showed an inverse pattern with a pronounced increase in power around the time of the response . 10 . 7554/eLife . 21481 . 007Figure 3 . Response-aligned STN power changes . ( A ) Time frequency spectrum aligned to the onset of the motor response from −1 . 5 to +1 . 5 s averaged across conditions . ( B ) Spectra shown separately for the four conditions . The color map is identical to A . ( C ) Significant differences between conditions as revealed by cluster-based permutation tests . DOI: http://dx . doi . org/10 . 7554/eLife . 21481 . 00710 . 7554/eLife . 21481 . 008Figure 3—figure supplement 1 . STN channel selection . Example of channel selection for one representative patient ( patient #6 ) . Response-aligned time frequency spectra are shown for the three bipolar channels ( from left to right: ventral -> middle -> dorsal ) for both STNs ( left STN in upper row , right STN in lower row ) . The channel with the strongest LFO increase was picked if ( i ) it was not further than two contacts away from the channel with the strongest beta modulation ( only relevant for omnidirectional octopolar electrodes ) and ( ii ) also showed a pronounced beta modulation . In this patient , the most ventral contact was chosen for left STN and the middle contact for right STN . Note that the most ventral contact of right STN also showed LFO modulation , but only very little beta modulation . Therefore , the middle contact was chosen for the right STN . DOI: http://dx . doi . org/10 . 7554/eLife . 21481 . 00810 . 7554/eLife . 21481 . 009Figure 3—figure supplement 2 . Differences in STN power between correct and incorrect trials . Response-aligned time frequency spectra are shown in the left column , while spectra aligned to the onset of the moving dots are shown in the right column for correct ( upper row ) and incorrect ( lower row ) low coherence trials . There were no statistically significant differences between correct and incorrect responses using permutation testing ( 1000 permutations with p=0 . 05 as cluster-building and statistical threshold for cluster-based comparisons ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21481 . 009 We found two neural correlates of speed-accuracy adjustments . First , when comparing speed and accuracy conditions , LFO power was significantly stronger after speed compared to accuracy instructions in the time period preceding the response ( p<0 . 05 ) , while there was no difference between trials with low and high coherence ( p>0 . 05 , see Figure 3B–C ) . The pre-response increase in LFO power during speed instructions was also observed when aligning the spectra to the onset of the moving dots cue with a significant cluster starting ~500 ms after dots onset ( p<0 . 001 ) , see Figure 4A–C . To assess whether the observed changes in LFO power were more strongly related to the onset of the moving dots cue or the response , we directly compared the mean LFO power during the respective time windows ( from 750 ms before the response until the response vs . 500 ms until 1250 ms after the cue ) after speed and accuracy instructions in a 2*2 ANOVA . Except from the already known main effect of instruction ( F ( 1 , 10 ) = 10 . 941 , p=0 . 008 ) , this analysis did not show any significant differences between the cue- and response-related changes ( effect of alignment: F ( 1 , 10 ) = 0 , p=0 . 993 , interaction alignment*instruction: F ( 1 , 10 ) = 0 . 93 , p=0 . 358 ) . Furthermore , in the considered time periods , there was no significant phase consistency of LFO across trials ( intertrial-phase clustering , see Material and methods ) in the cue- or response-aligned data neither after speed nor accuracy instructions ( all p>0 . 05 ) . This indicates that the observed differences in LFO power did not have strong cue- or response-evoked components . 10 . 7554/eLife . 21481 . 010Figure 4 . STN power changes aligned to the onset of the moving dots . ( A ) Time frequency spectrum aligned to the onset of the cue from −0 . 5 to +1 . 5 s averaged across conditions . ( B ) Spectra shown separately for the four conditions . The color map is identical to A . ( C ) Significant differences between conditions as revealed by cluster-based permutation tests . ( D ) Cue-induced decrease in beta power ( 13–30 Hz ) averaged across conditions . The time series of each trial was capped at the time of the response before averaging . A decrease in beta power was evident from ~150 ms to 400 ms after the onset of the moving dots . ( E ) This beta decrease was stronger in speed vs . accuracy instructions ( but not high vs . low coherence ) . For D and E and solid traces represent mean , while shaded areas around the traces indicate standard error of the mean . * , significant at p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 21481 . 010 We also found a correlate of speed-accuracy adjustments in the beta frequency band . For the cue-aligned spectra , there was a significant effect of coherence in the beta band with increased beta power from ~500 ms to 1000 ms after the cue and decreased beta power from ~1000 ms to 1500 ms after the cue in the low compared to high coherence trials ( p<0 . 001 ) , see Figure 4C . However , there was a very strong RT difference between the low and high coherence conditions ( see dotted vertical lines in Figure 4B ) . Thus , the difference in the beta power might just reflect that responses took place later in low coherence trials . To address this effect of RT differences on STN beta power , we conducted an additional analysis based on ( Hanks et al . , 2014 ) , in which we focused on the early cue-induced reduction in beta power ( see Materials and methods for more details ) . This decrease was evident in all conditions ( Figure 4D ) and took place well before the second decrease in beta around the time of the response . ANOVA showed that the cue-induced beta decrease was more pronounced after speed compared to accuracy instructions ( 11 . 8% ± 5 . 8% vs . 6 . 9% ± 3 . 3% , main effect of instruction: F ( 1 , 10 ) = 6 . 619 , p=0 . 028 , Figure 4E ) , while there was no effect of coherence ( F ( 1 , 10 ) = 0 . 177 , p=0 . 683 ) or an interaction of instruction*coherence ( F ( 1 , 10 ) = 2 . 525 , p=0 . 143 ) . Together , we detected two correlates of speed-accuracy adjustments in the STN as reflected by task-related power changes; an enhanced cue-induced decrease in beta power from ~150 ms to 400 ms after onset of the moving dots followed by a stronger increase in LFO power starting ~750 ms before the response in speed compared to accuracy instructions . To further investigate these two features , we extracted single trial estimates of pre-response STN LFO power ( from 750 ms before the response until the response ) and the cue-induced beta power decrease ( from 150 ms to 400 ms after the cue ) and z-scored these values by subtracting the mean and dividing by the standard deviation for each subject ( Frank et al . , 2015; Herz et al . , 2016 ) , as in Figure 5A . Interestingly , these single trial changes in the LFO and beta band were not statistically related to each other as indicated by correlation analysis ( mean rho = −0 . 006 ± 0 . 042 , t ( 10 ) = −0 . 45 , p=0 . 662 , visualized in Figure 5B ) . Next , we aimed to test whether these changes were related to adjustments of decision thresholds as proposed by computational models of decision-making . 10 . 7554/eLife . 21481 . 011Figure 5 . STN power changes predict adjustments of decision thresholds . ( A ) Histograms of the z-scored single trial values of the increase in LFO before the response ( −750 ms before the response until the response ) and cue-induced decrease in beta power ( 150 ms to 400 ms after the cue ) for all subjects combined . Black bars represent trials with accuracy instructions , green bars trials with speed instructions . ( B ) Scatter plot of LFO and beta power single trial values illustrating the lack of a correlation between the two . Note that the statistical test for this analysis was based on a one-sample t-test of Fisher r-to-z-transformed within-subject correlation coefficients . ( C ) Posterior probability density for the effect of LFO on decision thresholds ( left column ) , and the effect of LFO on thresholds separately for speed ( green ) and accuracy ( black ) instructions ( middle column ) , which is further illustrated in the right column . ( D ) Posterior probability density for the effect of the cue-induced beta decrease on decision thresholds , which is further illustrated in the right column . Solid lines indicate significant results ( ≥95% of posterior distribution different from zero ) , while dotted lines indicate non-significant results . DOI: http://dx . doi . org/10 . 7554/eLife . 21481 . 011 Since we hypothesized that STN power changes related to speed-accuracy adjustments would modulate decision thresholds we entered single-trial estimates of STN LFO and beta power as predictors of decision thresholds into the HDDM regression analysis ( see Materials and methods ) . This revealed a significant effect of LFO ( 99% posterior probability ) as well as an interaction of LFO*instruction ( 95% posterior probability ) , see Figure 5C . Post-hoc tests showed that LFO predicted elevated decision thresholds only after accuracy instructions ( 96% posterior probability ) , but not after speed instructions ( 63% posterior probability ) . Thus , even though LFO was increased after speed instructions , it only predicted modulations of decision thresholds when subjects responded more cautiously after accuracy instructions ( right column in Figure 5C ) . Cue-induced beta power decreases predicted lower decision thresholds ( 95% posterior probability ) , but there was no interaction of beta*instruction ( 59% posterior probability , Figure 5D ) . Thus , in contrast to LFO power changes , the relationship between beta power decreases and decision thresholds did not depend on speed vs . accuracy instructions . Importantly , the relationship between beta power decreases and thresholds was specific to the early cue-induced decrease in beta , since the averaged pre-response beta power ( computed analogously to pre-response LFO power ) did not predict threshold adjustments ( 54% posterior probability ) , which is in line with previous observations ( Herz et al . , 2016 ) . As a further control analysis , we assessed whether STN power changes were specific to modulations of decision thresholds by conducting the same regression analysis using drift rates as dependent variable . This analysis showed that there were no relationships between STN LFO or beta power changes and trial-by-trial fluctuations in drift rates ( effect of LFO: 69% posterior probability , LFO*coherence: 73% posterior probability , beta-decrease: 53% posterior probability , beta-decrease*coherence: 53% posterior probability ) . Finally , the HDDM incorporating STN power changes also improved model performance compared to the HDDM not containing any neural data ( difference in deviance information criterion: 485 ) . To test whether the relationship between trial-wise changes in STN activity and decision thresholds were also reflected in simple behavioral measures , we conducted additional regression analyses using reaction times as dependent variable and STN activity ( LFO and beta power , see above ) as predictors . We found that the pre-response increase in LFO power predicted increased RT only after accuracy ( slope: 0 . 038 , p=0 . 022 ) , but not after speed instructions ( slope: −0 . 006 , p=0 . 366 ) . In addition , stronger cue-induced decreases in beta power predicted shorter RT across instructions ( slope: −0 . 022 , p=0 . 035 ) . These findings show that similar effects of STN activity can be detected using either DDM or simple behavioral measures . However , applying DDM allowed us to relate STN activity changes to a specific latent decision-making parameter , that is , the decision threshold . This is not possible through correlations with behavioral measures alone , since RT is not only influenced by changes in decision thresholds , but also the rate of evidence accumulation and processes not directly related to the decision process ( including motor execution delays ) . Together , combining recordings of STN power changes during speed-accuracy adjustments with HDDM demonstrated that the observed changes in the LFO and beta band both significantly contributed to trial-by-trial modulations of decision thresholds . Since STN oscillations with different spectral properties have been related to distinct oscillatory cortico-STN networks ( Litvak et al . , 2011 ) , we aimed to investigate whether the observed changes in STN LFO and beta power were related to modulations of different cortico-STN connections in the final part of the analysis . While cortical oscillatory activity in the beta band is a hallmark of motor networks ( Litvak et al . , 2011; Pfurtscheller , 1981; 2003 ) , cortical LFO activity during ‘cognitive’ motor tasks has mainly been observed in PFC ( Cavanagh et al . , 2011 , 2012; Cohen , 2014b ) . Therefore , we assessed whether task-related modulations of the beta and LFO rhythms were predominantly expressed at scalp electrode Fz ( over PFC ) vs . scalp electrodes C3/C4 ( over or close to motor cortex ) as a first step ( Figure 6A and B ) . EEG activity recorded at Fz showed a stronger pre-response LFO increase compared to electrodes C3/C4 ( 2 . 7% ± 4 . 4% vs . −1 . 1 ± 1 . 3% , p<0 . 05 ) . Conversely , activity at C3/C4 displayed a more pronounced cue-induced decrease in beta power compared to Fz ( 12% ± 8% vs . 4 . 3% ± 1 . 8% , p<0 . 001 ) . 10 . 7554/eLife . 21481 . 012Figure 6 . Task-related changes in cortical activity and cortico-STN connectivity . ( A ) Time frequency spectrum for EEG at electrode Fz aligned to the motor response ( from −1 . 5 to +1 . 5 s ) and cue-onset ( from −0 . 5 to 1 . 5 s ) averaged across conditions . The pre-response increase in LFO was significantly stronger at Fz compared to electrode C3/C4 ( p<0 . 05 ) . ( B ) Time frequency spectrum for EEG at electrode C3/C4 aligned to the motor response ( from −1 . 5 to +1 . 5 s ) and cue-onset ( from −0 . 5 to 1 . 5 s ) averaged across conditions . The cue-induced beta decrease was significantly stronger at C3/C4 compared to Fz ( p<0 . 001 ) . ( C ) Inter-site phase clustering ( ISPC ) between LFO at Fz and STN . The left column shows ISPC separately for speed ( green ) and accuracy instructions ( black ) . In C and D , the shaded rectangles indicate the time windows of interest , which were based on the analysis of STN power changes . Solid traces represent mean and shaded areas around the traces indicate standard error of the mean . The middle and right columns show correlation analyses between Fz-STN LFO ISPC and STN LFO power changes for accuracy and speed instructions , respectively . Each observation corresponds to one patient ( n = 8 ) . ( D ) Same as C , but for C3/C4-STN ISPC in the beta band . ** , significant at p<0 . 01; ns , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 21481 . 012 To directly assess whether oscillations in the beta and LFO band observed in the STN might be related to changes in cortico-STN connectivity , we then computed the inter-site phase clustering ( ISPC ) , a phase-based connectivity measure , which indicates how reliably the phases of oscillatory activity in two areas are aligned across trials ( Cohen , 2014a; Herz et al . , 2016; Zavala et al . , 2014 ) . We assessed whether cortico-STN phase coupling was different after speed vs . accuracy instructions and whether inter-individual differences in cortico-STN connectivity were correlated with differences in STN power during speed and accuracy instructions . The latter analysis was important in order to test if the context-specific relationship between STN-LFO power and decision thresholds , which was only observed after accuracy instructions , might be related to an increased influence of the PFC underlying Fz on the STN during modulation of decision thresholds when participants are more cautious ( i . e . after accuracy instructions ) . This is predicted by neurobiological models of decision-making ( Frank , 2006; Ridderinkhof et al . , 2011 ) and suggested by previous studies ( Frank et al . , 2015; Herz et al . , 2016 ) . Fz-STN ISPC in the LFO band was not significantly different between speed and accuracy instructions in the pre-response period ( p>0 . 1 ) , see Figure 6C . However , correlations between Fz-STN ISPC and STN LFO power were significantly different between speed and accuracy instructions ( z ( 7 ) = −2 . 33 , Pcorrected = 0 . 04 ) . While there was no relationship between Fz-STN ISPC and STN power after speed instructions ( rho = −0 . 301 , Pcorrected = 0 . 938 ) , there was a significant positive correlation between Fz-STN coupling and STN power in the LFO band after accuracy instructions ( rho = 0 . 823 , Pcorrected = 0 . 024 ) , see Figure 6C . Thus , only after accuracy instructions , in which single-trial STN LFO power predicted adjustments of decision thresholds , were inter-individual differences in STN LFO power closely related to differences in Fz-STN phase coupling . This relationship between Fz-STN LFO phase coupling and STN LFO power was absent after speed instructions , similar to the lack of a relationship between single-trial STN LFO power and trial-wise modulations of decision thresholds when speed was emphasized . C3/C4-STN ISPC in the beta band showed a pronounced decrease after onset of the cue only after speed , but not accuracy instructions ( Figure 6D ) , leading to a significant difference between conditions ( p<0 . 01 ) . Importantly , this difference is unlikely to be a simple consequence of the stronger cue-induced STN beta power decrease after speed vs . accuracy instructions , since analyzing ISPC 300 ms after the onset of moving dots , in which there was no difference between instructions in STN beta power ( see Figure 4E ) , still yielded a significant difference in C3/C4-STN ISPC ( p<0 . 05 ) . Correlation analyses revealed that the relationship between C3/C4-STN beta ISPC and STN beta power was significantly different between speed and accuracy instructions ( z ( 7 ) = −2 . 65 , Pcorrected = 0 . 016 ) . This was driven by a significant positive correlation between C3/C4-STN ISPC and STN beta power after speed instructions ( rho = 0 . 832 , Pcorrected = 0 . 02 ) , while there was no significant correlation after accuracy instructions ( rho = −0 . 447 , Pcorrected = 0 . 534; Figure 6D ) when STN may have been hijacked by coupling with Fz .
In the current study , we combined recordings of STN LFP and cortical EEG during perceptual decision-making with computational modeling of the latent parameters underlying the decision process . This allowed us to demonstrate neural correlates of the involvement of STN and of cortico-STN connectivity in controlling the trade-off between fast and accurate decisions . We found two distinct , statistically-uncorrelated mechanisms comprising modulations of a Fz-STN network in the LFO band and a C3/C4-STN network in the beta band . These two networks differed in their cortical topography , relative timing of recruitment and spectral characteristics as well as in their precise relationship with decision thresholds . The Fz-STN network predicted increased thresholds only after accuracy instructions , while the C3/C4-STN network predicted decreased thresholds irrespective of instructions , but was more strongly modulated during speed emphasis . Electrophysiological recordings of STN LFPs and EEG have shown that LFO modulations during ‘cognitive’ motor tasks are mainly related to cortical PFC activity and PFC-STN connectivity ( Cavanagh et al . , 2011 , 2012; Cohen , 2014b; Herz et al . , 2016; Pastötter et al . , 2012; Zavala et al . , 2014 ) . More specifically , Pastötter et al . ( 2012 ) demonstrated that changes in cortical LFO power are mainly localized in the medial PFC using EEG in healthy participants performing a speed-accuracy trade-off task . Furthermore , Frank et al . ( 2015 ) showed that trial-by-trial variations in Fz LFO measured using EEG correlated with activity of the medial PFC measured using fMRI and that both signals were related to increases in decision thresholds . These findings add to the increasing evidence that the PFC contributes to decision-making and modulations of the speed-accuracy trade-off ( Aron et al . , 2016; Bogacz et al . , 2010; Frank , 2006; Heekeren et al . , 2008 ) . Within the PFC , mainly – albeit not exclusively - dorsomedial PFC seems to play a pivotal role during context-dependent adaptations of decision-making and motor control ( Forstmann et al . , 2010 , 2008; Frank et al . , 2015; Herz et al . , 2014; Ivanoff et al . , 2008; van Veen et al . , 2008; Wenzlaff et al . , 2011 ) . Importantly , dorsomedial PFC has anatomical connections to the STN via the indirect and hyperdirect pathway ( Alexander et al . , 1986; Aron et al . , 2007; Forstmann et al . , 2010; Lambert et al . , 2012 ) . Here , we found that STN LFO power was increased prior to the response after speed compared to accuracy instructions . However , it was only predictive of elevated thresholds after accuracy instructions and , similarly , the extent to which Fz and STN LFO fell into register ( as reflected by phase coupling ) only correlated with variations in STN LFO power after accuracy instructions . These observations are in line with the idea that PFC increases its influence over STN activity when caution is warranted ( Frank , 2006; Ridderinkhof et al . , 2011 ) leading to increased decision thresholds in order to delay the response ( Frank et al . , 2015; Herz et al . , 2016 ) . Further evidence for this hypothesis has been provided by previous studies assessing ‘conflict’ between competing responses . Using the Flanker task Zavala and colleagues ( 2013 , 2015 , 2016 ) found increased STN LFO power in trials with distracting stimuli compared to trials without such conflict . Similar changes in STN LFO power have been observed during a Stroop task ( Brittain et al . , 2012 ) . Since response conflict has been shown to increase decision thresholds ( Ratcliff and Frank , 2012 ) , these previously reported LFO changes might be related to threshold adjustments . However , LFO activity – at least when measured over PFC using EEG - cannot be simply mapped to one specific process or mechanism , since it is not only modulated by conflict , but various processes including novelty , punishment , error , memory and learning ( Cavanagh et al . , 2012; Cohen , 2014b ) . Thus , the increased STN LFO power after speed instructions observed in the current study could be related to a variety of processes , such as the implementation of a non-default mode of responding ( Fleming et al . , 2010 ) assuming that participants’ natural tendency was to weigh accuracy over speed ( Forstmann et al . , 2008 ) . Importantly , however , this change in LFO power after speed emphasis was not related to adjustments of decision thresholds nor to PFC-STN phase coupling indicating either that cortical areas other than dorsomedial PFC were increasing their influence over STN or that communication between PFC and BG were mainly relayed through the striatum instead of STN for threshold adjustments during the speed regime ( Forstmann et al . , 2008; Ivanoff et al . , 2008; van Veen et al . , 2008 ) . We found a separate neural correlate of speed-accuracy adjustments , namely an immediate cue-induced decrease in STN beta power , which was more pronounced during speed emphasis and predicted lower decision thresholds irrespective of the type of instruction . Furthermore , variations in STN beta power after speed instructions were closely related to the extent to which beta phase coupling between C3/C4 and STN decreased after the cue indicating that the stronger STN beta power decrease during speed emphasis was related to modulations of C3/C4-STN connectivity . It should be noted that concomitant changes in power can affect measures of phase coupling , since low power can render phase estimates unreliable underestimating the consistency of phase-alignment between areas . Thus , the decreased phase coupling after speed instructions should be interpreted with caution , even though we found that C3/C4-STN beta phase coupling was already decreased when STN beta power showed no differences between speed and accuracy instructions ~300 ms after the cue . At the cortical level , beta power modulations during movements ( Pape and Siegel , 2016; Pfurtscheller , 1981; 2003 ) , decision-making ( Donner et al . , 2009; Wyart et al . , 2012 ) and speed-accuracy trade-offs ( Pastötter et al . , 2012 ) have been localized to motor areas . Furthermore , STN beta oscillations are coherent to motor cortical areas during rest ( Hirschmann et al . , 2011; Litvak et al . , 2011 ) . Thus , even though conclusive evidence can only be provided through simultaneous STN and focal cortical recordings , for example , using electrocorticography , the observed beta band changes in STN and at C3/C4 in the current study are likely to be related to modulations of a motor cortical-STN network adding to the increasing evidence that ‘decision-related’ signals can be observed in areas involved in motor preparation and execution ( Cisek and Kalaska , 2010; Donner et al . , 2009; Klein-Flügge and Bestmann , 2012; Pastötter et al . , 2012; Thura and Cisek , 2016 ) . One role of motor cortical-STN beta modulations during speed-accuracy adjustments might be related to the amount of vigor invested in the response . In the current study , we did not measure movement velocity or force . However , previous studies have provided strong evidence that the BG in general ( Desmurget and Turner , 2010; Yttri and Dudman , 2016 ) , and variations in STN beta power in particular ( Tan et al . , 2015 , 2016 ) , are related to encoding movement vigor . During speed-accuracy adjustments speed emphasis not only decreases decision times , but also increases movement vigor to indicate the choice ( Spieser et al . , 2016; Thura et al . , 2014 ) . Furthermore , the longer is spent on deliberation the less time is spent on performing the motor response ( Thura and Cisek , 2016; Thura et al . , 2014 ) . Such close relationships between ‘cognitive’ and ‘motor’ aspects of a decision have led to the hypothesis that a common signal might underlie adjustments in the speed of decision and vigor of movements ( Thura and Cisek , 2016; Thura et al . , 2014 ) . Future studies are needed to test the intriguing possibility that modulations of cortico-STN networks during speed-accuracy adjustments do not only mediate changes in decision time , but also related changes in movement vigor . How can changes in cortico-STN connectivity lead to adjustments between fast and accurate decisions ? Recordings of cortical single-unit activity in non-human primates have shown that when speed is emphasized , neurons integrating sensory evidence exhibit an increase in the baseline and gain of firing rates during deliberation , rather than terminating at lower levels of firing rates before the choice is executed ( Hanks et al . , 2014; Heitz and Schall , 2012; Thura and Cisek , 2016 ) . These findings indicate that a shift in baseline , but not a change in the threshold ( which is mathematically equivalent , see also Figure 2A ) , underlies speed-accuracy adjustments at the neural level . Such changes in cortical activity and excitability could be mediated through the BG by modulating tonic feedback inhibition of the cortex ( Alexander et al . , 1986; Bogacz et al . , 2010; Kravitz et al . , 2010 ) putatively in the form of a dynamic ‘urgency’ signal ( Churchland et al . , 2008; Cisek and Kalaska , 2010; Drugowitsch et al . , 2012; Thura and Cisek , 2016; Thura et al . , 2014 ) . Interestingly , changes in cortical firing rates during speed-accuracy adjustments have been observed in multiple areas comprising motor and premotor cortex ( Thura and Cisek , 2016 ) , parietal cortex ( Hanks et al . , 2014 ) and frontal eye field ( Heitz and Schall , 2012 ) , all of which share connections with the STN ( Lambert et al . , 2012 ) . Thus , context-dependent changes in STN activity could adjust cortical activity ( e . g . baseline firing rates ) simultaneously in multiple , spatially remote areas according to current task demands . As mentioned earlier , cortex and basal ganglia are interconnected through circuits , which comprise not only cortico–STN connections , but involve a multitude of subcortical areas ( Alexander et al . , 1986; Kravitz et al . , 2010; Lambert et al . , 2012 ) . Therefore , we do not assume that during speed-accuracy trade-offs neural activity changes within the basal ganglia are limited to the STN . Behavioural adjustments might ultimately be implemented by modulations of cortico-basal ganglia network dynamics through multiple pathways . Since spike-based communication can be optimized regarding efficiency and selectivity through synchronized oscillations ( Fries , 2015 ) , we propose that the changes in oscillatory activity observed in this study reflect modulations of distinct neural communication channels facilitating adjustments between hasty and cautious decision regimes . Together , our results suggest that distinct mechanisms are employed in the brain to facilitate speed-accuracy adjustments in cortico-basal ganglia networks . A better understanding of such mechanisms might render it possible to focus therapeutic interventions on specific neural circuits in order to improve treatment of neurological disorders in the future .
Eleven patients with Parkinson’s disease ( PD , ten males , mean disease duration 7 . 8 years ± 2 . 9 standard deviation ( SD ) , mean age 56 . 7 years ± 11 . 7 SD ) , who had undergone deep brain stimulation ( DBS ) surgery of the bilateral subthalamic nucleus ( STN ) 2–5 days prior to the recordings , were enrolled in the study . For more clinical details of the patients and specifications of the inserted electrodes please see supplementary file 1 . Lead localization was verified by stereotactic intraoperative magnetic resonance imaging or by monitoring the clinical effect and side effects during operation and immediate postoperative stereotactic computerized topography . Recordings from bilateral STN were performed through externalized electrode extension cables in the time period between electrode insertion and implantation of the subcutaneous pacemaker approximately 1 week after the first operation . In order to approximate physiological STN function as well as possible all patients were tested on their normal dopaminergic medication . All patients had a good dopamine response as indicated by a pronounced improvement of motor function after a levodopa challenge in the preoperative assessment ( mean improvement in Unified Parkinson’s Disease Rating Scale-III 61 . 9% ± 12 . 6 , see supplementary file 1 ) . Of note , decision thresholds - the main interest of the current study – are not modulated by dopamine in healthy people ( Winkel et al . , 2012 ) or PD patients ( Huang et al . , 2015 ) . To assess whether patients’ behavior was comparable to that of healthy controls , we also recruited 18 healthy age-matched control participants ( eight males , mean age 60 . 5 ± 16 . 1 years ) . In accordance with the declaration of Helsinki , participants gave written informed consent to participate in the study , which was approved by the local ethics committee ( Oxfordshire REC A ) . We used a modified version of the moving dots task; see Figure 1A . The task was presented on a MacBook Pro ( OS X Yosemite , version 10 . 10 . 3 , 13 . 3 inch Retina display , 60 Hz refresh rate ) using PsychoPy v1 . 8 ( Peirce , 2007; RRID:SCR_006571 ) . The display was viewed from a comfortable distance allowing the subjects to interact with the keyboard . At the beginning of each trial , a text cue indicated whether participants should respond as quickly ( ‘Fast ! ’ ) or as accurately as possible ( ‘Accurate ! ’ ) . The duration of this cue was randomly jittered between 0 . 75 and 1 . 25 s with an average duration of 1 s . Then , a cloud of 200 randomly moving white dots was presented on a black background . The diameter of the cloud was 14 cm and dot size was 10 pixels . Each dot moved in a straight line at a rate of 0 . 14 mm per frame for 20 frames before moving to another part of the cloud where it moved in a new direction chosen pseudorandomly between −180° and 180° . While some of the dots were moving randomly , the remaining dots moved coherently in one direction , which made the cloud of dots appear to move to the left or right . Participants were instructed to press a key with their right index finger ( ‘/’ on the keyboard of the laptop ) if they perceived that the cloud was moving to the right and to press a key with their left index finger ( ‘z’ ) when they perceived a leftwards movement . Between responses both index fingers rested on the respective keys . The percentage of dots moving coherently in one direction was either 50% ( high coherence ) or 8% ( low coherence ) . These two cues were pseudorandomly presented with equal probability so that participants could not predict whether the next trial would contain dots with high or low coherence . The trial was terminated by a response or after a 3 s deadline in case participants did not respond followed by immediate visual feedback , which was shown for 500 ms . During accuracy instruction ‘incorrect’ was shown as feedback both for errors of commission and errors of omission , while ‘correct’ was shown for all correct trials . During speed instructions ‘in time’ was shown for all responses within the 3 s window , while ‘too slow’ was shown if patients did not respond within the 3 s deadline . Similar to previous studies of speed-accuracy adjustments in PD , we did not impose a more restricted time window for responding during speed instruction ( Green et al . , 2013; Huang et al . , 2015 ) , since motor function varies considerably between PD patients . While trials with different coherence levels were randomly interspersed , accuracy and speed trials alternated in blocks of 20 trials ( Green et al . , 2013 ) . These blocks were repeated six times each resulting in 240 trials for the whole experiment ( Figure 1B ) , which lasted approximately 10 min . Before commencement of the experimental recordings , patients could practice the task for as long as they wished ( usually approximately 40 trials ) . Prior to statistical analyses , trials without responses ( errors of omission ) or reaction times ( RT ) <0 . 25 s were excluded ( combined 3 . 1% of all trials ) . The trials differed regarding the type of instruction ( accuracy vs . speed ) and coherence ( high vs . low ) ; see Figure 1C . Accordingly , RT were analyzed using a 2*2 ANOVA using SPSS statistics v22 ( IBM , New York , USA; RRID:SCR_002865 ) . In case of significant interactions , post-hoc tests were conducted using paired samples t-tests . Since the RT distribution of error trials is important for inferences on latent decision-making parameters in the drift diffusion model ( DDM ) framework ( see below ) , we additionally analyzed whether errors were primarily observed during fast or slow responses . To this end , we divided erroneous responses during low coherence trials ( there were only ~5% errors in high coherence trials , see Results ) into fast and slow trials after a median split . We then calculated the % change in accuracy by computing ( Accuracyslow – Accuracyfast ) / Accuracyfast separately for Accuracy and Speed instructions and tested whether the resulting values were different from 0 using one-sample t-tests . To directly compare patients and healthy participants , we computed the effect of instruction ( accuracy vs . speed ) and coherence ( low vs . high ) on RT and accuracy and calculated independent samples t-tests . For all tests , the statistical threshold was set to p=0 . 05 . Bonferroni correction was applied when appropriate and corresponding p-values marked as Pcorrected for clarity . Normality assumptions were tested using Shapiro-Wilk tests . Since accuracy rates violated the normality assumption , we applied an arcsine transform ( inverse of the sine function ) before statistical testing , but report non-transformed % values in figures and main text for readability . Values for statistical analyses are given as mean ± standard deviation throughout the article unless stated otherwise . We recorded bilateral STN LFPs from the implanted DBS electrodes . Additionally , in 8 of the 11 patients , electroencephalography ( EEG ) was recorded over two regions of interest: dorsomedial prefrontal cortex ( PFC ) and motor cortex areas . These regions were recorded from Fz as well as C3 ( left hemisphere ) and C4 ( right hemisphere ) according to the international 10–20 system . We also recorded from midline electrodes Cz , Pz and Oz whenever possible , but due to surgical wounds and dressings we were not able to cover a broader area . Accordingly , although we maintain that Fz , C3 and C4 recorded activity from our regions of interest , we do not assume that they do so exclusively . Electrooculography ( EOG ) was recorded for eye movement artifact rejection . Data were sampled at 2048 Hz , band-pass filtered between 0 . 5 and 500 Hz and amplified ( TMSi porti , TMS International , Enschede , The Netherlands ) . Further analyses were conducted offline using custom-written Matlab scripts ( R2015a , The MathWorks , Natick , MA , USA; RRID:SCR_001622 ) . Artifacts related to eye movements were removed by subtracting the filtered and adaptively scaled EOG data ( 40 Hz low-pass filter ) . Scaling was performed with an optimization algorithm ( Matlab function fminocn , initial value = 1 ) minimizing the sum of squared errors ( Fischer et al . , 2016 ) . Trials with residual artifacts were discarded after visual inspection . After removal of trials based on behavioral data ( see above ) and artifacts in the electrophysiological data on average 201 trials ( 83 . 4% ) remained per patient resulting in 2209 trials combined . Data were downsampled to 200 Hz , high-pass filtered at 1 Hz and low-pass filtered at 100 Hz . For quadripolar contacts ( five patients , see supplementary file 1 ) , LFPs were converted to a bipolar montage between adjacent contacts ( three channels per hemisphere ) to limit effects of volume conduction ( Herz et al . , 2016 ) . Similarly , for octopolar non-directional contacts ( three patients ) , bipolar channels were computed between adjacent contacts leading to seven channels per hemisphere . For octopolar directional contacts ( three patients ) , bipolar montages were created between the dorsal omnidirectional contact and its three adjacent directional contacts as well as between the ventral omnidirectional contact and its three adjacent directional contacts resulting in six bipolar channels per hemisphere . Power and phase of LFPs were computed using the continuous wavelet transform with two cycles per frequency for the lowest considered frequency ( 2 Hz ) which linearly increased to five cycles per frequency for the highest considered frequency ( 30 Hz ) in 1 Hz steps . We chose this frequency range , because we a-priori expected low-frequency oscillations ( LFO ) between 2 and 8 Hz and beta oscillations between 13 and 30 Hz to be modulated during the task ( Herz et al . , 2016; Zavala et al . , 2014 ) . Power of each frequency was normalized to the mean signal of that frequency across the whole experiment . Of note , we applied normalization to account for between-subject differences unrelated to the task ( e . g . signal-to-noise ratio ) , however , observed qualitatively highly similar spectra when omitting this step . The resulting time frequency spectra were aligned to the onset of the moving dots and time of the response , respectively . In order to detect the bipolar STN channel , which was most strongly modulated by the task , we employed the following procedure . First , for each patient and hemisphere , response-aligned time-frequency spectra ( averaged across all conditions to avoid circularity ) were visualized for all channels and the channel with the most pronounced beta modulation ( reduction of beta power in the peri-response time window ) was noted . For further analyses , we then selected the channel showing the strongest peri-response LFO increase if this channel ( i ) also showed a strong beta-modulation ( >15% change from baseline ) and ( ii ) was not further than two channels remote from the best beta channel to avoid including activity likely recorded outside of the STN ( contact length is 1 . 5 mm with 0 . 5 mm spacing between contacts for all implanted electrodes ) . In 53% of hemispheres , the bipolar channel showing the strongest beta modulation was identical to the channel showing the strongest LFO modulation and in 47% the best LFO channel was localized ventrally to the best beta channel . Figure 3—figure supplement 1 shows an example of STN channel selection for one representative patient . When using the best beta channel instead of the best LFO channel for analyses of task-related modulations of STN activity , we obtained highly similar results regarding modulations in the beta band to the ones presented in the paper ( data not shown ) . After channel selection , normalized spectra of the selected left and right STN were averaged resulting in one STN channel per patient . Preprocessing and time-frequency transformation of EEG channels were identical to the procedure for STN LFPs described above except that EEG channels were not converted to a bipolar montage , but referenced to the average of all EEG channels connected to the TMSi porti during recordings ( www . tmsi . com ) . Analysis of C3 activity was performed for right hand movements and of that from C4 for left hand movements , while for analysis of PFC activity , the midline electrode Fz was used . Of note , since EEG activity in prefrontal areas often has a clearer lower boundary at ~4 Hz - in contrast to STN activity , which typically extends into lower frequencies ( Cavanagh et al . , 2011; Herz et al . , 2016; Zavala et al . , 2014 ) - it is often referred to as theta power ( Cavanagh et al . , 2011 , 2012; Cohen , 2014 ) . Here , however , we use the same 2–8 Hz band for both STN and EEG power for consistency , and term this LFO activity . For the statistical analysis , we compared time-frequency spectra during speed vs . accuracy instructions ( effect of instruction ) and low vs . high coherence trials ( effect of coherence ) using cluster-based permutation testing ( Maris and Oostenveld , 2007 ) by shuffling between condition labels for each subject . We applied 1000 permutations and used p=0 . 05 as cluster-building threshold and as statistical threshold for cluster-based comparisons . The time window of interest was 1 . 5 s prior to until 1 . 5 s after the movement for the response aligned spectra and 0 . 5 s prior to until 1 . 5 s after the onset of moving dots for the cue-aligned spectra . We also compared differences between error and correct trials for the low coherence condition . This control analysis showed no significant differences and is shown in Figure 3—figure supplement 2 . In order to take into account RT differences between conditions in the cue-aligned spectra we applied an additional analysis based on ( Hanks et al . , 2014 ) for changes in the beta band . This was necessary , because the well-known strong beta-decrease during movements will necessarily induce strong differences in cue-aligned beta power changes between conditions with short and long RT , simply because the response will take place earlier in the condition with short RT . Thus , for each trial , we computed the change in beta power over time until the response was executed , that is , the time series of each trial was capped at the time of the response . We then averaged these time series across trials from 500 ms before dots onset until the point in time when 50% of trials in the condition with lowest RT ( high coherence during speed instructions ) contributed to the average . Thereby we ensured the inclusion of at least 50% of trials at all considered time points for all conditions . For statistical analysis , we identified the time window in which beta showed a cue-induced decrease ( Oswal et al . , 2012 ) based on the average across all conditions ( see Figure 4D ) and computed the change in beta between the start of the beta decrease ( 150 ± 25 ms post-cue ) and the end of the beta decrease ( 400 ± 25 ms post-cue ) for each patient and condition . Importantly , this time period was well before the average RT ( ~1250 ms post-cue ) . We then conducted a 2*2 ANOVA with instruction and coherence as factors after ensuring that assumptions of normality were met using Shapiro-Wilk test . Since we found an effect of instruction on LFO power both for response- and cue-aligned spectra ( see Results ) we additionally analyzed whether changes in LFO power differed between alignments by extracting the mean LFO power from the relevant windows ( −750 ms before the response until the response vs . 500 ms to 1250 ms postcue ) for speed and accuracy instructions separately and conducted a 2*2 ANOVA with factors alignment ( response vs . cue ) and instruction ( speed vs . accuracy ) . Finally , we also computed the intertrial phase clustering ( ITPC ) for STN LFO in order to test whether the LFO phase during the above mentioned windows of interest was time-locked to the cue or response . To this end , we projected the phase at each time point of each trial onto the complex plane , averaged across trials and computed its absolute value ( Cohen , 2014a; Zavala et al . , 2013 , 2016 ) , as follows:ITPC ( t ) =|1N∑n=1Neiϕn , t| where ϕn , t is the phase angle at trial n and time t . We computed the mean LFO ITPC for the above mentioned 750 ms time windows for the cue- and response-aligned data after speed and accuracy instructions separately . The resulting ITPC is bound between 0 and 1 , where 0 indicates phase inconsistency across trials and one means that the phase at a given time point is identical for each trial . These values were then averaged across participants and compared against a critical ITPC value given byITPCcritical= −ln ( p ) n where n is the average number of trials ( 201 ) and p the critical p-value of 0 . 05 ( Cohen , 2014a ) . In the DDM framework , perceptual decision-making between two alternatives is reflected by a continuous integration of sensory evidence over time until sufficient evidence has been accumulated and the choice is executed . DDM has been widely applied over the last decades and has been shown to accurately predict behavior over a range of different tasks ( Ratcliff and McKoon , 2008 ) . There are three main parameters in DDM . First , the drift rate v reflects the rate of evidence accumulation . If a cue clearly favors one over the other choice the drift rate is high resulting in fast and accurate decisions , while ambiguous cues will lead to low drift rates and thus slow and error-prone choices . In our experiment , we expected the level of dots coherence to modulate the drift rate ( Ratcliff and McKoon , 2008 ) . Second , the decision threshold a defines how much evidence is accumulated before committing to a choice . Thus , the decision threshold constitutes a decision criterion , which transforms a continuous variable ( sensory evidence ) into a categorical choice ( option A or B ) . There is ample evidence that speed-accuracy adjustments are mediated through modulations of the decision threshold so that decreased thresholds ( or mathematically equivalent an increased baseline , see Figure 2A ) during speed emphasis lead to faster responses at the expense of accuracy . This leads to increased fast errors during speed emphasis , that is , errors are typically faster than correct trials ( Ratcliff and McKoon , 2008 ) . Conversely , during conditions with high thresholds errors are typically slower than correct trials ( for a more detailed explanation of this finding see [Ratcliff and Rouder , 1998] ) . In the current study , we hypothesized that the decision threshold would be modulated by speed vs . accuracy instructions . The third parameter in DDM is the non-decision time t , which is thought to be related to afferent delay , sensory processing and motor execution . We applied a Bayesian hierarchical estimation of DDM ( HDDM ) , which is particularly suited for experiments with low trial counts ( Wiecki et al . , 2013 ) , implemented in Python 2 . 7 . 10 ( RRID:SCR_008394 ) . Another advantage of HDDM is that it allows regression analyses between trial-by-trial variations of model parameters and fluctuations of neural activity , such as STN power modulations . In other words , model parameters and neural activity are not analyzed separately and then correlated afterwards ( which would result in n = 11 observations ( patients ) in the current study ) , but the neural variables are directly entered into the model and allowed to modulate the latent decision-making parameters at each trial ( resulting in n = 2209 observations ( trials ) in our study ) . The hierarchical design assumes that parameters from individual participants are not completely independent , but drawn from the group distribution while allowing variations from this distribution given sufficient evidence to overwhelm the group prior . Prior distributions were informed by 23 previous studies ( Wiecki et al . , 2013 ) . The starting parameter ( bias parameter ) z was fixed to 0 . 5 , because leftwards and rightwards movements were equally likely . We a-priori assumed that the decision threshold a was modulated by instruction ( speed vs . accuracy ) and the drift rate v by level of coherence ( low vs . high ) allowing for overall trial-by-trial variability in drift . In addition , we also assessed whether the non-decision time t was affected by changes in instructions or dots coherence . Since this was not the case ( see Results ) , we used a simple model with drift rates affected by coherence and thresholds altered by instructions for model checks and regression analyses ( see below ) . Markov chain Monte Carlo sampling was used for Bayesian approximation of the posterior distribution of model parameters generating 20 , 000 samples and discarding 10 , 000 samples as burn-in ( Herz et al . , 2016 ) . To ensure model convergence we inspected traces of model parameters , their autocorrelation and computed the R-hat ( Gelman-Rubin ) statistics ( Wiecki et al . , 2013 ) . To assess model performance , we computed quantile probability plots ( Ratcliff and McKoon , 2008 ) , in which predicted and observed RT for the 10 , 30 , 50 , 70 and 90 percentile were plotted against their predicted and observed cumulative probability for each condition . Error trials were only plotted for the low coherence condition due to the paucity of errors in the high coherence condition ( ~5% ) . Parameters of the model were analyzed by Bayesian hypothesis testing . We considered posterior probabilities ≥95% of the respective parameters being different than zero significant ( Frank et al . , 2015; Herz et al . , 2016 ) . In other words , model parameters were significant if ≥95% of samples drawn from the posterior were different from zero ( or different from the distribution they were compared to , for example , of healthy participants ) . Even though such posterior probabilities are distinct from classical p values ( e . g . in a t-test ) they can be interpreted in a similar manner ( Wiecki et al . , 2013 ) . After having conducted the HDDM analysis for patients , we conducted the identical analysis for healthy participants to assess whether task effects on latent decision-making parameters were different between both groups . After estimating the HDDM not containing any neural data , we entered neural variables , which were related to speed-accuracy adjustments as observed in the analysis of electrophysiological data , into the model . In particular , we fitted a model assuming that a threshold on a given trial depended on measured neurophysiological data . These data comprised LFO power in the pre-response period and the cue-induced decrease in beta power ( see Results ) . Based on the trial-averaged group comparison , single trial LFO power was extracted from −750 ms until the response or from –RT until the response in case RT were <750 ms ( Herz et al . , 2016 ) , that is , locked to the response . The beta-power decrease was extracted using the change from 150 ms to 400 ms ( ±25 ms ) after cue-onset ( see above ) . Then single trial values were averaged across the respective time window and frequencies . Trials with RT <425 ms ( 1 . 7% ) were excluded from the HDDM analysis to ensure that the dots-induced beta decrease did not fall into the time of the motor response and the remaining single trial values were z-scored by subtracting the mean and dividing by the standard deviation for each subject . We assessed a putative statistical relationship between the cue-induced beta decrease and pre-response LFO increase , by conducting Spearman correlations for each subject , Fisher r-to-z-transforming the correlation coefficients and then testing the resulting values against 0 using a one-sample t-test on the second level . Since we hypothesized that the changes in STN activity observed during speed-accuracy adjustments were related to modulations of decision thresholds , we regressed these neural variables against estimates of thresholds at each trial during model estimation . In other words , the regression coefficients between STN activity and decision thresholds were estimated within the same model , which was used to estimate the decision-making parameters themselves . For example , on a given trial the threshold a was defined by: a = b0 + b1Instr + b2LFO + b3LFO*Instr + b4Betadecrease + b5Betadecrease*Instr , where Instr refers to the type of Instruction ( 0 for Accuracy , one for Speed ) , LFO indicates the pre-response increase in LFO , Betadecrease the cue-induced decrease in beta power , and b1-5 are the estimated regression coefficients . Of note , since we included single-trial power in LFO and beta in the same model , we accounted for effects of beta on thresholds for the LFO regression estimates and vice versa . Even though we hypothesized that changes in STN power were related to modulations of decision thresholds , we additionally computed an identical regression analysis with drift rate estimates as dependent variable in the same model as a control analysis . Posteriors of regression coefficients for trial-wise regressors were estimated only at the group level to address potential collinearity among model parameters , for regularizing parameter estimates and to prevent parameter explosion ( Frank et al . , 2015; Wiecki et al . , 2013 ) . Statistical inference on regression coefficients was based on the distribution of the posterior probability densities ( see above ) . Finally , we computed the deviance information criterion ( DIC ) for the HDDM not containing any neural data and the HDDM containing neural activity in order to assess whether including STN activity improved model performance . DIC is mainly used for comparisons of hierarchical models where other measures , such as the Bayesian information criterion ( BIC ) , are not appropriate ( Frank et al . , 2015; Wiecki et al . , 2013 ) . Lower DIC values indicate improved model fits taking into account model complexity . Traditionally , DIC differences >10 are considered significant ( Herz et al . , 2016 ) . In order to assess whether the observed relationships between STN activity and HDDM parameters ( see Results ) were also evident in simple behavioral measures , we applied additional regression analyses using STN power ( LFO and beta ) as predictors and RT as dependent variable . We used a linear mixed-effects regression model , in which the intercept of the regression was allowed to vary between participants ( random effect ) , while the slope of the regressions was treated as fixed effect ( consistent across subjects ) , similar to the HDDM regression analysis described above . Since we tested specific directional effects based on the HDDM regression analyses , for example , stronger cue-induced beta-decreases predicted lower thresholds ( see Results ) , we used Pone-tailed = 0 . 05 as statistical threshold . EEG was recorded in addition to STN LFPs for computation of cortico-STN connectivity . More specifically , we wanted to assess whether modulations of STN activity in the LFO and beta band were related to changes in Fz-STN connectivity and C3/C4-STN connectivity , respectively , and whether this relationship was modulated by speed vs . accuracy instructions . In a first step , we plotted response- and cue-aligned spectra for Fz and C3/C4 separately . To analyze whether task-related changes in LFO and beta were predominantly expressed over Fz or C3/C4 , we compared the pre-response increase in LFO and cue-induced decrease in beta ( see Statistical analysis of STN LFPs ) between these two sites using permutation testing ( 1000 permutations ) by shuffling between electrode labels for each subject . After having established the spatial selectivity of cortical LFO and beta activity , we then computed the inter-site phase clustering ( ISPC ) between Fz and STN in the LFO band as well as between C3/C4 and STN in the beta band . ISPC is a phase-based measure of connectivity reflecting to what extent oscillations in two different areas are phase-locked at specific time points ( Cohen , 2014a; Herz et al . , 2016; Zavala et al . , 2013 , 2014 , 2016 ) . This measure is independent of power changes in that it only considers the phase information of the signal and will be only affected by power in cases where power is very low ( because phase estimates deteriorate ) or zero ( because phase cannot be estimated with zero power ) ( Cohen , 2014a ) . Note that we did not compute single trial estimates of connectivity , because connectivity estimates are far more robust when averaging across trials . After extracting the phase from the wavelet transform , the difference between the phase of the STN LFP and respective EEG channel was calculated at each time point of each trial , and averaged across trials . Note that this computation is identical to ITPC described above , except that the phase difference between the STN and EEG signal were used rather than phase angles from STN alone . The resulting values were then integrated over time using a sliding window decreasing linearly from ±200 ms ( corresponding to ±1 cycle for 5 Hz , the median frequency of LFO ) to ±100 ms ( roughly corresponding to ±2 cycles for 22 Hz , the median of beta ) . For specifically assessing task-related changes in ISPC , we calculated the % change from baseline . The baseline was defined as the time from 500 ms until 250 ms before onset of the moving dots cue . The time window of interest was defined based on the modulations of STN power during speed-accuracy adjustments: a 750 ms time window prior to the response for LFO and a 500 ms time window starting 150 ms after dots onset for beta . For LFO , this was computed for Fz-left STN and Fz-right STN and then averaged across hemispheres . For beta , C3-left STN was used for right-handed responses and C4-right STN for left-handed responses before averaging across hemispheres . This resulted in one ISPC value per patient ( n = 8 ) for each cortico-STN connection ( Fz-STN and C3/C4-STN ) . To assess putative effects of speed-accuracy adjustments , we then compared ISPC for Fz-STN and C3/C4-STN between speed and accuracy instructions . We applied permutation testing by shuffling between condition labels using 1000 permutations . Finally , we assessed whether differences in STN power , which were shown to predict trial-by-trial adjustments of decision threshold ( see Results ) , were related to differences in cortico-STN connectivity across subjects and whether this relationship depended on the type of instruction . To this end , we entered the ISPC values into a Pearson correlation analysis with the corresponding STN power ( i . e . pre-response STN LFO power for Fz-STN ISPC and cue-induced STN beta power decrease for C3/C4-STN ISPC ) separately for speed and accuracy instructions . The resulting rho values for speed and accuracy were Fisher r-to-z transformed and compared between the speed and accuracy condition . In case of a significant difference the correlations within a condition , that is , during accuracy and speed conditions separately , were tested for significance . Prior to conducting Pearson correlations , we ensured that assumptions of parametric correlations were met including absence of outliers using Grubbs test and normality using Shaprio-Wilk test . Please note that single subject estimates of decision thresholds were not used for any correlation analyses , since the hierarchical design of HDDM violates the assumption of independence of observations ( Wiecki et al . , 2013 ) . For all statistical tests of EEG data and cortico-STN connectivity , a statistical threshold of p=0 . 05 was applied correcting for multiple comparisons using the Bonferroni method whenever applicable . For permutation tests , p-values are given as p<0 . 05 , p<0 . 01 or p<0 . 001 instead of precise p-values , since these can change slightly when repeating permutation tests . | In everyday decisions , we have to balance how quickly we need to make a decision with how accurate we want our decision to be . For example , if you plan your next holiday you might want to make sure that you pick the best destination without caring too much about the time it takes to arrive at that decision . On the other hand , in your lunch break you might want to quickly choose between the different meals on the menu to make sure you are back at work on time , even though you might overlook a dish that you would have preferred . This effect – that decisions we make in haste are more likely to be suboptimal than slower , more deliberate decisions – is known as the speed-accuracy trade-off . One theory suggests that the activity of a brain area termed the subthalamic nucleus reflects whether people will prioritize speed or accuracy during decision-making . This area is seated deep inside the brain , meaning that it is normally difficult to record its activity . Herz et al . have now recorded the activity of the subthalamic nucleus in individuals with Parkinson’s disease who underwent brain surgery as part of their treatment . When these individuals switched between fast and cautious decision-making , the activity in the subthalamic nucleus changed , as did its relationship with the activity seen in other brain areas . Furthermore , these activity changes predicted how much information participants acquired before committing to a choice . Deep brain stimulation of the subthalamic nucleus is now a standard treatment for Parkinson’s disease . It will be important to assess whether this treatment affects the changes in subthalamic activity that are related to decision-making , and whether this affects whether an individual is more likely to make fast or accurate decisions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | Distinct mechanisms mediate speed-accuracy adjustments in cortico-subthalamic networks |
Many aspects of social behavior are controlled by sex-specific pheromones . Gender-appropriate production of the sexually dimorphic transcription factors doublesex and fruitless controls sexual differentiation and sexual behavior . miR-124 mutant males exhibited increased male–male courtship and reduced reproductive success with females . Females showed a strong preference for wild-type males over miR-124 mutant males when given a choice of mates . These effects were traced to aberrant pheromone production . We identified the sex-specific splicing factor transformer as a functionally significant target of miR-124 in this context , suggesting a role for miR-124 in the control of male sexual differentiation and behavior , by limiting inappropriate expression of the female form of transformer . miR-124 is required to ensure fidelity of gender-appropriate pheromone production in males . Use of a microRNA provides a secondary means of controlling the cascade of sex-specific splicing events that controls sexual differentiation in Drosophila .
In animals , the performance of the individual in social behaviors such as mate recognition , courtship and aggression are important determinants of reproductive fitness . These behaviors are modulated in part by chemical cues , pheromones , used for intraspecific communication . In Drosophila melanogaster , the courtship and aggressive behaviors exhibited by male flies are influenced by a cocktail of pheromones produced by males and females ( Jallon , 1984; Fernandez et al . , 2010; Wang and Anderson , 2010 ) . Detection of pheromones is mediated by specific receptors that detect compounds spread by volatile diffusion and transferred during physical contact ( Kurtovic et al . , 2007; Vosshall , 2008; Stowers and Logan , 2010; Wang et al . , 2011; Thistle et al . , 2012; Toda et al . , 2012 ) . Pheromones in Drosophila melanogaster are strikingly sexually dimorphic in expression and in their effects on male and female behavior ( Jallon , 1984; Ferveur and Sureau , 1996 ) . Long-chained hydrocarbons present on the cuticular surface of the abdomen constitute a major class of Drosophila sex pheromones . The hydrocarbons are synthesized by specialized cells called oenocytes ( Billeter et al . , 2009 ) . Female pheromones are largely comprised of cis , cis-7 , 11-heptacosadiene and cis , cis-7 , 11-nonacosadiene , both of which are known to serve as aphrodisiacs for males ( Antony et al . , 1985 ) . Males primarily produce hydrocarbons bearing a single double bond ( e . g . , cis-7-tricosene , cis-7-pentacosene and cis-9-pentacosene ) , although these compounds are also produced by females ( Jallon and David , 1987 ) . The male-predominant cis-7-tricosene acts as an aphrodisiac for females but an anti-aphrodisiac for males . Members of the oenocyte-produced pentacosene family can also act as male aphrodisiacs , when present at high levels ( Scott and Richmond , 1988; Siwicki et al . , 2005 ) . Drosophila males also produce a different class of pheromones in the ejaculatory bulb , which are transferred during mating and mediate chemical communication ( Guiraudie-Capraz et al . , 2007; Yew et al . , 2009 ) . 11-cis-Vaccenyl-Acetate ( cVA ) , an oxygenated lipid , is thought to have an aphrodisiac effect on females , stimulating receptivity towards copulation , and acting as an anti-aphrodisiac for males ( Jallon , 1984; Cobb , 1996; Kurtovic et al . , 2007 ) . CH503 ( 3-acetoxy-11 , 19-octacosadien-1-ol ) , a second lipid made in the male ejaculatory bulb , also acts as an anti-aphrodisiac for males after being transferred to the female during mating ( Yew et al . , 2009 ) . Sexually dimorphic behavior and chemical communication are under the control of the sex determination pathway ( Burtis and Baker , 1989; Ryner et al . , 1996; Kimura et al . , 2005; Villella et al . , 2005; Vrontou et al . , 2006; Kimura et al . , 2008; Siwicki and Kravitz , 2009 ) . Expression of the splicing factor Sex-lethal ( Sxl ) in genetically female animals promotes sex specific splicing of the sexually dimorphic transformer transcript to produce the female splice form ( TraF ) . TraF in turn promotes splicing to produce the female form of Doublesex ( DsxF ) . In the absence of TraF , the default male form of Dsx ( DsxM ) is produced , along with the male form of Fruitless ( FruM ) . Dsx proteins direct male vs female sexual differentiation , including pheromone production , as well as sexual behavior ( Waterbury et al . , 1999; Rideout et al . , 2010 ) , whereas FruM controls male sexual behavior but not pheromone production ( Demir and Dickson , 2005; Manoli et al . , 2005 ) . MicroRNAs have previously been implicated in the control of gene expression noise , acting as a backup mechanism to minimize the consequences of leaky expression of transcripts whose primary regulation is under transcriptional control ( Stark et al . , 2005; Karres et al . , 2007; Bushati et al . , 2008; Shkumatava et al . , 2009; Weng and Cohen , 2012 ) , reviewed in ( Herranz and Cohen , 2010; Ebert and Sharp , 2012 ) . miRNAs are also well suited to buffer the effects of inappropriate splicing . For example , miR-1 can limit expression of the cytoplasmic splice form of tropomyosin , while sparing muscle specific splice forms ( Stark et al . , 2005 ) . miR-124 is abundantly expressed in the Drosophila brain , where it has been shown to limit leaky expression of an inhibitor of neuronal stem cell proliferation during larval development ( Weng and Cohen , 2012 ) . Here we present evidence that miR-124 acts to limit the impact of leaky regulation of splicing in the sexual differentiation pathway . miR-124 mutant males showed reduced mating success when paired with female flies , and elicited courtship by normal males . These effects were traced to aberrant pheromone production . We identified the sex-specific splicing factor transformer as the functionally significant target of miR-124 in this process , suggesting a role for miR-124 in the control of male sexual differentiation , by limiting inappropriate expression of the female form of transformer .
Drosophila males engage in a complex set of courtship behaviors to induce receptiveness of females to mating . miR-124 mutant males exhibited a normal repertoire of behaviors when paired with sexually mature Canton S ( CS ) female virgins in a standard courtship assay ( including orientation toward the female , courtship song , tapping , licking , abdomen curling , and attempted copulation ) . However , miR-124 mutant males achieved copulation significantly less often than CS controls during the 30-min observation period ( Figure 1A , **p<0 . 01 ) . miR-124 mutant females and CS females did not show a significant difference in receptiveness to courtship by CS males ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 00640 . 003Figure 1 . Male courtship behavior . ( A ) Percentage of males achieving copulation in a 30-min observation period . Genotypes as indicated . Control males were CS . Rescue indicates the miR-124 RMCE allele with miR-124 reintegrated at the endogenous locus ( 34 ) . Data represent the average of five independent experiments ± SEM . ( B ) Courtship initiation latency measures time ( in minutes ) to initiate courtship for CS control and miR-124 flies . Data represent the average of four independent experiments ± SEM . ns: no significant difference . ( C ) Percentage of males achieving copulation in 30 min , comparing CS control and miR-124 mutant flies before and after removal of the wings . Data represent the mean of more than 20 movies per genotype ± SD . ( D ) Courtship index compares the proportion of the measurement period males spent courting . CS control and miR-124 mutant males were tested using decapitated CS females as targets . Data represent 56 trials conducted in 4 batches of 14 pairs each . The horizontal line represents the median . Although the variance was high , the difference in the medians was borderline significant ( p=0 . 042 comparing for the 56 pairs ) . However , when the data were analyzed as the average of four independent experiments ( n = 14 in each experiment ) the difference in the means was not significant . In all figures: *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ns: not statistically significant . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 00310 . 7554/eLife . 00640 . 004Figure 1—figure supplement 1 . Receptivity of miR-124 mutant females . 5-day-old socially naive CS males were paired individually with 5-day-old CS or miR-124 virgin females and the number of females that accepted copulation over an observation period of 20 min was scored . No significant difference was observed in receptivity of 5-day-old control ( CS ) or miR-124 virgin females towards 5-day-old socially naïve CS males . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 004 To determine the basis for the reduced mating efficiency we examined a number of courtship behavioral parameters . Initiation latency , the time taken by the male to recognize the female and begin courtship , was unaffected ( Figure 1B ) . Males use a courtship song produced by wing vibration to elicit receptivity in female flies . If a defect in courtship song is responsible for the poor mating success of mir-124 mutant males , removal of the wings should eliminate the observed difference in receptivity of females to courtship . Under these conditions , miR-124 mutants were also less successful in mating than control males ( Figure 1C ) . Thus , courtship song does not appear to be an important contributor to the difference between control and mutant males . Progression from courtship to copulation involves behavioral input from female flies ( Rezaval et al . , 2012 ) . To remove female behavioral response from the assay , we tested decapitated target flies . We did not observe a reduction in the level of courtship activity by mutant males compared to that of control males under these conditions ( Figure 1D ) . Thus the failure to achieve copulation is unlikely due to reduced activity of the mutant male . Reduced copulation therefore likely reflects rejection of the miR-124 mutant male’s advances by the female . This defect was rescued when miR-124 expression was restored in the miRNA expressing cells of the mutant ( Figure 1A ) . Drosophila males normally pay little sexual attention to other sexually mature males . Males with altered sexual orientation elicit a behavior called chaining , in which groups of males follow each other while attempting courtship ( Hall , 1978 ) . We observed chaining among groups of miR-124 mutant males . Male–male courtship could result from altered sexual orientation or from a change in the expression of inhibitory or stimulatory cues , or from an inability to recognize inhibitory courtship cues . To distinguish among these possibilities , we quantified the courtship behavior of mutant and control males when placed with mutant or control male targets . There was no difference in the amount of time that miR-124 mutant or CS control males devoted to courtship of CS target males ( Figure 2A ) . However , miR-124 mutant targets elicited more courtship activity from both CS control and miR-124 males ( Figure 2A , **p<0 . 01 ) . This effect was suppressed when miR-124 expression was restored in its endogenous domain ( Figure 2B , **p<0 . 01 ) . Next , a courtship choice assay was performed in which test males were presented with a choice of decapitated control or miR-124 target males . Wild-type CS males devoted more than twice as much time to courting the miR-124 target as they did to the control target ( Figure 2C , **p<0 . 01 ) . Thus , CS males appeared to be more attracted by miR-124 males than by other CS males . 10 . 7554/eLife . 00640 . 005Figure 2 . Male–male courtship . ( A ) Courtship index comparing CS control and miR-124 mutant flies using decapitated CS or miR-124 mutant males as targets . The number of animals used for each sample is indicated ( n: ) . Scores for many control flies were very close to zero , overlapping the X axis , and so are not visible as individual data points in the scatter plot . Data represent one of four independent trials performed with comparable results . ( B ) Courtship index for CS control males toward decapitated targets . The target genotypes used are CS control , miR-124 mutant and rescued mutant . Data represent one of four independent trials performed with comparable results . ( C ) Courtship choice assay comparing the time CS control males courted decapitated CS control and miR-124 mutant targets , when presented together . Data represent the mean of more than 20 movies per genotype ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 005 The behavior of the control and mutant males in each of these assays depended on the genotype of the targets , not on the genotypes of the test males themselves . This suggests that the male–male courtship phenotype is unlikely to reflect a change in neuronal circuitry of the mutant males that could affect their sexual orientation or their ability to recognize normal inhibitory cues . Rather , the observation that behaviorally inert mutant males elicited courtship behavior from control males suggested a change in chemical cues provided by miR-124 mutant males . Cuticular hydrocarbon profiles were generated for sexually mature miR-124 mutant and control male flies using gas chromatography/mass spectrometry ( GC-MS ) . GC-MS analysis showed that the level of cVA was significantly reduced in miR-124 mutant males ( Table 1 and Figure 3A , ***p<0 . 001 ) , and was partially restored in rescued mutants ( Table 1 and Figure 3A ) . Conversely , pentacosenes were recovered at elevated levels on miR-124 mutant males by GC-MS ( Table 1 and Figure 3B , *p<0 . 05 ) and found near normal levels in the rescue mutants ( Table 1 and Figure 3B , **p<0 . 01 ) . These results suggest that miR-124 mutant males produce elevated levels of compounds that behave as male aphrodisiacs , and lower levels of compounds that have anti-aphrodisiac activity on males , leading to increased male–male courtship . 10 . 7554/eLife . 00640 . 006Table 1 . GC-MS analysis of cuticular hydrocarbon extracts from control , mir-124 mutant , and rescued mutant malesDOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 006Compound and elemental composition*Control† ( n = 6 ) mir-124 mutant† ( n = 6 ) Rescued mutant† ( n = 6 ) C21:0 ( nC21 ) 0 . 46 ± 0 . 080 . 32 ± 0 . 050 . 76 ± 0 . 11C22:10 . 24 ± 0 . 010 . 27 ± 0 . 020 . 35 ± 0 . 02cVA ( cis-vaccenyl acetate ) 9 . 36 ± 3 . 401 . 75 ± 0 . 57***6 . 60 ± 2 . 17***C22:00 . 74 ± 0 . 060 . 62 ± 0 . 020 . 95 ± 0 . 057 , 11-C23:20 . 13 ± 0 . 010 . 07 ± 0 . 0010 . 12 ± 0 . 029-C23:1 ( 9-tricosene ) 1 . 39 ± 0 . 131 . 76 ± 0 . 251 . 84 ± 0 . 147-C23:1 ( 7-tricosene ) 23 . 52 ± 1 . 1724 . 92 ± 1 . 7432 . 80 ± 2 . 03***5-C23:1 ( 5-tricosene ) 2 . 71 ± 0 . 113 . 06 ± 0 . 203 . 01 ± 0 . 18C23:0 ( nC23 ) 10 . 57 ± 0 . 4011 . 21 ± 0 . 2512 . 66 ± 0 . 63**C24:10 . 32 ± 0 . 110 . 37 ± 0 . 090 . 30 ± 0 . 07C24:00 . 36 ± 0 . 020 . 43 ± 0 . 040 . 35 ± 0 . 032-MeC241 . 44 ± 0 . 081 . 58 ± 0 . 152 . 03 ± 0 . 12C25:20 . 52 ± 0 . 060 . 71 ± 0 . 070 . 70 ± 0 . 049-C25:1 ( 9-pentacosene ) 4 . 80 ± 0 . 616 . 33 ± 0 . 65*4 . 11 ± 0 . 747-C25:1 ( 7-pentacosene ) 22 . 99 ± 1 . 5525 . 62 ± 0 . 63***11 . 61 ± 1 . 16***5-C25:1 ( 5-pentacosene ) 1 . 10 ± 0 . 330 . 79 ± 0 . 022 . 38 ± 0 . 01C25:0 ( nc25 ) 2 . 34 ± 0 . 153 . 13 ± 0 . 032 . 52 ± 0 . 152-MeC266 . 75 ± 0 . 495 . 37 ± 0 . 086 . 55 ± 0 . 139-C27:10 . 16 ± 0 . 020 . 19 ± 0 . 030 . 12 ± 0 . 047-C27:10 . 97 ± 0 . 100 . 77 ± 0 . 070 . 29 ± 0 . 06**C27:0 ( nC27 ) 1 . 66 ± 0 . 332 . 42 ± 0 . 601 . 86 ± 0 . 392-MeC285 . 90 ± 0 . 815 . 95 ± 0 . 716 . 18 ± 0 . 77C29:00 . 37 ± 0 . 110 . 78 ± 0 . 260 . 54 ± 0 . 172-MeC300 . 64 ± 0 . 160 . 99 ± 0 . 270 . 87 ± 0 . 25*The elemental composition is listed as the carbon chain length followed by the number of double bonds; 2-Me indicates the position of methyl branched compounds . †The normalized signal intensity for each compound and SEM is indicated; *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 when compared to control ( ANOVA followed by post-hoc Tukey HSD test ) . 10 . 7554/eLife . 00640 . 007Figure 3 . Aberrant pheromone production by miR-124 mutant males . ( A ) Normalized cVA level measured by GC-MS in extracts from control , miR-124 mutant , and rescued mutant males . Data represent the average of six independent preparations ± SEM . n = 15 in each preparation . ( B ) Normalized 9-pentacosene level measured by GC-MS from control , miR-124 mutant , and rescued mutant males . Data represent the average of six independent preparations ± SEM . n = 15 in each preparation . ( C ) Percentage of males achieving copulation in 30 min , comparing miR-124 mutant flies with or without cVA perfuming . Hexane perfuming was used as a control . Data represent the mean of >20 movies per genotype ± SD . ( D ) Courtship index ( CI ) using CS test males and miR-124 mutant target males perfumed with hexane solvent alone as a control , or with hexane containing cVA . No significant difference was observed between CI of CS males towards miR-124 target males perfumed with hexane ( average CI = 0 . 320 ) or with cVA ( average CI = 0 . 315 ) . n = 30 in each experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 00710 . 7554/eLife . 00640 . 008Figure 3—figure supplement 1 . Abundance of cVA on perfumed flies . DART mass spectrometry was used to assess the efficiency of the perfuming method . miR-124 mutant males perfumed with cVA exhibited more cVA than solvent-perfumed miR-124 mutant males and approximately 50% the amount of cVA found on control flies . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 008 To ask whether the changes in pheromone levels were sufficient to account for the increased male–male courtship elicited by miR-124 mutant males , we carried out perfuming experiments . Mutant males perfumed with cVA showed a significant improvement in their ability to achieve copulation with control females ( Figure 3C , *p<0 . 05 ) . We also examined the effects of perfuming on male courtship behavior . Decapitated miR-124 mutant males were perfumed with cVA and used as targets in the male–male courtship assay . There was no significant difference between courtship of targets perfumed with cVA or with the hexane solvent alone ( Figure 3D; the perfuming protocol restored cVA to <50% the level on control flies , Figure 3—figure supplement 1 ) . The cVA-perfumed miR-124 mutant target males also have elevated levels of the pentacosene pheromones . Thus , the perfumed mutant males are expected to give mixed excitatory and inhibitory courtship signals . In this context , the level of cVA reached by perfuming may be insufficient to fully rescue male–male courtship , while being sufficient to restore male–female courtship . However , we do not exclude the possibility that cVA might be more effective at inhibiting male courtship if presented at a higher local concentration . cVA is normally concentrated on the tip of the male ejaculatory apparatus . The perfuming experiment distributes cVA over the entire body . Although miR-124 mutant males showed less mating success in the courtship assay , they are fertile in laboratory conditions . The aberrant pheromone production might be expected to confer a disadvantage in a competitive situation , where the female has a choice of mates . To test this , single CS female virgins were placed in mating chambers with one CS control male and one miR-124 mutant or rescued mutant male . miR-124 mutant males were rarely selected in the presence of a wild-type male , but females did not distinguish between CS males and rescued mutant males ( Figure 4A ) . Mutant males would likely be at a disadvantage in a natural competitive setting . 10 . 7554/eLife . 00640 . 009Figure 4 . Comparison of other social behaviors . ( A ) Female mate choice was monitored by videotaping in chambers containing single females and two males of the indicated genotypes . The genotype of the male that succeeded in copulating was recorded . More than 95% of control male achieved copulation , in the presence of miR-124 mutant males ( left bar ) compared with ∼50% in the presence of rescued mutant males ( right bar ) . ( B ) Fighting latency was monitored by videotaping encounters between pairs of males in chambers containing a patch of food . Latency is the number of encounters that do not elicit aggressive behavior prior to the first fight . Data represent the mean of more than 16 movies per genotype ± SD . ( C ) Fighting frequency was monitored by videotaping encounters between pairs of males in chambers containing a patch of food . Frequency records the number of aggressive encounters in 30 min . Data represent the mean of more than 16 movies per genotype ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 00910 . 7554/eLife . 00640 . 010Figure 4—figure supplement 1 . Locomotion assay . The total distance travelled by single 5-day-old males of the indicated genotypes in a 10 mm courtship chamber was traced and measured for 10 min . The velocity of each genotype was calculated and normalized to control level . 14 flies were recorded per genotype . There was no significant difference between the control and mutant flies . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 010 Aggression is another social behavior commonly observed among Drosophila males , and is promoted by chemical cues such as cVA ( Wang and Anderson , 2010 ) . To ask if the loss of miR-124 influences male aggressiveness , the fighting behavior between pairs of mutant or wild-type males was analyzed . In this setting , wild-type males typically fight for sole occupancy of the food patch , resulting in the establishment of a hierarchy ( Chen et al . , 2002 ) . miR-124 mutant males exhibited overall lower levels of aggression based on several parameters . First , mutant males experienced more encounters before any fighting took place ( latency , Figure 4B , **p<0 . 01 ) . Mutant males exhibited lower frequency of fighting behaviors , including lunging and fencing ( Figure 4C , **p<0 . 01 ) and were often observed sharing the food patch after a few encounters . There was no obvious difference in overall activity levels , based on observation during the assay and results of a locomotion assay ( Figure 4—figure supplement 1 ) . Lower cVA production in the miR-124 mutant may contribute to the lowered intensity of aggressive behaviors observed in these flies . Sexually dimorphic behavior and chemical communication are under the control of the sex determination pathway ( Figure 5A ) . To ask whether miR-124 might act in the sex determination pathway , we used a microRNA sponge to deplete miR-124 in doublesex-expressing cells . Doublesex expression is sexually dimorphic in the brains of males and females ( Rideout et al . , 2010; Robinett et al . , 2010 ) . In the male , DsxM is required for differentiation of FruM-expressing neurons ( Rideout et al . , 2010 ) . To increase efficacy , the sponge was expressed in males lacking one copy of the endogenous miR-124 gene . Depletion of miR-124 in dsx-expressing cells elicited male–male courtship at a level comparable to that elicited by homozygous miR-124 null mutant target males ( Figure 5B ) . 10 . 7554/eLife . 00640 . 011Figure 5 . miR-124 acts in the sex differentiation pathway . ( A ) Key components of the sexual differentiation system . ( B ) Courtship index comparing miR-124 mutants and flies expressing a miR-124 sponge under dsx-Gal4 control in males lacking one copy of the endogenous miR-124 gene with CS controls . n: number of animals per sample . Data represent one of four independent trials performed , with comparable results . The horizontal lines represent the median for each data set . ( C ) Courtship index comparing proportion of time CS control males spent courting decapitated male flies that had expressed the UAS-miR-124 sponge under elav-Gal4 control vs flies that carried the UAS-miR-124 sponge transgene without Gal4 and vs miR-124 heterozygous control males . n: number of animals per sample . Data represent one of four independent trials performed , with comparable results . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 011 Doublesex is expressed in both neuronal and non-neuronal tissues , whereas miR-124 is highly enriched in the CNS . To ask whether the CNS is the site of miR-124 action , we used the pan-neuronal elav-Gal4 driver to direct expression of the miR-124 sponge in males lacking one copy of the endogenous miR-124 gene . This resulted in increased courtship of these flies by wild-type males ( Figure 5C ) , suggesting that miR-124 activity in the CNS contributes to the male courtship phenotype , presumably by modulation of pheromone production . Computational target prediction datasets do not list any of the known components of the sex determination pathway among predicted miR-124 targets . To allow for the possibility that the prediction algorithms might miss sites with specific features , we scanned sex determination pathway transcripts using the RNAhybrid prediction tool ( Rehmsmeier et al . , 2004 ) and found two potential sites for miR-124 in the 3′ UTR of transformer ( Figure 6A , B ) . The first site is present in the 3′ UTR region common to both the female-specific and non-sex-specific tra transcripts , while the second one is located in sequences unique to the non-sex-specific form . Pairing to residues 2–8 of the miRNA , called the seed region , is important in miRNA target identification ( Brennecke et al . , 2005 ) . Each of the sites in tra would require 3 G:U base pairs with the miR-124 seed . G:U base pairs in the seed region are compatible with miRNA function , but reduce the efficiency of target regulation ( Brennecke et al . , 2005 ) . A luciferase reporter assay showed that these sites can mediate regulation by miR-124 ( Figure 6C ) . 10 . 7554/eLife . 00640 . 012Figure 6 . miR-124 targets transformer . ( A ) Predicted pairing of miR-124 to two sites in the traF transcript . ( B ) Sex-specific splicing results in the formation of a female-specific traF isoform . A non-sex-specific isoform is produced in males and females , traC . Exons are represented by black boxes , 5′ UTR and 3′ UTR by grey boxes . Sites for the primer-pairs used for detection of both isoforms , p1 and p2 , span an intron in both splice forms . The PCR product from the spliced mRNA is 87 bp ( unspliced primary transcript would produce a product of 154 bp ) . Primers p3 and p4 span the first intron of traF . Note that the positions of the primer pairs are approximate . The positions of the 2 miR-124 target sites are indicated . ( C ) Luciferase reporter assays . S2 cells were transfected to express a tra 3′UTR luciferase reporter or a control reporter with the SV40 3′UTR . Cells were co-transfected to express miR-124 or with a vector-only control , and a Renilla luciferase reporter as a control for transfection efficiency . Data show the mean ratio of firefly to Renilla luciferase activity based on three independent replicates . Error bars represent SEM . p<0 . 05 using two-tailed unpaired Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 012 As a first step to determine whether tra might be a functionally important target of miR-124 in vivo , we examined tra transcript levels by quantitative RT-PCR in RNA samples from control and miR-124 mutant male heads . The tra primary transcript undergoes sex-specific splicing in females to produce traF , which encodes a splicing factor ( Figure 6B ) . An alternate splice form is produced in both males and females , and is thought to produce a non-functional protein . Using primers that recognize the female-specific form , we observed that traF mRNA increased ∼2 . 5-fold in the mutant and returned to near normal levels in the rescued mutant ( Figure 7A , *p<0 . 05 ) . The female-specific traF splice form can be detected at low levels in control males by qPCR , at a few percent of the level found in females ( Figure 7—figure supplement 1 ) . This likely reflects a low level of improper splicing . 10 . 7554/eLife . 00640 . 013Figure 7 . miR-124 acts through regulation of transformer . ( A ) Elevated expression of traF transcript measured by quantitative real-time PCR using RNA isolated from male flies of the indicated genotypes ( primer pair p3 and p4 ) . actin 42A was used as an internal control for normalization . Data represent the average of five independent experiments ± SEM . Although traF transcript levels are low in control males , they were detected by quantitative real-time PCR ( traces are shown in Figure 6 ) . ( B ) Percentage of males achieving copulation with CS females in 30 min . Data represent the mean of more than 20 movies per genotype ± SD . Genotypes: CS: canton S control; miR-124 refers to the trans-heterozygous mutant combination miR-124Δ4/miR-124 Δ177; 124>traRNAi refers to the trans-heterozygous mutant combination miR-124Δ4/miR-124 Δ177 carrying the miR-124-promoter Gal4 transgene and UAS-tra RNAi . Depletion of tra significantly improved performance of the miR-124 mutant males . ( C ) Courtship index comparing proportion of time CS control males spent courting decapitated males of the indicated genotypes . n: number of animals per sample . Data represent one of three independent trials performed , with comparable results . Depletion of tra significantly reduced the attractiveness of the miR-124 mutant males to normal levels . ( D ) Quantification of cVA levels in males of the indicated genotypes by GC-MS . Knocking down of tra in using miR-124Gal4 driver significantly rescued the changes cVA levels in miR-124 mutant males . Data represent the average of two ( for miR-124 > traRNAi ) or three replicates ( CS and miR-124 ) ±SEM . n = 15 in each replicate . ( E ) Quantification of 9-pentacosene levels in males of the indicated genotypes by GC-MS . Depletion of tra lowered 9-pentacosene levels to within control levels . Data represent the average of two ( for miR-124 > traRNAi ) or three replicates ( CS and miR-124 ) ±SEM . n = 15 in each replicate . ( F ) Courtship index comparing proportion of time CS control males spent courting decapitated males of the indicated genotypes . N = 28 animals per sample . Data represent one of three independent trials performed , with comparable results . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 01310 . 7554/eLife . 00640 . 014Figure 7—figure supplement 1 . Amplification of traF shown by quantitative real-time RT-PCR . ( A ) Detection of traF transcript in heads from 5-day-old control females ( purple line ) and 5-day-old males ( orange line ) was shown by the amplification curves from real-time quantitative RT-PCR experiments . The difference was ∼4 cycles , or 32-fold . ( B ) No amplification was observed in controls not treated with reverse transcriptase ( nonRT ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 014 Consistent with previous reports ( Chan and Kravitz , 2007; Fernandez et al . , 2010 ) , increased expression of TraF in the male brain proved to be sufficient to reduce mating success and to elicit male–male courtship ( not shown ) . If elevated traF expression contributes to the miR-124 mutant phenotype , we would expect reducing traF levels to ameliorate the mutant phenotype . For these experiments , a UAS-traRNAi transgene was expressed under miR-124-Gal4 control in the miR-124 mutant background . The transgene targets a region common to both the female and non-sex-specific splice forms . Lowering traF levels in the miR-124 expressing cells was sufficient to increase male–female mating success ( Figure 7B ) ; to reduce male–male courtship ( Figure 7C ) , to improve production of cVA by several fold ( Table 2 and Figure 7D ) , and to lower levels of 9-pentacosene ( Table 2 and Figure 7E ) . Lowering traF levels in neurons by expressing UAS-traRNAi under elav-Gal4 control also proved to be sufficient to suppress male–male courtship ( Figure 7F ) . These findings indicate that upregulation of transformer in the CNS of the miR-124 mutant is causally linked to the pheromone production and behavioral abnormalities in the mutant males . 10 . 7554/eLife . 00640 . 015Table 2 . GC-MS analysis of cuticular hydrocarbon extracts from control , miR-124 mutant , rescued mutants , and miR-124>tra-RNAi malesDOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 015Compound and elemental composition*Control† ( n = 3 ) mir-124 mutant† ( n = 3 ) Rescued mutant† ( n = 3 ) mir-124> tra-RNAi† ( n = 2 ) C21:0 ( nC21 ) 0 . 28 ± 0 . 10 . 21 ± 0 . 010 . 51 ± 0 . 030 . 35 ± 0 . 04C22:10 . 22 ± 0 . 020 . 24 ± 0 . 010 . 31 ± 0 . 020 . 34 ± 0 . 03cVA ( cis-vaccenyl acetate ) 3 . 86 ± 0 . 430 . 48 ± 0 . 04***2 . 57 ± 0 . 47*2 . 09 ± 0 . 23*C22:00 . 61 ± 0 . 030 . 60 ± 0 . 010 . 87 ± 0 . 050 . 70 ± 0 . 057 , 11-C23:20 . 14 ± 0 . 010 . 07 ± 0 . 0010 . 17 ± 0 . 020 . 11 ± 0 . 029-C23:1 ( 9-tricosene ) 1 . 10 ± 0 . 051 . 20 ± 0 . 021 . 57 ± 0 . 121 . 94 ± 0 . 077-C23:1 ( 7-tricosene ) 21 . 68 ± 1 . 1421 . 04 ± 0 . 2929 . 07 ± 2 . 12***28 . 95 ± 2 . 20***5-C23:1 ( 5-tricosene ) 2 . 56 ± 0 . 052 . 62 ± 0 . 052 . 71 ± 0 . 253 . 11 ± 0 . 40C23:0 ( nC23 ) 9 . 84 ± 0 . 1510 . 66 ± 0 . 0611 . 33 ± 0 . 2*10 . 35 ± 0 . 33C24:10 . 09 ± 0 . 050 . 19 ± 0 . 010 . 16 ± 0 . 020 . 22 ± 0 . 01C24:00 . 41 ± 0 . 010 . 52 ± 0 . 010 . 40 ± 0 . 030 . 44 ± 0 . 012-MeC241 . 52 ± 0 . 131 . 24 ± 0 . 021 . 81 ± 0 . 151 . 78 ± 0 . 09C25:20 . 41 ± 0 . 020 . 54 ± 0 . 020 . 74 ± 0 . 010 . 76 ± 0 . 069-C25:1 ( 9-pentacosene ) 6 . 13 ± 0 . 127 . 78 ± 0 . 05**5 . 74 ± 0 . 216 . 86 ± 1 . 027-C25:1 ( 7-pentacosene ) 26 . 01 ± 0 . 6926 . 97 ± 0 . 2514 . 09 ± 0 . 46***23 . 23 ± 1 . 15***5-C25:1 ( 5-pentacosene ) 1 . 41 ± 0 . 680 . 75 ± 0 . 01023 ± 0 . 020 . 59 ± 0 . 03C25:0 ( nc25 ) 2 . 65 ± 0 . 063 . 79±0 . 052 . 85 ± 0 . 072 . 68 ± 0 . 262-MeC267 . 72 ± 0 . 285 . 22 ± 0 . 01***6 . 64 ± 0 . 275 . 23 ± 0 . 21***9-C27:10 . 20 ± 0 . 010 . 25 ± 0 . 010 . 20 ± 0 . 020 . 18 ± 0 . 057-C27:11 . 15 ± 0 . 090 . 92 ± 0 . 020 . 41 ± 0 . 010 . 60 ± 0 . 08C27:0 ( nC27 ) 2 . 38 ± 0 . 123 . 77 ± 0 . 08*2 . 72 ± 0 . 161 . 92 ± 0 . 212-MeC287 . 66 ± 0 . 137 . 53 ± 0 . 057 . 85 ± 0 . 375 . 76 ± 0 . 33**C29:00 . 62 ± 0 . 061 . 37 ± 0 . 010 . 92 ± 0 . 10 . 52 ± 0 . 052-MeC300 . 98 ± 0 . 061 . 59 ± 0 . 031 . 41 ± 0 . 130 . 66 ± 0 . 02*The elemental composition is listed as the carbon chain length followed by the number of double bonds; 2-Me indicates the position of methyl branched compounds . †The normalized signal intensity for each compound and SEM is indicated; *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 when compared to control ( ANOVA followed by post-hoc Tukey HSD test ) .
It is generally thought that the sex determination pathway acts in a binary fashion , with particular spliced forms of the pathway being turned on or off , depending on the genetic sex of the cell ( Cline and Meyer , 1996 ) . The Sxl splicing factor is produced in genetic females and competes with U2AF , an essential splicing factor , for binding to a splice site in the tra primary transcript . In the presence of sufficient Sxl , U2AF binds to a lower affinity site and promotes splicing to produce the female-specific traF transcript ( Valcarcel et al . , 1993 ) . Nonetheless , low-levels of the female-specific Sxl and traF transcripts have been observed in males ( this report; Tarone et al . , 2005 ) . In the case of traF , this might reflect a low-level of U2AF binding to the low affinity site , even in the absence of Sxl . Leaky low-level expression of Sxl in males could be another contributing factor . Under normal conditions , the level of traF transcript found in males appears to be innocuous . Inappropriate splicing to produce traF transcript in males is expected to increase production of dsxF at the expense of dsxM . Interestingly , the modest increase in the level of traF in miR-124 mutant males led to reduced splicing of dsx to produce dsxM , but we did not observe a corresponding increase in the production of the female splice form dsxF ( Figure 8 ) . Production of dsxF requires the assembly of a complex containing TraF protein along with Tra2 and SR proteins at a series of sites that comprise the female-specific splice enhancer ( Lynch and Maniatis , 1996 ) . Our findings suggest that a modest increase in the level of TraF protein can interfere with production of dsxM without leading to production of dsxF . If low levels of TraF protein can lead to assembly of non-functional complexes , it is possible that their binding to the female-specific splice enhancer , might compromise male splicing without effectively promoting female splicing . 10 . 7554/eLife . 00640 . 016Figure 8 . Transcript level of dsxM , but not dsxF , is affected by miR-124 loss-of-function . Expression of dsxM ( A ) dsxF ( B ) transcripts measured by quantitative real-time PCR using RNA isolated from male flies of the indicated genotypes . actin 42A was used as an internal control for normalization . Data represent the average of four independent experiments ± SEM . **p<0 . 05 , NS: not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 016 When expressed at high levels , traF or dsxF can compromise male sexual differentiation and behavior ( Mckeown et al . , 1988; Villella and Hall , 1996 ) . Our findings provide evidence that a modest increase in the level of traF in miR-124 expressing cells in the CNS can interfere with male pheromone production . In this scenario microRNA mediated regulation ensures that leakiness in the production of traF is kept at levels that are functionally insignificant in the male . A modest increase in traF is not expected to have much effect in females , where the endogenous level is higher . microRNAs are well suited to provide an additional layer of noise reduction to post-transcriptional regulation mediated by splicing . Pheromone production is controlled by the sex determination pathway . Genetic experiments have demonstrated the role of the Dsx protein in the regulation of male and female specific pheromone profiles . In females , DsxF ensures the production of female-specific hydrocarbons while suppressing the production of male-specific hydrocarbons and other male-specific pheromones such as cVA . The presence of DsxM protein in males ensures that synthesis of female-specific hydrocarbons are suppressed in males ( Baker and Belote , 1983; Waterbury et al . , 1999 ) . In animals lacking miR-124 , the level of tra transcripts increases . The presence of TraF is expected to affect sexual differentiation in males . Gal4-directed expression of DsxF in an otherwise wild-type male ( also expressing DsxM ) has been reported to reduce cVA levels , whereas DsxF expression in dsx mutant males abolished cVA production completely ( Waterbury et al . , 1999 ) . Ectopic expression of DsxF in XY males has also been shown to cause production in female-specific diene-hydrocarbons such as cis , cis-7 , 11-heptacosadiene and cis , cis-7 , 11-nonacosadiene ( Waterbury et al . , 1999 ) . We did not detect these compounds in cuticular extracts from miR-124 mutant males . The difference is likely due to the absence of miR-124 expression in the oenocytes where the TraF–DsxF cascade is thought to exert its effect on female hydrocarbon production . Regulation of male-specific hydrocarbons is probably more complex and is likely to involve modulation from the nervous system . Many of the characteristic male compounds are also synthesized by the oenocytes , since genetic ablation of these cells abolished all male hydrocarbon production , but does not affect levels of cVA , produced in the ejaculatory bulb ( Billeter et al . , 2009 ) . However , feminization of the nervous system in XY males led to significant elevation of characteristic male hydrocarbons such as cis-7-tricosene and cis-9-pentacosene , although no gain of female hydrocarbons was observed ( Fernandez et al . , 2010 ) . Brain specific depletion of desat1 , which encodes a desaturase enzyme involved in pheromone biosynthesis , was shown to alter pheromone production ( Bousquet et al . , 2012 ) . We noted the presence of unconventional sites that potentially could be targeted by miR-124 in the open reading frame and 5′ UTR of the desat1 mRNA ( Figure 9 ) . The function of these sites has not been tested . If functional , desat1 could be overexpressed in the miR-124 mutant . While the consequences of elevated Desat1 expression are not known , the possibility exists that miR-124 might act via multiple targets in the CNS to indirectly modulate pheromone production in peripheral tissues . In moths , the neuropeptide PBAN has been linked to control of pheromone production , suggesting a role for neuroendocrine control of sexual differentiation ( Jurenka and Rafaeli , 2011 ) . Our findings provide evidence that miR-124 regulation of transformer may act in the context of neuroendocrine control of male pheromone production . 10 . 7554/eLife . 00640 . 017Figure 9 . miR-124 sites on desat1 and elo68α transcripts . Left: sequences of two potential miR-124 sites in desat1 transcript . Top: a 6-mer site in the coding sequence common to all the isoforms; Bottom: an unconventional site with 2 GU base pairs in the 5′UTR of desat1-RC isoform . Right: sequences of two potential miR-124 sites on elo68α transcript . Seed pairing in both sites are weak . All of these sites are unconventional and it is uncertain whether they would show regulation by miR-124 . Their function has not been tested experimentally . DOI: http://dx . doi . org/10 . 7554/eLife . 00640 . 017
Flies were maintained on standard yeast-cornmeal-agar medium at 25°C , 60% relative humidity on a 12:12 light-dark cycle . Canton-S was used as the wild-type control . In all experiments , miR-124 mutants were a transheterozygous combination of two independently generated alleles . The miR-124Δ4 and miR-124 Δ177 targeted knockout alleles are described in ( Weng and Cohen , 2012 ) . The original knockout alleles contain a mini-white genetic marker flanked by LoxP sites . Because mini-white can affect behavior , the marker was excised from the original miR-124Δ177 , w+ and miR-124Δ4 , w+ alleles by crossing to Cre-expressing flies , as described ( Chen et al . , 2011 ) . mini-white-excised derivatives of miR-124Δ177 and miR-124Δ4 were each backcrossed to Canton S for six generations prior to behavioral tests . The deficiency line uncovering the miR-124 locus used in Figure 1A is Bloomington stock BL7837 . For genetic rescue experiments , the mini-white reporter in miR-124 Δ177 RMCE allele was replaced with a miR-124 hairpin fragment , as described ( Weng and Cohen , 2012 ) . The rescued mutant flies were homozygous for this chromosome ( Figures 2 , 4 and 6 ) . The miR-124 promoter Ga4 transgene is described in ( Weng and Cohen , 2012 ) . The UAS-tra-RNAi transgene was Bloomington Stock #28 , 512 . For courtship assays , males were collected at late pupal stage and aged individually for 5 days; target flies were collected at late pupal stage and aged for 5 days in groups of 20/vial . Behavior assays were performed 2–4 hr before lights off , 25°C , 60% relative humidity under normal ambient light . Courtship assays were carried out as described ( Demir and Dickson , 2005 ) . For male–female assays , Canton-S virgin females served as mating targets . 5-day-old socially naïve Canton-S , miR-124 mutants or miR-124-rescue males were tested . Courtship behavior was videotaped for 45 min after a virgin female and a test male were introduced into the courtship chamber by gentle aspiration . The courtship index is the proportion of time males spend courting within a 10-min observation period . Male–female courtship assays were carried out as described ( Demir and Dickson , 2005 ) . 5-day-old socially naïve Canton-S males were paired individually with either 5-day-old Canton-S or 5-day-old miR-124 virgin females . Courtship behavior was videotaped for 45 min after pairing . The percentage of females that accepted copulation by CS males was recorded for each genotype . Male–male courtship assays: on the day of the experiment , target males were briefly anesthetized on ice and decapitated with a razor blade before being introduced into courtship chambers . Individual intact test males were gently aspirated into the chamber containing a decapitated target and the behavior of the test males was recorded for 45 min . Round chambers of 10 mm diameter and 4 mm height were used for the mating competition assay . Mutants and wild-type male flies were collected at late pupal stage and isolated in standard food vials . On the fourth day post eclosion , mutants and controls were anaesthetized briefly and marked with acrylic paint at the back of the thorax . On the fifth day , a mutant and a wild-type with different colors were introduced into a courtship chamber containing a Canton-S virgin female and were videotaped for 70 min . The percentage of copulation success for both mutants and controls was measured . The fighting chamber was 16 mm in diameter and 9 mm in height . A food patch was introduced by pipetting 50 μl of melted standard fly food in the center of the chamber . Pairs of socially naïve 5 day-old male flies were aspirated gently into the fighting chamber . Behavior was recorded for 45 min . Experimental and control groups were videotaped simultaneously . Fighting latency measures the number of encounters until the first antagonistic encounter between the pair . Frequency reports the number of incidents , including lunging and fencing , observed in 30 min . 5-day-old socially naïve CS or miR-124 mutant males were individually aspirated into the courtship chamber used for the male–female courtship assay as described above . The activity of the fly was videotaped for 15 min by a Sony Camcorder and analyzed by ImageJ . The velocity of the fly in the first 10 min of observation was recorded . Flies were reared as for the behavior assays and aged in groups of 15–20 flies per vial . Six replicates of fifteen 5-day-old male flies were anaesthetized on ice and placed into 1 . 8 ml glass microvials with Teflon caps ( s/n 224740; Wheaton , Millville , NJ ) . 120 µl of hexane ( Fisher Chemicals , Pittsburgh , PA ) containing 10 μg/ml of hexacosane ( Sigma-Aldrich , St Louis , MO ) standard was added into each vial and incubated at room temperature for 20 min . 100 μl of solvent was transferred into a new vial and evaporated under a gentle stream of nitrogen . Extract was stored at −20°C until analysis . At least three biological replicates were prepared per genotype . Extracts were re-dissolved in 60 µl of hexane and transferred into GC-MS vials ( Supelco ) . Analysis was run in a 5% phenyl-methylpolysiloxane ( DB-5 , 30 m length , 0 . 32 i . d . , 0 . 25 μm film thickness , Agilent , Santa Clara , CA ) column and GCMS QP2010 system ( Shimadzu , Kyoto , Japan ) with an initial column temperature of 50°C for 2 min and increment to 300°C at a rate of 15 °C/min in splitless mode . The relative signal intensity for each hydrocarbon species was calculated by dividing the area under the chromatography peak by the total area under all of the peaks . The values from 3–6 replicate measurements were averaged . For application of synthetic compounds to target flies , 9 μg of synthetic cVA ( Cayman Chemical Company Ann Arbor , MI ) was diluted in 200 μl of hexane and introduced into a 1 . 8-ml glass microvial . The hexane was evaporated under a gentle flow of nitrogen , leaving the compound as a residue coating the bottom of the vial . Flies were briefly anaesthetized on ice , transferred to coated vials in groups of seven , and subjected to three vortex pulses lasting 20 s each , with 10 s pauses between each pulse . The perfumed flies were allowed to recover for about 1 hr in fresh vials with standard food . Six flies from each group were used for behavioral tests and the remaining fly was subjected to hydrocarbon analysis by Direct Analysis in Real Time mass spectrometry to monitor effective transfer of the test compound to the flies . The atmospheric pressure ionization time-of-flight mass spectrometer ( AccuTOF-DART , JEOL USA , Inc . ) was equipped with a DART interface and operated in positive-ion mode at a resolving power of 6000 ( FWHM definition ) . Mass accuracy is within ±15 ppm . The DART interface was operated using the following settings: the gas heater was set to 200°C , the glow discharge needle was set at 3 . 5 kV . Electrode 1 was set to +150 V and electrode 2 was set to +250 V . He2 gas flow was set to 2 . 5 l/min . Under these conditions , mostly protonated ( [M + H]+ ) molecules are observed . Using clean forceps , an anaesthetized fly was picked up by both wings , making sure not to damage the fly . The fly was placed in a stream of charged helium gas until peaks of triacylglycerides start to appear . All fly samples were placed approximately in the same location in the DART source for the same amount of time in order to obtain reproducible spectra . Six flies from each genotype were measured . Polyethylene glycol ( Sigma-Aldrich ) was used as calibrant . Relative quantification of compound abundance was performed by normalizing the areas under the signal corresponding to cVA ( [M + H]+ 311 . 29 ) to the tricosene signal ( [M + H]+ 323 . 36 ) . DART MS is unable to differentiate isoforms of tricosene therefore the tricosene signal represents the summed signal intensity from 5 , 7 , and 9-Tricosene . Tricosene was selected as the normalization peak due to the unaltered levels in mutants compared to CS controls in GC-MS . Statistical analysis for behavior assays and hydrocarbon quantification was done using Prism 4 ( GraphPad Software , La Jolla , CA ) . For behavior data , a nonparametric Mann–Whitney test was used to compare two samples . Kruskal–Wallis test followed by Dunn’s post-test was used to compare multiple samples . For hydrocarbon analysis , multi-way ANOVA followed by Tukey HSD post-test was performed . S2 cells were transfected in 24-well plates with 250 ng of miRNA expression plasmid or empty vector , 25 ng of firefly luciferase reporter plasmid , and 25 ng of Renilla luciferase DNA as a transfection control . Transfections were performed in triplicate in at least three independent experiments . 60 hr after transfection , dual luciferase assays ( Promega , Madison , WI ) were performed according to manufacturer’s instructions . | Like many animals , the fruit fly Drosophila uses pheromones to influence sexual behaviour , with males and females producing different versions of these chemicals . One of the pheromones produced by male flies , for example , is a chemical called 11-cis-vaccenyl-acetate ( cVA ) , which is an aphrodisiac for female flies and an anti-aphrodisiac for males . The production of the correct pheromones in each sex is genetically controlled using a process called splicing that allows a single gene to be expressed as two or more different proteins . A variety of proteins called splicing factors ensures that splicing results in the production of the correct pheromones for each sex . Sometimes , however , the process by which sex genes are expressed as proteins can be ‘leaky’ , which results in the wrong proteins being produced for one or both sexes . Small RNA molecules called microRNAs act in some genetic pathways to limit the leaky expression of genes , and a microRNA called miR-124 carries out this function in the developing brain Drosophila . Now , Weng et al . show that miR-124 also helps to regulate sex-specific splicing and thereby to control pheromone production and sexual behaviour . Mutant male flies lacking miR-124 were less successful than wild-type males at mating with female flies , and were almost always rejected if a female fly was given a choice between a mutant male and a wild-type male . Moreover , both wild-type and mutant male flies were more likely to initiate courtship behaviour towards another male if it lacked miR-124 than if it did not . The mutant male flies produced less cVA than wild-type males , but more of other pheromones called pentacosenes , which is consistent with the observed behaviour because cVA attracts females and repels males , whereas pentacosenes act as aphrodisiacs for male flies in large amounts . Weng et al . showed that these changes in the production of pheromones were caused by an increased expression of the female version of a splicing factor called transformer in the mutant males , but further work is needed to understand this process in detail . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"genetics",
"and",
"genomics"
] | 2013 | miR-124 controls male reproductive success in Drosophila |
Central thalamus plays a critical role in forebrain arousal and organized behavior . However , network-level mechanisms that link its activity to brain state remain enigmatic . Here , we combined optogenetics , fMRI , electrophysiology , and video-EEG monitoring to characterize the central thalamus-driven global brain networks responsible for switching brain state . 40 and 100 Hz stimulations of central thalamus caused widespread activation of forebrain , including frontal cortex , sensorimotor cortex , and striatum , and transitioned the brain to a state of arousal in asleep rats . In contrast , 10 Hz stimulation evoked significantly less activation of forebrain , inhibition of sensory cortex , and behavioral arrest . To investigate possible mechanisms underlying the frequency-dependent cortical inhibition , we performed recordings in zona incerta , where 10 , but not 40 , Hz stimulation evoked spindle-like oscillations . Importantly , suppressing incertal activity during 10 Hz central thalamus stimulation reduced the evoked cortical inhibition . These findings identify key brain-wide dynamics underlying central thalamus arousal regulation .
The thalamus plays an important role in coordinating global brain signals responsible for cognition and normal waking behavior ( Sherman and Guillery , 1996; Llinás et al . , 1998; Mitchell et al . , 2014 ) . The central thalamus and intralaminar nuclei , in particular , have been postulated to play a critical and unique function in regulating arousal , attention , and goal-directed behavior ( Schiff and Pfaff , 2009; Mair et al . , 2011 ) . This idea dates back to the first demonstrations that direct and indirect electrical stimulations of central thalamus control cortical electroencephalography ( EEG ) dynamics and elicit behavioral transitions between drowsiness/relaxation and wakefulness/attention ( Moruzzi and Magoun , 1949; Hunter and Jasper , 1949; Fuster , 1958 ) . Since the initial identification of central thalamus’s causal effect on brain state and behavior , significant support for its role in arousal regulation has come from anatomical and histological studies . Steriade and Glenn ( Steriade and Glenn , 1982 ) identified a monosynaptic pathway from the mesencephalic reticular formation to the central lateral ( CL ) and paracentral ( PC ) nuclei of central thalamus that projects to cerebral cortex and striatum . In addition to this input , the central thalamus receives projections from other arousal systems , including norepinephrinergic innervation from locus coeruleus ( Vogt et al . , 2008 ) and cholinergic innervation from the upper brainstem and basal forebrain ( Heckers et al . , 1992 ) . In combination with these inputs , the diffuse projections of central thalamus allow it to influence the overall excitability of cortex during states of attention . For example , virtually all relay cells of the CL nucleus project to both striatum and cerebral cortex ( Deschenes et al . , 1996 ) . Studies on the physiological properties of central thalamus also show that it is tightly coupled to arousal regulation . First , variations in the level of activity within the intralaminar nuclei are linked to changes in behavioral alertness ( Kinomura et al . , 1996; Shirvalkar et al . , 2006; Mair and Hembrook , 2008; Schiff et al . , 2013; Giber et al . , 2015 ) , including transitions during the normal sleep–wake cycle and acute cognitive enhancements such as improved working-memory and sustained attention ( Baker et al . , 2012 ) . Similarly , lesions of the central thalamus can produce enduring cognitive impairments , including reduced attentional processing and memory ( Guberman and Stuss , 1983; Mair et al . , 1998; Van Der Werf et al . , 1999; Newman and Burk , 2005 ) , hypersomnolence ( Bassetti et al . , 1996 ) , or even coma ( Castaigne et al . , 1981; Plum , 1991 ) . Indeed , neuronal loss across central thalamus has been associated with severely disabled and vegetative patients following severe traumatic brain injury ( Maxwell et al . , 2006 ) . In addition , electrical stimulation of the central thalamus at low frequencies is associated with absence seizures and behavioral arrest in animal models ( Hunter and Jasper , 1949 ) and human subjects ( Velasco , 1996 ) . Human imaging studies have also found that anesthesia-induced loss of consciousness is associated with disrupted thalamocortical functional connectivity in regions consistent with the intralaminar nuclei ( Akeju , 2014 ) . According to the mesocircuit hypothesis of forebrain dysfunction , the central thalamus , which has a strong activating role in driving cortical and striatal neurons ( Schiff , 2010 ) , is under tonic inhibition by GABAergic pallidal neurons ( Grillner et al . , 2005 ) . This GABAergic population is itself inhibited by the striatal neurons driven by central thalamus , creating a positive feedback loop . Thus , when the thalamostriatal and thalamocortical projections from central thalamus are partially lost due to brain injury , it causes disinhibition of the pallidum and increased inhibition of the remaining central thalamus neurons , which further reduces cortical activation . While this down-regulation is predicted to have broad modulatory impact on global dynamics , deep brain stimulation ( DBS ) of central thalamus has been explored as a potential means of reversing its effects and facilitating arousal regulation in the minimally conscious state ( Shirvalkar et al . , 2006; Schiff et al . , 2007; Mair and Hembrook , 2008; Smith et al . , 2009; Mair et al . , 2011; Schiff , 2012; Fridman et al . , 2014; Gummadavelli et al . , 2015 ) . Despite success in a single-subject clinical study ( Schiff et al . , 2007 ) , identification of circuit-level mechanisms that link therapeutic efficacy of central thalamus DBS to specific stimulation parameters remains challenging and at present limits the clinical efficacy of DBS in subjects with traumatic brain injury . Thus , while significant progress has been made in understanding the connections of central thalamus and its behavioral correlates ( Van der Werf et al . , 2002; Schiff , 2008 ) , relatively little is known about the dynamic function of these circuits , and no clear mechanism exists to explain how – or indeed , if – a single population in central thalamus can act as a switch on the global brain state . To overcome such obstacles and dissect the dynamic influence of central thalamus on global brain networks , we combined targeted optogenetic control of excitatory relay neurons within central thalamus with whole-brain fMRI readouts , EEG , and single-unit recordings . The combination of optogenetic stimulation with fMRI ( ofMRI ) has been demonstrated to be an effective method for mapping the functional role of specific genetically- and spatially-defined neuronal populations at various brain regions and different frequencies of stimulation ( Lee et al . , 2010; Desai et al . , 2011; Weitz et al . , 2015; Liang et al . , 2015; Takata et al . , 2015; Duffy et al . , 2015; Byers et al . , 2015 ) . In the present work , we sought to apply this approach to uncover the downstream effects of distinct firing patterns by central thalamus relay cells at the whole-brain level – a visualization uniquely possible with ofMRI . Prior work beginning with Morison and Dempsey ( Morison and Dempsey , 1942 ) , who envisioned “dissecting… the electrical activity of the cortex on the basis of its relations with the thalamus , ” has shown that high- and low-frequency electrical stimulation of thalamic nuclei can produce distinct cortical activation patterns . For example , early studies of the intralaminar nuclei per se showed that low-frequency stimulation evokes slow-wave activity and spindle bursts in cortical EEG , which are associated with primary generalized absence seizures , loss of consciousness , and drowsiness ( Hunter and Jasper , 1949; Jasper , 1949; Seidenbecher and Pape , 2001 ) . Conversely , high-frequency electrical stimulation has been shown to desynchronize the cortical EEG signal ( Moruzzi and Magoun , 1949 ) , which is associated with behavioral arousal . While these studies set early hypotheses on the mechanisms of arousal regulation , the non-selective nature of electrical stimulation has prevented the observed responses from being attributed specifically to relay cells , and not synaptic afferents or fibers of passage mixing together in a bulk activation effect . More importantly , although a picture of whole brain activity can be vaguely inferred from electrophysiology recordings , these techniques cannot provide a direct visualization of activity across individual brain regions over the entire brain . Because ofMRI can provide spatial and temporal information on the whole-brain scale during perturbations of specific neural circuitry , we applied this technique to study the causal role of central thalamus relay neurons in activating forebrain networks . In addition , following on novel results described below , we were led to examine the interplay between central thalamus and zona incerta ( ZI ) , a subcortical region implicated in the modulation of absence seizures in rats ( Shaw et al . , 2013 ) . The ZI , a mostly GABAergic region , has been shown to limit the transmission of ascending sensory information via feedforward inhibition of higher order thalamic nuclei ( Barthó et al . , 2002; Trageser and Keller , 2004; Lavallée et al . , 2005; Trageser et al . , 2006 ) . Such activity can induce a state of reduced sensory processing , similar to the behavioral quiescence induced by low-frequency central thalamus stimulation . These studies suggest a powerful control exerted by ZI over brain state and higher level processing , much like central thalamus . However , the possible involvement of ZI in central thalamus arousal circuits remains unexplored . Here , we investigated the electrophysiology responses of this region during targeted stimulation of central thalamus at frequencies that either facilitate or suppress attention and arousal . To determine whether ZI plays an active role in these circuits , we also used optogenetics to specifically inhibit neurons in this region during central thalamus stimulation . This experimental paradigm was used to infer ZI’s functional contribution to the central thalamus-driven brain circuit dynamics measured with fMRI and electrophysiology .
To investigate the specific role of central thalamus , we applied optogenetic techniques to control relay cells in a spatially and temporally precise manner . We performed a stereotactic injection in the right CL and PC intralaminar nuclei of central thalamus with adeno-associated virus carrying channelrhodopsin-2 ( ChR2 ) and the fluorescent reporter protein EYFP under control of the CaMKIIa promoter . This promoter is expressed primarily in excitatory neurons , the vast majority of which in thalamus are relay cells ( Smith , 2008; Ellender et al . , 2013 ) . Of cells identified within the bulk injection area , 35% were EYFP-positive , and 97% of EYFP-positive cells co-expressed CaMKIIa , indicating high sensitivity for stimulation of excitatory neurons ( n = 2 rats , 831 cells; Figure 1—figure supplement 1 ) . While ChR2-EYFP expression extended beyond these two nuclei ( Figure 1A ) , targeted stimulation of the intralaminar nuclei was achieved by ( a ) stereotactic placement of the implanted optical fiber , as confirmed with high-resolution T2-weighted structural MR images ( Figure 1B , C ) , and ( b ) spatially restricted illumination ( Figure 1A , B ) . We initially injected and cannulated 47 rats using the central thalamus as the stereotactic target ( -3 . 2 mm AP , +1 . 5 mm ML , -5 . 5 mm DV ) . However , the intralaminar nuclei are relatively small and difficult to accurately target . We therefore used only a subset of these animals based on the empirically observed distribution of optical fiber tip locations using T2-weighted MRI scans ( Figure 1B; <0 . 85 mm distance from target coordinate ) . Of the 18 rats that had an accurately localized implant location , two exhibited a general absence of fMRI activity – most notably at the site of stimulation – and were excluded , leaving 16 animals for further analysis . 10 . 7554/eLife . 09215 . 003Figure 1 . Targeted stimulation of central thalamus evokes positive BOLD changes and increases in neuronal firing at the site of stimulation . ( A ) Representative wide-field fluorescence image shows robust ChR2-EYFP expression throughout central thalamus , overlaid with the estimated cone of excited tissue shown to scale . ( B ) Empirically observed locations of fiber optic implants in initial cohort of 47 rats , estimated using high-resolution structural MRI scans . Of these animals , 18 had implant locations that were accurately localized to the central thalamus ( <0 . 85 mm from target site , shown as dashed circle and cross ) . Two were excluded based on lack of thalamic activation , leaving n = 16 rats for further analysis . Black dots indicate implant coordinates of 16 animals used for analysis . Gray dots indicate implant coordinates of 31 rejected animals . ( C ) Representative T2-weighted anatomical MRI scan used to estimate implant location , marked with arrow . ( D ) Schematic of 23 coronal slices acquired during ofMRI experiments . Slice numbers correspond to activation maps in Figure 2 . ( E ) Average time series of significantly modulated voxels within the ipsilateral thalamus ROI ( see Figure 2D ) exhibit robust positive BOLD responses during repeated 20 s periods of stimulation at 10 , 40 , and 100 Hz , indicated by blue bars . Values are mean ± s . e . m . across animals ( n = 16 , 10 , and 16 for each frequency , respectively ) . ( F ) Diagram of local in vivo optrode recordings during optical stimulation of central thalamus . Inset shows spike waveforms of a recorded neuron . ( G ) Representative peri-event time histogram of a recorded neuron showing the increase in firing rate within central thalamus during optical stimulation at each of the three frequencies tested . See also Figure 1—figure supplement 1 and Figure 1—source data 1 . BOLD: Blood-oxygen-level-dependent; ROI: Regions of interest . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 00310 . 7554/eLife . 09215 . 004Figure 1—source data 1 . Firing rates before , during , and after repeated 20 s stimulation periods for each of the five neurons recorded in central thalamus . Exact p values comparing thalamic firing rate before and during stimulation are provided . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 00410 . 7554/eLife . 09215 . 005Figure 1—figure supplement 1 . Specificity of ChR2 targeting for CaMKIIa-positive cells . Immunohistochemistry confirms the specific targeting of ChR2-EYFP to CaMKIIa-positive neurons in central thalamus . 35% of cells identified within the bulk injection area were EYFP-positive , and 97% of EYFP-positive cells co-expressed CaMKIIa ( n = 2 rats , 831 cells ) . Scale bar , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 005 In order to achieve a small volume of directly excited tissue limited to the intralaminar nuclei , we used a 62 . 5-μm diameter optical fiber . Assuming that an intensity of 1 mW/mm2 is required for ChR2 activation ( Aravanis et al . , 2007 ) , the specific power exiting from the fiber optic’s tip in these experiments ( 2 . 5 mW ) corresponds to a penetration depth of 1 . 08 mm and a volume of 0 . 08 mm3 over which ChR2+ neurons can be excited . Figure 1A illustrates this penetration depth and activation cone ( 11 . 7° half-angle of divergence ) to scale with the targeted nuclei , showing that stimulation is well restricted to the central thalamus . These two factors ( MR-validated stereotactic fiber placement and a small volume of excited tissue ) suggest that the effects reported here primarily derive from stimulation of excitatory relay neurons within the central thalamus . To explore the anatomical connectivity of transfected neurons in central thalamus , we collected ex vivo fluorescence microscopy images of ChR2-EYFP expression . Due to the spread of viral transfection ( Figure 1A ) , it is possible that the reported fluorescence reflects projections from adjacent thalamic nuclei as well . Nevertheless , in agreement with known projection systems of central thalamus , EYFP-expressing axons were observed throughout forebrain , including frontal cortex and striatum ( Figure 2—figure supplement 1 ) . In particular , the medial prefrontal , lateral prefrontal , cingulate , motor , and sensory cortices all received strong projections . This input was highly convergent at the superficial layers , with moderate but weaker projections present in middle layers as well . Furthermore , projections were significantly restricted to the hemisphere ipsilateral to virus injection for both cortex and striatum . While these anatomical connections provide a strong foundation for understanding how central thalamus can influence brain state , they do little to explain the dynamic nature of these circuits – for example , how stimulation of central thalamus at different frequencies can lead to distinct behavioral responses ( Hunter and Jasper , 1949; Velasco , 1996 ) . Therefore , to dissect the functional significance of these massive forebrain projections and visualize the large-scale spatial and temporal dynamics evoked by central thalamus stimulation , we combined optical stimulation with simultaneous in vivo whole-brain functional imaging ( Lee et al . , 2010 ) . During optogenetic fMRI experiments , 23 coronal slices with 0 . 5 × 0 . 5 mm2 in-plane resolution and 0 . 5 mm thickness were acquired at a frame rate of 750 ms using spiral k-space trajectories and a sliding window reconstruction algorithm to achieve high-spatiotemporal resolutions with whole-brain coverage ( bregma +5 . 2 to -5 . 3 mm; Figure 1D ) ( Fang and Lee , 2013 ) . Novel inverse Gauss-Newton methods were also used to correct for possible motion artifacts and optimize the robustness of detecting optogenetically evoked responses ( Fang and Lee , 2013 ) . For each experiment , we delivered 20 s periods of stimulation every minute for 6 min at 10 , 40 , or 100 Hz . This form of continuous steady-state stimulation mimics the approach used in clinical DBS and has been showed to evoke robust fMRI responses with optogenetic stimuli ( Lee , et al . , 2010; Duffy , et al . , 2015; Weitz et al . , 2015 ) . Indeed , stimulation at all three frequencies resulted in a robust positive blood-oxygen-level-dependent ( BOLD ) signal at the site of stimulation that was highly synchronized to light delivery , increased upon optical activation , and gradually returned to baseline following the end of stimulation ( Figure 1E ) . To confirm that this BOLD signal reflected underlying neuronal firing patterns , we next performed simultaneous single-unit recordings with stimulation using an optrode at the central thalamus ( Figure 1F ) . In agreement with the fMRI signal , stimulations at 10 , 40 , and 100 Hz all resulted in robust increases in the local neuronal firing rate ( Figure 1G; n = 5 neurons , p< 0 . 05 , Wilcoxon signed-rank test between the 20 s pre-stimulation and stimulation periods , 12 trials for each neuron ) . Both locally in the thalamus and at downstream , synaptically connected brain regions , the frequency of stimulation was a critical parameter in determining the extent of ipsilateral and contralateral BOLD activation – defined here as positive BOLD signals significantly synchronized to the block stimulation paradigm ( see ‘Materials and methods’ ) . In general , a much larger volume of brain tissue was activated by stimulation at 40 and 100 Hz compared to 10 Hz , with frontocortical areas and striatum being strongly activated at high frequencies ( Figure 2A–C; Videos 1–3 ) . To quantify these spatial differences in recruitment patterns , we calculated the total volume of positive and statistically significant BOLD signals evoked by stimulation in select region of interests ( ROIs ) ( Figure 2D ) . This difference in activation volume between low- ( 10 Hz ) and high- ( 40 or 100 Hz ) stimulation frequencies was significant at the thalamus , striatum , and medial prefrontal , lateral prefrontal , cingulate , motor , and sensory cortices ( Figure 2E–H ) . Striatal activity was primarily localized to the dorsal sector , with negligible activity occurring in the ventral region ( Figure 2B , C ) . Furthermore , BOLD activation was generally restricted to the ipsilateral hemisphere , although activation volumes in the contralateral striatum , lateral prefrontal cortex , motor cortex , and sensory cortex were all significantly greater during 100 Hz stimulation compared to 10 Hz stimulation ( Figure 2F–H ) . 10 . 7554/eLife . 09215 . 006Figure 2 . Spatial characterization of evoked fMRI signals . ( A-–C ) Average coherence maps of brain-wide activity during stimulation of excitatory central thalamus relay neurons at 10 , 40 , and 100 Hz . Warm colors indicate positive BOLD responses , while cool colors indicate negative BOLD responses ( see ‘Materials and methods’ ) . ( D ) Regions of interest ( ROIs ) used for quantitative analysis of spatial ofMRI activation patterns . ( E ) The amount of active volume ( positive signal with coherence > 0 . 35 ) in the ipsilateral thalamus is significantly greater during 40 and 100 Hz stimulations than 10 Hz stimulation . Thalamic recruitment is relatively limited on the contralateral side . ( F ) Activation of the ipsilateral striatum is significantly greater during 40 and 100 Hz stimulations than 10 Hz stimulation . Activation of the contralateral striatum is limited across frequencies , although there is an increase from 10 to 100 Hz . ( G ) Medial and lateral prefrontal cortex exhibit a significantly greater volume of activation during 40 and/or 100 Hz stimulation than 10 Hz stimulation . Activity in the contralateral hemisphere is limited across all tested frequencies , although there is an increase from 10 to 100 Hz . ( H ) Activation of cingulate , motor , and somatosensory cortex is each greater during 40 and 100 Hz stimulations than 10 Hz stimulation . The contralateral motor and sensory cortices are also activated to a greater extent during 40 and/or 100 Hz stimulation . Scale bars in panels A through C represent 2 mm . Asterisks in panels E through H indicate significant differences for 10 versus 40 Hz and 10 versus 100 Hz stimulations . *p < 0 . 05 , **p < 0 . 005 , ***p < 0 . 001 , one-sided Wilcoxon signed-rank tests , corrected for multiple comparisons . Error bars represent mean ± s . e . m . across animals . n = 16 , 10 , and 16 animals for 10 , 40 , and 100 Hz , respectively . Abbreviations are as follows: i- ( ipsilateral ) , c- ( contralateral ) , Cg ( cingulate cortex ) , MC ( motor cortex ) , MPFC ( medial prefrontal cortex ) , LPFC ( lateral prefrontal cortex ) , SC ( sensory cortex ) , Str ( striatum ) , Th ( thalamus ) . See also Figure 2—figure supplements 1/2/3 and Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 00610 . 7554/eLife . 09215 . 007Figure 2—source data 1 . Animal-specific activation volumes for the twelve regions of interest shown in Figure 2E–H . Exact p values comparing the volume activated during 10 versus 40 Hz stimulation and 10 versus 100 Hz stimulation are provided . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 00710 . 7554/eLife . 09215 . 008Figure 2—figure supplement 1 . Representative fluorescence images of ChR2-EYFP at remote targets illustrate the massive projections to forebrain from transfected relay neurons in the right central thalamus . The bottom two rows provide magnified images of cortical and striatal regions used for quantitative ofMRI analysis . The top row provides the whole-brain slices from which these magnified images come . EYFP-expressing axonal projections are primarily localized to the ipsilateral hemisphere and to superficial layers in cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 00810 . 7554/eLife . 09215 . 009Figure 2—figure supplement 2 . Widespread and frequency-dependent recruitment of forebrain with optogenetics is distinct to stimulation of intralaminar nuclei of central thalamus . ( A ) Volumes of striatal and cortical activation ( i . e . positive BOLD signals with coherence greater than 0 . 35 ) during 40 Hz stimulation of central thalamus , presented with activation volumes during stimulation of other thalamic nuclei . Central thalamus is the only target to result in significant recruitment of striatum and prefrontal and frontal cortical regions . ( B ) Comparison of frequency dependent effects of central thalamus stimulation with those of intermediate hippocampus ( IH ) stimulation . Unlike central thalamus stimulation , which recruits significantly more volume in striatum , motor cortex , and sensory cortex at 40 Hz than at 10 Hz ( *p < 0 . 05 , **p< 0 . 005 , ***p < 0 . 001; one-sided Wilcoxon signed-rank test ) , recruitment of these regions during hippocampal stimulation does not exhibit a significant dependence on frequency . Note that activation data was not available ( N/A ) at MPFC or LPFC for hippocampus stimulation due to differences in field of view . Abbreviations are as follows: i- ( ipsilateral ) , c- ( contralateral ) , MPFC ( medial prefrontal cortex ) , LPFC ( lateral prefrontal cortex ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 00910 . 7554/eLife . 09215 . 010Figure 2—figure supplement 3 . The frequency-dependent recruitment of forebrain by central thalamus and its control over cortical BOLD signal polarity are preserved when pulse width is held constant . ( A ) Representative coherence map of brain-wide activity during stimulation of excitatory central thalamus relay neurons at 10 , 40 , and 100 Hz using a constant pulse width of 3 ms . Warm colors indicate positive BOLD responses , while cool colors indicate negative BOLD responses ( see ‘Materials and methods’ ) . ( B ) Quantification of positive BOLD responses in cortex and striatum ( n = 3 animals ) . Gray lines indicate animal-specific results , normalized to 100 Hz stimulation . Black lines indicate the average across animals . All six regions exhibit an increase in recruitment with frequency , consistent with the study’s main results when pulse width was varied to keep the duty cycle and total amount of light delivery constant . Regions of interest are the same as those used in Figure 2 . ( C ) Hemodynamic response functions evoked in somatosensory cortex during 10 , 40 , and 100 Hz stimulation of central thalamus using a constant pulse width of 3 ms . Consistent with the study’s main results , a negative BOLD signal is evoked at 10 Hz , while slow and fast positive BOLD responses are evoked at 40 and 100 Hz , respectively . Blood-oxygen-level-dependent . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 01010 . 7554/eLife . 09215 . 011Video 1 . Spatiotemporal dynamics of ofMRI activity during 10 Hz stimulation of excitatory relay neurons of the central thalamus . Highlighted voxels are restricted to those significantly synchronized to the block stimulation paradigm , as determined by frequency domain analysis . Color coding reflects the instantaneous relative percent modulation of each voxel’s hemodynamic response function , thresholded over ± ( 0 . 2 to 1 . 5 ) % . Laser status indicates the 20 s period of stimulation ( 2–22 s ) . Abbreviations are as follows: SC ( sensory cortex ) , Th ( thalamus ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 01110 . 7554/eLife . 09215 . 012Video 2 . Spatiotemporal dynamics of ofMRI activity during 40 Hz stimulation of excitatory relay neurons of the central thalamus . Highlighted voxels are restricted to those significantly synchronized to the block stimulation paradigm , as determined by frequency domain analysis . Color coding reflects the instantaneous relative percent modulation of each voxel’s hemodynamic response function , thresholded over ± ( 0 . 2 to 1 . 5 ) % . Laser status indicates the 20 s period of stimulation ( 2–22 s ) . Abbreviations are as follows: Cg ( cingulate cortex ) , LPFC ( lateral prefrontal cortex ) , MC ( motor cortex ) , MPFC ( medial prefrontal cortex ) , SC ( sensory cortex ) , Str ( striatum ) , Th ( thalamus ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 01210 . 7554/eLife . 09215 . 013Video 3 . Spatiotemporal dynamics of ofMRI activity during 100 Hz stimulation of excitatory relay neurons of the central thalamus . Highlighted voxels are restricted to those significantly synchronized to the block stimulation paradigm , as determined by frequency domain analysis . Color coding reflects the instantaneous relative percent modulation of each voxel’s hemodynamic response function , thresholded over ± ( 0 . 2 to 1 . 5 ) % . Laser status indicates the 20 s period of stimulation ( 2–22 s ) . Abbreviations are as follows: Cg ( cingulate cortex ) , LPFC ( lateral prefrontal cortex ) , MC ( motor cortex ) , MPFC ( medial prefrontal cortex ) , SC ( sensory cortex ) , Str ( striatum ) , Th ( thalamus ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 013 These results provide a direct , region-specific visualization of the widespread driving effect that central thalamus has been shown to exert over forebrain , and link prior anatomical and physiological studies on arousal regulation to spatially precise and quantitative measures of cortical and striatal activation . For example , the evoked responses are consistent with the unilateral nature of thalamocortical projections ( Figure 2—figure supplement 1 ) , but reveal that the contralateral cortex can still be modulated by unilateral stimulation of central thalamus , particularly at high frequencies . Importantly , stimulation of other thalamic nuclei failed to evoke similarly widespread activity in the striatum and cortex ( Figure 2—figure supplement 2A ) . Furthermore , large differences in forebrain activation between 10 and 40 Hz stimulations were not observed for other forms of subcortical stimulation ( Figure 2—figure supplement 2B ) , suggesting this is a distinct property of central thalamus . Throughout these experiments , a constant duty cycle of 30% was used to maintain the total amount of light delivery across frequencies and control for possible heating artifacts ( Christie et al . , 2013 ) . Because we wished to keep a 20 s pulse train for all stimulation frequencies and avoid possible differences introduced by neuronal adaptation , maintaining a constant duty cycle required unique pulse widths for each frequency ( i . e . 30 , 7 . 5 , and 3 ms for 10 , 40 , and 100 Hz , respectively ) . To rule out the possibility that these changes in pulse width were the primary cause of the above differences in forebrain recruitment , we repeated stimulations while maintaining a 3 ms pulse width . Visualization and quantification of evoked fMRI responses show that the increase in cortical and striatal activation with frequency was preserved ( Figure 2—figure supplement 3A , B ) . These data suggest that stimulation frequency was the primary factor in modulating forebrain fMRI activation . We next examined the temporal dynamics of cortical responses evoked during low- and high-frequency central thalamus stimulation . Despite targeted activation of excitatory neurons , the somatosensory cortex exhibited a strong negative BOLD signal during 10 Hz stimulation , suggesting a suppression of baseline activity ( Figures 2A and 3A , B ) . In stark contrast , central thalamus stimulations at 40 and 100 Hz led to positive changes in the BOLD signal at the somatosensory cortex ( Figures 2B , C and 3A , B ) . Thus , stimulation of the same excitatory population at different frequencies resulted in completely opposite responses at a downstream target . Importantly , these responses were preserved when pulse width was held constant in control experiments , indicating that stimulation frequency was the primary factor controlling this effect ( Figure 2—figure supplement 3A , C ) . 10 . 7554/eLife . 09215 . 014Figure 3 . The sign of evoked cortical activity depends on the frequency of central thalamic stimulation . ( A , B ) 10 Hz stimulation of central thalamus evokes a strong negative BOLD signal throughout ipsilateral somatosensory cortex , while 40 and 100 Hz stimulations evoke positive responses . Time series come from the sensory cortex ROI defined in Figure 2D . Hemodynamic response function ( HRF ) shows the average response to a single 20 s period of stimulation , indicated by blue bar . Error bars represent mean ± s . e . m . across animals . n = 16 , 10 , and 16 for 10 , 40 , and 100 Hz , respectively . ( C ) Diagram of in vivo recordings at somatosensory cortex during stimulation of excitatory central thalamus relay neurons . Inset shows spike waveforms of a recorded neuron . ( D , E ) Representative peri-event time histogram of a recorded neuron , and corresponding quantification of firing rate during the 20 s periods before , during , and after stimulation . Neural firing rate decreased within the somatosensory cortex during 10 Hz central thalamus stimulation , but increased during 40 and 100 Hz stimulations ( n = 17 trials each , *p < 0 . 05 , ***p < 0 . 001 pre vs . ON , two-tailed Wilcoxon signed-rank test; see Table 1 for further analysis ) . Values are mean ± s . e . m . See also Figure 2—figure supplement 3C and Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 01410 . 7554/eLife . 09215 . 015Figure 3—figure supplement 1 . Cortical spikes that occur during periods of inhibition driven by 10 Hz central thalamus stimulation exhibit a non-uniform distribution over time . ( Top figure ) Average peristimulus time histograms ( PSTHs ) of spike events in somatosensory cortex during 10 Hz central thalamus stimulation for six animals . Analysis was restricted to the 5 s time bin with the greatest number of neurons inhibited for each animal . PSTHs were calculated by aligning spike counts to the onset of individual 30 ms light pulses , summing over the 50 pulses delivered during the 5 s stimulation period , averaging across trials , and binning at 5 ms intervals for each inhibited neuron . Histograms were normalized by the corresponding spike count value during the 20 s pre-stimulation baseline period ( represented by the dashed red line ) , and averaged across neurons for each animal . Blue rectangles represent the 30 ms light pulse . Note that spike events are reduced relative to baseline for the majority of the 100 ms inter-stimulus period , but spike events that do occur have a non-uniform distribution that peaks 6–34 ms after stimulus onset . These patterns suggest that some thalamic stimuli induce spike events in cortex , despite the net suppression of activity relative to pre-stimulation levels . Animals presented include two used for ChR2-electrophysiology experiments in Table 1 and four used for combined ChR2-eNpHR electrophysiology experiments in Figure 5 . ( Bottom table ) Summary of PSTH peak latencies and spike fidelity for inhibited neurons in somatosensory cortex . Peak latency was defined as the 5 ms bin with highest spike count for each neuron’s PSTH . Spike fidelity represents the percentage of light pulses in the given 5 s bin of inhibition that evoke at least one spike during the 30 ms pulse . Values represent mean +/- s . t . e . across cells in the figure and table . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 015 While previous studies have hinted at similar findings of frequency-dependent polarity changes ( Logothetis et al . , 2010; Weitz et al . , 2015 ) , downstream positive and negative BOLD signals that result from selective stimulation of excitatory neurons at distinct frequencies have not yet been visualized and validated with electrophysiology . To define the neuronal underpinnings of these signals , we therefore performed single-unit extracellular recordings in the somatosensory cortex during central thalamus stimulation ( Figure 3C ) . In agreement with the BOLD activity observed during ofMRI experiments , 10 Hz stimulation resulted in a decrease in neuronal firing rate between pre-stimulation and stimulation periods ( Figure 3D , E; n = 10 of 11 recorded neurons ) . Conversely , stimulations at 40 and 100 Hz both led to increases in neuronal firing ( Figure 3D , E; n = 11 of 11 recorded neurons ) . Because the evoked firing rates appeared to change over the course of stimulation , we specifically compared the pre-stimulation firing rate to the average firing rates during consecutive 5 s periods of the 20 s stimulus ( i . e . 0–5 s , 5–10 s , 10–15 s , and 15–20 s; uncorrected p < 0 . 05 , Wilcoxon signed rank test; 17 trials for each neuron ) . Interestingly , the decrease in firing rate during 10 Hz stimulation occurred primarily over the interval from 5 to 15 s after stimulation began , while the increase in firing rate during 40 Hz stimulation occurred primarily over the first 10 s ( Table 1 ) . On the other hand , the increase in neuronal firing rate during 100 Hz stimulation was generally maintained throughout the 20 s stimulation period ( Table 1 ) . Such differences may reflect short-term plasticity of the thalamocortical pathway , which has previously been shown to exhibit frequency-dependent properties ( Castro-Alamancos and Connors , 1996a; 1996b ) . Peri-stimulus time histograms also revealed that spike events occurring during inhibition had a non-uniform distribution over time , which peaked between 6 and 34 ms after light onset ( Figure 3—figure supplement 1 ) . These data suggest that the glutamatergic thalamocortical input at 10 Hz sometimes generated action potentials . Notably , however , not every light pulse resulted in an immediate action potential . 10 . 7554/eLife . 09215 . 016Table 1 . Electrophysiology results from sensory cortex single-unit recordings . See also Table 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 01610 . 7554/eLife . 09215 . 017Table 1—source data 1 . Firing rates before , during , and after repeated 20 s stimulation periods for each of the 11 neurons recorded in somatosensory cortex . Exact p values comparing the somatosensory cortex firing rate before and during stimulation are provided . The 20 s stimulation period was divided into four consecutive 5 s blocks to evaluate the change in firing rate over time . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 017Stimulation frequencyEffect on sensory cortex firing ratePercentage of neurons with significant change in firing rate ( n = 11 ) 0–5 s after stim . onset5–10 s after stim . onset10–15 s after stim . onset15–20 s after stim . onset10 HzIncrease0%0%0%0%Decrease0%91%82%9%40 HzIncrease100%91%36%55%Decrease0%0%0%0%100 HzIncrease100%82%82%82%Decrease0%0%0%0% Together , these ofMRI and electrophysiological data indicate that neuronal activity throughout somatosensory cortex is suppressed at low frequencies of central thalamus stimulation and increased at high frequencies of stimulation . Because our stimulations were restricted to excitatory neurons with cell bodies located in central thalamus , the causal relationship between stimulation frequency and cortical excitation/inhibition can be attributed to the neurons’ initial firing pattern . These results add to a growing body of literature in systems neuroscience suggesting that a neuronal population’s firing pattern can have vastly different – even opposite – effects on downstream regions depending on its specific temporal code ( Dempsey and Morison , 1943; Logothetis et al . , 2010; Mattis et al . , 2014; Weitz et al . , 2015 ) . Given that stimulation was restricted to excitatory neurons , we hypothesized that the suppression of cortex during 10 Hz stimulation might be related to the frequency-dependent modulation of a GABAergic population . We chose to investigate the response properties of the ZI , which has been implicated in providing a powerful GABAergic modulation of 10 Hz spike-wave activity in spontaneous absence seizures in the rat ( Shaw et al . , 2013 ) . Anatomically , ZI sends direct GABAergic projections to somatosensory thalamic nuclei and sensory cortex ( Nicolelis et al . , 1995; Kolmac and Mitrofanis , 1999; Barthó et al . , 2002 ) . Functionally , ZI has also been shown to selectively gate sensory information processing in higher order thalamic nuclei through GABAergic inhibition ( Trageser and Keller , 2004; Lavallée et al . , 2005; Trageser et al . , 2006 ) . To investigate the involvement of ZI , we performed single-unit and field potential electrophysiology recordings in this region during simultaneous optogenetic stimulation of central thalamus at 10 or 40 Hz ( Figure 4A ) . EEG recordings were simultaneously collected in frontal cortex to directly evaluate the relationship between ZI activity and whole-brain arousal state , which is typically measured with forebrain EEG . The ZI was targeted using stereotactic localization and the well-defined somatotopic representation of this region ( Nicolelis et al . , 1992 ) . The electrode was targeted to -3 . 96 mm AP , +2 . 2–2 . 6 mm ML , +6 . 7–7 . 2 mm DV from dura . The ZI was identified according to a compatible depth reading , spike latencies consistent with a polysynaptic response ( on the order of 10 ms; Figure 4B ) , and a receptive field that responds to contralateral whisker stimulation , which ZI is known to possess ( Nicolelis et al . , 1992 ) . The electrode was initially lowered through the dorsal part of the VP thalamus ( approximately 1 . 5 mm above ZI ) , which also responds to whisker stimulation , until the recorded neurons did not respond to such a stimulus . The electrode was then lowered for another ~1 . 5 mm until the recorded neurons fired in response to whisker stimulation , indicating the ZI had been reached . 10 . 7554/eLife . 09215 . 018Figure 4 . Frequency-dependent spindle-like oscillations are evoked in zona incerta ( ZI ) . ( A ) Diagram of in vivo recordings at ZI and simultaneous EEG recordings in frontal cortex during optical stimulation of central thalamus in anesthetized animals . ( B ) Representative peri-event time histograms of spiking activity from recorded ZI neurons aligned to the onset of individual light pulses , summed over all pulses and trials . Peak spike latencies are approximately 10 and 8 ms for 10 Hz ( left ) and 40 Hz ( right ) stimulations , suggesting that recordings are performed at least one synapse away from the stimulated population in central thalamus . Schematics at top illustrate the 30% duty cycle pulse trains which lasted 20 s for each frequency . ( C ) Representative peri-event time histograms over the 20 s period of stimulation show increases in ZI firing during 10 and 40 Hz stimulations . Among the 28 isolated single-units in ZI ( n = 2 animals ) , most exhibited a significant increase in firing rate during stimulation ( n = 26 and 22 out of 28 neurons , respectively; p < 0 . 05 , one-tailed Wilcoxon signed-rank test with 20 trials for each cell ) . ( D ) Representative field potential recordings from the same channel and trial number during 10 Hz ( top ) and 40 Hz ( bottom ) stimulation of central thalamus . Four amplitude-modulated , spindle-like oscillations ( SLOs ) are evoked during 10 Hz stimulation ( marked by black triangles ) , while none are evoked during 40 Hz stimulation . Inset shows a zoomed-in SLO . ( E ) The number of SLOs was greater during 10 Hz stimulation than 40 Hz stimulation across 11 of 12 available channels ( n = 2 animals , 20 trials each , p < 0 . 01 , one-tailed Wilcoxon rank sum test ) . ( F ) When more than one SLO was evoked within the same 20 s period of 10 Hz stimulation , the distribution of inter-event intervals was centered at 6 . 6 ± 0 . 2 s ( s . e . m . ) . ( G ) Representative EEG recordings collected in frontal cortex during central thalamus stimulation and simultaneous ZI recordings . 10 Hz stimulation evokes a spike-wave response , which is associated with loss of consciousness and perceptual awareness , while 40 Hz stimulation evokes a low voltage fast response indicative of arousal . ( H ) ChR2-positive processes were observed in ZI , providing a basis for its recruitment during stimulation of central thalamus . i . c . : internal capsule . See also Figure 4—source data 1 , 2 . EEG: Electroencephalography . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 01810 . 7554/eLife . 09215 . 019Figure 4—source data 1 . Firing rates before , during , and after repeated 20 s central thalamus stimulation periods for each of the 28 neurons recorded in zona incerta . Exact p values comparing zona incerta firing rate before and during central thalamus stimulation are provided . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 01910 . 7554/eLife . 09215 . 020Figure 4—source data 2 . Quantification of spindle-like oscillation ( SLO ) occurrences in the zona incerta during 10 and 40 Hz stimulation of central thalamus . Exact p values comparing SLO counts during 10 versus 40 Hz stimulation are provided . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 020 Out of 28 isolated ZI neurons , the majority exhibited increases in their firing rate during central thalamus stimulation at both 10 and 40 Hz ( Figure 4C; n = 26 and 22 , respectively; p < 0 . 05 , Wilcoxon signed rank test between the 20 s pre-stimulation and stimulation periods , 20 trials for each neuron ) . However , a key difference was that large , amplitude-modulated spindle-like oscillations ( SLOs ) in the field potential occurred significantly more often during 10 Hz stimulation than 40 Hz stimulation ( Figure 4D , E ) . These oscillations exhibited an inter-event interval centered around 6 . 6 ± 0 . 2 s ( s . e . m . ) , similar to those observed in thalamus during sleep onset ( Contreras et al . , 1997 ) ( Figure 4F ) . Consistent with this , simultaneous EEG recordings in frontal cortex revealed strong spike-wave modulation during 10 Hz stimulation and lower amplitude , fast oscillations during 40 Hz stimulation , which are associated with loss of consciousness and aroused brain states , respectively ( Figure 4G ) . EYFP-expressing axons were also observed in ZI ( Figure 4H ) , indicating that central thalamus relay neurons may have direct connections to ZI and providing a possible anatomical substrate for these responses . The observation of spindle-like oscillations in ZI during 10 , but not 40 , Hz central thalamus stimulation indicates that this region can be uniquely engaged by central thalamus-driven networks . However , it remains unknown whether the evoked activity in ZI plays a causal role in driving the frequency-dependent inhibition of somatosensory cortex . To address this question , we injected the inhibitory opsin halorhodopsin ( eNpHR ) fused to the mCherry fluorescent marker and controlled by the pan-neuronal hSyn promoter into ZI of four animals expressing ChR2-EYFP in central thalamus ( Figure 5A , B , Figure 5—figure supplement 1 ) . Two new stimulation paradigms were explored: ( 1 ) 20 or 30 s continuous eNpHR activation , and ( 2 ) 20 s , 10 Hz central thalamus stimulation performed within a 30 s period of continuous eNpHR activation . Single-unit recordings were performed simultaneously at the ZI and sensory cortex during concurrent activation of these two opsins ( Figure 5C ) . 10 . 7554/eLife . 09215 . 021Figure 5 . Cortical inhibition driven by 10 Hz central thalamus stimulation depends on normal incertal processing . ( A ) Wide-field fluorescence image shows robust eNpHR-mCherry expression spatially localized to the right zona incerta . Scale bar , 1 mm . ( B ) Confocal images show eNpHR-mCherry localized to somatic membrane of neurons in zona incerta . Scale bar , 10 µm . Two hundred and nine out of 882 DAPI-stained cells co-expressed mCherry in ZI ( 24% , n = 2 animals ) . ( C ) Schematic of cortical electrophysiology recordings during 10 Hz stimulation of central thalamus and continuous ( cont . ) inhibition of zona incerta using ChR2 and eNpHR , respectively . ( D ) Peri-event time histogram of a representative neuron in zona incerta whose firing rate is suppressed during activation of eNpHR with 593 nm light . ( E ) Peri-event time histogram of a representative neuron in zona incerta whose firing rate remains suppressed throughout the period of 10 Hz central thalamus stimulation during eNpHR activation ( compare to Figure 4C ) . ( F ) Activation of eNpHR in zona incerta significantly reduces the change in incertal firing rate evoked by 10 Hz central thalamus stimulation in 60 of 70 neurons ( p < 0 . 05 , one-sided Wilcoxon rank sum test ) . Changes in firing rate are normalized to pre-stimulation levels . ( G ) Peri-event time histograms from a representative cortical neuron show that the inhibitory response evoked by 10 Hz central thalamus stimulation is reversed by simultaneously suppressing activity in zona incerta . Firing rates are normalized to the average pre-stimulation values . ( H ) Quantification of evoked changes in cortical firing rate during 10 Hz central thalamus stimulation with and without concurrent eNpHR activation . 50 out of 76 cells exhibit reduced inhibition when central thalamus stimulation is paired with eNpHR activation ( p < 0 . 05 , Wilcoxon rank sum test over 1 s bins ) . Changes in firing rate are normalized to pre-stimulation levels . ( I ) Confocal images show mCherry-positive axonal projections from zona incerta in somatosensory cortex . Scale bar , 20 µm . See also Figure 5—source data 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 02110 . 7554/eLife . 09215 . 022Figure 5—source data 1 . Evoked changes in incertal firing rate during 10 Hz central thalamus stimulation with and without concurrent eNpHR activation . Changes are normalized to pre-stimulation baseline levels . Exact p values are provided . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 02210 . 7554/eLife . 09215 . 023Figure 5—source data 2 . Evoked changes in cortical firing rate during 10 Hz central thalamus stimulation with and without concurrent eNpHR activation in zona incerta . Values represent the 1 s bin that exhibited the greatest number of cells with reduced inhibition for each animal . Changes are normalized to pre-stimulation baseline levels . Exact p values are provided . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 02310 . 7554/eLife . 09215 . 024Figure 5—figure supplement 1 . Wide-field fluorescence image of eNpHR expression in zona incerta , overlaid with the estimated cone of activated eNpHR ( i . e . inhibited neurons ) shown to scale . Penetration depth and volume were calculated to be 0 . 64 mm and 0 . 024 mm3 , respectively , using the methods described in ( Aravanis et al . , 2007 ) and a threshold light intensity of 5 mW/mm2 ( Mattis et al . , 2012 ) . The optical fiber had a diameter of 105 μm , NA of 0 . 22 , and half-angle of divergence of 9 . 3° . Penetration depth and activation volume correspond to an optical power of 3 mW exiting the fiber optic’s tip . Stimulation coordinate corresponds to -3 . 96 mm AP , +2 . 4 mm ML , and -6 . 7 mm DV . The thalamic reticular nucleus , another region of dense GABAergic neurons , is shown for reference . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 024 Among the 70 neurons recorded in ZI , delivery of 593 nm light resulted in a decrease in firing for 62 cells ( p < 0 . 05 , Wilcoxon signed rank test between 20 s pre-stimulation period and 20 or 30 s stimulation period , 15–20 trials for each neuron ) , indicating that illumination of halorhodopsin was successful in suppressing incertal activity . The evoked decrease in neuronal firing rate typically lasted throughout the duration of 593 nm light delivery ( Figure 5D ) . When halorhodopsin activation in ZI was paired with 10 Hz stimulation of central thalamus , the previously described increase in incertal firing ( Figure 4C ) was disrupted . In 60 out of 70 neurons , the difference in incertal firing rate between the 20 s 10 Hz central thalamus stimulation period and the pre-stimulation period was significantly reduced with concurrent eNpHR activation ( Figure 5F; p < 0 . 05 , one-sided Wilcoxon rank sum test , n = 10–20 trials ) . Figure 5E illustrates the suppression of ZI activity throughout the 20 s period of 10 Hz central thalamus stimulation in a representative neuron . These data indicate that activation of halorhodopsin significantly suppressed the incertal firing evoked by 10 Hz central thalamus stimulation , and successfully disrupted incertal processing . To determine whether this suppression of ZI affected the cortical activity driven by central thalamus stimulation , we quantified the changes in somatosensory cortex firing rate evoked by ChR2 activation with and without illumination of eNpHR . Seventy-six somatosensory cortex neurons were recorded , and the 20 s period of central thalamus stimulation was divided into four 5 s bins as before . Consistent with the data presented in Figure 3 , 68 cells ( 89% ) exhibited a decrease in firing during 10 Hz stimulation of central thalamus ( uncorrected p < 0 . 05 , Wilcoxon signed rank test; 10–15 trials for each neuron ) . Strikingly , however , suppression of ZI activity with eNpHR reversed this effect . Across animals , 50 out of 76 neurons ( 66% ) exhibited reduced inhibition when central thalamus stimulation was paired with eNpHR activation ( Figure 5H; p < 0 . 05 , Wilcoxon rank sum test over 1 s bins; 10–20 trials for each neuron ) . Indeed , a fraction of cells switched from inhibitory responses to excitatory ones . Figure 5G illustrates the firing patterns of one cell that exhibited an inhibitory response during 10 Hz central thalamus stimulation that was eliminated when ZI was simultaneously suppressed with eNpHR . Collectively , these data suggest that incertal activity during 10 Hz central thalamus stimulation has a net inhibitory effect on somatosensory cortex . In support of this influence being through direct anatomical connections , mCherry-positive axons were observed in the sensory cortex ( Figure 5I ) , consistent with previous reports of incerto-cortical projections ( Lin et al . , 1990 ) . These findings present a conceptually novel role of ZI in central thalamus arousal circuits . Finally , to relate these findings more directly to behavior associated with central thalamus arousal circuits and previous electrical stimulation studies , we performed 10 , 40 , and 100 Hz stimulations in asleep , unanaesthetized animals with simultaneous video and EEG recordings ( see ‘Materials and methods’ ) . Control ( pre-stimulus ) activity was consistent across frequencies of stimulation , as quantified with EEG band power in delta , theta , alpha , and beta bands ( Figure 6—figure supplement 1 ) . During 10 Hz stimulation , the majority of animals exhibited behavior indicative of an absence seizure , including freezing and behavioral arrest throughout stimulation followed by a return to sleep ( Figure 6A; n = 4/7 ) . In addition , the most common EEG response was a transition to slow spike-wave discharges ( Figure 6B , C; n = 5/7 ) , which are typically associated with loss of consciousness ( Mirsky and VanBuren , 1965 ) . In stark contrast , stimulations at 40 and 100 Hz resulted in behavioral transitions to an awake state , reflected by exploration and goal-directed movement ( Figure 6A; n = 4/7 and 4/6 , respectively ) . Similarly , the most common EEG pattern evoked by these high-frequency stimulations was a low voltage fast response ( Figure 6B; n = 3/7 and 6/6 , respectively ) , indicative of cortical activation and desynchronization . Collectively , these phenomena are consistent with the patterns of cortical and striatal recruitment observed with ofMRI . Moreover , the slow spike-wave and low voltage fast EEG responses evoked during behavioral experiments ( Figure 6C , D ) match those recorded under anesthetized conditions ( Figure 4G ) , further linking the network activation patterns revealed by ofMRI to the arousal responses reported here , as well as those reported in early stimulation studies ( e . g . [Hunter and Jasper , 1949] ) . 10 . 7554/eLife . 09215 . 025Figure 6 . Optogenetic stimulation of central thalamus in asleep animals modulates brain state in a frequency-dependent manner . ( A ) Low-frequency stimulation ( 10 Hz ) in a majority of animals ( n = 4/7 ) evokes behavioral absence seizures , while high-frequency stimulations ( 40 and 100 Hz ) cause a majority of animals to awaken ( n = 4/7 and 4/6 , respectively ) . Dashed boxes indicate the most common response for each frequency , with arrows indicating the corresponding transition from sleep . ( B ) Low-frequency stimulation typically evokes spike-wave responses in EEG ( n = 5/7 ) , consistent with the behavioral reading of absence seizures . The most frequent EEG response during high-frequency stimulations is low voltage fast ( n = 3/7 and 6/6 ) , indicative of arousal . N , normal . SW , spike-wave . lvf , low voltage fast . s , spiking . e , evolving seizure . ( C , D ) Representative traces of EEG responses classified as spike-wave and low voltage fast . Insets show 4 s magnification . Importantly , these EEG patterns match those recorded under anesthetized conditions ( Figure 4G ) , further linking the responses visualized with ofMRI to the reported behavioral responses . See also Figure 6—figure supplement 1 . EEG: Electroencephalography . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 02510 . 7554/eLife . 09215 . 026Figure 6—figure supplement 1 . Pre-stimulus activity is consistent across frequencies of stimulation in asleep rats , as quantified with EEG bandpower in delta , theta , alpha , and beta bands . EEG: Electroencephalography . DOI: http://dx . doi . org/10 . 7554/eLife . 09215 . 026
Previously proposed mechanisms of arousal regulation have focused on the physiological and anatomical specialization of neurons within central thalamus ( Schiff , 2008 ) . Here , using a combination of ofMRI and electrophysiological recordings , we directly visualize whole-brain network activations produced by selective stimulation of central thalamus relay neurons and reveal novel insight into frequency-dependent gating of forebrain arousal . Since the earliest observations that electrical stimulation of central thalamus exerts frequency-dependent effects on behavior and EEG rhythms ( Moruzzi and Magoun , 1949; Hunter and Jasper , 1949 ) , behavioral arousal and cognition have been tightly linked with cortical activation ( i . e . low-amplitude , high-frequency oscillations ) , while behavioral arrest has been linked with cortical deactivation ( i . e . high-amplitude , low-frequency oscillations ) . More recently , it has also been shown that pharmacologically-induced changes in thalamic firing levels can switch cortical dynamics between activation and deactivation ( Hirata and Castro-Alamancos , 2010 ) . However , no studies have characterized the specific changes in activity that simultaneously occur across the whole intact brain during these events to explain how the same neuronal population can selectively switch arousal state . Here , we show that distinct firing patterns of excitatory neurons in the central thalamus drive these opposing EEG rhythms , lead to dramatic differences in the spatial extent of forebrain recruitment , and switch the region’s downstream influence on cortex from excitation to inhibition . Notably , high-frequency EEG patterns evoked during 40 and 100 Hz stimulation associate with robust activation of frontal cortex , motor cortex , somatosensory cortex , and striatum – regions that receive widespread glutamatergic projections from intralaminar nuclei ( Jones and Leavitt , 1974; Berendse and Groenewegen , 1990; Groenewegen and Berendse , 1994; Smith et al . , 2004; Hoover and Vertes , 2007; Hunnicutt et al . , 2014 ) . On the other hand , slow-wave oscillations evoked during 10 Hz stimulation are associated with limited forebrain activation and strong inhibition of somatosensory cortex . The frequency-dependent generation of spindle-like oscillations , which are known to underlie brain synchronization at the onset of sleep ( Steriade et al . , 1993 ) , suggest that these differences may in part be due to the engagement of thalamocortical networks responsible for sleep and loss of perceptual awareness during 10 Hz stimulation . In the context of our findings , it is also noteworthy that for some rat strains with absence seizures , a specific 10 Hz generator in somatosensory cortex has been independently proposed ( van Luijtelaar and Sitnikova , 2006 ) . The broadly contrasting cortical responses between low and high frequencies of central thalamus stimulation imply the frequency-dependent activation of a GABAergic population . In this study , we investigated the behavior of ZI – a region rich in GABAergic neurons that also sends direct inhibitory projections to sensory thalamus and sensory cortex ( Lin et al . , 1990; Benson et al . , 1992; Nicolelis et al . , 1995; Kolmac and Mitrofanis , 1999; Barthó et al . , 2002 ) . We found that spindle-like oscillations were uniquely evoked when central thalamus was stimulated at 10 Hz and somatosensory cortex was inhibited . Stimulation at this frequency in asleep rats also evoked absence seizure-like freezing and spike-waves in a majority of animals , a cortical pattern known to be modulated by ZI projections ( Shaw et al . , 2013 ) . Importantly , suppressing the incertal activity evoked by 10 Hz central thalamus stimulation with halorhodopsin reduced the cortical inhibition ( Figure 5H ) , suggesting a key role for ZI in modulating this response . Indeed , it has been previously suggested that rhythmic incertal activity contributes to membrane hyperpolarizations and sustained high-voltage cortical rhythms through GABAergic incertofugal pathways ( Shaw et al . , 2013 ) . Our data support this hypothesis , and link such a pathway to whole-brain , directly visualized fMRI activity patterns . Given the presence of GABAergic projections from ZI to central thalamus ( Barthó et al . , 2002 ) , activity in ZI may also act to limit forebrain activation , as observed with ofMRI during 10 Hz stimulation , through incertal-thalamic feedback . This incerto-thalamic pathway would parallel the previously reported gating of ascending sensory information at the level of thalamus by ZI ( Trageser and Keller , 2004; Lavallée et al . , 2005 ) . The hypothesized feedforward and feedback inhibition via ZI both suggest a direct projection from central thalamus to ZI , which our fluorescence imaging data support ( Figure 4H ) . However , we note that previous tracing studies failed to identify thalamic input specifically from intralaminar nuclei to ZI ( Shammah-Lagnado et al . , 1985 ) . In summary , our findings provide the first demonstration that arousal regulation driven by central thalamus has a causal and frequency-dependent influence on ZI , and that suppressing the recruitment of ZI modulates the brain-wide dynamics driven by central thalamus stimulation . Specifically , our results suggest that the frequency-dependent depression of cortical activity is in part mediated by extrinsic inhibitory signals originating from ZI . An additional mechanism that could contribute to the evoked suppression of cortex is feedforward thalamocortical inhibition – the process by which relay neurons drive inhibitory post-synaptic potentials ( IPSPs ) in pyramidal cells via fast spiking cortical interneurons ( Agmon and Connors , 1991; Porter et al . , 2001; Cruikshank et al . , 2007 ) . Low-frequency ( 10 Hz ) stimulations of certain thalamic nuclei in vivo yield strong hyperpolarization of cortical neurons putatively via this process ( Castro-Alamancos and Connors , 1996a; 1996c ) . It is also interesting to note that the ‘recruiting response’ , characterized by an enhanced cortical response during low frequency electrical stimulation of intralaminar nuclei ( Morison and Dempsey , 1942 ) , is hypothesized to originate from this inhibition ( Castro-Alamancos and Connors , 1997 ) . Our observation that individual stimuli sometimes trigger spikes in cortex ( Figure 3—figure supplement 1 ) is consistent with the possibility that these phenomena occur during the delivered 10 Hz optogenetic stimuli . However , intracellular and laminar recordings are needed to more conclusively resolve this issue . More recently , a study utilizing optogenetic targeting showed that non-specific ‘matrix’ thalamocortical neurons preferentially drive inhibitory interneurons in cortical layer I ( Cruikshank et al . , 2012 ) . Moreover , they found that IPSPs generated by stimulation of matrix neurons ( which constitute the nuclei targeted here [Jones , 2001] ) remain sustained during repeated stimuli compared to those evoked by stimulation of non-matrix neurons . Given the above findings , it is possible that interneuron-mediated thalamocortical inhibition , in addition to the demonstrated role of ZI , may contribute to the observed cortical responses . However , to the best of our knowledge , there have been no in vivo studies demonstrating that cortical interneuron-to-pyramidal cell inhibition is stronger at low frequencies of intralaminar thalamic stimulation compared to high frequencies . In addition to providing novel insight into the mechanisms of arousal regulation by central thalamus , our study offers important insight into the cellular origins of the fMRI BOLD signal . While there is a growing body of evidence suggesting that negative BOLD signals reflect local decreases in neuronal activity ( Shmuel et al . , 2006; Pasley et al . , 2007; Allen et al . , 2007; Devor et al . , 2007; Sotero and Trujillo-Barreto , 2007 ) , the nature of this signal remains a subject of debate and holds significant potential for the interpretation of functional imaging studies ( Schridde et al . , 2008; Ekstrom , 2010 ) . It has been shown that different sensory stimuli can evoke positive and negative BOLD signals in the same cortical area , which are linked to increases and decreases in neural activity , respectively ( Shmuel et al . , 2006 ) . Building upon these studies , we found that direct stimulation of central thalamus excitatory neurons at different frequencies leads to activation or suppression of neuronal activity in a downstream cortical location , which is coupled with positive and negative BOLD signals , respectively ( Figure 3 ) . These findings strongly support the hypothesis that a major component of the negative BOLD signal derives from decreases in neuronal activity and are consistent with previous reports of tight neural-hemodynamic coupling in the somatosensory cortex ( Huttunen et al . , 2008 ) . Our results are also consistent with a previous study by Logothetis et al . , which showed that low frequencies ( <50 Hz ) of electrical microstimulation in the thalamic lateral geniculate nucleus evoke negative BOLD responses in the monosynaptically connected V1 cortex , while higher frequencies ( 100–200 Hz ) evoke positive BOLD responses in the same region ( Logothetis et al . , 2010 ) . We observed similar results in the somatosensory cortex , which is monosynaptically connected to the stimulated intralaminar nuclei ( Van der Werf et al . , 2002 ) ( Figure 2—figure supplement 1 ) . The study by Logothetis et al . also found that cortical regions which are polysynaptically connected to the lateral geniculate nucleus , such as V2 , even exhibit negative BOLD responses at high frequencies of stimulation ( >60 Hz ) . It was proposed that these polysynaptic deactivations result from frequency-dependent disynaptic inhibition , the process by which pyramidal cells in cortex inhibit local and remote pyramidal cells via GABAergic interneurons . Unlike the study by Logothetis et al . , we did not observe significant negative BOLD signals in either mono- or polysynaptically connected regions of cortex during high frequencies of stimulation . Furthermore , given the bias of corticocortical disynaptic inhibition toward higher frequencies ( Silberberg and Markram , 2007 ) , this microcircuit is unlikely to be driving the observed suppression of cortex during 10 Hz central thalamus stimulation . In the context of electrical stimulation , our study helps dissociate the confounding effects of ( a ) delivering stimulation at a certain frequency ( which can preferentially recruit certain neuronal elements ( McIntyre and Grill , 2002 ) ) and ( b ) the excited neuronal population firing at a specific frequency . With electrical stimulation , it has been impossible to dissociate these two effects in vivo , since the frequency of stimulation and preferential recruitment of specific neuronal populations could not be decoupled ( McIntyre and Grill , 2002 ) . This made it difficult to explain , for example , the relationship between stimulation parameters and the therapeutic efficacy of DBS . Using targeted , temporally precise , optogenetic stimulation in the current study allowed us to selectively excite a single group of neuronal elements and identify their specific role in creating distinct modes of network function . The use of electrical stimulation instead would have prevented us from gaining this unique insight into the specific role of excitatory central thalamus neurons and their spiking frequency . Finally , in the context of central thalamus DBS , our study offers important insight into the identification of proper stimulation targets and parameters that are needed before the therapeutic application of central thalamus stimulation can reach its full clinical potential . In particular , the images from ofMRI experiments ( Figure 2 and Figure 2—figure supplement 3 ) reveal dramatic differences in global brain dynamics that can result from controlling one parameter of stimulation ( i . e . frequency ) . Furthermore , the widespread activation of cortex and striatum observed at high frequencies of stimulation adds to a growing body of evidence that the central thalamus is a highly appropriate target for the remediation of acquired cognitive disabilities via forebrain recruitment . In a more general context that extends beyond stimulation of central thalamus , the ofMRI techniques we employ here are generalizable and can be universally applied to study the mechanisms underlying DBS for other target regions and disorders . With this knowledge , stimulation paradigms can be optimized to accelerate clinical translation for a wide range of neurological disorders that currently lack such treatment , paving the way for the development of next-generation DBS therapies .
Female Sprague-Dawley rats ( >11 weeks old , 250-350 g ) were used as subjects for all thalamic injections . Animals were individually housed under a 12 hr light–dark cycle and provided with food and water ad libitum . Animal husbandry and experimental manipulation were in strict accordance with National Institute of Health , UCLA Institutional Animal Care and Use Committee ( IACUC ) , and Stanford University IACUC guidelines . pAAV5-CaMKIIa-hChR2 ( H134R ) -EYFP-WPRE plasmid was obtained from the Deisseroth lab at Stanford University . Concentrated virus was produced at the vector core of the University of North Carolina at Chapel Hill . Rats were anesthetized with isoflurane ( induction 5% , maintenance 2–3%; Sigma-Aldrich , St . Louis , MO ) and secured in a stereotactic frame . Standard procedures for sterile surgery were followed . Buprenorphine was administered to minimize pain . Artificial tears were applied to the eyes . The head was shaved , and 70% ethanol and betadine were applied to the bare scalp following a midline incision . A small craniotomy was performed with a dental drill above the targeted coordinate . Two microliters of virus were injected through a 34-gauge needle ( World Precision Instruments Inc . , Sarasota , FL ) at 150 nl/min with a micro-syringe pump controller at the desired coordinates in central thalamus or other subcortical targets for control experiments: I ) CL and PC nuclei of central thalamus ( -3 . 2 mm AP , +1 . 5 mm ML , -5 . 6 mm DV; n = 47 animals for imaging ) ; II ) ventral posteromedial nucleus ( -2 . 5 mm AP , +2 . 6 mm ML , -6 . 0 mm DV ) ; III ) anterior thalamic nuclei ( -3 . 1 mm AP , +1 . 8 mm ML , -5 . 3 mm DV ) ; IV ) posterior thalamic nuclei ( -4 . 6 mm AP , +1 . 8 mm ML , -5 . 0 mm DV ) ; V ) intermediate hippocampus ( -5 . 8 mm AP , +5 . 2 mm ML , -3 . 4 mm DV , n = 8 animals ) . All injections were made in the right hemisphere . The syringe needle was left in place for an additional 10 min before being slowly withdrawn . Custom-designed guide cannulas ( Plastics One ) or fiber-optic cannulas ( Doric Lenses Inc . ) were mounted on the skull and secured using metabond ( Parkell ) . Incisions were sutured , and animals were kept on a heating pad until recovery from anesthesia . Buprenorphine was injected subcutaneously twice daily for 48 hr post-operatively to minimize discomfort . The original cohort of 47 central thalamus animals was further refined to a group of 18 after screening for implant locations less than 0 . 85 mm away from the target coordinate ( estimated with T2 MRI scans; Figure 1B ) . Two additional animals were excluded due to lack of thalamic activation , leaving 16 animals for analysis . In a second cohort of rats , concentrated AAV5-hSyn-eNpHR3 . 0-mCherry virus produced at the University of North Carolina at Chapel Hill vector core was injected into the right ZI ( -3 . 96 mm AP , +2 . 8 mm ML , +7 . 4 mm DV , n = 4 animals ) after completion of the ChR2 injection into the central thalamus as described above . Both injections were performed during the same surgery . 0 . 5 μl of eNpHR virus were injected through a 34-gauge needle at 100 nl/min . Following the injection , the syringe needle was left in place for approximately 10 min before being slowly withdrawn . Recovery details were the same as those described above . fMRI scanning was performed using a 7T Bruker Biospec small animal MRI system at UCLA . Animals were initially anesthetized with 5% isoflurane and intubated before placement onto custom-made MRI-compatible cradles with ear and tooth securement . A 39 mm outer diameter , 25 mm inner diameter custom-designed transmit/receive single-loop surface coil was centered over the region of interest on the skull to maximize signal-to-noise ratio . An optical fiber of 62 . 5 μm core diameter was connected to a 473 nm laser source ( Laserglow Technologies , Toronto , Canada ) and coupled with the implanted fiber-optic cannula . A single ofMRI scan consisted of a block design with six 20 s pulse trains of light ( 10 , 40 , or 100 Hz in randomized order ) delivered once per minute over 6 min . Five to six consecutive scans were collected during each session . For all experiments , the optical fiber output power was calibrated to 2 . 5 mW . A duty cycle of 30% was used across frequencies to maintain the total amount of light delivery , resulting in unique pulse widths of 30 , 7 . 5 , and 3 ms for 10 , 40 , and 100 Hz , respectively . In a series of control experiments using a second cohort of animals with validated probe locations ( n = 3 ) , the duty cycle was varied while the pulse width was held constant at 3 ms ( Figure 2—figure supplement 3 ) . During fMRI scanning , animals were placed into the iso-center of the magnet while artificially ventilated ( 45~60 strokes/min ) under light anesthesia using a ventilator and calibrated vaporizer with a mixture of O2 ( 35% ) , N2O ( 63 . 5% ) , and isoflurane ( 1 . 3–1 . 5% ) . To ensure stable BOLD signals , expiratory CO2 was kept at 3–4% and body temperature was maintained at 36 . 5–37 . 5°C using heated airflow . T2-weighted high-resolution anatomical images were acquired with a fast spin echo sequence prior to fMRI scanning to check for brain damage and validate the optical fiber’s location ( 137 µm resolution in-plane resolution with 35×35 mm2 FOV , 0 . 5 mm slice thickness , 32 coronal slices ) . Gradient recalled echo ( GRE ) BOLD methods were used to acquire fMRI images during photostimulation . The fMRI image acquisition was designed to have 35×35 mm2 in-plane field of view ( FOV ) and 0 . 5×0 . 5×0 . 5 mm3 spatial resolution with a sliding window reconstruction to update the image every repetition time ( TR ) ( Fang and Lee , 2013 ) . The two-dimensional , multi-slice gradient-echo sequence used a four-interleave spiral readout ( Glover and Lee , 1995; Kim et al . , 2003 ) , 30° flip angle , 750 ms TR , and 12 ms echo time , resulting in 23 coronal slices ( 128 × 128 matrix size ) . The spiral k-space samples were reconstructed through a two-dimensional gridding reconstruction method ( Jackson et al . , 1991 ) . Finally , real-time motion correction was performed using a previously described GPU-based system ( Fang and Lee , 2013 ) . Scans with significant motion , identified by careful visual inspection for spiral artifacts and activations at the boundary of the brain , which indicates large motion , were excluded from analysis . This condition for exclusion was established prior to data collection . All fMRI data processing was performed using the Matlab software environment ( MathWorks , Inc . , Natick , MA ) and mrVista ( Stanford Vision and Imaging Science and Technology Laboratory , Stanford , CA; http://web . stanford . edu/group/vista/cgi-bin/wiki/index . php/MrVista ) . Motion-corrected images belonging to consecutive scans of the same stimulation paradigm and scanning session were first averaged together . The average 4D images were then aligned to a common coordinate frame , using a six degree-of-freedom rigid body transformation . If multiple scanning sessions were performed on the same animal at the same frequency ( typically 1 , at most 4 ) , the resulting images from each session were first averaged together before any inter-subject analysis to achieve maximum signal-to-noise ratio while weighting the images from all animal subjects equally . Time series were calculated for each voxel in these individual-animal images as the percent modulation of the BOLD signal relative to a 30 s baseline period collected prior to stimulation . Boxcar detrending with a window size of 1 min was also performed to correct for possible scanner drift . Next , a coherence value was calculated for each voxel’s time series as the magnitude of its Fourier transform at the frequency of repeated stimulation blocks ( i . e . 1/60 Hz ) divided by the sum-of-squares of all frequency components ( Engel et al . , 1997 ) . Voxels with a coherence value greater than 0 . 35 were considered to be significantly synchronized to stimulation . Assuming Gaussian noise and ~470 degrees of freedom ( computed using the SPM software environment ) , the Bonferroni-corrected p value for this threshold can be estimated to be less than 10–9 ( Bandettini et al . , 1993 ) . Activation volume ( Figure 2 ) was defined as the number of significant voxels that exhibited a positive response within a predefined region of interest , multiplied by the volume per voxel . Positive responses were identified as those having a phase in the interval [0 , π/2] ∪ [3π/2 , 2π] . Phase represents the temporal shift of the response when it is modeled as a sinusoid and was calculated as the argument of each voxel’s Fourier transform at the frequency of repeated stimulation blocks ( i . e . 1/60 Hz ) . Hemodynamic response functions ( HRFs ) were calculated as the average 60 s response of a voxel’s six-cycle , 6-min time series . Time series and HRFs displayed for figures were generated by averaging the mean time series or mean HRF of voxels with a coherence value greater than 0 . 35 in the corresponding ROI across animals . For Videos 1–3 , the first data point’s value was subtracted from each voxel’s HRF to define its relative percent modulation from the onset of stimulation . To generate average activation maps ( Figure 2 ) , the 4D fMRI images from experiments at the same stimulation location and frequency were normalized and averaged together across animals . The averaged images were then processed according to the above Fourier domain analyses . Coherence values were overlaid onto all voxels having a coherence above the 0 . 35 threshold . Warm and cool colormaps generated using Matlab’s ‘hot’ and ‘winter’ functions were used for positive and negative responses , respectively , to illustrate the localization of negative BOLD to the somatosensory cortex . These activation maps were overlaid onto corresponding T2-weighted anatomical images with a digital standard rat brain atlas ( Paxinos and Watson , 2005 ) . The same atlas was used to segment ROIs . An identical analysis pipeline was used for activation maps in Figure 2—figure supplement 3 with a representative animal . EEG electrodes were implanted upon completion of ofMRI experiments in a subset of animals . Surgical preparation and recovery details were the same as those used for virus injection . Stainless steel screws ( 0–80 , 1 . 5 mm diameter , Plastics One ) attached to 2 cm of insulated wire ( 30 gauge , R30Y0100 , Wire Wrapping Wire , O . K . Industries ) were used as EEG electrodes and secured on the skull using dental cement . The recording electrode was placed approximately 2 mm anterior of bregma and 2 mm to the right of midline . The reference electrode was located approximately 5 mm anterior of bregma and 3 mm to the left of midline ( Horner et al . , 2003 ) . Prior to video-EEG recording , animals were anesthetized under 5% isoflurane for approximately 5 min for optical fiber coupling and EEG wire connection . Animals were then transferred to a light- and sound-controlled experimental chamber where they were allowed to move freely . Behavioral experiments began after animals recovered from anesthesia and subsequently fell asleep for 15 min ( as indicated by lack of motion and real-time EEG output readings ) . For each experiment , the animal was video-recorded during 5 min of sleep , followed by 20 s of optical stimulation ( 10 , 40 , or 100 Hz , 473 nm laser , 2 . 5 mW laser power , 30% duty cycle ) , and then an additional 5 min post-stimulation period . EEG data was acquired throughout the experiment at 1 kHz with an MP150 data acquisition unit and EEG100C amplifier ( Biopac Systems Inc . , Santa Barbara , CA ) , using EL254S Ag-AgCl electrodes and Gel102 conductive EEG paste . A digital camera was used to video-record the experiment . All behavioral experiments were performed during the animals’ light cycle . EEG recordings were classified using the Biopac Acqknowledge software by an experienced electroencephalographer blind to treatment into a single best category: normal , low voltage fast , spikes , spike-waves , or evolving electrographic seizure . Video clips paired to each EEG recording were classified into one of the following categories to further assess the animal’s brain state: sleep ( i . e . no change ) , awakening ( animal is alert and exploring ) , absence seizure ( animal is immobile and appears frozen for the duration of stimulation , but returns to a sleeping state once stimulation ends ) , or convulsive seizure . All observed behavioral responses could be classified into one of these categories . Band power in Figure 6—figure supplement 1 was quantified using Matlab’s ‘bandpower’ function and normalized by the signal’s total power from 0 Hz to one half the sampling rate ( 500 Hz ) . Upon completion of ofMRI and EEG behavioral experiments , in vivo electrophysiology experiments were performed in a subset of animals . Animals were anesthetized with 5% isoflurane for induction and maintained at 2–3% until any craniotomies were complete . Isoflurane was kept at 0 . 8–1 . 2% during the recording session , and artificial tears were applied to the eyes . Recordings in Figures 4 and 5 were performed under ventilation conditions identical to fMRI experiments . After securing the animal within a stereotactic frame , small craniotomies were performed using a dental drill above the region of interest . For stimulation , the cannula implanted at central thalamus was connected to a 473 nm laser source ( Laserglow Technologies ) with an output power level of 2 . 5 mW via an optical fiber . The cannula implanted at ZI was connected to a 593 nm laser source ( Laserglow Technologies ) calibrated to 2 . 5–3 . 0 mW . An acute 16-channel microelectrode array was targeted to the recording site using stereotactic instruments ( NeuroNexus Technologies; A1x16 standard model linear electrode array for local and cortical recordings; V1-16-Poly2 polytrode array for ZI recordings; 0 . 35 ± 0 . 5 MOhm impedance ) . A stainless steel reference screw was placed above the cerebellum . Continuous field potential and single unit spiking events were recorded using the Plexon omniplex system with plexcontrol software ( Plexon Inc . , TX ) . When only ChR2 was activated , recordings were performed for 20 s without stimulation , followed by repeated stimulation cycles ( 20 s on , 40 s off ) at 10 , 40 , or 100 Hz with 30% duty cycle . When ChR2 and eNpHR were activated together , the same stimulation paradigm was followed , except that a 30 s period of continuous 593 nm light delivery began 5 s before the onset of ChR2 excitation . When only eNpHR was activated ( Figure 5D ) , a 20 or 30 s period of continuous 593 nm light delivery was used , with 40 or 30 s periods of no light delivery between repeated cycles , respectively . For single unit responses , the Plexon multichannel acquisition processor was used to amplify and band-pass filter the neuronal signals ( 150 Hz – 8 kHz ) . Signals were digitized at 40 kHz and processed to extract action potentials in real-time . To separate the field potential , we used a low-pass filter ( 200 Hz cutoff frequency , 4-pole Bessel filter ) and downsampled signals to 1 kHz . Simultaneous EEG data was collected at 1 kHz during ZI recordings in Figure 4 using the MP150 data acquisition unit and EEG100C amplifier ( Biopac ) . For the analysis in Figure 4 , field potential recordings were high pass filtered with a cutoff frequency of 2 Hz to eliminate respiratory artifacts . Spindle-like oscillations ( SLOs ) occurring during the stimulus were then quantified on a per trial basis using a post-hoc custom algorithm ( see Source code 1 ) . Briefly , an SLO was identified when the recording’s magnitude reached at least 6 standard deviations above its mean absolute value . If the recording did not exceed this value for the preceding 500 ms , and was above this value for at least 2% of samples over the next 500 ms , an SLO was counted . This method of quantification accurately captured the large-amplitude oscillations that could be visually discerned ( see Figure 4D ) . Upon completion of in vivo of MRI , behavioral , and electrophysiology experiments , rats were deeply anesthetized with isoflurane in a knockdown box and transcardially perfused with 0 . 1M phosphate-buffered saline ( PBS ) and ice-cold 4% paraformaldehyde ( PFA ) in PBS . Brains were extracted and fixed in 4% PFA overnight at 4°C . The brains were equilibrated in 10% , 20% , and then 30% sucrose in PBS at 4°C . Coronal sections ( 50 μm ) were prepared on a freezing microtome ( HM 430 Sliding Microtome , Thermo Scientific Inc . ) . Consecutive sections ( 500 µm apart ) were mounted and examined with a fluorescence microscope ( Leica EL6000 ) . For quantitative immunohistochemistry ( Figure 1—figure supplement 1 ) , free-floating sections were processed with 5% normal donkey serum , and 0 . 4% Triton X-100 for 60 min . Sections were then exposed at 4°C for 48 hr to primary antibodies against mouse monoclonal CaMKIIα ( CaMKIIα , 1:500 , 05–532 , Millipore , Billerica , MA ) . After washing with PBS , sections incubated for 2 hr at room temperature with Alexa Fluor 647-conjugated AffiniPure donkey anti-mouse IgG ( 1:250 , Jackson Laboratories , West Grove , PA ) . Slices were then washed and mounted ( DAPI-Fluoromount G , SouthernBiotech , Birmingham , AL ) . Immuno-fluorescence was assessed with a laser confocal microscope ( Leica CTR 6500 ) . For high-resolution , whole-brain fluorescence imaging ( Figures 1A , 4H , and Figure 2—figure supplement 1 ) , frozen brains were embedded using stainless steel Tissue-Tek base molds and Neg-50 embedding medium ( Richard-Allan Scientific [Thermo]; n = 2 animals ) ( Pinskiy et al . , 2013 ) . Post-freezing , the Neg-50 embedded brain was sectioned on a Microm HM550 cryostat using the tape-transfer method with all sections mounted directly onto slides . Alternating sections , cut at 20 μm , were separated to form two distinct series per brain . One slide series of the sectioned material was processed for Nissl cell body staining , using a thionin-based protocol and coverslipped with DPX mounting medium . The alternate series was dehydrated and directly coverslipped with DPX for fluorescence imaging . Whole-slide digital imaging was performed using a Hamamatsu NanoZoomer 2 . 0-HT system at 0 . 46 μm/pixel , with fluorescence scans at 12-bit depth using a tri-pass filter cube . Following data conversion to lossless jp2 ( JPEG2000 ) , individual brain sections were aligned and registered using rigid 2-D image transformation . All statistical tests were performed in Matlab . Non-parametric tests were used throughout the analysis . For in vivo electrophysiology measurements at thalamus and ZI , one-tailed Wilcoxon signed-rank tests were used to evaluate changes in firing rate between the pre-stimulation and stimulation periods . For measurements at sensory cortex in Figure 3 , a two-tailed version of the test was used to evaluate either increases or decreases in firing rate . For results in Table 1 , the average pre-stimulus firing rate ( 20 s bin ) was compared to the average firing rate of four 5 s bins over the 20 s period of stimulation using a one-tailed Wilcoxon signed rank test , uncorrected for multiple comparisons . One-sided Wilcoxon rank sum tests were used to evaluate differences in SLO occurrence ( Figure 4E ) , as well as changes in cortical or incertal firing when eNpHR activation was coupled with central thalamus stimulation ( Figure 5F , H ) . For electrophysiology results , independence was assumed between repeated trials . All other assumptions for these tests were satisfied . For volumetric comparisons in Figure 2 , one-sided Wilcoxon signed-rank tests were used to identify increases in the volume of BOLD activation between 10 and 40 Hz and 10 and 100 Hz ( corrected for multiple comparisons ) . Note that variance was generally similar across groups being compared . Significance was determined at the α = 0 . 05 cutoff level . No statistical methods were used to estimate sample size . All statistical tests used to compare changes with frequency ( Figure 2 and Figure 2—figure supplement 2B ) were performed pairwise , with an equal number of animals used for each frequency . | The ability to wake up every morning and to fall asleep at night is something that most people take for granted . However , damage to a brain region called the central thalamus can cause a range of consciousness-related disorders , including memory problems , excessive sleeping , and even comas . For example , cell death within the central thalamus has been associated with severely disabled patients following traumatic brain injury . Previous studies have found that electrically stimulating the neurons in the central thalamus can change whether an animal is drowsy or awake and alert . However , it was not clear whether a single group of neurons in the central thalamus was responsible for switching the brain’s state between sleep and wakefulness , or how this would work . Liu , Lee , Weitz , Fang et al . have now used a technique called optogenetics to stimulate specific neurons in the central thalamus of rats , by using flashes of light . Stimulation was combined with several techniques to monitor the response of other brain regions , including fMRI imaging that shows the activity of the entire brain . The results showed that rapidly stimulating the neurons in the central thalamus – 40 or 100 times a second – led to widespread brain activity and caused sleeping rats to wake up . In contrast , stimulating the neurons of the central thalamus more slowly – around 10 times a second – suppressed the activity of part of the brain called the sensory cortex and caused rats to enter a seizure-like state of unconsciousness . Further investigation identified a group of inhibitory neurons that the central thalamus interacts with to carry out this suppression . The results suggest that the central thalamus can either power the brain to an “awake” state or promote a state of unconsciousness , depending on how rapidly its neurons are stimulated . Future work will seek to translate these results to the clinic and investigate how stimulation of the central thalamus can be optimized to reduce cognitive deficits in animal models of traumatic brain injury . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Frequency-selective control of cortical and subcortical networks by central thalamus |
Squamous cell carcinomas occurring at transition zones are highly malignant tumors with poor prognosis . The identity of the cell population and the signaling pathways involved in the progression of transition zone squamous cell carcinoma are poorly understood , hence representing limited options for targeted therapies . Here , we identify a highly tumorigenic cancer stem cell population in a mouse model of transitional epithelial carcinoma and uncover a novel mechanism by which loss of TGFβ receptor II ( Tgfbr2 ) mediates invasion and metastasis through de-repression of ELMO1 , a RAC-activating guanine exchange factor , specifically in cancer stem cells of transition zone tumors . We identify ELMO1 as a novel target of TGFβ signaling and show that restoration of Tgfbr2 results in a complete block of ELMO1 in vivo . Knocking down Elmo1 impairs metastasis of carcinoma cells to the lung , thereby providing insights into the mechanisms of progression of Tgfbr2-deficient invasive transition zone squamous cell carcinoma .
Transition zones are distinct anatomical regions between two different types of epithelia , and function as stem cell niches in many regions of the body ( Runck et al . , 2010; San Roman et al . , 2011; Mcnairn and Guasch , 2011 ) . Populations of cells with stem cell characteristics , such as label-retention and self-renewal , and expression of stem cell markers such as the cell surface glycoprotein CD34 , have been found in the transition zones between the cornea and conjunctiva in the eye ( Zieske , 1994; Cotsarelis et al . , 1989 ) , between the esophagus and the stomach ( Kalabis et al . , 2008 ) , between the endocervix and ectocervix ( Elson et al . , 2000; Herfs et al . , 2012 ) , between the mesothelium and oviductal epithelium of the ovary ( Flesken-Nikitin et al . , 2013 ) and between the anal canal and the rectum ( Runck et al . , 2010 ) . Transition zones are uniquely susceptible to carcinogenesis , and the resulting tumors are typically highly malignant and associated with poor prognosis ( San Roman et al . , 2011; Mcnairn and Guasch , 2011; Flesken-Nikitin et al . , 2013; Grodsky , 1961 ) . Ocular surface squamous neoplasias are relatively rare , but most of them involve the limbus ( McKelvie et al . , 2002 ) . As many as 86% of esophageal tumors arise in association with Barrett’s esophagus at the esophageal-gastric junction ( Trudgill et al . , 2003; Yamamoto et al . , 2016 ) . The transition zone in the ovary was recently shown to be acutely sensitive to oncogenic transformation ( Flesken-Nikitin et al . , 2013 ) . Cervical cancers arise at the transition between the columnar epithelium of the endocervix and the squamous epithelium of the ectocervix ( Elson et al . , 2000; Herfs et al . , 2012; Petignat and Roy , 2007 ) . Highly malignant squamous cell carcinomas ( SCC ) develop between the stratified epithelium of the anal canal and the simple epithelia of the rectum ( Grodsky , 1961; Kim et al . , 2013; Guasch et al . , 2007 ) . To date , we still lack a clear understanding of the signaling and cellular mechanisms that drive transitional epithelial carcinogenesis . Deregulated TGFβ signaling seems to be a hallmark of aggressive transition zone cancers . In cervical cancer , mutations or loss of TGFβ downstream effectors SMAD2 and SMAD4 are common ( Maliekal et al . , 2003 ) , and loss of nuclear SMAD2 and SMAD4 is associated with poor survival ( Kloth et al . , 2008 ) . In genital SCC , TGFβ receptor ( TGFβRII ) protein expression is decreased and loss of phosphorylated SMAD2 is observed , even at early stages , suggesting that loss of TGFβ signaling may be an early event in carcinogenesis ( Guasch et al . , 2007 ) . Mouse models targeting components of the TGFβ signaling pathway have been generated ( Muñoz et al . , 2006 ) . Many epithelia develop normally despite the loss of a component of the TGFβ signaling pathway ( Guasch et al . , 2007; Muñoz et al . , 2006; Biswas et al . , 2004; Padua and Massagué , 2009 ) . However , tumorigenesis occurs rapidly when these epithelia are exposed to carcinogens ( Biswas et al . , 2004 ) , polyomavirus middle T antigen expression ( Forrester et al . , 2005 ) , oncogenic mutations , such as mutations in APC ( Muñoz et al . , 2006 ) , or activated Hras ( Guasch et al . , 2007; Schober and Fuchs , 2011; Lu et al . , 2006 ) or Kras ( Lu et al . , 2006 ) , or spontaneously in transition zones . Within the gastric transition zone , loss of SMAD3 ( Nam et al . , 2012 ) or BMP signaling ( Bleuming et al . , 2007 ) results in invasive carcinoma . Mice with a neuronal-specific deletion of Tgfbr1 develop spontaneous periorbital and perianal SCC ( Honjo et al . , 2007 ) . The backskin of mice lacking Tgfbr2 in all Keratin 14-expressing ( K14+ ) progenitors of the stratified epithelia is morphologically normal , but these mice develop spontaneous SCC in cervical and anorectal transition zones ( Guasch et al . , 2007 ) . RHO and RAC-guanine triphosphatases ( GTPases ) are small G proteins ( 21–25 kDa ) , and belong to the RAS superfamily ( Parri et al . , 2010 ) . They act as molecular switches to elicit rapid changes in cell shape , polarity , and migratory ability in response to external cues ( Parri et al . , 2010; Vega and Ridley , 2008; Sadok et al . , 2014; Alan and Lundquist , 2013 ) and are major players in malignant cell invasion . RAC exists in an inactive form , bound to GDP , and in an active form , bound to GTP ( Parri et al . , 2010; Sadok et al . , 2014; Laurin and Cote , 2014; Lazer and Katzav , 2011 ) . Guanine exchange factors ( GEFs ) are required to promote the active , GTP-bound form of RAC , and GTPase activating proteins ( GAPs ) return RAC to its inactive , GDP-bound state ( Parri et al . , 2010; Vega and Ridley , 2008; Sadok et al . , 2014; Laurin and Cote , 2014 ) . More than 70 GEFs have been described , which act downstream of many signaling pathways , including growth factor receptors , integrins , cadherins , and cytokine receptors ( Parri et al . , 2010 ) . Engulfment and cell motility ( ELMO ) proteins ( originally described as CED-12 in C . elegans ) participate in RAC1-dependent engulfment and apoptosis ( Côté and Vuori , 2007; Gumienny et al . , 2001 ) . ELMO proteins form a complex with DOCK proteins that serves as a GEF for RAC proteins . This complex plays important roles in chemotaxis , phagocytosis , neurite outgrowth , and cancer cell invasion ( Laurin and Cote , 2014; Côté and Vuori , 2007; Gumienny et al . , 2001; Grimsley et al . , 2004; Brugnera et al . , 2002; Jarzynka et al . , 2007; Sai et al . , 2008; Li et al . , 1706; Komander et al . , 2008 ) . Subsets of long-lived tumor-initiating stem cells or cancer stem cells ( CSCs ) are often resistant to cancer therapies and thus may be responsible for tumor recurrence ( Clevers , 2011; Malanchi et al . , 2012 ) . They sustain tumor growth through their ability to self-renew and to generate differentiated progeny , and they may play a role in metastasis ( Clevers , 2011; Malanchi et al . , 2012; Oskarsson et al . , 2014; Chaffer and Weinberg , 2011; Charafe-Jauffret et al . , 2010 ) . To date , the cellular and molecular mechanisms of Tgfbr2-deficient transition zone carcinoma development and metastasis are unknown . In this study , we used the murine Tgfbr2-deficient anorectal SCC model to study the consequences of loss of TGFβ signaling in CSC-driven tumor propagation and metastasis . We found that these Tgfbr2 cKO anorectal SCC , which spontaneously metastasize to the lungs , contain a unique population of epithelial cells with features of CSCs , including: expression of the CSC marker CD34 , clonogenicity in vitro , tumorigenicity in vivo , and upregulation of genes associated with invasion and metastasis . Using RNA-Sequencing and chromatin immunoprecipitation , we uncovered a novel mechanism linking loss of TGFβ signaling with invasion and metastasis via the RAC-activating GEF ELMO1 . We show that Elmo1 is a novel target of TGFβ signaling via SMAD3 and that restoration of Tgfbr2 results in complete block of ELMO1 in vivo . Knocking down Elmo1 impairs metastasis to the lung , providing a new therapeutic avenue to target the early phase of metastasis in highly aggressive transition zone tumorigenesis .
Mice lacking Tgfbr2 in stratified epithelia expressing Keratin 14 ( K14 ) develop spontaneous squamous cell carcinoma ( SCC ) at the transition zone between the anal canal and rectum ( Guasch et al . , 2007 ) . To lineage trace Tgfbr2-deficient cells within these transitional SCC and enable isolation of specific Tgfbr2-deficient tumor cell populations , we backcrossed these mice into mice containing loxP sites flanking a STOP sequence preceding eYFP inserted into the Rosa26 locus ( Figure 1—figure supplement 1 ) , such that all K14-positive epithelial cells , including the anorectal SCC cells , while conditionally null for Tgfbr2 expressed YFP ( cKO mice , Figure 1A–C ) . We had previously identified a population of cells with stem cell characteristics , including colocalization with known stem cell markers , such as CD34 , in the anorectal transition zone of wild-type mice ( Runck et al . , 2010 ) . We hypothesized that tumors arising at the anorectal transition zone in the Tgfbr2 cKO mice would contain a population of CD34-expressing cells , and that these cells would represent a population of tumor-propagating cells or so-called cancer stem cells ( CSCs ) . Based on the idea that CSCs should reside at the tumor–stroma border , we thought that CSCs of anorectal SCCs should express abundant integrins . To test this hypothesis , we first analyzed marker expression within histologic sections of Tgfbr2-deficient anorectal tumors . Immunofluorescence staining revealed that all YFP+SCC cells located at the tumor–stroma interface expressed high levels of the hemidesmosomal α6 integrin ( Figure 1D ) and the focal adhesion marker β1 integrin ( Figure 1E ) , and a fraction of these expressed CD34 ( Figure 1F ) . These three markers were used to isolate discrete populations of cells by fluorescence-activated cell sorting ( FACS ) . Tumors were dissociated into a single-cell suspension as previously described ( McCauley and Guasch , 2013 ) , stained with antibodies , and subjected to flow cytometry . Blood cells , endothelial cells , macrophages and dead cells were excluded , and live , K14+YFP+ epithelial cells were further purified upon α6-integrin and β1-integrin expression . Of these live , YFP+ , α6-integrin+ , β1-integrin+ cells , distinct CD34-negative ( CD34− ) and CD34-positive ( CD34+ ) populations of cells were observed and isolated ( Figure 1G ) . The frequency of epithelial CD34+ cells within the tumor varied between mice , from 7% to 34% . When plated on a feeder layer of irradiated fibroblasts , both YFP+CD34−and YFP+CD34+ cells sorted from Tgfbr2 cKO SCC formed colonies; however , CD34− colonies appeared to be differentiated paraclones and were unable to be passed more than once , whereas CD34+ colonies appeared to form holoclones , were able to proliferate extensively , and survived unlimited passage ( Figure 1H–I ) . CD34+ SCC cells did not respond to TGFβ stimulation , confirming the loss of TGFβRII , and aberrantly expressed Keratin 8 , a hallmark of SCC ( Ikeda et al . , 2008 ) , compared to keratinocytes isolated from the anal canal of wild-type mice ( Figure 1J ) . These data suggest that the CD34+ population of epithelial cells isolated from the Tgfbr2 cKO anorectal SCC has self-renewal properties in vitro . 10 . 7554/eLife . 22914 . 003Figure 1 . Tgfbr2 cKO anorectal SCC contain a population of epithelial CD34+ cells which are enriched for in vitro clonogenicity . ( A ) Triple transgenic mice were obtained by crossing Tgfbr2flox/flox mice with R26R-eYFPflox-STOP-flox mice and K14-Cre mice . ( See Figure 1—figure supplement 1 ) ( B , B’ , B’’ ) . Tgfbr2 cKO mice developed spontaneous anorectal tumors ( B ) which formed in the anal canal between the hair-bearing perianal skin and rectum ( B’ ) , and expressed YFP ( B’’ ) . ( C ) Hematoxylin and eosin ( H and E ) staining of Tgfbr2 cKO tumors revealed moderately- to poorly-differentiated squamous cell carcinoma ( SCC ) at the transition between the epithelium of the anal canal and the rectum . ( D–F ) Immunofluorescence staining of Tgfbr2 cKO anorectal SCC revealed YFP+ tumor cells expressed both α6-integrin ( D ) and β1-integrin ( E ) and that there was a distinct population of CD34+ tumor cells ( F ) . Boxed areas represent isolation and magnification of the red channel . DAPI counterstains nuclei in blue . Representative of 16 mice analyzed by histology and immunostaining and 37 analyzed by FACS ( G ) . CD34+ and CD34− cells were isolated from the Tgfbr2 cKO anorectal SCC by fluorescence-activated cell sorting ( FACS ) . After dissociation and staining , CD45+ blood cells , CD31+ endothelial cells and CD11b+ macrophages were excluded from the live ( 7AAD- ) , K14+YFP+ population . These epithelial tumor cells were further purified by gating for α6-integrin ( CD49f+ ) and β1-integrin ( CD29+ ) cells ( YFP +7AAD-CD11b-CD31-CD45-CD49f+CD29+ , abbreviated: YFP+ ) . Of these epithelial tumor cells , distinct CD34+ and CD34− populations were isolated ( see Figure 2—figure supplement 1 ) . ( H–I ) When plated on a feeder layer of irradiated fibroblasts , sorted YFP+CD34− and YFP+CD34+ tumor cells were able to form clones . CD34− clones appeared differentiated and were unable to survive multiple passages , whereas CD34+ clones formed holoclones and subsequently robust cell lines which survived unlimited passage ( n = 4 tumor-bearing mice ) . Panel H is a representative example of three different primary clones isolated from one of four distinct spontaneous tumors . ( J ) Tgfbr2 cKO anorectal CD34+ SCC cells are nonresponsive to TGFβ as they did not phosphorylate SMAD2 nor downregulate c-Myc compared to WT anal keratinocytes , and aberrantly expressed Keratin 8 . Abbreviations: bv , blood vessel; F , irradiated fibroblast; P0 , passage 0; P1 , passage 1; k , keratinocyte . Scale bars = 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 00310 . 7554/eLife . 22914 . 004Figure 1—figure supplement 1 . Conditional targeting of Tgfbr2 and lineage tracing in Keratin 14-positive tissues . Exon 4 of the Tgfbr2 locus , encoding the transmembrane domain ( TM ) and the intracellular phosphorylation sites ( P ) of the TGFβRII protein is flanked by loxP sites and has been backcrossed into a mouse reporter containing an Enhanced Yellow Fluorescent Protein gene ( eYFP ) inserted into the ROSA locus . Tgfbr2 is deleted and YFP is expressed in epithelial tissue upon breeding with K14-Cre mice . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 004 We next sought to determine whether the Tgfbr2-deficient CD34+ SCC cells were enriched for self-renewal properties and tumorigenicity in vivo . We previously developed an orthotopic transplantation assay in which anorectal SCC cells are injected specifically and reproducibly within the anorectal transition zone of immunocompromised Nu/Nu mice ( McCauley and Guasch , 2013 ) . Orthotopic injection of 200–106 CD34+ cells from Tgfbr2 cKO anorectal SCC into recipient Nu/Nu mice caused secondary tumor formation with 100% efficiency ( n = 89 ) ( Figure 2—figure supplement 1A ) . These secondary anorectal tumors were invasive , moderately- to poorly-differentiated SCC , characterized by cellular atypia , squamous nests with keratin pearls , intercellular bridges , aberrant mitoses , cellular disorganization and desmoplastic stroma ( Figure 2—figure supplement 1C ) , and were morphologically similar to the primary Tgfbr2 cKO tumors of origin . Just as in the primary Tgfbr2 cKO SCC , a population of YFP+ epithelial cells expressing CD34 could be identified by immunofluorescence staining ( Figure 2B ) and isolated by FACS using the same strategy as used previously ( Figure 2C ) . The frequency of epithelial CD34+ cells in the secondary tumors ranged from 7% to 22% . CD34 expression within this population of cells was confirmed at the mRNA level ( Figure 2—figure supplement 2 ) . We confirmed that loss of Tgfbr2 was maintained in YFP+ SCC cells by qPCR ( Figure 2D ) and immunofluorescence staining , whereas TGFβRII and nuclear pSMAD2 were still present in the surrounding stroma ( Figure 2E–F ) . 10 . 7554/eLife . 22914 . 005Figure 2 . Tgfbr2 cKO CD34+ SCC cells are enriched for in vivo tumorigenicity . ( A ) Strategy to generate secondary tumors from the triple transgenic mice K14-Cre; Tgfbr2flox/flox; R26R-eYFPflox-STOP-flox ( cKO ) . ( B–B’ ) Immunofluorescence staining of the secondary anorectal SCC revealed populations of YFP+CD34+ and YFP+CD34− tumor cells , preserving the hierarchy observed in the primary Tgfbr2 cKO tumor . White arrows show the clusters of YFP+CD34+ cells . Dotted lines delineate the tumor from stroma . DAPI counterstains nuclei in blue . See Figure 2—figure supplement 2 for histology and FACS profile of the secondary and tertiary tumors . ( C ) Using the same FACS strategy as employed for the primary Tgfbr2 cKO anorectal SCC , the secondary anorectal tumors were sorted and distinct CD34+ and CD34− epithelial populations were isolated . ( D ) FACS-isolated YFP+CD34+ and YFP+CD34− epithelial tumor cells were subjected to mRNA extraction and qPCR and compared to FACS-isolated YFP-negative cells for Tgfbr2 expression . Data represent the mean ± s . d . from three independent tumors; Student's t-test , *p=0 . 0313 . ( E–F ) Immunofluorescence staining of the secondary anorectal SCC confirmed the loss of TGFβRII ( E–E’ ) and phosphorylated SMAD2 ( F–F’ ) in the epithelial YFP+ cells while expression was maintained in the K14-YFP- stroma ( denoted by the white arrows ) . This is a representative example of 21 secondary tumors analyzed by histology , immunostaining and FACS . Abbreviation: bv , blood vessel . Scale bars = 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 00510 . 7554/eLife . 22914 . 006Figure 2—source data 1 . Values and statistics for Figure 2D using the Wilcoxon matched-pairs signed rank test . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 00610 . 7554/eLife . 22914 . 007Figure 2—figure supplement 1 . Orthotopic transplant of CD34+ cKO SCC cells results in secondary and tertiary tumor formation which recapitulate the hierarchy of the tumor of origin . ( A ) Orthotopic transplantation of 200–106 Tgfbr2 cKO CD34+ SCC cells into the anorectal transition zone of recipient nude mice resulted in secondary tumor formation with 100% efficiency ( n = 89 in sum; 200 cells , n = 4; 1000 cells , n = 6; 5000 cells , n = 2; 10 , 000 cells , n = 10; 100 , 000 cells , n = 44; 200 , 000 cells , n = 7; 300 , 000 cells , n = 6; 500 , 000 cells , n = 4; 750 , 000 cells , n = 2; 1 , 000 , 000 cells , n = 4 ) . Data represent the mean number of days after transplantation before palpable tumor formation ± standard deviation . ( B ) CD34+ SCC cells were enriched for tumor forming efficiency , compared to CD34− SCC cells or YFP+α6-β1- SCC cells , when transplanted orthotopically into a tertiary mouse . ( C–E ) H and E staining revealed that orthotopic transplant of cultured Tgfbr2 cKO CD34+ SCC cells into a secondary recipient ( C ) or orthotopic transplant of Tgfbr2 cKO CD34− SCC cells ( D ) or orthotopic transplant of Tgfbr2 cKO CD34+ SCC cells directly into a tertiary recipient ( E ) results in SCC formation which recapitulate the morphology of the Tgfbr2 deficient tumor of origin ( n = 1/13 CD34− , n = 8/13 CD34+ ) . Scale bars: 100 µm ( C–E ) , 50 µm ( C’–E’ ) . ( F–H ) Using the same FACS strategy as employed for the primary Tgfbr2 cKO anorectal SCC , the secondary and tertiary anorectal tumors were sorted and distinct CD34+ and CD34− epithelial populations were isolated , maintaining the tumor hierarchy of the Tgfbr2 cKO tumor of origin . Of the one tertiary mouse that developed a tumor from transplant of CD34− cells , CD34 was re-expressed within the tumor environment ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 00710 . 7554/eLife . 22914 . 008Figure 2—figure supplement 2 . CD34 mRNA expression correlates with CD34 protein expression in sorted cKO SCC cells . Quantitative real-time PCR validated that CD34 mRNA expression is increased in the CD34+ sorted cell populations ( two cell lines tested ) . ***p=0 . 000013 . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 008 To determine whether SCC CD34+ cells were tumorigenic , CD34+ and CD34− YFP+ epithelial cells were isolated from the secondary anorectal tumor by FACS and orthotopically transplanted into the anorectal transition zone of Nu/Nu mice . After tertiary transplant of CD34+ SCC cells , 8/13 mice developed anorectal tumors , compared to 1 mouse of 13 transplanted with CD34− SCC cells ( Figure 2—figure supplement 1B ) . YFP+ cells negative for α6-integrin and β1-integrin were unable to form tumors upon transplantation ( n = 14 ) , indicating that not every cell type within the Tgfbr2 cKO anorectal SCC is tumorigenic . Tertiary anorectal tumors maintained the tumor hierarchy of the primary and secondary Tgfbr2 cKO SCC , and selection for CD34+ CSCs resulted in an increased ratio of CD34+ cells ( 75% ) ( Figure 2—figure supplement 1D–E ) . In fact , in the single tertiary tumor that formed after transplant of CD34− SCC cells , CD34 was re-expressed in 49% of YFP+ epithelial tumor cells by FACS ( Figure 2—figure supplement 1G ) , indicating that CD34 expression is dynamic in vivo . This is in accordance with the dynamic CD34 expression found in a DMBA-induced model of Tgfbr2 deficient SCC of the backskin ( Schober and Fuchs , 2011 ) . Whereas CSC marker expression may be dynamic ( Clevers , 2011 ) , CD34 remains a useful marker to assay the CSC properties of a population of cells isolated from murine SCC . Taken together , the ability of CD34+ cells , which lack TGFβ signaling , to form secondary and tertiary tumors that recapitulate the hierarchy of the Tgfbr2 cKO primary tumors suggests that CD34+ SCC cells are able to self-renew and differentiate in vivo . Squamous cell carcinomas , including those occurring at transition zones , frequently metastasize to the lung . We analyzed the lungs of tumor-bearing mice and observed that Tgfbr2 cKO SCC indeed spontaneously metastasized to the lungs with 100% frequency ( n > 30 mice analyzed ) ( Figure 3A ) . These lung metastases expressed Keratin 5 ( Figure 3—figure supplement 1 ) , indicating their squamous epithelial origin . Furthermore , Tgfbr2 cKO lung metastases expressed YFP and contained populations of CD34+ cells ( Figure 3B–C ) , recapitulating the hierarchy of the primary tumor of origin . Sequencing of RNA from CD34+ and CD34- Tgfbr2 cKO SCC cells , isolated by FACS as described in Figure 2C , revealed that CD34+ SCC cells were enriched for an invasive and metastatic gene signature as well as for mRNA involved in the RAC/RHO/RAS pathway ( Figure 3D and Supplementary file 1 ) . Using ToppCluster to generate a network of genes shared between Tgfbr2 deficient CD34+ anorectal transitional SCC cells and published datasets of aggressive human cancers ( Figure 3—figure supplement 2 ) we found that a number of genes were commonly overexpressed in human cancers including cervical carcinoma . We validated a selection of the most upregulated genes in the CD34+ SCC cells by qPCR ( Figure 3E ) , including Cathepsin S , Fibrillin1 , Spp1 , Mmp9 and Tgfb2 , which are all implicated in ECM organization , invasion and metastasis , and members of the RHO GTPase pathway Rac2 , Rhoh , RhoJ , Vav1 , Dock2 , and Elmo1 . 10 . 7554/eLife . 22914 . 009Figure 3 . Tgfbr2 cKO CD34+ SCC cells display a metastatic transcriptional signature . ( A–C ) H and E staining of the lungs of Tgfbr2 cKO mice ( A ) and mice orthotopically transplanted with Tgfbr2 cKO CD34+ SCC cells ( B ) revealed metastatic nodules which are YFP+ and contain a population of CD34+ tumor cells ( C ) . The boxed area represents isolation and magnification of CD34 in the red channel . See also Figure 3—figure supplement 1 . ( D ) RNA-Seq comparison of CD34− and CD34+ cells isolated from Tgfbr2 cKO CD34+ SCC ( n = 2 tumors each from two distinct cell lines ) revealed that CD34+ SCC cells were enriched for an invasive and metastatic signature . This table represents a selected set of genes which are upregulated ( red ) or downregulated ( green ) by more than two fold with an FDR <0 . 05 in FACS-purified CD34+ cells compared to CD34− cells . Genes in bold were selected for validation by qRT-PCR . See Supplementary file 1 for the full table of differentially expressed genes and Figure 3—figure supplement 2 for comparison with human databases . ( E ) Selected genes which were upregulated in CD34+ cells compared to CD34− cells in the RNA-Seq analysis were selected for validation by qRT-PCR , including genes involved in ECM organization , adhesion , invasion and metastasis and the RAC/RHO/RAS pathway . Asterisks denote statistical significance using two-tailed , unpaired student’s t-test; Ctss p=0 . 050895 , Fbn1 p***=0 . 00003 , Spp1 p***=0 . 00035 , MMP9 p**=0 . 00801 , Tgfb2 p*=0 . 016721 , Rac2 p*=0 . 0296 , Rhoh p=0 . 177 , Rhoj p***=0 . 000057 , Vav1 p***=0 . 000032 , Dock2 p*=0 . 02782 , Elmo1 p***=0 . 00067 . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 00910 . 7554/eLife . 22914 . 010Figure 3—figure supplement 1 . Lung metastases express keratin 5 . ( A–B ) Immunofluorescence staining of lungs of Tgfbr2 cKO mice ( A ) and mice transplanted with CD34+ cKO SCC cells ( B ) revealed clusters of keratin 5-positive epithelial tumor cells , whereas the lung parenchyma of wild-type mice ( C ) is negative for keratin five with the exception of low expression levels in the larger conducting airways . Abbreviation: br , bronchi . DAPI counterstains nuclei in blue . Scale bars = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 01010 . 7554/eLife . 22914 . 011Figure 3—figure supplement 2 . Tgfbr2 cKO CD34+ SCC cells upregulate genes implicated in invasive human cancers . Using ToppCluster , the genes upregulated in CD34+ SCC cells were compared to previously published datasets and a network of genes shared between Tgfbr2 deficient transitional epithelial CD34+ SCC cells and aggressive human cancers was generated . Genes colored in yellow ( invasion and metastasis ) and blue ( RAC/RHO/RAS pathway ) correspond to the genes selected for qRT-PCR validation in Figure 3E . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 011 We confirmed that the GEF ELMO1 is co-expressed with CD34-positive tumor epithelial cells at the protein level by immunofluorescence staining ( Figure 4A ) and observed increased staining of this RAC-activating factor at the leading edge of the tumor . We also confirmed that RAC2 is co-expressed with CD34-positive cells ( Figure 4B ) and that RAC1 is strongly expressed at the tumor-stroma border of Tgfbr2 cKO SCC ( Figure 4C ) . A strong RAC activity was confirmed in vitro when we analyzed the amount of GTP-bound RAC in Tgfbr2 cKO SCC CD34+ cells ( Figure 4D ) . These results indicate that , while Rac1 mRNA was not found to be upregulated in CD34+ cells by RNA-Seq analysis , RAC1 protein activity may be elevated in CSCs by upregulation of GEFs . Taken together , these data implicate CD34+ CSCs in metastasis of Tgfbr2-deficient SCC , potentially through upregulation of the RHO/RAC GTPase pathway . 10 . 7554/eLife . 22914 . 012Figure 4 . Tgfbr2-deficient tumors upregulate RAC signaling . ( A–C ) Immunofluorescence staining with antibodies against YFP , ELMO1 , RAC2 , RAC1 and CD34 revealed that some CD34+ tumor cells co-express ELMO1 ( A , white arrows ) , RAC2 ( B , white arrows ) and RAC1 ( C , white arrows ) , with strong expression at the invading front at the tumor-stroma border ( dashed lines ) . DAPI counterstains nuclei in blue . n = 6 tumors tested . Abbreviations: bv , blood vessel . Scale bars = 20 µm . ( D ) RAC activity assay revealed the increased amount of GTP-bound RAC in the Tgfbr2 cKO SCC CD34+ cell line compared to wild-type anal keratinocytes ( K ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 012 We hypothesized that aberrant RAC signaling was , at least in part , responsible for the highly aggressive nature of transition zone cancers . We compared the profile of CD34+ CSCs from our Tgfbr2 cKO anorectal SCC to the CSCs purified from malignant skin SCC in mice from various genetic backgrounds treated with the chemical mutagen 7 , 12-dimethyl-benz[a] anthracene ( DMBA ) to induce an activating mutation in Hras ( Schober and Fuchs , 2011 ) . In DMBA-induced Tgfbr2 cKO skin SCC , Schober et al reported two CSC populations: CD34Hi and CD34Lo , with the CD34Lo CSC population demonstrating increased tumorigenicity over the CD34Hi population . Bioinformatics analysis revealed that only six genes were common between the CSC from Tgfbr2 cKO anorectal CD34+ cells and DMBA-induced Tgfbr2 cKO skin CD34Hi population ( ***p value=1 . 57×10−7 ) and seven genes were common in the DMBA-induced Tgfbr2 cKO skin CD34Lopopulation ( ***p value=7 . 48×10−5 ) ( Figure 5—figure supplement 1 ) . As expected in malignant cancers , these genes are involved in ECM organization , epithelial to mesenchymal transition ( EMT ) and metastasis but none was related to the RAC/RHO/RAS pathway . When we compared the skin CSC population from DMBA-induced Tgfbr2 and focal adhesion kinase ( FAK ) double KO mice , in which SCC initiation is delayed compared to Tgfbr2 cKO skin , nine genes were common between the Tgfbr2 cKO anorectal CD34+ CSCs and the CD34Lopopulation ( ***p value=1 . 98×10−10 ) . Similarly , only 11 genes involved in ECM organization and other functions were common between the Tgfbr2 cKO anorectal CD34+ CSCs and the DMBA-induced skin CD34Lopopulation in genetically wild-type mice ( ***p value=1 . 49×10−11 ) . The analysis of skin SCC signatures from FAK single KO mice , which are more refractory to DMBA-induced SCC formation , showed 21 common genes between Tgfbr2 cKO anorectal CD34+ CSCs and the CD34Lopopulation ( ***p value=5 . 35×10−26 ) and one gene related to the RAC/RHO/RAS pathway was found in common ( Pdgfrb ) . We also compared our Tgfbr2 cKO anorectal SCC CD34+ CSC gene signature with DMBA-induced skin SCC CD34+ cells with overexpression of VEGF , which accelerates tumor growth ( Beck et al . , 2011 ) . We found 40 genes that are upregulated in Tgfbr2-deficient anorectal CD34+ CSCs yet downregulated by VEGF in comparison to DMBA-induced backskin CD34+ CSCs ( ***p value=2 . 69×10−53 ) , and we found 45 genes commonly upregulated between Tgfbr2-deficient anorectal CD34+ CSCs and DMBA-induced SCC CD34+ skin CSCs when mice overexpressed VEGF ( ***p value=2 . 62×10−39 ) . Among these data sets , we found overlap of three RAC/RHO/RAS family genes with Tgfbr2-deficient anorectal CD34+ CSCs , suggesting that dysregulation of the RAC/RHO/RAS pathway may be associated with CSCs within highly aggressive tumors . Immunofluorescence staining confirmed that TGFβ-deficient DMBA-induced skin SCC does not express RAC1 and RAC2 in contrast to anorectal SCC , where these proteins are found at the invasive front of the tumor ( Figure 5 ) . These analyses suggest that aberrant RAC signaling may be a hallmark of highly aggressive tumors arising spontaneously in tumor-prone transition zones . 10 . 7554/eLife . 22914 . 013Figure 5 . RAC1 and RAC2 are uniquely expressed in Tgfbr2-deficient transition zone tumors compared to DMBA-induced Tgfbr2-deficient skin SCC . ( A–D ) Immunofluorescence staining with antibodies against RAC1 , RAC2 and CD34 revealed strong expression of RAC1 and RAC2 in the anorectal SCC tumor compared to the backskin SCC . Skin tumors come from cKO mice treated for 16 weeks topically with the chemical mutagen 7 , 12-dimethyl-benz[a] anthracene ( DMBA ) as previously described ( Guasch et al . , 2007 ) . Some RAC1-positive cells ( A ) and clusters of RAC2-positive tumor cells ( C ) correlate with CD34+ tumor cells ( white arrows ) in the anorectal SCC but not in the skin SCC ( B , D ) . All images have been acquired using the same laser parameters and exposure time . DAPI counterstains nuclei in blue . Abbreviation: bv , blood vessel . Scale bars = 20 µm ( A–A’’–B–B’’ ) , 10 µm ( C–C’’–D–D’’ ) . n = 3 different skin and anorectal tumors tested for each antibody . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 01310 . 7554/eLife . 22914 . 014Figure 5—figure supplement 1 . Venn diagrams of cell-type specific signatures in various skin SCC . CD34+ and CD34− anorectal TZ SCC TGFβ-deficient signatures ( Supplementary file 1 ) have been compared with all databases in Table 1 of Schober and Fuchs ( 2011 ) that are in the same genetic drivers Tgfbr2-deficient with and without loss of Focal Adhesion Kinase ( FAK ) using R program available through the Comprehensive R Archive Network . CD34+ and CD34− anorectal TZ SCC TGFβ-deficient signatures have been also compared with other transcriptional profiles of CSC from skin cancers in a TGFβ-intact background ( DMBA/TPA treated ) in a WT background ( Schober and Fuchs , 2011 ) and VEGF gain of function ( from Beck et al . , 2011 ) , https://www-ncbi-nlm-nih-gov . gate2 . inist . fr/geo/query/acc . cgi ? acc=GSE31465 ) . The overlaps represent commonly enriched genes in corresponding populations compared to all others . In a TGFβ-deficient background , only six genes were found significantly commonly highly expressed between the Tgfbr2 cKO CD34+ SCC anorectal TZ cells and the Tgfbr2 cKO CD34High SCC skin cells ( ***p value=1 . 57×10-7 ) , seven genes were found significantly commonly highly expressed between the Tgfbr2 cKO CD34+ SCC anorectal TZ cells and the Tgfbr2 cKO CD34Low SCC skin cells ( ***p value=7 . 48×10-5 ) and nine genes were found significantly commonly highly expressed between the Tgfbr2 cKO CD34+ SCC anorectal TZ cells and the Tgfbr2/FAK double KO CD34Low SCC skin cells ( ***p value=1 . 98×10-10 ) . In a WT background , 11 genes were found common between the Tgfbr2 cKO CD34+ SCC anorectal TZ cells and the CD34Low SCC skin cells ( ***p value=1 . 49×10-11 ) and 21 genes when FAK is deficient ( ***p value=5 . 35×10-26 ) . In skin SCC with a gain of function of VEGF ( Beck et al . , 2011 ) , 45 genes upregulated in CD34+VEGF versus CD34+ skin SCC were found also significantly highly expressed in the Tgfbr2 cKO CD34+ SCC anorectal TZ cells ( ***p value=2 . 62×10-39 ) and 40 genes downregulated in CD34+VEGF versus CD34+ skin SCC were found significantly highly expressed in the Tgfbr2 cKO CD34+ SCC anorectal TZ cells ( ***p value=2 . 69×10-53 ) . The genes have been listed and classified in five categories: ECM organization , epithelial to mesenchymal transition , adhesion/invasion/metastasis , Rac/Rho/Ras pathway and others . All p values have been calculated with the hypergeometric ( HG ) test and are represented in the graph . The orange vertical bar indicates the statistical significance at 5% . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 014 ELMO proteins form a complex with DOCK proteins that serves as a GEF for RAC proteins in many cellular processes , including cancer cell invasion ( Laurin and Cote , 2014; Côté and Vuori , 2007; Gumienny et al . , 2001; Grimsley et al . , 2004; Brugnera et al . , 2002; Jarzynka et al . , 2007; Sai et al . , 2008; Li et al . , 1706; Komander et al . , 2008 ) . Because Dock2 and Elmo1 mRNA were both upregulated in Tgfbr2 cKO CD34+ SCC cells , we wondered whether the loss of TGFβ signaling was responsible for the upregulation of these RAC pathway activators . We probed the promoter regions of the mouse Dock2 and Elmo1 genes for the consensus SMAD-binding element ( SBE ) GTCT ( Derynck et al . , 1998; Massague and Wotton , 2000 ) and the consensus TGFβ-inhibitory element ( TIE ) GNNTTGGNGN ( Kerr et al . , 1990; Frederick et al . , 2004; McCauley et al . , 2014 ) . The Dock2 promoter contained one TIE located 266 bp upstream of the transcriptional start site ( TSS ) and four SBEs located 207 , 497 , 559 and 709 bp upstream of the TSS ( Figure 6—figure supplement 1A ) , and the Elmo1 promoter contained one TIE located 902 bp upstream of the TSS and one SBE located 153 bp upstream of the TSS ( Figure 6A ) . These sites were evolutionarily conserved and not found in repeat regions of the genome . Moreover , these sites were not found in the promoter of Elmo2 , despite its high degree of similarity with Elmo1 . We used chromatin immunoprecipitation ( ChIP ) to determine whether SMAD3 , a canonical effector of TGFβ signaling , bound any of these potential SBEs or TIEs . SMAD3 bound to the TIE on the Elmo1 promoter , but not to the SBE ( Figure 6B ) , and did not bind to any of the sites identified on the Dock2 promoter ( Figure 6—figure supplement 1A ) . These results indicated that Elmo1 is a previously unidentified direct target of TGFβ signaling via SMAD3 . 10 . 7554/eLife . 22914 . 015Figure 6 . The GEF ELMO1 is a novel target of TGFβ signaling via SMAD3 . ( A ) Promoter analysis using MatInspector ( Genomatrix ) revealed two putative SMAD-responsive elements in the Elmo1 promoter . The consensus SMAD-binding element ( SBE ) sequence GTCT was identified 153 base pairs ( bp ) upstream of the Elmo1 transcriptional start site , and the consensus TGFβ-inhibitory site ( TIE ) GNNTTGGNGN was identified 902 bp upstream of the Elmo1 transcriptional start site . Primers were designed to flank these sites ( purple arrows ) . ( B ) Chromatin immunoprecipitation with an anti-SMAD3 antibody was used to isolate DNA fragments that were amplified by the primers designed to flank the Elmo1 TIE , but not the Elmo1 SBE , after overexpressing SMAD3 in NIH3T3 cells and treating with TGFβ1 ( 2 ng/ml ) for 24 hr . Non-template ( NTC ) and no-antibody ( no Ab ) controls were used to verify the specificity of binding . ( C–E ) Lentiviral infection of Tgfbr2 cKO CD34+ SCC cells with the full-length Tgfbr2 gene inserted into the pLVX-IRES-mCherry vector resulted in rescue of Tgfbr2 mRNA by more than 4000-fold ( C–D ) and phosphorylated SMAD2 ( p-SMAD2 ) in response to treatment with TGFβ1 ( 2 ng/ml ) for 1 hr ( E ) , compared to Tgfbr2 cKO CD34+ SCC cells infected with the empty pLVX-IRES-mCherry vector . No Tgfbr2 mRNA was detected in Tgfbr2 cKO CD34+ SCC cells infected with the empty pLVX-IRES-mCherry vector . Data represent the mean ± standard deviation; student’s t-test , ***p=0 . 000112 . ( F–I ) Tumors generated from orthotopic transplantation of 100 , 000 Tgfbr2 cKO CD34+ SCC cells infected with empty vector or with full-length Tgfbr2 were dissociated and YFP+mCherry+ , YFP+mCherry-CD34+ and YFP+mCherry-CD34− cells were isolated by FACS and subjected to RNA extraction . ( F ) Approximately 7 . 7% of the cKO SCC + empty vector total tumor bulk expressed mCherry , whereas 0 . 5% of the cKO SCC + Tgfbr2 expressed mCherry at the time of analysis . See also Figure 6—figure supplement 2 . The frequency of YFP+mCherry+CD34+ cells was significantly reduced in the rescued cKO SCC + Tgfbr2 tumor . ( G ) YFP+mCherry+ cells isolated from tumors generated from Tgfbr2 cKO CD34+ SCC cells infected with full-length Tgfbr2 expressed Tgfbr2 mRNA 250-fold over YFP+mCherry- cells isolated from the same tumor or YFP+mCherry+ cells isolated from tumors generated from Tgfbr2 cKO CD34+ SCC cells infected with empty vector ( ***p=0 . 000005 ) . No Tgfbr2 mRNA was detected by qRT-PCR in YFP+mCherry- cells isolated from tumors generated from Tgfbr2 cKO CD34+ SCC cells infected with full-length Tgfbr2 or YFP+mCherry+ cells isolated from tumors generated from Tgfbr2 cKO CD34+ SCC cells infected with empty vector . ( H ) Rescue of TGFβRII abolished Elmo1 mRNA expression in YFP+mCherry+ cells isolated from tumors generated from Tgfbr2 cKO CD34+ SCC cells infected with full-length Tgfbr2 compared to YFP+mCherry-CD34+ cells isolated from the same tumor . Data represent the mean ± standard deviation . Asterisks denote statistical significance using two-tailed , unpaired student’s t-test; p***=9×10−24 . Three different tumors for each condition have been analyzed . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 01510 . 7554/eLife . 22914 . 016Figure 6—figure supplement 1 . Dock2 is not a direct target of TGFβ signaling via SMAD3 . ( A ) Promoter analysis using MatInspector ( Genomatrix ) revealed five putative SMAD- responsive elements in the Dock2 promoter . The consensus SMAD-binding element ( SBE ) sequence GTCT was identified 207 , 497 , 559 and 709 base pairs ( bp ) upstream of the Dock2 transcriptional start site , and the consensus TGFβ-inhibitory site ( TIE ) GNNTTGGNGN was identified 266 bp upstream of the Dock2 transcriptional start site . Primers were designed to flank these sites ( purple arrows ) . ( B ) Chromatin immunoprecipitation with an anti-SMAD3 antibody failed to isolate DNA fragments that were amplified by the primers designed to flank any of these sites on the Dock2 promoter , after overexpressing SMAD3 in NIH3T3 cells and treating with TGFβ1 ( 2 ng/ml ) for 24 hr . Non-template ( NTC ) and no-antibody ( no Ab ) controls were used to verify the specificity of binding . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 01610 . 7554/eLife . 22914 . 017Figure 6—figure supplement 2 . Restoration of Tgfbr2 caused a two-fold delay in tumor formation . Tumors were generated from orthotopic transplantation of 100 , 000 Tgfbr2 cKO CD34+ SCC cells infected with empty vector pLVX-IRES-mCherry or with full-length Tgfbr2 inserted into the pLVX-IRES-mCherry vector . Three tumors for each condition were generated . The latency represents the day when the tumor is visible and palpable . The survival represents the day when the mouse is sacrificed due to tumor burden . Mice are sacrificed when the tumor exceeds the authorized size and volume according to Institutional Animal Care and Use Committee . Asterisks denote statistical significance using two-way ANOVA and Bonferroni post tests . p**<0 . 01 , p*<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 017 Because we identified a direct link between SMAD3 and the promoter region of Elmo1 , we hypothesized that rescue of TGFβRII in Tgfbr2-deficient cells would reduce ELMO1 expression . We cloned the full-length Tgfbr2 gene into the pLVX-IRES-mCherry lentiviral vector and infected Tgfbr2 cKO CD34+ SCC cells with this construct or the empty vector . Infection of YFP+CD34+ SCC cells with the rescue construct , but not the empty vector , efficiently restored Tgfbr2 mRNA expression ( Figure 6C–D ) . Rescued CD34+ SCC cells became sensitive to TGFβ treatment and phosphorylated SMAD2 ( Figure 6E ) . Orthotopic transplantation of rescued CD34+ SCC cells resulted in a two-fold delay in tumor latency compared to Tgfbr2 cKO CD34+ SCC cells infected with the empty vector ( Figure 6—figure supplement 2 ) , although all mice eventually developed tumors due to the inefficient infection rate of the rescue vector ( Figure 6F ) . Tumors generated from the orthotopic transplantation of Tgfbr2 cKO CD34+ SCC cells infected with the empty vector or rescued with full-length Tgfbr2 were dissociated and YFP+mCherry+ , YFP+mCherry-CD34+ and YFP+mCherry-CD34− cells were isolated by FACS ( Figure 6F ) and subjected to RNA extraction . YFP+mCherry+ cells isolated from tumors generated from rescued Tgfbr2 cKO CD34+ SCC cells infected with full-length Tgfbr2 expressed Tgfbr2 mRNA 250-fold over YFP+mCherry- cells isolated from the same tumor or YFP+mCherry+ cells isolated from tumors generated from Tgfbr2 cKO CD34+ SCC cells infected with empty vector ( Figure 6G ) . No Tgfbr2 mRNA was detected by qRT-PCR in YFP+mCherry- cells isolated from tumors generated from Tgfbr2 cKO CD34+ SCC cells infected with full-length Tgfbr2 or YFP+mCherry+ cells isolated from tumors generated from Tgfbr2 cKO CD34+ SCC cells infected with empty vector ( Figure 6G ) , indicating that mCherry expression faithfully represented cells in which TGFβ signaling had been restored . Restoration of Tgfbr2 in Tgfbr2 cKO CD34+ SCCs dramatically reduced the frequency of mCherry+CD34+ CSCs from 6 . 5% to 2 . 5% ( Figure 6F ) , demonstrating that loss of TGFβ signaling is a requirement for CSC maintenance in transition zone carcinoma . Importantly , rescue of Tgfbr2 abolished Elmo1 mRNA expression in YFP+mCherry+ cells isolated from tumors generated from Tgfbr2 cKO CD34+ SCC cells infected with full-length Tgfbr2 compared to YFP+mCherry-CD34+ cells isolated from the same tumor ( Figure 6H ) , confirming that the RAC-activating GEF Elmo1 is a novel target of TGFβ repression . Given that Tgfbr2-deficient anorectal SCC expressed the GEF ELMO1 , we wondered whether ELMO1 might also be expressed in human anorectal cancers . We performed immunohistochemistry on a series of human anorectal biopsies that ranged from normal mucosa to invasive grade three carcinomas . We found that ELMO1 was expressed in 5 of the 15 of anorectal tumors tested ( Table 1 ) . Concomitant with this expression was a corresponding loss of phosphorylated SMAD2 within the tumor tissue in 5 out of 6 invasive SCC ( Table 1 and Figure 7C–E , Figure 7—figure supplement 1 ) . Interestingly , none of the early stage tumors tested ( anal intraepithelial neoplasia or SCC in situ ) expressed ELMO1 , and these specimens stained strongly for nuclear phosphorylated SMAD2 similarly to normal anorectal mucosa ( Figure 7A–B , Figure 7—figure supplement 1 ) . Taken together , these data support a role for ELMO1 in invasive TGFβ-deficient transition zone SCC . 10 . 7554/eLife . 22914 . 018Table 1 . ELMO1 is expressed in human TGFβ-deficient invasive anorectal SCC . Human anorectal tumor biopsies from male and female patients , aged 32–70 , were analyzed for phosphorylated ( activated ) SMAD2 ( pSMAD2 ) and ELMO1 by immunohistochemistry ( IHC ) ( see Figure 7 and Figure 7—figure supplement 1 ) . Loss of pSMAD2 correlated with increased ELMO1 expression in 5 out of 6 SCC samples . Scoring: ( ++ ) , strong positive staining; ( + ) , positive staining; ( - ) , negative staining . Abbreviations: SCC: squamous cell carcinoma; AIN3: Anal intraepithelial neoplasia ( early stage tumor ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 018DiagnosisSexAgepSMAD2ELMO1Picturesnormal anorectal mucosaM58++−Figure 7ASCC in situM47++−Figure 7—figure supplement 1ASCC in situ + focus of invasionM37++−SCC in situ with microinvasionM69++−SCC in situ + invasive SCCF70++ and −−AIN3F32++−Figure 7BAIN3M52++−invasive SCC + AIN3M61++ and −+invasive SCC + AIN3M52++ and −+Figure 7—figure supplement 1Binvasive SCC grade 2M51++−invasive SCC grade 2M59++−invasive SCC grade 2F51++−invasive SCC grade 3M53++−invasive SCC grade 3F55++ and −+Figure 7Cinvasive SCC grade 3F53+ and −++Figure 7Dinvasive SCC grade 3M48−++Figure 7E10 . 7554/eLife . 22914 . 019Figure 7 . ELMO1 is expressed in human TGFβ-deficient invasive anorectal SCC . Human anorectal tumor biopsies were analyzed for phosphorylated ( activated ) SMAD2 ( pSMAD2 ) and ELMO1 by immunohistochemistry ( IHC ) . See also Figure 7—figure supplement 1 for additional tumor biopsy sections and antibody controls . Examples of IHC staining from three invasive anal SCC grade 3 ( C–E ) show reduced or absent nuclear pSMAD2 staining in contrast to normal anorectal mucosa ( A ) and early stage tumor , anal intraepithelial neoplasia ( B ) . Loss of pSMAD2 correlated with increased ELMO1 expression in 5 out of 6 SCC samples ( see Table 1 ) . Scoring: ( ++ ) , strong positive staining; ( + ) , positive staining; ( − ) , negative staining . Dashed lines delineate the epidermis from dermis in ( A ) and delineate the tumor from stroma in ( B–E ) . Hematoxylin counterstains nuclei in blue . Scale bars = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 01910 . 7554/eLife . 22914 . 020Figure 7—figure supplement 1 . ELMO1 is expressed in human TGFβ-deficient invasive anorectal SCC . Human anorectal tumors were analyzed for phosphorylated ( activated ) SMAD2 ( pSMAD2 ) and ELMO1 by immunohistochemistry ( IHC ) ( A–B ) . Secondary antibodies alone were used as controls for the IHC ( C ) . Examples of IHC staining from one SCC in situ and one invasive anal SCC . Scoring: ( ++ ) , strong positive staining; ( + ) , positive staining; ( − ) , negative staining . Dashed lines delineate the tumor from stroma . Hematoxylin counterstains nuclei in blue . Scale bars = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 020 To determine whether a reduction in ELMO1 could reduce invasion and metastasis in Tgfbr2 cKO SCC , we knocked down Elmo1 in Tgfbr2 cKO CD34+ SCC cells using two shRNA constructs ( Figure 8 ) . Treatment of Tgfbr2 cKO CD34+ SCC cells with Elmo1 shRNA resulted in 40% ( construct #1 ) or 50% ( construct #2 ) reduction in endogenous Elmo1 mRNA levels , compared to cells infected with control shRNA ( SH02 ) ( Figure 8A ) . Western blot analysis confirmed this reduction at the protein level ( Figure 8B ) . To confirm the specificity of the shRNA , we used a hairpin-resistant ELMO1 cDNA ( ELMO1* ) in which we had introduced three base mutations in the target sequence of the Elmo1 shRNA #2 without affecting the function of ELMO1 ( Figure 8—figure supplement 1 ) . We cloned this construct into the pLVX-IRES-mCherry lentiviral vector and infected Tgfbr2 cKO CD34+ SCC Elmo1 shRNA expressing cells with ELMO1* or the empty vector . Western blot analysis confirmed that overexpression of the hairpin-resistant ELMO1* construct restores similar level of expression of ELMO1 ( Figure 8C ) . We performed an in vitro wound healing assay to show that knocking down Elmo1 in Tgfbr2 cKO CD34+ SCC cells affected their ability to migrate and close the wound . This effect was rescued when Elmo1 knockdown cells expressed the hairpin-resistant ELMO1* construct ( Figure 8D–E and Video 1 ) . We confirmed that the effect in cell migration was not due to a difference in proliferation as measured by flow cytometry ( Figure 8F ) . 10 . 7554/eLife . 22914 . 021Figure 8 . Knockdown of Elmo1 in vitro affects cell migration . ( A ) shRNA knockdown of Elmo1 in cKO SCC CD34+ cells with two constructs resulted in 40–50% reduction in endogenous Elmo1 mRNA expression compared to cKO SCC CD34+ cells infected with control shRNA ( SH02 ) . Asterisks denote statistical significance using paired-sample Wilcoxon Signed Rank test; p*=0 . 0350 ( Elmo1 shRNA#1 ) , p*=0 . 0355 ( Elmo1 shRNA#2 ) . ( B ) Western blot analysis confirmed the reduction in endogenous ELMO1 protein in cKO SCC cells compared to cells infected with control shRNA . ( C ) Western blot analysis confirmed that overexpression of the hairpin-resistant ELMO1 construct ( ELMO1* ) in the Elmo1 shRNA cKO SCC cells restored expression of Elmo1 . See also Figure 8—figure supplement 1 . ( D–E ) Wound healing assay in vitro showed that knockdown of Elmo1 in cKO SCC cells impaired their ability to migrate 5 hr and 10 hr after wounding . This effect was rescued when the hairpin-resistant Elmo1 construct ( ELMO1* ) is expressed . See also Video 1 . ( E ) Quantification of the wound healing assay showing the slope of linear trend curve of wound widths normalized to SH02 control . For each construct three different cell lines have been tested and the experiments have been repeated five times . Asterisks denote statistical significance using two-tailed , unpaired student’s t-test; p***<0 . 0001 ( Elmo1 shRNA#2 ) , p***=0 . 0002 ( Elmo1 shRNA#2 + ELMO1* ) . ( F ) Elmo1 knockdown did not affect cell proliferation . Histograms show the fluorescence of the dye eFluor 670 in the APC channel at 0 hr , 24 hr , 48 hr and 72 hr . Decrease of the fluorescence reflected proliferation by the dilution of the dye over time . Three separate experiments have been done and the mean of the geometric mean for each sample is represented in the graph . There was no statistical difference in cell proliferation between samples calculated by two-way ANOVA and Bonferroni post tests . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 02110 . 7554/eLife . 22914 . 022Figure 8—source data 1 . Values and statistics for Figure 8A using paired-sample Wilcoxon Signed Rank test . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 02210 . 7554/eLife . 22914 . 023Figure 8—source data 2 . Values and statistics for Figure 8E using two-tailed , unpaired student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 02310 . 7554/eLife . 22914 . 024Figure 8—source data 3 . Values and statistics for Figure 8F using two-way ANOVA and Bonferroni post tests . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 02410 . 7554/eLife . 22914 . 025Figure 8—figure supplement 1 . Expression of the hairpin-resistant ELMO1* construct . To validate the specificity of the anti-ELMO1 antibody ( white ) and the expression of the ELMO1 hairpin resistant construct ( ELMO1* ) , Tgfbr2 cKO CD34+ EYFP+ SCC cells were infected for 48 hr by the vector pLVX-IRES-mcherry control or the vector expressing a mutant form of ELMO1 ( ELMO1* ) that blocks the effect of the shRNA ELMO1 without affecting ELMO1 function . Immunofluorescence staining reveals that ELMO1 is recognized only in the pLVX-IRESmcherry-ELMO1* and correlated with the mcherry expression . Scale bars: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 02510 . 7554/eLife . 22914 . 026Video 1 . In vitro wound healing assay showing the migration of cKO SCC SHO2 control , knockdown of Elmo1 and the hairpin-resistant Elmo1 construct . Images were taken every 10 min for 10 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 026 We transplanted these cells orthotopically into the anorectal transition zone of recipient mice and observed tumor formation with no difference in latency . Consistent with this result , we observed that Elmo1 knockdown does not affect the proportion of CSC CD34+YFP+mCherry+ cells analyzed by FACS ( Control SH02: 4 . 1% compared to Elmo1 shRNA: 3 . 3% , Figure 9—figure supplement 1A ) We isolated epithelial YFP+CD34+ cells from these tumors , using the same FACS strategy as described in Figure 1 , and observed a 60% reduction in Elmo1 mRNA compared to CD34+ cells isolated from SH02 tumors , validating in vivo the loss of Elmo1 ( Figure 9—figure supplement 1B ) . Infection of the cells with the pLVX-mCherry vector did not affect the efficiency of the shRNA , as we observed a similar reduction in Elmo1 mRNA compared to CD34+ cells isolated from SH02 tumors that did not express the mCherry ( 50% reduction for construct #1 and 60% reduction for construct #2 ( Figure 9—figure supplement 1C ) . Using immunofluorescence we confirmed reduction in ELMO1 expression at the protein level in these tumors ( Figure 9—figure supplement 2A ) compared to SH02 tumors . Infection of CD34+ Elmo1 shRNA SCC cells with the hairpin-resistant ELMO1* construct efficiently restored Elmo1 mRNA and ELMO1 protein expression in the resulting tumors ( Figure 9—figure supplement 2B’’ ) , indicating that mCherry expression faithfully represented cells in which ELMO1 expression had been restored . Consistent with a migration defect in vitro ( Figure 8D ) , we observed a dramatic reduction in RAC1 staining at the tumor-stroma border in tumors with Elmo1 knockdown ( Figure 9A–A’ ) , which was rescued in cells expressing the hairpin-resistant ELMO1* construct ( Figure 9A’’ ) . 10 . 7554/eLife . 22914 . 027Figure 9 . Knockdown of Elmo1 diminishes RAC1 expression and markers of invasion in Tgfbr2-deficient SCC . ( A ) Immunofluorescence staining with antibodies against YFP and RAC1 revealed a reduction in RAC1 staining at the tumor-stroma border in Tgfbr2 cKO tumors with knockdown of Elmo1 ( A’ ) , compared to SH02 control tumors ( A ) . Tumors expressing the hairpin-resistant ELMO1* construct show restoration of RAC1 compared to tumors with Elmo1 knockdown ( A’’ ) . All pictures have been taken at the same exposure time in the RAC1 channel . Three tumors for each condition have been analyzed . DAPI counterstains all nuclei in blue . Scale bars = 10 μm . ( B ) qPCR analysis of genes implicated in EMT , invasion and metastasis revealed a significant reduction in their mRNA expression in CD34+ cells isolated from Tgfbr2 cKO tumors with knockdown of Elmo1 compared to SH02 control tumors ( see Figure 9—figure supplement 1 and 2 ) . Data represent the mean ± standard deviation . Asterisks denote statistical significance using two way-ANOVA and Bonferroni post tests to compare each Elmo1 shRNA to SHO2 control . All p-values are <0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 02710 . 7554/eLife . 22914 . 028Figure 9—source data 1 . Values and statistics for Figure 9B using two way-ANOVA and Bonferroni post tests . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 02810 . 7554/eLife . 22914 . 029Figure 9—figure supplement 1 . Knockdown of Elmo1 in vivo . ( A ) FACS analysis of tumors generated from orthotopic transplantation of 100 , 000 Tgfbr2 cKO CD34+ SCC cells infected with SH02 control or Elmo1 shRNA selected with puromycin and infected again with pLVX-IRES-mCherry or with the hairpin-resistant Elmo1 construct ( ELMO1* ) and selected based on mCherry expression . Three different tumors for each construct were analyzed . Approximately 32% of the cKO SCC SH02 or Elmo1 shRNA + empty vector total tumor bulk expressed mCherry , whereas less than 1 . 5% of the cKO SCC Elmo1 shRNA + ELMO1* expressed mCherry at the time of analysis . ( B ) Tumors were dissociated and YFP+mCherry-CD34+ and YFP+mCherry-CD34− cells were isolated by FACS and subjected to RNA extraction and qPCR analysis . We observed a 60% reduction in Elmo1 mRNA compared to CD34+ cells isolated from SH02 tumors . Infection of CD34+ Elmo1 shRNA SCC cells with the hairpin-resistant ELMO1* construct efficiently restored Elmo1 mRNA . Asterisks denote statistical significance using two-way ANOVA and Bonferroni post tests; p***<0 . 001 . ( C ) 100 , 000 CD34+ cKO SCC cells which were infected with SH02 control , Elmo1 shRNA construct#1 or Elmo1 shRNA construct #2 were orthotopically transplanted into recipient mice and YFP+CD34+ cells were isolated from the resulting tumors using the same FACS strategy as employed in Figure 1 , and subjected to qPCR . Elmo1 mRNA was reduced by 53–61% in CD34+ cells isolated from tumors with knockdown of Elmo1 compared to SH02 tumors . *p=0 . 045 for construct #1 and **p=0 . 004 for construct #2 . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 02910 . 7554/eLife . 22914 . 030Figure 9—figure supplement 2 . Knockdown of Elmo1 diminishes markers of invasion in Tgfbr2- deficient SCC . ( A ) Infection of CD34+ Elmo1 shRNA SCC cells with the hairpin-resistant ELMO1* construct efficiently restored ELMO1 protein expression in the generated tumors as shown by immunofluorescence with an antibody against ELMO1 . Scale bars: 10 μm . ( B ) 100 , 000 CD34+ cKO SCC cells which were infected with SH02 control , Elmo1 shRNA construct #2 or Elmo1 shRNA construct #2 with the hairpin-resistant ELMO1* were orthotopically transplanted into recipient mice and CD34+ YFP+mCherry+ cells were isolated from the resulting tumors using the same FACS strategy as employed in Figure 1 , and subjected to qPCR . qPCR analysis of genes implicated in EMT , invasion and metastasis revealed a significant reduction in their mRNA expression in CD34+ cells isolated from Tgfbr2 cKO tumors with knockdown of Elmo1 compared to SH02 control tumors . See also Figure 9—figure supplement 1 . The level of mRNA expression of these genes is restored or increased in the cells expressing the hairpin-resistant ELMO*1 construct . Data represent the mean ± standard deviation . Asterisks denote statistical significance using two-way ANOVA and Bonferroni post tests compared to SHO2; p*<0 . 05 , p**<0 . 01 , p***<0 . 001 . . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 030 Because RAC signaling plays a role in tumor invasion , we hypothesized that a number of markers of EMT and/or invasion would be altered upon Elmo1 knockdown , and indeed observed reduction in the mRNA expression of Snail , αSma , Vimentin , Zeb2 and Mmp9 in CD34+ cells isolated from Tgfbr2 cKO tumors infected with both Elmo1 shRNA constructs #1 and #2 compared to those infected with SH02 control ( Figure 9B ) . The level of mRNA expression of these genes was restored or increased in the cells expressing the hairpin-resistant ELMO1* construct ( Figure 9—figure supplement 2B ) . This increase can be explained by the fact that the level of Elmo1 in the rescued cells was higher than endogenous levels in the SH02 control cells . These results indicate that Elmo1 reduction alters localization of RAC in Tgfbr2-deficient SCC in vivo and reduces the expression of invasive markers in CD34+ CSCs . To determine whether Elmo1 knockdown altered tumor progression and metastasis in Tgfbr2 cKO SCC , we screened the lungs of tumor-bearing mice for YFP+ metastases by FACS . We observed a dramatic overall reduction in the number of YFP+ cells in the lungs of mice orthotopically transplanted with Tgfbr2 cKO CD34+ cells infected with Elmo1 shRNA knockdown compared to those infected with SH02 control ( n = 6 mice in each condition ) ( Figure 10A–B ) . We detected 1 . 4% YFP+ cells in the lungs of mice transplanted with the SH02 control cells , whereas we detected a dramatic reduction to 0 . 4% YFP+ cells in mice transplanted with the ELMO1 shRNA#1 and failed to detect any YFP+ cells in the lungs of mice transplanted with ELMO1 shRNA #2 . We serially sectioned and stained the whole lungs from additional tumor-bearing mice , and observed YFP+ metastatic nodules in mice transplanted with SH02 cells , but never in mice transplanted with Elmo1 shRNA-infected cells ( Figure 10C ) , validating the sensitivity of FACS in providing a quantitative method to screen for lung metastases . Taken together , these data demonstrate that upregulation of the GEF ELMO1 is required for Tgfbr2-deficient SCC CD34+ CSCs to metastasize . 10 . 7554/eLife . 22914 . 031Figure 10 . Knockdown of Elmo1 inhibits Tgfbr2-deficient SCC metastasis . ( A–B ) Whole lungs from mice bearing tumors from orthotopic transplantation of Tgfbr2 cKO CD34+ SCC cells infected with SH02 control and Elmo1 shRNA were dissociated and total YFP+ cells were quantified by FACS ( n = 6 different mice analyzed for each construct ) . Per-CP was used to exclude auto-fluorescent cells . Data represent the mean ± standard deviation . Asterisks denote statistical significance using two-tailed , unpaired student’s t-test; *p=0 . 0318 . ( C ) Upon serial section of entire lungs , YFP+ lung metastases were only observed microscopically in mice bearing tumors from orthotopic transplantation of Tgfbr2 cKO CD34+ SCC cells infected with SH02 control , but not infected with Elmo1 shRNA#1 or #2 . DAPI counterstains nuclei in blue . Scale bars = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 03110 . 7554/eLife . 22914 . 032Figure 10—source data 1 . Values and statistics for Figure 10B using two-tailed , unpaired student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 22914 . 032
Transition zones are found throughout the body and are preferential sites for malignant tumor formation . Despite the clinical importance of these tumors , the cellular and molecular mechanisms defining their growth and progression remain unknown . Here , we have taken advantage of a mouse model that develops spontaneous tumors at their anorectal transition zone upon loss of TGFβ signaling , which recapitulates invasive human anogenital squamous cell carcinoma , to define the cells and mechanisms by which these tumors grow and to identify the signaling pathways upon which they rely when undergoing metastasis . We have shown that epithelial CD34+ cells of Tgfbr2 cKO SCC , which are clonogenic in vitro , able to self-renew in vivo , and generate differentiated progeny in serial transplantation assay are potential CSCs . We have further shown that restoration of TGFβRII in Tgfbr2 cKO SCC reduced the frequency of these CD34+ CSCs , demonstrating that loss of TGFβ signaling is a requirement for CSC maintenance in transition zone carcinoma . This is in accordance with what has been described in Hras mutated DMBA-induced skin SCC , where the frequency of CD34+ CSCs was 10 fold higher in Tgfbr2 cKO mice than in WT mice ( Schober and Fuchs , 2011 ) . However , we find little overlap between gene signatures of CD34+ CSCs from Tgfbr2 cKO transition zone SCC and from Tgfbr2 cKO + Hras backskin SCC . Upregulated genes in the CD34+ CSC signature of Tgfbr2 cKO + Hras backskin SCC include those involved in cell cycle progression and DNA repair ( Schober and Fuchs , 2011 ) , whereas the CD34+ CSC signature in our Tgfbr2 cKO anorectal SCC is primarily comprised of genes involved in invasion , metastasis , and activation of the RAC signaling pathway . Interestingly , tumor cells with low CD34 expression displayed greater tumorigenicity than cells with high CD34 expression in the Tgfbr2 cKO + Hras backskin SCC ( Schober and Fuchs , 2011 ) , whereas we observed a dramatic enrichment for tumorigenicity in CD34+ SCC cells from transition zone tumors . We found very few genes in common between Tgfbr2-deficient anorectal CSCs and CSCs from DMBA-induced Hras skin SCC from WT mice or mice deficient in FAK , which is expected as these mice are refractory to DMBA-induced skin SCC compared to the accelerated skin SCC development in Tgfbr2 deficient mice . We found more overlap between transition zone SCC CSCs and backskin SCC CSCs when we compared the gene profile of Tgfbr2-deficient anorectal CSCs to CSCs from Hras + VEGF overexpressing backskin SCC . When mice were administered the DMBA skin carcinogenesis protocol while overexpressing VEGF , the resulting skin tumors were more aggressive and the stemness profile of CD34+ CSCs was qualitatively altered compared to wild-type mice ( Beck et al . , 2011 ) . We found 45 genes commonly upregulated between Tgfbr2-deficient anorectal CD34+ CSCs and Hras + VEGF overexpressing SCC CD34+ skin CSCs , and 40 genes that were upregulated in Tgfbr2-deficient anorectal CD34+ CSCs but downregulated by VEGF in skin CSCs . Three of these genes ( Rgs16 , Rassf3 and Fam13a ) are members of the RAC/RHO/RAS pathway , and dysregulated Rgs16 has been associated with poor prognosis in colorectal cancer ( Miyoshi et al . , 2009 ) . Our gene expression analysis is evidence that the requirements of CSCs within distinct tumor types vary , and that their growth and invasive properties are heterogeneous and context-dependent . Furthermore , our data show that deregulated expression of RAC1 and RAC2 is evident in Tgfbr2 deficient transition zone tumors but not in skin SCC in various genetic backgrounds analyzed . While we found three RAC/RHO/RAS family members which were common between Tgfbr2-deficient anorectal CSC and genes up- or down-regulated in DMBA-induced backskin CSCs in the aggressive VEGF gain-of-function mouse model , none of these three genes are RAC-activating GEFs . This suggests that the mechanism linking loss of TGFβ signaling with de-repression of RAC-activating GEFs is unique to CSCs within transition zone tumors , which may explain their particular aggressiveness and malignancy . Our findings identify a novel mechanism whereby loss of TGFβ signaling in transitional epithelial carcinoma drives an invasive and metastatic program specifically in cancer stem cells . We established the metastatic gene signature of CD34+ CSCs in Tgfbr2 cKO anorectal SCC and found that the GEF ELMO1 is a direct target of SMAD3 . While more than 70 GEFs have been described to date ( Parri et al . , 2010 ) , ELMO1 has been specifically identified as oncogenic in a variety of studies ( Lazer and Katzav , 2011; Jarzynka et al . , 2007; Li et al . , 1706; Jiang et al . , 2011; Ungefroren et al . , 2011 ) . Upregulation of GEFs is a frequent mechanism of increased RAC activity and cancer cell invasion and metastasis ( Lazer and Katzav , 2011; Ensign et al . , 2013 ) . We have shown here that loss of TGFβ-mediated repression is one mechanism that may lead to this upregulation . There have been many reports of RAC GTPases functioning downstream of TGFβ signaling ( Atfi et al . , 1997; Meriane et al . , 2002; Faherty et al . , 2013; Brown et al . , 2008 ) , but never in the context of elevated GEF expression as a result of loss of TGFβ signaling . In pancreatic ductal adenocarcinoma cells that retain TGFβ signaling , RAC1 regulates SMAD2 and SMAD3 at the post-transcriptional level , shifting TGFβ from a tumor-suppressive role to a tumor-promoting role ( Ungefroren et al . , 2011 ) , suggesting that the relationship between the TGFβ and RHO/RAC GTPase pathways is complex and context-dependent but nevertheless central . TGFβ signaling is involved in many cellular processes and participates in oncogenesis in several types of cancer , especially aggressive cancers such as pancreatic ( Jaffee et al . , 2002 ) , colon ( Jones et al . , 2008 ) , and oral carcinomas ( Sivadas et al . , 2013 ) . This participation may be through a loss of activity and loss of tumor suppressive function . However , TGFβ has a complex role in tumor progression . It can act as a tumor promoter or suppressor depending on the tumor type and stage . Identifying the consequences of TGFβ signaling alterations in cancer and the players in tumor progression and metastasis for each type of cancer is a prerequisite to derive appropriate therapeutic tools . RAC inhibitors exist , but are poorly specific . Identification of key downstream components of TGFβ signaling , such as ELMO1 , offers new perspectives and opportunities . In conclusion , this is the first report of a mechanism connecting loss of TGFβ signaling with loss of repression of a RAC-activating GEF . This mechanism may sustain invasion and metastasis in aggressive cancers that lack TGFβ signaling .
The conditional knockout Tgfbr2flox/flox x K14-Cre mouse model ( Guasch et al . , 2007; Leveen et al . , 2002 ) has been derived in a pure C57BL/6N background and backcrossed into a mouse reporter containing an Enhanced Yellow Fluorescent Protein gene ( eYFP ) inserted into the Gt ( ROSA ) 26Sor locus ( Srinivas et al . , 2001 ) and called R26R-eYFPflox-STOP-flox ( Jackson Laboratory ) ( McCauley and Guasch , 2013; McCauley et al . , 2014 ) . Control mice were either Tgfbr2flox/flox x R26R-eYFPflox-STOP-flox or Tgfbr2 +/+ x R26R-eYFPflox-STOP-flox x K14-Cre , all in a C57BL/6N background . Transplantation assays were carried out in homozygous Nu/Nu female mice , approximately six to eight weeks old , as previously described ( McCauley and Guasch , 2013 ) . The chemical mutagen 7 , 12-dimethyl-benz[a] anthracene ( DMBA ) was administered topically to the backskin of cKO mice for 16 weeks as previously described ( Guasch et al . , 2007 ) . Mice are housed in a sterile barrier facility as previously described ( McCauley and Guasch , 2013 ) . All experiments were approved by the Cincinnati Children’s Hospital Research Foundation Institutional Animal Care and Use Committee ( protocol # 1D10087 ) and in agreement with European and national regulation ( protocol # 4572 ) and carried out using standard procedures . The identification of each allele was performed by PCR on DNA extracted from clipping the ear of the mice as previously described ( Guasch et al . , 2007; McCauley et al . , 2014 ) . Tumors were dissociated into a single cell suspension according to protocol established in our lab ( McCauley and Guasch , 2013 ) and CD34+ cancer stem cells were isolated as described at Bio-protocol ( McCauley and Guasch , 2017 ) . Tumors were dissociated into a single cell suspension according to protocol established in our lab ( McCauley and Guasch , 2013 ) . Cells in suspension were labeled with the following antibodies at the dilutions indicated: PE-Cy7 conjugated to rat-anti-mouse CD11b ( BD Pharmingen , 1/200 , RRID:AB_2033994 ) , PE-Cy7 conjugated to rat-anti-mouse CD31 ( BD Pharmingen 1/100 , RRID:AB_10612003 ) , PE-Cy7 conjugated to anti-mouse CD45 ( eBioscience 1/200 , RRID:AB_469625 ) , PE conjugated to CD49f , ( BD Biosciences , 1/50 , RRID AB_396079 ) , Pacific Blue conjugated to CD29 ( Biolegend , 1/100 , RRID:AB_2128079 ) , biotin conjugated to anti-mouse CD34 ( eBioscience , 1/50 , RRID:AB_466426 ) and APC conjugated to streptavidin ( BD Pharmingen , 1/200 , RRID:AB_10050396 ) . Immediately before sorting , cells were incubated with 7-amino-actinomycin D ( 7-AAD , eBioscience , 20 µl 0 . 05 mg/ml stock per 106 cells ) to exclude dead cells . Tumor cell populations were sorted for RNA extraction , tissue culture or transplantation using sterile practices using a FACS Aria II ( BD Biosciences ) and FACSDiva software ( BD Biosciences ) in the Research Flow Cytometry Core at CCHMC . Cells isolated for RNA extraction were collected directly into cell lysis buffer containing beta-mercaptoethanol , vortexed and stored at −80°C until RNA extraction . Cells isolated for tissue culture were collected in epithelial cell culture media ( Nowak and Fuchs , 2009 ) ( E media ) containing 0 . 05 mM calcium , centrifuged at 1000 RPM for 5 min at 4°C , resuspended in E media containing 0 . 3 mM calcium , and plated on a feeder layer of irradiated fibroblasts . Cells isolated for transplantation were collected in E media without serum , centrifuged at 1000 RPM for 5 min at 4°C , resuspended in 30% Matrigel and transplanted as previously described ( McCauley and Guasch , 2013 ) . Tumors were dissected and portions of each tumor were processed for embedding in paraffin as well as embedding in OCT . Pieces of tumor were fixed in formalin for 24 hr at 4°C , then dehydrated and embedded in paraffin in the Pathology Core Facility at CCHMC . Deparaffined sections were then rehydrated and stained with antibodies or Hematoxylin and Eosin in the Pathology Core at CCHMC . Alternatively , using a protocol optimized to preserve YFP expression , pieces of tumor were fixed in 4% paraformaldehyde for 24 hr at 4°C , then washed thoroughly in 1x PBS and soaked in 30% sucrose at 4°C for 24 hr , then incubated in a slurry of 2:1 fresh 30% sucrose:OCT at 4°C for 24 hr , then embedded in OCT compound ( Tissue-Tek , Sakura , Torrance , CA ) and stored at −80°C as previously described ( McCauley and Guasch , 2013 ) . Deparaffined tissue sections ( 5 µm ) were subjected to antigen retrieval and immunostaining as previously described ( Tompkins et al . , 2009 ) . Frozen tissue sections ( 10 µm ) were subjected to immunofluorescence labeling as previously described ( Runck et al . , 2010 ) . Primary antibodies against the following proteins were used at the dilution indicated: green fluorescent protein , conjugated to Alexa Fluor 488 ( Invitrogen , 1/1 , 000 ) ; α6 integrin/CD49f ( BD Biosciences , 1/100 , RRID:AB_396079 ) , β1-integrin/CD29 ( Millipore , 1/100 , RRID:AB_2128202 ) , biotin conjugated to rat-anti-mouse CD34 ( eBioscience , 1/50 , RRID:AB_466426 ) , Keratin-5 ( Seven Hills Bioreagents , Rabbit 1/250 or Guinea Pig 1/5 , 000 ) , ELMO1 ( Abcam , 1/100 , for immunofluorescence , Sigma , 1/50 , RRID:AB_1848128 for immunohistochemistry ) , RAC1 ( Cell Signalling , 1/100 and Santa-Cruz , 1/50 , RRID:AB_2238100 ) , RAC2 ( Millipore , 1/100 , RRID:AB_2176134 ) , pSMAD2 ( Cell Signaling Technology , Danvers , MA , 1/100 , RRID:AB_331673 ) . 4’ , 6-diamidino-2-phenylindole ( DAPI ) was utilized as a marker of cell nuclei ( Sigma Chemical Co . , St . Louis , MO , 1/5 , 000 ) . Secondary antibodies conjugated to Alexa Fluor 488 or 540 or 649 ( Jackson ImmunoResearch , West Grove , PA ) were used at a dilution of 1/1 , 000 . For immunohistochemistry , slides were stained with the ABC kit ( Vector Laboratories , Burlingame , CA ) and counterstained with nuclear fast red ( Sigma Chemical Co . , St . Louis , MO , USA ) according to manufacturers’ instructions . Confocal images were acquired by capturing Z-series with 0 . 3 μm step size on a Zeiss LSM 880 laser scanning confocal microscope . Images in different focal planes were combined using the Zen software . Tumor cells were sorted by FACS and CD34− and CD34+ epithelial populations were collected in E media containing 0 . 05 mM calcium . Cells were centrifuged at 1000 RPM for 5 min at 4°C , resuspended , and plated at equal densities on a feeder layer of irradiated mouse embryonic fibroblasts ( MEFs ) in E media containing 0 . 3 mM calcium . MEFs were isolated from wild-type CD-1 mice at embryonic day 13 . 5 and cultured in DMEM containing 10% serum and 1% penicillin-streptomycin . After expanding MEFs , confluent plates were trypsinized with 0 . 05% Trypsin-EDTA ( Gibco ) and irradiated with 60Gy by the CCHMC Comprehensive Cancer Core Facility . Irradiated MEFs were replated at 100% confluency in DMEM containing 10% serum and 1% penicillin-streptomycin one day before sorting and plating tumor cells . On the day of the sort , the media on the irradiated MEFs was changed to E media containing 0 . 3 mM calcium . Clones begin to appear after 7–10 days of culture , and were passaged by transferring individual clones of cells on Whatman paper to a new plate on a feeder layer of irradiated MEFs . After the third passage , CD34+ SCC cells were grown on plastic without feeders in E media containing 0 . 05 mM calcium . These cells have been tested mycoplasma free . Wild-type anal keratinocytes were isolated from newborn C57BL/6 mice at postnatal day one by dissecting the anal canal , dissociating epidermis from dermis by incubating in dispase overnight at 4°C , extracting keratinocytes using 0 . 12% Trypsin-EDTA diluted in versene containing 0 . 1% glucose , and plating on a feeder layer of irradiated MEFs in E media containing 0 . 3 mM calcium as described above . Keratinocytes were passaged to a new feeder layer of irradiated MEFs in E media containing 0 . 3 mM calcium once confluent . After the third passage , wild-type anal keratinocytes were grown on plastic in E media containing 0 . 05 mM calcium . Proteins were detected by Western blotting as previously described ( McNairn et al . , 2013 ) . Briefly , cells were lysed and proteins were separated by SDS-PAGE , transferred to nitrocellulose membranes , and subjected to immunoblotting using antibodies to the following proteins at the dilutions indicated: p-Smad2 ( Cell Signaling Technology , 1/1 , 000 , RRID:AB_331673 ) , c-Myc ( Cell Signaling Technology , 1/1 , 000 , RRID:AB_2151827 ) , Smad2/3 ( BD Biosciences , 1/500 , RRID:AB_398161 ) , Keratin 8 ( NICHD Developmental Studies Hybridoma Bank maintained by the University of Iowa , 1/1000 , RRID:AB_531826 ) , β-actin ( Sigma , 1/2 , 000 , RRID:AB_476744 ) , RAC1 ( Cell Signaling , 1/2000 ) , ELMO1 ( AbCam , 1/2000 ) , α-Tubulin ( Sigma , 1/5000 , RRID:AB_477593 ) , GAPDH ( Santa Cruz , 1/5000 , RRID:AB_477593 ) . HRP-coupled secondary antibodies were used at 1/2 , 000 in 5% nonfat milk , and IRDye-conjugated secondary antibodies ( Li-COR , Lincoln , NE ) were used at 1/10 , 000 in 5% nonfat milk . Immunoblots were developed using standard ECL ( Amersham ) and Luminata TM crescendo and classico ( Millipore ) as previously described ( McNairn et al . , 2013 ) or the Odyssey CLx Infrared Imaging System ( Li-COR , Lincoln , NE ) . Total RNA was isolated using a Qiagen Rneasy Micro Kit and used to produce cDNA using the Maxima first strand cDNA synthesis kit ( Fermentas , San Jose , CA ) . Reverse transcription ( RT ) reactions were diluted to 10 ng/μl and 1 μl of each RT was used for real-time PCR . Real-time PCR was performed using the CFX96 real-time PCR System , CFX Manager Software and the SsoFast EvaGreen Supermix reagents ( Biorad , Hercules , CA ) or StepOne Plus real-time PCR system and the Power Sybr Green PCR Master Mix reagents ( Applied Biosystems , Grand Island , NY ) . All reactions were run in triplicate and analyzed using the ΔΔCT method with relative expression normalized to Gapdh . Primers used: GapdhFCGTAGACAAAATGGTGAAGGTCGGRAAGCAGTTGGTGGTGCAGGATGCD34FACCACAGACTTCCCCAACTGRCGGATTCCAGAGCATTTGATCtssFTGCTAGTTATTGCTCTTACCCAGRGTAACTACACATTGATCACGACACFbn1FATTGTTCACCGAGTCGATCTGRACGAGAAGCCTGAGAAAGTGSpp1FTGCACCCAGATCCTATAGCCRCTCCATCGTCATCATCATCGMmp9FTCCGTGTCCTGTAAATCTGCRCTTTTCCTAGCCCAGTCACTAAGTGFb2FTTTCTGCGTCAGTGTGAGTCRCTTTTCCTAGCCCAGTCACTAAGRac2FCACAGCCCACACGACAGRCACACGGAGAAACAGCAATTCRhohFCGATCACCTTTTCTACACCCTGRCATACAACCCCTCTACAGTGCRhojFATATGCTGGTGAGGTGTTGGRAAGACATGAACTAAGGCCACCVav1FCCATGAACTGTCCTCACCAGRCATCTCTGGGCTTTATCCTGGDock2FTCGGTGGAGAACTTTGTGAGRACGGTTGTCTTTGCTCTCATCElmo1FACTTTGGTCTCACTTGTAGCAGRCAGTGTGATAGAGGGATTGGTCElmo2FGATACTTCCCCTTGCCTCAGRGCTTCCTGAGACCTACAATGGTgfbr2FGCAAGTTTTGCGATGTGAGARTCCGTGTTGTGGTTGATGTTSnailFCTCCTACCCCTCAGTATTCATGRAGGGAGGTAGGGAAGTGGαSmaFGTGAAGAGGAAGACAGCACAGRGGGAGTAATGGTTGGAATGGGVimentinFATGGACAGGTGATCAATGAGACRCAGTAAAGGCACTTGAAAGCTGZeb2FGCAACATACTCTTTCTCCCCAGRTCTGAGCCTTCCTGTGAAAAG All genomic analysis was performed in GeneSpring NGS . Samples were sequenced using the HiSeq 2000 ( Illumina , CA ) with 50 bp , single-end reads . Following primer and barcode removal , sequences were aligned to the mm9 mouse genome using Ensembl transcripts . Following alignment , reads were quantified to generate computing reads per kilobase per million reads ( RPKM ) , then normalized using the DESeq algorithm and baselining to the median of all samples . We applied a filter , requiring at least 20 reads in at least one of the four samples . We generated a list of differentially regulated genes by comparing CD34-high samples to CD34-negative samples , with a fold change cutoff of 2 . 0 ( n = 896 entities ) . Entities were exported to ToppCluster in order to identify enrichment in previously published microarray datasets ( Coexpression Ontologies ) . A network consisting of genes and associated studies was generated through ToppCluster and Cytoscape . Chromatin immunoprecipitation was performed as previously described ( McCauley et al . , 2014 ) . Briefly , NIH3T3 cells were seeded in 10 cm plates at 80% confluence and transiently transfected with a pCMV-driven mouse SMAD3 ( Sino Biological Inc . , Daxing , China ) using X-treme Gene transfection reagent ( Roche Applied Science , Indianapolis , IN , USA ) for 24 hr , then treated with recombinant human TGFβ1 ( R & D Systems , Minneapolis , MN , USA , 2 ng/ml ) for an additional 24 hr . Cells were cross-linked with 1% formaldehyde and subjected to ChIP using an antibody against SMAD3 ( Abcam , Cambridge , MA , USA , RRID:AB_2192903 ) using a ChIP assay kit ( Millipore , Billerica , MA , USA ) according to manufacturer’s instructions . After purification , DNA obtained from the ChIP assay was used as PCR templates to verify the interaction between DNA and protein , using primers designed to amplify distinct sites in the mouse Elmo1 and Dock2 promoters . Primers are described below . PCR products were then subjected to gel electrophoresis on a 3% agarose gel using a molecular weight marker to verify the size of migrating bands . Primers used: Elmo1SBE FgctttcttcagtccctcataggaSBE RagcttccatttcagggaaactccTIE FgtgcaaccaggagaatttaaagcaTIE RtgagatgcccgaatccggagDock2SBE4 FcaacttgtgctgtcagaaactgaaSBE4 RtctcagggctaccatcacaatgSBE2/3 FcattgtgatggtagccctgagaSBE2/3 RttctgtgccatgaacccaactgSBE1 FgtcactaacagggttcagaagtcaSBE1 RatttggagaccaccctcatttgtcTIE FgtcactcactgagtacaggttcttTIE Rtgacttctgaaccctgttagtgac Using EcoRI and XhoI restriction enzymes , the full length Mus musculus Tgfbr2 gene ( 1 . 7 kb ) was isolated from a pcDNA3 expression vector ( Guasch et al . , 2007 ) , sequenced and subcloned into the multi-cloning site of the pLVX-IRES-mCherry lentiviral vector ( Clontech , Mountain View , CA ) using the NEB Quick Ligation Kit ( New England Biolabs , Ipswitch , MA ) , according to manufacturer’s instructions . The resulting ligation was transformed into DH5α competent bacteria and selected on LB-amp plates overnight . DNA was extracted using a Maxi Prep DNA kit ( Qiagen , Venlo , Limburg ) according to manufacturer’s instructions . Colonies were subjected to enzymatic digestion followed by sequencing to confirm the integration . Vector control ( pLVX-IRES-mCherry ) , rescue ( pLVX-TGFβRII-IRES-mCherry ) and hairpin resistant ELMO1 construct ( pLVX-IRES-mCherry-ELMO1* ) constructs were produced by the Cincinnati Children’s Lentiviral Core and the lentivectors production facility/SFR BioSciences Gerland – Lyon Sud . 2 mL of viral supernatant were used for each 10 cm plate , after being washed in E Media with 0 . 05 mM Ca++ and concentrated by three centrifugations at 4000 rpm for 15 min at 4°C using a Vivaspin 20MWCO 30 kDa column ( GE Healthcare , Pittsburgh , PA ) . Concentrated virus was combined with 8 µg/ml polybrene ( Hexadimethrine bromide , Sigma . St . Louis , MO ) and 3 ml E Media with 0 . 05 mM Ca++ , applied to Tgfbr2 cKO CD34+ SCC cells seeded at 60% confluency in 10 cm plates , and incubated at 37°C 5% CO2 for 24 hr . 24 hr after infection , plates were washed three times with sterile 1X PBS and given fresh E Media with 0 . 05 mM Ca++ . 48 hr after infection , YFP+mCherry+ cells were selected by FACS and used directly for in vivo orthotopic transplantation or re-plated for in vitro experiments . To confirm the rescue , cells were treated with recombinant human TGFβ1 ( R and D Systems , Minneapolis , MN , USA , 2 ng/ml ) for 1 hr at 37°C before cells were trypsinized and processed for RNA extraction for qPCR or lysed for protein extraction and Western blot . Experiments were performed three times in triplicate and statistical significance was determined using paired two-tailed Student’s t-test . Two MISSION shRNA pLKO . 1-puro bacterial constructs against the mouse Elmo1 gene ( TRCN0000112655 , TRCN0000112656 ) and the SH02 control shRNA were purchased from Sigma Aldrich via the Cincinnati Children’s Robotic Lenti-Library Core and lentivirus for SH02 control , TRCN0000112655 ( ELMO1 shRNA construct #1 ) and TRCN0000112656 ( ELMO1 shRNA construct #2 ) was produced by the Cincinnati Children’s Lentiviral Core and the lentivectors production facility/SFR BioSciences Gerland – Lyon Sud . 2 mL of viral supernatant were used for each 10 cm plate , after being washed in E Media with 0 . 05 mM Ca++ and concentrated by three centrifugations at 4000 rpm for 15 min at 4°C using a Vivaspin 20MWCO 30 kDa column ( GE Healthcare , Pittsburgh , PA ) . Concentrated virus was combined with 8 µg/ml polybrene ( Hexadimethrine bromide , Sigma ) and 3 ml E Media with 0 . 05 mM Ca++ , applied to Tgfbr2 cKO CD34+ SCC cells seeded at 60% confluency in 10 cm plates , and incubated at 37°C 5% CO2 for 24 hr . 24 hr after infection , plates were washed three times with sterile 1X PBS and given fresh E Media with 0 . 05 mM Ca++ . Beginning 48 hr after infection , 1 µg/ml puromycin was added with fresh E Media with 0 . 05 mM Ca++ every other day to select for infected puromycin-resistant clones . To create a construct that is not recognized by the Elmo1 shRNA construct #2 ( TRCN0000112656 ) , we created three bases mutations in its target sequence without affecting ELMO1 function . The Geneart site-direct mutagenesis kit ( Thermofisher Scientific ) was used on pCMV-Sport6 expressing the full-length Mus musculus Elmo1 gene ( 2 , 2 Kb ) ( GE Dharmacon ) , sequenced and subcloned into the multicloning site of the pLVX-IRES-mcherry lentiviral vector ( clonetech , Mountain View , CA ) using the NEB Quick Ligation Kit ( New England Biolaps , Ipswitch , MA ) . Lentiviral vectors were produced by SFR Biosciences Gerland Lyon Sud . Primers used to create the mutagenesis: ELMO1 mut F AACTTGCTTTCTCCATCTTGTATGATTCAAATTGCCAACTGAACT R AGTTCAGTTGGCAATTTGAATCATACAAGATGGAGAAAGCAAGTT Lungs from tumor-bearing mice were dissociated into a single cell suspension according to a modified protocol based on one previously established in our lab ( McCauley and Guasch , 2013 ) . Briefly , lungs were inflated with a cocktail containing dispase ( Sigma , St . Louis , MO ) , 20% collagenase ( Sigma , St . Louis , MO ) and Hank’s buffered saline solution ( Gibco , Waltham , MA ) and dissociated in the same cocktail for 30 min while shaking at 37°C . Dissociated lung tissue was washed and filtered according to our tumor dissociation protocol ( McCauley and Guasch , 2013 ) , with the added treatment of red blood cell lysis buffer according to manufacturer’s instructions ( eBioscience , San Diego , CA ) . Whole lungs were sorted for YFP and analyzed using a FACS Aria II ( BD Biosciences ) and FACSDiva software ( BD Biosciences ) in the Research Flow Cytometry Core at CCHMC . Percent YFP+ cells was calculated by dividing number of YFP+ cells by total cells after gating for appropriately sized single cells by forward scatter and side scatter . Experiments were performed twice in triplicate and statistical significance was determined using paired two-tailed Student’s t-test . The wound-healing assay was used to determine cell migration ability . 5 × 104 cells were plated in ibidi culture insert onto the 24-well plate ( ibidi cat-80241 ) for 24 hr to reach 90–95% of confluence . A wound was created by removing inserts . Cells were washed with PBS twice to remove detached cells and incubated with medium E low Ca2+ containing puromycin ( 1 µg/ml ) . The cells were observed under an inverted light microscope ( Carl Zeiss ) equipped with a CCD camera ( Ropper ) at X10 objective . Images were taken by MetaMorph software every 10 min for 10 hr . The wound widths of different area at each time points were measured with MetaMorph software . Data result from calculating the slope of linear trend curve of wound widths as a function of time and are representative of three experiments . Quantitative data are presented as the mean and significant difference was determined by two-tailed Student’s t-test . Cell culture dishes at 80% confluence were washed with ice-cold PBS1X , lysed with 500 µl of lysis buffer ( 50 mM Tris , pH 7 , 2 , 350 mM NaCl , 1% Triton X-100 , 0 . 5% Na deoxycholate , 0 . 1% SDS , 10 mM MgCl2 , protease inhibitor cocktail complete tablets ( Roche ) ) and centrifuged for 5 min at 13 , 000 RPM at 4°C . The supernatant was incubated with bacterially produced glutathione-S-transferase ( GST ) -PAK-CD fusion protein , containing the RAC and Cdc42-binding region from human PAK1B ( Sander et al . , 1998 ) bound to glutathione-coupled Sepharose beads at 4°C for 45 min . The beads and proteins bound to the fusion protein were washed three times with a wash buffer ( 50 mM Tris , pH7 . 2 , 1% Triton X-100 , 150 mM NaCl , 20 mM MgCl2 , protease inhibitor cocktail complete tablets ( Roche ) ) and eluted in Laemmli sample buffer ( 60 mM Tris pH 6 , 8 , 2% SDS , 10% glycerine , 0 , 1% bromophenol blue ) and then analysed for bound RAC1 molecules by western blot using a RAC1 antibody . Cells were stained with 10 μM Cell Proliferation dye eFluor 670 ( Affymetrix eBioscience ) according to the manufacturer’s guidelines and analyzed by flow cytometry at 0 hr , 24 hr , 48 hr and 72 hr with a Fortessa instrument ( 3 lasers 405/488/630 ) ( Becton Dickinson ) . DAPI staining was used and excluded to acquire 104 cells in the live population . Three separate experiments have been done and the mean of the geometric mean for each sample has been calculated using Prism software ( RRID:SCR_002798 ) . | Many different types of cells make up the tissues and organs throughout our bodies . There are locations throughout the body where two different types of cells meet – called transition zones – and these regions are susceptible to cancer formation . Many of these tumors are particularly aggressive , including those that arise in the transition zone in the cervix , the junction between the esophagus and the stomach , and the transition zone between the anus and rectum . Aggressive tumors such as these frequently spread and form tumors in other organs , such as the lung , in a process called metastasis . We still lack a clear understanding of what makes transition zones prone to forming tumors or why the tumors that form are so aggressive . However , we do know about some differences between these tumor cells and healthy cells . For example , in healthy cells , the “transforming growth factor beta” ( TGFβ ) signaling pathway , is crucial for regulating many different processes , including cell growth . By contrast , many of the cells in aggressive transition zone tumors are unable to correctly regulate TGFβ signaling . Mice that have been genetically engineered so that their cells are deficient in TGFβ signaling spontaneously develop aggressive transition zone tumors . By studying these mice , McCauley et al . found that only a small fraction of the tumor cells are responsible for the growth of the tumor . These cells express genes that enhance their ability to migrate and invade , including one called Elmo1 . When McCauley et al . blocked the activity of this gene the aggressive tumor cells lost their ability to metastasize to the lung . This is a new link in understanding how a particular genetic mutation or the inability to regulate a cell signaling pathway , such as TGFβ , can drive tumor growth and metastasis . Based on this knowledge , it may be worth investigating whether blocking the activity of ELMO1 could help to prevent metastasis from transition zone tumors . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"cancer",
"biology"
] | 2017 | De-repression of the RAC activator ELMO1 in cancer stem cells drives progression of TGFβ-deficient squamous cell carcinoma from transition zones |
HIV-1 replication commences inside the cone-shaped viral capsid , but timing , localization , and mechanism of uncoating are under debate . We adapted a strategy to visualize individual reverse-transcribed HIV-1 cDNA molecules and their association with viral and cellular proteins using fluorescence and correlative-light-and-electron-microscopy ( CLEM ) . We specifically detected HIV-1 cDNA inside nuclei , but not in the cytoplasm . Nuclear cDNA initially co-localized with a fluorescent integrase fusion ( IN-FP ) and the viral CA ( capsid ) protein , but cDNA-punctae separated from IN-FP/CA over time . This phenotype was conserved in primary HIV-1 target cells , with nuclear HIV-1 complexes exhibiting strong CA-signals in all cell types . CLEM revealed cone-shaped HIV-1 capsid-like structures and apparently broken capsid-remnants at the position of IN-FP signals and elongated chromatin-like structures in the position of viral cDNA punctae lacking IN-FP . Our data argue for nuclear uncoating by physical disruption rather than cooperative disassembly of the CA-lattice , followed by physical separation from the pre-integration complex .
Retroviral replication involves reverse transcription of the viral RNA genome and requires nuclear entry of the subviral complex to allow for chromosomal integration of the viral cDNA mediated by the viral integrase ( IN ) ( Lusic and Siliciano , 2017 ) . Although the gammaretrovirus murine leukemia virus requires nuclear envelope breakdown during mitosis for productive replication , HIV-1 and other lentiviruses infect non-dividing cells , implying that the subviral complex can pass through the intact nuclear envelope ( Suzuki and Craigie , 2007 ) . Reverse transcription mediated by the viral reverse transcriptase ( RT ) is initiated in the cytoplasm , but recent evidence indicates that cDNA synthesis is completed inside the nucleus ( Burdick et al . , 2020; Dharan et al . , 2020; Selyutina et al . , 2020 ) , at least in the case of HIV-1 . The cytoplasm is a hostile environment for retroviral genome replication: exposure of cytoplasmic DNA to cellular nucleic acid sensors would lead to induction of innate immunity ( Doitsh et al . , 2014; Monroe et al . , 2014 ) , thereby aborting viral infection . The viral cone-shaped capsid apparently plays a central role in guiding and shielding ( Rasaiyaah et al . , 2013; Sumner et al . , 2020 ) the genome through the cytosolic environment ( Campbell and Hope , 2015; Novikova et al . , 2019 ) . It consists of ~1200–1500 CA molecules assembled into hexamers and pentamers ( Briggs et al . , 2003 ) , which have been shown to interact with components of the nuclear pore complex ( NPC ) ( Bhattacharya et al . , 2014; Matreyek et al . , 2013; Price et al . , 2014 ) , implying a role for the CA-lattice in nuclear entry ( Di Nunzio et al . , 2013; Lelek et al . , 2015; Matreyek et al . , 2013 ) . However , the HIV-1 capsid , with a length of ~120 nm and a width of ~60 nm at its wide end ( Mattei et al . , 2016 ) , is presumed to exceed the dimensions of the NPC channel with a reported maximal diameter of ~40 nm ( von Appen et al . , 2015 ) . This implies that capsid uncoating should occur – at least partially – prior to nuclear entry , and various publications reported uncoating either in the cytoplasm ( Cosnefroy et al . , 2016; Hulme et al . , 2011; Mamede et al . , 2017; Xu et al . , 2013 ) or at the nuclear pore ( Arhel et al . , 2007; Burdick et al . , 2017; Francis and Melikyan , 2018 ) , with some evidence for cell-type-dependent differences . On the other hand , nuclear HIV-1 pre-integration complexes ( PIC ) were found to retain varying amounts of CA molecules ( Bejarano et al . , 2019; Burdick et al . , 2020; Chin et al . , 2015; Hulme et al . , 2015; Stultz et al . , 2017; Zila et al . , 2019 ) , at least in certain cell types ( Zila et al . , 2019 ) , and recent reports indicating the presence of intact capsid lattice ( Dharan et al . , 2020; Selyutina et al . , 2020 ) and capsid-like structures ( Zila et al . , 2021 ) inside the nucleus challenged the current models of early HIV-1 replication . Accordingly , the timing , subcellular localization , trigger and mechanism of HIV-1 capsid uncoating are still under debate . Studying early HIV-1 replication is hampered by the fact that most cytoplasmic entry events appear to be non-productive in tissue culture ( Klasse , 2015; Sanjuán , 2018 ) . Therefore , characterization of individual subviral complexes containing viral cDNA with respect to their content , subcellular distribution and trafficking is required to shed light on the pathway of productive replication . Viral cDNA can be visualized in fixed cells using fluorescence in situ hybridization ( FISH ) ( Marini et al . , 2015 ) or its derivatives using branched probes ( Chin et al . , 2015; Puray-Chavez et al . , 2017 ) , but the harsh assay conditions destroy the native cellular environment and impair immunofluorescence analysis . Incorporation of the modified nucleoside 5-ethynyl-2´-deoxyuridine ( EdU ) allowed the detection of actively reverse transcribing HIV-1 complexes by visualizing de novo synthesized viral DNA via click chemistry ( Peng et al . , 2015; Stultz et al . , 2017 ) , but this approach is also limited to fixed cells and cellular extraction precludes high-resolution analysis ( Müller et al . , 2019 ) . To overcome these limitations , we adapted a live cell compatible genetically encoded system ( ANCHOR ) that allows single-molecule gene labeling ( Germier et al . , 2017; Saad et al . , 2014 ) , and has been applied for visualization of DNA from Adenovirus ( Komatsu et al . , 2018 ) , Cytomegalovirus ( Mariamé et al . , 2018 ) , and HIV-1 ( Blanco-Rodriguez et al . , 2020 ) . This system is based on the prokaryotic chromosomal partitioning system ParB-parS , where ParB ( designated OR ) specifically binds the parS seed sequence ( designated ANCH ) . Multiple copies of parS introduced into the HIV-1 genome act as nucleation sites to oligomerize the fluorescently labeled OR protein when the reverse transcribed ANCH cDNA sequence becomes accessible to the fusion protein . Here , we show that HIV-1 cDNA containing subviral complexes associated with CA are detected in the nucleus of infected cells , including primary CD4+ T cells . Over time , cDNA containing complexes segregate from CA and the bulk of viral replication proteins , confirming nuclear uncoating . Using 3D correlative light and electron microscopy ( CLEM ) , we detected capsid-like structures at the position of nuclear IN-FP-containing complexes , whereas elongated , chromatin-like densities were observed at the position of viral cDNA punctae . Importantly , strong CA signals were observed on nuclear HIV-1 complexes in all cell types analyzed , indicating that prior failure to detect nuclear CA by immunofluorescence was largely due to masked epitopes .
To test for retention of the ANCH sequence and efficiency of visualizing HIV-1 cDNA following reverse transcription , we stably transduced HeLa-based TZM-bl cells with different amounts of an ANCH3 containing HIV-1-based vector . These cell populations were subsequently transduced or transfected with an expression vector for eGFP . OR3 ( Figure 1—figure supplement 1a ) . Confocal microscopy revealed distinct eGFP . OR3 punctae in the nuclei of >90% of transfected cells ( Figure 1—figure supplement 1b–d ) . At low multiplicities of transduction with the ANCH-vector , where the majority of cells is expected to originate from a single integration event , we observed an average of 1 . 6 ± 0 . 29 and 1 . 4 ± 0 . 34 eGFP . OR3 punctae per nucleus ( Figure 1—figure supplement 1d ) . The number of punctae correlated with the multiplicity of transduction over a wide range ( Figure 1—figure supplement 1d ) . Of note , the eGFP . OR3 signal was stable for more than 4 weeks ( when unintegrated viral cDNA species are expected to be degraded ) suggesting that integrated viral DNA can be detected . Thus , the ANCH sequence is retained during reverse transcription , and this approach can be used to detect HIV-1 cDNA during the early replication phase . Next , we introduced the ~1 , 000 bp ANCH3 sequence into the HIV-1 proviral plasmid pNLC4-3 ( Bohne and Kräusslich , 2004 ) ( HIV ANCH ) replacing part of the env gene ( Figure 1a ) . Virus-like particles were pseudotyped with the vesicular stomatitis virus G protein ( VSV-G ) or HIV-1 Env as indicated . They also contained exogenously expressed IN tagged with a fluorescent marker ( IN-FP ) ( Albanese et al . , 2008 ) , in addition to wild-type IN encoded by the virus , for visualization of subviral replication complexes . These particles were used to infect polyclonal TZM-bl cell populations stably transduced to express OR3 fused with either eGFP , mScarlet ( Bindels et al . , 2017 ) or the stainable SNAP-tag ( Keppler et al . , 2003 ) ; these cells also stably expressed fluorescently tagged Lamin B1 ( LMNB1 ) to clearly distinguish nuclear and cytoplasmic events . Figure 1b shows TZM-bl eBFP2 . LMNB1 and eGFP . OR3 expressing cells infected with HIV ANCH . Distinct infection-induced eGFP punctae were clearly detected in the nuclei of these cells , but were not observed in the cytoplasm where eGFP . OR3 was diffusely distributed ( Figure 1b ) . Distinct nuclear eGFP punctae were not detected in uninfected cells ( Figure 1—figure supplement 2a ) . To determine whether HIV ANCH cDNA became chromosomally integrated , we infected the human SupT1 T-cell line with HIV ANCH and analyzed the copy number of integrated proviral genomes by semi-quantitative Alu-PCR; this experiment could not be performed in TZM-bl cells , since these cells carry multiple HIV-1 LTR copies from prior lentiviral vector transductions . Integrated proviral DNA was readily detected in HIV ANCH infected SupT1 cells , but was not observed when infection was performed in the presence of an RT- or IN-inhibitor ( Figure 1c ) . Similar to TZM-bl cells , SupT1 cells also showed nuclear eGFP . OR3 punctae following infection with HIV ANCH ( see below , Figure 7—figure supplement 1a ) . We then determined the number of eGFP . OR3 punctae in TZM-bl cells using confocal microscopy of cells fixed at different time points after infection with HIV ANCH ( Figure 1d ) and performed parallel quantitation of total HIV-1 cDNA and 2-LTR circles ( representing unintegrated nuclear HIV-1 cDNA ) using digital droplet PCR ( ddPCR ) ( Figure 1e ) . Total cDNA levels became saturated at 10 hr post infection ( h p . i . ) and 2-LTR circles peaked at 24 h p . i . ; both species strongly declined over the following 5 days ( Figure 1e ) . eGFP punctae , on the other hand , increased over the first 72 hr , but then remained stable over the following 3 days despite the observed loss of HIV-1 cDNA species ( Figure 1d ) . Taken together , these results clearly indicated that integrated HIV-1 DNA can be detected using the ANCHOR system . To determine whether detection of integrated proviral copies may be influenced by RNA transcription at the respective site , we compared the number of eGFP . OR3 punctae in HIV ANCH infected TZM-bl cells treated with the CDK9/p-TEFb inhibitor Flavopiridol or solvent control . No difference was observed ( Figure 1—figure supplement 3a–c ) , suggesting that the dynamic nature of OR3 recruitment does not interfere with transcription . Next , we generated a non-infectious derivative of HIV ANCH termed NNHIV ANCH to allow for live cell imaging outside the BSL3 facility . NNHIV ANCH is based on the previously reported plasmid NNHIV that carries point mutations in the active site of IN and a deletion in the tat gene ( IND64N/D116N tatΔ33-64bp ) ( Zila et al . , 2021 ) . This derivative retains reverse transcription and nuclear import ability , whereas integration and transcription are blocked . TZM-bl eGFP . OR3 cells infected with VSV-G pseudotyped NNHIV ANCH also showed nuclear eGFP . OR3 punctae ( Figure 1f–h ) indicating that unintegrated HIV-1 cDNA is detected by the ANCHOR system as well . eGFP . OR3 punctae were not detected when cells were treated with NNHIV ANCH lacking VSV-G and were absent or strongly reduced in the presence of the RT inhibitors efavirenz ( EFV ) or azidothymidine ( AZT ) ( Figure 1f and Figure 1—figure supplement 4 ) . To investigate the dynamics of appearance of eGFP . OR3 punctae in NNHIV ANCH infected TZM-bl cells , we performed live cell imaging experiments using spinning disk confocal microscopy ( SDCM ) . The onset of marker recruitment to viral cDNA in the nucleus was observed at 7–8 h p . i . , while the half-maximal signal was reached between 13 and 15 h p . i . ( Figure 1g , h; Figure 1—video 1 ) . Again , no infection-induced eGFP . OR3 punctae were detected in the cytosol of infected cells . The onset of nuclear HIV-1 2-LTR detection using ddPCR coincided with the appearance of eGFP . OR3 punctae ( Figure 1h ) . Formation of both 2-LTR circles ( dependent on nuclear NHEJ components and ligase IV [Li et al . , 2001] ) and eGFP . OR3 punctae requires viral cDNA to be synthesized and accessible to proteins not present in the subviral replication complex . Accordingly , the lack of cytoplasmic eGFP . OR3 punctae may be due to incomplete cDNA synthesis in the cytoplasm and/or to shielding of the cDNA from the fusion protein until full capsid uncoating in the nucleus . To address this question , we focused on the timing and quantification of reverse transcription in the described system . Quantitation of reverse transcription products was performed using FISH analysis . A large proportion ( ~80% ) of RT products detected by FISH was localized in the cytoplasm of infected cells , whereas no eGFP . OR3 punctae were observed in this compartment ( Figure 2a; note that the harsh sample treatment required for FISH detection did not preserve eGFP . OR3 punctae , precluding analysis of both markers in the same specimen ) . The number of nuclear FISH signals per cell ( ~14 ) ( Figure 2a , right ) was similar to the number of nuclear IN . SNAP signals ( ~12 ) ( Figure 2b ) . A similar distribution of late RT products was also observed by ddPCR analysis under the same infection conditions: ~130 gag reverse transcripts and 2–3 2-LTR circles were detected per cell ( Figure 2c ) ; the slightly lower absolute number of DNA molecules detected by FISH ( ~80/cell ) is likely due to the high density of signals , with some diffraction limited punctae representing more than one reverse transcript . In contrast , only ~4 nuclear eGFP . OR3 clusters were detected per cell ( Figure 1f ) . These results clearly showed that the majority of late RT products – including all cytoplasmic products – were not associated with eGFP . OR3 . Progress of reverse transcription was also assessed at the single particle level . For this , we infected cells with NNHIV ANCH carrying fluorescently tagged IN-FP in the presence of EdU followed by fluorescent click labeling of newly synthesized DNA at different time points post infection . Cellular DNA synthesis was inhibited by the DNA polymerase α/δ inhibitor aphidicolin ( APC ) ( Figure 2d–f , Figure 2—figure supplement 1 ) . Co-localization with EdU was observed for 7% of IN-FP-positive structures in the cytoplasm ( 95 % CI of mean: 4–9% ) and for 36% of IN-FP-positive complexes in the nucleus ( 95 % CI of mean: 29–43% ) . Importantly , 81% of nuclear eGFP . OR3 punctae were positive for EdU ( 95 % CI of mean: 70–93%; Figure 2d , e ) . While some nuclear EdU-positive complexes were positive for both the IN-FP and eGFP . OR3 , we also observed HIV-1 cDNA containing complexes that were only positive for either IN-FP or eGFP . OR3 ( Figure 2d , panels i and ii ) . The average EdU signal , as a correlate for reverse transcript length , was lower on cytoplasmic than on nuclear HIV-1 complexes and was highest on eGFP . OR3 punctae . When setting the EdU signal for eGFP . OR3 punctae to 100% , the relative EdU signal was significantly reduced to 23% ( p<0 . 0001 ) on cytoplasmic IN-FP-positive complexes and to 52% ( p<0 . 0001 ) on nuclear IN-FP-positive complexes ( Figure 2f ) . These observations support recent reports that reverse transcription is completed inside the nucleus ( Burdick et al . , 2020; Dharan et al . , 2020; Francis et al . , 2020; Selyutina et al . , 2020 ) and indicate that the viral cDNA only becomes detectable to eGFP . OR3 when reverse transcription has been completed . The low degree of co-localization between the fluorescent IN fusion protein and eGFP . OR3 on EdU-positive nuclear punctae at 24 h p . i . ( Figure 2d ) prompted us to analyze the relative distribution of both fluorescent proteins in a time resolved manner after NNHIV ANCH infection . At 8 h p . i . , 70 ± 11% of nuclear eGFP . OR3 punctae were also positive for IN . SNAP ( Figure 3a–c ) . This co-localization was largely lost at 24 h p . i . with only 14 ± 6% of nuclear eGFP . OR3 punctae positive for IN . SNAP ( Figure 3a–c ) . Strikingly , IN . SNAP punctae were often observed in close vicinity of eGFP . OR3 punctae at this later time point ( Figure 3a , right panel ) , suggesting that they may have separated from a common complex . Similar results were observed for HIV-1 ANCH ( Figure 3—figure supplement 1a ) as well as for an integration competent lentiviral vector containing ANCH ( Figure 3—figure supplement 1b ) , and when particles were pseudotyped with HIV-1 Env instead of VSV-G ( Figure 3—figure supplement 1c ) . These results showed that the observed phenotype was not dependent on the cytosolic entry pathway or on integration competence . To directly address the possibility of separation of the proviral cDNA from IN . SNAP , we performed live cell imaging of infected cells . We observed gradual loss of the IN . SNAP signal correlating with increased eGFP . OR3 recruitment and eventual separation of eGFP . OR3 punctae and IN . SNAP containing complexes ( Figure 3d , e , Figure 3—video 1 ) . Of note , we occasionally observed consecutive appearance of two individual eGFP . OR3 punctae and their subsequent separation from the same IN . SNAP complex ( Figure 3—figure supplement 2 , Figure 3—video 2 ) , suggesting that single diffraction limited IN . SNAP punctae may correspond to multiple cDNA containing subviral HIV-1 complexes , indicating clustering of nuclear complexes as has been recently noted by others ( Francis et al . , 2020; Rensen et al . , 2021 ) . The observation that the IN . SNAP signal remained as a distinct cluster after separation of the eGFP . OR3-associated viral cDNA suggested that these clusters constitute a stable complex potentially containing other viral and cellular proteins held together by a scaffold . The viral capsid or a capsid-derived structure would be an obvious candidate for such a scaffold . Earlier studies reported a wide range of CA amounts at nuclear complexes ( Chen et al . , 2016; Chin et al . , 2015; Francis et al . , 2020; Hulme et al . , 2015; Peng et al . , 2015; Zhou et al . , 2011 ) , while a recent study showed strong nuclear signals for an eGFP . CA fusion protein ( Burdick et al . , 2020 ) in HeLa-derived cells . We decided to revisit this issue , since we and others had previously observed that CA immunostaining efficiency in the nucleus strongly depended on sample treatment conditions ( Chin et al . , 2015; Zila et al . , 2019 ) . While PFA fixation alone did not result in CA immunostaining of nuclear complexes ( Figure 4a ) , methanol extraction ( following PFA fixation ) of parallel samples consistently yielded clear CA-specific signals co-localizing with most IN-FP-positive punctae inside the nucleus of TZM-bl cells ( Figure 4b–c and Figure 4—figure supplement 1a ) . Furthermore , IN-FP-positive punctae were also strongly positive for the host protein cleavage and polyadenylation specificity factor 6 ( CPSF6 ) that binds specifically to the hexameric CA lattice ( Bhattacharya et al . , 2014; Price et al . , 2014; Figure 4d ) and is involved in nuclear import ( Bejarano et al . , 2019; Burdick et al . , 2020; Chin et al . , 2015 ) , trafficking to nuclear speckles ( Francis et al . , 2020; Rensen et al . , 2021 ) and integration-site targeting ( Achuthan et al . , 2018; Francis et al . , 2020; Sowd et al . , 2016 ) . We reasoned , that this dense coat of CPSF6 might mask the underlying capsid lattice , precluding antibody detection . This hypothesis was tested by treating infected cells - after the subviral complexes had entered the nucleus – with the small molecule PF74 , which acts as a competitive inhibitor of CA-CPSF6 interaction ( Price et al . , 2014 ) and may thus displace CPSF6 from the subviral complex . TZM-bl cells infected with NNHIV ANCH for 24 hr were treated with different concentrations of PF74 , followed by fixation and IF staining ( Figure 4d–g ) . We observed a dose-dependent loss of CPSF6 signal from nuclear HIV-1 subviral complexes upon PF74 treatment , while IN-FP punctae stayed intact ( Figure 4e and g , left panel ) . PF74-mediated removal of CPSF6 from the nuclear subviral complexes revealed a strong CA IF signal ( Figure 4f ) , with inverse correlation of CPSF6 and CA signal intensities ( Figure 4g , right panel ) . CPSF6 displacement occurred rapidly , within 30 min of PF74 treatment ( Figure 4h ) . Thus , efficient detection of nuclear CA signals in HeLa-derived cells can be achieved by either general extraction using methanol , or by specific displacement of CPSF6 from subviral complexes . CA immunostaining intensities of particles outside of the nucleus were not affected by PF74 treatment ( Figure 4—figure supplement 1b ) , suggesting that the enhanced CA signal on nuclear complexes was not due to structural changes in the capsid lattice induced by the drug . This conclusion was further supported by using a monoclonal antibody with a defined epitope close to the CPSF6 binding interface ( Gorny and Zolla-Pazner , 1991 ) . This epitope was readily accessible and not affected by PF74 treatment on cytosolic capsids , whereas CA detection on nuclear subviral complexes by this antibody again required CPSF6 displacement ( Figure 4j , k ) . We conclude that CPSF6 shields CA from antibody detection inside the nucleus and CA signal intensity on nuclear complexes is comparable to that on cytoplasmic complexes in HeLa-derived cells . In the previous experiment ( Figure 4c ) , TZM-bl cells had been treated with APC to block cell division , suggesting that nuclear entry of viral complexes in these HeLa-derived cells had occurred through the intact NPC , as observed in terminally differentiated macrophages ( Bejarano et al . , 2018; Stultz et al . , 2017 ) . In order to prove that nuclear HIV-1 complexes had entered through intact nuclear pores and did not enter during nuclear envelope breakdown in cells that had escaped cell cycle arrest under our conditions ( Figure 4—figure supplement 1e , f ) , we observed eBFP2 . LMNB1 expressing TZM-bl cells over the whole 12 hr time course of infection by live imaging ( Figure 4l and m , Figure 4—video 1 ) before PF74 treatment and IF staining . We clearly detected HIV-1 complexes in the nuclei of cells that had not undergone division during the observation period . These nuclear complexes displayed similar CA intensities compared to extranuclear particles ( Figure 4n and o ) , supporting the conclusion that subviral complexes comprising most or all of the CA complement of the viral capsid can pass through intact nuclear pores in HeLa-derived cells . To further analyze the ultrastructure of these nuclear CA-containing complexes , we employed stimulated emission depletion ( STED ) nanoscopy and CLEM in combination with electron tomography ( ET ) . The mature HIV-1 capsid contains ca . 50% of the total CA content of the intact virion and post-fusion cytoplasmic capsids therefore exhibit a weaker CA signal than complete virions ( Briggs et al . , 2004; Zila et al . , 2019 ) . Accordingly , CA specific immunofluorescence would be expected to be lower for nuclear complexes compared to cell-associated particles that represent a mixture of post-fusion particles and complete particles at the plasma membrane or endocytosed in the cytosolic region . However , the observed intensity of the CA signal on nuclear HIV-1 complexes was equal to the intensities observed for cell-associated particles ( Figure 4b ) . This may be explained by clustering of capsid-derived structures in a single diffraction limited spot . In order to investigate this possibility , we used STED nanoscopy to resolve individual subviral structures with a resolution of <50 nm . Multiple individual CA signals in close vicinity to each other could be resolved within the area of a single focus detected in confocal mode ( Figure 5a and b ) . For a more detailed analysis of the associated structures , we performed CLEM-ET as described in the following section . IN . SNAP-positive and eGFP . OR3-negative nuclear punctae detected by fluorescence microscopy could be correlated with single or multiple electron-dense cone-shaped structures , whose shapes and dimensions closely resembled mature HIV-1 capsids ( Figure 5c , Figure 5—video 1 ) . STED nanoscopy of nuclear SNAP . OR3-positive punctae corresponding to viral cDNA identified elongated structures ( Figure 5d ) ; in this case , only a single object was resolved by STED nanoscopy at each diffraction-limited position . The findings described above suggest that apparently intact conical HIV-1 capsids can access the nucleoplasm , where reverse transcription is completed followed by separation of the viral cDNA from the bulk of viral proteins including CA . To determine the ultrastructure of the observed subviral complexes , we performed CLEM-ET analysis . For this , we employed a TZM-bl mScarlet . OR3 cell line ( Figure 3—figure supplement 1d ) , because the mScarlet fluorescence signal was best retained upon plastic embedding . Cells were infected with VSV-G pseudotyped NNHIV ANCH carrying IN . SNAP . SiR , thus allowing direct high-pressure freezing of the sample without pre-fixation . Cells were vitrified at 24 h p . i . and thin sections were prepared after freeze-substitution and plastic embedding . Samples retained fluorescence for mScarlet . OR3 and IN . SNAP . SiR . Multi-channel fluorescent Tetraspeck markers were used for correlation and we identified positions corresponding to mScarlet . OR3 and IN . SNAP . SiR signals , respectively ( Figure 6a–c ) . These positions were imaged using electron tomography ( Figure 6d–i ) . A total of 21 individual structures were identified and visualized by CLEM-ET covering the volume of six individual nuclei ( Figure 6—figure supplement 1 ) . The majority of structures ( 20/21; 95% ) was found in clusters of two or more structures . At positions correlating to IN-positive punctae lacking mScarlet . OR3 ( 18/21; 86% ) , we detected 10 cone-shaped structures resembling intact HIV-1 capsids ( Figures 5c and 6g ii . and 6i . ) and eight structures with less defined morphology consistent with deformed tubular structures or remnants of capsids ( Figure 6g ii and 6h ii; Figure 6—video 1 ) . Dense material most likely corresponding to the viral nucleoprotein complex was visible inside of most conical structures ( Figures 5c and 6g ii . and 6i top left panel , black arrowheads ) , whereas tubular and capsid remnant-like structures mostly lacked interior densities ( Figure 6g ii . , 6h ii . and 6i bottom left panel , open white arrowheads ) . In contrast , electron tomograms that correlated to positions of mScarlet . OR3 signals ( lacking IN . SNAP . SiR signal; n = 7 ) showed no defined conical or capsid remnant-like structures . Instead , elongated dense objects of ~100–300 nm in length were observed ( Figure 6g i . and 6h i . ) , in line with the findings from STED nanoscopy ( compare Figure 5d ) . These structures consisted of linked globular densities with a diameter of ~30 nm resembling the appearance of chromatin ( Figure 6h i . ) . One of the visualized objects correlated to both IN . SNAP . SiR and mScarlet . OR3 . The corresponding electron tomogram revealed a dense cluster of three capsid-related structures ( Figure 6i ) . One of these structures lacked interior density ( Figure 6i , bottom left panel , open arrowhead ) and appeared to be connected with an adjacent elongated density ( filled white arrowheads ) that seemed to protrude from the narrow end of the cone ( Figure 6i right panel; Figure 6—video 2 ) . Taken together with the observations from live cell imaging , we speculate that this structure might represent a subviral complex captured in the process of capsid uncoating and genome release . In order to validate our findings in more relevant cell types , we adapted the system to the T cell line SupT1 and to primary CD4+ T cells and primary monocyte-derived macrophages ( MDM ) . Of note , nuclear CA immunofluorescence signals have been detected previously in MDM ( Bejarano et al . , 2019; Francis et al . , 2020 ) , but not or only weakly in T cell lines ( Zila et al . , 2019 ) or primary T cells . Infection of an eGFP . OR3 expressing SupT1 cell line ( Figure 7—figure supplement 1 ) or of primary activated CD4+ T cells transduced to express eGFP . OR3 ( Figure 7a ) with NNHIV ANCH showed nuclear OR3 punctae and separation of IN-FP and OR3 punctae as observed for TZM-bl cells . In accordance with previous observations ( Zila et al . , 2019 ) , no or weak CA signals were detected co-localizing with IN . SNAP punctae in T cells ( Figure 7a ) , even upon methanol extraction . However , similarly to observations in HeLa-based cells ( Figure 4d ) , strong signals for CPSF6 were detected at positions of IN-FP ( Figure 7b top panel ) . Displacing this CPSF6 coat from the nuclear subviral particles using PF74 ( Figure 7b–c ) as described above , strikingly revealed a strong CA IF signal ( Figure 7d–f ) . Of note , detection of nuclear CA in T cells depended on a combination of PF74 treatment and methanol extraction of specimens ( Figure 7—figure supplement 2 ) , while either treatment alone was sufficient to expose CA in HeLa-derived cells ( Figure 4—figure supplement 1a–b ) . In agreement with results from HeLa-based cells , no difference was observed for CA IF signals on extracellular particles and on cytoplasmic or nuclear envelope-associated subviral complexes upon PF74 treatment , while the nuclear CA signal became strongly enhanced ( Figure 7e–f ) . Thus , the failure to detect nuclear CA by IF in CD4+ T cells is due to shielding of epitopes by the accumulation of CPSF6 rather than to CA being lost upon nuclear entry in these cells , in accordance with our recent cryo-ET analyses in SupT1 cells ( Zila et al . , 2021 ) . Compared to extracellular virions , subviral particles in the cytosol and at the nuclear envelope displayed a reduced CA signal due to loss of free CA molecules from the post-fusion complex . This phenotype was most notable for nuclear envelope-associated complexes at later time points whose CA signal intensity corresponded to ~50% of that of complete particles ( Figure 7f ) . Of note , nuclear subviral complexes showed a higher mean CA signal compared to complete virions , consistent with nuclear clustering of CA containing complexes ( Francis et al . , 2020; Rensen et al . , 2021 ) also in this cell type; this was most evident at earlier time points ( Figure 7e–f ) . After PF74 treatment we could also observe some nuclear eGFP . OR3 punctae representing HIV-1 cDNA that were associated with IN-FP and strongly CA positive ( Figure 7g ) . Upon separation from eGFP . OR3 punctae , CPSF6 clusters remained associated with the IN-FP , confirming that the viral cDNA separates from an IN-FP/CA/CPSF6-positive nuclear structure in the nucleus of infected primary CD4+ T cells as well ( Figure 7h ) . Finally , we adapted the ANCHOR system to primary MDM by transducing these cells with an eGFP . OR3 expressing lentiviral vector . Three days post transduction , MDM were infected with VSV-G pseudotyped NNHIV ANCH containing IN . mScarlet for 24 hr . A similar phenotype as shown above for TZM-bl cells and primary CD4+ T cells was observed in MDM: the majority of eGFP . O3 punctae was detected separated from , but often in close vicinity of IN . mScarlet punctae ( Figure 7i ) .
In this study , we adapted a single molecule labeling method to study the dynamics of HIV-1 cDNA in living cells . Using this system , we showed that the HIV-1 ANCH dsDNA recognizing OR3 marker is only recruited to the viral cDNA inside the nucleus , while no OR3 punctae were observed associated with viral structures in the cytosol despite the presence of abundant reverse transcription products . Both , integrated and unintegrated HIV-1 cDNA were detected by OR3 in the nucleus . Metabolic labeling of nascent DNA revealed that nuclear eGFP . OR3 punctae contained significantly higher DNA amounts compared to cytoplasmic or nuclear subviral complexes lacking the eGFP . OR3 signal . Together with recent reports employing indirect RT inhibitor time-of-addition assays , which showed that reverse transcription remains sensitive to inhibition until after nuclear import ( Burdick et al . , 2020; Dharan et al . , 2020; Francis et al . , 2020 ) , and an elegant experiment showing that positive and negative strand-specific HIV DNA hybridisation probes only co-localize inside the nucleus ( Dharan et al . , 2020 ) , our data support completion of HIV-1 reverse transcription in the nucleus . Furthermore , HIV-1 cDNA separated from an IN fusion protein ( IN-FP ) , often used as a marker for the HIV-1 replication complex , inside the nucleus; IN fusions are thus not suitable for tracking HIV-1 cDNA in the nucleus . We expect that unfused IN , also present in the replication complex , will remain – at least partially – with the cDNA to mediate chromosomal integration and thus separates from the bulk of IN-FP . Nuclear IN-FP punctae were strongly positive for the viral CA protein and CA , CPSF6 and IN-FP signals stayed together after separation from the viral cDNA . OR3 recruitment to the HIV-1 cDNA requires the dsDNA to be accessible to the 66 kDa eGFP . OR3 fusion protein , which depends on loss of integrity of the capsid shell . CA signal intensity on nuclear IN-FP-positive and eGFP . OR3-negative structures was equal to or higher than observed for cytoplasmic complexes , suggesting that the bulk of CA stays associated with the viral replication complex and the viral DNA remains encased inside a closed capsid or capsid-like structure until after nuclear entry . Using CLEM-tomography of IN-positive and eGFP . OR3-negative nuclear complexes , we observed morphologically intact cone-shaped structures with internal density representing the nucleoprotein complex , which closely resembled HIV-1 capsids inside the cytosol or in complete virions . This finding is consistent with our recent study showing that the nuclear pore channel is sufficiently large to accommodate the HIV-1 core and apparently intact cone-shaped HIV-1 capsids can enter the nucleus through intact nuclear pores ( Zila et al . , 2021 ) . Taken together , these results indicate that reverse transcription initiates in the cytoplasm inside a complete or largely complete capsid , and this capsid-encased complex traffics into the nucleus , where reverse transcription is completed; subsequently , it must be uncoated for integration to occur . Formation of eGFP . OR3 punctae requires both , completion of reverse transcription and – at least partial – uncoating , and these two events may conceivably occur in a coordinated manner . IN-FP ( and CA ) positive nuclear complexes at later time points were often observed in close vicinity but clearly separated from eGFP . OR3 punctae , and live cell microscopy confirmed separation of the two markers from a single focus over time . Both markers retained their focal appearance and could thus be analyzed by CLEM . Electron tomography of late IN-FP-positive complexes revealed electron-dense structures that resembled broken HIV-1 capsids or capsid-like remnants lacking the density of the nucleoprotein complex . In contrast , eGFP . OR3-positive and IN-negative nuclear subviral structures never exhibited an electron-dense lining resembling the capsid shell , and these complexes were always negative for CA by immunofluorescence . These results indicate that viral cDNA associated with some replication proteins emanates from the broken capsid , which retains the bulk of CA and capsid-associated proteins . Uncoating therefore does not appear to occur by cooperative disassembly of the CA lattice , but by physically breaking the capsid shell and loss of irregular capsid segments . We often observed clustering of capsids or capsid-remnants inside the nucleus indicating preferred trafficking routes of subviral complexes , as described by others ( Francis et al . , 2020; Rensen et al . , 2021 ) . The broken capsid-remnant structures inside the nucleus of infected cells closely resembled ruptured HIV-1 cores observed in a recent study analyzing HIV-1 cDNA formation and integration in an in vitro system using purified virions ( Christensen et al . , 2020 ) . These authors reported partially broken capsid shells with irregular defects at time points when endogenous cDNA formation was largely completed; they also detected polynucleotide loops emanating from the holes in the capsid lattice . Theoretical models and AFM-studies had suggested that the volume of double-stranded HIV-1 b-DNA cannot be accommodated inside the intact capsid and that the resulting pressure might mechanically trigger uncoating ( Rankovic et al . , 2017; Rouzina and Bruinsma , 2014 ) . Taken together , these results suggest that the growing dsDNA inside the viral capsid in the nucleus may eventually lead to local rupture of the capsid lattice , concomitantly allowing completion of reverse transcription and triggering uncoating of the proviral DNA . It must be kept in mind , however , that lentiviral vectors with much shorter length of the vector RNA efficiently transduce cells; On the other hand , the effect might be offset by discontinuities present within the growing dsDNA chain in case of the full length genome ( Miller et al . , 1995 ) . Nuclear import is not required for completion of cDNA synthesis and loss of capsid integrity since similar structures were detected in the in vitro system ( Christensen et al . , 2020 ) . The observation that the viral cDNA was not fully released from the viral core in vitro suggests , however , that the nuclear environment may play a role in this process . The described pathway appears to be conserved in HeLa reporter cells and primary HIV-1-sensitive CD4+ T-cells and MDM: separation of IN/CA complexes from the OR3-positive cDNA inside the nucleus of infected cells was observed in all cell types , and the IN-positive subviral complexes exhibited a strong CA signal in all cases . Previous failure to detect CA on nuclear complexes in T cells has been due to epitope masking by the cellular CPSF6 protein and the current results thus indicate a common pathway for early HIV-1 replication in different cell types including primary target cells of HIV-1 infection . The efficiency of CA immunostaining in the nucleus appears to be partially dependent on the cell type . In MDMs ( Bejarano et al . , 2018; Stultz et al . , 2017; Zila et al . , 2019 ) and THP-1 cells ( Rensen et al . , 2021 ) , nuclear CA signals were detected by immunofluorescence upon copper-click-mediated cellular extraction without CPSF6-removal . In HeLa-derived cells , immunostaining of nuclear CA could be recovered by either methanol extraction or PF74-mediated CPSF6 displacement , whereas a combination of methanol extraction and PF74 treatment was necessary in T-cells . This difference might be explained by differences in nuclear architecture , the architecture of the CPSF6 cluster or/and the presence of other cell-type-specific proteins in the cluster and warrants further investigation . Nevertheless , strong CA signals resembling the intensity on cytoplasmic subviral HIV-1 complexes were detected in all cell types indicating a common pathway for nuclear entry of capsid-containing HIV-1 replication structures . STED and CLEM analysis revealed elongated structures with regularly spaced globular densities at the position of eGFP . OR3-positive punctae that had separated from the IN fusion protein and CA . These structures resembled chromatinized DNA ( Miron et al . , 2020 ) , in line with biochemical evidence that HIV-1 cDNA is rapidly chromatinized when it becomes accessible to the nucleoplasm ( Geis and Goff , 2019 ) . Detection of a cone-shaped structure lacking the electron-dense internal nucleoprotein signal and directly associated with an elongated chromatin-like structure at a position that was positive for both IN-FP and eGFP . OR3 may have captured a subviral complex in the process of uncoating . We suggest that chromatinization of HIV-1 cDNA emerging from the broken capsid may facilitate complete uncoating of the genome , which could explain why viral cDNA remained largely associated with the capsid structure in the in vitro system . In conclusion , our results indicate that complete or largely complete HIV-1 capsids enter the nucleus of infected cells , where reverse transcription is completed and the viral cDNA genome is released by physical disruption rather than by cooperative disassembly of the capsid lattice ( Figure 8 ) . The viral capsid thus plays an active role in the entire early phase of HIV-1 replication up to chromosomal integration and appears to be important for cytoplasmic trafficking , reverse transcription , shielding of viral nucleic acid from the innate immune system , nuclear entry , and intranuclear trafficking . The cone-shaped HIV-1 capsid with its fullerene geometry thus is the key orchestrator of early HIV-1 replication .
PrimerSequenceLinearize NL4-3/NNHIV fwCAGTTTTAATTGTGGAGGGGLinearize NL4-3/NNHIV rvttaAGGTACCCCATAATAGACSNAP_Bam_fwccgcgcgggatccagggatggacaaagactgcgaaatgSNAP_Not_rvgccgcccgcggccgctttacagcccaggcttgcccagtcteBFP2-LMNB1-10 into pWPI_BLR fwtttccgatcacgagactagcctcgaggtttGCCACCATGGTGAGCAAGeBFP2-LMNB1-10 into pWPI_BLR rvtttactagtacgcgtgcgatcgccccggggCTACATAATTGCACAGCTTCTATTGGU1a Fwd primerACATCAAGCAGCCATGCAAAAU1a Rev primerCAGAATGGGATAGATTGCATCCAU1a probeAAGAGACCATCAATGAGGAANuc1b Fwd primerCGTCTGTTGTGTGACTCTGGTAACTNuc1b Rev primerCACTGCTAGACATTTTCCACACTGANuc1b probeATCCCTCAGACCCTTTAluI ( first round Alu PCR ) TCCCAGCTACTGGGGAGGCTGAGGLM667 ( first round Alu PCR ) ATGCCACGTAAGCGAAACTCTGGCTAACTAGGGAACCCACTGλT ( second round Alu qPCR ) ATGCCACGTAAGCGAAACTLR ( second round Alu qPCR ) TCCACACTGACTAAAAGGGTCTGAZXF-P ( probe; second round Alu qPCR ) TGTGACTCTGGTAACTAGAGATCCCTCAGACCC Plasmids were cloned using standard molecular biology techniques and verified by commercial Sanger sequencing ( Eurofins Genomics , Germany ) . Gibson assembly was performed using the NEB HiFi Mastermix ( New England Biolabs , USA ) and 30 bp overlap regions . PCR was performed using Q5 High-Fidelity DNA Polymerase ( New England Biolabs ) according to manufacturer’s instructions with primers purchased from Eurofins Genomics . E . coli DH5α and Stbl2 ( Trinh et al . , 1994 , p . 2 ) ( Thermo Fisher Scientific , USA ) were used for amplification of standard plasmids or LTR containing plasmids , respectively . To facilitate the cloning procedure , EcoRI/XhoI fragments comprising the env region of HIV were subcloned from pNLC4-3 ( Bohne and Kräusslich , 2004 ) and its non-replication competent derivative pNNHIV ( Zila et al . , 2021 ) into pcDNA3 . 1 ( + ) ( Thermo Fisher Scientific ) . These constructs were PCR linearized , deleting a ~ 1000 bp region ( nt 130–1113 ) within the env coding sequence . The ANCH3 1000 bp sequence was PCR amplified from pANCH3 ( NeoVirTech , France ) , introducing a stop codon directly upstream of ANCH3 , and transferred into the linearized vector fragments using Gibson assembly . The modified fragments were transferred into pNL4-3 or pNNHIV backbones using EcoRI/XhoI . The SNAP-tag gene was PCR amplified from pSNAP-tag ( m ) ( Addgene #101135 ) and cloned into pVpr-IN . eGFP ( Albanese et al . , 2008 ) using BamHI/NotI , substituting the eGFP gene for the SNAP-tag coding region . To generate the Vpr-IND64N/D116N . SNAP mutant , the IND64N/D116N from Vpr-IND64N/D116N . eGFP ( gift from D . A . Bejarano ) was cloned into Vpr-IN . SNAP using BamHI/NotI . The ANCH3 1000 bp sequence was PCR amplified from pANCH3 ( NeoVirTech ) and cloned by Gibson assembly into pWPI IRES puro ( Trotard et al . , 2016 ) linearized with NotI . The eGFP . OR3 gene was PCR amplified from peGFP-OR3 ( NeoVirTech ) and transferred via Gibson assembly into PmeI/BamHI linearized pWPI IRES puro to create the expression cassette EF1-alpha eGFP . OR3 IRES puro . The SNAP gene was amplified from pVpr . IN . SNAP and the mScarlet ( WT ) ( Bindels et al . , 2017 ) gene from the mScarlet C1 vector ( Addgene #85042 ) and placed N-terminal to OR3 into PCR linearized pWPI EF1-alpha OR3 IRES puro backbone by Gibson assembly , substituting eGFP . The eBFP2 . LMNB1 gene was amplified from pEBFP2-LaminB1-10 ( Addgene #55244 ) and transferred via Gibson assembly into PmeI/BamHI linearized pWPI IRES BLR ( Trotard et al . , 2016 ) . HEK293T ( Pear et al . , 1993 ) ( RRID:CVCL_0063 ) , HeLa TZM-bl ( Wei et al . , 2002 ) ( RRID:CVCL_B478 ) , and SupT1 ( Smith et al . , 1984 ) ( RRID:CVCL_1714 ) cell lines were authenticated using STR profiling ( Eurofins Genomics ) and monitored for mycoplasma contamination using the MycoAlert mycoplasma detection kit ( Lonza Rockland , USA ) . Cells were cultured at 37°C and 5% CO2 in Dulbecco's Modified Eagle's Medium ( DMEM; Thermo Fisher Scientific ) containing 4 . 5 g l−1 D-glucose and L-glutamine supplemented with 10% fetal calf serum ( FCS; Sigma Aldrich , USA ) , 100 U ml−1 penicillin and 100 µg ml−1 streptomycin ( PAN Biotech , Germany ) ( adherent cell lines ) or in RPMI 1640 ( Thermo Fisher Scientific ) containing L-glutamine supplemented with 10% FCS , 100 U ml−1 penicillin and 100 µg ml−1 streptomycin ( SupT1 cells ) . Primary CD4+ T cells were cultured in RPMI 1640 containing L-glutamine supplemented with 10% heat-inactivated FCS , 100 U ml−1 penicillin and 100 µg ml−1 streptomycin . Monocyte-derived macrophages ( MDM ) were cultured in RPMI 1640 containing 10% heat-inactivated FCS , 100 U ml−1 penicillin , 100 µg ml−1 streptomycin and 5% human AB serum ( Sigma Aldrich ) . CD4+ T cells were enriched from blood of healthy donors using RosetteSep Human CD4+ T cell enrichment cocktail ( Stemcell Technologies , Canada ) according to the manufacturer’s instructions followed by Ficoll density gradient centrifugation . Subsequently , cells were activated using human T-Activator CD3/CD28 Dynabeads ( Thermo Fisher Scientific ) and 90 U/ml IL-2 for 48–72 hr . MDMs were isolated from buffy coats of healthy blood donors as described previously ( Bejarano et al . , 2019 ) . Lentiviral vector particles were produced by co-transfection of packaging plasmid psPAX2 ( Addgene #12260 ) , the respective lentiviral transfer vector pWPI , the envelope expression plasmid pCMV-VSV-G ( Addgene #8454 ) and pAdvantage ( Promega , USA ) in a ratio of 1 . 5: 1 . 0: 0 . 5: 0 . 2 µg into HEK293T cells ( 4 × 105 cells/well seeded the day before in six well plates ) using polyethylenimine ( PEI; 3 µl of 1 mg/ml PEI per µg DNA ) . At 48 hr post transfection , the tissue culture supernatant was harvested and filtered through 0 . 45 µm mixed cellulose ester ( MCE ) filters . SupT1 ( 1 ml of freshly 1:4 diluted cells ) or TZM-bl ( 5 × 104 cells/well seeded the day before in 12 well plates ) cells were transduced using 50–500 µl supernatant . At 48 hr post transduction , selection with 1 µg/ml puromycin or 5 µg/ml blasticidin was initiated . For transduction of MDM , lentiviral vectors were produced with Vpxmac239 ( Bejarano et al . , 2018 ) by calcium phosphate transfection of packaging plasmid pΔR8 . 9 NSDP ( Pertel et al . , 2011 ) , containing a Vpx interaction motif in Gag , pWPI eGFP . OR3 IRES puro , Vpx expression plasmid pcDNA . Vpxmac239 ( Sunseri et al . , 2011 ) and pCMV-VSV-G at a ratio of 1 . 33: 1 . 00: 0 . 17: 0 . 33 µg ( 68 µg / T175 flask ) . MDM were differentiated in human AB serum ( Sigma Aldrich ) from monocytes ( Bejarano et al . , 2019 ) in 15-well µ-Slides Angiogenesis ( ibidi , Germany ) for 10 days and transduction was performed 2 days prior to infection . Production of viral particle stocks pNLC4-3 or pNNHIV ANCH , a Vpr-IN plasmid ( Vpr- ( SNAP/eGFP/mScarlet ) . IN or Vpr- ( SNAP/eGFP/mScarlet ) . IND64N/D116N ) and pCMV-VSV-G or pCAGGS . NL4-3-Xba ( Bozek et al . , 2012 ) were transfected in a ratio of 7 . 7: 1 . 3: 1 . 0 µg into HEK293T cells using calcium phosphate ( 70 µg / T175 flask ) . Medium was changed at 6–8 hr and cells were further incubated for 48 hr . Supernatant was harvested and filtered through 0 . 45 µm MCE before ultracentrifugation through a 20% ( w/w ) sucrose cushion ( 2 hr , 107 , 000 g ) . Pellets were resuspended in phosphate-buffered saline ( PBS ) containing 10% FCS and 10 mM HEPES ( pH 7 . 5 ) , and stored in aliquots at - 80°C . Virus was quantified using the SYBR Green based Product Enhanced Reverse Transcription assay ( SG-PERT ) ( Pizzato et al . , 2009 ) . MOI of infectious particles was determined by titration on TZM-bl cells and immunofluorescence staining against HIV CA at 48 h p . i . The proportion of positive cells was counted in >10 randomly selected fields of view . 3 . 33 × 103 TZM-bl cells were seeded into 15-well µ-Slides Angiogenesis ( ibidi ) the day before infection . Stock solutions of SNAP-Cell TMR-Star or SNAP-Cell 647-SiR ( New England Biolabs ) in DMSO were diluted to 4 µM in complete medium , mixed 1:1 with IN . SNAP particles and incubated at 37°C for 30 min . Labeled particles were added to cells at 5–30 µUnits RT/cell in 50 µl . For VSV-G pseudotyped pNL4-3 ANCH , 30 µUnits RT per TZM-bl cell corresponds to ~MOI six in TZM-bl cells . Infection of MDM was performed with NNHIV ANCH ( 50 µl , 120 µUnits RT/cell ) . Infection of suspension cells was performed with 2 × 104 cells per 15 well µ-Slide in 96-well v-bottom plates ( 40 µl; 30 µU RT/cell ) . For PF-3450074 ( PF74; Sigma Aldrich ) experiments in primary CD4+ T cells , medium was changed at 5 or 22 hr to medium containing 15 µM PF74 or DMSO , for 1 hr before transfer to PEI coated ( with 1 mg/ml PEI for 60 min ) µ-Slides . Slides were incubated for 1 hr for cell attachment prior to fixation . Efavirenz ( EFV; Sigma Aldrich ) , Raltegravir ( Ral; AIDS Research and Reference Reagent Program , Division AIDS , NIAID ) and Azidothymidine ( AZT ) were added at time of infection . Flavopiridol ( Sigma Aldrich ) and 5 , 6-dichloro-1-beta-D-ribofuranosylbenzimidazole ( DRB; Sigma Aldrich ) were added 8 hr prior to fixation or RNA extraction . 10 µM EdU ( Thermo Fisher Scientific ) and 6 µM APC ( Sigma Aldrich ) were added at the time of infection . Samples were washed with PBS and fixed ( 15 min , 4% PFA ) , washed again three times using PBS , permeabilized with 0 . 5% Triton X-100 ( TX-100 ) for 10 or 20 min and washed again . In indicated experiments , cells were extracted using ice-cold methanol for 10 min . Afterwards , cells were washed two times using 3% bovine serum albumin ( BSA ) /PBS and blocked for 30 min with 3% BSA . Primary antibody in 0 . 5% BSA was added for 1 hr at room temperature . After washing three times with 3% BSA/PBS , secondary antibody in 0 . 5% BSA was added for 1 hr at room temperature and samples were washed and stored in PBS . For EdU incorporation experiments , cells were click-labeled for 30 min at room temperature using the Click-iT EdU-Alexa Fluor 647 Imaging Kit ( Thermo Fisher Scientific ) according to the manufacturer’s instructions . Biotinylated HIV-1 FISH probes were prepared with the Nick Translation Kit ( Roche , Germany ) according to the manufacturer’s instructions . Probes were purified with Illustra Microspin G-25 columns ( GE Healthcare , UK ) according to the manufacturer’s instructions and ethanol precipitated with human Cot-1 DNA ( Roche , Germany ) and herring sperm DNA ( Sigma Aldrich ) . Probes were resuspended in 10 μL formamide , incubated at 37°C for 15–20 min and 10 μL of 20% dextran/4X saline-sodium citrate ( SSC ) buffer was added . 1 . 25 × 104 TZM-bl cells/well were seeded on PEI ( 1 mg/ml ) coated glass cover slips and infected with VSV-G pseudotyped IN . SNAP . SiR labeled NNHIV ANCH ( 30 µU/cell ) . At 24 hr , cells were fixed for 10 min with 4% PFA/PBS , permeabilized with 0 . 5% TX-100/0 . 1% Tween/PBS for 10 min at room temperature , and washed in 0 . 1% Tween/PBS . Following 30 min blocking in 4% BSA/PBS , cells were incubated with rabbit anti-GFP antibody ( ab6556; Abcam , UK ) , diluted ( 1:2000 ) in 1% BSA/PBS overnight at 4°C . Cells were washed in 0 . 1% Tween/PBS and incubated with secondary Alexa Fluor antibody ( Thermo Fisher Scientific ) for 1 hr at room temperature . Cells were fixed for 10 min with 0 . 5 mM ethylene glycol bis ( succinimidyl succinate ) ( EGS ) /PBS , washed in 0 . 1% Tween/PBS and permeabilized with 0 . 5% TX-100/0 . 5% saponin/PBS for 10 min . Cells were incubated for 45 min in 20% glycerol/PBS and subjected to four glycerol/liquid N2 freeze-thaw cycles . Samples were rinsed , incubated in 0 . 1 M HCl for 10 min , equilibrated in 2X SSC for 20 min and left in hybridization buffer ( 50% Formamide/2X SSC ) for 30 min . Samples were then washed in PBS , treated with 0 . 01 N HCl/0 . 002% pepsin ( 3 min , 37°C ) and quenched by addition of 1X PBS/1 M MgCl2 . Fixation with 4% PFA/PBS and PBS wash was followed by treatment with 100 µg/ml RNase A ( PureLink , Invitrogen , USA ) in 2x SSC for 30 min at 37°C , washing and overnight storage in hybridization buffer . One to 10 μL of heat-denatured FISH probe ( 7 min at 95°C ) was loaded onto glass slides covered with coverslips coated with prepared cells . Slides were sealed in a metal chamber heated at 80°C for 7 min , and incubated for 48 hr at 37°C . Samples were washed in 2X SSC at 37°C , followed by 3 washes in 0 . 5 X SSC at 56°C . FISH detection was performed using anti-biotin antibody ( SA-HRP ) and a FITC/Cy5 coupled secondary antibody with the TSA Plus system ( Perkin Elmer , USA ) . Coverslips were stained with Hoechst , mounted on glass slides and imaged using the Nikon/Andor SDCM system described below . Spinning disc confocal microscopy ( SDCM ) was performed on an inverted Nikon Eclipse Ti2 ( Nikon , Japan ) microscope equipped with a Yokogawa CSU-W1 Spinning Disk Unit ( Andor , Oxford Instruments , United Kingdom ) and an incubation chamber ( 37°C , 5% CO2 ) . Imaging was performed using a 100 × oil immersion objective ( Nikon CFI Apochromat TIRF 100X Oil NA 1 . 49 ) and either single or dual-channel EMCCD camera setup ( ANDOR iXon DU-888 ) recording the eBFP2 ( 405/420–460 ) , eGFP ( 488/510–540 nm ) , mScarlet ( 568/589–625 nm ) and SiR ( 647/665–705 nm ) channels with a pixel size of 0 . 13 µm . 3D stacks of 10–30 randomly chosen positions were automatically recorded with a z-spacing of 0 . 3–0 . 5 µm using the Nikon Imaging Software Elements v5 . 02 . For CLEM experiments a Perkin Elmer Ultra VIEW VoX 3D spinning disk confocal microscope ( Perkin Elmer , United States ) with a 100 x oil immersion objective ( NA 1 . 4; Perkin Elmer ) was used . Medium was exchanged for 50 µl imaging medium containing FluoroBrite DMEM ( Thermo Fisher Scientific ) , 10% FCS , 4 mM GlutaMAX ( Gibco Life Technologies ) , 2 mM sodium pyruvate ( Gibco Life Technologies ) , 20 mM HEPES pH 7 . 4 , 100 U/ml Penicillin and 100 µg/ml Streptomycin ( PAN-Biotech ) . Samples were transferred to the SDCM setup described above . 3D stacks were recorded up to 24 hr ( time interval of 3–30 min , z-spacing 0 . 5 µm ) . Data from Figure 1g , f and Figure 4h was fit to a four-parametric logistic population growth model with variable slope using Prism 5 . 01 ( Graphpad ) . y=a+ ( b−a ) ∗ ( 1+10 ( Log ( t1/2−t ) −n ) ) −1with y = normalized eGFP . OR3 events per cell , a = Y value at the bottom plateau , b = Y value at the top plateau , t = h p . i . , t1/2 = time at half maximal signal and n = slope factor . 3 . 33 × 103 SNAP . OR3 expressing cells were seeded on 15-well µ-Slides Angiogenesis ( ibidi ) and infected as described above . Prior to fixation , cells were incubated with 2 µM SNAP-Cell 647-SiR ( New England Biolabs ) for 30 min at 37°C , washed three times and fixed with 4% PFA ( 15 min ) . STED microscopy was performed using a 775 nm STED system ( Abberior Instruments GmbH , Germany ) equipped with a 100 x oil immersion objective ( NA 1 . 4; Olympus UPlanSApo ) . STED Images were acquired using the 590 and 640 nm excitation laser lines while the 405 and 488 laser lines were acquired in confocal mode . Nominal STED laser power was set to 80% of the maximal power ( 1250 mW ) with 20 µs pixel dwell time and 15 nm pixel size . STED Images were linearly deconvolved with a Lorentzian function ( FWHM 50 nm ) using the Richardson-Lucy algorithm and the software Imspector ( Abberior Instruments GmbH ) . The images were filtered in Fiji/ImageJ ( Schindelin et al . , 2012 ) with a mean filter ( kernel size: 0 . 26 × 0 . 26 µm ) to reduce noise . For visualization some of the low signal-to-noise 3D movies were denoised using content-aware image restoration ( CARE ) ( Weigert et al . , 2018 ) as indicated . Convolutional neuronal networks were trained with a set of fixed cell images recorded with high laser power/long camera exposure ( ground truth ) and low laser power/short camera exposure ( noisy training input ) using the Python toolbox CSBdeep ( Weigert et al . , 2018 ) . The model was then applied to reconstruct raw movies in Fiji using the CSBdeep CARE plugin . Quantification of fluorescent spot intensities was performed using Icy ( de Chaumont et al . , 2012 ) . Raw 3D stacks were used to detect volumes of IN objects using the spot detector plugin . Methanol-induced shrinkage of cells in z orientation and infection-induced invaginations of the nuclear envelope renders automated nuclei detection difficult . To ensure that only truly nuclear objects ( and not still NE-associated ones ) were classified as such , they were manually curated and ambiguous particles were excluded . Objects displaying positive signals in the lamin channel were classified as nuclear envelope ( NE ) associated in the TZM-bl experiments . For measurements in primary CD4+ T cells , we applied a more stringent classification , manually excluding objects that did not colocalize with lamin in the major part of the signal or were localized above/below the focal planes of the nucleus . Cell-specific local background was subtracted for CA quantification . For quantification of CPSF6 intensities , the diffusive nuclear expression level of the cell was measured ( using a ROI without punctae ) and intensities of CPSF6 accumulations at IN objects were normalized to the expression level of the cell . Nuclear OR3 punctae were counted if their intensity was ≥20% above the diffuse nuclear expression level of the respective cell . Colocalization was scored when the pixel areas of the respective fluorescent spots ( partially ) overlapped . Tracking was performed using the Fiji plugin Manual Tracking . Fiji standard ‘Fire’ lookup table ( LUT ) was used for visualization of single channel images . Statistical tests were performed using Prism v5 . 01 ( GraphPad Software Inc , USA ) . Data were plotted using Prism v5 . 01 or the Python statistical data visualization package matplotlib v3 . 1 . 3 ( Hunter , 2007 ) and seaborn v0 . 10 . 0 ( Waskom et al . , 2020 ) . Graphs show mean with error bars defined in the figure legends . Particle preparations filtered through 0 . 45 µm CME were treated with 15 U/mL DNaseI ( Sigma Aldrich ) /10 mM MgCl2 for 3–4 hr at 37°C prior to ultracentrifugation . 5 × 104 TZM-bl cells were seeded into 24-well plates and infected the following day using 10–30 µU RT/cell . At the indicated h p . i . cells were washed , scraped and lysed using 50 µl of 10 mM Tris-HCl pH 9 . 0 , 0 . 1% TX-100 , 400 µg/mL proteinase K ( Thermo Fisher Scientific ) at 55°C overnight . Proteinase K was inactivated at 95°C for 10 min and lysates were stored at −20°C . Alternatively , DNA was purified from cells using the DNeasy Blood and Tissue Kit ( Qiagen , Germany ) according to the manufacturer’s instructions . Lysates were directly used as input for ddPCR; purified DNA was prediluted to ~20 ng/µl . For gag cDNA detection , this input was additionally diluted 1:20 to prevent saturation . ddPCR was performed using the QX200 droplet generator/reader ( BioRad , USA ) and analyzed using QuantaSoft v1 . 7 . 4 ( BioRad ) as described earlier ( Bejarano et al . , 2018; Zila et al . , 2019 ) . Quantification of viral RNA transcripts eBFP2 . LMNB1 and eGFP . OR3 expressing TZM-bl cells were infected with VSV-G pseudotyped NL4-3 ANCH ( 5 µUnits RT/cell; MOI ~ 1 ) for 55 hr . Twenty µm RAL was added at the time of infection . 8 hr prior to RNA extraction and purification using the Invitrap spin universal RNA mini kit ( Stratec biomedical , Germany ) , medium was changed for medium containing 1–25 µM flavopiridol ( Sigma Aldrich ) or 1–25 µM 5 , 6-dichloro-1-beta-D-ribofuranosylbenzimidazole ( DRB; Sigma Aldrich ) as indicated . Quantitative reverse transcription PCR was performed as previously described ( Marini et al . , 2015 ) . Briefly , messenger RNA ( mRNA ) levels were quantified by TaqMan quantitative RT-PCR ( qRT-PCR ) . First , the RT reaction was performed using M-MLV RT ( Invitrogen ) and a random primer set ( Invitrogen , cat . no . : 48190011 ) , followed by qPCR using HIV-1 primers and probes ( specific for transcription of the first nucleosome nuc1a or gag/u1a ) and the housekeeping genes 18S and GAPDH ( both containing VIC/TAMRA fluorescent probe; Applied Biosystems , USA ) as controls ( see list of primers ) . A total of 2 × 105 SupT1 cells were infected using VSV-G pseudotyped HIV-1 NL4-3 ANCH ( 10 µU RT/cell ) for 24 or 48 hr . Twenty µm Ral or 20 µm EFV was added at the time of infection . Cells were washed , lysed and genomic DNA was extracted using the DNeasy Blood and Tissue Kit ( Qiagen ) according to the manufacturer’s instructions . Nested Alu-LTR PCR was performed as described before ( Tan et al . , 2006 ) . Briefly , Alu-LTR fragments were amplified starting from 100 ng of genomic DNA . The product of the first amplification was diluted 1:50 in H2O and amplified by qPCR . The B13 region of the housekeeping gene lamin B2 was amplified from 10 ng of genomic DNA for normalization ( Livak and Schmittgen , 2001 ) . The copy number ( Log10 ) of integrated HIV-1 DNA per million cells was calculated using a standard curve obtained by serially diluting DNA from HIV-1 infected and sorted p24+ cells with DNA from uninfected cells as described before ( Shytaj et al . , 2020 ) . 1 . 2 × 105 TZM-bl mScarlet . OR3 cells were grown on 3 mm sapphire discs in a 35 mm glass-bottom dish ( MatTek , USA ) . Cells were infected with VSV-G pseudotyped IN . SNAP . SiR labeled NNHIV ANCH ( 30 µU RT/cell ) and incubated for 24 hr at 37°C . Subsequently , cells were cryo-immobilized by high pressure freezing ( HPF ) ( HPM010; Abra Fluid , Switzerland ) and transferred to freeze-substitution ( FS ) medium ( 0 . 1% uranyl acetate , 2 . 3% methanol and 1% H2O in Acetone ) tempered at −90°C . Freeze-substituted samples were embedded in Lowicryl HM20 resin ( Polysciences , USA ) inside a FS device ( AFS2 , Leica , Germany ) equipped with a robotic solution handler ( FSP , Leica ) . FS and embedding into Lowicryl resin was performed according to Kukulski et al . , 2011 with modifications ( Zila et al . , 2021 ) . Temperature was raised to −45°C at 7 . 5 °C/hr . Samples were washed with acetone ( 4 × 25 min ) and infiltrated with increasing concentrations of Lowicryl in acetone ( 25 , 50% and 75%; 3 hr each ) , while raising temperature to −25°C ( 3 . 3 °C /hr ) . The acetone-resin mixture was replaced by Lowicryl HM20 for 1 hr and the resin was exchanged three times ( 3 , 5 and 12 hr ) . Samples were polymerized under UV light for 24 hr at −25°C . Polymerization continued for an additional 24 hr while the temperature was raised to 20°C at 3 . 7 °C/hr . Thick resin sections ( 250 nm ) were cut using a microtome ( EM UC7 , Leica ) and placed on a slot ( 1 × 2 mm ) EM copper grid covered with a formvar film ( FF2010-Cu , Electron Microscopy Sciences , USA ) . Sections were covered by 0 . 1 μm TetraSpeck microsphere fiducials ( Thermo Fisher Scientific ) . Nuclear regions were stained with 50 μg/ml Hoechst ( Thermo Fisher Scientific ) . For SDCM , grids were transferred to 25 mm glass coverslips mounted in a water-filled ring holder ( Attofluor cell chamber , Thermo Fisher Scientific ) . Z stacks of cell sections were acquired using the PerkinElmer UltraVIEW VoX 3D Spinning-disc Confocal Microscope described above ( z-spacing 200 nm ) . To identify mScarlet . OR3 and IN . SNAP . SiR signals in cell sections , images were visually examined using Fiji ( Schindelin et al . , 2012 ) . Subsequently , both sides of EM grids were decorated with 15 nm protein-A gold particles for tomogram alignment and contrasted with 3% uranyl acetate ( in 70% methanol ) and lead citrate . Individual grids were placed in a high-tilt holder ( Fischione Model 2040 ) and loaded into the Tecnai TF20 ( FEI , Eindhoven , Netherlands ) electron microscope operated at 200 kV , equipped with a field emission gun and a 4 K by 4 K pixel Eagle CCD camera ( FEI , USA ) . To identify positions for ET , a full grid map was acquired using SerialEM ( Mastronarde , 2005 ) and acquired electron micrographs were pre-correlated with imported SDCM images in SerialEM using fiducials as landmark points ( Schorb et al . , 2017 ) . Single-axis electron tomograms of selected regions were then carried out . Tomographic tilt ranges were from −60° to 60° with an angular increment of 1° and pixel size 1 . 13 nm . Alignments and 3D reconstructions of tomograms were done using IMOD ( Kremer et al . , 1996 ) . For high precision post-correlation , tomograms of cell sections were acquired at lower magnification with 4° increment and 6 . 3 nm pixel size . Post-correlation was performed using the eC-CLEM plugin ( Paul-Gilloteaux et al . , 2017 ) in Icy ( de Chaumont et al . , 2012 ) . The length and diameter of capsids were measured in IMOD ( Kremer et al . , 1996 ) . Segmentation and isosurface rendering were performed in Amira ( Thermo Scientific ) . | When viruses infect human cells , they hijack the cell’s machinery to produce the proteins they need to replicate . Retroviruses like HIV-1 do this by entering the nucleus and inserting their genetic information into the genome of the infected cell . This requires HIV-1 to convert its genetic material into DNA , which is then released from the protective shell surrounding it ( known as the capsid ) via a process called uncoating . The nucleus is enclosed within an envelope containing pores that molecules up to a certain size can pass through . Until recently these pores were thought to be smaller than the viral capsid , which led scientists to believe that the HIV-1 genome must shed this coat before penetrating the nucleus . However , recent studies have found evidence for HIV-1 capsid proteins and capsid structures inside the nucleus of some infected cells . This suggests that the capsid may not be removed before nuclear entry or that it may even play a role in helping the virus get inside the nucleus . To investigate this further , Müller et al . attached fluorescent labels to the newly made DNA of HIV-1 and some viral and cellular proteins . Powerful microscopy tools were then used to monitor the uncoating process in various cells that had been infected with the virus . Müller et al . found large amounts of capsid protein inside the nuclei of all the infected cells studied . During the earlier stages of infection , the capsid proteins were mostly associated with viral DNA and the capsid structure appeared largely intact . At later time points , the capsid structure had been broken down and the viral DNA molecules were gradually separating themselves from these remnants . These findings suggest that the HIV-1 capsid helps the virus get inside the nucleus and may protect its genetic material during conversion into DNA until right before integration into the cell’s genome . Further experiments studying this process could lead to new therapeutic approaches that target the capsid as a way to prevent or treat HIV-1 . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2021 | HIV-1 uncoating by release of viral cDNA from capsid-like structures in the nucleus of infected cells |
The origin of angiosperms has been a long-standing botanical debate . The great diversity of angiosperms in the Early Cretaceous makes the Jurassic a promising period in which to anticipate the origins of the angiosperms . Here , based on observations of 264 specimens of 198 individual flowers preserved on 34 slabs in various states and orientations , from the South Xiangshan Formation ( Early Jurassic ) of China , we describe a fossil flower , Nanjinganthus dendrostyla gen . et sp . nov . . The large number of specimens and various preservations allow for an evidence-based reconstruction of the flower . From the evidence of the combination of an invaginated receptacle and ovarian roof , we infer that the seeds of Nanjinganthus were completely enclosed . Evidence of an actinomorphic flower with a dendroid style , cup-form receptacle , and angiospermy , is consistent with Nanjinganthus being a bona fide angiosperm from the Jurassic , an inference that we hope will re-invigorate research into angiosperm origins .
Despite the importance of , the great interest in and intensive effort spent on investigating angiosperms , a controversy remains as to when and how this group came into existence . Since the time of Darwin , some scholars have proposed that angiosperms existed before the Cretaceous ( Smith et al . , 2010; Clarke et al . , 2011; Zeng et al . , 2014; Buggs , 2017 ) , although the conclusion ‘there are no reliable records of angiosperms from pre-Cretaceous rocks’ made almost 60 years ( Scott et al . , 1960 ) seemed to be recently re-confirmed ( Herendeen et al . , 2017 ) . Such uncertainty makes answers to many questions about the phylogeny and systematics of angiosperms tentative . Some reports of early angiosperms ( i . e . , Monetianthus ( Friis et al . , 2001 ) ) are based on a single specimen , which restricts further testing and confirming . Better and more specimens of early age and with features unique to angiosperms are highly sought-after to test related evolutionary hypotheses . Here , we report an unusual actinomorphic flower , Nanjinganthus gen . nov . , from the Lower Jurassic based on the observations of 264 specimens of 198 individual flowers on 34 slabs preserved in various orientations and states ( Supplementary file 1 ) . The abundance of specimens allowed us to dissect some of them , thus demonstrate and recognize a cup-form receptacle , ovarian roof , and enclosed ovules/seeds in Nanjinganthus . These features are consistent with the inference that Nanjinganthus is an angiosperm . The origin of angiosperms has long been an academic ‘headache’ for many botanists , and we think that Nanjinganthus will shed a new light on this subject .
Nanjinganthus gen . nov . Flowers subtended by bracts . Bracts fused basally . Flowers pedicellate , actinomorphic , epigynous , with inferior ovary . Sepals 4–5 , rounded in shape , each with usually 4–6 longitudinal ribs in the center and two lateral rib-free laminar areas , attached to the receptacle rim with their whole bases , surrounding the petals when immature , with epidermal cells with straight cell walls . Petals 4–5 , cuneate , concave , each with usually 5–6 longitudinal ribs in the center and two lateral rib-free laminar areas , with rounded tips , surrounding the gynoecium when immature , with epidermal cells with straight cell walls . Gynoecium in the center , unilocular , fully closed by a cup-form receptacle from the bottom as well as sides and by an integral ovarian roof from the above . Style centrally attached on the top of the ovarian roof , dendroid-formed . One to three seeds inside the ovary , elongated oval , hanged on the ovarian wall by a thin funiculus , with the micropyle-like depression almost opposite the chalaza . Nanjinganthus dendrostyla gen . et sp . nov . Nanjing- for Nanjing , the city where the specimens were discovered , and -anthos for ‘flower’ in Latin . Wugui Hill , Sheshan Town , Qixia District , Nanjing , China ( N32˚08″ 19′ , E118˚58″ 20′ ) ( Figure 1—figure supplement 1 ) . The South Xiangshan Formation , the Lower Jurassic . Nanjinganthus dendrostyla gen . et sp . nov . the same as the genus . The flowers are frequently concentrated and preserved in groups on certain bedding surfaces ( Figures 1a–g and 2a–b ) , although many of them are preserved as isolated individuals on other slabs . Figure 2d ( PB22222B ) . Figure 6a , f ( PB22222a ) , Figure 7e–i ( PB22281 ) , Figure 5h ( PB22279 ) . PB22222-PB22229 , PB22236 , PB22238 , PB22241-PB22243 , PB22245-PB22247 , PB22256-PB22260 , PB22278-PB22282 , PB22489 . dendrostyla , for ‘tree-like’ ( dendri- ) and ‘style’ ( -stylus ) in Latin . The receptacle is ‘the axis of a flower on which the perianth , androecium and gynoecium are borne’ ( Stevens , 2018 ) . This is the definition followed here . The important characteristic of the receptacle in Nanjinganthus is its cup form , a form frequently seen in more derived angiosperms according to the APG system . A dendroid style is seen in ten flowers ( four in PB22224 , Figures 2h and 3a–b; four in PB2222a , Figures 2i , 5f–g , 6a and 7d; one in PB22282 , Figure 7a–c; one in PB22489 , Figure 5i–j ) . The repeated occurrences of such an unexpected feature in the specimens of Nanjinganthus underscore its truthful existence . The dendroid-form distal portion of the gynoecium may be branched stigmas in Nanjinganthus . But it is possible that these lateral appendages on the style are actually pollen sac complexes , as are similarly attached on the style in extant Malvaceae ( Judd et al . , 1999 ) . We have performed a meticulous fluorescence microscopic examination of this structure and found no trace of pollen grains , reducing the possibility that these lateral branches are clusters of pollen sacs , which is the case seen in some angiosperms ( Malvaceae ) . A branched distal projection is apparently lacking in all known gymnosperms , but it has been seen some derived angiosperms , such as Passifloraceae , Poaceae and Euphorbiaceae ( Heywood , 1978 ) . One of the advantages of a branched style is the increased receptive area , which is conducive to anemophilous pollination . The occurrence of such feature in Nanjinganthus might suggest that Nanjinganthus had yet not established a close cooperation with animals ( insects ) . However , it is noteworthy that this feature is not seen among extant basal angiosperms sensu APG ( Chase et al . , 2016 ) . Considering the extremely early age of Nanjinganthus , we refrain from correlating Nanjinganthus with assumed derived taxa ( Malvaceae and Rosaceae ) . We hope the future research may shed more light the nature of this part of Nanjinganthus . We have not seen any trace of the carpels typical of Magnoliales , which were previously believed by some to represent ancestral angiosperms . The seeds are physically enclosed by the cup-form receptacle and ovarian roof in Nanjinganthus . This constitutes the foundation based on which we justify our interpretation of Nanjinganthus as an angiosperm . The lack of carpel typical of Magnoliales cannot prevent Nanjinganthus from being an angiosperm as many angiosperms are actually ‘acarpellate’ ( Heads , 1984; Sattler and Lacroix , 1988 ) . It is noteworthy that , at least in some of basal angiosperms such as Nymphaea ( Nymphaeales ) ( Taylor , 1991; Taylor , 1996 ) and derived angiosperms such as Cactaceae ( Boke , 1964 ) , the ovary is inferior and the seeds are attached to the ovarian walls . Whether the ovaries in these taxa share similar derivation pathway is a question worthy of further investigation . Four terms are used to describe the foliar parts in Nanjinganthus , namely , bract , scale , sepal , and petal . These terms are used according to the following demarcations and definitions . Bracts designate the foliar parts subtending the ovary . The scales are the foliar parts attached to the sides of the ovary . The sepals are those foliar parts attached to the rim of the receptacle with their whole bases . And the petals are foliar parts with narrowing bases attached to the receptacle rim and inside the sepals . Similar occurrence of bracts , sepals and petals is seen in some extant flowers ( Figure 5—figure supplement 1 ) . The enclosure of the seeds is fulfilled by the cup-form receptacle from the bottom and the structure here-called ‘ovarian roof’ ( preserved complete in Figures 4c , 5h and 7e–g , but partially preserved in Figures 2f , 5c and 6f , j , l ) from the above . The intact ovarian roof is clearly seen in the side view ( Figure 7f–g ) and in surface view ( Figures 4c and 5h ) , in the latter case the seeds inside ovary are fully eclipsed by the ovarian roof . The ovarian roof is partially lost in Figure 6l , in which a central portion of the ovarian roof broke off revealing one of the seeds inside the ovary . The ovarian roof is almost completely lost ( but still with some of its residue ) in Figure 6f , and finally fully lost in Figures 2f and 6j–k , in which the seeds are plainly visible . This series of varying preservation status of ovarian roof suggests that the ovarian roof has fully enclosed the ovules in its original status , and the loss of ovarian roof and exposure of seeds are artifacts due to preservation . We cannot recognize the maturity of the ovules/seeds in Nanjinganthus , the length about 1 mm suggests that they are most likely to be seeds rather than ovules , therefore we prefer to use the term ‘seed’ rather ‘ovule’ throughout this paper . The number of seeds in Nanjinganthus is variable . According to our observation , it may be one ( not shown ) , two ( Figure 6f–i ) , or even three ( Figure 6j–k ) .
The Mesozoic was an age of gymnosperms , so the Jurassic age of Nanjinganthus makes it necessary to compare Nanjinganthus with common fossil gymnosperms frequently seen in the Mesozoic first . The potential candidates for Nanjinganthus include Caytoniales , Corystospermales , Ginkgoales , Czekanowskiales , Coniferales , Iraniales , Pentoxyales , Bennettitales , and Gnetales . Caytonia is a very intriguing fossil plant that has been frequently compared with angiosperms ( Thomas , 1925; Doyle , 2006 ) . Regardless of its ultimate gymnospermous affinity ( Thomas , 1925; Harris , 1933; Harris , 1940; Nixon et al . , 1994 ) , Caytonia can be easily distinguished from Nanjinganthus by its cupule with an adaxial basal opening , bilateral reproductive organs , and lack of both a dendroid style and foliar appendages in its reproductive organs . Corystospermales is usually considered as a Mesozoic seed fern group , unlike Caytonia , the cupules in most Corystospermales open on the abaxial base and are rarely compared with angiosperms ( Taylor et al . , 2009 ) . Corystospermales can be easily distinguished from Nanjinganthus by their cupule which has an abaxial basal opening , bilateral reproductive organs , and the lack of both a dendroid style and foliar appendages in the reproductive organs . Ginkgoales diversified greatly during the Mesozoic , and unlike extant Ginkgo , the Mesozoic relatives of Ginkgo are well represented by their reproductive organs , which are composed of seeds in clusters ( Zhou and Zheng , 2003 ) . Ginkgoales can be easily distinguished from Nanjinganthus by their clustered naked seeds , and lack of a dendroid style in the reproductive organs . Czekanowskiales are a unique group of fossil plants restricted to the Mesozoic . Their reproductive organs are bivalvate cupules containing two rows of seeds . Czekanowskiales can be easily distinguished from Nanjinganthus by their bivalvate cupules , bilateral reproductive organs , and lack of both a dendroid style and foliar appendages in the reproductive organs . Irania is the single genus of the Iraniales , which is assumed to have borne clusters of pollen sacs and fruits , from the Triassic-Jurassic ( Schweitzer , 1977 ) . Although no seeds have been observed in Irania , it is suspected to be an angiosperm . Its female and male parts are not concentrated on the same axis , and do not constitute a flower-like structure , and it is unknown whether the seeds are enclosed . These features distinguish Irania from Nanjinganthus . Pentoxylales are Mesozoic woody fossil plants characterized by a stem with five steles ( Taylor et al . , 2009 ) . Their reproductive organs are cones with numerous naked orthotropous seeds helically arranged around the axes of their cones . Pentoxylales can be easily distinguished from Nanjinganthus by their cones which are devoid of any foliar appendages and the lack of a dendroid style . Bennettitales are important Mesozoic gymnosperms that are frequently related to angiosperms ( Crane , 1985; Rothwell et al . , 2009 ) . Their reproductive organs are characterized by orthotropous ovules with micropylar tubes dispersed among interseminal scales , and these parts are helically arranged along the cone axis . Bennettitales can be easily distinguished from Nanjinganthus by their cones with ovules bearing micropylar tubes , and lack of a dendroid style ( Taylor et al . , 2009 ) . Gnetales are important gymnosperms that diversified once in the Mesozoic , among which Gnetum has leaves that are difficult to distinguish from eudicots ( Martens , 1971; Biswas and Johri , 1997 ) . A characteristic feature of Gnetales is their decussate arrangement of leaves and cone parts . Like in Bennettitales , the reproductive organs of Gnetales are characterized by orthotropous ovules with micropylar tubes surrounded by scales . Like Bennettitales , Gnetales can be easily distinguished from Nanjinganthus by their cones with ovules with micropylar tubes and lack of a dendroid style . Besides the above comparison among female organs of seed plants , it is noteworthy that male cones in some conifers demonstrate a certain resemblance to Nanjinganthus , although such a comparison appears spurious when the presence of seeds in Nanjinganthus is taken into consideration . The bud-like Nanjinganthus ( Figures 2g , 4a–b , g and 5i ) appears similar to male cones of Microbiota decussata ( a3 in Figure 2 of Schulz et al . , 2014 ) and Thuja plicata ( f2 in Figure 2 of Schulz et al . , 2014 ) . Mature Nanjinganthus ( Figures 2h–i , 3a–b , 5f–g , 6a and 7a , e , i ) appear like the male cone of Sequoia sempervirens , Taxus floridana , and Tsuga canadensis ( e2 , e5 , f4 , respectively , in Figure 2 of Schulz et al . , 2014 ) . The sporangiophores in these taxa are arranged around the cone axis and appear dendrioid , and the scales at the base appear like the sepals/petals in Nanjinganthus . But the cup-form receptacle with seeds inside plus the lack of pollen grains in the distal dendroid part distinguish Nanjinganthus from all these male cones . As in female cones , these male cones also have cone axes penetrating the cones from the bottom to the tip and thus are different from Nanjinganthus in which the pedicel stops at the bottom of the organ ( ovary ) ( Figure 6b ) and the style starts above the ovarian roof ( Figure 4c , 5h , 6a-c , 7a-c ) . In addition , the spatial distribution and furcation of the vascular bundles in the sepals and petals of Nanjinganthus ( Figures 2c–d , 4e–f , 9a–b ) are distinct from those seen in bracts and scales in coniferous cones . From the above comparison , we infer that none of the known gymnosperms , fossil or extant , are comparable to Nanjinganthus . The enclosed seeds distinguish Nanjinganthus from gymnosperms , which are not supposed to enclose their ovules in such a way ( Table 1 ) . There are several reports of Jurassic angiosperms , including Schmeissneria ( Wang et al . , 2007 ) , Xingxueanthus ( Wang and Wang , 2010 ) , Juraherba ( Han et al . , 2016 ) , and Euanthus ( Liu and Wang , 2016 ) . These genera are from the Middle-Late Jurassic of northeastern China . The cup-form receptacle , inferior ovary , and dendroid style distinguish Nanjinganthus from all these Jurassic peers , and justify Nanjinganthus as a new genus of angiosperm . Although multiple characters have been suggested to identify fossil angiosperms ( Herendeen et al . , 2017 ) , angio-ovuly before pollination is the key character that distinguishes an angiosperm from other seed plants ( Tomlinson and Takaso , 2002; Wang , 2010a; Wang , 2018 ) . This criterion has been repeatedly applied to identify fossil angiosperms ( i . e . Archaefructus ( Sun et al . , 1998 ) , which initially had no other features ( stamen , venation , pollen grains ) but enclosed seeds to support their angiospermous affinity ) . The integral ovarian roof of Nanjinganthus has no opening ( Figures 4c and 5h ) . After burial , this ovarian roof can block the sediment from entering the ovarian locule ( Figure 7e–g ) . That this space remained free of sediment suggests a full enclosure of the ovules/seeds , securing an angiospermous affinity for Nanjinganthus . The radial arrangement of two whorls of foliar parts ( sepals and petals ) in Nanjinganthus is very similar to those of flowers in extant angiosperms ( Figures 1a–b and 2a–f ) , while the above comparison with known gymnosperms emphasizes that Nanjinganthus cannot be interpreted as a gymnosperm . Furthermore , Nanjinganthus satisfies all thirteen definitions of flowers advanced by various authors ( Bateman et al . , 2006 ) . These features again are consistent with the angiospermous affinity of Nanjinganthus ( Figure 11 ) . There have been several suggested models of ancestral angiosperms ( Arber and Parkin , 1907; Cronquist , 1988; Endress and Doyle , 2015; Sauquet et al . , 2017 ) . These models were drawn more or less after the assumed basalmost living angiosperms , either Magnolia or Amborella . The common features of these model plants include apocarpy , superior ovary , lack of obvious style , etc . However , these features are rarely seen in Nanjinganthus or other early angiosperms ( Wang et al . , 2007; Wang and Zheng , 2009; Wang , 2010b; Wang , 2018; Han et al . , 2013; Han et al . , 2016; Han et al . , 2017; Liu and Wang , 2016; Liu and Wang , 2017; Liu et al . , 2018; Liu and Wang , 2018 ) . Instead , an inferior ovary , a feature unexpected by , at least , most theories of angiosperm evolution , is clearly seen in Nanjinganthus and quite many Early Cretaceous flowers ( Friis , 1984; Friis , 1990; Friis et al . , 2011 ) . This discrepancy between fossil observation and theories suggests EITHER that inferences based on living plants have limited capability of ‘predicting’ past history , OR that angiosperms originated polyphyletically , each lineage has followed a different evolution route , and Nanjinganthus represents one of the many , OR that angiosperms have a history that dates back to a time much earlier than the Cretaceous , and Nanjinganthus is one of the many derived from such assumed ancestor , OR a combination of these . Whatever the implications are , the currently dominant theories of angiosperm evolution apparently need to be reassessed . Most Nanjinganthus specimens are concentrated on certain bedding surfaces , and over 50 individual flowers are preserved on a single slab ( Figures 1a–g and 2a–b ) , suggesting that Nanjinganthus may have flourished and dominated a particular niche , although Nanjinganthus played only an inferior role in the broader Jurassic ecosystem . The concentrated preservation of delicate flowers is more likely a result of autochthonous preservation , suggesting a habitat very close to water for Nanjinganthus . Various studies ( including palaeobotany ) on the South Xiangshan Formation in the last century ( Hsieh , 1928; Li et al . , 1935; Sze and Chow , 1962; Zhou and Li , 1980; Cao , 1982; Cao , 1998; Cao , 2000; Wang et al . , 1982; Huang , 1983; Huang , 1988; Ju , 1987 ) and our palynological analysis as well as U/Pb dating ( Figure 1—figure supplement 1; Supplementary file 2; Santos et al . , 2018 ) suggest a latest Early Jurassic age for Nanjinganthus . Together with the ‘unexpectedly’ great diversity of angiosperms in the Early Cretaceous ( Sun et al . , 1998; Sun et al . , 2001; Sun et al . , 2002; Leng and Friis , 2003; Leng and Friis , 2006; Ji et al . , 2004; Wang and Zheng , 2009; Wang and Zheng , 2012; Wang and Han , 2011; Han et al . , 2013; Han et al . , 2017; Wang , 2015; Liu et al . , 2010a ) , pollen grains indistinguishable from angiosperms in the Triassic ( Hochuli and Feist-Burkhardt , 2004; Hochuli and Feist-Burkhardt , 2013 ) , a bisexual flower from the Jurassic ( Liu and Wang , 2016 ) , and an herbaceous angiosperm from the Middle Jurassic ( Han et al . , 2016 ) , the unexpectedly early age of Poaceae ( Prasad et al . , 2011; Wu et al . , 2018 ) and Solanaceae ( Wilf et al . , 2017 ) , Nanjinganthus with over 200 specimens is consistent with a pre-Cretaceous origin of angiosperms . The systematic position of Nanjinganthus is now apparently open to further investigation , although it demonstrates a certain resemblance to Pentapetalae sensu ( Judd et al . , 2016 ) . We cannot determine whether Nanjinganthus stands for a Jurassic stem group of angiosperms that started their radiation later in the Cretaceous or a lateral branch leading to an evolutionary dead end . It is premature to determine its phylogenetic position before more information of contemporaneous peers is available , although we welcome all phylogeneticists to evaluate Nanjinganthus in their own ways and perspectives . Nanjinganthus is recognized based on at least 198 individual flower fossils from the Early Jurassic that are preserved in various states and orientations . We infer that the seeds were enclosed by a cup-form receptacle and an ovarian roof , traits which suggest an angiospermous affinity for Nanjinganthus . It would be intriguing to figure out in future research whether Nanjinganthus represents a stem group , a group derived from more ancient ancestors , or an evolutionary dead end of polyphyletic angiosperms . We hope that the discovery of Nanjinganthus will re-invigorate research on the origin and early history of angiosperms .
Initially , what is now known as the Xiangshan Group was called the ‘Nanking Sandstein’ that was put in the Upper Carboniferous by Richthofen ( 1912 ) . Liu and Chao ( 1924 ) thought that the Nanking Sandstein belonged to the Jurassic and renamed it the ‘Tsung Shan Formation’ . Hsieh ( 1928 ) subdivided the Tsungshan Formation into six units , namely , in ascending order , Huang Ma Ching Shale , Quartzitic Conglomerate , Tzu Hsia Tung Series , Lingkusssu Shale , Light Yellow Sandstone , Variegated Sandstone and Shale . Li et al . ( 1935 ) found fossil plants including Equisetites , Neocalamites , Cladophlebis , Otozamites , Pterophyllum , Dictyophyllum , Pagiophyllum , Baiera guilhaumatii , and Podozamites lanceolatus . They also changed the ‘Tsungshan Formation’ to ‘Xiangshan Layers’ , and regarded its age as being the Early Jurassic . Sze and Chow ( 1962 ) used the term ‘Xiangshan Group’ for the previous ‘Xiangshan Layers’ , and this has been the convention followed ever since . The age of the Xiangshan Group has been concluded by various authors to range from the Late Triassic to the Middle Jurassic . Ju ( 1987 ) divided the Xiangshan Group into a lower South Xiangshan Formation and an upper North Xiangshan Formation , respectively . The standard section of the lower part of the Xiangshan Group ( the South Xiangshan Formation , the Lower Jurassic ) is 394 meters thick in South Xiangshan in Nanjing , and has yielded abundant fossil plants . The standard section of the upper part of the Xiangshan Group ( the North Xiangshan Formation , the Middle Jurassic ) is in North Xiangshan in Nanjing . It is 1005 meters thick , and has only yielded a few stem fossils ( Ju , 1987 ) . Based on fossil plants , Cao ( 1982 ) thought that the age of the South Xiangshan Formation could not be later than the Early Jurassic , and considering that the early Early Jurassic flora ( in the middle and lower parts of the Guanyintan Formation in southwest Hunan ) ( Zhou and Li , 1980 ) is biostratigraphically below the South Xiangshan Formation , Cao ( 1982 ) regarded the age of the lower part of Xiangshan Group ( =South Xiangshan Formation ) as the middle-late Early Jurassic . The South Xiangshan Formation ( lower part of the Xiangshan Group ) has yielded abundant bivalve and plant fossils . Its outcrops are scattered in Jiangning , Longtan , and Zhenjiang ( all in the suburbs of Nanjing ) . In these areas , the outcrops are well exposed and especially fossiliferous near the South Xiangshan and Cangbomen regions . The formation includes sandstones , siltstones , shales , carbonaceous shales , and coal seams . There are abundant plant fossils in the South Xiangshan Formation , and almost all of the plants in the Xiangshan Group are from this formation . Various authors have collected fossil plants of the Xiangshan Flora ( Cao , 1982; Cao , 1998; Cao , 2000; Wang et al . , 1982; Huang , 1983; Huang , 1988; Ju , 1987 ) . According to Cao ( 1982 ) , Cao ( 1998 ) , Cao ( 2000 ) , Wang et al . ( 1982 ) , and Ju ( 1987 ) , the Xiangshan Group includes at least 46 genera of plants and is very similar to the flora of the Hsiangchi Group in western Hubei . Cycadophytes ( 34% ) dominate the flora , and ferns are the second most dominant group ( 20% ) , among which Dipteridaceae play an important role . Ginkgoales are also abundant ( 19% ) ( Ju , 1987 ) . The important and frequently observed taxa include Hysterites , Selaginellites , Equisetites cf . lateralis , E . aff . multidentatus Oishi , E . sarrani ( Zeiller ) Halle , Neocalamites hoerensis ( Schimper ) Halle , N . dangyangensis Chen , Marattiopsis asiatica Kawasaki , M . hoerensis ( Schimper ) Schimper , Todites goeppertianus ( Münster ) Krasser , T . princeps ( Presl ) Gothan , Osmundopsis Harris , Cladophlebis denticulata ( Brongniart ) Fontaine , C . goeppertianus ( Münster ) Krasser , C . raciborskii Zeiller , Spiropteris Schimper , Phlebopteris polypodioides Brongniart , Danaeopsis Heer ex Schimper , Thaumatopteris pusilla ( Nathorst ) Oishi et Yamasita , Dictyophyllum nathorstii Zeiller , D . nilssonii ( Brongniart ) Goeppert , Clathropteris meniscioides Brongniart , Cl . platyphylla Goeppert , Cl . obovata Oishi , Coniopteris hymenophylloides ( Brongniart ) Seward , Thinnfeldia Ettingshausen , Augustiphyllum yaobuensis Huang , Scoresbya dentata Harris , Pterophyllum firmifolium Ye , Pt . propinquum Goeppert , Pt . subaequale Hartz , Nilssonia complicatis Li , N . orientalis Heer , N . minor Harris , N . cf . compta ( Schenk ) Ye , N . cf . polymorpha Schenk , N . pterophylloides Nathorst , N . cf . saighanensis Seward , N . taeniopterioides Halle , N . parabrevis Huang , N . moshanensis Huang , Nilssoniopteris vittata ( Brongn . ) Florin , Ctenis Lindley et Hutton , Ctenozamites cf . ptilozamioides Zhou , C . cf . cycadea ( Berger ) Schenk , Cycadolepis corrugata Zeiller , Anomozamites cf . minor Nathorst , A . cf . major ( Brong ) Huang , A . cf . inconstans ( Goeppert ) Schimper , A . quadratus Cao , Tyrmia nathorstii ( Schenk ) Yeh , T . latior Ye , T . lepida Huang , T . susongensis Cao , Otozamites minor Tsao , Ot . hsiangchiensis Sze , Ot . mixomorphus Ye , Ot . tangyanensis Sze , Ptilophyllum hsingshanense ( Wu ) Cao , Pt . contiguum Sze , Pt . pecten ( Philips ) Morris , Hsiangchiphyllum trinervis Sze , Ginkgoites cf . tasiakouensis Wu et Li , G . cf . sibiricus ( Heer ) Seward , G . cf . magnifolius Du Tiot , Baiera cf . furcata ( L . et H . ) Braun , B . asadai Yabe et Oishi , B . guilhaumatii Zeiller , B . multipartita Sze et Lee , B . cf . gracilis Bunbury , Sphenobaiera huangii ( Sze ) Hsu ex Li , S . spectabilis ( Nath . ) Florin , Czekanowskia rigida Heer , C . hartzii Harris , Phoenicopsis Heer , Ginkgodium Yokoyama , Desmiophyllum Lesquereux , Stenorachis ( Nathorst ) Saporta , Vittifoliolum multinerve Zhou , Pityophyllum longifolium ( Nathorst ) Möller , Podozamites lanceolatus ( L . et H . ) Braun , Ferganiella Prynada , Elatocladus Halle , Swedenborgia cryptomerioides Nathorst , Taeniopteris cf . richthofenii ( Schenk ) Sze , T . inouyei Tateiwa , Conites and Carpolithus ( Figure 3—figure supplement 1; Figure 4—figure supplement 1 ) . Preliminary analysis of the strata yielding Nanjinganthus has recognized abundant palynomorphs . The palynoflora includes Anapiculatisporites sp . , Annulispora folliculosa ( Rogalska ) De Jersey , Contignisporites sp . , Cyathidites australis Couper , C . minor Couper , Deltoidospora sp . , D . minor Pocock , Dictyophyllidites harrisii Couper , D . mortonii ( De Jersey ) Playford and Dettmann , Gleicheniidites sp . , G . senonicus Ross , Ischyosporites sp . , I . variegatus ( Couper ) Schultz , Leptolepidites verrucatus Couper , Manumia delcourtii ( Pocock ) Dybkjær , Neoraistrickia ramosus ( Balme and Hennelly ) Hart , Osmundacidites wellmanii Couper , Polycingulatisporites triangularis ( Bolchovitina ) Playford and Dettmann , Punctatosporites sp . , Retitriletes austroclavatidites ( Cookson ) Döring et al . , R . clavatoides ( Couper ) Döring et al . , Sestrosporites pseudoalveolatus ( Couper ) Dettmann , Striatella scanica ( Nilsson ) Filatoff and Price , S . seebergensis Mädler , Alisporites sp . , A . robustus Nilsson , Callialasporites dampieri ( Balme ) Dev , C . minus ( Tralau ) Guy , C . trilobatus ( Balme ) Dev , C . turbatus ( Balme ) Schulz , Cerebropollenites macroverrucosus ( Thiergart ) Pocock , Chasmatosporites sp . , C . apertus ( Rogalska ) Nilsson , C . hians Nilsson , Classopollis chateaunovi Reyre , C . classoides ( Pflug ) Pocock and Jansonius , C . meyeriana ( Klaus ) De Jersey , C . simplex ( Danzé-Corsin and Laveine ) Reiser and Williams , Cycadopites sp . , C . follicularis Wilson and Webster , Ephedripites sp . , Monosulcites sp . , M . minimus Cookson , Perinopollenites elatoides Couper , Platysaccus sp . , Podocarpidites sp . , Quadraeculina anellaeformis Maljavkina , Q . enigmata ( Couper ) Xu and Zhang , Q . minor ( Pocock ) Xu and Zhang , Spheripollenites psilatus Couper , Vitreisporites pallidus Nilsson ( Figure 2—figure supplement 1 ) ( Santos et al . , 2018 ) . This palynological assemblage suggests a latest Early Jurassic age for Nanjinganthus . The samples were processed by crushing , initial heavy liquid and subsequent magnetic separation at Langfang Yuneng Rock Mineral Separation Technology Service Co . , Ltd . in Langfang City . More than 1000 grains of zircons were hand-picked under a binocular microscope . More than 200 grains of representative zircons for each sample were coined in epoxy resin mounts , ground and polished to expose the central part of zircons , and then photographed under microscope in transmitting light and reflected light . Afterward , the internal structure of the zircons was studied by means of cathodoluminescence ( CL ) imaging at the Beijing Gaonianlinghang Technology Co . , Ltd . in Beijing City . U-Pb dating of these samples were carried out using laser ablation multicollector inductively coupled plasma mass spectrometry ( LA-MC-ICP-MS ) at the Tianjin Institute of Geology and Mineral Resources . The laser beam was 35 μm in diameter . Concentrations of U , Th , and Pb elements were calibrated using SRM 610 as the external reference standard . For the analysis method please see refer Li et al . , 2009 . Repeated analyses of standards yielded precisions at better than 10% for most elements . 207Pb/206Pb , 206Pb/238U , 207Pb/235U and 208Pb/232Th ratios and apparent ages were calculated using ICPMSDataCal software ( Liu et al . , 2010a; Liu et al . , 2010b ) and corrected for both instrumental mass bias and depth dependent elemental and isotopic fractionation using zircon GJ-1 as the external standard . U-Pb age Concordia diagram and histograms apparent ages diagram were drawn by using ISOPLOT ( ver . 3 ) ( Ludwig and Ludwig , 2003 ) . There was no previous isotopic age for the Xiangshan Group . We sampled the layers above the fossiliferous layers ( Figure 1—figure supplement 1 ) and picked zircon grains for U/Pb dating . The zircon grains appeared to be reworked ( Supplementary file 2 ) , with ages ranging from 2738 Ma to 207 Ma ( 67 zircon grains with the concordance >90% from 168 zircon grains ) , and 207 Ma ( two zircon grains ) is the youngest age ( Figure 1—figure supplement 1 ) . Most of the zircon grains were of magmatic origin with internal oscillation belts and high Th/U values , implying a granitic provenance . So the upper limit age of Nanjinganthus is 207 Ma ( the Late Triassic ) . Taking all dating information into consideration , we think that the age of Nanjinganthus falls in the scope ranging from 174 to 207 Ma and is closer to 174 Ma ( the latest Early Jurassic ) . Such a conclusion on absolute age of Nanjinganthus is in agreement with megafossil biostratigraphical analysis ( Cao , 1982; Huang , 1983; Huang , 1988; Ju , 1987 ) , although Neocalamites horridus was previously known only in the Late Triassic ( Zan et al . , 2012 ) . The fossils studied here were collected from an outcrop of the South Xiangshan Formation at a quarry owned by the Xiaoyetian Cement Company Ltd . in the northeastern suburb of Nanjing , Jiangsu , China ( N32˚08′ 19′′ , E118˚58′ 20′′ ) ( Figure 1—figure supplement 1 ) . Plant fossils of the formation have been extensively studied by various scholars ( Sze and Chow , 1962; Cao , 1982; Cao , 1998; Cao , 2000; Wang et al . , 1982; Huang , 1983; Huang , 1988; Ju , 1987 ) , and our collection from the local outcrop indicates that the fossil plants closely associated with Nanjinganthus constitute a flora dominated by Dipteridaceae ( Clathropteris ) and various cycadophytes ( mainly Nilssonia , Ptilophyllum , and Pterophyllum ) , which is consistent with previous works . Some of these associated plant fossils are shown in Figure 3—figure supplement 1 and Figure 4—figure supplement 1 . The specimens were initially photographed using a Sony ILCE-7 digital camera . The sediment covering the specimens was dégaged using a JUN-AIR pneumatic drill , and the details of the fossils were observed and photographed using a Nikon SMZ1500 stereomicroscope equipped with a Digital Sight DS-Fi1 camera . Organically preserved sepals and petals were processed with 40% peroxide for cuticle analysis according to routine methods , and the processed cuticles and cleaned organic material of the sepals and petals were observed and photographed using the Rhod fluorescent light in a Zeiss Z2 Imager with an AxioCam HRc camera . Extended-focus images were generated using the Z-stack function in an AxioVs40 × 64 V4 . 9 . 1 . 0 . The removed cuticles were coated with gold and observed using a Leo 1530 VP scanning electron microscope ( SEM ) , and serial pictures were obtained after the internal details of the flower were exposed through grinding with a pneumatic drill . One of the organically-preserved petals was embedded in resin and sectioned for light microscopy and transmission electron microscopy ( TEM ) . One fragment of the distal portion of a flower embedded in sediments was observed by Micro Computed Laminography ( Micro-CL ) ( Wei et al . , 2017 ) to show the dendroid style embedded in the sediments . All photographs were saved in TIFF format and assembled for publication using Photoshop 7 . 0 . | From oranges to apples , flowering plants produce most of the fruits and vegetables that we can see on display in a supermarket . While we may take little notice of the poppy fields and plum blossoms around us , how flowers came to be has been an intensely debated mystery . The current understanding , which is mainly based on previously available fossils , is that flowers appeared about 125 million years ago in the Cretaceous , an era during which many insects such as bees also emerged . But not everybody agrees that this is the case . Genetic analyses , for example , suggest that flowering plants are much more ancient . Another intriguing element is that flowers seemed to have arisen during the Cretaceous ‘out of nowhere’ . Fossils are essential to help settle the debate but it takes diligence and luck to find something as fragile as a flower preserved in rocks for millions of years . In addition , digging out what could look like a bloom is not enough . It is only if the ovules ( the cells that will become seeds when fertilized ) of the plant are completely enclosed inside the ovary before pollination that researchers can definitely say that they have found a ‘true’ flower . Now , Fu et al . describe over 200 specimens of a new fossil flower that presents this characteristic , as well as other distinctive features such as petals and sepals – the leaf-like parts that protect a flower bud . Called Nanjinganthus , the plant dates back to more than 174 million years ago , making it the oldest known record of a ‘true’ flower by almost 50 million years . Contrary to mainstream belief , this would place the apparition of flowering plants to the Early Jurassic , the period that saw dinosaurs dominating the planet . This discovery may reshape our current understanding of the evolution of flowers . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"evolutionary",
"biology"
] | 2018 | An unexpected noncarpellate epigynous flower from the Jurassic of China |
Due to their strict maternal inheritance in most animals and plants , mitochondrial genomes are predicted to accumulate mutations that are beneficial or neutral in females but harmful in males . Although a few male-harming mtDNA mutations have been identified , consistent with this ‘Mother’s Curse’ , their effect on females has been largely unexplored . Here , we identify COIIG177S , a mtDNA hypomorph of cytochrome oxidase II , which specifically impairs male fertility due to defects in sperm development and function without impairing other male or female functions . COIIG177S represents one of the clearest examples of a ‘male-harming’ mtDNA mutation in animals and suggest that the hypomorphic mtDNA mutations like COIIG177S might specifically impair male gametogenesis . Intriguingly , some D . melanogaster nuclear genetic backgrounds can fully rescue COIIG177S -associated sterility , consistent with previously proposed models that nuclear genomes can regulate the phenotypic manifestation of mtDNA mutations .
The acquisition of the mitochondria by the ancestral eukaryote is one of the most remarkable instances of symbiosis in biology ( Sagan , 1967; Schwartz and Dayhoff , 1978 ) . This symbiosis gave the eukaryotic cell the ability to perform oxidative phosphorylation ( Williams et al . , 2013 ) . In modern eukaryotes , oxidative phosphorylation is carried out by the electron transport chain , which comprises subunits encoded by both nuclear and mitochondrial genomes ( mtDNA ) . Oxidative phosphorylation plays a fundamental role in many eukaryotes and is responsible for meeting most cellular energy demands . In humans , dysfunction of oxidative phosphorylation is associated with many diseases including cancer , diabetes , infertility , and neurodegenerative disorders ( Wallace , 2001 ) . Despite the appearance of a symbiotic relationship , the evolutionary interests of mitochondria can be in conflict with those of nuclear genomes ( Partridge and Hurst , 1998 ) . This conflict arises from the differences in transmission between the two genomes . Whereas the nuclear genome is transmitted in a Mendelian fashion through both sexes , mtDNA is exclusively maternally inherited in most metazoans . Indeed , there are elaborate mechanisms to prevent the inheritance of sperm mitochondria ( Birky , 1995; Al Rawi et al . , 2011; Sato and Sato , 2011; Zhou et al . , 2011; DeLuca and O'Farrell , 2012 ) . Due to this uniparental inheritance , natural selection for mtDNA fitness can operate in females but not in males . As a result of this relaxed selection , 'male-harming' mtDNA mutations are expected to accumulate as long as they are beneficial , neutral or nearly neutral in females , a scenario referred to as the ‘Mother’s Curse’ ( Frank and Hurst , 1996; Gemmell et al . , 2004 ) . However , since decreased male fitness is detrimental for the evolutionary success of the nuclear genome , outbred populations subject to accumulation of male-harming mtDNA mutations can rapidly evolve nuclear genome-encoded suppressors to restore male fitness ( Rand et al . , 2004; Dowling et al . , 2008 ) . Thus , evolutionary theory predicts a conflict-driven molecular arms race between mtDNA and the nuclear genome . These evolutionary predictions have spurred interest into the molecular basis of how mtDNA mutations can manifest as specifically ‘male-harming’ particularly since mitochondrial function would be expected to be similar in most organs of both sexes . One of the best-understood examples of such ‘male-harming’ mtDNA mutations are cytoplasmic male sterility ( cms ) mutants described in many plant species ( Budar et al . , 2003; Chase , 2007 ) . Such cms mtDNA can induce male sterility in hermaphroditic plants that can be cross-pollinated . By reallocating resources for pollen production instead for production of seeds , cms mtDNA increase female fitness – and consequently increase their own fitness ( Budar et al . , 2003 ) . Molecular characterization of cms mutants shows that their ‘male-harming’ ( and female-beneficial ) properties are caused by novel chimeric genes that are created via recombination between mtDNA molecules . Nuclear-encoded suppressors of cms belong to the pentatricopeptide repeat ( PPR ) protein family and use RNA editing to effectively neutralize these male-harming chimeric genes in mtDNA ( Budar et al . , 2003; Chase , 2007; Castandet and Araya , 2012 ) . Animal mtDNA genomes are relatively small compared to those in plants , and the presence of novel chimeric genes is extremely rare in animal mtDNA . Despite these limitations , there is evidence that mutations in animal mtDNA genes might be male-harming . For example , mtDNA haplogroup T in humans is associated with reduced sperm motility ( Ruiz-Pesini et al . , 2000 ) . Polymorphisms in mtDNA that cause this phenotype are hypothesized to be significant contributors to untreatable male subfertility , known to affect 7–10% of men ( Baker , 1994 ) . Recent findings implicate mtDNA variation as a significant contributor to shorter male lifespans in many taxa ( Clancy , 2008; Camus et al . , 2012 ) . Although the phenotypic consequences of these mtDNA mutations are severe in males , it remains unclear whether they are benign in females , i . e . , whether they are specifically ‘male-harming’ mtDNA mutations as predicted by the Mother’s Curse hypothesis . Indeed , in spite of the theoretical predictions , male-harming mtDNA mutations have proven difficult to detect in animals . Several factors account for this difficulty . First , in natural populations , there can be significant indirect selection on mtDNA to maintain male fitness when the fitness of females is dependent on related males ( those with the same mtDNA ) ( Wade and Brandvain , 2009; Hedrick , 2012 ) . Such indirect selection against male-harming mtDNA mutations may be especially stringent in populations with high rates of inbreeding ( Wade and Brandvain , 2009; Hedrick , 2012 ) . Such indirect selection for male fitness might keep male-harming mtDNA mutations at low frequencies in natural populations , making them difficult to detect without deep surveys of natural polymorphism . A second factor that may impede the discovery of male-harming mtDNA mutations in natural populations is their suppression by nuclear genome-encoded loci , which are under stringent selective pressure to restore male fitness ( Rand et al . , 2004; Dowling et al . , 2008 ) . Effects of detrimental mtDNA variants could thus only be evident if they were separated from their nuclear-encoded suppressors . With this goal , many researchers have created combinations of nuclear and mtDNA genomes in D . melanogaster , either from different strains or species ( Rand et al . , 2001; Maklakov et al . , 2006; Rand et al . , 2006 , Dowling et al . , 2007 , 2008; Montooth et al . , 2010; Clancy et al . , 2011; Correa et al . , 2012 ) . Such combinations have revealed instances of dramatic incompatibility , even within species , which results in male but not female lifespan defects ( Camus et al . , 2012 ) . Variation in mtDNA also affects sperm competitiveness in D . melanogaster ( Yee et al . , 2013 ) . While these elegant experiments have revealed mtDNA-dependent affects on male lifespan and fertility , they did not comprehensively assess affects on female life-history traits; it is thus unclear whether these mutations are specifically male-harming . One further limitation of such mtDNA ‘swaps’ is that introduced mtDNA often have several co-inherited mutations . This linkage of different mutations makes it nearly impossible to implicate single mutations as being causal for the incompatibility or to understand their biological mechanism . An exception is a single point mutation in D . melanogaster mtDNA that has been causally linked to male sterility ( Clancy et al . , 2011 ) . However , recent demonstration of this mtDNA haplotype’s negative pleiotropic effects on fertility and aging , both within and between sexes suggests it is not a specifically male-harming mutation ( Camus et al . , 2015 ) . Thus , we currently lack clear examples of specifically male-harming mtDNA mutations in animals . This has also left unanswered the larger question of molecular mechanisms by which mtDNA mutations can manifest their ‘male-harming’ phenotypes . Here , we isolated a mtDNA variant that harbors a single non-synonymous change in subunit II of cytochrome c oxidase ( COIIG177S ) . We find that COIIG177S males have an age– and temperature–dependent decrease in fertility . This decrease in fertility is correlated with a drop in COII enzymatic activity , which remarkably does not result in defects in any other male or female phenotypic traits we measured . Cellular characterization reveals decreased sperm production and function in the mutant males . By combining evolutionary principles with detailed functional characterization , our study thus provides one of the clearest examples of a male-harming mtDNA mutation in animals , and provides insights into the stringent requirements for optimal mtDNA function in sperm development . We further show that the fertility defect in COIIG177S males can be completely suppressed by diverse nuclear backgrounds derived from various D . melanogaster strains , as predicted by the theory of cyto-nuclear genetic conflict .
To isolate male-harming mtDNA mutations , we devised an experimental evolution strategy that was partly inspired by previous strategies to decouple male versus female evolution in D . melanogaster ( Rice , 1996; Rice , 1998 ) . In our strategy , D . melanogaster females are prevented from mating with their male siblings . Instead , virgin females are mated to naïve males from an external stock every generation ( Figure 1A ) . This strategy is predicted to eliminate indirect selection against male-harming mtDNA mutations because the fitness of the population is not dependent on the males carrying the mtDNA mutations; mtDNA are effectively absolved from supporting male fitness ( Wade and Brandvain , 2009 ) . Moreover , continuous replacement of a large fraction of the nuclear genome every generation via crosses to external males practically eliminates the likelihood of evolution of nuclear suppressors during the course of the experiment . In contrast , selection of mtDNA function for female fitness is stringently maintained over the entire course of the experiment . Thus , we anticipated this experimental evolution strategy might provide a permissive environment for accumulation and phenotypic manifestation of mtDNA mutations that are beneficial , neutral , or nearly neutral ( only slightly deleterious ) in females but cause defects in males . 10 . 7554/eLife . 16923 . 003Figure 1 . An experimental evolution strategy to recover male-harming mtDNA mutations . ( A ) In 12 D . melanogaster lines undergoing experimental evolution , 300 virgin female progeny in every generation were prevented from mating with sibling males ( shown as being ‘crossed out’ ) and instead were mated with 100 males from the original stock . ( B ) In six lines undergoing coevolution , we allowed 300 virgin female and 100 male siblings to mate in every generation . Crosses for the 12 experimental and six coevolving lines were carried out for 35 generations . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 00310 . 7554/eLife . 16923 . 004Figure 1—figure supplement 1 . Status of Wolbachia infection in w1118 derived stocks . Using PCR for a Wolbachia-specific WSP gene , we find no evidence of Wolbachia infection in the re-isolated stocks of wildtype and COIIG177S mtDNA . As a positive control for Wolbchia infection , we used the cinnabar brown ( cnbw ) strain of D . melanogaster . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 004 From an ‘original stock’ of w1118 flies , we established 18 independent lines , each consisting of 300 females and 100 males . In 12 of these lines ( experimental lines ) , virgin females were collected every generation and mated with males from the original stock ( Figure 1A ) . Female flies were subjected to this mating scheme continuously for 35 generations over approximately 70 weeks . For the remaining 6 lines ( coevolving lines ) , we allowed sibling males to mate with the females over the course of 35 generations ( Figure 1B ) . Based on published mtDNA mutation rate of 6 . 2 × 10−8 per site per fly generation ( Haag-Liautard et al . , 2008 ) , we anticipated that multiple mutations per mtDNA would be sampled in 35 generations across the 18 lines . Any mtDNA mutations fixed at the end of the experiment must have either been pre-existing heteroplasmic mtDNA mutations that went to fixation ( although the rate of fixation of such heteroplasmic mutations would be slow [Solignac et al . , 1987] ) or arose de novo in the experimental lines . After 35 generations , we subjected flies to phenotypic analyses to assess whether any male-harming mutations had been sampled in the course of our experimental evolution strategy . To assess male viability , we first measured sex ratio in crosses between females with mtDNA subjected to experimental evolution ( either ‘experimental’ or ‘coevolving’ lines ) and control males ( original w1118 stock maintained as a separate population ) . We did not observe a change in male to female ratio in any of the 18 lines , suggesting that any mutations in mtDNA acquired over the 35 generations did not alter male viability ( Figure 2A ) . Next , we measured male fertility by mating males derived from the starting , experimental , and coevolving lines with females from the original stock . For 17 of 18 lines , we did not detect any significant differences in male fertility . However , we found that males from experimental line 7 ( hereafter referred to as ‘EL7’ ) sired significantly fewer progeny compared to males from all the other experimental , coevolved and original lines ( Figure 2B ) . In contrast , EL7 females do not have any significant decline in fertility ( Figure 2C ) . This suggested the possibility of a male-harming mtDNA mutation in the EL7 line that was not deleterious to female fertility . We investigated whether the decline in male fertility we observed manifested over the entire adulthood of EL7 male flies . We found no obvious reduction in male fertility over the first few days of adulthood ( note that F1 progeny do not emerge until day 12 ) in EL7 compared to the other experimental or control lines . Instead , we observed that the reduction in male fertility was age-dependent ( Figure 2D ) . Our findings are consistent with previous observations in which mitochondrial dysfunction is more severe in aged individuals ( Camus et al . , 2012; Tower , 2015 ) . 10 . 7554/eLife . 16923 . 005Figure 2 . A single experimental line with male-specific fertility defects . ( A ) We measured average sex ratio of progeny by mating five females from each of the original , experimental , or coevolving lines , with three males from the original stock . Average sex ratio was found to be close to 50:50 in each of the lines , suggesting that there were no gross viability differences between male and female progeny in any of the lines . All error bars represent standard error of the mean . All experiments were done in replicates of 10 per group . ( B ) We measured male fertility by mating three males from each of the original stock , experimental , or coevolving lines with five females from the original stock , and calculating average number of resulting progeny . Male fertility is represented as a normalized percentage of progeny relative to the original stock . Only one line , experimental line 7 ( EL7 ) showed significant reduction relative to the others . All error bars represent standard error of the mean . All experiments were done in replicates of 10 per group . ( C ) We measured female fertility by mating five females from each of the original , experimental , or coevolving lines with three males from the original stock , and calculating average number of resulting progeny . Like in ( B ) , female fertility is represented as a normalized percentage of progeny relative to the original stock . EL7 female fertility is not significantly different from the original stock . All error bars represent standard error of the mean . All experiments were done in replicates of 10 per group . ( D ) To gain further insight into altered male fertility in EL7 ( B ) , we plotted the cumulative number of progeny sired by ( three ) males as a function of time since initial mating ( 10 replicates per line ) . We find that the cumulative number of progeny sired by EL7 males is normal until day 18 but is subsequently significantly lower than for all other lines . This finding suggests an age-dependent decline in male fertility in the EL7 line . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 005 To identify the putative mutation responsible for decreased male fertility , we performed whole-genome sequencing of the original and EL7 D . melanogaster lines using DNA isolated from a pool of 100 flies for each line . We were able to achieve average coverage of over 1000-fold for the entire mtDNA , except for the highly repetitive , AT-rich ‘control’ region ( Figure 3—figure supplement 1 ) . Analysis of these mtDNA sequences revealed a single missense mutation in EL7 mtDNA , resulting in a glycine to serine substitution at position 177 in subunit II of cytochrome c oxidase ( COIIG177S ) , the fourth complex in the mitochondrial electron transport chain ( Figure 3A–B ) . Interestingly , the COIIG177S mutation did not arise de novo during the course of our experiment . Although 98% of the sequencing reads from EL7 harbored the COIIG177S mutation , 59% of the reads from the original stock also corresponded to the mutant allele . Sanger sequencing confirmed this difference in proportion of wildtype and mutant mtDNA ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 16923 . 006Figure 3 . A single missense mutation ( G177S ) in subunit II of cytochrome c oxidase underlies lower male fertility in EL7 . ( A ) The five complexes of the electron transport chain are schematized . These complexes are comprised of subunits encoded by both nuclear and mitochondrial genomes; the latter are indicated in pink . Subunit II of cytochrome c oxidase ( COII ) with the glycine to serine mutation at position 177 ( G177S ) in Experimental Line 7 is highlighted in bright pink . ( B ) Crystal structure of Bos taurus COII ( in gold ) in complex with COI ( in blue ) ( PDB number: 2OCC ) . Glycine with its side chains at position 177 ( G177 ) in COII is indicated . ( C ) Partial amino acid sequence alignment of COII from representative animal species highlights the conserved glycine residue at position 177 ( boxed in yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 00610 . 7554/eLife . 16923 . 007Figure 3—figure supplement 1 . Whole genome mtDNA sequencing coverage . Coverage of the mtDNA genome sequencing from the original w1118 stock compared to the EL7 strain . With the exception of the AT-rich ‘control region’ ( which has relatively modest coverage ) , the rest of the mtDNA has good coverage in both the original w1118 and the EL7 strains , allowing unambiguous detection of mutations . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 00710 . 7554/eLife . 16923 . 008Figure 3—figure supplement 2 . COIIG177S is present in many experimental and coevolving lines . ( A ) 15 females from each of the coevolving , and ( B ) experimental lines were pooled . From these , DNA was extracted and the COII gene was PCR amplified and sequenced . Sequence traces of the pooled PCR products highlight the nucleotide position with G to A mutation responsible for the G177S mutation in COII ( indicated by a vertical line ) . These reveal a qualitative measure of the degree of heteroplasmy at G177S . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 00810 . 7554/eLife . 16923 . 009Figure 3—figure supplement 3 . COIIG177S is present at variable levels in flies of the original w1118 stock . We extracted DNA from 6 individual females from the original w1118 stock and carried out PCR and sequencing of the COII mtDNA gene . We present sequence traces highlighting the nucleotide position where the G to A mutation in COII occurs ( indicated by a vertical line ) . Note that first two individuals are clearly heteroplasmic . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 009 A mixture of wildtype and mutant mtDNA can exist in the same population in two ways . First , a population can consist of homoplasmic ( carrying genetically identical mtDNA ) wildtype and mutant individuals . Alternatively , the two mtDNA can be present in the same individual in a state of heteroplasmy . Sanger sequencing of single individuals from our original line revealed the presence of heteroplasmic individuals as well as flies that were homoplasmic wildtype and mutant ( Figure 3—figure supplement 3 ) . Therefore , 59% reflects the average combined frequency of the COIIG177S mutation from both heteroplasmic and homoplasmic mutant flies in the population . The mutant allele’s 59% frequency among reads from the original stock suggested that the COIIG177S mutation was present before initiation of the experimental lines . If this model is correct , we would expect that the COIIG177S mutation should also be found in some of the other control and experimental lines in addition to EL7 . Indeed , we found this to be the case ( Figure 3—figure supplement 2 ) . However , EL7 is unique in being the only population in which the COIIG177S mutation is nearly fixed , whereas it varies from 0% to more than 50% in other experimental populations , which do not suffer any overt signs of male sterility at least at a population-level ( Figure 2B , Figure 3—figure supplement 2 ) . These data are consistent with the hypothesis that COIIG177S mutation causes defects in male fertility but only when present at very high levels . Our findings are consistent with similar observations in pathogenic mtDNA variants in human disease ( Sobenin et al . , 2014 ) . Our finding that EL7 nearly fixed a mtDNA mutation that was already pre-existing provided another opportunity to test the association of the COIIG177S mutation with decrease in male fertility . We reasoned that homoplasmic mutant males re-isolated from the original stock should also have reduced fertility despite the fact that these males have not undergone 35 generations of experimental evolution like EL7 . Consistent with this expectation , we were able to re-isolate homoplasmic COIIG177S males , which sire fewer progeny compared to homoplasmic wildtype males derived from the same original stock ( Figure 4A ) ( mutant and wildtype lines re-established from single females ) . These data demonstrate that the homoplasmic COIIG177S mutation is necessary and sufficient to explain the observed decreased male fertility in EL7 males . These re-isolated COIIG177S homoplasmic mutant mtDNA males also recapitulated the age-dependent decline in male fertility ( Figures 2D , 4B ) . 10 . 7554/eLife . 16923 . 010Figure 4 . Males with COIIG177S mutation have decreased fertility at elevated temperatures . ( A ) To measure male fertility , we compared the fertility of males homoplasmic for either wildtype or COIIG177S mutant mtDNA reestablished from the original stock . For each experiment , three males from homoplasmic stocks were mated with five females homoplasmic for wildtype mtDNA . Male fertility is presented as a normalized percentage of progeny produced , relative to the number produced by wildtype mtDNA males . COIIG177S mutant mtDNA males produce fewer progeny at 25°C , but this is further reduced at 29°C . To measure female fertility , five females from homoplasmic wildtype or COIIG177S mutant mtDNA stocks were mated with three males with wildtype mtDNA . Female fertility is presented as a normalized percentage of progeny produced , relative to wildtype mtDNA females . All error bars represent standard error of the mean . All experiments were done in replicates of 10 per group . ( B ) Number of progeny sired by wildtype or COIIG177S mutant males at 25°C is cumulatively plotted as a function of time since initial mating ( 10 replicates per line ) . Actual number of cumulative progeny from each replicate is indicated by dots at given time points . ( C ) Number of progeny sired by wildtype or COIIG177S mutant males at 29°C is cumulatively plotted as a function of time since initial mating ( 10 replicates per line ) . The reduction in male fertility is more significant at the higher temperature . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 01010 . 7554/eLife . 16923 . 011Figure 4—source data 1 . Near homoplasmy of wildtype and COIIG177S mtDNA in re-isolated lines . Summary of data from duplex sequencing of COIIG177S mutant and wildtype mtDNA strains re-isolated from the ancestral , heteroplasmic w1118 stock . Individual fly heads from the mutant COIIG177S mtDNA stock ( Mut Fly 1 , 2 , 6–10 ) were subjected to duplex sequencing . Pooled samples represent sequencing data generated by sequencing combined DNA from 10 individuals from either the mutant or wildtype mtDNA strains . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 011 We characterized the nature of the heteroplasmy in the re- isolated COIIG177S mutant mtDNA lines by sequencing a pool of COIIG177S mutant mtDNA and a separate pool of re-isolated wildtype mtDNA adult fly heads using a Duplex Sequencing strategy ( Kennedy et al . , 2014 ) followed by hybrid capture to significantly enrich for mtDNA ( Figure 4—source data 1 ) . Under this strategy , mtDNA is sequenced at very high depth of coverage , labeling individual DNA molecules and sequencing each one multiple times in order to distinguish true mutations from sequencing errors . This strategy allows for sensitive evaluation of heteroplasmic mtDNA mutations . We found no evidence for wildtype mtDNA sequence in a pool of flies that we had re-isolated to enrich for the COIIG177S mutant mtDNA ( 0 duplex consensus reads out of >7000 ) . In contrast , we found no evidence for the COIIG177S mutant mtDNA in flies ( 0 reads out of >7000 ) that we had re-isolated from the same original w1118 stock to enrich for wildtype mtDNA . We were cognizant that although the pooled COIIG177S mutant flies showed no evidence of wild-type sequence , some heteroplasmy may nevertheless persist in individual flies . We therefore also sequenced 7 individual flies from the COIIG177S mutant mtDNA pool to high depth of coverage . For six flies , we found no reads ( out of >8 , 000 ) corresponding to wildtype mtDNA , whereas in one fly we uncovered seven out of >11 , 000 reads corresponding to wildtype mtDNA ( <0 . 1% ) . We therefore conclude that the re-isolated strains are almost completely homoplasmic for either COIIG177S mutant mtDNA or wildtype mtDNA . We therefore used these re-isolated strains for all subsequent phenotypic analyses . Although we had set out to recover de novo mtDNA mutations that are male-harming using our experimental evolution strategy , we instead recovered what appears to be a pre-existing heteroplasmic mtDNA mutant , which fixed in only one of the 12 experimental lines . Therefore , we cannot attribute the isolation of this mutant to our scheme . Future work will be needed to determine the effectiveness of our experiment evolution scheme in recovering male-harming mutations , perhaps aided by experimentally increasing mtDNA mutation rates . For the rest of the manuscript , we focus on detailed characterization of the COIIG177S mutation to determine its cellular consequences and to determine whether it is specifically male-harming . All our analyses so far suggest that COIIG177S is a specifically male-harming mtDNA mutation . Hence , we decided to further characterize the cellular and molecular basis underlying its detrimental effects . Previous experiments have shown that the phenotypic effects of mtDNA mutation can be exacerbated with increased stress , including higher temperatures ( Hoekstra et al . , 2013; Chen et al . , 2015 ) . We therefore investigated the effect of higher temperature on the fertility defect in COIIG177S males . We found that COIIG177S males are almost completely sterile when raised at 29°C instead of at 25°C ( Figure 4A ) . Furthermore , we found that this defect manifests in both old and young males ( Figure 4C ) . Thus , higher temperature provides a more sensitized condition to evaluate the various consequences of the COIIG177S mtDNA mutation as previously observed ( Hoekstra et al . , 2013 ) . Next , we investigated the molecular consequences of the COIIG177S mutation . COII is a subunit of cytochrome C oxidase ( COX ) , which oxidizes the reduced form of cytochrome c . The glycine residue at position 177 is found in a loop of COII’s structure where it comes in very close proximity to subunit I of the enzymatic complex ( Figure 3B ) . Based on the fact that G177 is highly conserved across metazoans ( Figure 3C ) , we hypothesized that the G177S mutation affects either an intrinsic function of COII or its interaction with other proteins in complex IV of the electron transport chain . To understand the biochemical consequences of the COIIG177S mutation , we measured COX activity from whole fly lysates of flies grown at 25°C . Although the COX activity in COIIG177S mutant flies was slightly lower than flies with wild-type mtDNA , this defect was not statistically significant , even in old flies ( Figure 5A ) . We therefore measured COX activity from flies raised at 29°C where male fertility is most significantly impaired . Our analyses revealed an approximately 20% decrease in COX enzymatic activity in COIIG177S mutants grown at 29°C in both male and female flies ( Figure 5B ) . Given that the activity of the electron transport chain is coupled to ATP synthesis , we also measured ATP levels at 29°C . In contrast to COX activity , we did not observe a decrease in ATP levels in the COIIG177S mutants ( Figure 5C ) . However , we did observe a mild but significant decrease in reactive oxygen species ( ROS ) levels ( Figure 5D ) . Thus , impaired COX activity and reduced ROS levels , but not ATP production , appear to correlate with loss of male fertility . 10 . 7554/eLife . 16923 . 012Figure 5 . Biochemical consequences of COIIG177S mtDNA mutation . We measured COX activity from young ( 3–4 days ) or old ( 21 days ) flies raised at 25°C ( A ) , or young flies raised at 29°C ( B ) . ( C ) We measured ATP levels from 3–4 day old males raised at 29°C . ( D ) We measured ROS levels from 3–4 day old males raised at 29°C . 4–5 replicates per group for all experiments . Data is normalized to wildtype flies for each group in all experiments . Error bars represent standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 012 Our finding that COIIG177S mutants result in lower COX activity of both males and females prompted us to evaluate several other phenotypes commonly seen in flies with mitochondrial impairment . Mitochondrial dysfunction is often associated with aging defects in many species ( Cho et al . , 2011; Tower , 2015 ) . Naturally occurring variation in mtDNA is also known to affect aging in D . melanogaster ( Rand et al . , 2006; Clancy , 2008; Camus et al . , 2012 ) . We therefore assayed the lifespan of wildtype and COIIG177S mutant males and females at both 25°C and 29°C . We did not observe any differences in lifespan , even at the sensitized higher temperature between wildtype and mutant flies in either males or females ( Figure 6 ) . 10 . 7554/eLife . 16923 . 013Figure 6 . COIIG177S mtDNA mutation does not significantly affect male or female lifespan . We measured average lifespan of males and females at 25°C ( A , B ) or 29°C ( C , D ) . N = 100 flies per group . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 01310 . 7554/eLife . 16923 . 014Figure 6—figure supplement 1 . COIIG177S mutation does not affect heat tolerance of flies raised at 29°C . Time it takes males ( A ) or females ( C ) to become paralyzed within allotted time of 6 min upon 39°C heat shock . Time it takes males ( C ) or females ( D ) to recover after 39°C heat shock for 6 min . All flies raised at 29°C . Error bars represent standard error of the mean . N = 16 flies per group . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 01410 . 7554/eLife . 16923 . 015Figure 6—figure supplement 2 . COIIG177S mutation does not affect bang sensitivity . Amount of time it takes males raised at 25°C ( A ) or ( B ) 29°C to right themselves after being vortexed at maximum speed for 10 s . Amount of time it takes females raised at 25°C ( C ) or ( D ) 29°C to right themselves after being vortexed at maximum speed for 10 s . Error bars represent standard error of the mean . N = 18–20 flies per group . Note that there is no difference in females carrying wildtype or COIIG177S mutant mtDNA . There is a trend to increased bang sensitivity in males carrying COIIG177S mutant mtDNA ( B ) but this is not statistically significant . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 015 Sensitivity to physiological stress and neuronal function are also associated with mitochondrial dysfunction ( Fergestad et al . , 2006; Pieczenik and Neustadt , 2007; Ugalde et al . , 2007; Distelmaier et al . , 2009; Celotto et al . , 2011 ) . We therefore assayed heat intolerance , and ‘bang sensitivity’ , a measure of neuronal dysfunction ( Burman et al . , 2014 ) . There were no statistically significant differences in these traits in either males or females , even in aged flies grown at 29°C , the highly sensitized and susceptible condition ( Figure 6—figure supplements 1 , 2 ) . Thus , despite the 20% reduction in COX activity in both males and females , the phenotypic consequences of COIIG177S in D . melanogaster appear to be largely benign , with deleterious effects confined to male fertility alone . We next investigated the biological basis of the decreased male fertility in COIIG177S mtDNA-bearing males . Male sterility could either result from defects in the ability of males to mate or can be a result of defective sperm development/function or a combination of both . In order to distinguish between these two possibilities , we took advantage of the fact that transfer of male accessory gland peptides during mating induces egg-laying in females ( Wolfner , 1997 ) . We find that mating wildtype females with COIIG177S males induces egg-laying just as robustly as in females mated with males carrying wildtype mtDNA ( Figure 7A ) . However , almost all of the eggs laid by females mated with COIIG177S mutant males were unfertilized and failed to hatch ( Figure 7A ) . Taken together , these data suggest that the decreased fertility in COIIG177S males occurs due to defects in sperm development or function , and not due to mating ability . We therefore sought to determine the nature of cellular defects in COIIG177S male testes . 10 . 7554/eLife . 16923 . 016Figure 7 . Normal mating but defective sperm development in COIIG177S mutant mtDNA males . ( A ) We measured whether the COIIG177S mtDNA mutation affected the mating success of males . We measured eggs laid by virgin wildtype females that were either unmated , or mated with 2–5 day old wildtype , or COIIG177S mutant males at 29°C . We determined the fraction of eggs hatched by counting unhatched eggs and larva 24 hrs after eggs were laid . All experiments were done in replicates of 6 per group . Error bars represent standard error of the mean . Our results show that the number of eggs laid after mating to wildtype mtDNA males is not significantly different from those mated to COIIG177S mutant mtDNA males; in the latter case , most of the eggs are unfertilized and do not hatch . ( B , C ) We present maximum projection representative images of DAPI stained testis from 2–5 day old wildtype mtDNA male flies grown at 29°C at early ( B ) , middle ( B’ ) , and late ( B’’ ) needle stage of sperm development . Note that the sperm are organized during early needle stage ( arrow ) and then break up into individual sperm by late needle stage ( arrowheads ) . We also present maximum projection images of DAPI stained testis from COIIG177S mutant mtDNA male flies grown at 29°C at early ( C ) , middle ( C’ ) , and late ( C’’ ) needle stage of sperm development . Note also that the sperm in COIIG177S mutant are ‘clumped’ and disorganized early in the needle stage and remain so through remainder of spermiogenesis ( arrow ) . Scale bar , 20 μm . ( D , E ) Representative DAPI stained images of whole testis ( outlined in dotted line ) from 2–5 day old virgin wildtype mtDNA ( D ) and COIIG177S mutant mtDNA males ( E ) raised at 29°C . For orientation , in both images , we identify the tip of the testis ( where germ stem cells reside ) as well as the seminal vesicle ( the storage organ for mature sperm ) . Note the much smaller seminal vesicle size in the mutant males . Scale bar , 50 μm . ( F ) Quantification of the seminal vesicle size , as measured by cross-sectional area , normalized to wildtype . Average calculated from 5–7 testes . Error bars represent standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 01610 . 7554/eLife . 16923 . 017Figure 7—figure supplement 1 . Aged males recapitulate the male sterility defects at 25°C . ( A ) Representative DAPI stained images of seminal vesicles ( outlined in dotted line ) from w1118 males containing either wildtype mtDNA or COIIG177S mutant mtDNA . We analyzed both young virgin males ( 3–4 day adults ) raised at 25°C or 29°C , as well as aged virgin males ( 21 day old adults ) raised at 25°C . Note the much smaller seminal vesicle sizes in COIIG177S mutant mtDNA containing aged males raised at 25°C as well as young males raised at 29°C . Scale bar , 100 μm . ( B ) Quantification of the seminal vesicle size , as measured by cross-sectional area , normalized to aged wildtype mtDNA males raised at 25°C . Average calculated from 5–7 testes . Error bars represent standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 017 Sperm development occurs within a cyst in which 64 sperm nuclei share a common cytoplasm . These sperm undergo individualization during the needle stage near the terminal epithelium of the testis . Examination of testes stained with the DNA marker DAPI revealed clear late needle-stage defects in COIIG177S males . These mutant sperm failed to individualize properly and instead formed tangled clumps ( Figure 7B–C ) . Consistent with this developmental defect , there is a significant reduction in the number of mature sperm that are stored in the seminal vesicles of COIIG177S males raised at 29°C , as indicated by the decrease in the size of the seminal vesicles ( Figure 7D–F ) . In addition , sperm that could be isolated from COIIG177S males have reduced motility ( Videos 1 , 2 ) . 10 . 7554/eLife . 16923 . 018Video 1 . Sperm motility assays in w1118 males carrying wildtype mtDNA . From a representative male grown at 29°C . Sperm stained with mitotracker Green ( green ) , which stains immotile sperm and mitotracker CMS Rox ( red ) , which stains motile sperm preferentially ( related to Figure 8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 01810 . 7554/eLife . 16923 . 019Video 2 . Sperm motility assays in w1118 males carrying COIIG177S mutant mtDNA . From a representative male grown at 29°C . Sperm stained with mitotracker Green ( green ) , which stains immotile sperm and mitotracker CMS Rox ( red ) , which stains motile sperm preferentially ( related to Figure 8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 019 Although we used the higher temperature to provide a more sensitized assay to measure male fertility , we were aware of the possibility that the decreased fertility and the reduced seminal vesicle size could be solely the result of the elevated temperature ( 29°C ) . We therefore carried out similar experiments in aged males at 25°C , which also had displayed a reduction in fertility . Just like young COIIG177S males at 29°C , we found that aged COIIG177S males at 25°C also had reduced seminal vesicle size compared to aged males with wildtype mtDNA ( Figure 7—figure supplement 1 ) . In contrast , the seminal vesicle sizes of young males with wildtype mtDNA were the same , whether they were raised at 25°C or 29°C . Together , our results suggest that the decreased male fertility and reduced seminal vesicle size is not due to a confounding effect of higher temperature , but rather the result of the COIIG177S mtDNA mutation . Examining the male infertility phenotype further , we found that mitotracker green , a mitochondrial membrane potential-independent dye , preferentially stains immotile sperm whereas mitotracker CMXRos , a mitochondrial membrane potential-dependent dye , preferentially stains motile sperm in wild type testes . Sperm from COIIG177S mutant males showed similar mitotracker green staining but significantly reduced CMXRos staining overall compared with wildtype sperm ( Figure 8 ) , suggesting that the lack of motility likely results from dysfunctional mitochondrial activity . Taken together , these data demonstrate that decreased COX activity in COIIG177S males correlates with the cellular defects in sperm development and motility in COIIG177S mutants . 10 . 7554/eLife . 16923 . 020Figure 8 . Lower motility of sperm from COIIG177S mutant mtDNA males . Sperm from wildtype mtDNA ( A ) and COIIG177S mutant mtDNA ( B ) males raised at 29°C stained with mitotracker Green ( green ) , which stains immotile sperm and mitotracker CMS Rox ( magenta ) , which stains motile sperm preferentially . Scale bar , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 020 Male-harming mtDNA mutations like COIIG177S are predicted to be detrimental for the evolutionary success of the nuclear genome . In the face of such mtDNA mutations , it is expected that nuclear genomes might have evolved suppressors to restore male fitness ( Rand et al . , 2004; Dowling et al . , 2008; Wolff et al . , 2014 ) . To test this evolutionary hypothesis , we generated males heterozygous for varied nuclear genomes by crossing COIIG177S mtDNA-bearing females with males from a number of D . melanogaster strains , collected from different global populations . We then assayed the fertility of the resulting male progeny that carried the COIIG177S mtDNA but were heterozygous for the nuclear genome . To maximize the sensitivity of our assays , we performed these fertility assays in young males at 29°C . Surprisingly , we found that the nuclear genomes from many of the strains we tested were able to completely restore male fertility in COIIG177S males ( Figure 9A , Figure 9—figure supplement 1 ) . 10 . 7554/eLife . 16923 . 021Figure 9 . Dominant suppression of COIIG177S associated male fertility defects by diverse nuclear genome backgrounds . ( A ) We measured male fertility of heterozygous males derived from crosses of w1118 females ( carrying wildtype or COIIG177S mutant mtDNA ) to males from a variety of different nuclear backgrounds ( i . e . , Oregon R , Arizona , Congo isoline 18 , Congo isoline 26 ) . In each case , three heterozygous males were mated with 5 w1118 females at 29°C ( 10 replicates per group ) . Male fertility was measured as average number of resulting progeny , normalized as a percentage of progeny of corresponding males with wildtype mtDNA . Raw progeny numbers are given in Figure 9—figure supplement 1 . Most nuclear backgrounds show mild to complete suppression of the male fertility defects observed in w1118/w118 males carrying COIIG177S mutant mtDNA ( **p<0 . 05 ) . Error bars represent standard error of the mean . ( B ) We examined lifetime male fertility in the w1118/Oregon R background at 25°C ( note that male fertility is generally higher at lower temperatures ) to quantify the suppression of male sterility . ( C ) We measured male fertility of heterozygous males derived from crosses of w1118 females ( carrying wildtype or COIIG177S mutant mtDNA ) to males from six DGRP strains , at 29°C ( 10 replicates per group ) as in ( A ) . At least five of the tested DGRP strains show partial to complete suppression of male sterility . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 02110 . 7554/eLife . 16923 . 022Figure 9—figure supplement 1 . Dominant suppression of COIIG177S-associated male fertility by diverse nuclear genome backgrounds . ( A–C ) We measured male fertility of heterozygous males derived from crosses of w1118 females ( carrying wildtype or COIIG177S mutant mtDNA ) to males from a variety of different nuclear backgrounds . In each case , 3 heterozygous males were mated with 5 w1118 females at 29°C ( 10 replicates per group ) except in panel ( B ) when fertility was measured at 25°C . Male fertility was measured as average number of resulting progeny; this data is presented as normalized to the wildtype mtDNA background in Figure 9 . Most nuclear backgrounds show mild to complete suppression of the male fertility defects observed in w1118/w118 males carrying COIIG177S mutant mtDNA ( **p<0 . 05 ) . Error bars represent standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 022 Male fertility is highly sensitive to temperature , with different strains exhibiting different threshold of tolerance ( Rohmer et al . , 2004; David et al . , 2005; Hoekstra et al . , 2013 ) . Hence , the restoration of fertility that we observe could simply reflect greater tolerance to high temperature independent of the COIIG177S mtDNA . To address this possibility , we measured total fertility of males until 21 days of age at 25°C . We found that the Oregon R nuclear background suppresses not only the young male sterility at 29°C but also male fertility at 25°C ( Figure 9B , Figure 9—figure supplement 1 ) . Thus , we conclude that the Oregon R nuclear background encodes a bona fide suppressor of COIIG177S mtDNA-mediated male infertility . We currently don’t know the identity of the suppressor loci or whether these are specific only to the COIIG177S mtDNA mutation . For instance , these suppressors might employ a general mechanism that allows for rescue of male fitness defects caused by other mtDNA mutations . Under either scenario , we hypothesized that COIIG177S suppressor loci may be more abundant across D . melanogaster strains . We used lines from the DGRP collection ( Mackay et al . , 2012 ) , a set of fully sequenced inbred lines derived from a single natural population , to ask whether there is a lot of standing variation in the ability of nuclear genomes to suppress effects of the COIIG177S mutation . We found that DGRP line 861 was able to completely restore male fertility and most other lines were able to at least partially restore male fertility in males carrying COIIG177S mutation ( Figure 9c , Figure 9—figure supplement 1 ) . These data suggest that genetically dominant nuclear suppressors of the COIIG177S male fertility defects are widespread in natural populations . The different penetrance of the rescue also suggests that the suppression of COIIG177S associated male infertility is likely to involve multiple loci . We examined in more detail the cellular basis of ‘rescued’ male fertility in OregonR-w1118 heterozygous males bearing the COIIG177S mtDNA . We found that sperm development and motility were partially rescued in young males raised at 29°C ( Figure 10A–E , Figure 10—figure supplement 1 , Video 3 ) and aged males at 25°C ( Figure 7 —figure supplement 1 ) , consistent with the restoration of fertility that we previously observed ( Figure 9A , B ) . We also evaluated the biochemical consequence of such nuclear suppression of COIIG177S mtDNA’s male sterility to distinguish between two possible mechanisms of suppression . Under the first alternative , restoration of fertility in COIIG177S males might be due to higher baseline COX activity in the suppressor background , which can compensate for the hypomorphic COIIG177S and therefore be sufficient for sperm development and motility . Under this scenario , there would still be a difference in COX activity between wildtype and COIIG177S flies . A second alternative is that there is no difference in COX activity in flies carrying wildtype or COIIG177S mtDNA . When we measured COX activity in the suppressed nuclear background , we found the second scenario to be true i . e . , COX activity is the same in wildtype and COIIG177S mtDNA-bearing males ( Figure 10F ) . This result was also observed for three additional suppressor lines we tested ( Figure 10—figure supplement 2 ) . These data suggest that in the suppressed background , COIIG177S mtDNA is as functionally effective as wildtype mtDNA and is indicative of a compensatory mechanism that does not rely simply on higher baseline levels of COX activity . 10 . 7554/eLife . 16923 . 023Figure 10 . w1118/OreR heterozygous males carrying COIIG177S mutant mtDNA show rescued sperm development and COX activity . Representative DAPI stained images of the late needle stage of sperm development from w1118/w1118 males carrying either wildtype mtDNA ( A ) , or COIIG177S mtDNA ( B ) compared to heterozygous w1118/OreR males carrying COIIG177S mtDNA ( C ) . Numbers of examined males with defects in sperm development are indicated . Arrow points to ‘clumped’ sperm . Note the presence of many individualized sperm ( arrowheads ) in A and C but few in B . Scale bar , 20 μm . We stained sperm from homozygous w1118/w1118 males ( D ) and w1118/OreR heterozygous males ( E ) , both carrying COIIG177S mutant mtDNA with mitotracker Green ( green ) and mitotracker CMS Rox ( magenta ) , which stain immotile and motile sperm respectively . All flies were raised at 29°C . Scale bar , 20 μm . ( F ) COX activity measured from young ( 3–4 day old ) w1118/w1118 males and w1118/OreR heterozygous males both carrying COIIG177S mutant mtDNA . All flies were raised at 29°C . Data is normalized to flies with wildtype mtDNA in the corresponding nuclear background . Error bars represent standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 02310 . 7554/eLife . 16923 . 024Figure 10—figure supplement 1 . Seminal vesicle size is restored in aged 'suppressor' backgrounds in males containing COIIG177S mutant mtDNA raised at 25°C . ( A ) Representative DAPI stained images of seminal vesicles ( outlined in dotted line ) from either w1118/w1118 or w1118/OregonR males containing either wildtype mtDNA or COIIG177S mutant mtDNA . All males were raised at 25°C and allowed to age before dissection . Note the much smaller seminal vesicle sizes in COIIG177S mutant mtDNA containing aged w1118/w1118 males are completely restored in the w1118/OregonR males . Scale bar , 100 μm . ( B ) Quantification of the seminal vesicle size , as measured by cross-sectional area , normalized to aged wildtype mtDNA males raised at 25°C ( data from Figure 7—figure supplement 1 is reproduced here for easy comparison ) . Average calculated from 5–7 testes . Error bars represent standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 02410 . 7554/eLife . 16923 . 025Figure 10—figure supplement 2 . COX activity is specifically restored in ‘suppressor’ nuclear backgrounds . We measured COX activity from young ( 3–4 day old ) w1118/w1118 males carrying COIIG177S mutant mtDNA; the latter showed a 20% reduction in COX activity . We also measured COX activity in three heterozygous nuclear backgrounds ( similar to Figure 10F ) and found there was no reduction in COX activity in flies carrying COIIG177S mutant mtDNA . Instead , in one instance , in w1118/Congo26 males , we observed a significant enhancement of COX activity associated with the COIIG177S mutant mtDNA . All flies were raised at 29°C . Data is normalized to flies with wildtype mtDNA in the w1118/w1118 nuclear background . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 02510 . 7554/eLife . 16923 . 026Video 3 . Sperm motility assays in w1118/Oregon R males ( 'nuclear suppressor' background ) carrying COIIG177S mutant mtDNA . From a representative male grown at 29°C . Sperm stained with mitotracker Green ( green ) , which stains immotile sperm and mitotracker CMS Rox ( red ) , which stains motile sperm preferentially ( related to Figure 10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16923 . 026
Evolution by natural selection rests on a simple but fundamental premise that the phenotypic manifestation of a genotype occurs in the same individual that transmits it to the next generation . Remarkably , the mitochondrial genome ( mtDNA ) does not abide by this rule since it is inherited and phenotypically manifests in both sexes , but is exclusively transmitted through females in most animals and plants . Thus , the phenotypic manifestation of mtDNA is decoupled from its transmission in males . Consequently , natural selection is ineffective at directly removing mtDNA mutations that are specifically harmful to males but not females . While the mtDNA phenotype/transmission decoupling clearly predicts existence of male-harming mutations ( Mother’s Curse ) ( Frank and Hurst , 1996; Gemmell et al . , 2004 ) , a systematic study of such mutations and their underlying mechanisms is hampered by the relative paucity of such mutations in animals . We devised an experimental evolution scheme intended to absolve mtDNA of supporting male function while both maintaining selection on female mtDNA function as well as reducing the possibility of nuclear suppression of male-harming mtDNA . While we did recover a bona fide male-harming COIIG177S mtDNA mutation via this scheme , we were subsequently surprised to discover that this was a pre-existing mtDNA mutation that was already present at moderate frequencies in our starting laboratory strain of w1118 . It is possible that the experimental evolution scheme might have created the permissive conditions that allowed this mtDNA mutation to rise to fixation in one strain . However , the fact that it did not rise to fixation in the other experimental populations means that we cannot conclude that the experimental strategy itself was causally important for the fixation of this male-harming mtDNA mutant . Future work on fine-tuning the strategy can address its effectiveness . For example , the scheme can be improved by increasing mtDNA mutation rates to improve sampling of de novo mutations , perhaps by employing a mutator mtDNA polymerase before initiating the experimental evolution scheme ( Trifunovic et al . , 2004 ) . Alternatively , random chemical mutagenesis might also be readily applied here in our scheme . Although such chemical mutagenesis will affect both mtDNA and nuclear genomes initially , backcrossing to ‘original stock’ males in our scheme will rapidly replace most mutated nuclear genes with wildtype versions , while enabling the transmission of mtDNA mutations that do not affect female fitness . Finally , combining the crossing scheme we propose with the targeted restriction endonuclease strategy to generate mtDNA mutations ( Xu et al . , 2008 ) might be a facile means to discriminate specifically ‘male-harming’ mtDNA mutations from those that grossly impair mtDNA function . Our phenotypic assays did not provide any evidence for the COIIG177S mutation being beneficial in females . Furthermore , the mutation rose to high frequency in only one out of the twelve experimental lines , suggesting that the mutation is likely neutral or nearly neutral in females and therefore became nearly fixed through drift rather than selection . These data are consistent with the ‘selective sieve’ model , which predicts that even mutations that are neutral or nearly neutral in females but deleterious in males might be common because they cannot be effectively removed by natural selection ( Innocenti et al . , 2011 ) . Our survey of the sequenced genomes from a large panel of 290 D . melanogaster strains and isofemale lines did not reveal any other lines besides our w1118 stock that harbored COIIG177S ( Richardson et al . , 2012 ) . Thus , COIIG177S might be rare in natural populations . The homoplasmic COIIG177S mutation results in a 20% decline in COX activity at 29°C . Remarkably , we observed no phenotypic effects of this decline in COX activity on other phenotypic traits in both males and females , even at high temperatures . The COX activity decline specifically impaired sperm production and function . This establishes COIIG177S as one of the first bona fide male-harming mtDNA mutations in animals . We considered the possibility that the phenotypic effects we have observed on male fertility are attributable to linked but unassayed changes in the 4-kb long AT-rich D loop control region of mtDNA . Indeed , the D-loop is one of the most rapidly evolving segments of mtDNA in D . melanogaster , but its highly repetitive nature challenges sequence characterization . However , our analyses allow us to directly implicate the COIIG177S mtDNA mutation as being causally linked to the phenotypes we have observed . Not only do we observe a perfect correlation of the male fertility phenotype with the reduction of COX activity , but we also observe a near complete restoration of male fertility upon replenishment of COX activity in the suppressor strains . Surprisingly , the decline in COX activity in COIIG177S flies was not associated with a corresponding decline in ATP production . Since we were unable to reliably measure either COX activity or ATP production in testes , we cannot rule out the possibility that ATP production is specifically impaired in the male germline . However , our findings suggest a different molecular consequence of lowered COX activity may be responsible for defects in sperm development; we did observe decreased reactive oxygen species ( ROS ) production in COIIG177S mutants . Our findings suggest that perhaps alterations in ROS levels might underlie the defects we see in sperm production and function . ROS have been shown previously to act as a signaling molecule to control the cell cycle checkpoint as well as the differentiation of hematopoietic progenitors during development in D . melanogaster ( Owusu-Ansah et al . , 2008; Owusu-Ansah and Banerjee , 2009 ) . These data leave open the possibility of ROS similarly acting as a signaling factor during sperm development . We hypothesize that male fertility may be a common target of male-harming mtDNA mutations because the reproductive tissues are highly sexually dimorphic . This hypothesis is consistent with previous data , which showed that naturally occurring variation in D . melanogaster mtDNA largely affects expression of male-expressed genes in the testis and the accessory gland ( Innocenti et al . , 2011 ) . Two separate mutations in mtDNA are known to cause male sterility in D . melanogaster ( Xu et al . , 2008; Clancy et al . , 2011 ) . A single amino acid mtDNA mutation ( A278T ) in Cytochrome B of complex III renders males sterile; the primary defect appears to be at the level of spermatid individualization ( Clancy et al . , 2011 ) . In contrast , a single amino acid mutation in Cytochome Oxidase I ( R301L ) causes male sterility primarily due to a sperm storage defect . However , the effects of these mutations on female fitness have not yet been comprehensively addressed so we cannot attribute them to be specifically ‘male-harming’ . How might mtDNA function be different in testis compared to other tissues ? It is possible that the testis has a differential requirement for COX activity . According to this hypothesis , although all tissues suffer from the relative reduction in COX activity , testis function is specifically impacted because it has a lower threshold of tolerance for a relative reduction in COX activity . In particular , COX activity might be required to facilitate the dramatic morphological changes that mitochondria undergo during sperm development in D . melanogaster ( Fuller , 1998 ) . If this hypothesis is correct , mildly hypomorphic mutations like COIIG177S might generally impair male fertility and thus represent a common mechanism of male-harming mtDNA mutations . Alternatively , the existence of a number of nuclear-encoded testis-specific components of the electron transport chain , including subunits of cytochrome C oxidase , provide a number of interacting partners that might exhibit testis-specific genetic incompatibility with the COIIG177S mutation ( Tripoli et al . , 2005; Gallach et al . , 2010 ) . Since we were unable to reliably measure COX activity in dissected testes of wildtype and mutant flies , we are unable to address whether the decrease in COX activity is more severe in testes than other tissues . Our study has identified one mtDNA mutation , COIIG177S , whose phenotypic effects appear to be restricted to the male germline . However , other forms of sexual dimorphism might lead to sex-specific phenotypic manifestations of other mtDNA mutations beyond reproductive tissues . For instance , there is already evidence that mtDNA might harbor mutations that affect aging and lifespan , which is highly dimorphic between the sexes ( Camus et al . , 2012 ) . At the molecular level , COX activity was shown to be differentially affected in male versus female flies based on their mtDNA haplotype ( Sackton et al . , 2003 ) . Furthermore , almost 90% of the transcriptome in D . melanogaster is differentially expressed between the sexes ( Ayroles et al . , 2009 ) . Amongst those differentially expressed genes , male-biased transcripts are enriched for mitochondrial energy metabolism , thus providing ample opportunities for sex-specific effects of mtDNA ( Ayroles et al . , 2009 ) . Consistent with the nuclear background exerting an effect on the phenotypic manifestation of mtDNA mutants , we found that the COIIG177S -mediated sterility was suppressed in many of the D . melanogaster nuclear backgrounds tested . Thus , our study not only shows the existence and biological basis of a male-harming mutation , but also demonstrates importance of the intimate relationship between nuclear and mtDNA genomes for the phenotypic manifestation of mtDNA mutations . Our finding that the phenotypic consequences of a specific male-harming mtDNA mutation are dependent on the nuclear genome also re-emphasizes the exciting possibility of uncovering the genetic basis of nuclear-mitochondrial incompatibilities and the molecular basis of buffering by nuclear genomes . Such buffering provides an important insight into the genetic basis of human mtDNA disease in which male-harming mtDNA mutations might initially increase in frequency in a 'suppressor' background while avoiding any negative fitness consequences . The phenotypic consequences of male-harming mtDNA mutations would only be exposed in a naïve nuclear genetic background upon introduction into new populations or even ( in the laboratory setting ) new species . Such nuclear-mitochondrial incompatibilities have been proposed to potentially serve as the basis of speciation ( Burton and Barreto , 2012 ) . Although only a few such nuclear-mitochondrial incompatibilities have been mapped in molecular detail ( Ellison and Burton , 2006; Lee et al . , 2008; Chou et al . , 2010; Clancy et al . , 2011; Meiklejohn et al . , 2013 ) , studies in which mtDNA swaps were made within and between species have revealed that many such incompatibilities exist ( James and Ballard , 2003; Sackton et al . , 2003; Ellison and Burton , 2006; Ballard et al . , 2007; Lee et al . , 2008; Montooth et al . , 2010 ) . Future genetic mapping experiments will help identify the nature and mechanism of nuclear suppression of COIIG177S and reveal whether they are specific or general suppressors of COIIG177S and other male-harming mtDNA mutations . Strategies such as the one we have outlined here , together with new advances in manipulating mtDNA genomes ( Xu et al . , 2008 ) , provide an exciting means to uncover the dark side of one of the most ancient symbioses on the planet .
All D . melanogaster strains were maintained on standard molasses-cornmeal medium at 25°C . We obtained the w1118 strain of D . melanogaster free of Wolbachia contamination ( kind gift of Susan Parkhurst , FHCRC ) . We obtained the D . melanogaster DGRP lines from the Bloomington Drosophila Stock Center ( Bloomington , IN ) . The other D . melanogaster lines were either obtained from the Drosophila Species Stock Center ( San Diego , CA ) or from our colleagues ( Daven Presgraves , John Pool , Chip Aquadro ) . We confirmed absence of Wolbachia using PCR in the original w1118 stock . Briefly , six flies from each of the assayed strains were homogenized in 49 μl squish buffer ( 10 mM Tris-CL pH 8 , 1 mM EDTA , 25 mM NaCl ) and 1 μl Proteinase K ( 200 μg/ml ) , following which the samples were incubated at 37°C for 30 min , heated to 95°C for 10 min and cooled to room temperature for 10 min . We then used WSP ( Wolbachia surface protein ) primers ( 5’- GCATTTGGTTAYAAAATGGACGA-3’ and 5’- GGAGTGATAGGCATATCTTCAAT-3’ ) and performed PCR using Taq DNA polymerase ( New England Biolabs , Ipswich , MA ) . We also assessed the status of Wolbachia infection after 35 generations of experimental evolution , in the re-isolated wt and COIIG177S mtDNA lines ( Figure 1—figure supplement 1 ) . All samples tested were found to be negative for Wolbachia . An original stock was established by expanding the w1118 strain from one bottle to 36 bottles . We established 12 experimental lines and 6 coevolving lines in bottles by adding 300 virgin females and 100 males from the original stock . The bottles were flipped three times a day for two days after which the flies were discarded . To set up every subsequent generation , virgin females were collected over a course of seven days and then crossed with males from the original stock in the case of the experimental lines . For coevolving lines , males from the corresponding coevolving line were used . After 35 generations , all lines were maintained by flipping bottles allowing sibling mating . All fertility and sex ratio assays were done in vials with five virgin females and three virgin males that were all 2–5 days old . All assays were done in 10 replicates per group . Flies were transferred to new vials every 2–3 days and flies were discarded after the fiveth’ ‘flip’ . In order to assess overall fertility , we counted all adult progeny to emerge from all of the vials . 100 females from the original stock and experimental line 7 were lysed in 100 μl squish buffer ( 10 mM Tris-HCl pH 8 , 1 mM EDTA , 25 mM NaCl , 200 μg/ml Proteinase K ) with RNase A at final concentration of 30ng/ml . Lysate was incubated at 55°C for 1 hr followed by 95°C for 10 min . Total DNA was then phenol-chloroform extracted from the lysates and the DNA pellet was dissolved in 25 μl H2O . Library preparation of samples and paired-end sequencing was performed by the Genomics Shared Resource core at Fred Hutchinson Cancer Research Center on an Illumina HiSeq 2500 . Sequence data are available through NCBI's SRA database ( project SRP057279 ) . Whole genome sequencing data was analyzed by aligning reads to a modified version of the D . melanogaster reference genome assembly ( BDGP Release 5 ) , where we masked a region of the unassembled chromosome ( chrU:5288528–5305749 ) that harbors an alternative version of the mitochondrial genome sequence ( in order to prevent ambiguous mapping of mitochondrial reads to the chrU region ) . Reads were filtered , quality- and adapter-trimmed , and aligned to our custom reference genome using GSNAP ( Wu and Nacu , 2010 ) . SNPs were called in the non-repetitive region of the mitochondrial genome ( chrM:1–14196 ) using GATK's HaplotypeCaller algorithm ( DePristo et al . , 2011 ) with a 'sample_ploidy' setting of 20 . A G-to-A change at position chrM:3611 , encoding COIIG177S ( Figure 3 ) , was present in 98% of Experimental Line 7 reads and 59% of reads from the original stock . No other variants had greater than 5% higher read frequency in Experimental Line 7 than in the original w1118 sample . Duplex Sequencing of fly mtDNA was performed as previously described ( Kennedy et al . , 2014 ) with several modifications . Total DNA was purified from individual fly heads using a QIAamp DNA micro kit ( Qiagen Inc . , Germantown MD ) . The DNA was sonicated using a Covaris AFA S2 ultrasonicator ( Covaris Inc . , Woburn MA ) with the following settings: Duty cycle: 10%; Intensity: 5; Cycles/burst: 100; Time: 15 s × 3 . The sheared DNA was then end-repaired and ligated using the NEBNext Ultra 2 end-repair and dA-tailing kit ( New England Biolabs , Ipswich MA ) according to the manufacturer’s instructions . Duplex Sequencing adapters , described previously ( Kennedy et al . , 2014 ) , were ligated to the DNA library using the NEBNext Ultra 2 ligation kit ( New England Biolabs , Ipswich MA ) according to the manufacturer’s instructions . 1 . 5 ng of total DNA was then PCR amplified using KAPA HiFi DNA polymerase ( Roche Inc . , Basel Switzerland ) according to the manufacturer’s recommendations . After amplification , the mtDNA was enriched by targeted capture using xGen target capture probes ( Integrated DNA Technologies Inc . , Coralville IA ) specific to the fly mitochondrial genome . The samples were then sequenced on a NextSeq500 machine ( Illumina Inc . , San Diego , CA ) to generate 150 bp paired-end reads . The data were processed as previously described ( Kennedy et al . , 2014 ) . 15 virgin females were collected from the original w1118 stock . Each female was individually mated with males from the same stock . Status of the G177S allele was assessed by Sanger sequencing of the females after they were allowed to have progeny . One line that appeared to be homoplasmic wildtype and one line that appeared to be homoplasmic mutant were kept and constituted the re-established stocks . six females or eight males from each group were gently homogenized on ice in 50 μl of sodium phosphate buffer with 0 . 05% Tween-20 . Lysates were centrifuged at 4°C at 4000 × g for 1 min . Supernatant was collected and 20 μl were used to measure COX activity using a kit ( ScienCell , Carlsbad , CA ) . COX activity was normalized to total protein concentration as determined by Pierce BCA Protein Assay Kit ( Pierce Biotechnology , Rockford , IL ) . Data shown represent averages of 4–5 replicates per group . 10 flies 3–4 days old and raised at 29°C were homogenized on ice in 200 μl PBS with 0 . 1% Tween-20 . Lysates were centrifuged at 4°C at 13 , 000 × g for 10 min . The supernatant was collected and 100 μl was incubated with 50 μM H2DCF ( Molecular Probes ) . Fluorescence intensity was measured in an Infinite M1000Pro ( Tecan , Switzerland ) microplate reader using 490 nm wavelength excitation and 520 nm wavelength emission . ROS levels were normalized to total protein concentration as determined by Pierce BCA Protein Assay Kit . Data shown represent averages of eight replicates per group . Five males 3–4 days old and raised at 29°C were homogenized on ice in 100 μl guanidine extraction buffer ( 6 M guanidine HCl , 100 mM Tris-HCl , pH 7 . 3 ) . Samples were frozen in liquid nitrogen for 5 min and then incubated at 95°C for 5 min . Lysates were centrifuged at 4°C for 10 min at 12 , 000 × g . Supernatant was collected and 5 μl was diluted in 95 μl H2O . 10 μl of the diluted lysate was used to measure ATP levels using the ATP determination kit ( Cat #: A22066 , Molecular Probes ) . ATP levels were normalized to total protein concentration as determined by Pierce BCA Protein Assay Kit . Data shown represent averages of four replicates per group . Five virgin wildtype females were either kept alone , or mated with three wildtype or COIIG177S mutant males at 29°C in vials with grape plates . Flies were flipped into new vials every day for eight days . 24 hr after the flies were removed from a vial , number of unhatched eggs and larvae on the grape plate were counted . Data shown represents averages of six replicates per group . Immunofluorescence staining of testes was performed as described previously ( Cheng et al . , 2008 ) . Briefly , testes were dissected in PBS , transferred to 4% formaldehyde in PBS and fixed for 30–60 min . The testes were then washed in PBS-T ( PBS containing 0 . 1% Triton-X ) for at least 30 min and mounted in VECTASHIELD with DAPI ( Vector Labs ) . Imaging of whole testis was performed on a Leica SP8 confocal microscope . Unfixed seminal vesicles were dissected in PBS mounted onto slides with PBS containing 1 µM Mitotracker Green and Mitotracker CMXRos ( Lifetechnologies ) . Sperm were extruded from seminal vesicles using a tungsten needle . Imaging was performed immediately upon addition of a cover slip to minimize the effects of hypoxia . Still images were taken on a Leica SP8 confocal microscope and movies were captured using a resonant scanner on a Leica SP5 confocal microscope . Males from wild strains including the DGRP collection were crossed with w1118 females homoplasmic for either wildtype or COIIG177S mtDNA . Resulting heterozygous male progeny were assayed for fertility , sperm development and motility , and COX activity . An assay for lifespan was performed as previously described ( Burman et al . , 2014 ) . Briefly , 100 flies from each group were divided into five vials with 20 flies each . Vials were flipped every other day and the number of dead flies was counted until all the flies had died . An assay for bang sensitivity was performed as previously described ( Burman et al . , 2014 ) . Briefly , vials with two flies each were mechanically stimulated by placement in a bench-top vortex for 10 s at the maximum setting . The time for each fly to right itself after vortexing was recorded . Data shown represents average from 18–20 flies per group . Assay for resistance to heat-induced paralysis was performed as previously described ( Burman et al . , 2014 ) . Briefly , flies were assayed for heat-induced paralysis by placing them into pre-warmed vials maintained at 39°C . The time for flies to become paralyzed was recorded . After exposure to 39°C for 6 min the animals were then placed in new room-temperature vials ( ~20°C ) , and the recovery time from paralysis was recorded . Data shown represents average from 16 flies per group . | Cell compartments called mitochondria are responsible for producing much of the energy that animal and plant cells need . Most of the proteins in mitochondria are produced from genes found in another compartment called the nucleus . However , some mitochondrial proteins are made from genes found in the mitochondria themselves . Unlike the genes in the nucleus , which animals and plants inherit from both their mother and father , the mitochondrial “genome” is only passed on along the female line . Therefore , males represent an evolutionary dead-end for mitochondrial genes . Evolutionary theory predicts that this should result in the evolution and spread of mutations that can be harmful to males , providing they do not reduce the ability of females to survive and reproduce . Although such ‘male-harming’ mutations have been well studied in plants , it is less clear how common they are in animals . Patel , Miriyala , Littleton et al . used fruit flies as a model system to identify and characterize male-harming mutations in the mitochondrial genome . The experiments isolated a mitochondrial genome with a single mutation in a gene that encodes an enzyme called cytochrome oxidase II . The mutation is said to be “hypomorphic” because it lowers the activity of the gene . The fertility of male flies with this mutation rapidly declined as they aged . However , the mutation did not appear to lower the fertility of female flies . In fact , apart from the lower male fertility , the mitochondrial mutation did not seem to affect any other traits in males or females . Further experiments revealed that this hypomorphic mutation specifically impairs the development of sperm . Patel , Miriyala , Littleton et al . also found that the effect of the mutation on the fertility of the males depended on the genes in the nucleus of their cells , as some nuclear genomes were able to partially or completely suppress the mutation . This supports previous findings that the effect of mitochondrial mutations in animals and plants may be complex and can be strongly influenced by the genes in their nucleus . Patel , Miriyala , Littleton et al . ’s findings suggest that sperm development is particularly susceptible to defects in mitochondria , and that hypomorphic mutations may represent a broader category of ‘male-harming’ mutations in animals . A future challenge will be to find out whether such mutations occur in humans and whether they are associated with infertility in men . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"genetics",
"and",
"genomics"
] | 2016 | A mitochondrial DNA hypomorph of cytochrome oxidase specifically impairs male fertility in Drosophila melanogaster |
The nucleoporin Nup98 is frequently rearranged to form leukemogenic Nup98-fusion proteins with various partners . However , their function remains largely elusive . Here , we show that Nup98-HoxA9 , a fusion between Nup98 and the homeobox transcription factor HoxA9 , forms nuclear aggregates that frequently associate with facultative heterochromatin . We demonstrate that stable expression of Nup98-HoxA9 in mouse embryonic stem cells selectively induces the expression of Hox cluster genes . Genome-wide binding site analysis revealed that Nup98-HoxA9 is preferentially targeted and accumulated at Hox cluster regions where the export factor Crm1 is originally prebound . In addition , leptomycin B , an inhibitor of Crm1 , disassembled nuclear Nup98-HoxA9 dots , resulting in the loss of chromatin binding of Nup98-HoxA9 and Nup98-HoxA9-mediated activation of Hox genes . Collectively , our results indicate that highly selective targeting of Nup98-fusion proteins to Hox cluster regions via prebound Crm1 induces the formation of higher order chromatin structures that causes aberrant Hox gene regulation .
The nucleoporin Nup98 is a mobile component of the nuclear pore complex ( NPC ) ( Griffis et al . , 2002; Rabut et al . , 2004; Oka et al . , 2010 ) , a sole gateway for selective nucleocytoplasmic macromolecular traffic . Nup98 is essential for such fundamental functions of NPC as selective nucleocytoplasmic transport ( Radu et al . , 1995; Powers et al . , 1997; Zolotukhin and Felber , 1999; Oka et al . , 2010 ) and maintenance of the permeability barrier ( Hulsmann et al . , 2012; Laurell et al . , 2011 ) . Besides , Nup98 is known as a multifunctional nucleoporin; it has been shown that Nup98 is involved in gene regulation ( Capelson et al . , 2010; Kalverda et al . , 2010; Liang et al . , 2013; Light et al . , 2013 ) , posttranscriptional regulation of specific sets of messenger RNAs ( mRNAs ) ( Singer et al . , 2012 ) , mitotic spindle assembly ( Cross and Powers , 2011 ) , mitotic checkpoint ( Jeganathan et al . , 2005; Salsi et al . , 2014 ) , and NPC disassembly ( Laurell et al . , 2011 ) . In leukemia , Nup98 is frequently found in the form of Nup98-fusions , which consist of N-terminal half of Nup98 containing multiple phenylalanine-glycine ( FG ) repeats and C-terminus of various partner proteins ( Gough et al . , 2011 ) . More than 30 different proteins with various physiological functions have been reported as Nup98 fusion partners ( reviewed in ( Gough et al . , 2011 ) ) . However , the molecular mechanism of Nup98-fusion mediated leukemogenesis is still largely unknown . Nup98-HoxA9 is one of the most frequent Nup98-fusion resulting from t ( 7;11 ) ( p15;p15 ) chromosomal translocation associated with acute myeloid leukemia , myelodysplastic syndrome , and chronic myeloid leukemia ( Nakamura et al . , 1996; Borrow et al . , 1996; Nishiyama et al . , 1999; Yamamoto et al . , 2000 ) . Indeed , the ectopic expression of Nup98-HoxA9 induces leukemia in mice ( Kroon et al . , 2001; Iwasaki et al . , 2005; Dash et al . , 2002 ) . It also has been shown that Nup98-HoxA9 inhibits hematopoietic cell differentiation ( Kroon et al . , 2001; Calvo et al . , 2002; Takeda et al . , 2006; Chung et al . , 2006; Yassin et al . , 2009 ) and enhances symmetric division of hematopoietic precursor cells in vitro ( Wu et al . , 2007 ) , suggesting that Nup98-HoxA9 contributes to leukemogenesis most likely by impairing cellular differentiation . With regard to its molecular function , Nup98-HoxA9 was shown to act as a transcriptional regulator ( Kasper et al . , 1999; Ghannam et al . , 2004; Bei et al . , 2005; Yassin et al . , 2009 ) . In addition , genome-wide gene expression analysis revealed that ectopic expression of Nup98-HoxA9 causes upregulation , rather than downregulation , of numerous genes ( Ghannam et al . , 2004; Takeda et al . , 2006 ) . Mechanistically , the FG repeat of Nup98 is known to associate with CREB-binding protein ( CBP ) /p300 ( Kasper et al . , 1999 ) , a histone acetyltransferase that functions as a transcriptional co-activator , and histone deacetylase ( HDAC ) 1 ( Bai et al . , 2006 ) . However , the interaction of Nup98 FG repeats with p300 by itself is not sufficient to promote self-renewal of hematopoietic stem cells ( Yung et al . , 2011 ) . Thus , the exact function of Nup98-HoxA9 still remains unclear . Here , we demonstrate that Nup98-HoxA9 is specifically recruited to the vicinity of Hox cluster genes via chromosomally bound Crm1 , a nuclear export factor that was originally identified as a protein required for maintaining the chromosomal structure in yeast ( Adachi and Yanagida , 1989 ) , to activate Hox cluster regions .
Previous reports have demonstrated that Nup98-HoxA9 fusion protein localizes as distinct small dots in the nucleus ( Kasper et al . , 1999; Bai et al . , 2006; Xu and Powers , 2010; Oka et al . , 2010 ) . To address the role of Nup98-HoxA9 , we first characterized in detail the intranuclear distribution of these dots . When expressed in HeLa cells , the enhanced green fluorescent protein ( EGFP ) -tagged Nup98-HoxA9 was easily distinguished as small nuclear dots . In contrast , Nup98 FG repeats were detected as large dots mostly associated with the nucleolus , and HoxA9 C-terminus containing the homeodomain , HoxA9-Ct , showed an even nucleopolasmic distribution ( Figure 1A ) , consistent with earlier reports ( Oka et al . , 2010; Xu and Powers , 2010 ) . We noticed that Nup98-HoxA9 dots were not randomly localized within the nucleus , but were frequently adjacent to 4' , 6-diamidino-2-phenylindole ( DAPI ) -dense heterochromatin regions ( Figure 1B , Figure 1—figure supplement 1 ) as revealed by co-staining with DAPI . Furthermore , immunocytochemical analysis using monoclonal antibodies specific to histone modifications ( Kimura et al . , 2008 ) ( Figure 1C ) revealed that those dots were frequently associated with H3K9me2 and H3K27me3 , facultative heterochromatin markers ( Trojer and Reinberg , 2007 ) , but not with the marker for constitutive heterochromatin H3K9me3 ( Peters et al . , 2003 , Rice et al . , 2003 ) . These results indicate that Nup98-HoxA9 dots are not randomly positioned in the nucleus but are associated with specific chromatin domains . 10 . 7554/eLife . 09540 . 003Figure 1 . Nup98-HoxA9 dots associate with facultative heterochromatin . ( A ) Subcellular localization of Nup98-HoxA9 and its truncated mutants . EGFP , EGFP-Nup98FG , EGFP-Nup98-HoxA9 , or EGFP-HoxA9-Ct expressing plasmids were transfected into HeLa cells for 24 hr and observed . Bar , 10 μm . ( B ) Confocal microscopy analysis of EGFP-Nup98-HoxA9 in HeLa cells . A merged image shows EGFP-Nup98-HoxA9 ( green ) and DAPI ( red ) . Bar , 10 μm . ( C ) Association of Nup98-HoxA9 dots with specific histone modifications . HeLa cells were transfected with the EGFP-Nup98-HoxA9 expressing plasmid . Twenty-four hours after transfection , cells were fixed and stained with antibodies against indicated histone modifications . DAPI staining was used to visualize the nuclei . Bar , 5 μm . DAPI , 4' , 6-diamidino-2-phenylindole; EGFP , enhanced green fluorescent protein . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 00310 . 7554/eLife . 09540 . 004Figure 1—figure supplement 1 . Serial z-sectioning of EGFP-Nup98-HoxA9 in HeLa cell . Serial z-sectioning was performed at a depth of 0 . 48 μm per section using confocal microscopy . DAPI staining was used to visualize nuclei . Bar , 5 μm . DAPI , 4' , 6-diamidino-2-phenylindole; EGFP , enhanced green fluorescent protein . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 004 We hypothesized that if Nup98-HoxA9 associates with specific chromatin modifications , its intranuclear distribution pattern may differ depending on the cell type . Therefore , we next expressed Nup98-HoxA9 in mouse embryonic stem ( ES ) cells , whose chromatin modifications are quite different from other cell types ( Meshorer and Misteli , 2006 ) . We found that although Nup98-HoxA9 nuclear dots could also be observed in ES cells , they were fewer in number and more heterogeneous in size compared with dots revealed in HeLa cells or NIH 3T3 cells ( Figure 2A ) . Immunocytochemical analysis in ES cells ( Figure 2—figure supplement 1 ) also showed the partial co-localization of Nup98-HoxA9 dots with H3K9me2 and H3K27me3 , but less with H3K9me3 . These findings indicated that the intranuclear distribution of Nup98-HoxA9 indeed changes depending on the cell type , although we cannot exclude the possibility that the variation in the transgene expression level caused these differences . Together , these results suggested that Nup98-HoxA9 might be involved in a cell type-specific gene and/or chromatin regulation . 10 . 7554/eLife . 09540 . 005Figure 2 . Nup98-HoxA9 evokes the expression of Hox cluster genes in ES cells . ( A ) Differential intranuclear localization of Nup98-HoxA9 in mouse ES , HeLa and NIH3T3 cells . Cells were transfected with the EGFP-Nup98-HoxA9 expressing plasmid for 24 hr , fixed , and stained with DAPI . Samples were analyzed using confocal microscopy . Bar , 10 μm . ( B ) Subcellular localization of Nup98-HoxA9 and its truncated mutants in ES cells . ES cell clones expressing FLAG-tagged Nup98FG , Nup98-HoxA9 , or HoxA9-Ct were fixed and stained with an anti-FLAG ( polyclonal ) antibody . Nuclei were stained with DAPI . Merged images show FLAG staining ( green ) and DAPI ( red ) . Bar , 10 μm . ( C ) Cell morphology of stable ES cell lines expressing FLAG-tagged Nup98-HoxA9 , Nup98FG , or HoxA9-Ct . Bar , 100 μm . ( D ) Differentiation assay of stable ES cell clones . ES cells stably expressing FLAG ( control ) , FLAG-tagged Nup98FG , Nup98-HoxA9 , HoxA9-Ct , HoxA9 , Nup153-HoxA9 , or Nup214-HoxA9 were plated at a density of 103 cells per well in 12-well plates either in the presence or absence of LIF . After 5 d , the plates were fixed and stained with alkaline phosphatase , a marker for undifferentiated stem cells . ( E ) Gene expression profiling of ES cell lines stably expressing FLAG-tagged Nup98-HoxA9 ( #1 and #9; two independent clones ) and HoxA9-Ct is compared with that of parental ES cells . A greater than 10-fold upregulation of gene expression was commonly observed in both Nup98-HoxA9 clone #1 and Nup98-HoxA9 clone #9 cells . ( F ) Hox gene expression profiling . The log2 fold ratios of the normalized signal value of Hox cluster genes from Nup98-HoxA9 or HoxA9-Ct expressing ES cells relative to signals from parental control ES cells are indicated . ( G ) Upregulation of Hox cluster genes in FLAG-Nup98-HoxA9 expressing ES cell lines was confirmed using semi-quantitative polymerase chain reaction . DAPI , 4' , 6-diamidino-2-phenylindole; EGFP , enhanced green fluorescent protein; ES , embryonic stem; LIF , leukemia inhibitory factorDOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 00510 . 7554/eLife . 09540 . 006Figure 2—figure supplement 1 . Co-staining of Nup98-HoxA9 with various histone marks . ES cell lines expressing FLAG ( control ) , FLAG-Nup98FG , or FLAG-Nup98-HoxA9 either cultured in the absence or presence of TSA ( 50 nM , 24 hr ) were fixed and stained with antibodies against FLAG ( polyclonal ) and indicated histone modifications . DAPI staining was used to visualize nuclei . Merged picture of histone modification ( red ) and FLAG staining ( green ) are shown . Bar , 5 μm . DAPI , 4' , 6-diamidino-2-phenylindole; ES , embryonic stem; TSA , trichostatin A . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 00610 . 7554/eLife . 09540 . 007Figure 2—figure supplement 2 . Protein expression levels of Nup98-HoxA9 and Nup98FG . Immunoblotting was performed with antibodies against FLAG ( M2 ) , Nup98 , and GAPDH using lysates from ES cell lines ( equivalent to 105 cells ) expressing FLAG ( control ) , FLAG-Nup98FG , or FLAG-Nup98-HoxA9 . Can Get Signal Immunoreaction Enhancer Solution ( Toyobo , Osaka , Japan ) was used to enhance the signal . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 00710 . 7554/eLife . 09540 . 008Figure 2—figure supplement 3 . Differentiation assay of stable embryonic stem cell clones . Alkaline phosphatase assay was performed as in Figure 2D . Data are mean values ± standard error of the mean of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 008 To gain more insights into the function of Nup98-HoxA9 , which most likely associates with the impairment of cell differentiation ( see Introduction ) , we tried to obtain stable cell lines expressing the FLAG-Nup98-HoxA9 fusion in ES cells using transposon vectors ( Kawakami and Noda , 2004; Urasaki et al . , 2006; Oka et al . , 2013 ) . Immunocytochemical analysis using stable ES clones demonstrated that FLAG-Nup98-HoxA9 had a similar localization pattern to transiently expressed EGFP-Nup98-HoxA9 in ES cells ( Figure 2B ) . Immunoblotting using monoclonal antibody that is raised against the N-terminal ( A . A . 1–466 ) of Nup98 ( Fukuhara et al . , 2005 ) revealed that FLAG-Nup98-HoxA9 was not overexpressed compared with endogenous Nup98 ( Figure 2—figure supplement 2 ) . Unexpectedly , we found that FLAG-Nup98-HoxA9 expressing stable ES cell clones ( Nup98-HoxA9 ES ) had distinct flat colonies , which was in contrast to tighter colonies of spherical shape detected in parental ES , Nup98FG , or HoxA9-Ct expressing ES cells ( Figure 2C ) . Further experiments revealed that Nup98-HoxA9 ES cells expressed stem cell marker genes , and had proliferation rates comparable to those of parental ES cells , indicating that they were not differentiated ( data not shown ) . In addition , Nup98-HoxA9 ES cells show more resistance to spontaneous cell differentiation in the absence of leukemia inhibitory factor ( LIF ) , compared with parental ES , Nup98FG , or HoxA9-Ct expressing ES clones ( Figure 2D and Figure 2—figure supplement 3 ) . Next , to elucidate whether Nup98-HoxA9 affected global gene expression , we performed DNA microarray analysis using the following four clones: two independent Nup98-HoxA9 ES clones ( clone #1 and clone #9 ) , parental ES cells , and the HoxA9-Ct ES clone . We found that a small fraction of genes ( less than 1% ) were strongly induced ( more than 10-fold ) in Nup98-HoxA9 ES cells compared with their expression levels in parental or HoxA9-Ct ES cells ( Figure 2E ) . Interestingly , we noticed that Hox cluster genes were highly activated in Nup98-HoxA9 ES cells ( Figure 2F ) . In particular , more than a half of the top 32 upregulated genes , which were common in both of Nup98-HoxA9 clone #1 and Nup98-HoxA9 clone #9 , were Hox cluster genes belonging to Hox-A , -B , and -C clusters ( Table 1 ) , which were validated by reverse-transcription polymerase chain reaction ( RT-PCR ) ( Figure 2G ) . These results suggest that there is a good correlation between the intranuclear distribution of Nup98-HoxA9 and its function in transcription , since Hox gene cluster regions are known to be organized as facultative heterochromatin ( Trojer and Reinberg , 2007 ) . 10 . 7554/eLife . 09540 . 009Table 1 . List of the top 32 genes that were upregulated both in Nup98-HoxA9 clone #1 and Nup98-HoxA9 clone#9 . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 009Fold change ( log2 ) Description98H#1 /EB398H#9 /EB3HoxA9-Ct#22/EB3Mus musculus homeobox A3 ( Hoxa3 ) , mRNA [NM_010452]9 . 159 . 370 . 05Mus musculus homeobox A10 ( Hoxa10 ) , transcript variant 1 , mRNA [NM_008263]8 . 177 . 610 . 05Mus musculus 2 days pregnant adult female oviduct cDNA , RIKEN full-length enriched library , clone:E230011F24 [AK053996]7 . 607 . 710 . 08Mus musculus homeobox A7 ( Hoxa7 ) , mRNA [NM_010455]7 . 697 . 440 . 32Mus musculus homeobox A11 ( Hoxa11 ) , mRNA [NM_010450]7 . 876 . 950 . 06Mus musculus homeobox C6 ( Hoxc6 ) , mRNA [NM_010465]8 . 465 . 990 . 05Mus musculus homeobox A5 ( Hoxa5 ) , mRNA [NM_010453]7 . 087 . 110 . 06Mus musculus blastocyst blastocyst cDNA , RIKEN full-length enriched library , clone:I1C0031F10 product [AK145700]6 . 167 . 101 . 26Mus musculus homeobox A9 ( Hoxa9 ) , mRNA [NM_010456]6 . 606 . 640 . 09predicted gene 3395 [Source:MGI Symbol;Acc:MGI:3781573] [ENSMUST00000172100]5 . 956 . 771 . 45Mus musculus blastocyst blastocyst cDNA , RIKEN full-length enriched library , clone:I1C0015F22 product [AK145555]5 . 876 . 760 . 51Mus musculus HOXA11 antisense RNA ( non-protein coding ) ( Hoxa11as ) , non-coding RNA [NR_015348]6 . 595 . 78-0 . 34Mus musculus blastocyst blastocyst cDNA , RIKEN full-length enriched library , clone:I1C0027E24 product [AK167004]5 . 546 . 460 . 75Mus musculus homeobox B8 ( Hoxb8 ) , mRNA [NM_010461]6 . 565 . 380 . 06Mus musculus blastocyst blastocyst cDNA , RIKEN full-length enriched library , clone:I1C0015H22 product [AK166824]5 . 536 . 361 . 07Mus musculus RCB-0559 K-1 . F1 cDNA , RIKEN full-length enriched library , clone:G430049J08 product [AK144159]5 . 196 . 190 . 37Mus musculus homeobox A4 ( Hoxa4 ) , mRNA [NM_008265]5 . 245 . 490 . 29Mus musculus blastocyst blastocyst cDNA , RIKEN full-length enriched library , clone:I1C0037K09 product [AK145750]4 . 935 . 720 . 83Mus musculus homeobox C4 ( Hoxc4 ) , mRNA [NM_013553]6 . 194 . 291 . 28Mus musculus homeobox B4 ( Hoxb4 ) , mRNA [NM_010459]5 . 274 . 61-0 . 33Mus musculus homeobox A6 ( Hoxa6 ) , mRNA [NM_010454]4 . 954 . 82-1 . 14Mus musculus homeobox B7 ( Hoxb7 ) , mRNA [NM_010460]5 . 404 . 16-0 . 75Mus musculus homeobox B2 ( Hoxb2 ) , mRNA [NM_134032]4 . 774 . 300 . 44Mus musculus cystatin 13 ( Cst13 ) , mRNA [NM_027024]3 . 725 . 312 . 99Mus musculus brain and acute leukemia , cytoplasmic ( Baalc ) , mRNA [NM_080640]3 . 575 . 441 . 17Mus musculus homeobox A2 ( Hoxa2 ) , mRNA [NM_010451]4 . 474 . 370 . 16Mus musculus CAP , adenylate cyclase-associated protein , 2 ( yeast ) ( Cap2 ) , mRNA [NM_026056]4 . 484 . 301 . 16RIKEN cDNA 5730446D14 gene [Source:MGI Symbol;Acc:MGI:1913890] [ENSMUST00000155922]4 . 304 . 440 . 06Mus musculus olfactory receptor 161 ( Olfr161 ) , mRNA [NM_146860]3 . 505 . 060 . 56Mus musculus carbonyl reductase 2 ( Cbr2 ) , mRNA [NM_007621]4 . 253 . 970 . 06Mus musculus homeobox A1 ( Hoxa1 ) , mRNA [NM_010449]3 . 864 . 13-0 . 11Mus musculus Scm-like with four mbt domains 2 ( Sfmbt2 ) , transcript variant 3 , mRNA [NM_001198809]3 . 583 . 652 . 69cDNA , complementary DNA; mRNA , messenger RNA . To find out how Hox genes could be selectively induced by the Nup98-HoxA9 fusion in ES cells , we first focused on Nup98FG repeats , since all Nup98-fusion proteins , including Nup98-HoxA9 , contain dense FG repeats of Nup98 at their N-termini . On the other hand , it is known that other nucleoporins containing such dense FG repeat domains are also implicated in oncogenic transformation ( Kasper et al . , 1999 , Xu and Powers , 2009 ) . Therefore , we created chimeric Nup-fusions by replacing Nup98 FG repeats with those from hNup153FG ( A . A . 1118–1475 ) and hNup214FG ( A . A . 1605–2090 ) , which have been demonstrated to be functionally exchangeable in experiments that addressed the transforming ability in NIH3T3 cells ( Kasper et al . , 1999 ) . When these chimeric proteins ( Nup153-HoxA9 and Nup214-HoxA9 ) were expressed in HeLa cells , they were not distributed as distinct fine nuclear dots observed in the case of the Nup98-HoxA9 fusion ( Figure 3—figure supplement 1 ) , in agreement with a previous report ( Xu and Powers , 2013 ) . 10 . 7554/eLife . 09540 . 010Figure 3 . Functional characterization of various NupFG-HoxA9 fusions . qPCR analysis of Hox-A cluster gene expression in various ES cell lines expressing the NupFG-HoxA9 fusion , Nup98FG , or HoxA9 . GAPDH was used as a reference gene . ES , embryonic stem ; qPCR , quantitative polymerase chain reaction . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 01010 . 7554/eLife . 09540 . 011Figure 3—figure supplement 1 . Subcellular localization of various NupFG-HoxA9 fusions . Subcellular localization of various FLAG-NupFG-HoxA9 fusions or HoxA9 in transiently transfected HeLa cells . Bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 011 We next isolated stable ES cell clones expressing these fusions . As shown in Figure 4—figure supplement 2 , we found that , in contrast to the case with Nup98-HoxA9 , other Nup-fusions were distributed more evenly in the nucleus ( although note that a few dots within the nucleus were seen with the Nup214-HoxA9 fusion ) . Next , the effect of these Nup-fusions on the expression of Hox genes was monitored in established stable ES cell lines . As shown in Figure 3 , the expression of Nup153-HoxA9 or Nup214-HoxA9 did not affect the expression of Hox cluster genes . Of note , we constantly observed substantial numbers of alkaline phosphatase-positive colonies of Nup153-HoxA9-expressing ES cells in the absence of LIF ( Figure 2D and Figure 2—figure supplement 3 ) , which may be related to the function of Nup153 in maintaining stem cell pluripotency in ES cells , as demonstrated in a recent study ( Jacinto et al . , 2015 ) . We also established stable clones expressing full-length HoxA9 . Gene expression analysis revealed that HoxA9 expression did not induce the activation of Hox cluster genes ( Figure 3 ) . These results underscore the physiological significance of Nup98 FG repeats in the dysregulation of Hox cluster gene expression . 10 . 7554/eLife . 09540 . 012Figure 4 . Association between Nup98-HoxA9 and Crm1 is critical for the Hox Gene activation mediated by Nup98-HoxA9 . ( A ) Top panel: Nup98-HoxA9 interacts and sequesters Crm1 onto Nup98-HoxA9 dots . HeLa cells were transfected with the EGFP-Nup98-HoxA9 expressing plasmid . After 24 hr , cells were fixed and stained with an anti-Crm1 antibody . Arrows indicate the cells transfected . Bottom panel: Nup98-HoxA9 ES cells were fixed and co-stained with anti-FLAG ( M2 ) and anti-Crm1 antibodies . Merged image of FLAG ( green ) and Crm1 ( red ) is shown . Bar , 5 μm . ( B ) The effect of LMB treatment on the cellular localization of Nup98-HoxA9 . Nup98-HoxA9 ES cells were cultured either in the presence or absence of 5 nM LMB for 2 hr , fixed and stained with antibodies against FLAG ( M2 ) and Crm1 . Merged images of FLAG ( green ) and Crm1 ( red ) are shown . Nuclei were stained with DAPI . Bar , 10 μm . ( C ) Effect of LMB treatment on the regulation of Hox cluster genes . Nup98-HoxA9 ES cells were cultured in the presence or absence of 5 nM LMB for 3 or 6 hr and the expression of indicated genes was analyzed by qPCR . GAPDH was used as a reference gene . EGFP , enhanced green fluorescent protein; LMB , leptomycin B; qPCR , quantitative polymerase chain reaction . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 01210 . 7554/eLife . 09540 . 013Figure 4—figure supplement 1 . Effect of LMB treatment on the FLAG-Nup98-HoxA9 protein level . Cell extracts of control ( FLAG#1 ) , Nup98-HoxA9 ( clone#1 ) , and Nup98-HoxA9 ( clone#9 ) ES cells ( equivalent of 105 cells ) either incubated with LMB ( 5 nM , 2 hr ) or left untreated were , loaded onto corresponding lanes . Immunoblotting was performed using an anti-FLAG or an anti-GAPDH antibody . ES , embryonic stem; LMB , leptomycin B . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 01310 . 7554/eLife . 09540 . 014Figure 4—figure supplement 2 . Effect of LMB treatment on the cellular localization of various NupFG-HoxA9 fusions . ES cell clones expressing FLAG-tagged Nup153-HoxA9 , Nu214-HoxA9 , or HoxA9 were cultured in the presence or absence of 5 nM LMB for 2 hr , fixed and stained with an anti-FLAG ( M2 ) antibody . Nuclei were stained with DAPI . Bar , 10 μm . DAPI , 4' , 6-diamidino-2-phenylindole; ES , embryonic stem; LMB , leptomycin B . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 014 What makes Nup98FG repeats differ from Nup153FG or Nup214FG ? We speculated that cohesive properties of Nup98FG ( Patel et al . , 2007; Xu and Powers , 2013; Yamada et al . , 2010; Schmidt and Gorlich , 2015 ) required to form nuclear aggregates ( Griffis et al . , 2002; Oka et al . , 2010; Patel et al . , 2007 ) containing Crm1 ( Takeda et al . , 2010 , Oka et al . , 2010 ) may be important for the pathogenic function . Indeed , the ectopic expression of Nup98-HoxA9 in HeLa cells caused sequestration of nucleoplasmic and nuclear envelope-localized Crm1 on Nup98-HoxA9 dots within the nucleus ( Figure 4A , top ) . When we stained Nup98-HoxA9 ES cells , we also noticed partial recruitment of Crm1 to FLAG-Nup98-HoxA9 dots ( Figure 4A , bottom ) . We have previously demonstrated that treatment with leptomycin B ( LMB ) , a specific inhibitor of the cargo binding of Crm1 ( Kudo et al . , 1998 ) , causes a dynamic redistribution of Nup98 nuclear dots ( Oka et al . , 2010 ) . This finding indicated that Nup98-Crm1 association is dependent on the cargo binding status of Crm1; therefore , we examined whether such properties of Nup98FG were still preserved in Nup98-HoxA9 . When Nup98-HoxA9 ES cells were treated with LMB for 2 hr , the dots disappeared ( Figure 4B ) . The disappearance of dots was concomitant with a modest decrease in the protein level ( Figure 4—figure supplement 1 ) . In contrast , LMB treatment did not affect the localization of full-length HoxA9 , Nup153-HoxA9 , and the majority of Nup214-HoxA9 , except for a few nuclear dots of Nup214-HoxA9 that disappeared upon LMB treatment ( Figure 4—figure supplement 2 ) . These results indicate that Crm1 plays a critical role in the maintenance of Nup98-HoxA9 dots in ES cells . Next , to examine whether the association between Nup98-HoxA9 and Crm1 is required for the Nup98-HoxA9-mediated Hox gene activation , we treated Nup98-HoxA9 ES cells with LMB and monitored its effects on gene expression . As shown in Figure 4C , LMB treatment caused a significant downregulation of Hox cluster genes , while the expression of other genes , such as Oct4 and Nanog , was not much affected . Therefore , these results indicate that the interaction of Nup98-HoxA9 with Crm1 is necessary for the selective Hox gene activation . To further investigate how Nup98-HoxA9 is involved in the Hox cluster gene regulation , we performed chromatin immunoprecipitation sequencing ( ChIP-seq ) analysis using Nup98-HoxA9 ES cells . Interestingly , we found a strong accumulation of FLAG-Nup98-HoxA9 on chromosomal regions of four Hox clusters , Hox-A , B , C , and D . Especially , three of these , Hox-A , Hox-B and Hox-C cluster regions showed the highest peaks in the whole genome ( Figure 5A ) . In particular , the Hox-A cluster region showed a single dominant peak in chromosome 6 ( Figure 5C ) . Examination of individual Hox cluster regions further revealed a selective accumulation of FLAG-Nup98-HoxA9 in the four Hox cluster regions ( Figure 5D ) . It is noteworthy that , as most of the genes in the Hox-D cluster were not upregulated by Nup98-HoxA9 , the significance of its binding to this region remains unknown . ChIP-qPCR ( quantitative polymerase chain reaction ) analysis ( Figure 5E ) confirmed that a significantly increased amount of FLAG-Nup98-HoxA9 was bound to the Hox-A cluster , but not to the restricted regions , as represented by the intergenic region ( between HoxA2 and HoxA3 ) , promoter ( HoxA4 ) , or 5'UTR ( HoxA9 ) regions of Hox-A cluster genes . 10 . 7554/eLife . 09540 . 015Figure 5 . Nup98-HoxA9 is highly selectively targeted to Hox cluster regions through the chromatin-associated Crm1 . ( A–D ) ChIP-seq analysis of Nup98-HoxA9 or Crm1 ( A , B: whole genome; C: chromosome 6; D: Hox cluster regions ) . The parental ES or Nup98-HoxA9 ES cells were cultured either in the absence or presence of LMB ( 5 nM , 2 hr ) or TSA ( 50 nM , 24 hr ) and used for ChIP-seq analysis . Also in ( D ) are regions that show Crm1 signals only when Nup98-HoxA9 is expressed ( eg . green arrows point to regions next to Hox-A and Hox-C ) . ( E ) ChIP-qPCR analysis of FLAG-Nup98-HoxA9 and Crm1 . Data are mean values ± standard error of the mean of three independent experiments . ChIP-seq , ChIP sequencing; ES , embryonic stem; LMB , leptomycin B; qPCR , quantitative polymerase chain reaction; TSA , trichostatin A . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 01510 . 7554/eLife . 09540 . 016Figure 5—figure supplement 1 . Binding of Nup98FG or Nup98HoxA9 to Hox-A cluster region . ChIP-qPCR analysis of control ( FLAG ) , Nup98FG ( 98FG ) , and Nup98-HoxA9 ( 98H ) expressing ES cells . ChIP-qPCR analysis was performed using an anti-FLAG antibody . Data are mean values ± standard error of the mean of three independent experiments . ChIP-seq , ChIP-sequencing; qPCR , quantitative polymerase chain reaction . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 016 Next , we examined whether Crm1 could bind Hox cluster regions using ChIP-seq analysis . Unexpectedly , we found that in parental ES cells , the endogenous Crm1 bound to all Hox regions including Hox-A , B , C , and D ( Figure 5B–D ) , which were almost the same sites as those occupied by Nup98-HoxA9 . Moreover , the binding of Crm1 to Hox cluster regions was significantly elevated in Nup98-HoxA9 ES cells ( Figure 5B–E ) . Furthermore , LMB treatment induced a drastic decrease of Nup98-HoxA9 binding to Hox clusters ( Figure 5B–E ) . These results indicate that Nup98-HoxA9 forms higher order nuclear structures containing Crm1 on Hox cluster regions to induce Hox gene activation . Next , we characterized the chromatin binding sites of FLAG-Nup98-HoxA9 and their overlap with Crm1-binding sites on a genome-wide scale . Aggregation plot revealed that the peak of FLAG-Nup98-HoxA9 binding site was severely diminished but not shifted by LMB treatment ( Figure 6A , left top ) . Furthermore , FLAG-Nup98-HoxA9 was preferentially targeted to the chromatin sites that originally bound Crm1 ( the Crm1 binding site in control ES cells ) ( Figure 6A , right top ) . Aggregation plots of the Crm1 binding signal revealed that the peak of Crm1 binding was not shifted , but rather significantly enhanced by the expression of FLAG-Nup98-HoxA9 ( Figure 6A , left bottom ) . In addition , the Crm1 signal also accumulated at the FLAG-Nup98-HoxA9 binding site ( Figure 6A , right bottom ) . Statistical analyses confirmed these results ( Figure 6B ) . We also performed an analysis of the target genes that bound FLAG-Nup98-HoxA9 or Crm1 ( Figure 6C , Figure 6—figure supplement 1 ) . As expected , the number of FLAG-Nup98-HoxA9 target genes was severely decreased by LMB treatment ( Figure 6—figure supplement 1 , top left ) . On the other hand , the number of Crm1 target genes were significantly increased by the expression of FLAG-Nup98-HoxA9; however , the majority of these genes were not overlapping ( Figure 6C , Figure 6—figure supplement 1 , top right ) , examples of which were observed in the vicinity of Hox-cluster regions ( Figure 5D , see green arrows near Hox-A ( Skap2 ) or Hox-C cluster ) . In these regions , it is likely that FLAG-Nup98-HoxA9 first binds to chromatin loci where Crm1 is not present , and then , Crm1 is recruited to FLAG-Nup98-HoxA9-bound sites , which results in the increase of Crm1-target genes on a whole genome level ( Figure 6C , Figure 6—figure supplement 1 , compare bottom two panels ) . Therefore , these results demonstrate that both chromatin-bound Crm1 and FLAG-Nup98-HoxA9 can recruit each other . 10 . 7554/eLife . 09540 . 017Figure 6 . Genome-wide analysis of FLAG-Nup98-HoxA9 and Crm1 binding sites . ( A ) Aggregation plots of FLAG-Nup98-HoxA9 and Crm1 binding sites . FLAG-Nup98-HoxA9 binding signals in untreated , LMB-treated , and TSA-treated cells are mapped against either FLAG-Nup98-HoxA9 ( top , left ) or Crm1 binding sites ( top , right ) of untreated cells . Crm1 binding signals in control , Nup98-HoxA9 ES cells , and TSA-treated Nup98-HoxA9 ES cells were mapped against either Crm1 ( bottom , left ) or FLAG-Nup98-HoxA9 binding sites ( bottom , right ) of untreated cells . ( B ) Box plot showing FLAG-Nup98-HoxA9 signals ( RPKM , reads per kilobase per million mapped reads ) in untreated , TSA-treated , and LMB-treated cells either around FLAG-Nup98-HoxA9 binding sites ( upstream 2 kb to downstream 2 kb ) in untreated Nup98-HoxA9 ES cells or around Crm1 binding sites ( upstream 2 kb to downstream 2 kb ) in control ES cells , or Crm1 signals ( RPKM ) in control ES , Nup98-HoxA9 ES , and TSA-treated Nup98-HoxA9 ES cells around Crm1 binding sites ( upstream 2 kb to downstream 2kb ) in control ES cells , or around FLAG-Nup98-HoxA9 binding sites in untreated Nup98-HoxA9 ES cells . ( C ) Venn diagrams showing the overlap between genes associated with FLAG-Nup98-HoxA9 , Crm1 in control ES , and Crm1 in Nup98-HoxA9 ES . Genes that contained FLAG-Nup98-HoxA9 or Crm1 binding site ( s ) within 5000 bp upstream and 1000 bp downstream of the TSS were analyzed . The numbers indicate the number of genes in each category . ES , embryonic stem; LMB , leptomycin B; TSA , trichostatin A; TSS , transcription start site . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 01710 . 7554/eLife . 09540 . 018Figure 6—figure supplement 1 . Venn diagrams showing the overlap between genes associated with FLAG-Nup98-HoxA9 and/or Crm1 in various cell lines or at different culture conditions . Genes that contained FLAG-Nup98-HoxA9 or Crm1 binding site ( s ) within 5000 bp upstream and 1000 bp downstream of the TSS were analyzed . ( a ) Overlap of FLAG-Nup98-HoxA9-bound genes in untreated , LMB- , or TSA-treated cells . The number of genes for each category is indicated . ( b ) Overlap of Crm1-bound genes in control , Nup98-HoxA9 , and TSA-treated Nup98-HoxA9 ES cells . ( C and D ) Overlap of FLAG-Nup98-HoxA9 and Crm1-bound genes in control ( c ) or Nup98-HoxA9 ( d ) ES cells . ES , embryonic stem; LMB , leptomycin B; TSA , trichostatin A; TSS , transcription start site . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 018 Our results so far suggest that Nup98-HoxA9 is targeted to Crm1 that associates with specific chromatin modifications . Therefore , we examined the effects of various epigenetic inhibitors on nuclear Nup98-HoxA9 dots . While most of these agents only showed a minor effect on the localization pattern of Nup98-HoxA9 ( Figure 7A ) , trichostatin A ( TSA ) treatment caused a drastic redistribution of FLAG-Nup98-HoxA9 , a significant increase in the number and a significant decrease in the size of dots , which mimicked that observed in HeLa cells ( Figure 7B ) , concomitant with the redistribution of endogenous Crm1 ( Figure 7—figure supplement 1 ) . We also confirmed that TSA treatment caused an upregulation of the FLAG-Nup98-HoxA9 protein and mRNA ( Figure 7—figure supplement 2 ) , which occurred presumably due to the reactivation of partially silenced transgenes cloned in the Tol2 transposon vector . A similar phenomenon was also observed in the case of transgenes created by the Sleeping Beauty transposon ( Garrison et al . , 2007 ) . Next , we examined the effects of TSA treatment on the genome-wide chromatin binding of Crm1 or FLAG-Nup98-HoxA9 . Unexpectedly , despite the increase in the FLAG-Nup98-HoxA9 protein level , TSA treatment caused a drastic decrease in the binding of Nup98-HoxA9 and Crm1 to chromatin ( Figure 6A , B ) , which was also obvious in Hox cluster regions ( Figure 5B–D ) . Furthermore , our ChIP-qPCR analysis showed that the binding of endogenous Crm1 to Hox-A cluster in control ES cells was also sensitive to TSA treatment ( Figure 7—figure supplement 3 ) . 10 . 7554/eLife . 09540 . 019Figure 7 . Epigenetic status and chromatin binding of Nup98-HoxA9/Crm1 . ( A ) Effect of various epigenetic inhibitors on the localization of Nup98-HoxA9 . Epigenetic effectors; 5-aza-dC ( an inhibitor of DNA methyltransferase ) , TSA ( an inhibitor of histone deacetylases ) , BIX-1294 ( an inhibitor of the G9a histone methyltransferase ) , and C646 ( an inhibitor of the histone acetyltransferase p300 ) . Immunostaining of FLAG-Nup98-HoxA9 ES cells was performed upon treatment with either 5-aza-dC ( 2 μM ) , TSA ( 100 nM ) , BIX-01294 ( 10 μM ) or C646 ( 25 μM ) for 24 hr . Nuclei were stained with DAPI . Bar , 10 μm . ( B ) Time-course of the Nup98-HoxA9 localization upon TSA ( 50 nM ) treatment . Nuclei were stained with DAPI . Bar , 10 μm . ( C ) The effect of Nup98-HoxA9 expression on the histone modification of the Hox-A cluster region . ChIP-qPCR analysis was performed using an anti-H3K4me3 or an anti-H3K27me3 antibody . Data are mean values ± standard error of the mean of three independent experiments . Statistical significance was evaluated with the Student t test . ∗p < 0 . 05 . ( D ) RNAPII recruitment at Hox cluster region . ChIP-qPCR analysis was performed using anti-PolII ( S2P ) or anti-PolII ( S5P ) antibody . Data are mean values ± standard error of the mean of three independent experiments . 5-aza-dC , 5-Aza-2'-deoxycytidine; ChIP , chromatin immunoprecipitation; DAPI , 4' , 6-diamidino-2-phenylindole; qPCR , quantitative polymerase chain reaction; TSA , trichostatin A . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 01910 . 7554/eLife . 09540 . 020Figure 7—figure supplement 1 . Effect of TSA treatment on the subcellular localization of Crm1 . Control ES or Nup98-HoxA9 expressing ES cells were cultured in the presence or absence of 50 nM TSA for 24 hr . Then , the cells were fixed and stained with antibodies against FLAG ( M2 ) and Crm1 . Merged images of FLAG ( green ) and Crm1 ( red ) are shown . Nuclei were stained with DAPI . Bar , 10 μm . DAPI , 4' , 6-diamidino-2-phenylindole; ES , embryonic stem; TSA , trichostatin A . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 02010 . 7554/eLife . 09540 . 021Figure 7—figure supplement 2 . Effect of TSA treatment on the expression of FLAG-Nup98-HoxA9 . Immunoblotting ( A ) or qPCR ( B ) indicated upregulation of FLAG-Nup98-HoxA9 by TSA treatment at a low dose . ( A ) Cell extracts of untreated or TSA-treated ( 50 nM , 24 hr ) Nup98-HoxA9 ES cells ( equivalent of 105 cells ) were loaded into corresponding lanes . Immunoblotting was performed using an anti-FLAG or an anti-GAPDH antibody . ( B ) qPCR analysis of Nup98-HoxA9 mRNA expression levels . The values were normalized to GAPDH mRNA expression levels . mRNA , messenger RNA; qPCR , quantitative polymerase chain reaction; TSA , trichostatin A . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 02110 . 7554/eLife . 09540 . 022Figure 7—figure supplement 3 . Effect of TSA treatment on Crm1 binding to the Hox-A cluster region in control ES cells . Control ES cells ( FLAG #1 ) were cultured in the presence or absence of 50 nM TSA for 24 hr . ChIP-qPCR analysis was performed using an anti-Crm1 antibody . Data are mean values ± standard error of the mean of three independent experiments . ChIP , chromatin immunoprecipitation; ES , embryonic stem; qPCR , quantitative polymerase chain reaction; TSA , trichostatin A . DOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 02210 . 7554/eLife . 09540 . 023Figure 7—figure supplement 4 . Effect of TSA treatment on histone modifications of the Hox-A cluster region . Control ES cells ( FLAG #1 ) were cultured either in the absence or presence of 50 nM TSA for 24 hr . ChIP-qPCR analysis was performed using an anti-H3K4me3 or an anti-H3K27me3 antibody . ChIP , chromatin immunoprecipitation; ES , embryonic stem; qPCR , quantitative polymerase chain reaction; TSA , trichostatin ADOI: http://dx . doi . org/10 . 7554/eLife . 09540 . 023 It is known that TSA treatment causes differential effects on the epigenetic status of chromatin depending on the dosage , duration , cell-type , and target genes . Interestingly , it is known that treatment of ES cells with TSA at a concentration of 50 nM drastically affects bivalent histone modifications H3K4me3 ( upregulation ) and H3K27me3 ( downregulation ) , in addition to the acetylation ( Karantzali et al . , 2008 ) . Since bivalent histone marks are critical for the regulation of Hox cluster genes in ES cells ( Mikkelsen et al . , 2007 , Pan et al . , 2007 , Zhao et al . , 2007 ) , we speculate that TSA treatment may dramatically modulate the epigenetic status of Hox cluster regions . Indeed , as shown in Figure 7—figure supplement 4 , TSA treatment caused a moderate upregulation of H3K4me3 and a robust decrease of H3K27me3 on the promoter region of Hox-A cluster genes , which is in line with a previous report ( Karantzali et al . , 2008 ) . These results suggest that inactive histone marks , including H3K27me3 , are important for the recruitment of Crm1 and FLAG-Nup98-HoxA9 to the Hox-A cluster region . We then compared the epigenetic status of the Hox-A region in Nup98-HoxA9 and parental ES cells . We revealed that Nup98-HoxA9 ES cells exhibited significantly higher levels of H3K4me3 , an active histone modification , in all tested Hox regions ( Figure 7C ) . On the other hand , enhanced levels of H3K27me3 were significant only on the HoxA9 5'UTR exon ( Figure 7C ) . Furthermore , ChIP-qPCR analysis of phospho-CTD isoforms of RNA polymerase II ( RNAPII ) revealed that the binding of both S5P ( initiating ) and S2P ( elongating ) forms of RNAPII to the Hox-A cluster regions increased in the Nup98-HoxA9 ES cells compared with parental ES cells ( Figure 7D ) . Collectively , our results suggest that Crm1 may associate with Hox genes via inactive histone marks , such as H3K27me3 , and subsequently , Nup98-HoxA9 may form a local higher order chromatin structure on Crm1 binding sites to further recruit active histone modifier ( s ) and RNAPII to induce a selective upregulation of Hox genes .
Our findings reveal that Nup98-HoxA9 is selectively recruited to Hox cluster regions via its interaction with Crm1 to induce gene expression in ES cells . Although activation of several Hox genes by Nup98-HoxA9 has been previously reported in other types of cells ( Calvo et al . , 2002; Takeda et al . , 2006; Ghannam et al . , 2004 ) , those effects were relatively weak and not highly specific , which is in contrast to the pronounced changes observed in our present study . We believe that such variable outcomes could stem from two possible reasons . First , the chromatin structure , epigenetic status or chromatin-bound protein complexes of Hox cluster regions may be dissimilar in different cell types , causing varying susceptibility to Nup98-HoxA9-mediated chromatin modifications . Second , the association status of Crm1 with Hox cluster gene loci , which is a prerequisite for the selective deregulation of Hox cluster genes , may be differentially modulated in different cell types , causing cell type-specific variability in the Nup98-HoxA9-mediated gene expression . Notably , although both FG repeats of Nup153 and Nup214 are known to interact with Crm1 ( Roloff et al . , 2013; Nakielny et al . , 1999; Fornerod et al . , 1997 ) , we found that these fusions , Nup153-HoxA9 and Nup214-HoxA9 , failed to activate Hox cluster genes . Thus , our results suggest that the ability of Nup98 FG repeats to form nuclear dot-like structures is critical in Hox gene activation . We propose a model whereby the interaction of Nup98-HoxA9 with prebound Crm1 induces the alteration of chromatin structures specifically on Hox cluster regions and this modification , in turn , causes robust activation of gene expression . It is known that Hox clusters in ES cells harbor a distinctive long-range histone modification signature called the bivalent motif , which possesses both transcriptionally active H3K4me3 and repressive H3K27me3 marks ( Bernstein et al . , 2006; Mikkelsen et al . , 2007 ) . This modification is maintained through the activity of polycomb complexes ( Schwartz and Pirrotta , 2007 ) . Our results demonstrate that Nup98-HoxA9/Crm1 binding to Hox region is sensitive to TSA treatment , which results in a robust decrease of the H3K27me3 mark on Hox cluster genes . Moreover , the expression of Nup98-HoxA9 causes a significant increase of the active histone mark H3K4me3 on Hox cluster region genes . Previously , it was reported that the FG repeat region of Nup98 interacts with histone acetyltransferases CBP and p300 ( Kasper et al . , 1999 ) . In addition , MBD-R2/NSL and Trx histone modifying complexes have been demonstrated to interact with full-length Nup98 in fly ( Pascual-Garcia et al . , 2014 ) . Thus , we speculate that Nup98-HoxA9 first binds to facultative heterochromatin and then recruits histone and/or chromatin modifying enzymes to induce active histone modifications and a subsequent activation of Hox cluster genes . Importantly , in CD133+ hematopoietic stem cells , histone modifications of Hox-A and Hox-B loci are also bivalent ( Cui et al . , 2009 ) , suggesting that Hox cluster genes are similarly regulated both in ES and hematopoietic progenitor cells . Selective activation of Hox genes has also been reported in other Nup98-fusions , Nup98-NSD1 and Nup98-Jarid1a , in myeloid progenitor cells ( Wang et al . , 2007 , Wang et al . , 2009 ) . Mechanistically , Nup98-NSD1 binds to the promoter region of Hox-A cluster genes and modifies histones via its H3K36 methyltransferase activity ( Wang et al . , 2007 ) . Similarly , Nup98-Jarid1a binds to di- or tri-methylated H3K4 of the Hox-A cluster region through its plant homeodomain ( PHD ) finger domain and causes deregulation of Hox genes activation ( Wang et al . , 2009 ) . However , it remained unclear how the expression of these Nup98-fusions actually causes selective activation of Hox genes . Our results suggest that chromosome prebound Crm1 may recruit Nup98-NSD1 , Nup98-Jarid1a , and possibly other Nup98-fusions to specific Hox cluster genes loci through its interaction with the Nup98FG repeat , which commonly exists in all Nup98-fusions . Unexpectedly , our data showed that Nup98FG could only weakly bind to Hox-A cluster region ( Figure 5—figure supplement 1 ) . However , since Nup98FG dots are frequently localized within the nucleolus ( data not shown ) , and Nup98FG is more prone to form nuclear aggregates than Nup98-HoxA9 with only a weak nucleoplasmic diffuse staining ( see Figure 2B ) , we speculate that Nup98FG is hardly accessible to the genomic DNA , including the Hox cluster region , when expressed by itself . Notably , a recent report ( Conway et al . , 2014 ) also demonstrated that Crm1 binds to the HoxA9 and HoxA10 gene regions , both in human leukemia cell line and immortalized mouse embryonic fibroblast cell line . Conway et al . ( 2014 ) further showed that Crm1 recruits CALM-AF10 fusion through its interaction with the nuclear export signal on CALM to activate Hox gene expression . Thus , the association of Crm1 with Hox loci could be a common molecular basis for aberrant Hox gene dysregulation mediated by numerous leukemic fusions .
HeLa cells and NIH3T3 cells were cultured in the Dulbecco's modified Eagle’s medium ( DMEM; Sigma-Aldrich , St . Louis , MO ) supplemented with 10% fetal bovine serum ( FBS ) . The parental EB3 ES cells ( Niwa et al . , 2002; Ogawa et al . , 2004 ) and their derivatives were cultured on gelatin-coated dishes in DMEM supplemented with 10% FBS , 10 mM MEM non-essential amino acids ( Life Technologies , Carlsbad , CA ) , 100 mM sodium pyruvate ( Life Technologies ) , 0 . 1 mM β-mercaptoethanol ( Life Technologies ) , and LIF at 37°C in 5% CO2 atmosphere . The following mouse monoclonal antibodies against specific histone modifications were used in this study: anti-H3K4me3 , anti-H3K27me3 , anti-H3K9me2 , and anti-H3K9me3 ( Kimura et al . , 2008 ) . Rat monoclonal Pol II antibodies to site specific phosphorylation used in this study were as follows; Anti-Pol II ( S2P ) , Anti-Pol II ( S5P ) ( Odawara et al . , 2011 ) . The anti-FLAG ( M2 , #F1804 ) and anti-FLAG ( polyclonal , #F7425 ) antibodies were purchased from Sigma-Aldrich . The anti-Crm1antibody ( #NB100-79802 ) was purchased from Novus Biologicals ( Littleton , CO ) . The anti-Nup98 antibody is as described ( Fukuhara et al . , 2005 ) . To generate EGFP-Nup98FG , EGFP-Nup98-HoxA9 , and EGFP-HoxA9-Ct mammalian expression vectors , the coding sequences for N-terminus of human Nup98 ( A . A . 1–469 ) , C-terminus of HoxA9 ( A . A . 164–272 ) , Nup98-HoxA9 ( combination of N-terminus of human Nup98 and C-terminus of HoxA9 were amplified and cloned into the pEGFP-C1 vector ( Clontech , Mountain View , CA ) . Sequences were verified by sequencing analysis . The Tol2-based Nup-fusion expression vectors were generated by inserting coding sequences for Nup98-HoxA9 , Nup98FG ( A . A . 1–469 ) , HoxA9-Ct ( A . A . 164–272 ) , HoxA9 ( full length ) , Nup214-HoxA9 ( a fusion between Nup214 ( A . A . 1605–2090 ) and HoxA9-Ct ) , Nup153-HoxA9 ( a fusion between Nup153 ( A . A . 1118–1475 ) and HoxA9-Ct ) into the pT2A-CMH , a Tol2 transposon-based vector containing a multiple cloning site ( Oka et al . , 2013 ) . The coding regions of the inserted cDNA molecules were verified by sequencing analysis . For transient transfection assays , cells were plated into 35-mm dishes and transfected with plasmid DNA using lipofectamine 2000 ( Life Technologies ) . Generation of stable ES cell lines expressing a transgene was performed as described previously ( Oka et al . , 2013 ) . Briefly , ES cells were co-transfected with pCAGGS-m2TP , which is an expression vector for Tol2 transposase , and the Tol2 transposon-based pT2A-CMH expression vector containing various inserted cDNA sequences . After 2 d , cells were re-plated into the ES-LIF medium containing hygromycin B ( 200 μg/ml ) . The medium was changed every other day until the appearance of colonies . Colonies were picked and the expression product of the transgenes was confirmed by western blotting and immunofluorescence staining . Cells were grown on coverslips and fixed with 3 . 7% formaldehyde in phosphate-buffered saline ( PBS ) for 15 min at room temperature . After treatment with 0 . 5% Triton X-100 in PBS for 5 min , the cells were incubated in blocking buffer ( 3% skim milk in PBS ) for 30 min and again incubated with primary antibodies overnight at 4°C . After washing with PBS , cells were incubated with Alexa Flour 488– and/or 594– conjugated secondary antibodies ( Life Technologies ) for 30 min . The cells were then stained with DAPI and the coverslips were mounted with Vectashield ( Vector Laboratories , Burlingame , CA ) . Images were acquired using FV1000 ( Olympus ) or SP8 ( Leica ) confocal microscopes . Microarray analysis was performed using total RNA prepared from indicated ES cell clones using the TRIzol reagent ( Invitrogen Carlsbad , CA ) as follows: cyanine-3 ( Cy3 ) labeled complementary RNA ( cRNA ) was prepared from 0 . 1 μg total RNA using the Low Input Quick Amp Labeling Kit ( Agilent Technologies , Santa Clara , CA ) according to the manufacturer's instructions , followed by RNAeasy column purification ( Qiagen , Valencia , CA ) . Cy3-labeled cRNA ( 0 . 6 μg ) was fragmented and hybridized to SurePrint G3 Mouse GE 8×60K Microarray ( Agilent Technologies ) for 17 hr at 65°C in a rotating Agilent hybridization oven . After hybridization , microarrays were washed for 1 min at room temperature with GE Wash Buffer 1 ( Agilent Technologies ) , then for 1 min with GE Wash buffer 2 ( Agilent Technologies ) at 37°C , and dried immediately by brief centrifugation . Slides were scanned immediately on the Agilent DNA Microarray Scanner ( G2505C ) using one color scan setting for 8× 60k array slides ( scan area 61 × 21 . 6 mm , scan resolution 3 µm , dye channel was set to Green and Green PMT was set to 100% ) . The scanned images were analyzed with the Feature Extraction Software 10 . 10 . 1 . 1 ( Agilent Technologies ) using default parameters to obtain background subtracted and spatially detrended processed signal intensities . Total RNA was extracted from ES cells using the TRIzol reagent ( Invitrogen ) and used for cDNA synthesis with the Transcriptor First Strand cDNA Synthesis kit ( Roche Applied Science , Mannheim , Germany ) or PrimeScript 1st strand cDNA Synthesis Kit ( Takara Bio , Otsu , Japan ) . All procedures were conducted according to the manufacturer's recommendations . RT-PCR was performed using KOD plus DNA polymerase ( Toyobo , Osaka , Japan ) . qPCR analysis was performed on a 384-well plate with QuantStudio 6 Flex Real-Time PCR System ( Life Technologies ) using Power SYBR® Green PCR Master Mix ( Applied Biosystems , Foster City , CA ) . As for mRNA expression levels , the relative expression levels were normalized using GAPDH mRNA levels as control . The primer sequences are listed in Supplementary file 1 . Cells were fixed in a medium containing 0 . 5% formaldehyde at room temperature for 5 min . After washing with ice-cold PBS twice , cells were resuspended in the ChIP buffer ( 10 mM Tris-HCl [pH 8 . 0] , 200 mM KCl , 1 mM CaCl2 , 0 . 5% NP40 ) containing protease inhibitors ( aprotinin , leupeptin , pepstatin A at a concentration of 1 μg/ml each ) and briefly sonicated ( Branson 250D sonifier , Branson Ultrasonics , Danbury , CT ) . After centrifugation , the supernatants containing chromatin were digested with 3 units/ml micrococcal nuclease ( Worthington Biochemical , Lakewood , NJ ) for 40 min at 37°C , and the reaction was stopped with ethylenediaminetetraacetic acid ( EDTA; final concentration of 10 mM ) . Enzyme-treated supernatants were incubated with anti-mouse or anti-rabbit IgG magnetic beads ( Dynabeads , Life Technologies ) preincubated with anti-FLAG ( M2 ) , anti-Crm1 , or anti-histone modification antibodies ( 2 μg ) for 6 hr . ( For Pol II ChIP , we first incubated the beads with rabbit anti-Rat IgG antibody [Jackson Immuno Research , West Grove , PA] before its incubation with phospho-specific Pol II antibodies . ) The beads were washed thoroughly twice with each of the following three buffers , ChIP buffer , ChIP wash buffer ( 10 mM Tris-HCl [pH 8 . 0] , 500 mM KCl , 1 mM CaCl2 , 0 . 5% NP40 ) , and TE buffer ( 10 mM Tris-HCl [pH 8 . 0] , 1 mM EDTA ) , and eluted in the elution buffer containing 50 mM Tris-HCl ( pH 8 . 0 ) , 10 mM EDTA , and 1% sodium dodecyl sulfate overnight at 65°C . DNA was recovered using the DNA gel extraction kit ( Promega , Madison , WI ) and used for ChIP-qPCR analysis or ChIP-seq analysis . The ChIP library was prepared according to the Illumina protocol and sequenced on the Illumina HiSeq1500 system . The sequence reads were uniquely mapped to the reference mouse genome ( mm9 ) using Bowtie 2 software ( version 2 . 2 . 2 ) with default parameter ( Langmead and Salzberg , 2012 ) . PCR duplicates were removed from mapped reads using SAMtools ( version 0 . 1 . 19 ) . The mapped reads of ChIP and input DNA control data were counted in non-overlapping 200 base windows on the genome and the counts were normalized as RPM ( read per million ) . We then calculated normalized ChIP-Seq signal intensities on each window as RPM difference between ChIP and input DNA control data ( RPMChIP – RPMinput ) . We used MACS ( version 2 . 0 . 10 ) with the default parameters to detect ChIP-Seq signal enriched regions . The microarray data have been deposited in the Gene Expression Omnibus ( GEO ) database under the series entry GSE67967 . The ChIP-Seq data are accessible through GEO Series accession number GSE68783 . | The nucleus of a eukaryotic cell ( which includes plant and animal cells ) contains most of the cell’s genetic material in the form of carefully packaged strands of DNA . Genes are stretches of DNA that contain the instructions needed to produce the proteins and RNA molecules that the cell needs to survive . These molecules move across the membrane that surrounds the nucleus through pores made of proteins . One of these pore-forming proteins is called Nup98 . The gene that produces Nup98 is frequently mutated in leukemia , where part of it becomes fused to regions of other unrelated genes . The proteins made from these combined genes are known as “fusion proteins” . The Nup98-HoxA9 fusion protein has been well studied , and appears to cause leukemia by interfering with the process called ( “cell differentiation” ) by which stem cells specialize to form different types of blood cells . During cell differentiation , cells change which sets of genes they activate to become specific types of cells . A family of genes called Hox genes ( to which the gene for HoxA9 belongs ) is critical in cell differentiation and thus must be fine-tuned . It is also known that the Hox genes form clusters , and its activation is partly controlled by how tightly the DNA is packaged . Previous studies have shown that the Nup98-HoxA9 fusion protein takes on the form of small dots in the nucleus . Oka et al . have now tracked how these proteins are distributed inside of the nucleus , and examined which part of the DNA they bind to , in more detail . This revealed that the dots of Nup98-HoxA9 tend to associate with tightly packed DNA , especially on Hox cluster genes , and activate these genes . Oka et al . further found that a protein called Crm1 , which is well known as a nuclear export factor that carries molecules out of the nucleus through the pore , is already bound to the Hox cluster genes in the nucleus and recruits the Nup98-HoxA9 protein . This interaction may change how the Hox gene is packaged in the nucleus . A future challenge will be to reveal how the Nup98-HoxA9 fusion protein and Crm1 on Hox cluster genes control gene expression . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"cell",
"biology"
] | 2016 | Chromatin-prebound Crm1 recruits Nup98-HoxA9 fusion to induce aberrant expression of Hox cluster genes |
The kinesin-3 family member Unc-104/KIF1A is required for axonal transport of many presynaptic components to synapses , and mutation of this gene results in synaptic dysfunction in mice , flies and worms . Our studies at the Drosophila neuromuscular junction indicate that many synaptic defects in unc-104-null mutants are mediated independently of Unc-104’s transport function , via the Wallenda ( Wnd ) /DLK MAP kinase axonal damage signaling pathway . Wnd signaling becomes activated when Unc-104’s function is disrupted , and leads to impairment of synaptic structure and function by restraining the expression level of active zone ( AZ ) and synaptic vesicle ( SV ) components . This action concomitantly suppresses the buildup of synaptic proteins in neuronal cell bodies , hence may play an adaptive role to stresses that impair axonal transport . Wnd signaling also becomes activated when pre-synaptic proteins are over-expressed , suggesting the existence of a feedback circuit to match synaptic protein levels to the transport capacity of the axon .
Synapse development , maintenance and plasticity involve highly orchestrated trafficking events in both pre and postsynaptic cells . In contrast to postsynaptic receptors , whose trafficking and organization has been studied extensively in many different synapse types ( Choquet and Triller , 2013 ) , much less is known about the mechanisms that regulate the assembly and maintenance of the neurotransmitter release machinery in the presynaptic neuron . This machinery includes the active zone ( AZ ) , an electron-dense complex of structural proteins that scaffold both calcium channels and synaptic vesicles ( SV ) for the coordination of calcium-regulated exocytosis ( Südhof , 2012 ) . The protein components of the AZ are synthesized in cell bodies and trafficked together in association with vesicles ( known as piccolo-bassoon transport vesicles ( PTVs ) ( Ahmari et al . , 2000; Maas et al . , 2012; Shapira et al . , 2003 ) . SV precursors are also synthesized in cell bodies , and carried by kinesin motors to synapses ( Hall and Hedgecock , 1991; Okada et al . , 1995 ) . Regulation of synapse development likely involves a global coordination of the synthesis and transport of both AZ and SV components . However the mechanisms that regulate these important steps in synapse development and maintenance are poorly understood . A critical role in synapse development has been assigned to the kinesin-3 family of motor proteins ( Hall and Hedgecock , 1991; Kern et al . , 2013; Niwa et al . , 2016; Pack-Chung et al . , 2007; Yonekawa et al . , 1998 ) . Mutations in mammalian Kif1a and its unc-104 orthologues in C . elegans and Drosophila ( also known as imac , Klp53D and bris in Drosophila ) cause severe defects in synapse development . In Drosophila unc-104-null mutants , synaptic boutons fail to form , SV and AZ components fail to traffic to nascent synapses , and concomitantly , SV and AZ associated proteins accumulate in the cell body ( Pack-Chung et al . , 2007 ) . It is broadly accepted that Unc-104 protein functions as a molecular motor to physically deliver presynaptic components to their destinations in the synaptic terminal ( Goldstein et al . , 2008 ) . However , while there is biochemical evidence that KIF1A can interact with and ‘carry’ SV precursors ( Okada et al . , 1995 ) , there is little evidence that KIF1A ( or Unc-104 ) carries AZ components . The mechanistic role of Unc-104 in AZ transport and assembly remains unclear . In this study we found that synaptic defects in embryonic unc-104-null mutants , including the failure to form synaptic boutons and AZs , arise not from direct loss of Unc-104 transport function , but via an indirect mechanism , which involves activation of the Wnd/DLK axonal damage signaling pathway . The Wnd/DLK mixed lineage kinase has recently received intense interest for its roles in regulating both regenerative and degenerative responses to axonal damage in vertebrate and invertebrate neurons ( Gerdts et al . , 2016; Hao and Collins , 2017; Li and Collins , 2017; Tedeschi and Bradke , 2013 ) . We found that the Wnd/DLK signaling pathway becomes activated when Unc-104’s function is impaired , and then promotes synaptic dysfunction by restraining expression of multiple pre-synaptic AZ and SV protein components . This restraint concomitantly reduces protein buildup in cell bodies , which may play an adaptive role to stresses that disrupt intracellular transport , and contribute to pathologies that arise when transport is disrupted .
Previous studies of NMJ development in unc-104-null mutant animals have revealed essential roles for this kinesin in synaptic maturation ( Hall and Hedgecock , 1991; Kern et al . , 2013; Pack-Chung et al . , 2007; Yonekawa et al . , 1998 ) . At the Drosophila neuromuscular junction ( NMJ ) , unc-104-null mutants are severely defective in the formation of presynaptic boutons , fail to localize SVs to NMJ terminals and show strong reductions in AZ localization ( Pack-Chung et al . , 2007 and Figure 1 ) . We found that disrupting the axonal damage signaling kinase Wnd ( wnd3 single mutants ) had no significant effect on bouton formation , AZ number , or presynaptic protein localization , but double mutants with unc-104-null alleles ( unc-104P350 , unc-104170 , and unc-10452 ) , gave a very informative phenotype: in unc-104null;wnd3/3 double mutants the synaptic bouton formation ( Figure 1A and C ) , AZ number ( Figure 1A and D ) , and synaptic levels of the AZ protein Brp ( Figure 1A and E ) were restored to a wild type phenotype . In contrast , the synaptic levels of SV proteins ( VGlut , SytI and CSP ) remained negligible , as in the unc-104null single mutants ( Figure 1B and E ) . These results are consistent with previous findings that Unc-104 functions as an essential molecular motor to transport SV precursors to synaptic terminals . However our findings indicate that transport of AZ precursors and bouton formation can occur independently of Unc-104 . These defects are mediated by a second and separable mechanism , which depends upon the function of the Wnd kinase . To further study the effect of Wnd upon AZ assembly we utilized several hypomorphic loss-of-function mutations of unc-104 , whose ability to survive to the third instar larval stage has allowed for extensive characterization of synaptic defects associated with Unc-104 ( Barkus et al . , 2008; Cao et al . , 2014; Kern et al . , 2013; Zhang et al . , 2016; 2017 ) . We used the EMS-generated alleles bris/null ( Kern et al . , 2013 ) and O3 . 1/null ( Barkus et al . , 2008 ) , and knockdown in motoneurons via independent RNAi lines . All of these mutants share similar characteristics: Post Synaptic Densities ( PSDs ) form ( identified by the presence of a core receptor subunit GluRIII ) , however ~50% lack apposing presynaptic AZ components , as identified by Brp ( Figure 2A and C ) , Liprin-α ( Figure 2B and D ) and the voltage gated calcium channel Cac-GFP ( Figure 2—figure supplement 2 ) . This defect is accompanied by a reduction in the levels of Brp within individual synapses and also across entire NMJ terminals ( Figure 2E ) . Similarly to the unc-104-null mutants ( Figure 1 ) , mutations in wnd fully rescued these presynaptic AZ assembly defects ( Figure 2A–E ) . RNAi knockdown of unc-104 using either pan-motoneuron or single motoneuron driver lines led to AZ defects and concomitant knockdown of wnd in neurons led to rescue , ( Figure 2—figure supplement 1 ) , indicating a cell-autonomous role for both Unc-104 and Wnd in the synaptic defects . In axonal regeneration and synaptic overgrowth , Wnd has been shown to act in a signaling pathway consisting of a cascade of MAP kinases and transcription factors ( Collins et al . , 2006; Nakata et al . , 2005; Xiong et al . , 2010 ) . We found that inhibition of the downstream MAP Kinase JNK ( Bsk ) and Fos transcription factor , via expression of dominant-negative ( DN ) isoforms in neurons , could also rescue the presynaptic defect shown in the unc-104 single mutants ( Figure 2C and Figure 2—figure supplement 2 ) . These results suggest that a signaling pathway consisting of Wnd , JNK and Fos mediates synaptic defects observed in unc-104 mutants . In contrast with unc-104-null mutants , we observed that wnd mutations caused a striking increase in quantity of VGlut at unc-104-hypomorphic mutant NMJs ( Figure 2F ) . The dramatic rescue of structural defects as well as SV protein localization at NMJs suggested that Wnd pathway mutations may also rescue synaptic transmission defects in unc-104-hypomorph mutants . Unc-104-hypomorph mutant NMJs have severely reduced mini frequency , and modestly reduced mEJP amplitude , EJP amplitude and quantal content ( Zhang et al . , 2017 ) . We observed that mutations in wnd , jnk ( bsk ) or fos fully rescued the mEJP frequency defects ( Figure 3A and C and Figure 3—figure supplement 1 ) . mEJP amplitude , EJP amplitude and quantal content were also rescued by wnd mutations ( Figure 3 ) , while Bsk ( JNK ) and Fos inhibition rescued some aspects of these defects ( Figure 3—figure supplement 1 and Figure 3—source datas 1–3 ) . Through distribution analysis , we further confirmed the change in mEJP amplitude and validated the coverage of mEJP events for frequency analysis ( Figure 3—figure supplement 2 ) . The suppression of these phenotypes by mutations in wnd suggests that the Wnd pathway impairs multiple aspects of synaptic structure and transmission . The total levels of Unc-104 protein remained low in the unc-104hypomorph;wnd double mutants ( Figure 3—figure supplement 3 ) . Moreover , defects in larval motility and lethality of unc-104-hypomorph mutants largely remained in unc-104hypomorph;wnd double mutants , implying the persistence of major defects from the loss of Unc-104’s function ( Figure 3—figure supplement 4 ) . Hence the extensive suppression of the synaptic defects at the larval NMJ is unlikely to be the result of increased stability of the residual Unc-104 protein . Instead , as indicated by the suppression of unc-104-null defects ( Figure 1 ) , the synaptic phenotypes reflect an activity of the Wnd pathway rather than a direct consequence of impaired Unc-104 driven transport . To test for Wnd signaling activation in unc-104 mutants , we utilized a transcriptional reporter of JNK signaling , the puckered ( puc ) -lacZ enhancer trap ( Martín-Blanco et al . , 1998 ) , which has been previously shown to report Wnd signaling activity in motoneurons ( Valakh et al . , 2013; Xiong et al . , 2010 ) . The basal expression of puc-lacZ is very low in wild type animals but was significantly increased in all unc-104 LOF mutant backgrounds ( including bris/null , O3 . 1/null , and 2 independent RNAi lines ) ( Figure 4A and B ) . This increase was abolished when wnd was concomitantly knocked-down by RNAi ( Figure 4B ) , hence reflects activation of a Wnd-mediated nuclear signaling cascade . Additional evidence for Wnd signaling activation was revealed by the presynaptic nerve terminal morphology of unc-104 hypomorphic mutant NMJs , which show increased numbers of synaptic branches and boutons which are smaller in size ( [Figure 4C and D] , and Kern et al ) . These features of presynaptic nerve terminal overgrowth have previously been described for Wnd pathway activation ( Brace et al . , 2014; Collins et al . , 2006; Valakh et al . , 2013; Wu et al . , 2007 ) . We confirmed that this overgrowth phenotype in unc-104 mutants requires the function of Wnd , JNK and Fos in presynaptic motoneurons ( Figure 4C and D ) . Finally , we observed that unc-104-hypomorph mutants showed an enhanced ability to regrow axons after injury ( Figure 4E–G ) . A potentially similar role in restraining axonal growth has recently been noted for unc-104 during development in C . elegans ( Stavoe et al . , 2016 ) . Since previous studies have shown that activation of Wnd/DLK signaling enhances the ability of axons to initiate regenerative axonal growth after injury ( Hammarlund et al . , 2009; Shin et al . , 2012; Xiong et al . , 2010; Yan et al . , 2009 ) , the activation of Wnd signaling in unc-104 mutants may explain this kinesin’s paradoxical function in restraining rather than promoting axonal growth . Previous studies have suggested links between JNK signaling and the regulation of axonal transport ( Fu and Holzbaur , 2014; Verhey and Hammond , 2009 ) . It is therefore interesting that other mutations that disrupt axonal transport , including mutations which inhibit kinesin-1 , dynein and dynactin , did not significantly affect the expression of puc-lacZ ( Figure 4B and Xiong et al . , 2010 ) . Altogether , our results suggest a unique functional interdependence between Wnd pathway and the Unc-104 kinesin , and that the Wnd signaling pathway becomes activated when Unc-104’s function is lost . Previous studies have established that the protein levels of Wnd and its DLK homologues in mammals and C . elegans are strictly regulated , and positively correlate with the activity of Wnd signaling pathway ( Feoktistov and Herman , 2016; Hao et al . , 2016; Huntwork-Rodriguez et al . , 2013 ) . A highly conserved ubiquitin ligase domain protein Hiw/Rpm-1/Phr1 restricts the levels of Wnd/DLK protein and inhibits Wnd/DLK signaling activation in wild type animals ( Babetto et al . , 2013; Collins et al . , 2006; Nakata et al . , 2005; Xiong et al . , 2010 ) . Mutations in hiw lead to strongly elevated levels of Wnd protein whose localization can be detected in neurites and at synapses ( Collins et al . , 2006 ) . We therefore examined Wnd protein in unc-104 mutants , for comparison to the previous characterized changes in hiw mutants . In contrast to hiw , mutations in unc-104 did not lead to a detectable increase in global levels of endogenous Wnd protein by Western Blot ( Figure 5A ) , or in synaptic Wnd protein level ( Figure 5B ) . In addition , unc-104 mutants do not share the previously reported phenotype of hiw mutants of delayed axonal degeneration ( Xiong et al . , 2012 , and data not shown ) . These observations suggest that the mechanism of Wnd signaling activation in unc-104 mutants is unlikely to occur via Hiw . We then looked further at Wnd’s localization using available reagents . By immunocytochemistry ( IHC ) with anti-Wnd antibodies , Wnd protein remained below the detection threshold in both wild type and unc-104 mutant animals ( data not shown ) . Therefore , to further evaluate potential effects upon Wnd protein we utilized two different wnd-GFP fusion proteins . First , a MiMIC-wnd-GFP line ( Venken et al . , 2011 ) , in which a GFP tag is inserted via an exon trap within the wnd genomic locus , creates another tool for detecting Wnd as expressed from its endogenous promoter . We verified that Wnd protein from this line was expressed and tagged ( Figure 5—figure supplement 1 ) . By IHC , this reagent revealed Wnd enrichment in neuronal cell bodies in unc-104 mutants ( Figure 5C and E ) . Second , an ectopically expressed UAS-GFP-wndKD transgene ( which was kinase dead to avoid pathway activation ) has been established as a sensitive method to detect Wnd protein change ( Collins et al . , 2006; Hao et al . , 2016; Xiong et al . , 2010 ) . This highly expressed transgene also revealed an increase in cell body localized Wnd in unc-104 mutants ( Figure 5D and F ) . This correlation of Wnd localization with signaling activation is interesting to note , since Wnd and its DLK homologue are known to physically associate with vesicles that are transported in axons ( Holland et al . , 2016; Xiong et al . , 2010 ) , and this transport appears to be important for its ability to mediate axon-to-cell body retrograde signaling . Does the cell body localization of Wnd reflect a role for Unc-104 in transporting Wnd-associated vesicles ? From live imaging analysis of UAS- GFP-wndKD vesicles we observed no impairment in transport kinetics or flux in unc-104 mutants ( Figure 5—figure supplement 2 ) , and no co-localization for Wnd and Unc-104 ( data not shown ) . Therefore we found no evidence for Wnd as a direct cargo of Unc-104 . We then considered the possibility that Unc-104 regulates Wnd signaling indirectly via the transport of another cargo . Previous work has suggested that other presumed cargo of Unc-104 , including Liprin-α , Rab3 and Rab3-GEF , play important early roles in AZ assembly ( Graf et al . , 2009; Niwa et al . , 2008; Shin et al . , 2003; Südhof , 2012 ) . We therefore asked whether defects in these synaptic ‘cargos’ of Unc-104 led to activation of Wnd signaling . We observed that liprin-α mutant NMJs contain a large portion of unapposed GluRIII-labeled PSDs ( Figure 5G ) , resembling the defects in unc-104 mutants . Similar defects were reported for rab3 and rab3-gef mutants ( Bae et al . , 2016; Graf et al . , 2009 ) . However , in contrast to unc-104 , the liprin-α and rab3 synaptic defects were not suppressed by mutations in wnd ( Figure 5G–I ) . Furthermore , the increased Brp intensity per AZ and reduced AZ number due to liprin-α and rab3 mutations was not suppressed by wnd mutations ( Figure 5—figure supplement 3 ) . In fact , the Brp intensity per AZ was slightly enhanced in liprin-α;wnd double mutants . These data suggest that Liprin-α and Rab3 regulate presynaptic assembly via pathways that are either independent or downstream of Wnd , and that defects in synapse assembly and structure per se do not cause activation of Wnd signaling . If Wnd signaling activation mediates synaptic defects in unc-104 mutants , then activation of this pathway via other means should also induce unc-104-like synaptic phenotypes . Indeed , over-expression of wnd alone in motoneurons resulted in cell-autonomous presynaptic defects that are comparable to unc-104 mutants . First , many individual synapses ( identified by GluRIII puncta ) lacked Brp ( Figure 6A and C ) , and synapses that contained Brp had reduced Brp intensity , which resulted in a global 70% reduction in Brp intensity across the entire NMJ terminal ( Figure 6B ) . Second , mEJP frequency , along with other aspects of synaptic transmission ( amplitude of EJP and mEJP ) , was significantly impaired ( Figure 6D–I ) . Third , synaptic localization of SV-associated proteins , measured by VGlut intensity within NMJ terminals , was also reduced ( Figure 6B and Collins et al . , 2006 ) . Taken together with the rescue of unc-104 synaptic defects by wnd mutations ( Figures 1–3 ) , these data indicate that activation of Wnd signaling leads to strong perturbations in presynaptic structure and function . As noted above , Wnd signaling also becomes activated when an upstream negative regulator , Hiw , is mutated ( Collins et al . , 2006; Nakata et al . , 2005 ) . We found that hiw mutants displayed strikingly similar presynaptic defects to unc-104 mutants or overexpression of Wnd ( Figure 6—figure supplements 1 and 2 ) . These included 60% of synapses lacking the AZ proteins Brp , Liprin-α and Cac ( Figure 6—figure supplement 1A , B and E ) , a reduction of total Brp intensity at individual synapses and across the NMJ terminal ( Figure 6—figure supplement 1C–D ) , a reduction in VGlut intensity at the NMJ terminal ( Collins et al . , 2006 ) , and reduced mEJP frequency ( Collins et al . , 2006 ) and Figure 6—figure supplement 2 ) . All these presynaptic defects were rescued by wnd mutations . Meanwhile , a previously described Wnd-independent defect in quantal content remains in hiw;wnd double mutants ( Figure 6—figure supplement 2 , and Collins et al . , 2006 ) . Taken together , these observations indicate that activation of the Wnd signaling pathway in multiple scenarios—in unc-104 mutants , in hiw mutants , or when Wnd is ectopically over-expressed—all lead to a shared set of characteristic defects in presynaptic structure and synaptic transmission . The above observations suggest that activation of the Wnd signaling pathway in unc-104 mutants is the cause of many of the synaptic defects associated with Unc-104 . However how Wnd signaling affects synapses , including both AZ and SV protein localization , is unknown . One of the hallmarks of the unc-104 mutant phenotypes is the aggregation of protein components of both SVs and AZs in neuronal cell bodies ( Hall and Hedgecock , 1991; Pack-Chung et al . , 2007 ) . Strikingly , wnd mutations enhanced this phenotype: SV and AZ components were dramatically increased in cell bodies of unc-104; wnd double mutants ( Figure 7A , C , E , and Figure 7—figure supplement 1 ) . We observed a similar increase when either bsk or fos was inhibited in the unc-104 mutant background ( Figure 7—figure supplement 2 ) . The exacerbated SV and AZ component accumulation in the neuronal cell bodies of unc-104; wnd double mutants seemed at first glance hard to reconcile with the suppression of synaptic defects in these animals . However , they may be explained by an action of the Wnd pathway within a feedback circuit to reduce the build-up of SV and AZ components in neuronal cell bodies . As a nuclear signaling cascade , we hypothesized that Wnd signaling may achieve this effect by down-regulating the expression levels of SV and AZ components . In this case , such reduction in global levels could account for the reduced levels at synaptic terminals in unc-104 mutants . To test this hypothesis , we carried out quantitative immunohistochemistry to estimate the total cellular levels of Brp and VGlut in motoneurons based on respective intensities in cell body , axonal and synaptic compartments ( Figure 7F ) . We found that compared with wildtype animals , the total levels of Brp and VGlut are reduced by 65% and 80% respectively in unc-104 mutants ( Figure 7F ) . The total levels of both proteins were restored in unc-104; wnd double mutants , with increases observed in all of the compartments ( cell bodies , axons , and synapses ) . ( Figure 7F ) . We suspect that the fact that motoneurons are only a fraction of neurons in the Drosophila nervous system prohibited our detection of global changes in these proteins by Western blot ( data not shown ) . However we observed this trend in total levels of VGlut ( Figure 7—figure supplement 3 ) . This may be facilitated by the fact that most of the glutamatergic neurons in larval VNCs are motoneurons . These observations , combined with the observations that Wnd activation reduces the levels of Brp and VGlut at NMJs , suggest a model in which activated Wnd signaling leads to a down-regulation of the expression levels of multiple pre-synaptic proteins in unc-104 mutants . This response may serve to counteract their buildup in neuronal cell bodies while causing the observed defects in synapse structure and function . In support of this , unc-104 mutations led to decreased expression of a vglut-DsRed transcriptional reporter in motoneurons , in a Wnd-dependent manner ( Figure 7G and H ) . This , together with the involvement of transcription factor Fos in restraining VGlut and Brp buildup in cell bodies ( Figure 7—figure supplement 2 ) , implies that Wnd signaling may inhibit presynaptic protein expression at the transcriptional level . We then asked whether Wnd signaling may also play a role during early synaptic development , when precise temporal coordination of presynaptic protein expression is critical for establishing synaptic contacts . Previous studies suggest that while Wnd/DLK signaling becomes activated in injured neurons , it is normally highly restrained in uninjured animals ( Collins et al . , 2006; Xiong et al . , 2010 ) , and , in contrast to its essential role in responses to axonal injury , roles for Wnd in developmental axonal outgrowth or synapse formation have not been previously described ( Collins et al . , 2006 ) . We investigated this more carefully via analysis of wnd-null mutants during embryonic stages of axonal outgrowth and synapse formation ( during embryonic stages 14 through 17 ) , and observed that NMJ synapse development progressed normally ( Figure 8—figure supplement 1 ) . However , wnd mutants showed a premature onset of SV protein expression ( Figure 8 ) . In wild type embryos VGlut protein expression was undetectable until late embryonic stage 15 , which coincides with the time at which motoneuron axons first reach their target muscles ( Johansen et al . , 1989 ) . VGlut first appears in cell bodies ( Figure 8A–C ) and then becomes detectable at synapses at stage 16 and beyond ( Figure 8D ) , consistent with continued expression and transport to synaptic terminals as the NMJ terminal expands . In wnd mutants VGlut expression levels resemble wild type at stage 16 and beyond , however its initial expression in cell bodies became apparent at an earlier time point , in early embryonic stage 15 ( Figure 8B–C ) . SytI intensity was also increased in wnd mutants at early embryonic stage 16 ( Figure 8E–F ) . These results suggest that endogenous Wnd signaling may function in pacing the onset of expression of SV proteins . This function may prevent unwanted buildup at inappropriate time points before synaptic contacts are established . Across our cumulative observations , we noticed an interesting correlation between the activity of Wnd signaling and the appearance of presynaptic proteins localized in motoneuron cell bodies . During development , the role of Wnd in restraining VGlut was most significant immediately after the onset of VGlut expression in cell bodies and before its transport to synaptic terminals ( Figure 8B–D ) . The same Wnd dependent pattern was observed for SytI ( Figure 8E–F ) . Similarly , in unc-104 mutants , the highly elevated activity of Wnd signaling to restrain the expression of presynaptic components coincided with their accumulation in neuronal cell bodies ( Figure 7 ) . If mislocalized or misregulated presynaptic proteins play a causal role in the induction of Wnd signaling , then over-expression of these proteins should lead to an activation of Wnd signaling independently of their impact on synaptic function . To test this , we individually overexpressed three different components pan-neuronally: Brp , SytI and VGlut . Each caused a significant induction of the Wnd responsive puc-lacZ reporter ( Figure 8G ) . In contrast , over-expression of other proteins ( Luciferase ( Figure 8G ) and membrane localized GFP ( Figure 4B ) , had no effect . Over-expression of VGlut can also lead to increased synaptic transmission ( Daniels et al . , 2011 ) , however we observed that over-expression of a non-functional VGlut transgene , VGlutA470V , which has no effect upon synaptic physiology ( Daniels et al . , 2011 ) , caused a similar induction of puc-lacZ expression ( Figure 8G ) . These results , taken together with the cell autonomous nature of Wnd activation in unc-104 mutants ( Figure 2—figure supplement 1 ) , suggest that Wnd signaling is sensitive to accumulations of presynaptic proteins in cell bodies .
The kinesin-3 family member Unc-104/KIF1A is known to be an important mediator of synaptic development and maintenance: mutations in unc-104 and its homologues inhibit the localization of SV and AZ precursors to nascent synapses , causing profound synaptic defects ( Barkus et al . , 2008; Hall and Hedgecock , 1991; Kern et al . , 2013; Li et al . , 2016; Niwa et al . , 2016; Otsuka et al . , 1991; Pack-Chung et al . , 2007; Yonekawa et al . , 1998; Zhang et al . , 2016 ) . While these synaptic defects have been considered logical outcomes of defective transport , we found that major aspects , including impaired AZ addition and development of synaptic boutons , are not mediated by a direct transport role for the Unc-104 protein . Rather , the unc-104null;wnd double mutants reveal separable functions for Unc-104: ( 1 ) Transport of SV precursors to synaptic terminals is likely a direct function , since the failure to deliver adequate VGlut , Syt1 and CSP persists in unc-104null;wnd double mutants , and the Unc-104 homologue KIF1A physically interacts with SV precursors ( Okada et al . , 1995 ) . ( 2 ) Regulation of bouton formation and localization of AZs is an indirect function , since this defect can be rescued in unc-104null;wnd double mutants ( in the complete absence of Unc-104 function ) . With the knowledge that the Wnd/DLK signaling pathway is activated in unc-104 mutants , it is now worth considering whether it contributes to other phenotypes previously described for Unc-104 and its homologues in other species . These include impaired dendritic branching ( Kern et al . , 2013 ) , increased microtubule dynamics ( Chen et al . , 2012 ) , and failed neuronal remodeling ( Park et al . , 2011 ) . It now becomes likely that these defects are mediated by Wnd/DLK , since Wnd/DLK signaling has previously been shown to influence microtubule growth ( Hirai et al . , 2011; Lewcock et al . , 2007 ) , neuronal remodeling ( Kurup et al . , 2015; Marcette et al . , 2014 ) and dendrite growth ( Wang et al . , 2013 ) . Unc-104/Kif1a mutants also show accelerated motor circuit dysfunction in aging animals ( Li et al . , 2016 ) , impaired BDNF-stimulated synaptogenesis ( Kondo et al . , 2012 ) and neuronal death ( Yonekawa et al . , 1998 ) . These phenotypes may also be facilitated by activation of DLK , which impairs synaptic development and function ( this study and Nakata et al . , 2005 ) , and has also been shown to mediate neuronal death in some contexts ( Chen et al . , 2008; Pozniak et al . , 2013; Welsbie et al . , 2013; Welsbie et al . , 2017 ) . Human mutations in KIF1A have been associated with hereditary spastic paraplegia ( SPG30 ) ( Fink , 2013 ) , and hereditary sensory and autonomic neuropathy type IIC ( HSN2C ) ( Rivière et al . , 2011 ) . The possibility that DLK activation mediates deleterious aspects of these disease pathologies becomes an interesting future question . How does the Wnd pathway become activated in unc-104 mutants ? The mechanism ( s ) that lead to activation of Wnd and its DLK homologues are of general interest for their roles in axonal regeneration as well as degeneration and neuronal death . In addition to axonal injury ( Watkins et al . , 2013; Welsbie et al . , 2013; Xiong et al . , 2010 ) , disruption of microtubule and/or actin/cortical cytoskeleton can lead to activation of DLK ( Valakh et al . , 2013 , 2015 ) . Moreover , many studies have noted a role for DLK signaling in mediating structural changes in neurons downstream of manipulations that disrupt cytoskeleton ( Bounoutas et al . , 2011; Marcette et al . , 2014; Massaro et al . , 2009 ) . Since the cytoskeleton is a closely functioning partner of all motor proteins , and is also implicitly affected by axonal injury , it is possible that these manipulations share a similar underlying mechanism with that of unc-104 mutations . While disruption of cytoskeleton should impair transport by many motor proteins , mutations that impair kinesin-1 and dynein do not lead to activation of Wnd ( Figure 4 and Xiong et al . , 2010 ) . This specificity suggests that disruption of Unc-104 mediated transport , potentially via mislocalization of Unc-104’s cargo , mediates Wnd/DLK’s activation after cytoskeletal disruption and potentially after axonal injury . This line of reasoning leads to further consideration of Unc-104’s cargo . Our live imaging data do not support a simple model that Wnd is a cargo of Unc-104 ( Figure 7—figure supplement 2 ) . Known cargo of Unc-104 are important for the assembly and function of synapses ( Goldstein et al . , 2008 ) , so does Wnd activation occur in response to an impairment in synaptic assembly or function ? We think this is unlikely , since mutations in rab3-gef , liprin-α and vglut , which impair presynaptic assembly and function , do not cause activation of Wnd ( data not shown ) . Instead , we note an intriguing correlation between the localization and abundance of presynaptic proteins with Wnd’s activation: Wnd signaling becomes activated in unc-104 mutants , which accumulate presynaptic proteins in the neuronal cell body . We noticed a similar role for endogenous Wnd in wild type neurons during the onset of embryonic NMJ development ( Figure 8 ) . These stages correspond to the onset of synaptic protein expression , before substantial transport to synaptic terminals , hence represent a time in which levels are high in the cell body . Consistent with the idea that Wnd signaling is sensitive to mislocalized presynaptic proteins , ectopic overexpression of several different presynaptic proteins caused an elevation in Wnd signaling in uninjured neurons ( Figure 8G ) . These observations suggest a model that accumulations of presynaptic proteins , as a feature of aberrant cargo transport , are ‘sensed’ by Wnd signaling ( Figure 9 ) . While previous studies in C . elegans ( Nakata et al . , 2005; Yan et al . , 2009 ) have suggested that DLK activation may impair synaptic development ( altering the size and spacing of active zones ) , the regulation of total levels of presynaptic proteins and the relationship with Unc-104 provides a new view into Wnd/DLK’s function and mechanism . In addition to VGlut , SytI , CSP-1 , Brp , Liprin-α and Cac , we suspect that Wnd signaling restrains the expression of a cohort of presynaptic proteins . Regulation of multiple targets required for synapse development and maintenance can explain the severe defects in unc-104-hypomorph mutants , and the dramatic suppression by disruption of the Wnd pathway ( Figures 2 and 3 ) . In support of this idea , presynaptic defects in unc-104-hypomorph mutants can be partially rescued by overexpressing Brp ( Kern et al . , 2013 ) or Rab3 ( Zhang et al . , 2016 ) . It is interesting to consider that the targets of Wnd regulation are also abundant presynaptic proteins , and are thought to be major cargo for axonal transport ( Figure 9 ) . Down-regulation of these proteins in response to defects in their transport or after axonal damage may comprise a stress response mechanism to prevent unwanted buildup or wasted cellular resources . This role may allow neurons adapt to stresses that impair axonal transport by counteracting cargo buildup . But the resulting reduction in synaptic protein levels can also be maladaptive , leading over time to synapse failure and/or loss . This feature may contribute to synaptic pathologies associated with defects in axonal transport . How does Wnd signaling regulate presynaptic proteins ? The regulation of the vglut-promoter-DsRed reporter and requirement for the Fos transcription factor suggests the involvement of transcriptional regulation ( Figure 7G–H ) , and we also noticed mildly increased levels of Brp , VGlut and Cac transcripts in unc-104;wnd double mutants ( data not shown ) . However we are limited in our ability to detect total changes in mRNA and protein levels from whole nerve cord preparations by the fact that the Wnd signaling pathway may not be acting in all cell types . It is also possible that additional post-transcriptional mechanisms , such as regulation of protein stability or translation , or bulk turnover via autophagy , factor into the regulation of presynaptic proteins by Wnd . A recent study has suggested that DLK may activate the integrated stress response pathway in mammalian neurons ( Larhammar et al . , 2017 ) . On the other hand , previous studies have separately linked Unc-104 , Wnd signaling and AP-1 ( Fos ) to autophagy ( Guo et al . , 2012; Shen and Ganetzky , 2009; Stavoe et al . , 2016 ) . It will be interesting to determine whether translation and/or autophagy contribute to the regulation of presynaptic proteins by Wnd signaling . Many previous studies have reported links between JNK signaling and kinesin-driven transport ( Verhey and Hammond , 2009 ) , with some observations suggesting that JNK signaling may directly regulate the function of kinesin-1 , modulating its cargo binding , affinity for microtubules and its processivity ( Fu and Holzbaur , 2013; Horiuchi et al . , 2007; Morfini et al . , 2006; Stagi et al . , 2006; Sun et al . , 2011 ) . Our finding of a separate role for JNK signaling in regulating the abundance of transported cargo adds a new layer of complexity to interpreting phenotypes of axonal transport defects . A commonly described defect is the presence of accumulations of cargo within axons , referred to as ‘traffic jams’ . These defects have been noted for many different mutations , including kinesin-1 and dynein subunits ( Gindhart et al . , 1998; Hurd and Saxton , 1996; Martin et al . , 1999 ) , and also in mutants for wnd and other members of JNK signaling pathways ( Bowman et al . , 2000; Horiuchi et al . , 2005 , 2007 ) . Does a failure to regulate excess protein cargo contribute to the presence of the jams ? Intriguingly , unc-104 mutations do not cause this type of ‘traffic jams’ in axons , but instead leads to accumulations of synaptic proteins in cell bodies , which correlates with the activation of Wnd signaling . Finally , it is interesting to compare Wnd’s role in tuning levels of presynaptic proteins with previously identified roles for DLK in promoting cell death ( Chen et al . , 2008; Huntwork-Rodriguez et al . , 2013; Pozniak et al . , 2013; Watkins et al . , 2013; Welsbie et al . , 2013 ) . While Kif1a mutant mice show early signs of neuronal death and degeneration , which may potentially be mediated by activation of DLK , unc-104 mutants in C . elegans and Drosophila lack hallmarks of cell death and synaptic degeneration ( Hall and Hedgecock , 1991; Kern et al . , 2013; Pack-Chung et al . , 2007 ) . In analogy with other stress response pathways , regulation of presynaptic proteins may comprise a first order response that may facilitate adaptation to stress , while cell death could be the consequence when compensatory mechanisms fail . Inhibition of synaptic proteins is , alone , a pathology that becomes relevant for long term maintenance of synapses and their function over time , since synthesis and transport of new synaptic proteins likely needs to occur throughout the long lifespan of a neuron . Interestingly , previous studies have linked activation of JNK signaling to synapse loss in aged animals ( Ma et al . , 2014; Sclip et al . , 2014; Voelzmann et al . , 2016 ) . Exciting future work lies ahead to further understand DLK’s activation and its consequences in different models of neuronal injury , disease , and aging . Since acceptance of this work , a new publication has shown that DLK signaling contributes to pathology in multiple mouse models of neurodegenerative diseases ( Le Pichon et al . , 2017 ) . This study provides further suggestive evidence that DLK signaling activation can contribute to synapse loss in disease conditions .
The following strains were used in this study: Canton-S , hiwΔN ( Wu et al . , 2005 ) , wnd1 , wnd3 , wnddfED228 ( Collins et al . , 2006 ) , UAS-wndkinase dead-GFP ( Xiong et al . , 2010 ) , MiMIC-wnd-GFP ( Venken et al . , 2011 ) , unc-104O3 . 1 , unc-104P350 ( Barkus et al . , 2008 ) , unc-104bris ( Medina et al . , 2006 ) , unc-104d11204 ( Thibault et al . , 2004 ) , unc-10452 ( Pack-Chung et al . , 2007 ) ; UAS-cacophony-GFP ( Kawasaki et al . , 2004 ) , OK6-Gal4 ( Aberle et al . , 2002 ) , OK319-Gal4 , OK371-Gal4 ( Mahr and Aberle , 2006 ) , m12-Gal4 ( Ritzenthaler et al . , 2000 ) , BG380-Gal4 ( Budnik et al . , 1996 ) , elav-Gal4C155 ( Lin and Goodman , 1994 ) , UAS-fosDN ( Eresh et al . , 1997 ) , UAS-bskDN ( Weber et al . , 2000 ) , khc8 , khc27 ( Brendza et al . , 1999 ) , khck13314 ( Spradling et al . , 1999 ) , Liprin-αF3ex15 , Liprin-αR60 ( Kaufmann et al . , 2002 ) , UAS-VGlut-GFP , UAS-VGlutA470V-GFP ( Grygoruk et al . , 2010 ) , UAS-Brp-GFP ( Bloomington ( BL ) 36291 and 36292 ) , UAS-SytI-GFP ( BL6925 and 6926 ) , Rab7-liprin-α-GFP ( Zhang et al . , 2016 ) , Rab3rup ( Graf et al . , 2009 ) , UAS-YFP-Rab3 , UAS-YFP-Rab3Q80L , UAS-YFP-Rab3T35N ( Zhang et al . , 2007 ) , uas-mcd8-ChRFP ( Schnorrer , 2009 . 5 . 11 ) , puc-lacZE69 ( Martín-Blanco et al . , 1998 ) , vglut promoter-DsRed ( gifts from Daniels and Diantonio ) , RNAi lines: moody RNAi ( control ) , Octβ2R RNAi ( vdrc 104524 , control ) , unc-104 RNAi ( vdrc 23465 , I and TRiP BL43264 , II ) , wnd RNAi ( vdrc 103410 and vdrc 26910 ) , Rab3 RNAi ( TRiP BL31691 and BL34655 ) , UAS-Dcr2 was a gift from Stephan Thor ( Linköping Université , Linköping Sweden ) . Flies were raised at 25°C or 29°C ( as indicated for certain RNAi knock-down ) on standard Semidefined yeast-glucose media ( Backhaus et al . , 1984 ) . To generate vglut-DsRed reporter flies , genomic sequence spanning 5 . 3 upstream of the ATG start codon for vglut ( CG9887 ) was cloned into a plasmid derived from pCaSpeR-AUG-bGal ( Thummel et al . , 1988 ) , in which lacZ was replaced with DsRed . T4-NLS ( Barolo et al . , 2004 ) coding sequence , such that the expressed DsRed would concentrate in the nucleus . Third-instar larvae were dissected in ice-cold PBS , then fixed in 4% formaldehyde ( FA ) in PBS/HL3 solution for 3 min for Cac-GFP , 10 min for Brp and GluRIII/GluRIIC staining or 20 min for other antibody staining , followed by blocking in PBS with 0 . 1% Triton ( PBT ) containing 5% Normal Goat Serum ( NGS ) block for 30 min . Control and experimental animals were always dissected , fixed and stained in the same condition and imaged in parallel using identical confocal settings . Embryos were dissected , fixed and stained as described in ( Featherstone et al . , 2009; Lee et al . , 2009 ) . In brief , embryos were collected for 30–60 min on Molasses plates and kept in 18°C ( for stage 14 to 16 ) or 25°C ( for stage 17 ) overnight . Early-stage embryos ( 14-16 ) were dechorionated , sorted ( based on GFP ) , staged ( based on gut morphology [Hartenstein , 1993] ) and dissected ( tungsten needles ) on negatively charged slides . Stage 17 embryos ( 20–21 hr AEL ) were dissected with Vet glue ( Vetbond ) in PBS ( PH = 7 . 3 ) on Sylgard-coated coverslips . Bouin’s fixation for 5 min was used for all antibodies staining but Synapsin staining ( 4% PFA for 25 min ) . The examined unc-104-null alleles include P350/P350 and 52/52 . Primary antibody and secondary antibody incubations were conducted in PBT containing 5% NGS at 4°C overnight and at room temperature for 2 hr , respectively , with three 10 min washes in PBT after each antibody incubation . The following primary antibodies and dilutions were used: ms anti-Brp ( DSHB Cat# nc82 Lot# RRID:AB_2314866 ) , 1:200; ms anti-Synapsin ( DSHB Cat# 3C11 ( anti SYNORF1 ) Lot# RRID:AB_528479 ) , 1:50; ms anti-CSP-1 ( DSHB Cat# DCSP-1 ( ab49 ) Lot# RRID:AB_2307340 ) , 1:100; ms anti-DLG ( DSHB Cat# 4F3 anti-discs large Lot# RRID:AB_528203 ) , 1:1000; Rb anti-GluRIII ( gift from Diantonio lab ) , 1:2500; Rb anti-SytI ( gifts from Noreen Reist , Mackler et al . , 2002 ) , 1:400; Rb anti-Unc-104 ( Pack-Chung E; Nat Neurosci . 2007 Cat# unc-104 Lot# RRID:AB_2569094 ) , 1:500; ms anti-lac-Z ( DSHB Cat# 40-1a Lot# RRID:AB_528100 ) , 1:100; Rb anti-Phospho-Smad1/5 ( Cell signaling ) , 1:100; Rb anti-DsRed ( Clontech Laboratories , Inc . Cat# 632496 Lot# RRID:AB_10013483Clontech ) , 1:1000; Rat anti-elav ( DSHB Cat# Rat-Elav-7E8A10 anti-elav Lot# RRID:AB_528218 ) , 1:50; A488 rabbit anti-GFP ( Molecular Probes Cat# A-21311 also A21311 Lot# RRID:AB_221477 ) , 1:1000 and Alexa488/cy3/Alexa647 conjugated Goat anti-HRP ( Jackson ImmunoResearch ) , 1:300 . Rabbit anti-VGlut ( gift from Diantonio lab ) , 1:10000 , staining was carried as described in Daniels et al . ( 2008 ) . For secondary antibodies we used Cy3- or A488-conjugated goat anti-rabbit or anti-mouse 1:1000 ( Invitrogen ) . Confocal images were collected as described in ( Füger et al . , 2012; Xiong et al . , 2010 ) . Similar settings were used to collect all compared genotypes and conditions . The identification and quantification of the % unapposed PSD was based on manual counts of the total number of individual GluRIII-labeled puncta ( on either muscle 4 or 26/27/29 , where indicated ) , scored for the presence or absence of an apposing AZ component ( Brp or Liprin-α-GFP ) . To affirm that AZ components were indeed completely absent , confocal settings and brightness levels were optimized for the weakest signals in unc-104 mutants . Since the same settings were used for all genotypes some pixels for AZ components were necessarily over-exposed in the wt controls . For measurements of intensity levels , using Volocity software , only raw images acquired together using the same confocal settings were compared . To measure VGlut and Brp levels in axons and NMJs , we used staining for HRP ( which labels neuronal membrane ) to define the region of interest . For cell bodies , we selected the signal above a specified threshold , ( To specify the threshold , we checked the background signal and examined the fluorescence signal distribution from multiple images . The threshold was chosen to be as at least 3 fold higher than the background and 1 SD higher than the center of the distribution . The same threshold was applied to all compared images ) . To estimate the total VGlut or Brp level within a single motoneuron ( Figure 6 ) , we summed measurements of: ( a ) total intensity for individual cell bodies ( located in the dorsal midline of the ventral nerve cord ) , ( b ) total intensity for individual NMJ nerve terminals at muscle 4 , and ( c ) estimated total intensity within a motoneuron axon , calculated from mean intensity in axonal segments , based on the assumption of 32 motoneuron axons per nerve and an average axon length of 1 mm . When imaging nerve cord , we used 0 . 8 μm step size for the z-stack and focused on the posterior and central nerve cord , corresponding to A4-A8 . When imaging axons or the NMJ , we used 0 . 4 μm step size for z-stacks . Axonal segments were imaged 900 μm away from the nerve cord . NMJ images were collected at segment A3 for muscle 4 or ( when using the m12Gal4 driver ) for muscle 26 , 27 and 29 , which are innervated by the SNc neuron . puc-lacZ level was measured within the nucleus region for motoneurons , selected by P-smad staining in the dorsal regions of A4-A8 in the ventral nerve cord . For live imaging analysis of GFP-wnd-KD transport , third instar larvae were dissected in the center of a circle reinforcement label ( Avery ) in HL3 solution ( Stewart et al . , 1994 ) with 0 . 45 mM calcium . Larvae were pinned at the head , tail and 2 lower corners , the pins were then pushed into sylgard so that the coverslip could lie directly on top of the reinforcement label ( and the larva ) , and excess HL3 solution was removed before imaging on an inverted microscope . Images were collected at 0 . 3 Hz for 5 min at 40x magnification in segmental nerves at a location 900 μm distal to the nerve cord . The images were then processed in imageJ with a kymograph plugin ( Jens Rietdorf and Arne Seitz ) and further analyzed in MATLAB with a program written to determine vesicle segmental speed and duration ( described in Ghannad-Rezaie et al . , 2012 ) . 25–30 3rd brains were collected in PBS and homogenized for each sample . The following antibodies and dilutions were used: rb anti-Wnd 4–3 ( Collins et al . , 2006 ) 1:700; rb anti-VGlut ( Daniels et al . , 2008 , 1:10000; ms anti-Brp ( NC82 ) , 1:100; ms anti-β-tubulin ( 1E7 , DSHB ) , 1:1000; and rb anti-unc-104 ( Gift from Tom Schwarz lab ) , 1:500 . The blots were probed with HRP conjugated secondary antibodies: Gt at-ms and Gt at-rb at the dilution of ( 1:5000 ) and imaged with either film or an Odyssey CLx imager ( LI-COR ) . The nerve crush assay was carried out as described ( Xiong et al . , 2010 ) , and animals were fixed either 9 or 18 hr after the injury . Axonal regeneration ( sprouting ) was quantified by measuring the number of injured axons that contained more than 5 branches at 9 hr , and the length of the longest branch at 18 hr . Third instar larvae were dissected within 3 min in HL3 solution containing 0 . 65 mM Calcium at 22°C . Muscle 6 at segment A3 was located by the use of an OLYMPUS BX51WI scope with a 10x water objective and then recorded intracellularly with an electrode made of thick wall glass ( 1 . 2 mm x 0 . 69 mm ) pulled by SUTTER PULLER P-97 . Amplifier GeneClamp 500B and digitizer Digidata 1440A were used . The recording was only used if the resting potential was negative to −60 mV and muscle resistance was >5 mΩ . A GRASS S48 STIMULATOR was used to obtain a large range of stimulation voltage range ( 1–70V ) . We noticed that hiw mutants and unc-104 mutants required a higher stimulus to recruit the 2nd axon that intervenes Muscle 6 ( 10–40V were required in hiw and unc-104 mutants , as opposed to 2–8V in wild type ) . To ensure that we could always recruit both axons , for each muscle we tested a range of stimulation voltage ( 1–70V ) to find the threshold which triggered the largest response within the testing range . A stimulus slightly larger than this threshold was then used at a frequency of 0 . 2 Hz and duration of 1 ms for EJP measurements . Axon Laboratory software was used for acquisition and the Mini Analysis program ( Synaptosoft Inc ) was used for analysis of mEJP frequency and amplitude . Parameters for Mini Analysis were set as: 0 . 2 ( threshold ) , 1 ( area threshold ) , 30 , 000 ( period to search a local maximum ) , 40 , 000 ( period to search a decay time ) , 40 , 000 ( time before a peak for baseline ) , 20 , 000 ( period to average a baseline ) , 0 . 6 ( fraction of peak to find a decay time ) and detect complex peak . Quantal content and EJP amplitudes were corrected for non-linear summation using the revised Martin correction factor as described in ( Kim et al . , 2009; Morgan and Curran , 1991 ) . In brief , to accurately estimate the potential change if units sum linearly , the following equation was applied when EJP amplitude was larger than 15% of Eresting potential: corrected EJP = EJP/ ( 1 f ( EJP/Edriving force ) ) , where Edriving force=Eresting potential-Ereversal potential and f = the membrane capacitance factor ( Δt/τ ) . At the Drosophila NMJ , Ereversal potential is estimated around 0 mV , Eresting potential from our recordings is around −70 mV and f = 0 . 55 ( Kim et al . , 2009; Morgan and Curran , 1991 ) . The minimum sample size for animal lethality and motility was determined by power analysis using: sample size = 2*SD2 ( Zα/2+Zβ ) 2/d2 , where SD = standard deviation , d = effect size and Zα/2 or Zβ are constants when α = 0 . 5 and β = 80% ( Jung , 2010 ) . For other assays ( including synapse morphology , electrophysiology , puc-lacZ expression and axonal sprouting ) , the sample size and biological replicate number was determined for comparison with previous studies ( Xiong et al . , 2010; Xiong et al . , 2012; Collins et al . , 2006; Kern et al . , 2013; Ghannad-Rezaie et al . , 2012 ) . For axonal regeneration ( sprouting ) , a total of 30 axons were measured from 8 individual animals . For comparison of intensities via confocal imaging and for structural analysis of synapse number and bouton number , at least 12 clusters of motoneurons within dorsal abdominal segments 2–5 ( which contain 10 cell bodies per cluster ) , taken from at least 6 animals ( 2–3 clusters per animal ) or 12 NMJs from at least 6 animals ( 2 NMJs per animal ) were examined per genotype for each criteria quantified . For analysis of axonal transport , kymographs were generated and analyzed from a total of 60 axons from 10 individual animals . For electrophysiology , for each genotype we analyzed 35 individual EJP traces and a 45s-long mEJP trace per muscle for at least 11 muscles ( derived from at least 6 independent animals – 2 muscles per animal ) . Data was analyzed by either Student’s t-test ( two groups ) or one-way ANOVA followed by Tukey test ( multiple groups ) . Normality of datasets for parametric/ANOVA was confirmed using the D’Agostino-Pearson-omnibus K2 test . p values smaller than 0 . 05 were considered statistically significant . All p values are indicated as *p<0 . 05 , **p<0 . 01 , and ***p<0 . 001 and ****p<0 . 0001 . Data are presented as mean ±SEM . | Each nerve cell , or neuron , has a long nerve fiber – called an axon – that forms specialized sites for information exchange – called synapses – with other cells . Many molecules work at synapses to coordinate the exchange of information . These molecules are largely made in the central part of the neuron – known as the cell body – and are then transported along the axon to the synapses . The transport of these molecules is carried out by proteins known as molecular motors . One molecular motor , called KIF1A in humans and Unc-104 in fruit flies , is thought to be a major transporter of synaptic molecules . Mutations that hinder this molecular motor result in neurons failing to form synapses and , instead , synaptic components accumulate in the cell body . However , it was not clearif Unc-104 does actually carry all of the components needed to assemble synapses along axons , or if it influences synapse formation in another way . Now , Li , Zhang et al . report new evidence that supports the second of these two hypotheses . The experiments made use of fruit flies in which the gene for Unc-104 had been deleted , and revealed that inhibiting enzymes in a specific signaling pathway could reverse the synaptic problems caused by the loss of Unc-104 . The signaling pathway , which is conserved between flies and humans , involves an enzyme that is called Wnd in flies and DLK in humans . The Wnd/DLK signaling pathway was previously known to regulate how neurons respond when their axons are damaged ( either by growing new axons or dying , depending on the context ) . Further investigation by Li , Zhang et al . revealed that signaling via the Wnd enzyme becomes triggered whenever the Unc-104 molecular motor is impaired . This activation correlates with the build-up of synaptic proteins in the cell body . Once activated , the pathway then reduces the total amount of synaptic proteins that the cell makes . This reduction matches the neuron’s reduced ability to transport them along the axon , and may help the neuron to adapt when axonal transport is impaired . However , the reduction in synaptic proteins also impaired the exchange of information at the synapses . These findings suggest how DLK could be behind problems with synapses in diseases in which transport along axons is impaired . These diseases include hereditary spastic paraplegia , which has been linked to mutations in human KIF1A , and may also include ALS and Alzheimer’s disease , which have recently been linked to DLK . DLK has received recent attention as a candidate drug target because it contributes to the deterioration of damaged neurons . These new findings further expand that interest by suggesting that inhibiting DLK may help neurons to maintain working synapses , which is more useful than simply preventing damaged neurons from dying . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2017 | Restraint of presynaptic protein levels by Wnd/DLK signaling mediates synaptic defects associated with the kinesin-3 motor Unc-104 |
Two parallel pathways produce cholesterol: the Bloch and Kandutsch-Russell pathways . Here we used stable isotope labeling and isotopomer analysis to trace sterol flux through the two pathways in mice . Surprisingly , no tissue used the canonical K–R pathway . Rather , a hybrid pathway was identified that we call the modified K–R ( MK–R ) pathway . Proportional flux through the Bloch pathway varied from 8% in preputial gland to 97% in testes , and the tissue-specificity observed in vivo was retained in cultured cells . The distribution of sterol isotopomers in plasma mirrored that of liver . Sterol depletion in cultured cells increased flux through the Bloch pathway , whereas overexpression of 24-dehydrocholesterol reductase ( DHCR24 ) enhanced usage of the MK–R pathway . Thus , relative use of the Bloch and MK–R pathways is highly variable , tissue-specific , flux dependent , and epigenetically fixed . Maintenance of two interdigitated pathways permits production of diverse bioactive sterols that can be regulated independently of cholesterol .
Cholesterol is an essential structural component of vertebrate cell membranes ( Maxfield and Tabas , 2005 ) and a precursor of vital end products such as bile acids ( Russell , 2009 ) and steroid hormones ( Sih and Whitlock , 1968 ) . First identified as a crystalline component of gallstones more than 200 years ago , cholesterol consists of a rigid , planar tetracyclic nucleus and a flexible , iso-octyl side-chain at carbon 17 ( Nes , 2011 ) . The molecule is synthesized entirely from acetate through a complex series of over 30 enzymatic reactions that are clustered into four major processes: condensation of acetate to isoprene , polymerization of isoprene to squalene , cyclization of squalene to lanosterol , and finally the conversion of lanosterol to cholesterol ( Figure 1 ) . 10 . 7554/eLife . 07999 . 003Figure 1 . Schematic representation of the Bloch and Kandutsch–Russell ( K–R ) pathways for the enzymatic conversion of squalene to cholesterol . The Bloch pathway , indicated by solid black arrows , is shown on the left . The Kandutsch–Russell pathway , indicated by red arrows , is shown on the right . Additional potential sites of crossover from the Bloch to K–R pathway are indicated by broken arrows . Sterol intermediates that were not measured using deuterium water labeling are shown in gray . DOI: http://dx . doi . org/10 . 7554/eLife . 07999 . 003 Two intersecting pathways have been described for biosynthesis of cholesterol from lanosterol . The two pathways use the same catalytic steps and are distinguished by the stage at which the double bond at C24 in the side chain is reduced . Bloch et al . ( 1965 ) proposed that reduction of the double bond in the side chain ( Δ24 ) is the last reaction in the pathway ( Figure 1 , black arrows ) . Thus , in the Bloch pathway , cholesterol synthesis proceeds via a series of side-chain unsaturated intermediates to desmosterol , which is reduced by DHCR24 to cholesterol . Subsequently , Kandutsch and Russell ( Kandutsch and Russell , 1960a , 1960b ) reported that the preputial glands of mice synthesized dihydrolanosterol and other side-chain saturated intermediates that were different from those in the Bloch pathway . They proposed an alternative pathway , the Kandutsch–Russell ( K–R ) pathway , in which the Δ24 bond of lanosterol is reduced and the conversion of dihydrolanosterol to cholesterol proceeds via 7-dehydrocholesterol ( Figure 1 , red arrows ) using the same enzymes as the Bloch pathway . Later studies showed the presence of saturated side chain intermediates in brain , skin , and eventually in all tissues examined ( Schroepfer , 1982 ) . The catalytic mechanisms and regulation of the enzymes that catalyze the Bloch and K–R pathways have been extensively investigated but much less is known about their relative use and physiological significance . Cholesterol biosynthesis is usually measured using radioisotope methods ( Dietschy and Spady , 1984 ) that are highly sensitive but do not provide information about the turnover of intermediate sterols in the biosynthetic pathway . Flux through the Bloch and K–R pathways has not been systematically studied in cultured cells or in vivo . Therefore , the relative use of the two pathways and their responses to changes in cholesterol availability are not known . The reason why both pathways have been maintained is also not known . The biosynthetic intermediates of the two pathways have powerful , but distinct , effects on cholesterol homeostasis ( Yang et al . , 2006; Lange et al . , 2008 ) , fatty acid synthesis ( Spann et al . , 2012 ) , and inflammation ( Spann et al . , 2012 ) . Varying the concentrations of specific cholesterol biosynthetic intermediates by differential use of the Bloch and K–R pathways may thus contribute to regulation of diverse cellular processes . The analytical challenges associated with distinguishing the multiple closely related sterol intermediates in each pathway , and interpreting the complex patterns of isotope enrichment generated by in vivo labeling have been major obstacles to in vivo studies of the Bloch and K–R pathways . Kelleher and Masterson ( 1992 ) demonstrated that stable isotope methods and isotopomer spectral analysis ( ISA ) could be used to measure the flux of cholesterol biosynthetic intermediates such as lathosterol in cultured cells and in living animals using gas chromatography–MS ( Lindenthal et al . , 2002 ) . More recently , McDonald et al . ( 2012 ) developed a liquid chromatography tandem mass spectrometry ( LC-MS/MS ) method to determine steady-state concentrations of most of post squalene cholesterol biosynthetic intermediates of the Bloch and K–R pathways in biological fluids . Here we combined these approaches to examine the flux of substrate though the Bloch and K–R pathways in cultured cells and in vivo in the tissues of mice .
As a first step towards characterizing the flux of precursor sterols through the cholesterol biosynthetic pathway , we used LC-MS/MS and deuterium oxide ( D2O ) labeling to analyze the turnover of a representative sterol , lanosterol , in immortalized human skin fibroblasts ( SV-589 cells ) . The isotopomer distributions of lanosterol before and 24 hr after addition of 5% D2O to the medium are shown in Figure 2A . In unlabeled cells , the isotopomer spectrum matched the predicted distribution based on a 1 . 1% natural abundance of 13C and a 0 . 015% natural abundance of deuterium ( Figure 2A , top , left panel ) . Lanosterol molecules containing exclusively 12C , 16O and 1H isotopes had an m/z peak of 409 Da ( M = 0 ) and comprised 72% of the total . Lanosterol molecules containing a single extra neutron ( i . e . , one atom of 13C or 2H ) had an m/z peak of 410 Da ( M = 1 ) comprised 24% of the total while those containing two extra neutrons ( M = 2 ) comprised 4% . 10 . 7554/eLife . 07999 . 004Figure 2 . Sterol biosynthesis in cultured cells . ( A ) Deuterated water ( D2O ) labeling of lanosterol in cultured cells . ( Top ) Isotopomer spectrum of lanosterol in SV-589 cells grown for 24 hr in the absence ( left ) or presence ( right ) of 5% D2O added to the medium . ( Bottom ) Turnover of lanosterol in SV-589 cells . Cells were grown in NCLPPS ( open circles ) or in NCLPPS plus 25-hydroxycholesterol ( 1 μg/ml ) ( closed circles ) to ∼60% confluence . After 16 hr , the medium was supplemented with 5% D2O and cells were harvested at 0 , 0 . 5 , 1 , 2 , 4 , 6 , 8 , 12 , and 24 hr ( last two points not shown , but used for modeling ) . Sterols were analyzed by LC-MS/MS and the fraction of lanosterol that was newly synthesized was determined using isotopomer analysis ( IA ) ( see Figure 2—figure supplement 1 ) and the results were fit to a first-order kinetic model ( solid lines ) as described in the ‘Materials and methods’ . The rates of synthesis of lanosterol in the presence or absence of sterols were calculated by multiplying the first-order rate constant by the concentration of lanosterol . Standard deviations are reported based on four independent replicates . ( B ) Biosynthetic rates of intermediary sterols in cultured cells . The rate of synthesis of intermediary sterols in the Bloch ( left ) and K–R ( right ) pathways was measured using D2O labeling . Cells were plated at a density of 500 , 000/60 mm dish and grown to ∼60% confluence . D2O was then added to the medium to a final concentration of 5% ( vol/vol ) . Cells were harvested at 0 , 0 . 5 , 1 , 2 , 4 , 6 , 8 , 12 , and 24 hr , and lipids were extracted using methanol-dichloromethane . Sterols were analyzed by LC-MS/MS and the fraction of each sterol that was newly synthesized was determined using IA and the results were fit to a first-order kinetic model as described in the ‘Materials and methods’ . Y1-BS1 cells , a mouse adrenal cell line ( Top ) were grown in 15%horseserum . SV-589 cells , an immortalized human skin fibroblast line ( bottom ) , were grown in either 10% FCS . Means and standard deviations are reported based on four independent replicate experiments in each cell line . ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 07999 . 00410 . 7554/eLife . 07999 . 005Figure 2—figure supplement 1 . Schematic representation of IA adapted from Kelleher and Masterson ( 1992 ) . The fractional abundance of each sterol peak in an unlabeled and labeled sample can be used to calculate the rate of newly synthesized sterols . In the unlabeled sample , the fraction abundance of M = 0 is 0 . 74 ( we will call this M0n ) ; for the labeled sample , it is 0 . 23 ( M0t ) . If we sample a sterol as the pool is turning over , the fractional abundance of M = 0 ( M0m ) will decrease from 0 . 74 to 0 . 23 . The corresponding values of M2m will increase from 0 . 04 to 0 . 29 . Using this data , the fraction of molecules that are newly synthesized ( g ) can be estimated using isotopic spectral analysis ( IA ) . Using IA , the value of M0m has two components: ( 1 ) M0n times the fraction of molecules not labeled ( 1 − g ) . ( 2 ) M0t times the fraction of molecules newly synthesized ( g ) . These two values can be summed to get the value of M0m . This same analysis can be done for any of the peaks . Then , by solving for g , the fraction of molecules newly synthesized can be determined based on M0m , M0n , and M0t . M0m is the measured value that is obtained from MS . M0n is determined from an unlabeled sample or from a natural abundance calculator . M0t can be determined from samples exposed to label long enough to ensure turnover of the entire pool , or using mass isotopomer distribution analysis ( MIDA ) ( Hellerstein and Neese , 1999 ) . The biosynthetic rate was calculated by fitting the relationship between g and time to a first-order kinetic model . The rate constant was multiplied by the concentration , yielding the biosynthetic rate , as demonstrated in Figure 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 07999 . 005 Cells grown in the presence of D2O incorporate deuterium into newly synthesized sterols , primarily via deuterated NADPH , causing a decrease in the abundance of M0 isotopomers and an increase in the proportion of heavier molecules . After cells were grown in 5% D2O long enough for the lanosterol pool to be replaced ( in this case , for 24 hr ) , only 25% of the lanosterol had not incorporated any deuterium atoms ( M = 0 ) . Over 75% of the lanosterol molecules had incorporated at least one heavy atom ( M1 + M2 + M3 ) ( Figure 2A , top , right panel ) . The largest fraction of lanosterol isotopomers ( 37% ) contained a single deuterium or 13C atom ( M = 1 ) . The shift in isotopomer distribution over time was used to infer the incorporation of deuterium , which was then used to measure the rate of lanosterol synthesis as described in the ‘Materials and methods’ and reviewed in Figure 2—figure supplement 1 . Next , we examined the effect of 25-hydroxycholesterol ( 25-OH Chol ) , a potent suppressor of cholesterol biosynthesis ( Kandutsch et al . , 1977 ) , on lanosterol turnover in cultured fibroblasts ( Figure 2A , bottom , left panel ) . Cells were grown for 16 hr in the presence or absence of 25-OH Chol ( 1 μg/ml ) prior to addition of 5% D20 to the medium . The rate of lanosterol biosynthesis was determined by sampling cells at the indicated time points . At each time point the fraction of newly synthesized lanosterol molecules ( termed ‘g’ ) was determined from the isotopomer spectrum . The relationship between time and g was fitted to a first-order kinetic model to determine the rate constant ( k ) , which was multiplied by the lanosterol concentration to determine the rate of synthesis ( ng/hr/μg protein ) . The addition of 25-OH Chol to the medium decreased the rate of lanosterol biosynthesis by 90% ( Figure 2A , bottom ) . To determine the relative utilization of the Bloch and K–R pathways in various cell types , we measured and compared the rates of deuterium incorporation from D2O into post-squalene cholesterol biosynthetic intermediates in cultured mouse adrenal cells ( Y1-BS1 cells ) ( Watt and Schimmer , 1981 ) and transformed human fibroblasts ( SV-589 cells ) ( Yamamoto et al . , 1984 ) ( Figure 2B ) . Deuterium was incorporated almost exclusively into Bloch pathway intermediates in Y1BS1 cells , in which the rates of incorporation were similar for lanosterol , ff-MAS , t-MAS , dehydrodesmosterol , and desmosterol ( Figure 2B , top left panel ) . Little turnover of K–R intermediates was detected in these cells ( Figure 2B , top right panel ) . In SV-589 fibroblasts ( Figure 2B , bottom panel ) , lanosterol was quantitatively converted to ff-MAS and t-MAS ( Bloch pathway ) with almost no detectable incorporation into the corresponding K–R intermediates ( dihydro-ff-MAS and dihydro-t-MAS ) . Incorporation into the downstream Bloch intermediates dehydrodesmosterol and desmosterol was minimal , but robust labeling of 7-dehydrocholesterol was observed , indicating a crossover from the Bloch to the K–R pathway between t-MAS and dehydrodesmosterol ( Figure 1 ) . The methylated biosynthetic intermediates between lanosterol and 7-dehydrocholesterol in the K–R pathway ( i . e . , dihydrolanosterol , dihydro-ff-MAS , and dihydro-t-MAS ) did not turnover at comparable rates to either lanosterol or 7-dehydrocholesterol , suggesting that the classical K–R pathway was not used to synthesize cholesterol in these cells . Instead , the cells used a hybrid pathway that we will refer to as the modified Kandutsch–Russell ( MK–R ) pathway . In this hybrid pathway , intermediates proceed down the Bloch pathway until demethylation of the sterol nucleus is complete , and then they undergo reduction of the double bond at C24 to enter the K–R pathway . The step at which sterol synthesis crosses over from the Bloch to the K–R pathway could not be pinpointed in this experiment since some intermediates ( shown in light gray in Figure 1 ) could not be measured due to either isobaric interference with cholesterol or because the levels were below the detection limits of the assay ( see ‘Material and methods’ ) . Our data suggest that in SV-589 cells , sterols are demethylated via the Bloch pathway and then undergo Δ24 reduction upstream of desmosterol . Therefore , in cultured fibroblasts the cross-over to the MK–R pathway occurs at either zymosterol , dehydrolathosterol , or dehydrodesmosterol . These experiments confirmed the conclusion of Kandutsch and Russell that significant differences exist between cells of different types in the pathways utilized for cholesterol synthesis ( Kandutsch and Russell , 1960a , 1960b ) ; however , they did not provide evidence that the K–R pathway as originally conceived ( Figure 1 , red arrows ) is utilized , at least in these two cell lines . To interrogate the factors that determine the relative utilization of the Bloch and MK–R pathway , we examined two cell culture systems in which rates of cholesterol biosynthesis were altered . First , we compared sterol flux in WT Chinese Hamster Ovary ( CHO-7 cells ) grown in cholesterol-containing medium ( 10% FCS ) or in cholesterol-depleted serum ( NCLPPS ) to upregulate cholesterol synthesis . Second , cholesterol synthesis was examined in a mutant line of CHO-7 cells ( SRD13A cells ) that lack SREBP cleavage activating protein ( SCAP ) , a protein required to activate the transcription of cholesterol biosynthesis genes ( Rawson et al . , 1999 ) . SRD13A cells have no active SREBPs and thus exhibit low rates of cholesterol synthesis ( Rawson et al . , 1999 ) . In WT CHO-7 cells , the Bloch pathway accounted for ∼90% of cholesterol synthesis ( Figure 3A ) . Deletion of SCAP reduced lanosterol biosynthesis by 80% , with similar reductions in the biosynthesis of other unsaturated side chain Bloch intermediates ( Figure 3A ) . Despite the massive reduction in net sterol synthesis in the SRD13A cells , the biosynthetic rate of 7-dehydrocholesterol was similar to that observed in the parental cell line ( Figure 3A , inset ) . This finding suggests the flux through the MK–R pathway is constitutive in these cells . 10 . 7554/eLife . 07999 . 006Figure 3 . Modification of cholesterol biosynthesis rate . ( A ) Biosynthetic rate of intermediary sterols of the Bloch ( left; black ) and K–R ( right; red ) pathways of CHO-7 ( wt , closed bars ) and SRD13A cells ( SCAP−/− , open bars ) . Cells were grown to ∼60% confluence in 10% NCLPPS before measuring sterol biosynthesis rates using D2O . The inset of the right panel rescales the dehydrocholesterol values shown below . ( B ) Sterol biosynthesis rate in HuH7 cells grown in FCS ( open bars ) or the cholesterol depleted medium NCLPPS ( solid bars ) . Cells were grown in their respective medium to ∼60% confluence before D2O labeled to measure biosynthetic rates . In panels A and B , means and standard deviations are reported based on four independent replicates of each cell line or condition . ( C ) Expression of genes encoding enzymes catalyzing the conversion of squalene to cholesterol ( shown in the Bloch sequence from left to right ) in HuH7 cells grown in either FCS or NCLPPS . Expression is normalized to 36B4 . Means and standard deviations are based on six replicates from two independent experiments . *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 07999 . 006 In cultured human hepatoma cells ( HuH-7 cells ) grown in cholesterol-replete serum ( FCS ) ( Figure 3B ) , the pattern of sterol turnover resembled that seen in SV-589 cells ( Figure 2B ) ; most of the sterol flux traversed the MK–R biosynthetic pathway ( Figure 3B , open bars ) . When cells were grown for 16 hr in cholesterol-depleted serum ( NCLPPS ) to stimulate cholesterol biosynthesis , sterol turnover doubled . Essentially all of the increase occurred via the Bloch pathway: flux through the MK–R pathway did not change ( Figure 3B , open bars ) . Therefore , cholesterol depletion shifted the relative utilization of the two pathways such that approximately equal amounts of lanosterol were metabolized via Bloch and MK–R intermediates . Cholesterol depletion also resulted in the incorporation of a small amount of deuterium into dihydrolanosterol , presumably reflecting conversion from lanosterol , but none of the other K–R intermediates prior to 7-dehydrocholesterol were labeled in these cells . To determine if the increase in Bloch pathway utilization associated with upregulation of cholesterol synthesis was coordinated at the transcriptional level , we compared levels of mRNAs encoding cholesterol biosynthetic enzymes in the HuH7 cells grown in FCS or NCLPPS ( Figure 3C ) . Despite a significant increase in Bloch pathway utilization in the cells that were grown in NCLPPS , no corresponding changes were observed in expression of the post-squalene cholesterol biosynthetic enzymes , including DHCR24 . Thus , the increase in flux through the Bloch pathway cannot be attributed simply to changes in the expression of SREBP-2 and the coordinated transcriptional upregulation of genes encoding enzymes in the pathway . Next , we examined the relationship between the level of expression of DHCR24 , which desaturates the side-chain of the sterol , and the relative use of the Bloch and MK–R pathways . To determine if changing DHCR24 expression altered the relative use of the Bloch and MK–R pathways , we expressed recombinant mouse DHCR24 in human embryonal kidney cells ( HEK-293 cells ) . In HEK-293 cells transfected with vector alone , 78% of cholesterol biosynthesis proceeded through the Bloch pathway ( Figure 4A , open bars ) . Overexpression of DHCR24 resulted in lanosterol being quantitatively converted into the Bloch pathway intermediate zymosterol , but did not increase incorporation of label into the early methyl sterols of the classic K–R pathway . Only about one-third of the labeled zymosterol was converted to dehydrodesmosterol and desmosterol ( Figure 4A , solid bars ) ; however , flux through 7-dehydrocholesterol was markedly increased in these cells , indicating cross-over of post-zymosterol intermediates to the MK–R pathway . Taken together , these data indicate that overexpression of DHCR24 can promote flux through the MK–R pathway , thus diverting flux through the terminal half of the Bloch pathway . No evidence was seen of utilization of the classical K–R pathway under these conditions . 10 . 7554/eLife . 07999 . 007Figure 4 . DHCR24 expression and sterol biosynthesis . ( A ) Sterol turnover in cells over-expressing DHCR24 . HEK-293 cells were cultured in DMEM +10% FCS for 2 days . On day 3 the cells were transfected with empty vector ( open bars ) or a plasmid encoding DHCR24 ( closed bars ) . After 40 hr , the medium was supplemented with 5% D2O and the incorporation of label into sterols was measured using LC-MS/MS . The rates of synthesis for intermediates in the Bloch ( left ) and K–R ( right ) pathways were determined using IA and fitted to a first order kinetic model . DHCR24 was measured by immunoblot analysis at the 0 hr time point ( blot in right panel ) . Means and standard deviations shown are based on three independent experiments . ( B ) HEK-293 cells were transfected with DHCR24 ( open ) or an empty vector ( closed ) . After 40 hr , 5 µg/ml of d6-lanosterol ( top ) or d5-zymosterol ( bottom ) conjugated to MCD was added to the medium . After 5 hr , cells were harvested and labeled sterols were measured by LC-MS/MS . Levels of labeled cholesterol are shown on the far right ( blue ) . The inset of the bottom right panel highlights level of d5-labeled zymostenol , lathosterol and 7-dehydrocholesterol . The signal intensities of the labeled sterols were normalized to total protein and to an internal standard . Means and standard deviations are based on 3 replicates . Similar results were observed in an independent experiment . *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 07999 . 007 As noted above , the flux of some sterol intermediates in the pathway could not be traced using deuterated water , which limited our ability to determine the point of cross-over from the Bloch to the MK–R pathway ( Figure 1 ) . To circumvent this problem , we incubated HEK-293 cells with synthetic isotopomers of lanosterol and zymosterol ( d6-lanosterol or d5-zymosterol ) that have isotopic distributions not found in natural sterols . By monitoring the conversion of these synthetic isotopomers into other sterols we were able to resolve all of the stable biosynthetic intermediate sterols in their labeled forms . In this experiment we were also able to examine cholesterol biosynthesis from desmosterol and from 7-dehydrocholesterol ( Figure 4B , blue bars ) . The design of these experiments differed from those shown previously . Here the labeled sterol was added to the medium and the distribution of label among the sterols of the two pathways was determined at a single time point ( 5 hr ) . In cells expressing vector alone , the d6 label appeared in unsaturated side chain sterols of the Bloch pathway , with little to no label appearing in the MK–R intermediates ( Figure 4B , open bars in upper panel ) . In cells expressing DHCR24 , the d6 label was detected in lanosterol , ff-MAS , t-MAS , and zymosterol , but not in the downstream Bloch intermediates ( Figure 4B , closed bars in upper panel ) . Instead , the label appeared in the demethylated saturated side chain sterols along the MK–R pathway ( zymostenol , lathosterol , and 7-dehydrocholesterol ) . These data suggest that zymosterol is the crossover point between the Bloch and MK–R pathways in these cells under these conditions , and that the methyl sterols upstream of zymosterol are poor substrates for ( or do not have access to ) DHCR24 . Under both conditions , the majority of the labeled sterol remained as d6-lanosterol . Overexpression of DHCR24 did not increase the formation of d6-dihydrolanosterol or of d6-cholesterol ( Figure 4B , upper panel ) , suggesting the DHCR24 is not rate-limiting for the conversion of lanosterol to cholesterol . To confirm the crossover point between the Bloch and K–R pathway , HEK-293 cells were incubated with d5-zymosterol . In cells expressing the empty vector alone , d5 label appeared in zymosterol , dehydrolathosterol , dehydrodesmosterol , and desmosterol ( Figure 4B , open bars in lower panel ) . A small amount of label was measured in lathosterol and 7-dehydrocholesterol , but not in other K–R intermediates . In cells transfected with DHCR24 , only a small fraction of the label was detected in intermediates of either pathway , presumably because they are rapidly converted to cholesterol . We found no label incorporated into dehydolathosterol or desmosterol . This result may reflect quantitative conversion of zymosterol to zymostenol , or rapid conversion of desmosterol to cholesterol . Label was detected in the downstream MK–R pathway intermediates zymostenol , lathosterol , and 7-dehydrocholesterol ( Figure 4B , closed bars in lower panel ) . These findings are consistent with those using labeled lanosterol and support the hypothesis that zymosterol is the first substrate for DHCR24 in the MK–R pathway used by these cells . The amount of d5-cholesterol derived from d5-zymosterol was 4 times higher when HEK-293 cells were transfected with DHCR24 relative to an empty vector ( Figure 4B , right ) , suggesting that DHCR24 is rate-limiting for the conversion of zymosterol to cholesterol in these cells . No signal corresponding to D5- or D6- labeled sterols was seen in cells treated with vehicle alone . To examine the pathways of post-squalene cholesterol biosynthesis in vivo , we assessed rates of sterol synthesis in different mouse tissues . Mice were labeled to approximately 5% D2O by administering an initial bolus of D2O ( 500 μl ) via intraperitoneal injection and by enriching their drinking water to 6% ( vol/vol ) D2O . Rates of sterol synthesis were calculated as described in the ‘Materials and methods’ and Figure 2—figure supplement 1 . The k-values and sterol concentrations are provided in Supplementary file 1 . In testes , deuterium label was only detected in Bloch pathway intermediates ( Figure 5 ) . The rates of synthesis of the methyl sterols ( lanosterol , ff-MAS and t-MAS ) in the Bloch pathway were ∼3 times higher than those of the demethylated intermediates ( zymosterol , dehydrolathosterol , dehydrodesmosterol and desmosterol ) in the pathway . Thus , a large fraction of the t-MAS synthesized was diverted from the cholesterol biosynthetic pathway to produce other , as yet unidentified sterols ( Byskov et al . , 1995 ) . The drop in flux along the Bloch pathway between t-MAS and zymosterol was not observed in any other tissue . 10 . 7554/eLife . 07999 . 008Figure 5 . Sterol biosynthesis in mice . ( A ) Sterol biosynthesis in selected mouse tissues . Mice were enriched to ∼5% D2O by intraperitoneal injection of 500 μl and supplementing the drinking water to 6% D2O . Tissues were collected at 0 , 1 , 2 , 3 , 4 , 6 , 8 , 12 , 18 , 24 , 48 , 72 , 120 , and 168 hr after injection ( 3 animals per time point ) , and analyzed by LC-MS/MS as described in the ‘Materials and methods’ . Rates of synthesis of the Bloch ( left ) and K–R ( right ) intermediates were calculated using IA and first-order kinetics . The experiment was repeated in an independent set of animals with similar results . ( B ) Percent utilization of the Bloch ( black ) and K–R ( red ) pathways for post-squalene cholesterol biosynthesis . The percentage of cholesterol derived from the Bloch pathway was determined by dividing the desmosterol synthesis rate by the sum of the desmosterol and 7-dehydrocholesterol synthesis rates . Similar results were obtained when the desmosterol synthesis rate was divided by the lanosterol synthesis rate , except in the testes ( not shown ) . ( C ) The fractional utilization of the Bloch pathway using the data shown in Panel B compared to the natural log of lanosterol synthesis rate measured in each tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 07999 . 008 In adrenal glands , cholesterol was also synthesized exclusively via the Bloch pathway ( Figure 5 ) with little or no deuterium being incorporated into any of the saturated side chain intermediates of the K–R pathway . The rates of lanosterol and desmosterol synthesis were similar . Thus , lanosterol was not converted to any products other than desmosterol , and presumably cholesterol , at an appreciable rate . In contrast to adrenal glands , the MK–R pathway predominated in skin . The synthesis of desmosterol was much less than either lanosterol or 7-dehydrocholesterol ( Figure 5 ) . There was measureable synthesis of dihydro-t-MAS , indicating that saturation of the side chain double bond can occur before total demethylation of t-MAS . In liver , the rates of synthesis of desmosterol and 7-dehydrocholesterol were similar . Thus , approximately half of the sterol flux followed the Bloch pathway , whereas the other half used the MK–R pathway . In contrast to all other tissues examined , dihydrolanosterol was synthesized in liver at ∼5–10% the rate of lanosterol synthesis ( 2 ng/μg tissue/day ) . Essentially no synthesis of dihydro-ff-MAS was detected . This result suggests that a fraction of hepatic lanosterol was converted to dihydrolanosterol but did not proceed down the K–R pathway . Consistent with this finding , cultured human hepatocytes ( HuH7 cells ) also synthesized dihydolanosterol that was not further metabolized into dihydro-ff-MAS ( Figure 3B ) . Thus ∼5–10% of the sterol synthesized in liver or in cultured hepatocytes is converted to dihydrolanosterol and exits the cholesterol biosynthetic pathway . We performed similar in vivo analyses on preputial gland and brain ( Figure 5 ) . The K–R pathway was first described in preputial gland , a modified sebaceous gland that is present in rodents but not in humans ( Kandutsch and Russell , 1960a , 1960b ) . We found that the turnover of 7-dehydrocholesterol in preputial glands was similar to that of lanosterol , which is consistent with the findings of Kandutsch and Russell ( 2 ) . Nevertheless , our data are not consistent with the K–R pathway as it was originally described . The flux from lanosterol proceeded via the Bloch pathway intermediates ff-MAS and t-MAS without any detectable incorporation into the corresponding saturated side chain sterols ( dihydrolanosterol and dihydro-ff-MAS ) . Therefore , even in preputial glands , lanosterol demethylation commences before side-chain saturation . A similar pattern of sterol flux through the MK–R pathway was measured in the skin and brain . The skin makes Vitamin D , but since we did not measure incorporation of label into cholesterol we do not know how much of the 7-dehydrocholesterol that is made in the dermis is converted to Vitamin D vs . cholesterol . In the brain , the absolute rate of cholesterol synthesis was low , corresponding to less than 2% of that observed in liver ( Figure 5 , bottom panel; note differences in scale ) . Like the skin and preputial gland , the MK–R pathway for cholesterol synthesis predominated in the central nervous system , even though the brain was reported previously to express very low levels of DHCR24 ( Nes , 2011 ) . Thus , none of the tissues examined in this study utilized the classic K–R pathway . Instead , sterols were demethylated , at least partially , before the side chain was saturated . To compare the relative pathway utilization across tissues , the fractional utilization of the Bloch pathway was estimated for each tissue by dividing the rate of synthesis of desmosterol by the sum of the rates of synthesis of desmosterol and 7-dehydrocholesterol . The values obtained ranged from 0 . 97 ( testes ) to 0 . 08 ( preputial gland ) ( Figure 5B ) . In general , tissues with a higher lanosterol synthesis rate had a higher fractional utilization of the Bloch pathway ( Figure 5C ) , with the exception of the skin and preputial gland , which had high sterol synthesis rates via the MK–R pathway . It has been suggested that relative usage of the Bloch and K–R pathway is determined by level of expression of DHCR24 ( Nes , 2011 ) . To determine if the tissue-specific differences we observed in usage of the MK–R pathway were caused by differences in DHCR24 expression , we measured levels of DHCR24 mRNA and protein in the different tissues ( Figure 6 ) . The tissue with the highest level of both DHCR24 mRNA and protein was the preputial gland , which also has highest fractional utilization of the MK–R pathway ( Figure 5 ) . Surprisingly , the tissue with the second highest level of DHCR24 mRNA and protein was the liver , which predominantly utilizes the Bloch pathway ( Figure 5A ) . Skin , which has the second highest fractional usage of the MK–R pathway for cholesterol biosynthesis , had mRNA and protein levels of DHCR24 that were lower than those found in the liver . In the brain the level of DHCR24 mRNA was higher than that detected in the skin and yet no protein was detected in this tissue . Thus , the level of DHCR24 mRNA in tissues did not always correlate with the fractional utilization of the MK–R pathway . 10 . 7554/eLife . 07999 . 009Figure 6 . Levels of DHCR24 mRNA and protein in mouse tissues . ( A ) The levels of DHCR24 mRNA relative to 36B4 in mouse tissues are arranged from highest to lowest fractional utilization of the Bloch pathway ( left to right ) . ( B ) Immunoblot analysis of DHCR24 . Tissues were collected from a single male mouse and a total of 5 μg of protein from each tissue was analyzed by immunoblotting using a polyclonal rabbit anti-mouse antibody as described in the ‘Material and methods’ . The experiment was performed 3 times with similar results . DOI: http://dx . doi . org/10 . 7554/eLife . 07999 . 009 In general , DHCR24 was expressed at very low levels in tissues that predominantly use the Bloch pathway ( testes , spleen , adrenal and adipose tissue ) and in those with low cholesterol biosynthetic rates ( heart and muscle ) ( Figure 5C ) . In mammals , liver is the major source of plasma cholesterol ( Dietschy and Turley , 2002 ) . To assess the feasibility of inferring rates of hepatic sterol biosynthesis from plasma samples , we simultaneously measured the turnover of sterols in the plasma and liver of mice ( Figure 7 ) . We found that the fractions of sterols with incorporated label in plasma and liver were remarkably similar for lanosterol , desmosterol , and 7-dehydrocholesterol ( Figure 4C ) . The only sterol for which the labeling pattern was substantially different between the two compartments was dihydrolanosterol , which readily incorporated label in liver but remained unlabeled in plasma . We were unable to determine whether dihydrolanosterol was being converted to another sterol , was quantitatively excreted into bile and thus did not enter the plasma compartment , or was obscured in the MS assay by an analyte that is present in plasma , but not liver . 10 . 7554/eLife . 07999 . 010Figure 7 . Sterol turnover in liver and plasma of mice . The fractional turnover of lanosterol , desmosterol , 7-dehydrocholesterol , and dihydrolanosterol was determined based on the incorporation of deuterium into sterols measured in liver ( closed boxes ) and plasma ( open circles ) from the same mice . DOI: http://dx . doi . org/10 . 7554/eLife . 07999 . 010
The Bloch and K–R pathways were described more than 50 years ago and are widely accepted as the two major pathways for cholesterol synthesis ( Kandutsch and Russell , 1960b; Bloch , 1965 ) . The present study represents the first systematic analysis of flux through these pathways in cell culture and in vivo . The major finding of the study is that the architecture of the cholesterol biosynthetic pathway shows striking differences among tissues . The testes and adrenal gland utilize the canonical Bloch pathway ( Bloch , 1965 ) almost exclusively ( Figure 8 , black arrows ) . None of the tissues examined in this study had a pattern of sterol turnover consistent with the reaction sequence proposed by Kandutsch and Russell ( 1960b ) ( Figure 1 , red arrows ) . A hybrid pathway that we have named the MK–R pathway exists in skin , preputial glands and brain , ( Figure 8 , red arrows ) . In these tissues , sterols undergo demethylation prior to side chain saturation ( Figure 5 ) . Whereas we were not able to localize the specific DHCR24 substrate in the D2O labeling studies , experiments using deuterium-labeled lanosterol and zymosterol in HEK-293 cells confirmed that zymosterol is the first sterol to undergo appreciable side-chain reduction , at least in these cells . Remarkably , immortalized cells in culture retained the specific pathways used by their tissues of origin . Regulation of cholesterol biosynthesis was achieved almost exclusively by changes in flux through the Bloch pathway . Our study reveals a strikingly varied pattern of sterol synthesis from lanosterol that allows for the generation of multiple , tissue-specific end-products in addition to cholesterol . 10 . 7554/eLife . 07999 . 011Figure 8 . A modified Kandutsch–Russell ( MK–R ) model of post squalene cholesterol biosynthesis . In this model , cholesterol biosynthesis proceeds from lanosterol to t-MAS . Black arrows denote the Bloch pathway . Red arrows denote the MK–R pathway . 24 , 25-double bond desaturation can occur at any step between lanosterol and desmosterol in this pathway , but in most tissues desaturation does not occur until after demethylation is complete ( after T-MAS ) . The pathway shown with blue arrows was only detected in the liver and did not contribute to cholesterol biosynthesis . Additional pathways involving sterol intermediates that do not result in the biosynthesis of cholesterol are also shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07999 . 011 The MK–R pathway is consistent with a reaction sequence proposed by Bae and Paik ( 1997 ) , who reported that zymosterol was preferred over lanosterol as a substrate for C-24-reduction in rat liver microsomes . Those authors hypothesized that the physiological pathway for conversion of lanosterol to cholesterol did not correspond to either the Bloch or the KR pathways , but rather to a reaction sequence in which C24-reduction occurred after demethylation but before the final Δ5-dehydrogenation and Δ7-reduction of the sterol nucleus . This proposal was not assessed in vivo and did not gain wide acceptance . The present data provide in vivo evidence for 24-reduction of zymosterol , but the predominance of the Bloch pathway in most tissues with high rates of cholesterol synthesis indicates that desmosterol is a major physiological substrate for the enzyme . Reduced cellular cholesterol content selectively upregulated the Bloch pathway in cultured CHO7 and HuH7 cells ( Figure 3 ) . Genetic ablation of SCAP , which is required for SREBP activation ( Rawson et al . , 1999 ) , dramatically reduced flux through the Bloch pathway , but had virtually no effect on the MK–R pathway . Similarly , depletion of cellular cholesterol by incubation in NCLPPS resulted in a major increase in flux through the Bloch pathway with no detectable change in flux through the MK–R pathway . These findings suggest that the MK–R pathway is constitutively active and that the Bloch pathway is used preferentially for regulated cholesterol biosynthesis in response to fluctuations in cholesterol availability and demand . Upregulation of the Bloch pathway also increases the synthesis of regulatory sterols that limit cholesterol accumulation in the cells: desmosterol activates LXR , which promotes cholesterol efflux from cells , and acts as a feedback inhibitor of cholesterol synthesis by inactivating SREBP ( Yang et al . , 2006; Spann et al . , 2012 ) . The testes and adrenal glands synthesized cholesterol predominantly via the Bloch pathway , but the fate of sterols moving through the pathway in the two organs differed markedly . In the adrenal glands , lanosterol was quantitatively converted to desmosterol , and then presumably to cholesterol , which provides the substrate for adrenal steroidogenesis . In contrast , only one-third of the lanosterol synthesized in the testes was converted to desmosterol , and virtually none entered the MK–R pathway . In this tissue , lanosterol was quantitatively converted to ff-MAS and then to t-MAS , but more than two-thirds of the t-MAS was diverted from the pathway before zymosterol ( Figure 5 ) . The metabolic fate of t-MAS in the testes is not known . t-MAS and its immediate precursor ff-MAS have been implicated as meiosis-activating sterols in the formation of male and female germ cells ( Byskov et al . , 1995 ) ; however , the role of these precursor sterols remains controversial . Germ-cell specific ablation of Cyp51a1 , the enzyme that converts lanosterol to ff-MAS ( Figure 1 ) , markedly reduced testicular t-MAS concentration but had no effect on reproductive function in male mice ( Keber et al . , 2013 ) . The reduction in flux from t-MAS to zymosterol that we observed in mouse testes ( Figure 5 ) may be due to t-MAS being converted to another sterol ( or steroid hormone ) , or to the secretion of t-MAS from the testes . The liver was the only tissue in which we observed appreciable dihydrolanosterol synthesis ( Figure 5 ) . Approximately 5–10% of the lanosterol synthesized in the liver was converted to dihydrolanosterol , but not to downstream intermediates in the K–R pathway . In contrast to other sterol intermediates synthesized in the liver , labeled dihydrolanosterol did not appear in the plasma ( Figure 7 ) . It is possible that dihydrolanosterol is rapidly excreted from the liver via the bile . Although we did not include biliary sterols in our flux studies , we have found that the concentration of dihydrolanosterol in bile is not enriched relative to the liver ( data not shown ) . This finding is not compatible with selective biliary excretion of dihydrolanosterol . Taken together , our data indicate that at least some of the dihydrolanosterol that is formed in the liver is converted to another sterol , the identity of which remains unknown . The function of dihydrolanosterol in the liver is also not known . Previous studies have demonstrated that dihydrolanosterol can promote degradation of HMG-CoA reductase ( Lange et al . , 2008 ) . This effect is specific to dihydrolanosterol: neither cholesterol nor the other biosynthetic intermediates tested , including lanosterol and lathosterol , showed this activity ( Song et al . , 2005; Lange et al . , 2008 ) . Therefore , one function of the dihydrolanosterol ( or a metabolic derivative thereof ) synthesized in the liver may be to provide rapid feedback inhibition of cholesterol biosynthesis through post-transcriptional regulation of HMG-CoA reductase . If this model is correct , then our data suggest that this regulation is likely only significant in liver . Three tissues predominantly used the MK–R pathway: brain , skin and preputial gland . Our finding that skin and preputial gland produced high levels of saturated side chain sterols is compatible with the findings of Kandutsch and Russell ( 1960b ) . The utilization of the MK–R pathway in the skin ensures a constant supply of 7-dehydrocholesterol for vitamin D synthesis ( DeLuca , 2008 ) . Why the MK–R is the predominant pathway in preputial gland is less clear . The major function of this gland is thought to be the synthesis of pheromones . The chemical nature of these pheromones has not been fully defined but squalene and cholesterol were the most abundant lipids identified in a GC/MS analysis of preputial glands from male rats ( Zhang et al . , 2008 ) . Thus , cholesterol , or one or more of its biosynthetic precursors may be used for pheromone synthesis . In the present study , a loss of sterol between lanosterol and 7-dehydrocholesterol was not observed , but we cannot exclude the possibility that some 7-dehydrocholesterol is diverted from cholesterol synthesis to pheromone synthesis . In contrast to skin and preputial gland , the brain has very low DHCR24 activity and was predicted to predominantly use the Bloch pathway ( Nes , 2011 ) . Although the brain is among the most cholesterol-rich organs , turnover , and hence synthesis of brain cholesterol is low ( Figure 5 ) ( Lund et al . , 2003 ) . Our data indicate other tissues with low rates of cholesterol synthesis , such as heart and skeletal muscle also predominantly used the MK–R pathway ( Figure 5 ) . These tissues are largely post-mitotic and do not synthesize steroid hormones or vitamin D . If the MK–R pathway is constitutive in vivo , as it appears to be in cultured cells , then predominant use of this pathway may ensure a constant rate of cholesterol synthesis in tissues that experience little variation in their cholesterol requirements . The relative utilization of the Bloch and MK–R pathways in cultured cells was strikingly similar to that observed in vivo in the tissue of origin . These findings suggest that the factors governing pathway selection are epigenetically fixed . Relative pathway utilization is presumably a function of the relative activities of the two enzymes at the branch point , which is tissue dependent . The relative function of the two enzyme could be determined by myriad factors including those that act directly on enzyme activity ( e . g . , transcription , post-translational modification ) , the expression of auxiliary proteins ( co-factors , inhibitors , proteins that present or sequester substrate ) or factors that affect the relative proximity of the enzymes to their substrates . Since DHCR24 activity is a major determinant of relative pathway utilization , cell-type specific control of pathway utilization may involve epigenetic control of DHCR24 expression . The pattern of DHCR24 expression in tissues was an imperfect predictor of relative pathway utilization . For example , the brain used the MK–R pathway despite low DHCR24 expression and activity ( Tint et al . , 2006 ) while liver and kidney predominantly used the Bloch pathway , despite relatively high expression of DHCR24 ( Figure 6 ) . DHCR24 is phosphorylated at multiple sites and may be regulated at the post-translational level ( Luu et al . , 2014 ) . The activity and substrate specificity of the enzyme may also be influenced by its intracellular location . Brown and his colleagues reported that DHCR7 and DHCR24 co-immunoprecipitate in CHO-7 cells ( Luu et al . , 2015 ) . Those authors proposed that the formation of a complex between the two proteins favors the efficient conversion of 7-dehydrodesmosterol to desmosterol , and then to cholesterol . This hypothesis is consistent with our finding that the Bloch pathway is predominantly used in steroidogenic cells . Further studies are required to elucidate the mechanistic basis for the wide differences in DHCR24 expression among tissues , and to identify other factors that govern the route of flux through the cholesterol biosynthetic pathways . The patterns of deuterium enrichment of biosynthetic precursor sterols in plasma were similar to those seen in the liver , but distinct from those observed in extrahepatic tissues . This finding indicates that the cholesterol pool in extrahepatic tissues is relatively isolated from the cholesterol in circulating lipoproteins . This hypothesis is supported by data from Dietschy and colleagues , who reported that 80% of cholesterol biosynthesis in mice takes place in extra-hepatic tissues , while 80% of plasma low density lipoprotein cholesterol is taken up by the liver ( Osono et al . , 1995 ) . Thus , most extrahepatic tissues obtain cholesterol primarily from de novo synthesis , with little contribution from circulating lipoproteins . This conclusion suggests that the effects of dietary and pharmacological interventions on cholesterol biosynthesis in the liver can be inferred from measurements of isotopic enrichment in the plasma . If extrahepatic tissues contribute little to the plasma sterol pool as shown in this study , how are sterols synthesized in peripheral tissues transported to the liver ( or gut ) for excretion ? The route by which cholesterol is transported from the peripheral tissues to the liver , a process termed reverse cholesterol transport ( Glomset , 1968 ) , remains to be fully defined ( Hellerstein and Turner , 2014 ) One possibility is that peripheral tissues release cholesterol , but not the biosynthetic precursor sterols , into the circulation . In this case we would fail to capture flux of peripheral tissue cholesterol since we did not measure cholesterol isotopomers in these experiments . An alternative possibility is that cholesterol from the periphery is transported in the circulation by cells rather than lipoproteins or other plasma components . Careful quantification of cholesterol fluxes in vivo will be required to elucidate the relative contributions of cellular and plasma elements to the centripetal transport of peripheral tissue cholesterol . The approach used in this study could potentially be confounded by the presence of two pools of an intermediate in the pathway that turn over at different rates . This problem would be particularly acute if a small pool of intermediate turns over rapidly while a second , larger pool sequestered from the biosynthesis pathway turned over slowly . Under these conditions the larger pool may significantly reduce the labeling of the total intermediate isolated from the cells resulting in falsely low estimates of flux . The observation that multiple independent intermediates showed essentially identical kinetics argues against this possibility , but we cannot formally exclude it . Metabolic pathways have traditionally been defined in two stages . First , the sequence of reactions that comprise the pathway is determined biochemically . These determinations are almost invariably qualitative in nature . Second , flux through the pathway is quantified using radioisotopes to trace the turnover of a single metabolite , typically the end product of the pathway . By combining LC-MS/MS methods with stable isotope tracing , flux can be monitored through multiple intermediates in a biosynthetic pathway simultaneously . As illustrated in the present study , this approach can reveal bifurcations in established pathways that imply novel biochemistry and physiology for molecules previously viewed simply as biosynthetic intermediates .
Immortalized cells from human fibroblasts ( SV-589 cells ) ( Yamamoto et al . , 1984 ) , human hepatoma ( HuH-7 cells ) ( Nakabayashi et al . , 1984 ) , mouse adrenocortical tumor ( Y1-BS1 cells ) ( Watt and Schimmer , 1981 ) , human embryonic kidney ( HEK-293 cells ) ( Graham et al . , 1977 ) , Chinese hamster ovaries ( CHO-7 cells ) , and SREBP cleavage activation protein deficient CHO-7 ( SRD13A; SCAP−/− ) cells were used in this study ( Rawson et al . , 1999 ) . Cells were plated at a density of 500 , 000 cells per 60 mm dish and maintained in monolayer culture at 37°C in 5% CO2 . SV-589 , HEK-293 , CHO-7 , SRD13A and HuH-7 cells were grown in Dulbecco's modified Eagle's medium ( DMEM ) supplemented with either 10% ( vol/vol ) fetal calf serum ( FCS ) or newborn calf lipoprotein poor serum ( NCLPPS ) . Y1-BS1 cells were grown in a 1:1 mixture of Ham's F-12 medium and DMEM supplemented with 15% horse serum ( HS ) . All media contained 100 units/ml penicillin and 100 μg/ml streptomycin sulfate . In experiments using NCLPPS and/or 25-OHC , the medium was changed 16 hr before beginning D2O labeling . For measurements of sterol synthesis , cells were plated and grown to ∼60% confluence either 2 days ( SV-589 , CHO-7 , SRD13A , and HEK-293 ) , 3 days ( HuH7 ) , or 5 days ( Y1BS1 ) . Then the medium was exchanged with medium containing 5% D2O . After 0 , 0 . 5 , 1 , 2 , 4 , 6 , 8 , 12 , and 24 hr , cells ( triplicate dishes at each time point ) were washed and then harvested in 3 ml Dulbecco's phosphate buffered saline ( PBS ) with 0 . 5% Triton X-100 . A 300 μl aliquot of each cell lysate was reserved for measurement of protein concentration by bicinchoninic acid assay ( BCA ) analysis ( Pierce BCA Protein Assay Kit ) . The remaining cell lysate was mixed with 3 ml methanol ( MeOH ) containing 20 ng of d6-sitosterol as an internal standard . The samples were sonicated for 5 min and stored at room temperature until they were prepared for LC-MS/MS analysis ( see below ) . Male C57BL/6J mice were obtained from the Jackson Laboratory ( Bar Harbor , ME ) , housed ( 4 per cage ) in a controlled environment ( 12-hr light/12-hr dark daily cycle , 23 ± 1°C , 50–70% humidity ) and fed ad libitium with standard chow ( Harlan; Teklad , 2016 ) for 4 weeks prior to experimentation . 3 days prior to experimentation the mice were entrained to a synchronized feeding cycle ( 12 hr fasting , 12 hr refeeding ) by removing food at the beginning of the light cycle and returning food at the onset of the dark cycle . All research protocols involving mice were reviewed and approved by the Institutional Animal Care and Use Committee at University of Texas Southwestern Medical Center . Sterol synthesis rates in mice were determined by measuring deuterium incorporation from D2O over time . At 9 AM ( 3 hr into the light cycle ) each mouse was injected intraperitoneally with 500 μl of D2O with 150 mM NaCl . After injection , the drinking water was supplemented with 6% D2O . Mice ( 3 per time point ) were sacrificed at 1 , 2 , 3 , 4 , 6 , 8 , 12 , 18 , 24 , 48 , 72 , 120 , and 168 hr after injection . Six mice that did not receive D2O were also sacrificed to measure the concentrations of sterols in their tissues . Immediately after sacrifice the adrenals , blood , brain , brown adipose tissue ( BAT ) from the nape of the neck , heart , kidneys , liver , preputial glands , skin , spleen , posterior hind limb muscle , testes , and epididymal adipose tissue ( WAT ) were removed . Plasma was isolated from blood obtained in an EDTA-coated tube after centrifugation . Fur was removed from the skin by applying Veet , thoroughly washing in PBS to remove residue , and then placed at −80°C . All other tissues were snap frozen in liquid N2 and stored at −80°C . Larger tissues were prepared for lipid extraction by weighing 50–100 mg pieces and immediately adding 5 ml of ice cold MeOH and 200 ng of d6-sitosterol . Tissues were thoroughly homogenized using an IKA pole rotor and bath sonication for 10 min at RT . The sample was then vortexed and centrifuged at 3000×g for 10 min . The supernatant was decanted and the pellet was resuspended in 5 ml MeOH and recentrifuged . The supernatant was pooled with the supernatant from the initial centrifugation . 1 ml of supernatant was removed and diluted with 2 ml of MeOH and 3 ml of PBS . Due to their small size , the adrenal and preputial glands were prepared by weighing the entire tissue , and then adding 3 ml MeOH and 20 ng of d6-sitosterol . The tissues were then homogenized and 3 ml PBS was added . Plasma was prepared by adding 100 μl to 6 ml of 1:1 MeOH/PBS while sonicating and adding 20 ng d6-sitosterol . From this point forward , cell culture , tissue , and plasma samples were handled identically . Lipids were extracted from the samples by a modified Bligh-Dyer extraction ( Bligh and Dyer , 1959 ) . Samples were saponified by adding 300 μl 45% ( wt/vol ) KOH and incubation for 2 hr at 60°C . After saponification , the samples were allow to cool to RT before 3 ml of dichloromethane ( DCM ) was added , inducing a 2-phase separation . The samples were vortexed and centrifuged at RT . The bottom phase ( containing primarily DCM ) was removed . An additional aliquot of DCM ( 4 ml ) was added to the top phase , vortexed , centrifuged , and the bottom phased was pooled with the previous bottom phase . The samples were evaporated under a light stream of N2 , resuspended in 300 μl of 9:1 MeOH/H2O , and transferred to vials for LC-MS/MS analysis . Sterols were analyzed as described by McDonald et al . ( 2012 ) . Lipid extracts were injected into a Shimadzu LC20A HPLC ( Kyoto , Japan ) with an Agilent Poroshell 120 EC-C18 column ( 2 . 1 × 150 mm , 2 . 7 micron beads , Santa Clara , CA ) and eluted using a solvent gradient that transitioned linearly from 93% MeOH/7% H2O to 100% MeOH in 7 min . The column was washed for 5 min in 100% MeOH and then returned to the initial solvent . Sterols were detected using an ABSciex ( Framingham , MA ) 4000 Qtrap MS/MS in positive mode with atmospheric pressure chemical ionization at a temperature of 350°C . The MS/MS detected mass to charge ratios ( m/z ) of 365–370 , 393–400 , 404 , and 409–414 , which spans the ion m/z plus 3 mass units for each sterol , along with the internal standard ( see Supplementary file 2 for details ) . For experiments that involve pre-labeled sterols , the mass ranges were extended to measure the M+5 and M+6 isotopomers . Standards were commercially available for all but four of the sterols in the cholesterol biosynthetic pathway . These standards were used for quantitation of sterol concentrations relative to d6-sitosterol , which was added as an internal standard . The four sterols that were not commercially available—dihydro-ff-MAS , dihydro-t-MAS , dehydrolathosterol , and dehydrodesmosterol—were identified by their unique m/z values and retention times . Based on their chemical structures , the m/z of dihydro-ff-MAS , dihydro-t-MAS , dehydrolathosterol , and dehydrodesmosterol are predicted to be 397 , 395 , 367 , and 365 Da , respectively . The retention times of these sterols was determined by analyzing sterol spectra of liver and feces from mice ( which have high concentrations of cholesterol biosynthetic intermediates relative to cholesterol ) consuming water supplemented with 10% D2O for 6 weeks . Sterol spectra in the D2O labeled mice were compared to those of unlabeled mice to identify peaks of the correct molecular mass that had an MS/MS fragmentation pattern characteristic of sterols and were endogenously synthesized ( based on deuterium incorporation ) . Only a single chromatographic peak met these criteria for each sterol . This strategy was tested using two additional sterols , ff-MAS and zymostenol , which were correctly identified when compared to the authentic standard . Three biosynthetic intermediates were not measured in this study . The concentrations of dehydrolathosterol were too low in all tissues and cell culture lines to reliably measure the isotopomer distribution . Lathosterol and zymostenol cannot be reliably measured in this system because they are isobaric with cholesterol , which saturates the signal at the m/z value of 369 Da . Cholesterol also saturated the isotopomers greater than M+1 for sterols with m/z = 367 , including desmosterol , 7-dehydrocholesterol , and zymosterol . Individual peaks for each sterol in Supplementary file 2 were integrated with Analyst software ( Framingham , MA ) . For each sterol , the fractional contribution of M+0 isotopomer ( M0m ) was calculated by dividing the intensity of the M+0 peak by the summed intensity of all measureable isotopomers of that sterol ( M+0 , M+1 , M = 2… ) . The fraction of each sterol that was newly synthesized ( g ) at each time point was determined by linearly deconvoluting M0m based on the fractional contribution of M+0 of natural ( M0n; ∼0 . 67 ) and fully labeled ( M0t; ∼0 . 35 ) sterols , expressed as:M0m=gM0t+ ( 1−g ) M0n . Which was solved for g:g=M0m−M0nM0t−M0n . This approach , which is called ISA or mass isotopomer distribution analysis ( MIDA ) , was developed by Kelleher and Masterson ( 1992 ) and Hellerstein and Neese ( 1992 ) ( see Figure 2—figure supplement 1 for a schematic representation ) . M0n was determined from the isotopic distributions of unlabeled samples ( Figure 2A ) , which were in good agreement with theoretical distributions based on the natural abundances of 13C , deuterium , and 18O . The value of M0t is dependent on both the enrichment of the labeling pool ( p; i . e . , the fraction of water that is D2O ) and the number of incorporation sites for label per molecule ( N; i . e . , the potential sites for deuterium incorporation ) . M0t was initially determined based on the experimentally determined asymptotic spectrum of isotopomers at the last time point of labeling ( Figure 2A ) ( i . e . , the distribution of isotopomers when all sterols were synthesized in the presence of D2O ) . The value was refined using MIDA as described by Hellerstein and Neese ( 1992 ) . For cell culture experiments , the media were enriched to 5% D20 thus p is 0 . 05 . The value of N was determined by regression of the observed M0t values relative to theoretical prediction . The calculated value of N was between 21 and 26 for all sterols in all cell lines , which is consistent with previously reported values ( Lee et al . , 1994 ) . M0t was determined by rounding N to the nearest integer . The concentrations of sterol intermediates were negligible in media and chow , except for t-MAS , which was present in chow at a concentration of 68 ng/mg . For mouse experiments , the value of p was determined by iteratively regressing N and p values against the experimentally determined asymptotic value of M0t to minimize the root mean square value across all tissues using MIDA . The value of N for lanosterol observed in cell culture experiments ( N = 21 ) was used as an initial value . This analysis yielded a value for p of 0 . 048 which was used for all tissues . The value of N for all sterols was then calculated based on the asymptotic value of M0t . The value of N ranged from 20 to 28 in all tissues . The rate of synthesis of each sterol was determined by assuming first-order kinetics , using the equation:g=g∞ ( 1−e−kt ) , where g∞ is the asymptotic value of g , t is time , and k is the rate constant . The value of g∞ and k were calculated using the Matlab Optimization Toolbox ( Natick , MA ) . The rate of synthesis of each sterol was determined by multiplying the rate constant ( k ) by the concentration . Total RNA from tissues and cells was isolated using commercial reagents ( RNA-STAT 60 ) . cDNA was synthesized from 2 μg RNA using Taqman ( Applied Biosystems , Grand Island , New York . ) with random hexamer primers and amplified by PCR in 2× SYBR Master Mix ( Applied Biosystems ) . The specific oligonucleotides for each transcript are shown in Supplementary file 3 . The levels of each mRNA were normalized to the level of 36B4 . Samples for immublotting were placed in sample buffer ( 10 mM HEPES , 1 . 5 mM MgCl2 , 10 mM KCl , 5 mM EDTA , 5 mM EGTA , 250 mM sucrose , pH 7 . 6 ) and homogenized using an Ultra-TURRAX homogenizer and then passed through a 22 G needle 35 times . Membranes were isolated from the homogenates by sequential centrifugation ( 7 min at 2200 rpm , followed by a second centrifugation at 14 , 000 rpm for 60 min ) . The pellet from the second centrifugation step was re-suspended in lysis buffer ( 10 mM tris–HCl , 100 mM NaCl , 1% SDS , 1 mM EDTA and EGTA , pH 6 . 8 ) and shaken for 1 hr at 4°C . After an hour , an equal volume of solubilization buffer ( 62 . 5 mM Tris–HCl , 15% SDS , 8 M urea , 10% glycerol , 100 mM DTT , pH 6 . 8 ) and 5× loading buffer was added ( Fermantas ) . Samples were size fractionated on 8% SDS-polyacrylamide gels and transferred to nitrocellulose ( GE Healthcare ) before incubating overnight at 4°C with a rabbit anti-mouse DHCR24 antibody ( 1:1000; Cell Signaling Technologies ) in PBST ( Sigma ) plus 5% ( wt/vol ) fat free milk . Anti-mouse calnexin ( 1:2000; MBD International ) was used to detect calnexin as a loading control . After incubation , the filter was washed and HRP-conjugated anti-rabbit IgG ( 1:5000; GE Healthcare ) was used as a secondary antibody . The filter was scanned using a LiCor Odyssey and visualized using iS ImageStudio software . Mouse DHCR24 was cloned into a PCDNA 3 . 1 TOPO TA vector using a commercial kit ( Invitrogen ) . The insert sequence was verified by sequencing and the plasmid was amplified with a Maxiprep kit ( Origene ) . The plasmid was prepared in a 1:3 ratio with Fugene 6 ( Promega ) ( wt/vol ) and diluted in Opti-Mem medium ( Life Technologies ) . Cells were transfected by exchanging the medium for Opti-Mem medium containing 8 μg plasmid DNA at a 1:3 ratio with Fugene 6 ( wt/vol ) . After 20 hr the medium was changed to Opti–Mem supplemented with 5% D2O . | Cholesterol is important for animals , both as an essential component of the membrane that surrounds cells and as a building block to make hormones and other biologically important molecules . However , cells limit how much cholesterol they make because an excess of this fatty molecule can cause serious health problems , including heart disease and stroke . Cholesterol is made via a complex process that involves more than 30 different steps , which can be organized into two biochemical pathways ( named the Bloch pathway and the Kandutsch–Russell pathway ) . The enzymes that carry out the steps in these pathways have been characterized in detail . Less is known about which of the two pathways is actually used in different cells and tissues , or how much cholesterol each pathway produces . This is partly because it is difficult to distinguish between the closely related intermediate molecules that are formed in each pathway . Mitsche et al . have now used mass spectrometry and isotope labeling techniques to analyze the relative contributions of the two cholesterol-making pathways in both cells grown in the laboratory and in mice . The experiments show that many cells use the Bloch pathway . However , no cells were found to use the Kandutsch–Russell pathway as it was originally described . Rather , some of the cells used a hybrid pathway where the production of cholesterol was started using the Bloch pathway and then after a certain number of steps , the process switched to using part of the Kandutsch–Russell pathway . Mitsche et al . referred to this mixed system as the ‘modified Kandutsch–Russell pathway’ . Mitsche et al . next examined the flow of molecules through these two pathways in different tissues and observed that the Bloch pathway is exclusively used in the testes and adrenal glands , which produce high levels of cholesterol . In contrast , the skin and brain use the modified Kandutsch–Russell pathway . In some tissues , a fraction of the building blocks that can be used to make cholesterol were instead diverted to make other products . This suggests that animals have maintained the two pathways over the course of evolution to enable them to generate a variety of products , which can be used to carry out different biological processes . One challenge following this work will be to use the newly developed methods to analyze other complex biochemical pathways . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2015 | Flux analysis of cholesterol biosynthesis in vivo reveals multiple tissue and cell-type specific pathways |
Bacterial chemotaxis systems are as diverse as the environments that bacteria inhabit , but how much environmental variation can cells tolerate with a single system ? Diversification of a single chemotaxis system could serve as an alternative , or even evolutionary stepping-stone , to switching between multiple systems . We hypothesized that mutations in gene regulation could lead to heritable control of chemotactic diversity . By simulating foraging and colonization of E . coli using a single-cell chemotaxis model , we found that different environments selected for different behaviors . The resulting trade-offs show that populations facing diverse environments would ideally diversify behaviors when time for navigation is limited . We show that advantageous diversity can arise from changes in the distribution of protein levels among individuals , which could occur through mutations in gene regulation . We propose experiments to test our prediction that chemotactic diversity in a clonal population could be a selectable trait that enables adaptation to environmental variability .
Escherichia coli uses a single chemotaxis protein network to navigate gradients of chemical attractants and repellents , as well as gradients of temperature , oxygen , and pH ( Sourjik and Wingreen , 2012 ) ( Figure 1A ) . The core of the network is a two-component signal transduction system that carries chemical information gathered by transmembrane receptors to flagellar motors responsible for cell propulsion . A second group of proteins allows the cells to physiologically adapt to changing background signal levels , enabling them to track signal gradients over many orders of magnitude . While different receptors allow cells to sense different signals , all signals are then processed through the same set of cytoplasmic proteins responsible for signal transduction and adaptation . This horizontal integration may impose conflicting demands on the regulation of these core decision-making components because signals can vary in time , space , and identity . In this study , we examine to what extent cell-to-cell variability in abundance of these core proteins may help resolve such conflicts . 10 . 7554/eLife . 03526 . 003Figure 1 . From proteins to fitness . ( A ) The cell receives extracellular ligand signals through transmembrane receptors . Changes in signal are rapidly communicated to the flagellar motors through the kinase CheA and response regulator CheY . CheZ opposes the kinase activity of CheA . At a slower timescale , the activity of the receptor complex physiologically adapts to its steady-state activity through the antagonistic actions of CheR and CheB . ( B ) Cartoon diagram of the response of the system to transient step-stimulus and definition of the key phenotypic parameters of the system . Without stimulation , the system has a steady-state clockwise bias , or fraction of time spent with motors in the clockwise state that results in tumbling . Upon stimulus with a step , CheY activity and therefore clockwise bias drops and the cell starts running more , then slowly adapts back to the steady-state with a characteristic timescale ( adaptation time ) . The steady-state clockwise bias and adaptation time are tuned by the concentrations of proteins in ( A ) . ( C ) Cells explore their environment by alternating between straight runs and direction-changing tumbles . When cells sense that they are traveling up a concentration gradient , they suppress tumbles to increase run length . Precisely how a cell navigates a gradient depends on its phenotypic parameters in ( B ) . ( D ) From a single genotype , noise in gene expression leads to a distribution of proteins expression levels ( blue shaded contours in protein space; left ) ; network design determines how proteins quantities map onto phenotypic parameters ( middle left ) ; the performance of all possible phenotypic parameter values across environments will determine the outer boundary of performance space ( middle right ) ; selection bestows a fitness reward based on performance and will reshape the performance front into the Pareto front , which , for optimal fitness , the population distribution should be constrained to ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03526 . 003 The cell uses flagella to explore its environment in a run-and-tumble fashion ( Berg and Brown , 1972 ) . Counterclockwise rotation of the flagella promotes the formation of a helical bundle that propels the cell forward in a run . Clockwise rotation tends to disrupt the bundle , interrupting runs with brief direction-changing tumbles . The fraction of time a motor spins clockwise , or clockwise bias , controls the frequency of tumbles and thus plays a central role in chemotactic behavior . Tumble frequency increases monotonically with clockwise bias until the latter reaches about 0 . 9 , at which point cells tumble nearly twice a second and are essentially stationary ( Alon et al . , 1998 ) . It has been observed that clonal cells , grown and observed under the same conditions without stimulation , will differ substantially in clockwise bias ( Park et al . , 2010 ) . The central logic of E . coli chemotaxis is to transiently decrease clockwise bias in response to an increase in attractant signal ( Figure 1B ) . This approach allows cells to climb gradients of attractants by lengthening runs up the gradient ( Figure 1C ) . The adaptation process that maintains receptor sensitivity is mediated by the covalent modification of the chemoreceptors through addition and subtraction of methyl groups by the enzymes CheR and CheB , respectively . Like clockwise bias , the timescale of this adaptation process has been observed to vary among clonal cells ( Spudich and Koshland , 1976 ) . The intracellular levels of these proteins are known to change both adaptation timescale and clockwise bias ( Alon et al . , 1999 ) . Chemoreceptor activity is communicated to the motors via phosphorylation of the response regulator CheY to form CheY-P by the receptor-associated kinase CheA . CheZ opposes the action of CheA by dephosphorylating CheY-P . Consequently , the balance of CheA and CheZ affects clockwise bias . The total amount of CheY in the cell determines the range of possible CheY-P levels , and due to noise in the expression of CheY ( Kollmann et al . , 2005 ) this dynamic range will likewise vary between clonal cells . These three phenotypic parameters—clockwise bias , adaptation time , and CheY-P dynamic range—are the main determinants of how E . coli performs chemotaxis . These in turn depend on the quantities of chemotaxis proteins within each individual cell . Hence , the copy numbers of these proteins directly determine the ability of the individual to navigate its environment . Since all signals are processed through the same core proteins , this dependency should be independent of the type of signal being followed . As such , cell-to-cell variability in protein abundance is likely a major contributor to the observed non-genetic behavioral diversity in clonal populations ( Figure 1D , First and second panels ) . Various mechanisms can contribute to such variability , including noise in gene expression ( Elowitz et al . , 2002 ) . Random segregation of proteins during cell division probably plays a role as well ( Huh and Paulsson , 2011 ) and may impose a lower bound on minimum variability attainable ( Lestas et al . , 2010 ) . Chemotaxis genes are chromosomally organized in operons—that is , expression of multiple genes are driven by common promoters . This genetic architecture ensures that noise in the activity of shared promoters will affect the expression of multiple genes in a correlated manner , conserving the ratios of proteins from cell to cell despite variations in total amounts ( Lovdok et al . , 2009 ) . Correlation in protein noise has been experimentally shown to be important in determining chemotactic performance ( Lovdok et al . , 2007 ) . Combined with the negative integral feedback design of the protein network , this conservation of protein ratios greatly reduces the occurrence of cells with unacceptable parameter values—for instance , those that only run or only tumble—and maintains the precision of the physiological adaptation process ( Alon et al . , 1999; Kollmann et al . , 2005; Barkai and Leibler , 1997; Yi et al . , 2000 ) . For these and other reasons ( Oleksiuk et al . , 2011; Endres and Wingreen , 2006; Sneddon et al . , 2012; Vladimirov et al . , 2010; Schulmeister etl . , 2011 ) , chemotaxis in E . coli is often said to be robust . Within this range of acceptable behaviors , however , substantial variability exists , and the fact that this variability has not been selected against raises the question of whether it might serve an adaptive function . Population diversity is known to be an adaptive strategy for environmental uncertainty ( Donaldson-Matasci et al . , 2008; Kussell and Leibler , 2005; Haccou and Iwasa , 2005 ) . In this case of chemotaxis , this would suggest that different cells in the population may hypothetically have behaviors specialized to navigate different environments ( Figures 1D , Second and third panels ) . Indeed , past simulations ( Vladimirov et al . , 2008; Jiang et al . , 2010; Dufour et al . , 2014 ) have shown that the speed at which cells climb exponential gradients depends on clockwise bias and adaptation time , and experiments ( Park et al . , 2011 ) using the capillary assay—an experiment that tests cells' ability to find the mouth of a pipette filled with attractant—have shown that inducing expression of CheR and CheB at different levels changes the chemotactic response . In order to understand the impact of these findings on population diversity , we must place them in an ecological context . Relatively little is known about the ecology of E . coli chemotaxis , but it is probable that they , like other freely swimming bacteria , encounter a wide variety of environments , from gradients whipped up by turbulent eddies ( Taylor and Stocker , 2012 ) to those generated during the consumption of large nutrient caches ( Blackburn et al . , 1998; Saragosti et al . , 2011 ) . In each case , variations in environmental parameters , such as in the amount of turbulence , the diffusivity of the nutrients , or the number of cells , will change the steepness of these gradients over orders of magnitude ( Taylor and Stocker , 2012; Stocker at al . , 2008; Seymour et al . , 2009 ) . Still other challenges include maintaining cell position near a source ( Clark and Grant , 2005 ) , exploration in the absence of stimuli ( Matthaus et al . , 2009 ) , navigating gradients of multiple compounds ( Kalinin et al . , 2010 ) , navigating toward sites of infection ( Terry et al . , 2005 ) , and evading host immune cells ( Stossel , 1999 ) . Each of these challenges can be described in terms of characteristic distances and times , for example the length-scale of a nutrient gradient , or the average lifetime of a nutrient source , or the characteristic time- and length-scales of a flow . Chemotactic performance , or the ability of cells to achieve a spatial advantage over time , will depend on how the phenotype of the individual matches the length- and time-scales of the environment . Considering the variety of scales in the aforementioned challenges , and the fact that all must be processed by the same proteins ( Figure 1A ) , it would seem unlikely that a single phenotype would optimally prepare a population for all environments , potentially leading to performance trade-offs ( Figure 1D , panel 3 ) wherein mutual optimization of multiple tasks with a single phenotype is not possible . Cellular performance will have an impact on fitness ( i . e . reproduction or survival ) depending on ‘how much’ nutrient or positional advantage is required to divide or avoid death . Therefore , selection that acts on chemotactic performance could transform performance trade-offs into fitness trade-offs ( Figure 1D , panels 3 and 4 ) , which are known to have direct consequences for the evolution of diversity ( Donaldson and Matasci et al . , 2008; Kussell and Leibler , 2005; Haccou and Iwasa , 1995; Shoval et al . , 2012 ) . Selection that favors top performers disproportionately to intermediate performers could hypothetically transform a weak performance trade-off into a strong fitness trade-off . Fitness trade-offs can lead to the development of multiple biological modules ( Reuffler et al . , 2012 ) . Some modules , like new limbs , may be permanent fixtures , while others , like metabolic pathways , may be switched on and off , either in response to the environment as it changes , or stochastically in anticipation of environmental fluctuation . This latter case , often called ‘bet-hedging ( Veening et al . , 2008 ) , ’ is a strategy used by bacteria to avoid extinction from antibiotic stress during infection ( Stewart and Cookson , 2012 ) and has evolved in the laboratory under fluctuating selection ( Beaumount et al . , 2009 ) . In these examples , environmental extremes lead to discrete partitioning of the population . Is there an intermediate case , a possible evolutionary stepping-stone , in which a single function is continuously diversified in the population without the formation of a wholly different state ? In this study , we seek to determine to what extent advantageous diversity can be created from a single biological network , as well as the possible mechanisms that may permit adaptation of such diversity in response to selective pressures . Due to the lack of quantitative information about the details of the natural environments of E . coli that would be relevant to chemotaxis , our goal is not the exact reconstruction of the distribution of challenges experienced by E . coli in the wild , but rather to use bacterial chemotaxis as a system to study the interactions between population diversity and environmental trade-offs . Although different signals are sensed by different receptors , the cell interprets all signals using the same set of proteins . Since we are interested in the relationship between cellular dynamics and the length- and time-scales of the environment , this allows us to simplify our study by focusing on different gradients of a single attractant type . Our findings should apply to different signal identities as well . For this study , we must be able to translate an individual cell’s protein concentrations into its fitness in different environments ( Figure 1D ) . Chemotaxis in E . coli is a system uniquely well-suited to this purpose . The wealth of molecular and cellular data that has been gathered by different research groups over the last several decades makes it one of the best-characterized systems in biology for the study of single-cell signal transduction and behavior . A key result of this past research is a molecular model of E . coli chemotaxis , which accounts for the interactions of all of the proteins in the network , that we will fit simultaneously to many experimental data sets . From this model , we will be able to calculate phenotypic parameters such as adaptation and clockwise bias as a function of protein concentrations ( Figure 1D , map from first to second panel ) . We will then simulate the performance of virtual E . coli cells with these phenotypic parameters to characterize any trade-offs that E . coli faces in performing fundamental chemotactic tasks ( Figure 1D , map from second to third panel ) . These tasks will be parameterized by characteristic lengths ( distance to source ) and times ( time allotted ) . A wide range of these environmental parameters will be explored to ensure that a full spectrum of cell–environment interactions are investigated . We will measure the performance of cells in the environments and apply different ecological models of selection to assign fitness . In doing so , we will examine how performance trade-offs give rise to fitness trade-offs ( Figure 1D , map from third to fourth panel ) . Finally , we will use a model of population diversity based on noisy gene expression to determine whether changing genetic regulation could allow populations to achieve a collective fitness advantage .
The first step in creating a single-cell conversion from protein levels into fitness was to build a model of the chemotaxis network . We began with a standard molecular model of signal transduction based explicitly on biochemical interactions of network proteins . We simultaneously fit the model to multiple datasets measured in clonal wildtype cells by multiple labs ( Park et al . , 2010; Kollmann et al . , 2005; Shimizu et al . , 2010 ) . Along with previous measurements reported in the literature , this fitting procedure fixed the values of all biochemical parameters ( i . e . reaction rates and binding constants ) , leaving protein concentrations as the only quantities determining cell behavior ( ‘Materials and methods’ , Supplementary file 1 ) . The fit took advantage of newer single-cell data not used in previous models that characterize the distribution of clockwise bias and adaptation time in a clonal population ( Park et al . , 2010 ) . In order to fit this data , we coupled the molecular model with a model of variability in protein abundance , adapted from Lovdok et al . ( Lovdok et al . , 2009; ‘Materials and methods’ ) . In this model , the abundance of each protein is lognormal-distributed and depends on a few parameters that determine the mean abundance and the extrinsic ( correlated ) and intrinsic ( uncorrelated ) noise in protein abundance ( details of the model discussed further below ) ( Elowitz et al . , 2002 ) . By combining these components , our model simultaneously fit the mean behavior of the population ( Kollmann et al . , 2005 ) and the noisy distribution of single-cell behaviors ( Park et al . , 2010 ) ( Figure 2—figure supplement 1 ) . In all cases , a single set of fixed biochemical parameters was used , the only driver of behavioral differences between cells being differences in protein abundance . Given an individual with a particular set of protein levels , we then needed to be able to calculate the phenotypic parameters: adaptation time , clockwise bias , and CheY-P dynamic range . To do so we solved for the steady state of the model and its linear response to small deviations in stimuli relative to background ( ‘Materials and methods’ ) . This produced formulae for the phenotypic parameters in terms of protein concentrations . For simplicity , we did not model the interactions of multiple flagella . Rather , we assumed that switching from counterclockwise to clockwise would initiate a tumble after a lag of 0 . 2 s that was required to account for the finite duration of switching conformation . A similar delay was imposed on switches from tumbles to runs . In this paper we only consider clockwise bias values below ∼0 . 9 , because above this value cells can spend many seconds in the clockwise state ( Alon et al . , 1998 ) . During such long intervals , non-canonical swimming in the clockwise state can occur . In this case , the chemotactic response is inverted and cells tend to drift away from attractants ( Khan et al . , 1978 ) . This behavior is therefore maladaptive for the cell; however , it is only observed in mutant cells ( Alon et al . , 1998 ) . In experiments with wildtype cells , this regime is not observed ( Park et al . , 2010 ) because of the robust architecture of the network ( Kollmann et al . , 2005 ) ( ‘Materials and methods’ ) . Using the definitions for adaptation time , clockwise bias , and CheY-P dynamic range , we reduced the molecular model into a phenotypic model written in terms of phenotypic parameters rather than protein levels ( ‘Materials and methods’ ) . Simulating the step-response of the molecular model with a given set of protein levels matched the behavior of the phenotypic model with corresponding phenotypic parameters ( Figure 2—figure supplement 2 ) . Because there are half as many phenotypic parameters as different proteins , the phenotypic model made it computationally possible to explore large ranges of behavior in the simulations we describe in the next section . To characterize chemotactic trade-offs faced by E . coli , we began by investigating which chemotactic phenotypes performed best in different ecological tasks . Here , we defined a phenotype as a particular set of values of the phenotypic parameters: adaptation time , clockwise bias , and CheY-P dynamic range . We used the phenotypic model to simulate the behavior of individual phenotypes in various environments and measured the performance of each phenotype based on metrics appropriate to each ecological challenge . In total , these steps provided us with a direct mapping from individual protein levels to chemotactic performance in the ecological tasks we describe below ( Figure 1D ) . E . coli , like other commensals and pathogens , must survive relatively nutrient-poor environments outside the host until it can colonize a new host . An important ecological parameter in this situation is the characteristic distance between resources , which sets the typical signal length-scale a bacterium must navigate . When a source is close , the challenge might be to climb steep gradients and to stay near the source . In contrast , when the environment is sparse , the ability to explore and navigate shallow gradients may be more important . Another important ecological parameter is the characteristic time-scale of changes in the environment , which dictates the allotted time a bacterium has to navigate its environment . As this time becomes shorter ( e . g . when bacteria must take advantage of nutrient patches in moving flows [Taylor and Stocker , 2012; Celani and Vergassola , 2010] ) chemotactic performance becomes more important . For these reasons , we parameterized environments in terms of distances and times . The range of values was chosen such that at one extreme , cells begin at the source , and at the other , the distance and time requirements are so stringent that reaching the source is only possible by swimming randomly ( i . e . pure diffusion ) . We considered two tasks . The first is a foraging challenge in which a spherical parcel of nutrient appears at a certain distance from the cell and immediately begins to diffuse away . This occurs , for instance , upon lysis of a unicellular eukaryote ( Blackburn et al . , 1998 ) . The location of the parcel is unpredictable and could be close or far . Each cell in the simulation accumulated nutrient by collecting an amount proportional to the concentration at its position at every time-step . Performance was defined as the amount of nutrient acquired ( Figure 2A ) within a certain time limit . For simplicity we assumed that consumption by an individual is small enough not to have an impact on the gradient itself . Feedback of populations onto the shape of the gradient certainly plays a role in many ecological scenarios and could be considered in this framework in the future . 10 . 7554/eLife . 03526 . 004Figure 2 . Performance of chemotactic phenotypes depends on environmental conditions . ( A ) Cartoon diagram ( not to scale ) of the foraging challenge: cells navigating a 3-D time-varying gradient created by diffusion of a small spherical drop of nutrient 100 µm in diameter with diffusion coefficient of 550 µm2/s and methyl-aspartate concentration of 100 mM . Inset: radial profile of the attractant concentration over time . ( B ) Average nutrient collected by each phenotype ( combination of clockwise bias and adaptation time ) in environment in A over 8000 replicates per phenotype . Because we are investigating optimal phenotypes and CheY-P dynamic range does alter the results as long as it is sufficiently high ( Figure 2—figure supplement 7 ) , results shown here use YTot = 13 , 149 mol . /cell . Clockwise bias and adaptation time were sample in log-spaced bins . Cells start near to the source ( 0 . 2 mm from its center ) , and are allowed to swim for 13 min while accumulating a small fraction of the nutrient they sense . ( C ) . Same as B except that cells start farther away from the source ( 1 mm from its center ) and 14 , 000 replicates per phenotype were used . ( D–F ) Similar to ( A–C ) but the environment consists of a colonization challenge: diffusion of ligand out of a spherical non-depleting source representing a colonization site; source methyl-aspartate concentration was 10 mM . Rather than nutrient collection , performance ( E and F ) was quantified as the reciprocal of the arrival time at the source averaged over all replicates ( 9000 and 36 , 000 for E and F respectively ) with a maximum time allotted of 15 min . DOI: http://dx . doi . org/10 . 7554/eLife . 03526 . 00410 . 7554/eLife . 03526 . 005Figure 2—figure supplement 1 . Comparing the model to single cell and population averaged measurements . The same set of model parameter values is used for all the plots . ( A ) Adaptation time and motor clockwise ( CW ) bias . Bottom: normalized histogram of motor clockwise bias in the population . Top: The mean and standard deviation of adaptation time in each bin of CW bias . Red lines: experimental data from ref . 4 . Black lines: model . Circles: Individual cells from the model . Color: probability density . ( B ) Population-averaged CW bias as a function of fold changes in mean expression level of all pathway proteins following ref . 7 . Red: data from ref . 7 . Black: model . ( C ) Population-averaged methylation rate as a function of population-averaged receptor activity obtained by exposing cells to exponential ramps of methyl-aspartate as described in ref . 42 . Red circles: data from ref . 42 . Black: simulation of model . DOI: http://dx . doi . org/10 . 7554/eLife . 03526 . 00510 . 7554/eLife . 03526 . 006Figure 2—figure supplement 2 . Agreement between protein model and parametric dynamics model . ( A ) Cartoon of step function of ligand delivered to immobilized cells in simulation to test response dynamics . ( B ) Direct comparison of response of molecular model ( blue ) and phenotypic model ( green ) with the same parameters to stimulus of the form in A illustrating close agreement . DOI: http://dx . doi . org/10 . 7554/eLife . 03526 . 00610 . 7554/eLife . 03526 . 007Figure 2—figure supplement 3 . Performance as a function of distance to source . ( A ) Cells with various phenotypes were challenged to forage a source presented at varying distances , r0 from 75 μm to 3 mm . Between 6000 and 30 , 000 replicates were simulated for each phenotype . 〈Ncol . 〉 : the average nutrient collected by all replicates of a given phenotype in μmol . ( B ) Same as A but for a colonization challenge; 〈1/Tarr . 〉: the average reciprocal-of-arrival-time of all of the replicates of a given phenotype in min−1 . ( C ) Data in A smoothed with a Gaussian filter and resampled on a higher resolution grid of phenotypic parameters . Diamond: phenotype with highest performance . ( D ) Same as C but with the data in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03526 . 00710 . 7554/eLife . 03526 . 008Figure 2—figure supplement 4 . Optimal phenotypes as a function of source distance . For each source distance and each task , the phenotype with highest performance was identified as shown in Figure 2—figure supplement 3 . The clockwise bias and adaptation time of these phenotypes are shown with the marker color corresponding to the distance to the source . Diamonds: foraging case . Circles: colonization case . DOI: http://dx . doi . org/10 . 7554/eLife . 03526 . 00810 . 7554/eLife . 03526 . 009Figure 2—figure supplement 5 . Effect of time restrictions on foraging performance . Cells were challenged to forage sources that appeared at distances of 200 , 5000 , or 1000 μm away ( columns from left to right ) . Different amounts of time were allotted to cells to accumulate ligand: 3 min , 5 min , 11 min , 15 min ( rows from top to bottom ) . 〈Ncol . 〉 : the average nutrient collected by replicates of a given phenotype in μmol . DOI: http://dx . doi . org/10 . 7554/eLife . 03526 . 00910 . 7554/eLife . 03526 . 010Figure 2—figure supplement 6 . Effect of time limits on near/far foraging trade-offs . Trade-offs in performance between foraging near and far sources are shown . From left to right ( cyan to magenta ) , the far case is progressively more distant compared to the near case: 1 mm , 1 . 5 mm , 2 mm , 3 mm . From top to bottom ( bright to dark colors ) , the time allotted is increasing: 3 min , 5 min , 11 min , 15 min . Reduced time allotment makes the front ( black line ) more concave for the same pair of environments . DOI: http://dx . doi . org/10 . 7554/eLife . 03526 . 01010 . 7554/eLife . 03526 . 011Figure 2—figure supplement 7 . Effect of source concentration on performance . Performance calculate and plotted as described in Figure 2—figure supplement 3 , but for different concentrations at the source . Left block ( ‘Foraging’ ) : foraging performance for increasing source distance ( columns ) and increasing source concentration ( rows ) : L1 = 1 mM , 10 mM , 100 mM . Right block ( ‘Colonization’ ) colonization performance for increasing source distance ( columns ) and increasing source concentration ( rows ) : L1 = 100 µM , 1 mM , 10 mM . DOI: http://dx . doi . org/10 . 7554/eLife . 03526 . 01110 . 7554/eLife . 03526 . 012Figure 2—figure supplement 8 . Effect of CheY-P dynamic range on performance . Left block ( ‘Foraging‘ ) : foraging performance for near ( 200 µm ) and far ( 1000 µm ) sources and increasing CheY-P dynamic range , which was changed through the total number of CheY molecules , Ytot , as described in the SI . Right block ( ‘Colonization’ ) same as left block but for colonization . DOI: http://dx . doi . org/10 . 7554/eLife . 03526 . 012 The second environment recapitulates a colonization task , in which a colonization site opens up at a random distance from the cell and immediately starts releasing an attractant signal by diffusion . This case is analogous to the classic capillary experiment ( Mesibov and Adler , 1972 ) and may have relevance to infection by species such as uropathogenic E . coli ( Lane et al . , 2005 ) . We approximated the site as a persistent spherical zone with a non-depleting concentration of attractant . Performance was defined by minimizing the time to reach the site , equivalent to maximizing the reciprocal of the arrival time , before a global time limit , which may be determined ecologically by the carrying capacity of the site or the periodic purging of the area around the site ( Figure 2D ) . Cells unable to reach the colonization site by that time were given an infinite arrival time and consequently a zero performance value . Later , we consider the reduction of this time limit as an ecological factor . For each ecological task , we scanned different environmental parameters ( the distance at which the source appears , the time allotted , and the source concentration [Figure 2—figure supplements 3 , 5 , 7 , respectively] ) and simulated the performance of different phenotypes . For each phenotype and environment , 6000–30 , 000 replicate trajectories were averaged together to quantify performance as a function of phenotype and environment . We began with no constraints or correlations between phenotypic parameters and scanned them independently; later we consider the effect of biological constraints on phenotypic distributions . When a nearby source appeared , cells in the foraging challenge immediately experienced high nutrient levels and were challenged to maintain their position despite having been exposed to a large increase in signal . Successful cells had high clockwise bias , which curtails long runs , and short adaptation time , which mitigates large responses ( Figure 2B ) . These cells in a way are defeating chemotaxis and motility both: to stay rooted , they tumble constantly and have a fast adaptation time that reduces the duration of response . If chemotactic populations are preparing for unexpected types of environments but have uniformly turned on expression of chemotaxis and motility genes , cells with these phenotypes could potentially function as if those processes were turned off , without having to introduce a genetic on-off switch . Conversely , when a source appeared farther away , cells had to use longer runs to reach the expanding front of the gradient and long adaptation times to integrate the weaker signals at its tails ( Figure 2C ) . If time is further limited , this far-source effect is exaggerated ( Figure 2—figure supplement 5 ) . The case of colonization was similar , except that shorter adaptation times were favored overall as compared to foraging . This was because the gradient geometry is much steeper in the vicinity of the source due to its persistently high concentration , and climbing that final part of the gradient was required for colonization ( Figure 2D inset vs . A inset ) . Climbing steep gradients requires fast adaptation to stay abreast of quickly changing background levels . The source concentration played a minor role in colonization; however , when foraging less concentrated sources , the favored strategy for far distances inverts from low to high clockwise bias , indicating that at that point little can be gained from motility—in fact higher motility may move the cell away from the source ( Figure 2—figure supplement 7 ) . The dynamic range of CheY-P has a negligible effect on cell performance so long as it is sufficiently high as to ensure that the response of CheY-P to kinase activity is linear and does not saturate ( Figure 2—figure supplement 8 ) . For this reason , when we discuss optimal performance in the subsequent analysis , we assume that the total amount of CheY molecules in the cell has been selected to be high enough to avoid these limitations ( ‘Materials and methods’ ) . In both challenges , the distance at which the source appeared substantially changed which phenotypes outperformed the others . In general , distant sources required lower clockwise bias and longer adaptation time than nearby ones ( Figure 2C compared to B and F compared to E ) . This becomes even more apparent if we plot the optimal phenotype as a function of source distance ( Figure 2—figure supplement 4 ) . These results are consistent with our recent study ( Dufour et al . , 2014 ) that used an analytical model to predict the velocity of cells climbing static one-dimensional gradients and detailed the mechanistic basis of performance differences between phenotypes . There , we demonstrated a trade-off wherein steep gradients required fast adaptation time and high clockwise bias for optimal velocity , whereas shallow gradients required slow adaptation time and low clockwise bias . Our present simulations of ecological tasks show that this trade-off also exists in more complex chemotactic scenarios . The dependence of the optimal phenotype on the environment follows the same trend in the previous analytical model as it does in our current simulation results , wherein simulations of distant sources are similar to simple shallow gradients and nearer sources are analogous to steeper gradients . Our results could be tested using several types of chemotactic performance experiments . The radial symmetry of our environments makes it possible to use the capillary ( Park et al . , 2011 ) and plug assays ( Lanfranconi et al . , 2003 ) , which present cells with a concentrated source of attractant . The soft agar swarm plate assay ( Lovdok et al . , 2007 ) could be used as well if modified to introduce nutrient solution to one spot of the plate instead of the whole . Microfluidic chemotaxis assays ( Kalinin et al . , 2010 ) could be constructed using soft lithography to reproduce these environments with a higher level of precision . In each of these cases , the distance between cells and the source and the duration for which the source is presented could be varied , as well as the source concentration . Cells with high performance should be selected , analyzed for their phenotypes and protein abundance , and re-grown , either under continual presentation of the same condition or switching between two or more conditions . Using these types of experiments , our theoretical results predict several specific outcomes . First , seeding the same clonal population in different assays that have different length- or time-scales should select for different optimal subpopulations with different phenotypic parameters and different levels of protein expression . Such measurements would make it experimentally possible to verify the chemotactic trade-offs we predict . Experimental work using the capillary assay already supports this claim ( Park et al . , 2011 ) . In the case of laboratory evolution with one selection condition , we predict an eventual shift toward genotypes that suppress population noise , as well as toward mutations in chemotaxis protein RBSs that allow the mean clockwise bias and adaptation time to specialize for this task . In this case , we predict that populations will reduce phenotypic diversity but run into a lower limit of protein noise . These outcomes could be measured by performing single cell phenotype analyses and by re-sequencing the operon . Conversely , alternating selection in different assays or different length- and time-scales may lead to enhanced phenotypic noise and still other RBS mutations . In these cases , whole genome re-sequencing may show alterations to the operon structure or to the master regulators of chemotaxis . Strains that are evolved in the lab could be compared to the wildtype ancestor in order to gain insight into the types of environments the latter evolved in . Furthermore , investigating phenotypic diversity in wild strains in comparison to domesticated and evolved laboratory strains may uncover differences that reflect the level of environmental diversity faced in their respective lifestyles .
The chemotaxis system exhibits significant plasticity in the shape of phenotypic distributions , which can provide fitness advantages in chemotactic trade-offs . Such trade-offs arise from environmental variability because the performance of a chemotactic phenotype is sensitive to the length- and time-scales of the environment it must navigate . This dependency is especially strong when time for navigation is limited . Though at this stage we cannot know what distribution of chemotactic challenges wildtype E . coli have faced , we do expect trade-offs to arise from the diversity of time- and length-scales in environmental encounters . Our simulations environments were simplified . They omitted many real-world factors for future studies , such as competition between multiple species , turbulence , and viscosity in environments such as soil or animal mucosa . As new data on these interactions emerge , the framework we introduced could be used to investigate trade-offs and resulting phenotypic distributions . Additionally , interactions with more than two environments are likely to occur and could be analyzed in the same way . Such cases will likely impose more constraints on navigation , giving rise to stronger trade-off problems . Increasing the number of phenotypic parameters would not necessarily alleviate these constraints , which would instead be primarily governed by the distance between the optimal phenotypes for these tasks in phenotypic parameter space . In our framework , we expanded the traditional genotype–phenotype relationship to consider protein levels separately . While genotype could be broadly defined to include both coding sequences and regulators of noise , separate treatment of protein levels permitted analysis of copy number variability apart from changes in the proteins themselves . This approach could be applied to other signal transduction systems , since variability in the levels of signaling proteins may change behavior as much as changing protein biochemistry . In this study , we tuned the distribution of protein levels using numerical parameters , but such changes would in fact occur through mutations . Mean expression levels could change via gene duplication , RBS point mutations , mRNA structures , or altered activity of upstream regulators . Phages and recombination events can reorganize genes , changing intrinsic noise relative to extrinsic noise by altering expression correlation . Regulators of promoters can incur mutations that result in negative feedback repression to reduce promoter noise . Protein localization affects partitioning noise , which is interesting since some chemotaxis proteins assemble into discrete membrane-bound clusters while others do not . In the future , it would be interesting to study the extent to which higher expression levels will result in fitness costs , possibly introducing trade-offs . For instance , physiological adaptation via the enzymatic actions of CheR and CheB consumes cellular resources , imposing metabolic costs that depend inversely on the adaptation timescale ( Lan et al . , 2012 ) . Different media and growth phases alter the expression levels of these proteins ( Li and Hazelbauer , 2004; Scott et al . , 2010 ) and will naturally change the distribution of phenotypes as well—this could be a mechanism for separating protein levels required for chemotaxis from those better suited for growth . In this study , challenges and regrowth occurred in discrete sequential steps and there was no direct inheritance of phenotype . The relative importance of these features will depend on the relationship between their time-scales and those of the environmental challenges ( Kussell and Leibler , 2005 ) . If the time-scale of environmental change is much slower than the time-scale of adaptation , for example , populations will adapt to their current environment rather than the statistics of environmental fluctuations . A new feature of our conceptual framework is the distinction between performance and fitness . Organisms exhibit many behaviors that , to researchers , are not directly connected to survival and reproduction . These gaps in our understanding inhibit our ability to understand the evolutionary significance of many organismal behaviors . Here , we demonstrated methods for broaching these questions quantitatively , and in so doing uncovered the relevant finding that nonlinearities in selection can strengthen or weaken trade-offs . This will be of general interest to those studying fitness trade-offs since the nature of selection can change the optimality of pure vs mixed population strategies . While we have used E . coli as a model system due to the wealth of experimental data , the framework developed here could be used to extend these questions to other human commensals and pathogens , with the hope of better understanding their ecology and pathogenesis . The closely related chemotaxis system in Salmonella enterica is required for virulence ( Stecher et al . , 2004 ) , as is the substantially different motility system of Borrelia burgdorferi ( Sze et al . , 2012 ) . On the other hand , pathogens such as Pseudomonas aeruginosa have multiple motility systems to tackle different environments during infection ( O'Toole and Kolter , 1998 ) . Phenotypic diversification within a single system may bridge the gap between one system and many by allowing populations to adapt to greater environmental variation without developing a new biological module . Multicellular organisms also exhibit different motion strategies in their constituent cells , from the singular approach of human sperm to the different motility patterns of neutrophils as they navigate the body to sites of infection and capture invading organisms ( Eisenbach and Lengeler , 2004 ) . Our framework could be used to investigate several open questions in such systems: How does behavioral diversity of single cells affect the fitness of the organism , and when is the diversification of a single cell type supplanted by the commitment of a new developmental cell lineage ? From the simplest two-component systems to the most elaborate signal transduction cascades , proteins responsible for sensing environmental signals are usually distinct from those involved in making behavioral decisions . Often , the output of many types of receptor proteins are fed into a much smaller number of signal transduction pathways . While cells can control their sensitivity to different signals by regulating the expression of different receptors , the integration of multiple signals through a central group of proteins will place conflicting demands on those core proteins . Thus , while horizontal integration is beautifully economical and a ubiquitous feature of biological pathways , our study illustrates that it is also likely to introduce trade-offs by design . Biology is replete with noise . Although the concept of non-genetic individuality may have been initially coined in reference to E . coli chemotaxis ( Spudich and Koshland , 1976 ) , we now know that many other biological systems exhibit substantial non-genetic cell-to-cell variability , including stem cell differentiation ( Huang , 2009 ) , bacterial sporulation ( Maamar . et al . , 2007 ) , and cancer cell response to chemotherapy ( Spencer et al . , 2009 ) . Different systems may have different mechanistic drivers that create , constrain , and adapt this variability . In all cases , however , it is conceivable that through genetic changes to drivers of non-genetic diversity , populations of cells may achieve higher collective success in tackling biological trade-off problems . This form of diversity may constitute an evolutionary stepping stone on the path from one to multiple biological modules .
To optimize population fitness , we first defined a general expression for population fitness beginning with the fitness of a single phenotype . Chemotaxis is non-deterministic , hence , in each environment g , an individual phenotype x⇀ had a distribution of performance V , or p ( V|x⇀ , g ) , where x⇀ is a vector of adaptation time , clockwise bias , and CheY-P dynamic range . Fitness was a function of single-cell performance f ( V ) . To calculate the fitness of a phenotype in a given environment , we took the expected value of its fitness over its distribution of performance 〈f〉x⇀ , g=∫f ( V ) p ( V|x⇀ , g ) dV . This should not be confused with the fitness of the average performance . We assume for simplicity that populations encounter challenges sequentially , all cells in the population experience each challenge simultaneously and in the same way , and populations must survive through all environments . Hence , within a given environment , a population consisting of many cells with different phenotypes has fitness equal to the average of its constituent cells 〈f〉P , g=P ( x⇀ ) 〈f〉x⇀ , gdx , where P ( x⇀ ) is the population distribution of phenotypes . Following this , population fitness from one environment to the next is multiplicative . In the long term this results in a geometric mean across environments , weighted by the probability of encountering each environment: ( 26 ) F=exp ( ∫log ( ∫〈f〉P , g ) h ( g ) dg ) , where h ( g ) is the distribution of environments . This formula is consistent with previous derivations ( Haccou and Iwasa , 1995 ) but has been extended to include stochastic performance of individual cells and a distinction between fitness and performance . While Equation ( 26 ) provides a general solution , in the specific cases analyzed in this study , the populations consist of a finite number of different phenotypes , there are a finite number of discrete simulation replicates , and for simplicity we show cases that compare two discrete environments g1 and g2 with occurrence probability h and ( 1—h ) . As such , the discrete calculation of population fitness becomes: ( 27 ) F= ( ∑ξ=1Npopfξ , g1Npop ) h ( ∑ξ=1Npopfξ , g2Npop ) ( 1−h ) , where ξ indexes the cells in the population , Npop is the number of cells in the population , and fξ , g is the fitness of the phenotype of cell ξ in environment g determined using a look-up table constructed from simulation data as described above . The trade-off problem itself is thus parameterized by: h , g1 , g2 , and the form and parameters of f ( V ) that gave rise to the look-up table . In the case of foraging these are the nutritional requirement K and the dependency n; for colonization there is only the time limit TL . These we collectively call the trade-off parameters . The population gene expression parameters generate a list of individuals with different phenotypes as described above . We can optimize the fitness of the population as a whole ( Figure 7 ) by first calculating population fitness F ( Equation ( 27 ) ) for a set of trade-off parameters . We then used MATLAB's pattern search optimization function on the population fitness formula , allowing only the gene expression parameters P0 , η , and ω to vary , but not the trade-off parameters , the biochemical parameters , or any other parameters . The constraints on these parameters are described below , and h was 0 . 8 . From this we obtained the optimized population parameters for strong and weak trade-offs ( performed separately ) . For each type of ecological task , the strong and weak trade-offs are between the same pair of near and far environments , with the same form of selection function , but each has a different set of selection function parameters . Since there is always some irreducible noise in biology , we used experimental observations to provide lower bounds for the noise parameters in our model . For a limit on the intrinsic component , we took the wildtype level of intrinsic noise , which we obtained by fitting the model to wildtype data ( described above ) . Multiple studies have described the advantage of reduced intrinsic noise in chemotaxis , so we assume wildtype cells are likely to be functioning at or near the minimum intrinsic noise . In order to apply this constraint , we ensure that the intrinsic noise scaling parameter and mean protein levels are constrained within the optimization algorithm such that the condition ηP0≥ηwtP0wt is maintained . There is also a lower bound on the minimum total protein noise , defined as the coefficient of variation squared , measured in single E . coli cells to be about 0 . 09 for proteins with a mean expression level of above 100 copies per cell ( Taniguchi et al . , 2010 ) . This constraint in practice acts more on extrinsic noise than on intrinsic noise since in our case the latter is typically fairly low . To enforce this constraint computationally , we ensure that P0 , η , and ω of Equation ( 18 ) are chosen by the optimization such that the squared coefficient of variation of every protein is above 0 . 09 . This typically has the effect of keeping ω above about 0 . 09 , depending on P0 and η . Increases in global expression levels of up to approximately 3 fold are observed for different strains and growth media[50] , and using mutations in flgM , increases up to 7 fold are possible ( Kollmann et al . , 2005 ) . Hence , we set our upper limit of mean expression levels at 5 fold to work within that range . | Bacterial colonies are generally made up of genetically identical cells . Despite this , a closer look at the members of a bacterial colony shows that these cells can have very different behaviors . For example , some cells may grow more quickly than others , or be more resistant to antibiotics . The mechanisms driving this diversity are only beginning to be identified and understood . Escherichia coli bacteria can move towards , or away from , certain chemicals in their surrounding environment to help them navigate toward favorable conditions . This behavior is known as chemotaxis . The signals from all of these chemicals are processed in E . coli by just one set of proteins , which control the different behaviors that are needed for the bacteria to follow them . Different numbers of these proteins are found in different—but genetically identical—bacteria , and the number of proteins is linked to how the bacteria perform these behaviors . It has been suggested that diversity can be beneficial to the overall bacterial population , as it helps the population survive environmental changes . This suggests that the level of diversity in the population should adapt to the level of diversity in the environment . However , it remains unknown how this adaptation occurs . Frankel et al . developed and combined several models and simulations to investigate whether differences in chemotaxis protein production help an E . coli colony to survive . The models show that in different environments , it can be beneficial for the population as a whole if different cells have different responses to the chemicals present . For example , if a lot of a useful chemical is present , bacteria are more likely to survive by heading straight to the source . If not much chemical is detected , the bacteria may need to move in a more exploratory manner . Frankel et al . find that different amounts of chemotaxis proteins produce these different behaviors . To survive in a changing environment , it is therefore best for the E . coli colony to contain cells that have different amounts of these proteins . Frankel et al . propose that the variability of chemotaxis protein levels between genetically identical cells can change through mutations in the genes that control how many of the proteins are produced , and predict that such mutations allow populations to adapt to environmental changes . The environments simulated in the model were much simpler than would be found in the real world , and Frankel et al . describe experiments that are now being performed to confirm and expand on their results . The model could be used in the future to shed light on the behavior of other cells that are genetically identical but exhibit diverse behaviors , from other bacterial species to more complex cancer cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"microbiology",
"and",
"infectious",
"disease"
] | 2014 | Adaptability of non-genetic diversity in bacterial chemotaxis |
Cell polarization underlies many cellular and organismal functions . The GTPase Cdc42 orchestrates polarization in many contexts . In budding yeast , polarization is associated with a focus of Cdc42•GTP which is thought to self sustain by recruiting a complex containing Cla4 , a Cdc42-binding effector , Bem1 , a scaffold , and Cdc24 , a Cdc42 GEF . Using optogenetics , we probe yeast polarization and find that local recruitment of Cdc24 or Bem1 is sufficient to induce polarization by triggering self-sustaining Cdc42 activity . However , the response to these perturbations depends on the recruited molecule , the cell cycle stage , and existing polarization sites . Before cell cycle entry , recruitment of Cdc24 , but not Bem1 , induces a metastable pool of Cdc42 that is sustained by positive feedback . Upon Cdk1 activation , recruitment of either Cdc24 or Bem1 creates a stable site of polarization that induces budding and inhibits formation of competing sites . Local perturbations have therefore revealed unexpected features of polarity establishment .
Cell polarization occurs in all kingdoms of life . In metazoa , it is critical for many cellular events including cell migration , embryogenesis , and cytokinesis . Polarization is dynamic , enabling cells to reorient to changing spatial and temporal cues . The central , conserved regulator of cell polarity in eukaryotes is the small Rho-family GTPase , Cdc42 ( Johnson and Pringle , 1990; Etienne-Manneville , 2004; Macara , 2004 ) . Establishment of an axis of polarity frequently involves the accumulation of active Cdc42 at a unique position on the cell cortex . An initial cue induces Cdc42-GTP accumulation at a unique site , where it then concentrates via an amplification system . This amplification system maintains the polarity site ( Thompson , 2013 ) . While this general scheme of polarity establishment is fairly well established , the dynamic molecular interactions required for establishing , maintaining , and enforcing a single axis of polarity are not well understood . One of the best studied cell polarization systems is bud site selection in Saccharomyces cerevisiae . Wild type yeast cells use landmark-directed cues to define the polarization axis ( Figure 1A ) . However , cells lacking these spatial cues or the GTPase that transduces these cues , Rsr1 , efficiently establish polarity in a process known as symmetry breaking . The current model of symmetry breaking suggests that stochastic accumulation of Cdc42-GTP induces a positive feedback loop mediated by a polarity complex containing the Cdc42 GEF , Cdc24 , the scaffold protein , Bem1 , and Cla4 , which binds directly to activated Cdc42 ( Howell et al . , 2012 ) ( Figure 1B ) . This tripartite complex is required for polarity establishment and its molecular features provide the requisite domains to result in positive feedback ( Chenevert et al . , 1992; Holly and Blumer , 1999; Bose et al . , 2001; Butty et al . , 2002 ) . Computational models and in vivo analyses of symmetry breaking have uncovered behaviors consistent with positive feedback , including traveling waves and oscillatory accumulation of polarity proteins ( Goryachev and Pokhilko , 2008; Ozbudak et al . , 2005; Howell et al . , 2012 ) . However , important assumptions of the models have not been directly tested . For instance , whenever Cdc42 is active it should induce accumulation of the polarity complex; however , the validity of this critical assumption has yet to be established . Similarly , although polarity establishment has been dissected in considerable detail , the events preceding these have attracted relatively little attention . For example , in symmetry breaking polarization , do events in early G1 influence the site of symmetry breaking ? To what extent are the functions of the polarity proteins influenced by the state of the cell cycle ? These omissions are largely due to the absence of tools that permit controlled perturbation of Cdc42 activation . In recent years , a number of genetically-encoded tools have been developed that allow control of diverse proteins using light ( Tischer and Weiner , 2014 ) . We have developed a set of small , highly engineered optogenetic protein tags , called TULIPs , that permit light-mediated control of protein-protein interactions ( Strickland et al . , 2012 ) . In this system , the optically-responsive protein ( LOVpep ) is tethered to the plasma membrane whereas its cognate binding partner ( ePDZ ) freely diffuses in the cytoplasm . Illumination induces a conformational change in LOVpep that allows it to bind ePDZ , causing the rapid ( <10 s ) relocation of ePDZ to the membrane . LOVpep spontaneously relaxes to the dark state with a half-time of 80 s ( Strickland et al . , 2012 ) . By fusing a protein of interest to ePDZ , we can potently and dynamically control its spatiotemporal localization using light . In previous proof of principle experiments , we demonstrated that local recruitment of Cdc24 directs formation of a mating projection in α-factor treated cells ( Strickland et al . , 2012 ) . Using optogenetic tools , we have probed the mechanism for yeast cell polarization . Our results provide direct evidence that positive feedback contributes to polarity establishment once Cdk1 is activated at START . However , while the current model predicts that the polarity components invariably function together in a potent positive feedback loop , we find that they do not do so prior to Cdk1 activation and that Cdc42 activation does not invariably induce the canonical positive feedback loop . Rather , we demonstrate the existence of a second positive feedback loop involving Cdc24 that precedes the canonical loop and appears to function independently of Bem1 . Finally , we show that multiple nascent sites of polarization compete , and that this competition is particularly potent upon activation of the Bem1-dependent positive feedback loop . We conclude that two alternative modes of positive feedback function in concert to promote polarity initiation and establishment , with their respective behaviors under strict cell cycle control .
In order to probe the mechanism of cell polarization , we developed optogenetic tools to recruit yeast polarity proteins to defined sites on the cell cortex . Specifically , we co-expressed a membrane-tethered variant of LOVpep ( Mid2-GFP-LOVpep ) with either Bem1-ePDZ or Cdc24-ePDZ in diploid rsr1∆ cells . In all conditions , the TULIPs-tagged proteins were expressed under the control of a β-estradiol inducible promoter ( Louvion et al . , 1993 ) , and the ePDZ-tagged polarity component was expressed in addition to its endogenous counterpart . We perturbed cells by illuminating at a single site with a diffraction-limited laser . The response to this perturbation was examined by following the recruitment of a Cdc42 biosensor derived from the Cdc42 effector protein Gic2 ( Tong et al . , 2007 ) and phase optics to observe bud emergence . We first examined the ability of Cdc24 and Bem1 to bias the site of polarization in unpolarized cells as a function of illumination frequency . We measured the angle , θ , between the site of illumination and the position of the nascent bud . These values were linearly scaled such that budding at the center of the laser target was assigned a score of 1 and budding opposite the target was assigned a score of −1; these scores were averaged for a population of cells ( Polarization Efficiency = average ( 1-2θ/π ) ) ( Figure 1D ) . Recruitment of Cdc24 or Bem1 recruitment was able to bias the bud site at very low light doses; Bem1 was slightly more efficient than Cdc24 ( Figure 1D ) . Additionally , recruitment of either component induced robust accumulation of active Cdc42 at the site of illumination , without altering the timing of polarization ( ~95 min between bud emergence events , regardless of photo-activation state or molecule recruited; data not shown ) . As the frequency of light increased , the ability of Cdc24 to bias the bud site remained roughly constant until the highest light dose , while the polarization efficiency of Bem1 dropped by ~50% once illumination increased to greater than 3 pulses per minute ( Figure 1F , Figure 1—figure supplement 1 ) . The reason for this drop is unclear though a similar drop is observed with ePDZ-mCherry ( Figure 1—figure supplement 2 ) ; this may result from light-induced rupture of the LOV2-flavin mononucleotide adduct ( Kennis et al . , 2004 ) . While symmetry breaking is predicted to be random relative to the previous bud site , the position of the new bud was not completely random . Additionally , targets were not randomly positioned , as the area around the previous bud site was underrepresented . These biases would cause a slight underestimation of polarization efficiency ( Figure 1—figure supplement 3 ) . Optogenetically recruited Cdc24 and Bem1 were also able to induce polarization of heterozygous RSR1/rsr1Δ diploids ( Figure 1—figure supplement 4 ) . 10 . 7554/eLife . 26722 . 003Figure 1 . Cdc24 recruitment can induce Cdc42 activation in polarized and unpolarized cells . ( A ) The endogenous cue is mediated by a system involving Rsr1 to yield patterned budding . Rsr1 directly interacts with Bem1 and Cdc24 to recruit and activate Cdc42 at an adjacent bud position . Rsr1 recruits Bem1 and/or Cdc24 ( 1 ) to activate Cdc42 ( 2 ) adjacent to the previous bud neck . Cdc42 undergoes positive feedback ( 3 ) by interacting with Cla4 to promote its own accumulation . ( B ) In the absence of Rsr1 , cells undergo a symmetry breaking event mediated by positive feedback . The symmetry breaking event may involve a stochastic accumulation of Cdc42-GTP at a unique location that then recruits the Cla4-Bem1-Cdc24 complex ( 1 ) . Cdc24 activates additional molecules of Cdc42 ( 2 ) to promote positive feedback ( 3 ) . ( C ) Light-induced symmetry breaking by recruiting the GEF Cdc24 to activate Cdc42 at a prescribed position and induce positive feedback . Localized photo-activation of LOVpep recruits Cdc24-ePDZb ( 1 ) to activate Cdc42 ( 2 ) . Activated Cdc42 interacts with Cla4 to induce positive feedback . ( D ) Polarization efficiency of a population of cells where each point represents an individual cell . The angle Θ is defined by the angle between the site of bud emergence and the laser position . Data are averages of all cells across multiple experiments ( n experiments >= 2 , N total cells > 15 for each group ) . Average and ± SEM is indicated . Statistical analysis in Figure 1—figure supplement 1 . Strains used: WYK8440 and WYK8301 . ( E ) Representative phase and inverted fluorescent images depicting the activation of Cdc42 in response to either Cdc24 or Bem1 recruitment in polarized or non-polarized cells . Difference overlay images place the subtraction of the two fluorescent images overlaid onto the phase contrast image . An increase in fluorescent signal is depicted by red pseudo-coloring on the overlay . Cells were treated to increasing doses of light . Shown from bottom to top , representative images for each group:1 pulse/60 s , 3 pulses/30 s , and 3 pulses/15 s . Each image is 16 . 2 µm x 16 . 2 µm . Strains used: WYK8440 and WYK8301 . ( F ) Box-and-Whisker plots depicting the relative change in mean fluorescence intensity of the Cdc42 biosensor at targeted regions . The relative intensity of polarized cells is the mean of the intensity at 10 min post initial illumination normalized to a control site ( typically 180° from target ) on the same time . The relative intensity for non-polarized cells compares accumulation of the biosensor 10 min before bud emergence normalized to a control site at the same time . Data is combined across multiple experiments ( n experiments >= 2 , N total cells > 15 for each group ) . Statistical analysis in Figure 1—figure supplement 5 . Strains used: WYK8440 and WYK8301 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 00310 . 7554/eLife . 26722 . 004Figure 1—figure supplement 1 . Statistical analysis of polarization efficiency as a function of light dose . ( A ) Comparative statistical analysis of polarization efficiency in response to Cdc24-ePDZ recruitment at various light doses ( Data shown in Figure 1D ) . Gray box indicates populations not statistically different at p=0 . 05 , orange box denotes statistically significant at p<0 . 05 , Mann-Whitney U test . ( B ) Comparative statistics for Polarization efficiency in response to Bem1 recruitment at various light doses as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 00410 . 7554/eLife . 26722 . 005Figure 1—figure supplement 2 . Recruitment of ePDZ-mCherry as a function of light dose . ( A ) Phase contrast and fluorescence images of GFP-LOVpep and ePDZ-mCherry in response to two light pulses per 60 s . Panels on the right indicate ePDZ-mCherry distribution prior to photo-illumination ( 0’ ) and after 2 min of photo-illumination to the indicated positions ( 2’ ) . Each image is 32 . 4 µm x 34 . 2 µm . Strain used: WYK8476 . ( B ) The relative change in mean intensity of ePDZ-mCherry at the targeted region after 2 min of illumination relative to the intensity at time 0 . Light gray indicates +/-SEM . Data is combined across multiple experiments ( n experiments >= 5 , N total cells > 75 for each group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 00510 . 7554/eLife . 26722 . 006Figure 1—figure supplement 3 . Bias in target position and new bud position relative to the previous bud . ( A ) Distribution of targeting position relative to new bud formation in mock-illuminated cells ( N cells > 120; aggregated from all mock illumination conditions ) . ( B ) Distribution of new bud site relative to the previous bud site in mock-illuminated cells ( N cells > 120 ) . ( C ) Distribution of target position relative to the previous bud site in mock-illuminated cells ( N cells > 120 ) . ( D ) Comparative polarization efficiency in two simulations . Model 1 assumes that there is no bias in target or new bud position . Model 2 approximates the biases the new bud and target positions as in B and C , respectively . Specifically , responding cells were simulated to respond with polarization efficiency = 0 . 75 . In model 1 , cells that do not respond were assumed to bud randomly in the range 46°−180° , with an average angle of 90° , corresponding to the angle expected if both targets and the new bud were random relative to the previous bud . In Model 2 , cells that do not respond were assumed to bud randomly in the range 46°−180° , with an average angle of 102° , as an average difference of 102° approximates the aggregate bias resulting from the experimental bias in target position and the bias in bud site selection . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 00610 . 7554/eLife . 26722 . 007Figure 1—figure supplement 4 . Local accumulation of either Cdc24 or Bem1 is sufficient to override the landmark-directed pathway . Polarization efficiency of a population of cells heterozygous for Rsr1 in response to recruitment of Cdc24-ePDZ or Bem1-ePDZ . Each point represents an individual cell . Average and +/- SEM is indicated . Polarization in response to both Cdc24 and Bem1 recruitment are statistically significant to their dark state controls . Strains used: WYK8598 and WYK8599 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 00710 . 7554/eLife . 26722 . 008Figure 1—figure supplement 5 . Statistical analysis of Cdc42 biosensor accumulation in polarized and non-polarized cells as a function of light dose . ( A ) Statistical analysis for Cdc42 biosensor accumulation in response to Cdc24-ePDZ recruitment in polarized and non-polarized cells ( data from Figure 1F ) . Gray box indicates populations not statistically different at p=0 . 05 , orange box denotes statistically significant at p<0 . 05 , Mann-Whitney U test . ( B ) Statistics for Cdc42 biosensor accumulation in response to Bem1 recruitment as in A . ( C ) Statistical comparison of Cdc42 biosensor accumulation in polarized vs . non-polarized cells at each light dose . Statistical analysis as in A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 008 We next assessed how the response varied as a function of both cell cycle position and the frequency with which cells were illuminated . At all doses of light , cells that were at the start of the cell cycle ( ~10 min before bud emergence ) activated Cdc42 in response to both Cdc24 and Bem1 recruitment . Some non-illuminated cells also exhibited Cdc42 activation at the target site , as the bud site occasionally coincided with the target position ( Figure 1D , F ) . To determine whether cells were constitutively responsive to optogenetic recruitment of Cdc24 or Bem1 , we illuminated polarized cells with small to medium-sized buds . At infrequent light pulses , polarized cells did not activate Cdc42 in response to Cdc24 or Bem1 recruitment ( 1–3 pulses , Figure 1E , F , Figure 1—figure supplement 5 ) . When cells were illuminated at a greater frequency ( >2 x per minute ) , those that expressed Cdc24-ePDZ activated detectable Cdc42 at all stages of the cell cycle . Conversely , Bem1 recruitment did not result in Cdc42 activation in polarized cells , even at higher illumination frequencies ( Figure 1E , F , Figure 1—figure supplement 5 ) . In summary , Cdc42 activation in unpolarized cells is readily induced in response to either Cdc24 or Bem1 recruitment; however , only high levels of Cdc24 recruitment can generate Cdc42-GTP in polarized cells , indicating that limitations to Cdc42 activation exist in polarized cells . From these data , we conclude that local recruitment of either Cdc24 or Bem1 is able to efficiently bias the bud site . Because our primary interests lie in the endogenous regulation of Cdc42 activation , we chose to dissect polarity establishment using a light dose ( 3 pulses per minute ) that efficiently biases the bud site , rather than the higher doses that might overcome the mechanisms that limit Cdc42 activation to one site in the cell . To understand the role of Bem1 in regulating polarization , we induced Bem1 recruitment in cells expressing either the Cdc42 biosensor , Cdc24-tdTomato expressed on a low-copy plasmid from its endogenous promoter , or cells in which one copy of Bem1 was tagged with tdTomato . Optogenetic recruitment of Bem1 was sufficient to promote activation of Cdc42 , accumulation of Cdc24 ( Figure 2A , B ) , and was able to efficiently position the bud site ( Figure 2D ) . Critically , we found that recruitment of Bem1 was able to induce the accumulation of endogenous Bem1 at the prescribed site , directly showing that polarization in yeast proceeds via positive feedback ( Figure 2A , B ) . These results raise the possibility that the observed activation of Cdc42 represents a combination of the direct activation of Cdc42 by the Cdc24 directly interacting with optogenetically-recruited Bem1 and amplification by endogenous mechanisms . Although Bem1 recruitment efficiently induced cell polarization , it did not induce precocious Cdc42 activation , or precocious accumulation of Cdc24 or Bem1 as compared to control cells ( Figure 2B , Figure 2—figure supplement 1 ) . Thus , the activity of the Cdc24-Bem1-Cla4 feedback loop appears sensitive to cell cycle regulation . 10 . 7554/eLife . 26722 . 009Figure 2 . Local accumulation of Bem1 directs bud site positioning via positive feedback . ( A ) Representative phase and fluorescence images and kymographs showing the position of the laser target and accumulation of the Cdc42 biosensor , Cdc24 , and endogenous Bem1 ( respectively ) . Each image is 16 . 2 µm x 16 . 2 µm . Kymographs are generated by iteratively linearizing the membrane at each time point ( schematic ) . Bud emergence occurs at Y = 0° and at time = 0 min . Arrows between the kymograph and fluorescent images indicate equivalent time points . Strains used: WYK8308 , WYK8318 , and WYK8576 . ( B ) Accumulation kinetics for each component in response to Bem1 recruitment . Red line is photo-activated cells , blue line is mock-illuminated cells . Bud emergence is time = 0 . Data combined across multiple experiments ( n experiments >= 2 , N total cells > 20 for each condition ) . ( C ) Domain schematic of Bem1-ePDZ , annotated with mutation sites and interactions . ( D ) Polarization efficiency of a population of cells as in 1F . Red dots are single photo-activated cells . Blue dots are single mock-illuminated cells . Data is combined across multiple experiments ( n experiments > 2 , N total cells > 20 for each group ) . Error bars indicate S . E . M . Polarization efficiency of Bem1 , Bem1 R369A , and Bem1 P355A in the light is statistically significant relative to Bem1 K482A light and dark , and their respective dark-state controls . Bem1 K482A was not statistically significant relative to its dark state control . p<0 . 05 , Mann-Whitney U test . Strains used: WYK8308 , WYK8434 , WYK8435 , and WYK8436 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 00910 . 7554/eLife . 26722 . 010Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 01010 . 7554/eLife . 26722 . 011Figure 2—figure supplement 1 . Photo-recruited Bem1 does not alter timing of accumulation of Cdc42-GTP , Cdc24 and Bem1 . Box-and-whisker plot of Cdc42-GTP and Cdc24 appearance time in response to Bem1 recruitment in photo-activated ( red ) or mock-illuminated ( blue ) cells . Outliers are depicted by black squares . Data are as in Figure 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 01110 . 7554/eLife . 26722 . 012Figure 2—figure supplement 2 . Recruitment of Cdc24-binding deficient Bem1 cannot induce accumulation of endogenous Bem1 . ( A ) Phase contrast and fluorescence images of Bem1-tdTomato accumulation in response to photo-recruitment of Bem1 ( K482A ) . Each image is 16 . 2 µm x 16 . 2 µm . Strain used: WYK8505 . ( B ) Polarization efficiency of population of cells where each point represents an individual cell from the condition depicted in A . Data are averages of all cells across multiple experiments ( n experiments >= 2 , N total cells > 20 for each group ) . Average and ± SEM is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 012 If polarization induction by Bem1 proceeds via a positive feedback loop , Bem1 likely functions by directly interacting with the GEF Cdc24 ( Kozubowski et al . , 2008 ) . To test this prediction , we introduced a mutation in the Bem1 PB1 domain that has previously been shown to abolish the Bem1-Cdc24 interaction ( K482A ) ( Figure 2C ) ( Ito et al . , 2001 ) and tested its ability to induce cell polarization . Mutational inactivation of the Bem1 PB1 domain ablated its polarization efficiency ( Figure 2D ) and its ability to induce recruit endogenous Bem1 ( Figure 2—figure supplement 2 ) . These results suggest that polarity induction by Bem1 requires direct interaction with Cdc24 . In addition to its Cdc24-interacting domain , Bem1 contains a PxxP motif and SH3 domain that interact , as well as a region demonstrated to interact with phosphoinositides ( PI ) ( Figure 2C ) . We introduced mutations known to abolish these activities ( Irazoqui et al . , 2003 ) and assayed each for their ability to bias the bud site . Both mutants exhibited partial activity ( Figure 2D ) . Although the PI-binding region is required for Bem1 function ( Irazoqui et al . , 2003 ) , light-mediated recruitment or indirect recruitment of endogenous Bem1 might compensate for this defect in this assay . We conclude that yeast polarization involves a positive feedback loop mediated by the Bem1-GEF complex . Furthermore , these data indicate that bud site selection requires activation of Cdc42 . To gain a mechanistic understanding of light-induced polarity establishment in response to recruitment of Cdc24 GEF , we determined which molecular features are required for induction of cell polarization ( Figure 3A , B ) . In addition to GEF activity , Cdc24 also contains a PB1 domain that enables it to bind to a corresponding PB1 domain on Bem1 ( Butty et al . , 2002 ) . Recently , it has been suggested that localization of Cdc24 is required for polarization via its ability to activate Cdc42 ( Woods et al . , 2015 ) . As expected , we find that mutation of conserved surface residues in the GEF-GTPase interface ( Rossman et al . , 2002 ) rendered Cdc24 inactive for polarization activity ( Figure 3A , B ) and Cdc42 activation ( data not shown ) . Deletion of the Bem1-binding domain of Cdc24 marginally reduced the polarization efficiency of the GEF ( Figure 3A , B ) . These results indicate that local recruitment of Cdc24 polarizes yeast cell growth through its ability to activate Cdc42 , and that the interaction of the recruited Cdc24 with other cellular proteins - including endogenous wild-type Cdc24 - is not sufficient to induce polarization . 10 . 7554/eLife . 26722 . 013Figure 3 . Light-mediated recruitment of Cdc24 directs bud site positioning . ( A ) Domain schematic of Cdc24-ePDZ and variants thereof . ( B ) Polarization efficiency of photo-illuminated ( red ) or mock-illuminated ( blue ) cells . Data are averaged across multiple experiments ( n experiments > = 2 , N total cells > 15 , for each group ) . Average and ± S . E . M . indicated . Polarization efficiency of Cdc24 and Cdc24ΔPB1 were statistically significant relative to Cdc24-GEF-dead light/dark , and statistically significant compared to their respective dark state controls . The difference between Cdc24 and Cdc24ΔPB1 were not statistically significant . The response to Cdc24-GEF dead was not statistically significant relative to its dark state control . p<0 . 05 , Mann-Whitney U test . Strains used: WYK8440 , WYK8437 , and WYK8439 . ( C ) Panels of representative phase and fluorescence images and kymographs showing the position of the laser target and accumulation of a Cdc42 biosensor or endogenous Bem1 in response to Cdc24 recruitment . Each image is 16 . 2 µm x 16 . 2 µm . Strains used: WYK8440 and WYK8301 . ( D ) Accumulation kinetics for the Cdc42 biosensor or endogenous Bem1 in response to Cdc24 recruitment . Data are combined across multiple experiments ( n experiments >= 2 , N total cells > 15 for each condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 01310 . 7554/eLife . 26722 . 014Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 01410 . 7554/eLife . 26722 . 015Figure 3—figure supplement 1 . Light-induced recruitment of Cdc24 induce precocious Cdc42 activation , but does not alter Bem1 kinetics . Box-and-whisker plot of Cdc42-GTP and Bem1 appearance in response to Cdc24-ePDZ in photo-activated ( red ) and mock-illuminated ( blue ) cells . Outliers are depicted by black squares . Bud emergence occurs at time = 0 . Data are as in Figure 3D . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 015 To observe the dynamics of Cdc42 activation in response to optogenetic Cdc24 recruitment , we used the biosensor of active Cdc42 to visualize GTPase activation ( Figure 3C , Video 1 ) . In control cells , 50% of cells displayed Cdc42 activation 12 min prior to bud emergence ( Figure 3D , Figure 3—figure supplement 1 ) . In experiments in which Cdc24 was continuously recruited , we observed precocious Cdc42 activation; 50% of cells exhibited polarized accumulation of active Cdc42 27 min prior to bud emergence ( Figure 3C , D ) . Precocious Cdc42 activation was less robust than that observed in the ~12 min prior to bud emergence . Thus , Cdc24 recruitment can induce Cdc42 activation at an earlier stage in the cell cycle than Bem1 . 10 . 7554/eLife . 26722 . 016Video 1 . Photo-recruitment of Cdc24 is sufficient to activate Cdc42 and bias the bud site . Representative phase contrast and fluorescent time-lapse images of the response to Cdc24 recruitment in cells expressing the Cdc42 biosensor . Left panel is the phase image with the position of the target defined by the black circle . Right panel is the Cdc42 biosensor . fps = 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 016 We next asked whether endogenous Bem1 precociously accumulated in a similar manner as activated Cdc42 . Strikingly , Bem1 did not accumulate until ~13 min prior to bud emergence ( Figure 3C , D ) . Indeed , despite accumulation of active Cdc42 as a consequence of Cdc24 recruitment , Bem1 accumulates at the same time relative to bud emergence in illuminated and mock-illuminated cells ( −13 min and −12 min , respectively ) ( Figure 3D , Figure 3—figure supplement 1 ) . Notably , the initial accumulation of Bem1 coincides with the time in which more robust Cdc42 activation is observed . These results suggest that , unlike previous models for the activity of the polarity proteins , Cdc42 activation does not invariably lead to recruitment of the intact polarity complex . The differential accumulation of Bem1 and Cdc42 in response to optogenetic recruitment of Cdc24 indicates that Cdc24 has the potential to activate Cdc42 independently of Bem1 , and that activated Cdc42 does not inevitably induce recruitment of the polarity complex . If this is the case , then Cdc24 , Bem1 , and Cdc42-GTP may not constitutively colocalize in non-perturbed cells . To test this hypothesis , we performed a detailed colocalization analysis of the three pairwise combinations of active Cdc42 , Bem1 , and Cdc24 . In all strains expressing fluorescent Bem1 , one copy of endogenous Bem1 was tagged . The Cdc42 biosensor was used to assess the pool of active Cdc42 . To visualize Cdc24 , we transformed cells with a low-copy plasmid encoding an extra copy of Cdc24-GFP under its own promoter . Though it localizes to the expected sites and it can be readily detected , Cdc24-GFP was dimmer than Bem1-tdTomato , Bem1-GFP , and the Cdc42 biosensor . Using imaging conditions optimized for detection of faint signals ( see Materials and methods ) , we acquired maximum intensity Z-projections of asynchronous cells . Using bud size and polarization as a guide , we sub-divided the cells into unbudded G1 cells , small-budded cells , and large-budded cells and characterized the localization pattern of each pair of probes in each cell cycle state . Unbudded cells were subdivided into two groups: early and late . Early cells were characterized by small distinct puncta of all three probes . Late cells featured a wide , cortically associated band in which all three probes localized . The categorization of these patterns as early and late is substantiated by previous studies ( Ozbudak et al . , 2005 ) and time-lapse imaging ( see below , Figure 4C ) . 10 . 7554/eLife . 26722 . 017Figure 4 . Cdc42-GTP , Cdc24 , and Bem1 do not constitutively colocalize prior to polarity establishment . ( A ) Inverted fluorescent images depicting the three pairwise combinations of Cdc42-GTP , Bem1 , and Cdc24 . All images are Z projections of 0 . 25 µm slices for the center 3 µm . Each image is 13 . 5 µm x 13 . 5 µm . Strains used: WYK8550 , WYK8551 , and WYK8552 . ( B ) Percentage of colocalization amongst puncta in early G1 or late G1 cells . Plots are separated by pairs as in A . Data are averages of all cells across multiple experiments ( n experiments = 2; N cells > 15 for each condition; N total cells > 100 ) . Error bars S . E . M . n . s indicates populations not statistically different at p>=0 . 05 , *p<0 . 05 , Mann-Whitney U test . ( C ) Time-lapse images capturing polarization in non-perturbed cells expressing either the Cdc42 Biosensor-tdTomato , endogenous Bem1-tdTomato , or Cdc24-tdTomato . Images are single planes of the center of the cell . Blue arrows denote the time and position of polarity establishment . Time = 0 is the onset of imaging , and images were captured for ten minutes at either 60 s or 30 s intervals as denoted . Strains used: WYK8301 , WYK8440 , and WYK8575 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 01710 . 7554/eLife . 26722 . 018Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 01810 . 7554/eLife . 26722 . 019Figure 4—figure supplement 1 . Pairwise analysis depicting the percent colocalization . Venn diagrams for early G1 and late G1 depicting the extent of colocalization between each pair . The numbers within each shape refer to the number of puncta , with the total number of puncta within each shape equaling 100 . The colocalization of all three components is presented in a model that is consistent with the pair-wise results . Related to results shown in Figure 4B . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 01910 . 7554/eLife . 26722 . 020Figure 4—figure supplement 2 . Validation of distinct pools of polarity proteins in early G1 . ( A ) Single plane snapshots feature similar distributions of Cdc24 , Bem1 , and Cdc42-GTP as in the Z-projections . Each image is 13 . 5 µm x 13 . 5 µm . Strains used: WYK8550 , WYK8551 , and WYK8552 . ( B ) Inverted fluorescent images depicting the pairwise combination of Bem1 and Cdc24 in early G1 . All images are single planes and the number in the upper left corresponds to the order in which the images were acquired . Each image is 13 . 5 µm x 13 . 5 µm . Quantification corresponds to the percentage of puncta that remain the same , appear , or disappear between 1 and 3 or 2 and 4 . Data are averages of all cells across multiple experiments ( n experiments = 2; N cells > 50 ) . Error bars S . E . M . n . s indicates populations not statistically different at p>=0 . 05 , *p<0 . 05 , Mann-Whitney U test . Strains used: WYK8551 . ( C ) Representative images corresponding to the pairwise combination of the Cdc42 biosensor and Cdc24 in early G1 . All images and quantification as in A . Data are averages of all cells across multiple experiments ( n experiments = 2; N cells > 50 ) . Error bars S . E . M . n . s indicates populations not statistically different at p>=0 . 05 , *p<0 . 05 , Mann-Whitney U test . Strains used: WYK8552 . ( D ) Representative images depicting the pairwise combination of the Cdc42 biosensor and Bem1 in early G1 . All images and quantification as in A . Data are averages of all cells across multiple experiments ( n experiments = 2; N cells > 50 ) . Error bars S . E . M . n . s indicates populations not statistically different at p>=0 . 05 , *p<0 . 05 , Mann-Whitney U test . Strains used: WYK8550 . ( E ) Representative fluorescent images of early G1 cells co-expressing Bem1-tdTomato and Bem1-GFP . All images are Z projections of 0 . 25 µm slices for the center 3 µm . Each image is 13 . 5 µm x 13 . 5 µm . Quantification is the percentage of colocalization amongst puncta in early G1 . Data are averages of all cells across multiple experiments ( n experiments = 2; N cells > 75 ) . Error bars S . E . M . n . s indicates populations not statistically different at p>=0 . 05 , *p<0 . 05 , Mann-Whitney U test . Strain used: WYK8554 . ( F ) Representative fluorescent images of early G1 cells co-expressing Cdc24-tdTomato and Cdc24-GFP . All images and quantification as in D . Data are averages of all cells across multiple experiments ( n experiments = 2; N cells > 75 ) . Error bars S . E . M . n . s indicates populations not statistically different at p>=0 . 05 , *p<0 . 05 , Mann-Whitney U test . Strain used: WYK8553 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 020 As expected , all three probes colocalize at the tips of small and nascent buds . However , at other stages of the cell cycle Bem1 , Cdc24 , and active Cdc42 do not completely colocalize ( Figure 4A ) . In large-budded cells , activated Cdc42 and Bem1 localize throughout the growing bud , while the GEF is limited to the growing bud tip . Additionally , puncta of the Cdc42 biosensor are detected in the mother cell . In early G1 cells , Bem1 and Cdc24 each localize in cortically associated puncta , but surprisingly , these puncta do not fully overlap . Furthermore , both Bem1 puncta and Cdc24 puncta displayed limited overlap with the Cdc42 biosensor , with only 35% and ~50% of puncta colocalizing , respectively ( Figure 4B , Figure 4—figure supplement 1 ) . Single plane image sequences confirm significant differences between two probes and limited differences between sequential images of a single probe . Further control experiments with co-expression of individual polarity proteins separately tagged with two fluorophores reveals extensive overlap ( Figure 4—figure supplement 2 ) . Notably , the localization pattern shifted in late G1 cells as all three components localize to the wide band , and the amount of reciprocal overlap between Bem1-Cdc24 and Bem1-Cdc42 biosensor increased to greater than 70% ( Figure 4B , Figure 4—figure supplement 1 ) . This more extensive overlap amongst Cdc24 , the Cd42 biosensor , and Bem1 in late G1 cells suggest that prior to the onset of polarization , Cdc24 has the potential to function independently of Bem1 . Of note , the punctate patterns of Bem1 , Cdc24 , and Cdc42 seen in early G1 cells are below the limit of detection in the assays involving optogenetic recruitment . Exposure times in those experiments were far lower in order to limit photobleaching during long-term ( >90 min ) imaging . To better understand the variation in protein localization at different cell cycle stages , we complemented the static imaging with time-lapse imaging . As proteins tagged with GFP tended to be dimmer than those tagged with tdTomato , we imaged cells expressing the tdTomato-tagged variants of the Cdc42 biosensor , Bem1 , or Cdc24 . Asynchronous populations of cells were imaged for a total of 10 min; images were acquired more frequently during the central four minutes . Each component was highly dynamic in the minutes leading up to the formation of a prominent wide band of accumulation , confirming that the punctate stage precedes polarization . While some puncta existed in the same spot for up to 3 min , appearance and disappearance of puncta were common ( Figure 4C ) . These results indicate that many Cdc24 and Bem1 molecules are not in a constitutive complex during early G1 , which is consistent with the finding that recruitment of Cdc24 , but not Bem1 , induces Cdc42 activation at this stage of the cell cycle . Cell cycle regulated assembly of the polarity complex could readily explain both results . To gain insight into cell cycle regulation of the polarity complex , we performed Cdc24 recruitment studies in strains co-expressing a marker of cell cycle entry , Whi5-tdTomato , and either the Cdc42 biosensor or Bem1-tdTomato . Nuclear import of Whi5 is a marker of Cdk1 inactivation . Whi5 is concentrated in the nucleus during the interval between mitotic exit until Cdk1 activation at Start ( Costanzo et al . , 2004; Skotheim et al . , 2008 ) ( ~25 min prior to bud emergence , Figure 5—figure supplement 1 ) . Mother cells that have not imported Whi5 into the nucleus are unresponsive to Cdc24 recruitment . Conversely , Cdc24 recruitment can induce Cdc42 activation in unbudded cells with nuclear Whi5 or unbudded cells in which Whi5 nuclear export has occurred ( Figure 5A , B ) , suggesting that mitotic exit is required for the activity of optogenetically recruited Cdc24 . In contrast , cortical recruitment of Bem1-tdTomato in response to Cdc24 recruitment did not occur until ~12 min after Whi5 nuclear export , corresponding to ~13 min prior to budding ( Figure 5A , B , Figure 5—figure supplement 2 ) . These results are consistent with a model in which Cdk1 activation promotes assembly of the Bem1-GEF complex to engage active Cdc42 . 10 . 7554/eLife . 26722 . 021Figure 5 . Cdk1 Activation is required for Bem1 accumulation , but dispensable for Cdc42 activation . ( A ) Representative panel of time-course images from Cdc24-ePDZ recruitment in cells co-expressing Whi5-tdTomato with either the Cdc42 biosensor or Bem1-tdTomato . Each image is 16 . 2 µm x 16 . 2 µm . Strains used: WYK8500 and WYK8502 . ( B ) Whi5 nuclear exit kinetics and accumulation kinetics for either Cdc42 activation or Bem1 at Cdc24-ePDZ recruitment sites . Purple line represents Whi5 exit , with data combined for both light- and dark-state conditions as they are not significantly different ( Figure 5—figure supplement 1 ) . Bud emergence occurs at time = 0 . Data are combined across multiple experiments ( n experiments > 2; N total cells > 25 for each condition ) . ( C ) Representative panels and sub-images of cells depicting the response to Cdc24 recruitment +/- Cdk1 activity . Each inset is 6 . 5 µm x 6 . 5 µm . Strains used: WYK8441 and WYK8442 . ( D ) Accumulation plots indicating the response of Cdc42 biosensor or Bem1 in response to Cdc24 recruitment +/- Cdk1 activity . Purple lines represent cells treated with 75 µM 1NM-PP1 . Green lines represent Vehicle-treated cells . Data are combined across multiple experiments ( n experiments = 2; N total cells > 25 for each condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 02110 . 7554/eLife . 26722 . 022Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 02210 . 7554/eLife . 26722 . 023Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 02310 . 7554/eLife . 26722 . 024Figure 5—figure supplement 1 . Neither probe expression nor illumination affects the timing of Whi5 nuclear exit . Timing for nuclear exit of Whi5 for each condition in Figure 5B plotted on a single plot . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 02410 . 7554/eLife . 26722 . 025Figure 5—figure supplement 2 . Photo-recruited Cdc24 activates Cdc42 prior to Whi5 nuclear exit . Correlation between Whi5 nuclear exit and either Cdc42 activation or Bem1 accumulation at Cdc24 recruitment sites . Dashed line represents simulataneous occurrence of the event and Whi5 nuclear exit; points below the dashed line indicate the event occurs before Whi5 nuclear exit and points above the line indicate the event occurs after Whi5 nuclear exit . Data are as in Figure 5B . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 02510 . 7554/eLife . 26722 . 026Figure 5—figure supplement 3 . Bem1 accumulation requires Cdk1 activity in response to light-induced recruitment of Cdc24 . Box-and-whisker plot of Cdc42-GTP and Bem1 appearance in cdk1-as cells . Green boxes are vehicle-treated cells , while purple boxes are 1NM-PP1-treated cells . Outliers are depicted by black squares . Vehicle or 1NM-PP1 is added at time = −20 min; time = 0 is the start of live-cell imaging and photo-activation . Data are as in Figure 5D . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 02610 . 7554/eLife . 26722 . 027Figure 5—figure supplement 4 . Addition of 1NM-PP1 to Cdk1 ( + ) does not adversely affect Cdc42 Biosensor , Bem1 accumulation , or bud emergence . Fluorescent and phase images depicting representative Cdk1 ( + ) treated with 1NM-PP1 . Each image is 16 . 2 µm x 16 . 2 µm . Strains used: WYK8440 and WYK8301 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 027 To directly test whether Cdk1 activation is required for Bem1 recruitment in response to active Cdc42 , we utilized an allele of Cdk1 , cdk1-as1 , that can be inhibited by an ATP-analog , 1NM-PP1 ( Bishop et al . , 2000; McCusker et al . , 2007 ) . Asynchronous cells were pre-treated with 1NM-PP1 for 20 min and we subsequently monitored for the ability of recruited Cdc24 to induce Cdc42 activation or Bem1 accumulation . We limited our analysis to mother cells with large-budded daughter cells , indicative of mother cells in early G1 . Consistent with previous reports ( McCusker et al . , 2007 ) , cells lacking Cdk1 activity could not undergo bud emergence nor could buds grow significantly ( Figure 5C , D ) . Nevertheless , Cdc24 recruitment induced accumulation of Cdc42 in 67% of Cdk1-inhibited cells ( data not shown ) . Indeed , Cdc42 was activated with the same kinetics irrespective of the presence or absence of the inhibitor ( Figure 5D ) . Strikingly , when Cdk1 activity was suppressed , Bem1 failed to accumulate within the time frame of the experiment ( greater than 1 . 5 hr after addition of 1NM-PP1 ) ( Figure 5C , D , Figure 5—figure supplement 3 ) . Furthermore , when Bem1 was recruited , cells were unable to activate Cdc42 within the time frame of the experiment ( data not shown ) . Cells expressing wild-type cdk1 were unaffected by the addition of 1NM-PP1 ( Figure 5—figure supplement 4 ) . These results demonstrate that Cdk1 activation promotes both Bem1 accumulation and bud emergence even in cells with locally activated Cdc42 . A prominent model for positive feedback posits that the Bem1-GEF complex is constitutive ( Bose et al . , 2001 ) . However , our data indicates that , prior to Cdk1 activation , optogenetically-recruited Cdc24 and Bem1 function differently . The ability of Cdc24 to activate Cdc42 in the apparent absence of Bem1 lacks precedent; therefore we studied the characteristics of these optogenetically-initiated sites . Specifically , we sought to determine the stability of these sites and whether these nascent sites of polarization interact . To answer these questions , we exploited the ability of the optogenetic system to dynamically reposition the site of protein recruitment . During the time window between the transition from isotropic growth and the time at which Cdc24 recruitment triggers Bem1 accumulation ( i . e . ~13 min prior to bud emergence ) , Cdc24 recruitment can induce Cdc42 activation . To test whether these sites are self sustaining , the targets were removed after Cdc42 activation was detected in early G1 . Optogenetic recruitment of ePDZ-tagged proteins is largely reversed within 3 min after discontinuing illumination ( Strickland et al . , 2012 ) , ensuring that any remaining signal is not due to Cdc24-ePDZ remaining at the target site due to the optogenetic tag . Frequently , the region retained a faint but consistent signal of active Cdc42 for up to 30 min and ultimately resulted in budding from the specified site ( Figure 6A , B ) . Conversely , Bem1 recruitment was only sufficient to bias the bud site if it was recruited within 15 min of bud emergence ( Figure 6A , B ) . This observation shows that Cdc42 activity can be maintained at a unique site in the apparent absence of Bem1-mediated positive feedback . Thus optogenetic recruitment of Cdc24 can induce a self-sustaining pool of Cdc42 activation . 10 . 7554/eLife . 26722 . 028Figure 6 . Cdc24 recruitment induces precocious Cdc42 activation and self sustaining Cdc24-tdTomato accumulation . ( A ) Phase and fluorescence images and kymographs of representative cells depicting the response to the transient recruitment of Cdc24-ePDZ or Bem1-ePDZ . Initial target site denoted by white stars; note time of target removal . Each image is 16 . 2 µm x 16 . 2 µm . Strains used: WYK8440 and WYK8308 . ( B ) Polarization efficiency following transient recruitment of Cdc24 and Bem1 . Time indicates when the target was removed relative to bud emergence ( bud emergence occurs at time = 0 ) . Black dots represent polarization of individual cells in response to transient Cdc24 recruitment . Blue dots represent polarization of individual cells in response to transient Bem1 recruitment . Lines represent averages ( +/- SEM ) of data binned in 10 min intervals , with the middle time point represented on the plot . Results are pooled across multiple experiments ( n experiments >= 2; N cells > 10 for each time interval; N total cells > 75 ) . Polarization Efficiencies of Cdc24 at −5 , –15 , and −25 min and Bem1 at −5 and −15 min were statistically significant relative to the corresponding earlier time points . Polarization Efficiencies of Cdc24 and Bem1 are statistically significant at −25 min . p<0 . 05 , Mann-Whitney U test . ( C ) Representative phase and fluorescence images and kymographs showing the position of the laser target and accumulation of Cdc24-tdTomato in response to Cdc24-ePDZ recruitment . Strain used: WYK8575 . ( D ) Accumulation kinetics for Cdc24-tdTomato . Data are combined across multiple experiments ( n experiments > 2 , N total cells > 30 for each condition ) . ( E ) Panels of representative phase and fluorescence images indicating the response of Cdc24-tdTomato to transient Cdc24-ePDZ recruitment . Target removal occurred at −20 min . Orange arrows denote sites of illumination without Cdc24-tdTomato accumulation . Strain used: WYK8575 . ( F ) Stacked bar chart indicating the percentage of cells that maintain Cdc24-tdTomato accumulation in response to Cdc24 recruitment and whether they polarize to the prescribed site . Buds that formed within 45° of the target were defined as budding on target . Data is binned by 10 min time intervals , with the middle time point represented on the plot . Data combined across multiple experiments ( n experiments > 3 , N total cells > 25 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 02810 . 7554/eLife . 26722 . 029Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 02910 . 7554/eLife . 26722 . 030Figure 6—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 03010 . 7554/eLife . 26722 . 031Figure 6—figure supplement 1 . Optogenetic Cdc24 recruitment induces precocious accumulation of Cdc24-tdTomato . Box-and-whisker plot of Cdc24-tdTomato appearance in photo-activated ( red ) and mock-illuminated ( blue ) cells . Outliers are depicted by black squares . Bud emergence occurs at time = 0 . Data as in Figure 6D . . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 03110 . 7554/eLife . 26722 . 032Figure 6—figure supplement 2 . Recruitment of catalytically inactive Cdc24 does not induce Cdc24-tdTomato accumulation . ( A ) Phase contrast and fluorescence images of Cdc24-tdTomato accumulation in response to photo-recruitment of Cdc24 ( GEF-dead ) . Each image is 16 . 2 µm x 16 . 2 µm . Strain used: WYK8504 . ( B ) Polarization efficiency of population of cells where each point represents an individual cell from the condition depicted in A . Data are averages of all cells across multiple experiments ( n experiments >= 2 , N total cells > 20 for each group ) . Average and +/- SEM is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 032 To determine how optogenetically-induced Cdc42 activation could be maintained in the absence of the optogenetic cue , we tested whether recruited Cdc24 induces a change in the localization of cytosolic Cdc24 . To that end , we recruited Cdc24 and visualized Cdc24-tdTomato . Indeed , in 50% of cells in which Cdc24-ePDZ was recruited , Cdc24-tdTomato appeared on the cortex ~25 min before bud emergence , whereas in mock-illuminated cells , Cdc24-tdTomato appeared on the cortex ~15 min before bud emergence ( Figure 6C–D , Figure 6—figure supplement 1 ) . The kinetics of accumulation for Cdc24-tdTomato parallels that of activated Cdc42 in response to Cdc24 recruitment ( Figure 3 ) . Furthermore , catalytically inactive Cdc24-ePDZ was unable to induce Cdc24-tdTomato recruitment ( Figure 6—figure supplement 2 ) . From these data we conclude that recruitment of Cdc24 can induce both Cdc42 activation and Cdc24 recruitment prior to Cdk1 activation , both of which appear to occur independently of Bem1 . Collectively , these results reveal the existence of a second pathway for positive feedback . The observed activation of Cdc42 likely represents a combination of direct Cdc42 activation by optogenetically recruited Cdc24 and its amplification by endogenous mechanisms . Given that Cdc24-tdTomato accumulates at nascent sites , we hypothesized that it would remain at sites after the optogenetic cue was halted by removal of the target . Intriguingly , we observed Cdc24 was maintained at the site for ~6 min following cessation of optogenetic perturbation ( Figure 6E ) . After this 6 min time window , the Cdc24 signal became more punctate and these puncta dynamically associated with the previously targeted region . Greater than 55% of cells budded from the targeted region when target removal occurred ~30 min prior to bud emergence ( Figure 6F ) . These data indicate that optogenetically-initiated sites of Cdc42 activation are stable and are maintained , at least in part , by the accumulation of Cdc24 , without accumulation of detectable Bem1 . To examine whether two sites of active Cdc42 in early G1 cells compete with one another , the target was repositioned within the same cell . Repositioning of the target caused Cdc42 accumulation at a new site and concomitant dissipation from the old site , with Cdc42 activity simultaneously detected at both sites for ~5 min ( Figure 7A , Video 2 ) . However , the site of Cdc42 accumulation could not be continuously repositioned . A qualitative change in behavior occurs prior to bud emergence . When targets are repositioned within ~13 min of bud emergence , accumulation of Cdc42 at the initial site remains and Cdc42 accumulates weakly at the new site . The cell eventually buds from the site where Cdc42 was active at the ~13 min transition point and Cdc42 signal from the alternate site dissipates upon bud emergence ( Figure 7B ) . Our previous results demonstrate that Bem1 begins to accumulate at the targeted site ~13 min before bud emergence . To confirm that the basis for the qualitative switch relied on the ability of Cdc24 recruitment to induce Bem1 accumulation , we performed the same experiment in cells expressing Bem1-tdTomato . As previously shown , Bem1 only detectably accumulated at the target position defined at the ~13 min transition point and it did not accumulate at new sites if the target was repositioned prior to bud emergence ( Figure 7C , D ) . These results indicate that activation of the Bem1-dependent positive-feedback loop stabilizes sites of Cdc42 activation . Furthermore , these data suggest that prior to Cdk1 activation , nascent sites of polarization influence Cdc42 activation at other sites . 10 . 7554/eLife . 26722 . 033Figure 7 . Local Cdc24 recruitment induces precocious activation of Cdc42 that is dynamically maintained in the apparent absence of Bem1 . ( A ) Panels of representative cells depicting the response to dynamically re-positioned Cdc24 recruitment . Upper panels consist of phase contrast and fluorescent images and the lower panel is a kymograph . The laser was moved every 10 ± 2 min throughout the cell cycle , as denoted by the white stars . Each image is 16 . 2 µm x 16 . 2 µm . Strain used: WYK8440 . ( B ) Percentage of targeting events ‘Outcompeted’ or ‘Not Outcompeted’ relative to bud emergence ( Time = 0 ) . Time indicates when the target was moved to a new position relative to bud emergence . Data is binned by 5 min time intervals , with the middle time point represented on the plot . The event was scored as ‘outcompeted’ if Cdc42 activity dissipated from the original position and accumulated at the new position . Conversely , the event was scored as ‘not outcompeted’ if Cdc42 activity remained at the initial position upon target repositioning . Data are combined across multiple experiments ( n total experiments > 2; n targeting events per time interval >10; N total targeting events > 100; N total cells > 20 ) . ( C ) Panels and a kymograph of a representative cell depicting the accumulation of Bem1 in response to dynamically re-positioned Cdc24 recruitment . Strain used: WYK8301 . ( D ) Box-and-whisker plot denoting Bem1 accumulation in CDK1 , cdk1-as + DMSO , and cdk1-as + 1 NM-PP1 cells . ( E ) Time-course images of cdk1-as cells challenged with Cdc24 dynamic reorientation . Panels consist of phase contrast and either Cdc42-GTP or Bem1 accumulation pseudo-colored as a heat map . Cells were treated with either DMSO or 1NM-PP1 as indicated . Strains used: WYK8441 and WYK8442 . ( F ) Quantification of dynamic reorientation in vehicle-treated or 1NM-PP1-treated cdk1-as cells . ( n total experiments > 2; n targeting events per time interval >10; N total targeting events > 100; N total cells > 20 ) Described in B . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 03310 . 7554/eLife . 26722 . 034Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 03410 . 7554/eLife . 26722 . 035Figure 7—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 03510 . 7554/eLife . 26722 . 036Video 2 . Cdc42 activity can be dynamically repositioned in response to mobile sites of Cdc24 recruitment . Representative phase contrast and fluorescent time-lapse images during dynamic repositioning experiments in cells expressing the Cdc42 biosensor . Left panel is the phase image with the position of the target defined by the black circle . Right panel is the Cdc42 biosensor . fps = 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 036 Given that Bem1 accumulation at Cdc24-prescribed sites requires Cdk1 activation , we hypothesized that the dynamic properties of Cdc24-generated sites would persist in the absence of Cdk1 activation . To test this prediction , we repeated the experiment in cells expressing cdk1-as treated with 1NM-PP1 ( Figure 7E ) . Again , limiting our analysis to mother cells with large-budded daughter cells , we found that Cdc42 activation could be dynamically repositioned for >70 min and that Bem1 was not detectably recruited within this time ( Figure 7E , F ) . These results confirm that Cdk1 activation is required for Bem1-mediated positive feedback activity , which functions to establish the axis of polarity . Combined , these results support three conclusions: ( i ) the ability of Cdc24 to induce Bem1 accumulation is cell cycle regulated , ( ii ) once active , the canonical positive feedback loop is highly stable and limits Cdc42 activation at competing sites , and ( iii ) maintenance of active Cdc42 and cytosolic Cdc24 before cell cycle entry appears independent of Bem1; therefore , the GTPase and the GEF participate in a positive feedback mechanism that functions earlier than the canonical Bem1-dependent positive feedback loop . This alternative positive feedback mechanism can be readily competed by a new site of Cdc24 recruitment . However , this activity may play a physiological role in establishing Cdc42 activity before Start in diploid cells . As F-actin has been proposed to play a role in polarity establishment ( Wedlich-Soldner et al . , 2003; Freisinger et al . , 2013; Jose et al . , 2013 ) , we examined whether actin depolymerization affects the response induced by local recruitment of Cdc24 . Cells expressing light-recruitable Cdc24 , Whi5-tdTomato and either the Cdc42 biosensor or Bem1-tdTomato were partially synchronized in G1 using a nocodazole block and release protocol and treated with Latrunculin A ( LatA ) to depolymerize F-actin . As previously shown , actin depolymerization partially inhibited cell polarization ( Jose et al . , 2013 ) . While 88% of cells polarize Cdc42-GTP in the presence of f-actin , only 32% of cells polarize when actin is depolymerized ( Figure 8 ) . Actin depolymerization has a similar effect on the efficiency of Bem1-tdTomato polarization . Drug treatment also slows cell cycle entry as judged by the efficiency of Whi5 exit from the nucleus ( Figure 8—figure supplement 1 ) , indicating that actin depolymerization affects several cellular processes . 10 . 7554/eLife . 26722 . 037Figure 8 . Localized recruitment of Cdc24 can overcome the polarity defect caused by actin depolymerization . Accumulation kinetics for Cdc42-GTP or Bem1 in response to Cdc24 recruitment in the presence or absence of polymerized actin . Cells were scored as ‘polarized’ if they retained accumulation of active Cdc42 or Bem1 for >15 min during the course of the experiment ( >100 min ) . Red lines represent cells exposed to light-induced Cdc24 recruitment . Blue lines represent mock-illuminated cells . Dark lines represent vehicle-treated cells , while lighter-colored lines represent latrunculin A-treated cells ( see schematic ) . Whi5 nuclear exit was used as a cell cycle marker and accumulation of Cdc42-GTP or Bem1 was scored relative to Whi5 exit . Data was binned by 5 min intervals and combined across multiple experiments ( n experiments = 2; N total cells > 20 ) . Stains used: WYK8500 and WYK8502 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 03710 . 7554/eLife . 26722 . 038Figure 8—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 03810 . 7554/eLife . 26722 . 039Figure 8—figure supplement 1 . Loss of f-actin slows cell cycle entry . Whi5 nuclear exit kinetics in Vehicle-treated ( left ) or Latrunculin A-treated ( right ) cells . Green lines refer to cells expressing the Cdc42 biosensor , while purple lines refer to cells expressing endogenous Bem1-tdTomato . Darker-colored lines represent mock-illuminated cells , while lighter-colored lines represent photo-activated cells ( see schematic ) . Whi5 nuclear exit was slowed in LatA-treated cells . Data were binned by 5 min intervals and combined across multiple experiments ( n experiments = 2; N total cells > 20 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 039 Filamentous actin has been suggested to facilitate delivery of Cdc42 to the cortex where it could be activated by Cdc24 ( Wedlich-Soldner et al . , 2003 ) . If true , Cdc24 recruitment would not be predicted to correct the polarization defect . However , light-directed recruitment of Cdc24 dramatically increased the fraction of cells that locally accumulate Cdc42-GTP in the absence of actin . Similarly , Bem1 polarization was rescued by Cdc24 recruitment in cells treated with LatA ( Figure 8 ) . These results suggest that actin depolymerization does not appear to affect the availability of Cdc42 , but rather it impacts the localization of the Bem1-Cdc24 complex , perhaps non-specifically . In the previous experiments , only one Cdc24-targeted site was active at the moment when Cdk1 was activated resulting in activation of Bem1-mediated positive feedback loop . Cdc24 recruitment to multiple sites at this critical time might result in recruitment of Bem1 at multiple sites , leading to the formation of multiple buds . Alternatively , sites may compete with each other as they form resulting in only one site becoming fully established . Therefore , to investigate whether nascent sites interact , we recruited Cdc24 to two sites simultaneously in unpolarized cells . Both sites generated activated Cdc42 ( Figure 9A , Video 3 ) and retained active Cdc42 until ~11 min before bud emergence . Subsequently , Cdc42 activity was limited to only one site ( Figure 9B ) and bud emergence occurred at that site . In a parallel experiment with recruited Cdc24 , we monitored accumulation of Bem1 and observed that it accumulates at only one of the two sites , which invariably predicted the site of bud emergence ( Figure 9A , B , Video 4 ) . Furthermore , despite recruitment of Bem1-ePDZ to two sites simultaneously , in the overwhelming majority of cases ( 31/33 cells ) , Cdc42 activation and Bem1 accumulation occurred at one site , which ultimately defined the nascent bud ( data not shown ) . These results indicate that multiple sites cannot coexist after activation of the Bem1-mediated positive feedback loop even under conditions in which they are simultaneously specified . 10 . 7554/eLife . 26722 . 040Figure 9 . Nascent sites undergo competition to establish a single axis of polarity . ( A ) Representative fluorescence and phase images in response to simultaneous recruitment of Cdc24 to two sites . Top panel depicts Cdc42-GTP response . Bottom panel depicts Bem1 response . Each image is 16 . 2 µm x 16 . 2 µm . Strains used: WYK8440 and WYK8301 . ( B ) Percentage of cells with signal at one or both sites at any given time relative to bud emergence . Dark gray lines depict percentage of cells with activation at two sites simultaneously . Light gray lines represent percentage of cells with accumulation at only a single site . Bud emergence occurs at time = 0 . Data are combined across multiple experiments ( n experiments >= 2; N total cells > 20 for each condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 04010 . 7554/eLife . 26722 . 041Figure 9—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 04110 . 7554/eLife . 26722 . 042Video 3 . Cdc42 activation can occur at two sites simultaneously in early G1Representative phase contrast and fluorescent time-lapse images in cells expressing the Cdc42 biosensor when challenged with two sites of Cdc24 recruitment . Left panel is the phase image with the position of the target defined by the black circle . Right panel is the Cdc42 biosensor . fps = 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 04210 . 7554/eLife . 26722 . 043Video 4 . Bem1 accumulation is limited to a single site of Cdc24 recruitment . Representative phase contrast and fluorescent time-lapse images in cells expressing Bem1-tdTomato when challenged with two sites of Cdc24 recruitment . Left panel is the phase image with the position of the target defined by the black circle . Right panel is the Cdc42 biosensor . fps = 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 043
In this work , we have studied the mechanism for Cdc42 activation throughout G1 . Traditional genetic and cell biological analyses have identified the key components and revealed many behaviors associated with cell polarization , which have led to a set of models that invoke positive feedback ( Wedlich-Soldner et al . , 2003; Irazoqui et al . , 2003; Goryachev and Pokhilko , 2008; Howell et al . , 2012; Howell and Lew , 2012 ) . However , the field has lacked the tools that can directly interrogate the spatio-temporal dynamics of signaling molecules and critically test the models . Here , we used optogenetics to probe the endogenous regulatory mechanisms that control Cdc42 activity . We used low light doses to recruit limited amounts of Cdc24 to reproducibly induce cell polarization; these conditions only resulted in Cdc42 activation at specific cell cycle stages . Rather than examining the direct consequences of experimentally induced GTPase activation ( Wagner and Glotzer , 2016 ) , we sought to generate small perturbations and define the conditions in which those perturbations were amplified by the endogenous pathways . Using this approach , we have demonstrated the existence of a previously postulated positive feedback loop which promotes bud emergence . In addition , we have demonstrated that potent mechanisms ensure that only a single site can undergo positive feedback at a given time . We have also found that the activity of this pathway is temporally confined by cell cycle regulation . Finally , we have demonstrated the existence of a second , previously uncharacterized , positive feedback mechanism that can maintain a focus of Cdc42 activity prior to Cdk1 activation and which may play a role in symmetry breaking polarization . The current model of symmetry breaking polarization posits that a stochastic accumulation of Cdc42-GTP activates a positive feedback loop mediated by the Cla4-Bem1-Cdc24 polarity complex , thereby generating a local focus of activated Cdc42 ( Howell and Lew , 2012 ) . A prediction of that model is that a seed of Cdc42-GTP would be sufficient to define the nascent bud site . We find that local , light-induced recruitment of either the GEF Cdc24 or the scaffold Bem1 is sufficient to bias the presumptive bud site ( Figure 1 ) . Furthermore , the ability of these proteins to induce polarization requires the molecular features necessary to generate active Cdc42; specifically , GEF activity for Cdc24 and Cdc24-binding ability for Bem1 ( Figure 3 and Figure 2 , respectively ) . These findings are consistent with the existing model for symmetry breaking , in which a positive feedback loop modulated by the polarity complex reinforces local activation of Cdc42 to promote bud emergence . We present direct evidence for positive feedback by optogenetically recruiting Bem1 and showing that endogenous Bem1 accumulates in response to that perturbation . Likewise , Cdc24 recruitment generates positive feedback by recruiting additional molecules of Cdc24 ( Figure 6 ) . In both cases , the ePDZ-tagged proteins must be capable of promoting Cdc42 activation to induce accumulation of their untagged counterparts . Unexpectedly , we find that the GEF Cdc24 functions in two distinct positive feedback pathways; the Bem1-dependent loop that is active upon Cdk1 activation , and an earlier , positive feedback mechanism that appears to be Bem1 independent . The molecular mechanism of positive feedback during early G1 remains to be characterized . Other models for symmetry breaking polarization invoke actin-dependent delivery of Cdc42 to the cortex ( Wedlich-Soldner et al . , 2003; Freisinger et al . , 2013; Jose et al . , 2013 ) . While depolymerization of actin does induce a polarity defect , that defect can be readily suppressed by light-mediated recruitment of Cdc24 to the plasma membrane ( Figure 8 ) . This result indicates that in the absence of polymerized actin , Cdc42 still associates with the plasma membrane . Thus , actin filaments are more likely to promote , either directly or indirectly , the localization of the Cla4-Bem1-Cdc24 complex in a subset of cells . Although recruitment of either Bem1 or Cdc24 efficiently biases the site of polarization , at certain stages of the cell cycle the molecular consequences of their recruitment are quite distinct . In late G1 , following commitment to the cell cycle , light-induced recruitment of either Bem1 or Cdc24 promotes accumulation of active Cdc42 , endogenous Bem1 , and cytosolic Cdc24 ( Figure 2 and Figures 3 and 6 , respectively ) . These results confirm prior models for polarization during the interval between Start and bud emergence . However , Bem1 recruitment does not induce precocious Cdc42 activation , Cdc24 accumulation , or endogenous Bem1 accumulation . In contrast , optogenetic recruitment of the GEF Cdc24 induces precocious activation of Cdc42 and accumulation of cytosolic Cdc24 , though neither are sufficient to induce accumulation of endogenous Bem1 prior to Cdk1 activation ( Figures 3 and 5 , respectively ) . Synthetic lethality precluded generation of a conditional Bem1 allele compatible with our experimental approach which would be necessary to directly test the Bem1 independence of Cdc24-mediated activation of Cdc42 during early G1 . Collectively , these results indicate that Cdk1 regulates the Bem1-Cdc24 complex; and , as a consequence of this regulation , previous models do not apply to the period prior to Start . Rather , a second mode of positive feedback operates in this interval . Consistent with that interpretation , we find that Cdc24 , Bem1 , and Cdc42-GTP do not invariably colocalize , with the lowest level of colocalization occurring prior to Cdk1 activation ( Figure 4 ) . We detect Bem1 and Cdc24 in small , mobile clusters that partially overlap in G1 cells . The nature of these clusters requires further analysis; though , oligomerization may contribute to their formation as the Cdc24 DH domain is capable of oligomerization ( Mionnet et al . , 2008 ) and Bem1 contains a PxxP motif and a SH3 domain which interact ( Irazoqui et al . , 2003; Endo et al . , 2003 ) and could mediate intermolecular association . The delay in Bem1-dependent positive feedback relative to Whi5 nuclear exit ( Figure 5 ) suggests that the polarity complex is regulated by Cln1/2 ( Skotheim et al . , 2008 ) . We propose that Cdk1 activity promotes assembly of the Bem1-Cdc24 complex , this interpretation is consistent with the accumulation of Bem1 at the bud neck in early G1 without accompanying Cdc42 activation ( Atkins et al . , 2013 ) ( Figure 4 ) , as well as a Cln2-dependent increase in exchange activity towards Cdc42 ( Howell et al . , 2009 ) . Cdk1 activity may regulate the association of both Cdc24 with Bem1 and Bem1 with Cla4 . If the former were constitutive , then Bem1 recruitment during early G1 should also result in Cdc24 recruitment which would induce Cdc42 activation ( Figure 10 ) . However , Bem1 recruitment in early G1 had no detectable effect on Cdc42 activation . Likewise , if the latter was constitutive , Cdc42 activation during G1 should induce Cla4 recruitment which would be predicted to induce Bem1 recruitment . However , Cdc24 recruitment in early G1 results in Cdc42 activation but does not result in detectable Bem1 recruitment . These results suggest that the canonical polarity complex is not assembled prior to Start . G1 CDK activity is implicated in directly down regulating Rga2 , a GTPase activating protein that plays a role in Cdc42 regulation ( Sopko et al . , 2007; Knaus et al . , 2007 ) . CDK activation at start may therefore promote Cdc42 activation by at least two parallel pathways . 10 . 7554/eLife . 26722 . 044Figure 10 . Working Model for polarity establishment . In early G1 , prior to Cdk1 activation , Cdc24 , Bem1 , and activated Cdc42 individually associate with the plasma membrane , but do not form stable complexes . With some frequency , Cdc24 , and perhaps other Cdc42 GEFs , activate Cdc42 which can participate in a weak positive feedback loop with one or both GEFs . It is unclear whether Cdc42-GTP in early G1 interact with downstream effectors , such as Cla4 , but if it does , it does not initiate Bem1-dependent positive feedback . Following Cdk1 activation , the Cla4-Bem1-Cdc24 complex assembles in late G1 . This complex may amplify the preexisting focus of Cdc42-GTP , and ultimately undergo strong positive feedback to generate a single focus of Cdc42-GTP that subsequently triggers bud emergence . DOI: http://dx . doi . org/10 . 7554/eLife . 26722 . 044 There also appears to be regulation of the positive feedback pathway after bud emergence . In polarized cells , Cdc24 recruitment at high light doses - but not Bem1 recruitment - activates Cdc42 , raising the possibility that the well studied Cdc24-Bem1-Cla4 complex functions as a unit for a limited fraction of the cell cycle surrounding polarization . What could be the function of the secondary positive feedback resulting in Cdc42 activity before Start ? Current models suggest that Cdc42 activation initiates upon Cdk1 activation ( Howell and Lew , 2012 ) . According to this model , stochastic activation of Cdc42 recruits the Cla4-Bem1-Cdc24 complex which amplifies this local inhomogeneity through positive feedback . Alternatively , polarization may be initiated by local association of Cla4-Bem1-Cdc24 via any of its myriad membrane association motifs . Indeed , our results show that local recruitment of members of this complex is indeed sufficient to induce symmetry breaking polarization and override the endogenous landmark-directed pathway . However , because Cdc24 can activate Cdc42 prior to Start , and because active Cdc42 is detected prior to Start ( Figure 4 ) , the site of polarization may not be dictated strictly by molecular noise or stochastic encounters of the Bem1-Cla4-Cdc24 complex with the membrane . Rather , a distinct mechanism ( s ) may exist that seeds the cortex with activated pools of Cdc42 during early G1 which pattern the ‘random’ choice of bud site upon passage through Start ( Figure 10 ) . Indeed , Bud3 , which is a validated Cdc42 GEF , induces an early wave of Cdc42 activation ( Kang et al . , 2014 ) . Though it is expressed in both haploid and diploid cells , Bud3 only impacts bud site selection in haploid cells ( Chant et al . , 1995 ) . The linking of two mechanisms for Cdc42 activation could allow for more rapid axis specification than could be achieved by a reaction-diffusion based mechanism alone ( Goryachev and Pokhilko , 2008; Kang et al . , 2014 ) . Indeed , such linked systems have been observed and modeled in a manner that can account for the speed of polarization events such as cell migration and the fixation of a single axis following an exploratory phase capable of reorientation ( Goryachev and Leda , 2017; Brandman et al . , 2005; Xiong et al . , 2010 ) . Wild-type yeast cells rarely form multiple buds in a single cell cycle , suggesting the existence of mechanisms by which nascent bud sites compete in a winner take all competition . Recent work has shown that prior to bud emergence , cells are capable of forming multiple nascent foci . Yet , by employing mechanism ( s ) that appear to involve negative feedback , only one focus matures into the site of polarization ( Howell et al . , 2012 ) , indicating competition between the nascent foci . Cells have been genetically manipulated to generate multiple buds . For example , expression of activated alleles of Cdc42 can induce the formation of multiple buds ( Caviston et al . , 2002 ) . Similarly , multiple buds result from membrane-tethering either Bem1 or a Cdc24 mutant that is resistant to negative feedback , while also hindering the membrane-cytoplasm exchange of Cdc42 by deleting the facilitating chaperone , Rdi1 . ( Wu et al . , 2015 ) . However , despite the fact that transient optogenetic membrane recruitment of Bem1 or Cdc24 can induce efficient polarization , neither proved robust enough to generate two buds , indicating that the competition mechanism ( s ) are sufficiently strong to extinguish multiple nascent sites so that only a single axis emerges the winner . Two sites containing active Cdc42 could be generated , but we did not detect Bem1 at two sites simultaneously , consistent with models in which the intact Cla4-Bem1-Cdc24 complex is tightly limited ( Wu et al . , 2015 ) . Nascent sites also compete during early G1 , prior to Start . Whereas a pool of Cdc42 can be induced in early G1 and maintained , if Cdc24 is recruited to a new site within the cell , the pool of active Cdc42 at the original site dissipates within a few minutes in response to the new site ( Figure 7 ) . As these sites appear to be Bem1 deficient , this suggests that a distinct mechanism of competition is active during this time . Indeed , whereas optogenetic recruitment of Cdc24 can induce two sites of Cdc42 activation during G1 that can co-exist , once Cdk1 is activated , the competition is more stringent and Bem1 only accumulates to detectable levels at one of the two sites . In this study we show the utility of optogenetics in dissecting cellular processes . The ability to control the spatiotemporal dynamics of signaling molecules allows novel perturbations of cells that can lead to new insights . By exogenously localizing a GEF , we find that yeast cells are capable of activating Cdc42 just after mitotic exit . We also uncovered a cell cycle regulated step in polarity establishment: activation of the canonical positive feedback loop by Cdk1 ( Figure 5 and Figure 6 ) . Furthermore , we discovered a second positive feedback loop that functions independently of the canonical positive feedback mechanism and is active prior to Cdk1 activation ( Figure 6 ) . Finally , because optogenetics allows proteins to be dynamically repositioned in one or more spots we found that yeast cells are remarkably resistant to formation of multiple sites of polarization .
DNA manipulations were simulated with SnapGene ( GSL Biotech ) . Plasmids were generated using a combination of conventional ligation , homologous recombination ( SLICE ) in bacteria ( Zhang et al . , 2012 ) , and Gibson Assembly ( Gibson et al . , 2009 ) . All plasmids were verified by DNA sequencing . All strains ( Supplementary file 1A ) were constructed in the W303 background ( leu2-3 112 ura3-52 can1-100 ade2-1 his3-11 trp1-1 ) . Haploid a and α cells were mated by incubating overnight in 500 µL YPD and then plating on selective media . All integrating plasmids were of the YIplac series . All low copy plasmids were of the pRS series ( Supplementary file 1B ) . Gene deletions were generated by one-step gene disruptions using standard procedure . Yeast were transformed using lithium acetate , single-stranded carrier DNA and polyethylene glycol . All strains were verified by colony PCR . Endogenous genes were epitope tagged by one-step PCR ( Longtine et al . , 1998 ) . PCR products for C-terminal td-Tomato fusions were amplified from a plasmid containing a tdTomato::HIS3MX cassette ( DLB3299 , a generous gift from Danny Lew ) . Cdc24 and Bem1 point mutants were assumed to accumulate at levels comparable to the wild-type proteins; most mutants were previously characterized ( Ito et al . , 2001; Irazoqui et al . , 2003; Butty et al . , 2002 ) , the Dbs:Cdc42 co-crystal structure ( Rossman et al . , 2002 ) was used to identify point mutations to inactivate Cdc24 GEF activity . The endogenous Cdc24 promoter sequence consisted of 600 bp upstream of the cdc24 locus that was amplified from genomic DNA and inserted by Gibson Assembly into Cen plasmids that encoded for either Cdc24-GFP or Cdc24-tdTomato . The cdc28-as1 allele was inserted by linearizing a hygromycin-resistant plasmid containing the cdc28-as1 coding sequence with an AflII restriction enzyme site adjacent to the F88G point mutation ( pKW50 ) ; the plasmid encoding the cdc28-as1 allele was generously provided by Eric Weiss ( pELW886 ) . Cells were grown in the dark at room temperature overnight in SC -His-Leu-Ura-Trp+Ade and diluted to OD600 = 0 . 1–0 . 2 . For optogenetic experiments , cells were treated with 50 nM of ß-estradiol after 2 hr of growth to induce expression of the optogenetic components . After 2 hr of induction , cells were concentrated 10-fold to 20-fold in fresh media + ß-estradiol and prepped for imaging . Cells co-expressing Cdc24-tdTomato and Cdc24-ePDZ , were induced for 90 min to limit deleterious overexpression of the GEF . For experiments less than 2 hr , cells were imaged on a 2 . 5% agar pad soaked in minimal media + ß-estradiol + drug ( where applicable ) for >20 min . For experiments longer than 2 hr , cells were imaged in a CellASIC Onix microfluidic perfusion chamber ( EMD Millipore Corporation ) to provide continuous nutrients . For non-optogenetic experiments , cells were concentrated 10-fold to 20-fold after 2 hr of growth and imaged on 2 . 5% agar pad soaked in minimal media . After 1 . 5 hr of induction with 50 nM ß-estradiol , cdk1-as cells were treated with 75 µM 1NM-PP1 or solvent ( 1% DMSO ) at room temperature , in the dark , and without shaking for 20 min . Subsequently , cells were imaged for >90 min , for a total time in 1NM-PP1 of approximately 2 hr . To synchronize diploid cells in early G1 , exponentially growing cells were treated with 15 µg/ml nocodazole for 2 hr . Cells were induced with 50 nM ß-estradiol and treated with a second dose of nocodazole at 7 . 5 µg/ml . After 1 hr , cells were washed three times with fresh media and released into minimal media + 50 nM ß-estradiol for 30 min . Cells were treated with 100 µM Latrunculin A ( Molecular Probes ) or solvent ( 1% ethanol ) for 20 min and then imaged for 90 min . Cells were imaged on an Axiovert 200M microscope ( Zeiss ) equipped with a spinning disk confocal ( CSU10 , Yokogawa ) , a 20 mW , 561 nm laser ( Cobolt ) , and an electron-multiplying charge-coupled device ( EMCCD ) camera ( Cascade 512B , Photometrics ) using a 63× , 1 . 4 numerical aperture objective ( Zeiss ) . The microscope was controlled using MetaMorph ( Molecular Devices ) . A 550 nm long-pass filter ( Edmund Optics ) was placed in the transmitted light path to avoid photoexciting the LOV domain when using phase optics . A galvanometer-steerable 440 nm dye laser ( Micropoint , Photonics Instruments ) for local photo excitation of Mid2-localized LOVpep . Illumination intensity was controlled by using an adjustable internal attenuator plate and an additional absorptive neutral density filter OD = 1 ( ThorLabs ) in the beam path . For polarization experiments , cells were photo-excited using the Micropoint laser . The coordinates of targeted sites ( x , y , t ) were recorded with each photo-excitation . Following illumination , a confocal tdTomato image and phase contrast image were acquired . For control ( dark state ) experiments , the experiment was performed identically except that the Micropoint laser was off . Exposure times were 500 msec for tdTomato and 100 msec for phase contrast . For non-optogenetic experiments , cells were imaged with a Zeiss Axioimager M1 equipped with a Yokogawa CSU-X1 spinning disk unit ( Solamere ) and illuminated with 50 mW , 488 nm and 50 mW , 561 nm lasers ( Coherent ) . Images were captured on a Cascade 1K electron microscope ( EM ) CCD camera controlled by MetaMorph ( Molecular Devices ) . For maximum intensity Z-projection snapshots , cells were imaged through the center 3 µm at 0 . 25 µm slices . The GFP and tdTomato images were acquired sequentially , followed by a phase contrast image . Maximum intensity projections were generated using Metamorph . Single plane snapshots were acquired at the mid-plane of the cell , with GFP and tdTomato images acquired sequentially . Exposure times were 900 msec for tdTomato , 900 msec for GFP , and 66 msec for phase contrast . Single plane snapshots of a confocal tdTomato image and phase contrast image were acquired at the mid-plane of the cell . Cells were imaged at a rate of once per minute for 3 min , followed by a rate of once per 30 s for 4 min , and finally imaged at a rate of once per minute for 3 min . Exposure times were 700 msec for tdTomato and 66 msec for phase contrast . All images were analyzed in ImageJ ( Schneider et al . , 2012 ) , with custom-written macros . To determine the angle between the laser position and the nascent bud site , we generated kymographs displaying the laser position , the site of bud emergence , and the intensity of the probe . To generate the kymographs , the cell outline was tracked using the plugin JFilament ( Smith et al . , 2010 ) , which created a series of XY-coordinates that was converted into an ROI for each image in the stack . The ROI was then overlaid on its cognate image , linearized , and normalized to 100 pixels in length by six pixels in width . This was repeated iteratively through each slice of the stack to build kymographs . The kymographs were annotated with the center coordinate of each target in each frame of the time-lapse , the time and position of bud emergence , and the position of the previous bud site . Polarization Efficiency was defined as ( 1-2θ/π ) ) where θ is the angle between the site of illumination and the position of the nascent bud . A population measure of polarization efficiency was found by taking the average of the polarization efficiency value for all cells . To quantify the appearance time of a polarity component , the time of bud emergence , and the polarization efficiency , we analyzed time-lapse images as follows: cells were only scored if they underwent both polarization and bud emergence within the time-course of the movie . The analysis was limited to mother cells; cells that budded within the first 20 min were excluded . Whi5 nuclear exit was defined as when nuclear Whi5 signal equaled that of the cytoplasm . To quantify the appearance of weak fluorescent signals , accumulation was scored by blinding the fluorescent image and scoring probe accumulation manually , by eye . To analyze colocalization , cells were separated into cell cycle stages depending on both the bud size and the distribution of the probes . Data were blinded and cells were pulled at random from each cell cycle subset . Puncta with colocalization ( GFP with tdTomato and tdTomato with GP ) were manually counted . For cdk1-as cells , we limited our analysis to large-budded mother cells that accumulated Bem1 or the Cdc42 biosensor to the bud neck , indicative of early G1 cells . A ‘large-bud’ was considered to have an area more than 4 . 5 µm2 . As cdk1-as cells treated with 1NM-PP1 do not undergo bud emergence , to be scored as polarized , they needed to maintain polarization for >15 min . To quantify competition in dynamic reorientation experiments , a targeting event was scored as ‘Outcompeted’ if the Cdc42 biosensor signal disappeared from the initial position and accumulated at the new position within the time that the target was maintained at the new position . The event was deemed ‘Not Outcompeted’ if the Cdc42 biosensor neither disappeared from the initial position nor accumulated at the new position . | Living cells are not always symmetrical . Instead they are often polarized , with a distinct front and back or top and bottom . Cell polarization influences many processes , including how a cell moves and grows , and where it will divide . Breaking symmetry – in other words , making one part of a cell different from the rest – regularly involves a small protein called Cdc42 , which can switch between an active and inactive form . This protein is found in a range of organisms from fungi to animals . Budding yeast is a valuable model to study cell polarization . This single-celled fungus polarizes in order to produce a daughter cell or ‘bud’ that emerges out of one end of the mother cell . To become polarized , the mother cell accumulates active Cdc42 in a small area of the cell membrane . This region then becomes the front of the cell , from where the future bud will emerge . However , it is not fully understood how active Cdc42 accumulates at only one place . One model proposed that some molecules of active Cdc42 that are already present on the membrane , recruit polarity proteins that in turn activate other , inactive Cdc42 molecules . This self-amplifying loop could eventually build up a local pool of active Cdc42 . However , it has proved challenging to directly test this model . Optogenetics is a technique in which a beam of light is used to manipulate proteins inside cells in a precise manner . The method was first developed in the field of neuroscience over a decade ago , and has more recently been applied to cell biology . Now , Witte et al . have used optogenetics to move polarity proteins to defined places on the membrane of yeast cells and analyse how this change affected the pattern of Cdc42 activation . The experiments showed that polarity proteins not only activate Cdc42 but they also recruit more polarity proteins to the same place . The resulting positive feedback loop leads to active Cdc42 accumulating at one site on the membrane . Further work showed that this mechanism only operates in this manner just before the mother cell replicates its DNA , which is when a yeast cell will normally polarize . These results provide a new perspective on how cells can make one part of the cell different from the rest . Beyond yeast , cell polarization plays a major role when animals , including humans , are developing as embryos or healing wounds . These processes are often controlled by a protein that is equivalent to Cdc42 or by other highly related switch-like proteins . This means that yeast will continue to provide a useful model to study these processes in the laboratory . Lastly , the optogenetics approach developed Witte et al . will be useful to dissect other processes that involve molecules being located at specific place in a cell at specific time . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2017 | Cell cycle entry triggers a switch between two modes of Cdc42 activation during yeast polarization |
In FoF1-ATP synthase , proton translocation through Fo drives rotation of the c-subunit oligomeric ring relative to the a-subunit . Recent studies suggest that in each step of the rotation , key glutamic acid residues in different c-subunits contribute to proton release to and proton uptake from the a-subunit . However , no studies have demonstrated cooperativity among c-subunits toward FoF1-ATP synthase activity . Here , we addressed this using Bacillus PS3 ATP synthase harboring a c-ring with various combinations of wild-type and cE56D , enabled by genetically fused single-chain c-ring . ATP synthesis and proton pump activities were decreased by a single cE56D mutation and further decreased by double cE56D mutations . Moreover , activity further decreased as the two mutation sites were separated , indicating cooperation among c-subunits . Similar results were obtained for proton transfer-coupled molecular simulations . The simulations revealed that prolonged proton uptake in mutated c-subunits is shared between two c-subunits , explaining the cooperation observed in biochemical assays .
FoF1-ATP synthase ( FoF1 ) is a ubiquitous enzyme that synthesizes or hydrolyzes ATP coupled with proton translocation at the inner mitochondrial membrane , chloroplast thylakoid membrane , and bacterial plasma membrane ( Boyer , 1997; Walker , 2013; Yoshida et al . , 2001 ) . FoF1 synthesizes ATP via rotation of the central rotor driven by the proton motive force across the membrane . The enzyme comprises two rotary motors that share the rotor , that is , the water soluble F1 , which has catalytic sites for ATP synthesis/hydrolysis ( Noji et al . , 2017 ) , and the membrane-embedded Fo , which mediates proton translocation ( Kühlbrandt , 2019 ) . The Fo motor consists of a c oligomer ring ( c-ring ) , which serves as the rotor , and the ab2 stator portion located on the c-ring periphery . Downgradient proton translocation through Fo drives rotation of the central rotor composed of a c-ring and γε subunits , thereby inducing conformational changes in F1 that result in ATP synthesis . Conversely , ATP hydrolysis in F1 induces reverse rotation of the rotor , which forces Fo to pump protons in the reverse direction . The c-ring is composed of 8–17 c-subunits depending on the species ( Watt et al . , 2010; Stock et al . , 1999; Mitome et al . , 2004; Meier et al . , 2005; Matthies et al . , 2009; Vollmar et al . , 2009; Pogoryelov et al . , 2009 ) . FoF1 from thermophilic Bacillus PS3 and yeast mitochondrial FoF1 contain 10 c-subunits in the c-ring , which is designated the c10-ring ( Stock et al . , 1999; Mitome et al . , 2004; Symersky et al . , 2012; Guo et al . , 2019; Figure 1 ) . The Fo-c subunit harbors an essential proton-binding carboxyl group ( c-Glu; cE56 in Bacillus PS3 , cE59 in yeast mitochondria ) located near the center of the membrane-embedded region; this group functions as the proton carrier ( Figure 1a ) . Protonation in the Glu allows the c10-ring to bind a proton , whereas proton release leads to Glu deprotonation . Accordingly , bacterial FoF1 activity is significantly decreased when the corresponding key residue is modified by the inhibitor N , N-dicyclohexylcarbodiimide ( DCCD ) ( Hermolin and Fillingame , 1989 ) or mutated to other amino acids ( Dmitriev et al . , 1995 ) , and Bacillus PS3 FoF1 carrying a single cE56Q mutation in the c10-ring does not catalyze ATP-driven proton pumping or ATP synthesis ( Mitome et al . , 2004 ) . The a-subunit comprises two separate half-channels , one connecting the c-ring to the periplasm side of the bacteria or the intermembrane space side of the mitochondria , the other connecting the c-ring to the cytoplasmic side of the bacteria or the matrix side of the mitochondria ( Figure 1b ) . Recent cryo-electron microscopy ( EM ) structural analyses of FoF1 at near-atomic resolution ( Allegretti et al . , 2015; Zhou et al . , 2015 ) have revealed two long tilted parallel α-helices in the a-subunit at the interface with the c10-ring . An essential Arg residue ( aR169 in Bacillus PS3 , aR176 in yeast mitochondria ) at the middle of the long parallel helices plays a critical role in separating the two half-channels by preventing proton leakage ( Mitome et al . , 2010 ) , and in the half-channels , two highly conserved Glu residues ( aE223 and aE162 in yeast mitochondria ) are regarded as proton-relaying sites ( Srivastava et al . , 2018; Figure 1a ) . Since the essential Arg ( a-Arg ) localizes near c-Glu in the c10-ring , the attractive interaction between a-Arg and deprotonated c-Glu is hypothesized to also contribute to Fo rotation ( Vik and Antonio , 1994 ) . In the Fo rotation models proposed based on experimental studies ( Vik and Antonio , 1994; Elston et al . , 1998; Kubo et al . , 2020 ) , the c-subunits facing the a-subunit perform three functions ( proton release , electrostatic interaction with a-Arg , and proton uptake ) depending on their positions relative to the a-subunit . A high-resolution structure analysis of yeast mitochondrial FoF1 showed four of the 10 c-subunits to be facing the a-subunit ( Srivastava et al . , 2018 ) . Three key residues , that is , aGlu162 , aR173 , and aGlu223 , localize between the c-Glu residues of the four c-subunits , suggesting that the c-Glu residues of adjacent c-subunits could cooperate through the a-subunit residues . A more recent theoretical study using a hybrid Monte Carlo/molecular dynamics ( MC/MD ) simulation based on a high-resolution structure showed that there can be two or three deprotonated c-Glu residues facing the a-subunit concurrently ( Kubo et al . , 2020 ) . This suggests that the waiting time for protonation of c-subunits is shared among two or three c-subunits . However , the relationship between a shared deprotonation time among multiple c-subunits and their cooperation in proton transport remain to be characterized . To directly investigate the cooperation among the c-subunits in the c10-ring , we used a genetically fused single-chain c-ring and analyzed the function of Bacillus PS3 FoF1 carrying hetero cE56D mutations . Biochemical assays showed that the ATP synthesis activity was reduced , but not completely inhibited , by a single cE56D mutation , and that it was further reduced by double cE56D mutations . Importantly , across all five double mutants , the activity tended to decrease further as the distance between the two mutation sites increased . To clarify the underlying molecular mechanisms , we performed proton transfer-coupled MD simulations of Fo , in which the mutations were mimicked , reproducing the characteristics of the biochemical experiment . From the analysis of the simulation trajectories , we found that prolonged duration times for proton uptake in the two mutated c-subunits can be shared . As the distance between the two mutation sites increases , the degree of time-sharing decreases . Taken together , these results reveal the functional coupling between neighboring c-subunits .
To investigate potential cooperation among the c-subunits in the c10-ring rotation driven by proton translocation , we generated FoF1 mutants harboring a hetero-mutated c10-ring from thermophilic Bacillus PS3 . We previously produced a fusion mutant , c10 FoF1 , in which 10 copies of the Fo-c subunit in the c10-ring were fused into a single polypeptide , and demonstrated that c10 FoF1 was active in proton-coupled ATP hydrolysis/synthesis ( Mitome et al . , 2004 ) . Starting with c10 FoF1 , we generated six mutant FoF1s harboring one or two hetero cE56D-mutated c-subunits . The single mutant carries a cE56D mutation in the c ( e ) -subunit ( dtesignated as mutant “e” ) , whereas the five double mutants , “ef , ” “eg , ” “eh , ” “ei , ” and “ej , ” harbor two cE56D mutations , each with its respective c-subunit ( Figure 1b ) . FoF1 mutants carrying one or two cE56D substitutions in the c10-ring were expressed in host Escherichia coli cell membranes at approximately one-tenth the level of wild-type ( WT ) FoF1 . Western blotting with anti-c-subunit antibodies showed c10-subunit expression in all mutants ( Figure 2a ) . Unlike in the WT , there was no band of monomer c-subunits in the mutants , and relatively stronger bands were seen at the position of the c10 subunit , indicating that the c10 subunit of the mutants was expressed in the membrane ( Figure 2a ) . First , ATP synthesis activity was measured using inverted membrane vesicles containing mutated FoF1s ( Figure 2b ) . The activity of mutant “e” was reduced to 35 . 6±8 . 8% of that of FoF1 , with fusion mutation only . The activity of the five double mutants with fusion mutation only was lower than that of the single mutation ( “ef”: 22 . 3±9 . 3%; “eg”: 18 . 8±8 . 8; “eh”: 13 . 0±8 . 9%; “ei”: 14 . 4±6 . 7%; “ej”: 12 . 0±4 . 7% ) . The ATP synthesis activity of mutant “ef” was significantly higher than that of mutants “eh , ” “ei , ” and “ej”; the corresponding p-values were 0 . 0357 , 0 . 0435 , and 0 . 0122 , respectively ( Table 1 ) . Since the ATP synthesis activity tended to decrease further as the distance between two mutation sites increased , a regression analysis among double mutations was performed between the distance between mutations , indicated by the number of c-subunits ( ef=1 , eg=2 , eh=3 , ei=4 , and ej=5 ) , and the ATP synthesis activity of these mutants . The regression confirmed that the ATP synthesis activity significantly decreased as the distance between the two mutations increased ( p=0 . 0039 ) . For comparison , the c10 ( E56Q ) -FoF1 with only a single E56Q mutation introduced into the first hairpin unit of the c10 did not catalyze ATP synthesis ( Mitome et al . , 2004 ) . Next , ATP-driven proton pump activity was assessed as a measure of the quenching of the fluorescence of 9-amino-6-chloro-2-methoxyacridine ( ACMA ) caused by proton influx into the inverted membranes ( Figure 2c ) . Mutants “e , ” “ef , ” and “eg” showed proton pumping , indicated by the slow quenching after ATP addition , while mutants “eh , ” “ei , ” and “ej” did not show any pumping . The proton pump activity of the single mutant “e” was higher than that of the double mutants “ef” and “eg . ” Thus , proton pump activity was high in the double mutants “ef” and “eg , ” in which the two mutations were located close to each other , but low in “eh , ” “ei , ” and “ej , ” in which the mutations were introduced farther apart . Although the mutant FoF1s showed ATP hydrolysis activity , approximately 90% of the activity was insensitive to DCCD , a compound that inhibits Fo ( Table 2 ) . DCCD-insensitive ATP hydrolysis indicates uncoupled FoF1 activity . All mutants showed 10–15% DCCD-sensitive ATP hydrolysis activity . Thus , a subtle difference in the structure of the proton-binding site induced by the cE56D mutation may have conferred resistance to DCCD binding or caused uncoupling . The rotation driven by ATP hydrolysis was affected to a greater extent by the threefold symmetry structure of F1 than by the rotation during synthesis , and the DCCD-sensitive ATP hydrolysis activity indirectly reflected the function of Fo . Biochemical assays showed that the decreased rotation speed of the double-mutant Fo motor depended on the distance between the two mutation sites; however , the underlying mechanism remains to be elucidated . To obtain mechanistic insights , we tested the mutated Fo motor rotations by proton transfer-coupled molecular simulations ( Kubo et al . , 2020 ) . Based on our previous simulation setup for the WT yeast mitochondrial Fo , we introduced the cE59D mutation in silico to one and two c-subunits corresponding to the biochemical assays ( see Materials and methods for more details ) . First , we demonstrated 10 trajectories for the single mutant “e” ( Figure 3a ) . Although the mutated c10-ring paused for a long period , the mutants still rotated in the synthesis direction , coupled with proton transportation . Next , we simulated all five double mutants ( “ef , ” “eg , ” “eh , ” “ei , ” and “ej” ) and calculated the average rotational velocities over 10 trajectories ( Figure 3b ) . Figure 3b shows the mean values and standard errors of the rotational velocities of the WT and all mutants . The rotational velocity of mutant “e” is almost two times slower than that of the WT . The rotational velocities of double mutants tend to decrease as the distance between the mutated chains increases . Thus , we were able to capture the characteristics of the experimental results in our simulations qualitatively , but not quantitatively . We then evaluated the molecular processes for the simulation . Each cE59 ( or cE59D ) is protonated when the corresponding c-subunit is far from the a-subunit . This is regarded as the resting state of cE59 ( Figure 4a ) . As counterclockwise rotation occurs , the c-subunit approaches the half-channel of the a-subunit , which is connected to the matrix ( the matrix half-channel ) . When cE59 comes close to aE162 , which is the relaying site to the matrix half-channel , proton transfer from cE59 to aE162 occurs via the Monte Carlo step . Depending on the transfer efficiency , several Monte Carlo steps may be required to achieve proton release from cE59 . We define the time from the first trial of the cE59-to-aE162 proton transfer to the success of transfer as “the duration for proton release” ( indicated in pink in Figure 4a ) . Once cE59 is deprotonated , the corresponding c-subunit can rotate counterclockwise further into the a-subunit facing region . After some rotation , the c-subunit approaches the other half-channel connected to the IMS ( the IMS half-channel ) . When cE59 comes close to aE223 , which is the relaying site for the IMS channel , cE59 attempts to take up a new proton from aE223 via the Monte Carlo step . We define the time from the success of proton release to the arrival at the rotation angle for proton uptake as “the duration for the deprotonated rotation” ( indicated in green in Figure 4a ) . Again , several Monte Carlo steps may be required to achieve this proton uptake . We define the time from the arrival at the proton uptake angle to success of proton uptake as the “the duration for proton uptake” ( blue in Figure 4a ) . Then , the c-subunit returns to the resting state . Thus , the entire time could be divided into three stages: stage 1 , the duration for proton release; stage 2 , the duration for deprotonated rotation; and stage 3 , the duration for proton uptake , in addition to the resting time . Note that these durations are defined for each c-subunit and that the durations in one c-subunit overlap with durations in other c-subunits . For each mutant and for the WT , for each of the 10 c-subunits , we analyzed these three durations . First , we examined the time course of a representative trajectory and the average durations for the WT for each c-subunit ( Figure 4b and c ) . The average durations for stages 1 , 2 , and 3 were approximately 500 , 1100 , and 200 MD frames , respectively . As expected , there were no significant differences in the durations among the 10 c-subunits . Next , we performed the same analysis using the single mutant “e” ( Figure 4d and e ) . We found that the mutation in the c ( e ) -subunit clearly affects the duration for this subunit; stages 1 and 2 did not differ much from those in the WT , whereas the duration of stage three was much longer than that in the WT , since cE59D has a lower rate of proton transfer , and since the pKa value of cE59D is lower than that of cE59 . The increased duration of stage two in the c ( g ) -subunit , and that of stage three in the c ( h ) - and c ( i ) -subunits , were caused by the delay in proton uptake of the c ( e ) -subunit . Because the c ( e ) -subunit scarcely received protons from the IMS channel , and because the c ( e ) -subunit often stopped around the IMS channel , the c ( g ) -subunit can rarely overcome the a-Arg barrier , thus increasing the duration of stage two in the c ( g ) -subunit . Similarly , the c ( h ) - and c ( i ) -subunits spend most of their time in the membrane and are unlikely to pass a proton to the a-subunit , but they approach the matrix channel momentarily in fluctuations and try to release a proton to the a-subunit . However , the acceptance ratio is so small that the duration of stage three increased . We then analyzed double mutants . For the “ef” mutant ( Figure 4f and g ) , similarly to that for the “e” mutant , the cE59D mutation in the c ( e ) -subunit prolongs the duration for proton uptake . Additionally , mutation in the c ( f ) -subunit also prolongs the duration for proton uptake . Interestingly , as shown in Figure 4f , these prolonged durations in c ( e ) - and c ( f ) -subunits are shared . Thus , by overlapping the delayed steps , the overall slowdown in the “ef” double-mutant system is lower than that if the effects of the two mutations were independent . In other words , sharing the delayed times of multiple subunits reduces the overall delay . In comparison , we examined the durations for the “ej” double-mutant ( Figure 4h ) . As expected , mutations in the c ( e ) - and c ( j ) -subunits slow proton uptake in these subunits , although the durations are not shared . Therefore , we expect that there is no coupling between the c ( e ) - and c ( j ) -subunit mutations resulting in additive effects of the two mutations . In summary , coarse-grained MD simulations qualitatively reproduced the effects of the single and double mutants found in biochemical assays and provided molecular interpretations of the coupling between two mutations . When the two mutations are in distant subunits of the c-ring , the effects of the two mutations are additive . In contrast , two mutations in neighboring subunits can result in overlapping of delays by the two mutations , leading to reduced effects of the two mutations .
In this study , we determined whether c-subunits function in a cooperative manner for the rotation of the FoF1 c10-ring and assessed the mechanistic role of c-Glu ( cE56 ) in this cooperation . We have demonstrated that the degree of cooperation between two c-subunits depends on the distance between the cE56D hetero-mutations at the proton-carrying site . The activity of FoF1 was significantly decreased , but not completely abolished , by a single cE56D mutation . This activity was further decreased by the second cE56D mutation; moreover , the activity was high when the two mutations were introduced into nearby c-subunits , and the activity decreased as the distance between the two mutations increased . To the best of our knowledge , this is the first study providing unambiguous evidence for the coupling between two c-subunits . Molecular simulations reproduced the major features of biochemical experiments on single and double mutants and further revealed the molecular mechanisms of the coupling . Sharing of the prolonged durations by mutations in neighboring c-subunits leads to coupling . When the cE56D substitution was introduced in one of the c-subunits , ATP synthesis activity decreased substantially . In E . coli FoF1 , ATP-driven proton pump activity was reported to be decreased after substitution of the conserved cAsp61 residue with Glu ( Miller et al . , 1990 ) . Here , after cE56D substitution , we detected partial retention of not only proton pump activity but also of ATP synthesis activity . In contrast , cE56Q substitution in one of the c-subunits was found to eliminate ATP synthesis activity , ATP-driven proton pump activity , and DCCD-sensitive ATP hydrolysis activity ( Mitome et al . , 2004 ) . In this study , ATP synthesis activity and ATP-driven proton pump activity were not completely lost when the carboxyl group of Glu was replaced with that of Asp . A comparison of this result with the cE56Q substitution results suggested that the presence of a carboxyl group capable of undergoing protonation and deprotonation is critical for rotation in the ATP synthesis direction coupled with proton transfer , and for the proton-transfer-coupled rotation induced by ATP hydrolysis . As changing the Glu side chain to an Asp side chain decreased activity , we concluded that subtle structural differences in the proton-binding site caused by the one-methylene-group difference in the sidechain length , together with the change in pKa , slowed the elementary process required for driving rotation . In the FoF1s , carrying the cE56D mutation in two c-subunits , ATP synthase activity was high when the two introduced Asp residues were close to each other , and the activity decreased as the distance between the two mutations increased . If the kinetic bottleneck in the c10-ring rotation was only in one step of one c-subunit , the same activity would appear among double mutations with different relative separations . Alternatively , even if the c-subunit plays multiple roles , if each role works independently , the same activity would be obtained , irrespective of the mutational position . However , the experimental results showed that the activity was decreased when the two mutations were introduced farther apart . Thus , the data unambiguously indicate that the kinetic bottleneck in the c-ring rotation contains multiple c-subunits . According to previously proposed models , proton release at c-Glu , electrostatic interaction between a-Arg and c-Glu , and proton binding at c-Glu drive c-ring rotation ( Vik and Antonio , 1994; Elston et al . , 1998 ) . Moreover , based on the crystal structure of mitochondrial FoF1 , the c-subunits that face the a-Glu223 residue bridging a proton from IMS , the a-Arg residue involved in electrostatic interaction , and the a-Glu162 residue bridging a proton to the matrix are located apart; therefore , we hypothesized that the c-subunits on the a-Glu223 side of a-Arg play a role in proton release , whereas the c-subunits on the a-Glu162 side of a-Arg play a role in proton uptake in the ATP synthesis rotation . MC/MD simulations based on the FoF1 atomic structure have revealed that proton transfer causes c10-ring rotation ( Kubo et al . , 2020 ) . Here , MD simulation of c10-ring rotation during ATP synthesis was performed based on the aforementioned hypothesis , that proton release and proton uptake are both affected by cE56D mutation . Our results indicated that the rotation speed is higher when the mutation is introduced at adjacent positions , and that the rotation speed decreases as the distance between the two mutants increases . These results , which are consistent with the findings of our biochemical experiments , indicate cooperative proton uptake during the rotation of the c10-ring . Further analysis revealed that the waiting times for proton uptake in multiple subunits are shared . However , as the distance between the mutations increases , the degree of sharing of waiting time decreases , resulting in lower rotation speeds . Overall , these findings suggest that at least three of the c-subunits on the a/c interface cooperate during c10-ring rotation in Fo . This is consistent with the presence of two or three deprotonated carboxyl residues facing the a-subunit in the MC/MD simulation of WT FoF1 ( Kubo et al . , 2020 ) . In the WT , the c-subunit with deprotonated cE56 is considered to be the c-subunit waiting for proton uptake during ATP synthesis . Since the WT prefers pathways with two or three c-subunits waiting to uptake protons rather than only one c-subunit , the waiting time for proton uptake can be shared between two or three c-subunits . With respect to double mutation activity , the “ef” and “eg” mutants tended to have higher activity than the “eh–ej” mutants . This is consistent with the upper limit of three deprotonated c-subunits obtained from the WT simulations . The waiting time for protonation can be shared among three of sequential c-subunits of c ( e ) , c ( f ) , and c ( g ) , but if they are located more than four subunits apart , the waiting time cannot be shared . Therefore , the activity of mutants “eh , ” “ei , ” and “ej” was lower than that of mutants “ef” and “eg . ” One limitation of this study is that we used the fusion mutation and the cE56D mutation . These mutations may affect not only our hypothesized driving force but also other activities . However , we consider our interpretations of the results to be valid based on the comparison with the results of the same mutation combination , and the results of MD simulations . Second , our MC/MD model includes only the a-subunit and c10-ring , whereas naturally occurring FoF1 also contains F1 and the b-subunit . As F1 exhibits threefold symmetry , which is mismatched with the tenfold symmetry in the c10-ring , the entire FoF1 is expected to exhibit more complex and asymmetric behaviors , which can represent a direction for future investigation of the enzyme .
Plasmids for the FoF1 mutants were generated from pTR19-ASDS ( Suzuki et al . , 2002 ) using the megaprimer method , and were then used for the transformation of a Fo-deficient E . coli strain , JJ001 ( Jones and Fillingame , 1998 ) . A plasmid for expressing the FoF1 mutant harboring a substitution of Fo-c Glu-56 with Asp ( cE56D ) was prepared from pTR19-ASDS ( Suzuki et al . , 2002 ) using the megaprimer method; this yielded pTR19-CE56D . The cE56D mutation sequence was verified through DNA sequencing . FoF1 carrying a hetero-mutation of cE56D in a fused c10-subunit prepared using Gly-Ser-Ala-Gly linkers ( Mitome et al . , 2004 ) was generated as follows . Briefly , an AvrII restriction site was introduced immediately after the initial c-subunit codon in the pTR19-CE56D expression plasmid , and new NheI and SpeI sites were introduced at downstream sites in the Fo-c gene ( to obtain pTR19-ACE56DN ) ; pTR19-ACE56DN was digested with EcoRI and NheI , and the 1 . 3 kb EcoRI-NheI fragment was ligated into an EcoRI-AvrII site in pTR19-AC1N or pTR19-ACE56DN ( to obtain pTR19-AC2DE or pTR19-AC2DD ) . Next , pTR19-AC2DE was digested with EcoRI and NheI , and the EcoRI-NheI fragment was ligated into an EcoRI-AvrII site in pTR19-AC1N or pTR19-ACE56DN ( to obtain pTR19-AC3DEE or pTR19-AC3DED ) . By using this procedure , uncE genes were singly fused to generate plasmids expressing six FoF1s containing tandemly fused decamers carrying the cE56D mutation at the first hairpin ( mutant “e” ) , first and second hairpins ( ef ) , first and third hairpins ( eg ) , first and fourth hairpins ( eh ) , first and fifth hairpins ( ei ) , and first and sixth hairpins ( ej ) . The multimer uncE genes of the mutants were verified through plasmid restriction mapping . Plasmids generated for the WT and mutant FoF1s were singly expressed in Fo-deficient E . coli strain JJ001 ( pyrE41 , entA403 , argHI , rspsL109 , supE44 , uncBEFH , recA56 , and srl::Tn10 ) ( Jones and Fillingame , 1998 ) . Transformants were cultured , and membrane vesicles were prepared as previously described ( Mitome et al . , 2004 ) . ATPase activity was measured using an ATP-regenerating system at 37°C in 50 mM Hepes-KOH buffer ( pH 7 . 5 ) , containing 100 mM KCl , 5 mM MgCl2 , 1 mM ATP , 1 μg/ml FCCP , 2 . 5 mM KCN , 2 . 5 mM phosphoenolpyruvate , 100 μg/ml pyruvate kinase , 100 μg/ml lactate dehydrogenase , and 0 . 2 mM NADH ( Mitome et al . , 2004 ) . One unit of activity was defined as hydrolysis of 1 μmol of ATP per minute; the slopes of decreasing 340 nm absorbance in the steady-state phase ( 400–600 s ) were used for calculating activity . The sensitivity of ATP hydrolysis activity to DCCD-induced inactivation was analyzed as previously reported ( Noji et al . , 2017 ) . The ATP hydrolysis activity in the presence of 0 . 1% lauryldimethylamine oxide was measured to estimate the amount of FoF1 in the membrane vesicles . ATP-driven proton pump activity was measured as the fluorescence quenching of ACMA ( excitation/emission: 410/480 nm ) at 37°C in 10 mM Hepes-KOH ( pH 7 . 5 ) , 100 mM KCl , and 5 mM MgCl2 , supplemented with membrane vesicles ( 0 . 5 mg protein/ml ) and ACMA ( 0 . 3 μg/ml ) ( Mitome et al . , 2004 ) . The reaction was initiated by adding 1 mM ATP , and quenching reached a steady level after 1 min; after 5 min , FCCP ( 1 μg/ml ) was added , and fluorescence reversal was confirmed . The magnitude of fluorescence quenching at 3 min relative to the level after FCCP addition was recorded as the proton pump activity . ATP synthesis activity was measured at 37°C using luciferase assays as previously described ( Mitome et al . , 2017; Suzuki et al . , 2007 ) . After incubating for 5 min at 37°C , we poured 1 . 6 ml PA3 buffer ( 10 mM Hepes-KOH [pH 7 . 5] , 10% glycerol , and 5 mM MgCl2 ) , 2 . 5 mM KPi ( pH 7 . 5 ) , 0 . 53 mM ADP ( Calbiochem , San Diego , CA ) , 26 . 6 μM P1 , P5-di ( adenosine-5ʹ ) pentaphosphate ( Sigma-Aldrich , St . Louis , MO ) , 50 μg/ml inverted membranes , and 0 . 125 volumes CLS II solution ( ATP Bioluminescence Assay Kit CLS II; Sigma-Aldrich ) into the cuvettes; 0 . 5 mM NADH was added after starting the measurement . Synthesized ATP amounts were calibrated using a defined amount of ATP at the end of the measurement . FCCP addition was confirmed to prevent ATP synthesis . Specific activity was calculated as follows . Using the data of the time course of the luminescence of luciferin-luciferase , the slope for 90 s before the addition of NADH was subtracted from the slope for 50 s after the addition of NADH . Defined amount of ATP was repeatedly added four times at 20-s intervals , and the average increase in luminescence per ATP amount over those times was calculated as the difference between the average luminescence for 5 s immediately after ATP addition and the average luminescence for 5 s immediately before ATP addition . The standard deviation and standard error were calculated . ATP synthesis activity was calculated as follows: the slope of ATP synthesis deducted at the baseline was divided by the increase in luminescence per amount of ATP and membrane protein ( mg ) added . Error propagation processing was performed to calculate the standard deviation . To estimate the amount of ATP synthase in the inverted membrane vesicle , the specific ATPase activity of the inverted membrane vesicles in 0 . 1% LDAO was calculated using the ATP regeneration system from the slope at 240–300 s after the addition of LDAO . Mean values and standard deviations were calculated from three or four data sets of each lot of the inverted membrane vesicles of mutants . Since the expression level of ATP synthase differs depending on the mutant , the ATP synthesis activity of the inverted membrane assuming the same expression level of ATP synthase should be estimated . The “ei” and “ej” mutants showed relatively low ATP hydrolysis activity in the presence of LDAO corresponding to the expression level . The ATP synthesis activity was calculated assuming that the expression level corresponded to the average of the four mutants , “e , ” “ef , ” “eg , ” and “eh . ” In the actual calculation , the ATP synthesis activity of the inverted membrane of each mutant was divided by the ATP hydrolysis activity of the mutant in the presence of LDAO , and the average of the ATP hydrolysis activity of the “e , ” “ef , ” “eg , ” and “eh” mutants in the presence of LDAO was multiplied . Error propagation processing was performed for calculating the standard deviation and standard error . Mean values and standard deviations were calculated from nine datasets of three lots of mutants c10 , “e , ” and “eg , ” from 10 data sets of three lots of mutants “ef , ” “eh , ” and “ei , ” and from 8 data sets of three lots of mutant “ej . ” Error propagation processing was performed for standard deviation and standard error calculation . The unbiased estimate of variance was calculated from the standard deviation , and the Student’s t-test was performed between the two mutants using the mean value and the unbiased estimate of variance to calculate the p-value . Protein concentrations were determined using a BCA Assay Kit ( Thermo Fisher Scientific , Waltham , MA ) , with bovine serum albumin serving as a standard . Membrane vesicles were separated using sodium dodecyl sulfate polyacrylamide gel electrophoresis ( SDS-PAGE ) with 15% gels containing 0 . 1% SDS , and proteins were stained with Coomassie Brilliant Blue R-250 . FoF1 expression was confirmed by immunoblotting with anti-β and anti-c polyclonal antibodies for FoF1 from the thermophilic Bacillus PS3 . To represent the proton transfer-coupled rotational motion of the c10-ring , protein motion and proton jump were modeled using MD and MC , respectively , and these dynamics were combined to reproduce c10-ring rotational motion with proton hopping ( Kubo et al . , 2020; Figure 3—figure supplement 1a ) . In our simulation system , we included the a-subunit and c10-ring ( Figure 1a ) structure models of yeast Fo based on the cryo-EM structure of a yeast mitochondrial ATP synthase ( PDB ID: 6CP6 ) ( Srivastava et al . , 2018 ) . The entire energy function was defined asVtotal ( R , H+ ) =Vnon−es ( R ) +Ves ( R , H+ ) +Vpka ( H+ ) where R represents all the coordinates of the protein , H+ collectively represents the protonated state of 12 protonatable sites . The first term on the right-hand side , Vnon-esR , is the whole-protein energy without the electrostatic interactions . We used the AICG2+ coarse-grained model to describe this energy function , where each amino acid is represented as a single particle located at the corresponding Cα atom . While the lipids were not explicitly modeled , the interactions between the protein residues and the lipid membranes were represented through an implicit membrane potential . Water solvents were also treated implicitly . The second term , Ves ( R , H+ ) , represents the electrostatic interaction that depends on both the protein coordinates and the protonated state . This is the sum of the regular Coulomb interaction , VC ( R , H+ ) , between all electrostatic residues and a term , Vmem ( R , H+ ) , that depends on the membrane environment . Primarily , the term Vmem ( R , H+ ) is a potential that applies for 10 cE59: When cE59 stays in the membrane region , the deprotonated state has a high energy . The validity of the membrane model , Vmem ( R , H+ ) , was discussed in the previous paper by Kubo et al . , 2020 . The last term , Vpka ( H+ ) , expresses the energy required to protonate a protonatable residue . This energy depends on environmental conditions , such as membrane potential and pH , and the inherent pKa value . In this paper , we treated the membrane potential was to be 150mV , and the pH value of the IMS side and the matrix side to be 7 . 0 and 8 . 0 , respectively . The hybrid MC/MD simulations consisted of the MC phase , at which protonation states of 12 protonatable sites ( the glutamic acid [or aspartic acid in the case of mutants] in 10 c-subunits , aE223 , and aE162 ) are updated , and the MD phase , when amino acid positions are updated by Langevin dynamics . Each round contained MC trial moves for all the protons involved , followed by 105 MD steps ( Figure 3—figure supplement 1a ) . All simulation setups were the same as those we have recently reported ( Kubo et al . , 2020 ) , except for the treatment of the cE59D mutation . In the hybrid MC/MD simulation , we mimicked cE59D mutations in the following manner . In the MD part , we simply changed the amino acid identity of the corresponding residue from glutamic acid to aspartic acid using the mutagenesis feature of PyMol ( Figure 3—figure supplement 1b ) . Given the nature of our coarse-grained representation , this results in minor changes . The MC move represents proton transfer , which must be largely affected by the cE59D mutations via two distinct mechanisms , that is , the change in transfer efficiency and the change in the free energy difference between protonated and deprotonated states . For the former , the proton transfer efficiency is markedly reduced by the cE59D mutation because aspartic acid has a shorter sidechain than glutamic acid by one methylene-group , and because the sidechain reorientation found in the corresponding glutamic acid ( Symersky et al . , 2012 ) may not occur in the aspartic acid mutant . In our model , the transfer efficiency contains exp-Ar-r0 factor , where r is the distance between Cα atoms of the donor and the acceptor , the offset distance r0 represents the sum of sidechain lengths of the donor and acceptor , and A is the decay rate . We used r0=0 . 8nm for cE59 ( the same value as reported previously Kubo et al . , 2020 ) and set r0=0 . 6nm for cE59D , representing its shorter sidechain of aspartic acid . The decay rate A was set to 2 . 5 ( 1/nm ) for cE59 ( the same value as reported previously Kubo et al . , 2020 ) and 9 . 0 ( 1/nm ) for cE56D , assuming the absence of sidechain reorientation in the mutant ( Figure 3—figure supplement 1c ) . Second , the free energy difference between the states before and after the proton transfer is modulated by pKa differences in the donor and the acceptor amino acids and thus is affected by the cE59D mutation . Although the pKa value specific to the corresponding site is unknown , we empirically chose pKa=8 . 0 for cE59 and 7 . 0 for cE59D considering the intrinsic difference in pKa values , a previous argument ( Srivastava et al . , 2018 ) , and computational estimates of pKa value by PROPKA ( Li et al . , 2005 ) . The validity of the decay rate A and the pKa values are further discussed below . The decay rate A and pKa of cE59 were set to 2 . 5 ( 1/nm ) and 8 . 0 , respectively , which are the same values as those used by Kubo et al . , 2020 . In this study , for the mutant cE59D , we set the decay rate A , and the pKa of cE59D as 9 . 0 ( 1/nm ) and 7 . 0 , respectively . Here , we discuss the validity of these parameter selections for cE59D . First , the parameter A appears in the weight wi→jr , θ estimation that determines the proton hopping probability , wi→jr , θ=frgθhR176 in which fr=exp-Ar-r0 represents the distance dependence . gθ is introduced to reflect the sidechain orientation of cE59 , and is a Gaussian function that depends on the angle , θ , between the vector from the midpoint of the two adjacent residues of cE59 to cE59 itself , and of the vector from cE59 to the half-channel of a-subunit ( aE223 and aE162 in gθ for the IMS and the matrix-side half-channels ) . The parameters in the Gaussian function are determined using the cryo-EM structure . hR176 is included to mimic the role of aR176 that inhibits the proton leakage between the two half channels ( Mitome et al . , 2010 ) . Next , we demonstrated the effect of cE59D on proton transfer probability . Since the cE59D mutation does not affect the role of aR176 , hR176 should not be changed . The angle dependence g ( θ ) may be affected by the mutation . However , there is no direct structural information on the side chain of the mutated aspartic acid in its protonated/deprotonated states . Thus , we applied the same function as that for glutamic acid . Symersky et al . , 2012 obtained the X-ray structure of the c-ring not embedded in the lipid bilayer and compared it with the previous c-ring structure embedded in the lipid bilayer ( Symersky et al . , 2012 ) . They found that the side chain of cE59 in the lipid environment has its tip facing inside of the c-ring ( closed conformation ) , whereas the side chain of cE59 in water aqueous solution has its tip facing outside of the c-ring ( open conformation ) . In addition , MD simulations confirmed that the orientation of the side chain of cE59 is reversible depending on the environment . Based on these results , they concluded that the orientation of the side chains of cE59 changes when cE59 moves from the environment in the lipid membrane to the environment facing the a-subunit , which would facilitate proton hopping . It should be noted that the sidechain reorientation depends on the proximity of cE59 with its proton-relaying partner ( aE223 or aE162 in our case ) . Thus , in our coarse-grained model , this effect can effectively be included in fr . For the case of the cE59D mutant , however , it is unlikely that the same degree of the reorientation occurs . Therefore , we decided to increase the decay rate A , of cE59D . To examine the impact of this parametrization , we conducted the same simulation with the parameter A of cE59D set at 2 . 5/nm , the same value as in cE59 . The results in Figure 3—figure supplement 2a show that the velocity of the single mutant ( e ) was lower than that of the WT , but the extent of decrease ( 30% ) was smaller than that of the experimental result ( 75% ) . Therefore , we set A=9 . 0/nm as the parameter of cE59D , which results in consistent results being obtained in the experiment for the single mutant ( e ) . Next , we assessed the pKa values of cE59 and cE59D . The intrinsic pKa of aspartic acid in aqueous solutions is smaller than that of glutamic acid . Also , the pKa values of the donor and the acceptor inside proteins are thought to be larger in water due to pKa shift ( Srivastava et al . , 2018 ) . Indeed , pKa estimations by a standard tool , PROPKA ( Li et al . , 2005 ) , on the WT c10-ring and the a-subunit complexes showed that the pKa values of the a-subunit-facing chain-a , -h , -i , and -j have particularly large pKa values , with the largest being 8 . 03 for chain-i ( Table 3 ) . Therefore , based on the study by ( Srivastava et al . , 2018 ) and the PROPKA result , we set the pKa of cE59 at 8 . 0 . Similarly , we used PROPKA for assessing the pKa values in the complex of the c10-ring with cE59D and the a-subunit , finding that the pKa value was , on average , 1 . 0 unit smaller than that in the cE59 . Therefore , we set the pKa for cE59D at 7 . 0 . Since the difference in the pKa values between aspartic acid and glutamic acid in the aqueous solution was 0 . 2 , we also considered an alternative parametrization; the pKa value of the cE59D at 7 . 8 ( Figure 3—figure supplement 2b ) . The results showed no significant difference in the rotation velocity between the single mutant ( e ) and the double mutant ( ej ) , which is inconsistent with the results of the experiment . Thus , we decided to consider the estimate made using PROPKA , and the pKa value for the cE59D was set at 7 . 0 . For each of the WT Foac10 and the six cE59D mutation patterns corresponding to the biochemical assay , we carried out 10 independent simulation runs with different stochastic forces . The mutants included the single mutant “e” and the five double-mutants “ef , ” “eg , ” “eh , ” “ei , ” and “ej . ” The single mutant “e , ” for example , has the cE59D substitution only in the “e” chain , whereas other chains contain the WT c-subunit sequence . The double-mutant “ef” harbors substitutions in the two neighboring subunits . Each simulation run contained 6000 rounds of MC/MD cycles ( twice as long as in our previous paper Kubo et al . , 2020 ) . Each round contained MC trial moves for all the protons involved and 105 MD steps . Thus , the entire trajectory corresponds to 6 . 0×108 MD ( 60 , 000 frames saved ) . Notably , due to limitations in the computation time , we could simulate only one to a few turns of 360° rotations for each trajectory . As the mutant systems show asymmetric arrangements , the unbiased estimate of average velocities requires the rotation of multiples of 360° . Thus , we used the cumulative rotation angle and the MD time step at which the c10-ring returned to the initial orientation for the last time in each trajectory . The rotation velocity was obtained as the ratio of the cumulative rotation angle to the MD time . This velocity was then averaged over 10 trajectories . | Cells need to be able to store and transfer energy to fuel their various activities . To do this , they produce a small molecule called ATP to carry the energy , which is then released when the ATP is broken down . An enzyme found in plants , animals and bacteria , called FoF1 ATP synthase , can both create and use ATP . When it does this , protons , or positive hydrogen ions , are transported across cellular boundaries called membranes . The region of the enzyme that is responsible for pumping the protons contains different parts known as the c-ring and the a-subunit . The movement of protons drives the c-ring to rotate relative to the a-subunit , which leads to producing ATP . Previous research using simulations and the protein structures found there are two or three neighbouring amino acids in the c-ring that face the a-subunit , suggesting that these amino acids act together to drive the rotation . To test this hypothesis , Mitome et al . mutated these amino acids to examine the effect on the enzyme’s ability to produce ATP . A single mutation reduced the production of ATP , which decreased even further with mutations in two of the amino acids . The extent of this decrease depended on the distance between the two mutations in the c-ring . Simulations of these changes also found similar results . This indicates there is coordination between different parts of the c-ring to increase the rate of ATP production . This study offers new insights into the molecular processes controlling ATP synthesis and confirms previous theoretical research . This will interest specialists in bioenergetics because it addresses a fundamental biological question with broad impact . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2022 | Cooperation among c-subunits of FoF1-ATP synthase in rotation-coupled proton translocation |
Dangerous damage to mitochondrial DNA ( mtDNA ) can be ameliorated during mammalian development through a highly debated mechanism called the mtDNA bottleneck . Uncertainty surrounding this process limits our ability to address inherited mtDNA diseases . We produce a new , physically motivated , generalisable theoretical model for mtDNA populations during development , allowing the first statistical comparison of proposed bottleneck mechanisms . Using approximate Bayesian computation and mouse data , we find most statistical support for a combination of binomial partitioning of mtDNAs at cell divisions and random mtDNA turnover , meaning that the debated exact magnitude of mtDNA copy number depletion is flexible . New experimental measurements from a wild-derived mtDNA pairing in mice confirm the theoretical predictions of this model . We analytically solve a mathematical description of this mechanism , computing probabilities of mtDNA disease onset , efficacy of clinical sampling strategies , and effects of potential dynamic interventions , thus developing a quantitative and experimentally-supported stochastic theory of the bottleneck .
Mitochondria are vital energy-producing organelles within eukaryotic cells , possessing genomes ( mitochondrial DNA , mtDNA ) that replicate , degrade and develop mutations ( Rand , 2001; Wallace and Chalkia , 2013 ) . MtDNA mutations have been implicated in numerous pathologies including fatal inherited diseases and ageing ( Lightowlers et al . , 1997; Wallace , 1999; Poulton et al . , 2009; Wallace and Chalkia , 2013 ) . Combatting the buildup of mtDNA mutations is of paramount importance in ensuring an organism's survival . Substantial recent medical , experimental , and media attention has focused on methods to remove ( Bacman et al . , 2013 ) or prevent the inheritance of ( Bredenoord et al . , 2008; Poulton et al . , 2009; Craven et al . , 2010; Poulton et al . , 2010; Burgstaller et al . , 2015 ) mutated mtDNA in humans . One means by which organisms may ameliorate the mtDNA damage that builds up through a lifetime is through a developmental process known as bottlenecking . Immediately after fertilisation , a single oocyte ( which may contain >105 individual mtDNAs ) may have a nonzero mtDNA mutant load or heteroplasmy ( the proportion of mutant mtDNA in the cell ) . As the number of cells in the developing organism increases , the intercellular population then acquires an associated heteroplasmy variance , that is , the variance in mutant load across the population of cells ( Figure 1A ) , allowing removal of cells with high heteroplasmy and retention of cells with low heteroplasmy . Intense and sustained debate exists as to the mechanism by which this increase of heteroplasmy variance occurs . Several experimental results in mice suggest that , during development , the copy number of mtDNA per cell in the germ cell line drops dramatically to ∼102 , reducing the effective population size of mitochondrial genomes ( Cree et al . , 2008; Wai et al . , 2008 ) . One postulated bottlenecking mechanism is that this low population size accelerates genetic drift and so increases the cell-to-cell heteroplasmy variance ( Bergstrom and Pritchard , 1998; Aiken et al . , 2008; Cree et al . , 2008; Wonnapinij et al . , 2010 ) , which was first observed to generally increase from primordial germ cells through primary oocytes to mature oocytes ( Jenuth et al . , 1996 ) . However , independent experimental evidence ( Wai et al . , 2008 ) suggests that heteroplasmy variance increases negligibly during this copy number reduction , though this interpretation has been debated ( Samuels et al . , 2010 ) . Wai et al . ( 2008 ) shows heteroplasmy variance rising during folliculogenesis , after the mtDNA copy number minimum has been passed . In yet another picture , supported by conflicting experimental results ( Cao et al . , 2007 , 2009 ) , heteroplasmy variance increases with a less pronounced decrease in mtDNA copy number ( a minimum copy number >103 in mice ) , solely through random effects associated with partitioning at cell divisions . Clearly a consensus on this important mechanism is yet to be reached . 10 . 7554/eLife . 07464 . 003Figure 1 . The mitochondrial bottleneck , and elements of a general model for bottlenecking mechanisms . ( A ) The mitochondrial DNA ( mtDNA ) bottleneck acts to produce a population of oocytes with varying heteroplasmies from a single initial oocyte with a specific heteroplasmy value . During development , mtDNA copy number per cell decreases ( by a debated amount , which we address; see Main text ) then recovers , suggesting a ‘bottleneck’ of cellular mtDNA populations . ( B ) Cellular mtDNA populations during the bottleneck are modelled as containing wildtype and mutant mtDNAs . MtDNAs can replicate and degrade within a cell cycle , with rates λ and ν respectively . ( C ) At cell divisions , the mtDNA population is partitioned between two daughter cells either deterministically , binomially , or through the binomial partitioning of mtDNA clusters . ( D ) Symbols used to represent quantities and model parameters used in the Main text , and their biological interpretations . DOI: http://dx . doi . org/10 . 7554/eLife . 07464 . 003 Important existing theoretical work on modelling the bottleneck has assumed a particular underlying mechanism ( Bergstrom and Pritchard , 1998; Wolff et al . , 2011 ) or derived statistics of mtDNA populations ( Chinnery and Samuels , 1999; Elson et al . , 2001; Wonnapinij et al . , 2008 , 2010 ) without explicitly considering changing mtDNA population size , or the discrete nature of the mtDNA population: effects which may powerfully affect mtDNA statistics . To capture these effects it is necessary to employ a ‘bottom-up’ physical description of mtDNA as populations of individual , discrete elements subject to replication and degradation , as in , for example , ( Chinnery and Samuels , 1999 ) and ( Capps et al . , 2003 ) . Exploring the bottleneck also requires explicitly modelling partitioning dynamics throughout a series of cell divisions , over which population size can change dramatically . While previous simulation work ( Cree et al . , 2008; Poovathingal et al . , 2009 ) has taken such a philosophy with specific model assumptions , we are not aware of such a study allowing for the wide variety of replication and partitioning dynamics proposed in the literature; we further note that replication-degradation-partitioning mtDNA models are yet to be fully described analytically . Nor is there a general quantitative framework under which different proposed bottleneck mechanisms can be statistically compared given extant data ( although statistical analyses focusing on particular mechanisms and individual sets of experimental results have been used throughout the literature , for example , using a Bayesian approach under a particular bottleneck model to infer model bottleneck size [Marchington et al . , 1998] ) . Combined developments in theory and inference are therefore required to make progress on this important question . We remedy this situation by constructing a general model ( features and parameters described in Figure 1 ) for the population dynamics of the bottleneck , able to describe the range of proposed mechanisms existing in the literature . Using experimental data on mtDNA statistics through development ( Jenuth et al . , 1996; Cao et al . , 2007; Cree et al . , 2008; Wai et al . , 2008 ) , we use approximate Bayesian computation ( Beaumont et al . , 2002; Toni et al . , 2009; Sunnåker et al . , 2013; Johnston , 2014 ) to rigorously explore the statistical support for each mechanism , showing that random mtDNA turnover coupled with binomial partitioning of mtDNAs at cell divisions is highly likely , and that the debated magnitude of mtDNA copy number reduction is somewhat flexible . Subsequently , we confirm the predictions of this model by performing new experimental measurements of heteroplasmy statistics in mice with an mtDNA admixture , including a wild-derived haplotype , that is genetically distinct from previous studies . We then analytically solve the equations describing mtDNA population dynamics under this mechanism and show that these results allow us to investigate potential interventions to modulate the bottleneck ( suggesting that upregulation of mtDNA degradation may increase the power of the bottleneck to avoid inherited disease; we discuss potential strategies for such an intervention ) and yield quantitative results for clinical questions including the timescales and probabilities of disease onset , and the efficacy of strategies to sample heteroplasmy in clinical planning .
We will consider three different classes of proposed generating mechanisms for the mtDNA bottleneck: those proposed in Cao et al . ( 2007 ) ; Cree et al . ( 2008 ) and Wai et al . ( 2008 ) . We will refer to these mechanisms by their leading author name . The Cree mechanism involves random replication and degradation of mtDNAs throughout development , and binomial partitioning of mtDNAs at cell divisions . The Cao mechanism involves partitioning of clusters of mtDNA at each cell division , thus providing strong stochastic effects associated with each division . We consider a general set of dynamics through which this cluster inheritance may be manifest , including the possibility of heteroplasmic ‘nucleoids’ of constant internal structure ( Jacobs et al . , 2000 ) , sets of molecules or nucleoids within an organelle , homoplasmic clusters , and different possible cluster sizes ( see Appendix 1 ) . The Wai mechanism involves the replication of a subset of mtDNAs during folliculogenesis . We note that this latter mechanism can be manifest in several ways: ( a ) through slow random replication of mtDNAs ( so that , in any given time window , only a subset of mtDNAs will be actively replicating ) or ( b ) through the restriction of replication to a specific subset of mtDNAs at some point during development . We will refer to these different manifestations as Wai ( a ) and Wai ( b ) respectively . The Wai ( a ) mechanism and the Cree model can both be addressed in the same mechanistic framework ( with potentially different parameterisations ) : if the rate of random replication in the Cree model is sufficiently low during folliculogenesis , only a subset of mtDNAs will be actively replicating at any given time during this period , thus recapitulating the Wai ( a ) mechanism ( see Appendix 1 ) . We will henceforth combine discussion of the Wai ( a ) and Cree mechanisms into what we term the birth-death-partition ( BDP ) mechanism . We seek a physically motivated mathematical model for the bottleneck that is capable of reproducing each of these mechanisms . Our general model for the bottleneck ( detailed description in ‘Materials and methods’ ) involves a ‘bottom-up’ representation of mtDNAs as individual intracellular elements capable of replication and degradation ( Figure 1B ) with rates λ and ν respectively . A parameter S determines whether these processes are deterministic ( specific rates of proliferation ) or stochastic ( replication and degradation of each mtDNA is a random event ) . These rates of replication and degradation of mtDNA are likely strongly linked to mitochondrial dynamics within cells , through the action of mitochondrial quality control ( Twig et al . , 2008; Hill et al . , 2012 ) modulated by mitochondrial fission and fusion ( Detmer and Chan , 2007; Youle and van der Bliek , 2012; Hoitzing et al . , 2015 ) , which can act to recycle weakly-perfoming mitochondria ( Mouli et al . , 2009; Twig and Shirihai , 2011 ) . This quality control can be represented through the degradation rates assigned to each mtDNA species , which may differ ( for selective quality control ) or be identical ( for non-selective turnover ) . The proportion of mtDNAs capable of replication is controlled by a parameter α in our model , dictating the proportion of mtDNAs that may replicate after a cutoff time T . Thus , if α = 1 , all mtDNAs may replicate; if α < 1 , replication of a subset proportion α of mtDNAs is enforced at this cutoff time . At cell divisions , mtDNAs may be partitioned either deterministically , binomially , or in clusters according to a parameter c ( Figure 1C ) . The copy number of mtDNA per cell is observed to vary dramatically during development , with dynamic phases of copy number depletion and different rates of subsequent recovery observed . Additionally , cell divisions occur in the germline at different rates during development , with cells becoming largely quiescent after primary oocytes develop . To explicitly model these different dynamic regimes , and the behaviour of mtDNA copy number during each , we include six different dynamic phases throughout development , each with different rates of replication and degradation ( labelled with subscript i labelling the dynamic phase: hence λ1 , ν1 , … , λ6 , ν6 ) , and allowing for different rates of cell division or quiescence . This protocol enables us to explicitly model effects of changing population size throughout development rather than assuming dependence on a single , coarse-grained effective population size; and to include the effects of specific and varying cell doubling times . A summary of symbols used in our model and throughout this article is presented in Figure 1D . Our model , with suitable parameterisation , can thus mirror the dynamics of the Cree and Wai ( a ) mechanisms ( stochastic dynamics and binomial partitioning , which we refer to as the BDP mechanism ) ; the Cao mechanism ( clustered partitioning ) ; and Wai ( b ) mechanism ( deterministic dynamics , restricted subset of replicating mtDNAs ) . The Cao mechanism , partitioning of clusters of mtDNA molecules , represents the expected case if mtDNA is partitioned in colocalised ‘nucleoids’ within each organelle ( or in other sub-organellar groupings ) . The size of mtDNA nucleoids is debated in the literature ( Bogenhagen , 2012; Kukat and Larsson , 2013; Wallace and Chalkia , 2013 ) ( although recent evidence from high-resolution microscopy suggests that nucleoid size is generally <2 ( Jakobs and Wurm , 2014 ) , consonant with recent evidence that individual nucleoids may be homoplasmic [Poe et al . , 2010] ) ; our model allows for inheritance of homoplasmic or heteroplasmic nucleoids of arbitrary characteristic size c , thus allowing for a range of sub-organellar mtDNA structure . We discuss the impact of mixed or fixed nucleoid content in Appendix 1 . We take data on mtDNA copy number in germ line cells in mice from three experimental studies ( Cao et al . , 2007; Cree et al . , 2008; Wai et al . , 2008 ) . We also use data from two experimental studies on heteroplasmy variance in the mouse germ line during development ( Jenuth et al . , 1996; Wai et al . , 2008 ) . These heteroplasmy variance studies employ intracellular combinations of the same pairing of mtDNA haplotypes ( NZB and BALB/c ) , modelling two different mtDNA types within a cellular population . These data , by convention ( Samuels et al . , 2010 ) , are normalised by heteroplasmy level h , giving ( 1 ) V′ ( h ) =V ( h ) E ( h ) ( 1−E ( h ) ) , where normalised variance V′ ( h ) is a quantity that will be often used subsequently . This normalised variance controls for the effect of different or changing mean heteroplasmy , and thus allows a comparison of heteroplasmy variance among samples with different mean heteroplasmies and subject to heteroplasmy change with time . We use a time of 100 dpc to correspond to mature oocytes ( see ‘Materials and methods’ ) . We take data on cell doubling times from a classical study ( Lawson and Hage , 1994 ) ( see ‘Materials and methods’ ) . A possible summary of these data ( although they provoke ongoing debate; see ‘Discussion’ ) is that , as shown in Figure 2A , the existing data on normalised heteroplasmy variance shows initially low variance until ∼7 . 5 dpc ( days post conception , which we use as a unit of time throughout ) , rising to intermediate values between 7 . 5 and 21 dpc , gradually rising further subsequently to become large in the mature oocytes of the next generation . In Figure 2A , and throughout this article , experimentally measured data will be depicted as circular or polygonal points , and inferred theoretical behaviour will be depicted as lines or shaded regions . 10 . 7554/eLife . 07464 . 004Figure 2 . Different mechanisms for the mtDNA bottleneck . ( A ) Trajectories of mean copy number E ( m ) and normalised heteroplasmy variance V ( h ) arising from the models described in the text , optimised with respect to data from experimental studies . Birth-death-partition ( BDP ) denotes the BDP model , encompassing Cree and Wai ( A ) mechanisms . Left plots show trajectories during development; right plots show behaviour in mature oocytes in the next generation . * denotes measurements in mature oocytes , modelled as 100 dpc ( see ‘Materials and methods’ ) . ( B ) Statistical support for different mechanisms from approximate Bayesian computation ( ABC ) model selection with thresholds ϵ1 , 2 , 3 , 4 = 75 , 60 , 50 , 45 . As the threshold decreases , forcing a stricter agreement with experiment ( thinner , darker columns ) , support converges on the BDP model . DOI: http://dx . doi . org/10 . 7554/eLife . 07464 . 004 Figure 2A shows mtDNA population dynamic trajectories resulting from optimised parameterisations of each of the mechanisms we consider ( see ‘Materials and methods’ ) . In Figure 2B we show posterior probabilities on each of these mechanisms . These posterior probabilities give the inferred statistical support for each mechanism , derived from model selection performed with approximate Bayesian computation ( ABC ) ( Beaumont et al . , 2002; Toni et al . , 2009; Sunnåker et al . , 2013; Johnston , 2014 ) using uniform priors . ABC involves choosing a threshold value dictating how close a fit to experimental data is required to accept a particular model parameterisation as reasonable . In our case , this goodness-of-fit is computed using a comparison of squared residuals associated with the trajectories of mean mtDNA copy number and normalised heteroplasmy variance ( see ‘Materials and methods’ and Appendix 1 ) . Each of the experimental measurements corresponds to a sample variance , derived from a finite number of samples of an underlying distribution of heteroplasmies , and therefore has an associated uncertainty and sampling error ( Wonnapinij et al . , 2010 ) . The reasonably small sample sizes used in these sample variance measurements are likely to underestimate the underlying heteroplasmy variance ( the target of our inference ) . Our ABC approach naturally addresses these uncertainties by using summary statistics derived from sampling a set of stochastic incarnations of a given model , where the size of this set is equal to the number of measurements contributing to the experimentally-determined statistic ( see ‘Materials and methods’ ) . Figure 2B clearly shows that as the ABC threshold is decreased , requiring closer agreement between the distributions of simulated and experimental data , the posterior probability of the BDP model increases , to dramatically exceed those of the other models . This increase indicates that the BDP model is the most statistically supported , and capable of providing the best explanation of experimental data ( which can be inutitively seen from the trajectories in Figure 2A ) . We note that ABC model selection automatically takes model complexity into account , and conclude that the BDP mechanism is the best supported proposed mechanism for the bottleneck . Briefly , this result arises because the BDP model produces increasing variance both due to early cell division stochasticity and later random turnover . By contrast , the Cao model alone only increases variance in early development when cell divisions are occurring . Qualitatively , this behaviour through time holds regardless of cluster ( nucleoid ) size and regardless of whether clusters are heteroplasmic or homoplasmic ( allowing heteroplasmic clusters decreases the magnitude of heteroplasmy variance but not its behaviour through time , see Appendix 1 ) . The Wai ( b ) model alone similarly only increases variance at a single time point ( later , during folliculogenesis ) . In Wai et al . ( 2008 ) , visualisations of cells after BrU incorporation show that a subset of mitochondria retain BrU labelling , which the authors suggest indicates that a subset of mtDNAs are replicating . In Appendix 1 , we show that the BDP model also results in the observation of only a subset of replicating mtDNAs over the timeframe corresponding to these experimental results . These observations thus correspond to results expected from the random turnover from the BDP model . We also note the mathematical observation that the Wai ( b ) mechanism requires the replication of <1% of mtDNAs during folliculogenesis to yield reasonable heteroplasmy variance increases ( Figure 2A shows the optimal case with α = 0 . 006; optimal fits to data generally show 0 . 005 < α < 0 . 01 ) , and the proportions of loci visible in Wai et al . ( 2008 ) are substantially higher than this required 1% value . We show in Appendix 1 that the heteroplasmy statistics corresponding to binomial partitioning also describe the case where the elements of inheritance are heteroplasmic clusters , where the mtDNA content of each cluster is randomly sampled from the population of the cell ( either once , as an initial step , or repeatedly at each division ) . This similarity holds broadly , regardless of whether the internal structure of clusters is constant across cell divisions or allowed to mix between divisions . The BDP model , in addition to describing the partitioning of individual mtDNAs , also thus represents the statistics of mtDNA populations in which heteroplasmic nucleoids are inherited ( Jacobs et al . , 2000 ) , or individual organelles containing a mixed set of mtDNAs or nucleoids are inherited , regardless of the size of these nucleoids ( see ‘Discussion’ ) . Having used ABC model selection to identify the BDP model as the most statistically supported , we can also use ABC to infer the values of the governing parameters of this model given experimental data . Figure 3A , B shows the trajectories of mean copy number and mean heteroplasmy variance resulting from model parameterisations identified through this process . Figure 3C shows the inferred behaviour of mtDNA degradation rate ν in the model , a proxy for mtDNA turnover ( as the copy number is constrained ) . Turnover is generally low during cell divisions , allowing heteroplasmy variance to increase due to stochastic partitioning . Turnover then increases later in germ line development , resulting in a gradual increase of heteroplasmy variance after birth until the mature oocytes form in the next generation . 10 . 7554/eLife . 07464 . 005Figure 3 . Parameterisation of the BDP model and inferred details of bottleneck mechanism . Trajectories of ( A ) mean copy number E ( m ) and ( B ) normalised heteroplasmy variance V′ ( h ) resulting from BDP model parameterisations sampled using ABC with a threshold ϵ = 40 . * denotes measurements in mature oocytes , modelled as 100 dpc ( see ‘Materials and methods’ ) . Note: the range in ( B ) does not correspond to a credibility interval on individual measurements , but rather on an expected underlying ( population ) variance , from which individual variance measurements are sampled . We thus expect to see , for example , several measurements lower than this range due to sampling limitations ( see text ) . ( C ) Posterior distributions on mtDNA turnover ν with time . ( D ) Posterior distribution on min E ( m ) , the minimum mtDNA copy number reached during development . ( E ) Posterior distribution on σ=∑i=36τ′iνi , a measure of the total amount of mtDNA turnover . DOI: http://dx . doi . org/10 . 7554/eLife . 07464 . 005 Figure 3D shows posterior distributions on the copy number minimum and total turnover ( see ‘Materials and methods’ ) resulting from this process; posteriors on all other parameters are shown in Appendix 1 . Substantial flexibility exists in the magnitude of the copy number minimum , illustrating that observed heteroplasmy variance can result from a range of bottleneck sizes from ∼200 to >103; going some way towards reconciling the conflict between Cao et al . ( 2007 ) and Cao et al . ( 2009 ) and Cree et al . ( 2008 ) and Wai et al . ( 2008 ) . The total amount of mtDNA turnover ( presented as σ=∑i=36νiτ′i , the product of turnover rate and the time for which this rate applies , summed over quiescent dynamic phases; for example , a turnover rate of 0 . 1 hr−1 for 30 days yields σ = 0 . 1 × 24 × 30 = 72 ) is constrained more than the specific trajectory of mtDNA turnover rates , showing that a variety of time behaviours of turnover are capable of producing the observed heteroplasmy behaviour . The bottleneck mechanism identified through our analysis has several characteristic features which facilitate experimental verification . Key among these are the prediction that heteroplasmy variance acquires an intermediate ( nonzero , but not maximal ) value as a result of the copy number bottleneck , then continues to increase due to mtDNA turnover in later development . Our theory also produces quantitative predictions regarding the structure of heteroplasmy distributions at arbitrary times . The existing data that we used to perform inference and model selection display a degree of internal heterogeneity , coming from several different experimental groups . Furthermore , these data represent statistics resulting from a single pairing of mtDNA types , and it is thus arguable how conclusions drawn from them may represent the more genetically diverse reality of biology . Burgstaller et al . ( 2014 ) recently addressed this issue of a limited number of mtDNA pairings by producing novel mouse models involving mixtures of standard and several new , unexplored , wild-derived haplotypes which capture a range of genetic diversity . To test the applicability and generality of our predictions , we have perfomed new experimental measurements of germline heteroplasmy variance in these model animals under a consistent experimental protocol ( see ‘Materials and methods’ ) . We use the ‘HB’ mouse line from Burgstaller et al . ( 2014 ) pairing a wild-derived mtDNA haplotype ( labelled ‘HB’ after its source in Hohenberg , Germany ) with C57BL/6N; we refer to this model as ‘HB’ . Heteroplasmy measurements were taken in oocytes sampled from mice at ages 24–61 dpc ( see ‘Materials and methods’ and Appendix 1; raw data in Figure 4—source data 1 ) . The statistics of these measurements yielded E ( h ) , V ( h ) and V′ ( h ) as previously . This age range was chosen to address the regions with most power to discriminate between the competing models; the existing V′ ( h ) data is most heterogeneous around 20–30 dpc and the later datapoints allow us to detect developmental heteroplasmy behaviour after the copy number minimum . Figure 4A shows these V′ ( h ) measurements . The qualitative behaviour predicted by the BDP mechanism is clearly visible: variance around birth ( after the copy number bottleneck ) is low but non-zero , subsequently increasing with time . The ability of the BDP model to account for the magnitudes and time behaviour of heteroplasmy variance more satisfactorily than the alternative models is shown by the model fits in Figure 4A . We explored these new data quantitatively through the same model selection approach used for the existing data . As shown in Figure 4B , the BDP mechanism again experiences by far the strongest statistical support in this genetically different system . 10 . 7554/eLife . 07464 . 006Figure 4 . Predictions and experimental verification of the BDP model . ( A ) New V′ ( h ) measurements from the HB mouse system , with optimised fits for the BDP , Wai ( b ) and Cao models . ( B ) Posterior probabilities of each model given this data under decreasing ABC threshold: ϵ = {50 , 40 , 30 , 25} . ( C ) All V′ ( h ) measurements from the HB model ( points ) with inferred V′ ( h ) behaviour from ABC applied to the BDP model ( red curves ) . As in Figure 3 , this range does not correspond to a credibility interval on individual measurements , but rather on an expected underlying ( population ) variance , from which individual variance measurements are sampled . The inferred behaviour strongly overlaps with the inferred behaviour for the BALB/c system ( blue curves ) , suggesting that the BDP model applies to a genetically diverse range of systems . ( D ) Heteroplasmy distributions . The transformation h′=−ln| ( h−1−1 ) E ( h ) / ( 1−E ( h ) ) | ( Burgstaller et al . , 2014 ) is used to compare distributions with different mean heteroplasmy . Red jitter points are samples from sets used to parameterise the BDP model; red curves show the 95% range on transformed heteroplasmy with time inferred from these samples . Blue jitter points are samples withheld independent from this parameterisation; their distributuions fall within the independently inferred range . Insets show , in untransformed space , distributions of the withheld heteroplasmy measurements ( blue ) compared to parameterised predictions ( red ) ; no withheld datasets show significant support against the predicted distribution ( Anderson-Darling test , p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07464 . 00610 . 7554/eLife . 07464 . 007Figure 4—source data 1 . Individual heteroplasmy measurements in the HB mouse model contributing to the new heteroplasmy variance data used to test our theory . DOI: http://dx . doi . org/10 . 7554/eLife . 07464 . 007 Figure 4C shows the result of our parameteric inference approach using these V′ ( h ) measurements coupled with the E ( m ) measurements used previously ( employing our assumption that modulation of copy number by heteroplasmy in this non-pathological haplotype is small ) . Strikingly , the quantitative behaviour of V′ ( h ) with time inferred from the HB model ( red ) matches the previous behaviour inferred from the NZB/BALB/c system ( blue ) very well , suggesting that our theory is applicable across a range of genetically distinct pairings . We note that the shaded region in Figure 4C corresponds to credibility intervals around the mean behaviour of V′ ( h ) , and the fact that individual V′ ( h ) datapoints ( subject to fluctuations and sampling effects ) do not all lie within these intervals is not a signal of poor model choice . An analogous situation is the observation of a scatter of datapoints outside the range of the standard error on the mean ( s . e . m . ) , which does not imply a mistake in the s . e . m . estimate . The difference between the trace in Figure 4A and the mean curve in Figure 4C arises because Figure 4A shows the behaviour of the model under a single , optimised parameterisation , whereas Figure 4C shows the distribution of model behaviours over the posterior distributions on parameters: the mean V′ ( h ) trace of this distribution is comparable but not equivalent to that from the single best-fit parameterisation . To confirm more detailed predictions of our model , we also examined the specific distributions of heteroplasmy in our new measurements . Given a mean heteroplasmy and an organismal age , the parameterised BDP model predicts the structure of the heteroplasmy distribution ( see ‘Materials and methods’ and next section ) . We parameterised the model using V′ ( h ) values from a subset of half of the new measurements ( chosen by omitting every other sampled set when ordered by time ) . Figure 4D shows a comparison of measured heteroplasmy distributions with a 95% bound from the parameterised BDP model . We then tested the predictions of the parameterised model against the other half of new measurements . 8 of the test measurements ( 2 . 4% ) fell outside the inferred 95% bound from the training dataset , illustrating a good agreement with distributional predictions . The Anderson-Darling test was used to compare the distribution of heteroplasmy in sampled oocytes with distributions predicted by our theory ( given age and mean heteroplasmy ) ; no set of samples showed significant ( p < 0 . 05 ) departures from the hypothesis that the two distributions were identical . Some example distributions are presented in Figure 4D ( i ) , ( ii ) , ( iii ) . Importantly , the BDP model yields analytic solutions for the values of all genetic properties of interest , using tools from stochastic processes ( detail in ‘Materials and methods’ and Appendix 1 ) . These results facilitate straightforward further study and fast predictions of timescales and probabilities of interest . The full theoretical approach is detailed in Appendix 1 , and equations for the mean and variance of mtDNA populations and heteroplasmy are given in the Methods . In Figure 2A we illustrate that these analytic results provide an excellent match to the numeric results of stochastic simulation , a result that holds across all BDP model parameterisations . It is also straightforward to calculate the fixation probability P ( m=0 ) , which allows us to characterise all heteroplasmy distributions that arise from the bottlenecking process , even when highly skewed ( see ‘Materials and methods’ and Appendix 1 ) . We have thus obtained analytic solutions for the time behaviour of mtDNA copy number and heteroplasmy throughout the bottleneck with no assumptions of continuous population densities or fixed population size , under a physical model with the most statistical support given experimental data . We can use our theory to explore the dependence of bottleneck dynamics on specific biological parameters . We first explore the effects of modulating mtDNA turnover by varying λ and ν in concert , corresponding to an increase in mtDNA degradation balanced by a corresponding increase in mtDNA replication . This increased mtDNA turnover increases the heteroplasmy variance ( see Figure 5A ) due to the increased variability in mtDNA copy number from the underlying random processes occurring at increased rates . We find that increasing mtDNA degradation ν without increasing λ also increases heteroplasmy variance , in addition to decreasing the overall mtDNA copy number ( Figure 5B ) . Applying this unbalanced increase in mtDNA degradation without a matching change in replication has a strong effect on mtDNA dynamics as it corresponds to a universal change in the ‘control’ applied to the system , analogous , for example , to changing target copy numbers in manifestations of relaxed replication ( Chinnery and Samuels , 1999 ) . The simple model we use does not include feedback , and controls mtDNA dynamics solely through kinetic parameters . Perturbing the balance of these parameters thus strongly affects the expected behaviour of the system . As we discuss later , elucidation of the specific mechanisms by which control is manifest in mtDNA populations will require further research , but these numerical experiments attempt to represent the cases where a perturbation is naturally compensated for ( matched changes , Figure 5A ) and where it is not ( unbalanced change , Figure 5B ) . 10 . 7554/eLife . 07464 . 008Figure 5 . Quantitative influences and clinical results from our bottlenecking model . ( A–C ) Trajectories of copy number E ( m ) and normalised heteroplasmy variance V′ ( h ) resulting from perturbing different physical parameters . Trajectory C labels the ‘control’ trajectory resulting from a fixed parameterisation; black dots show experimental data; * denotes measurements from primary oocytes , modelled at 100 dpc . ( A ) Increasing ( T+ ) and decreasing ( T− ) mtDNA turnover ( both mtDNA replication and degradation ) by 20% . ( B ) Increasing ( M+ ) and decreasing ( M− ) mtDNA degradation throughout development by a constant value ( 2 × 10−4 , in units of day−1 ) , while keeping replication constant . ( C ) Applying a positive ( S+ ) and negative ( S− ) selective pressure to mutant mtDNA by 5 × 10−6 day−1 . ( D ) Probability of crossing different heteroplasmy thresholds h* with time , starting with initial heteroplasmy h0 = 0 . 3 . ( E ) Probability distributions over embryonic heteroplasmy h given a measurement hm from preimplantation sampling ( ** hm = 0 . 1; *** hm = 0 . 4 ) at different times . DOI: http://dx . doi . org/10 . 7554/eLife . 07464 . 008 These results suggest that an artificial intervention increasing mitochondrial degradation may generally be expected to increase heteroplasmy variance during development . An increase in mtDNA degradation is expected to either directly increase heteroplasmy variance ( Figure 5B ) if mtDNA populations are weakly controlled , or to provoke a compensatory , population-maintaining increase in mtDNA replication , thus increasing mtDNA turnover , which also acts to increase variance ( Figure 5C ) if mtDNA populations are subject to feedback control . The increase in variance through either of these pathways will increase the power of cell-level selection to remove cells with high heteroplasmy and thus purify the population . For this reason , we speculate that mitochondrial degradation may represent a potential clinical target to address the inheritance of mtDNA disease ( more detail in Appendix 1 ) . Our model also allows us to explore the effect of different mtDNA types experiencing different selective pressures , by setting λ1 ≠ λ2 ( mutant mtDNA experiences a proliferative advantage or disadvantage ) . Such a selective difference causes changes in both mean heteroplasmy and heteroplasmy variance , as shown in Figure 5C ( e . g . , if heteroplasmy decreases towards zero , heteroplasmy variance will also decrease , as the wildtype is increasingly likely to become fixed ) . We do not focus further on selection in this study , noting that selective pressures are likely to be specific to a given pair or set of mtDNA types and are not generally characterised well enough to perform satisfactory inference . However , we do note that our theory gives a straightforward prediction for the functional form of mean heteroplasmy when nonzero selection is present , a sigmoid with slope set by the fitness difference ( see ‘Materials and methods’ ) . A key feature of mtDNA diseases is that pathological symptoms usually manifest when heteroplasmy in a tissue exceeds a certain threshold value , with few or no symptoms manifested below this threshold ( Rossignol et al . , 2003 ) . The probability and timescale with which cellular heteroplasmy may be expected to exceed a given value is thus a quantity of key interest in clinical planning of mtDNA disease strategies . In our model , the probability , as a function of time , of a cell containing m1 wildtype and m2 mutant mtDNAs can be straightforwardly derived . The resultant analytic expression involves a hypergeometric function , also an important mathematical element in expressions describing mtDNA statistics based on classical population genetics ( Kimura , 1955; Wonnapinij et al . , 2008 ) . The probability of obtaining a given heteroplasmy can therefore be computed as a sum over all copy number states that correspond to that heteroplasmy . However , as hypergeometric functions are comparatively unintuitive and computationally expensive , we here employ an approximation to the distribution of heteroplasmy based upon the above moments that are straightforwardly calculable from our model . This approximation involves fixation probabilities for each mtDNA type and a truncated Normal distribution for intermediate heteroplasmies ( see ‘Materials and methods’ ) . In Appendix 1 we show that this approximation corresponds well to the exact distributions calculated using the hypergeometric function . We underline that exact heteroplasmy distributions are straightfoward to compute using our approach: we use the truncated Normal approximation as it represents the exact distribution well , is more intuitively interpretable , and is computationally very inexpensive . Using this approach , the probability with time of a cell exceeding a threshold heteroplasmy h* can be straightforwardly computed for any initial heteroplasmy , allowing rigorous quantitative elucidation of this important clinical quantity ( see ‘Materials and methods’ ) . Figure 5D illustrates this computation by showing the analytic probability with which thresholds h* = 0 . 4 , 0 . 5 , 0 . 6 are exceeded at a time t , given the example initial heteroplasmy h = 0 . 3 . These results serve as a simple example of the power of our modelling approach: any other specific case can readily be addressed . Our theory thus allows general quantitative calculation of the probability ( and timescale ) that any given heteroplasmy threshold will be exceeded , given knowledge of the initial ( or early ) heteroplasmy . We next turn to the question of estimating heteroplasmy levels in a developed organism by sampling cells during development . This principle , clinically termed preimplantation genetic diagnosis ( Steffann et al . , 2006; Poulton et al . , 2009 ) , assists in clinical planning by allowing inference of the specific heteroplasmic nature of the embryo itself rather than a population average of an affected mother's oocytes ( Treff et al . , 2012 ) . However , the complicated and stochastic nature of the bottleneck makes this inference a challenging problem . Given a heteroplasmy measurement from sampling hm , accurate preimplantation diagnosis is contingent on knowledge of the distribution P ( h|hm ) , that is , the probability that the embryonic heteroplasmy is h given that a measurement hm has been made . We can use our modelling framework and Bayes' theorem ( see ‘Materials and methods’ ) to obtain a formula for this conditional probability , allowing a rigorous probability to be assigned to inferences from preimplantation sampling . Here , as above , we employ the truncated Normal approximation for the heteroplasmy distribution , noting that the exact treatment using hypergeometric functions is straightforward but more computationally expensive . Figure 5E illustrates this process by showing the probability distributions on embryonic heteroplasmy when measurements hm = 0 . 1 or 0 . 4 have been taken at different times during development . The increasing heteroplasmy variance through development means that substantially greater uncertainty is associated with heteroplasmy values inferred using measurements taken at later times . In conclusion , although care must be taken in applying this reasoning to cell types in which , for example , mitochondrial and cell turnover rates differ from those assumed here , or differentiation leads to tissue-specific selective factors acting on the mtDNA population , this formalism provides a general means of rigorously inferring embryonic heteroplasmy through genetic diagnosis sampling .
We have used a general stochastic model and approximate Bayesian computation with the available experimental data on developmental mtDNA dynamics to show that the bottleneck is most likely manifest through stochastic mtDNA dynamics and partitioning , with increased random turnover later during development , a mechanism which we can describe exactly and analytically ( Figure 6 ) . We emphasise that the bottom-up construction of our model from physical first principles both increases the flexibility and generality of our model , allowing different mechanisms to be compared together , and providing information on mtDNA dynamics throughout development rather than estimating an overall effect . We note that even though our model cannot represent the full microscopic truth underlying the mtDNA bottleneck , its ability to recapitulate the wide range of extant experimental measurements suggest that its study may yield useful insights into the effects of different treatments and perturbations on the bottleneck . 10 . 7554/eLife . 07464 . 009Figure 6 . Model for the mtDNA bottleneck . A summary of our findings . ( A ) There is most statistical support for a bottlenecking mechanism whereby mtDNA dynamics is stochastic within a cell cycle , involving random replication and degradation of mtDNA , and mtDNAs are binomially partitioned at cell divisions . ( B ) This mechanism results in heteroplasmy variance increasing both due to stochastic partitioning at divisions and due to random turnover . The absolute magnitude of the copy number bottleneck is not critical: a range of bottleneck sizes can give rise to observed dynamics . Random turnover of mtDNA increases heteroplasmy variance through folliculogenesis and germline development . DOI: http://dx . doi . org/10 . 7554/eLife . 07464 . 009 A key debate in the literature has focussed on the magnitude of the bottleneck . Some studies ( Aiken et al . , 2008; Cree et al . , 2008 ) have observed a depletion of mtDNA copy number during the bottleneck to minima around several hundred; other studies ( Cao et al . , 2007 , 2009 ) have observed that mtDNA copy number remains >103 . Our study shows that observed increases in heteroplasmy variance ( Jenuth et al . , 1996; Wai et al . , 2008 ) can be achieved across this range of potential minimal mtDNA copy numbers , meaning that the much-debated magnitude of mtDNA copy number reduction is not the sole critical feature of the bottleneck , in agreement with arguments from Cao et al . ( 2007 , 2009 ) ; Wai et al . ( 2008 ) . We find that the role of stochastic mtDNA dynamics can play a key role in determining heteroplasmy variance without additional mechanistic details , in keeping with approaches proposed by Cree et al . ( 2008 ) . The mechanism with the most statistical support is thus consistent with aspects from all existing proposals in the literature . We have shown that , of the models proposed in the literature , a BDP model , proposed after Cree et al . ( 2008 ) and compatible with an interpretation of Wai et al . ( 2008 ) , is the individually most likely mechanism , and capable of producing experimentally observed heteroplasmy behaviour . We cannot , given current experimental evidence , discount hybrid mechanisms , where BDP dominates the population dynamics but small contributions from other mechanisms provide perturbations to this behaviour , and propose experiments to conclusively distinguish between these cases ( see Appendix 1 ) . As the expected statistics of mtDNA populations undergoing inheritance of heteroplasmic mtDNA clusters is very similar to those undergoing binomial partitioning of mtDNAs ( see Appendix 1 ) , the inheritance of heteroplasmic nucleoids ( as opposed to individual mtDNAs ) is not excluded by our findings , though other recent experimental evidence suggests that this situation may be unlikely ( Poe et al . , 2010; Jakobs and Wurm , 2014 ) . We contend that the most likely situation may involve the partitioning of individual organelles , containing a mixture of homoplasmic nucleoids of characteristic size <2 . Notably , this case ( inheritance of heteroplasmic groups , likely with fluid structure due to mixing of organellar content and mitochondrial dynamics ) , gives rise to statistics which our binomial model reproduces ( see Appendix 1 ) . As mentioned in the model description , it is likely that mitochondrial dynamics ( fission and fusion of mitochondria ) ( Detmer and Chan , 2007 ) play a role in determining natural mtDNA turnover , and particularly mtDNA turnover in the presence of pathological mutations ( Nunnari and Suomalainen , 2012 ) , through the mechanism of mitochondrial quality control ( Twig et al . , 2008; Twig and Shirihai , 2011 ) . Mitochondrial dynamics may also influence the elements of partitioning , through changes in the connectivity of the mitochondrial network . In our current model , these influences are coarse-grained into descriptions of the dynamic rates of mtDNA replication and degradation , and the characteristic elements that are partitioned at divisions . These physical parameters , as opposed to the more microscopic details of mitochondrial dynamics , are expected to be the key determinants of heteroplasmy statistics through development . Accounting for how these parameters , which summarize the relevant outputs of mitochondrial dynamics , connect to details of microscopic models of mitochondrial dynamics is an important future research direction to be addressed when more quantitative data is available . The experimental data used to parameterise the first part of our study was taken from four studies in mice . Observation of similar dynamics in salmon ( Wolff et al . , 2011 ) points towards the bottleneck being a conserved mechanism in vertebrates . We also note that our results in mice are broadly consistent with findings from recent experiments in other organisms , suggesting that in primates and humans , heteroplasmy variance may increase at early developmental stages ( Monnot et al . , 2011; Lee et al . , 2012 ) , and that partitioning of mitochondria is binomial in HeLa cells ( Johnston et al . , 2012 ) . As more studies become available on human mtDNA behaviour during development we will test our model's applicability and its clinical predictions . We note that the results of a recent study of human preimplantation sampling ( Treff et al . , 2012 ) found that earlier measurements provided strong predictive power of mean heteroplasmy , about which substantial variation was recorded in the offspring—both of which results are consistent with the application of our model to theoretical sampling considerations . In addition , recent observations that the m . 3243A > G mutation in humans both increases mtDNA copy number during development ( Monnot et al . , 2013 ) , and displays a less pronounced increase of heteroplasmy variance ( Monnot et al . , 2011 ) than other mutations , are consistent with the link between heteroplasmy variance and mtDNA copy number in our theory . The combination of modern stochastic and statistical treatments that we have employed provides a generalisable and powerful way to recapitulate experimental data and rigorously deduce underlying biological mechanisms . We have used this combination to explore pertinent questions regarding the mtDNA bottleneck ( and others have used a similar philosophy to numerically explore mtDNA point mutations [Poovathingal et al . , 2009] ) : we hope to convince the reader that such methodology may be appropriate to explore other problems involving stochastic biological systems . We have used new experimental measurements to confirm our theoretical findings , illustrating the beneficial and powerful coupling of mathematical and experimental approaches to address competing hypotheses in the literature . Our detailed elucidation of the bottleneck allows us to propose further experimental methodology to address the current unknowns in our theory , including the specifics of mtDNA partitioning at cell division and the roles of selective differences between mtDNA types; importantly , we also propose a strategy to investigate our claim that our most supported model is compatible with the subset-replication picture of mtDNA dynamics . We list these experiments in full in Appendix 1 . Finally , we believe that the theoretical foundation for mtDNA dynamics that we have produced allows increased quantitative rigour in the predictions and strategies involved in mtDNA disease therapies , illustrated by the above application of our theory to problems in mtDNA sampling strategies , disease onset timescales , and interventions to increase the power of the bottleneck .
Our ‘bottom-up’ model represents individual mtDNAs as elements which replicate and degrade either randomly or deterministically according to the model parameterisation . Consonant with experimental studies showing that it is often a single mutant genotype that dominates the non-wildtype mtDNA population of a cell ( Khrapko et al . , 1999 ) , we consider two mtDNA types ( wildtype and mutant ) , though our model can readily be extended to more mtDNA types . We denote the number of ‘wild-type’ mtDNAs in a cell as m1 and the number of ‘mutant’ mtDNAs as m2 . The heteroplasmy of a cell is then h=m2m1+m2 , that is , the population proportion of mutant mtDNA . Individual mtDNAs are capable of replication and degradation , with rates denoted λ and ν respectively . According to a binary categorical parameter S , these events may be deterministic ( S = 0; the mtDNA population replicates and degrades by a fixed amount per unit time ) or Poisson processes ( S = 1; each individual mtDNA randomly replicates and degrades with average rates λ and ν ) . A parameter α controls the proportion of mtDNAs capable of replication: α = 1 allows all mtDNAs to replicate throughout development , α < 1 enforces a subset proportion α of replicating mtDNAs a time cutoff T after conception . A parameter c ( cluster size; a non-negative integer ) dictates the partitioning of mtDNAs at cell divisions . When c = 0 , partitioning is deterministic , so each daughter cell receives exactly half of its parent's mtDNA . For c > 0 , partitioning is stochastic . When c = 1 , partitioning is binomial: each mtDNA has a 50% chance of being inherited by either daughter cell . When c > 1 , the parent cell's mtDNAs are grouped in clusters of size c before division . Each cluster is then partitioned binomially , with a 50% chance of being inherited by either daughter cell . The mtDNA population changes in different ways as development progresses , first decreasing , then recovering , then slowly growing . We include the possibility of different ‘phases’ of mtDNA dynamics in our model to capture this behaviour . Each phase j has its own associated pairs of λj , νj parameters and may either be quiescent ( involving no cell divisions ) or cycling ( encompassing nj cell divisions ) . Thus , we may have an initial cycling phase with low mtDNA replication rates , so that copy number falls for several cell divisions , then a subsequent ‘recovery’ cycling phase with higher replication rates so that mtDNA levels are amplified , then quiescent phases as cell lineages are identified . We allow six different phases , with the first two fixed as cycling phases with the above doubling times , and the final phase fixed to include no mtDNA replication ( representing the stable , final occyte state ) . The initial conditions of our model involve an initial mtDNA copy number m0 ( the total number of mtDNAs in the fertilised oocyte ) and an initial heteroplasmy h0 ( the fraction of these mtDNAs that are mutated ) . We used three datasets for mtDNA copy number during mouse development: Cao et al . ( 2007 ) ; Cree et al . ( 2008 ) ; and Wai et al . ( 2008 ) . We use two datasets for heteroplasmy variance during development: Wai et al . ( 2008 ) and Jenuth et al . ( 1996 ) . By convention , we use the normalised versions of heteroplasmy variance ( i . e . , measured variance divided by a factor h ( 1 − h ) ) . Where the measurements were not given explicitly in these publications , we manually analysed the appropriate figures to extract the numerical data . For these values , we used data from correspondence regarding the Wai study ( reply to [Samuels et al . , 2010] ) , and manually normalise the Jenuth dataset . The Jenuth dataset contains measurements taken in ‘mature oocytes’ with no time given; we assume a time of 100 dpc for these measurements , though this time is generalisable and does not qualitatively affect our results . All values are presented in Appendix 1 . Data on cell doubling times in the mouse germ line is taken from Lawson and Hage ( 1994 ) , suggesting that doubling times start with an interval of every 7 hr , then after around 8 . 5 days post conception ( dpc ) increase to 16 hr , before the onset of a quiescent regime around 13 . 5 dpc ( roughly consistent with the estimate of ∼25 divisions between generations in the female mouse germ line [Drost and Lee , 1995] ) . We use Gillespie algorithms , also known as stochastic simulation algorithms ( Gillespie , 1977 ) , to explore the behaviour of our model of the bottlenecking process for a given parameterisation . For a given model parameterisation , the Gillespie algorithm is used to simulate an ensemble of 103 possible realisations of the time evolution of mtDNA content , and the statistics of this ensemble are recorded . The experimental data we use is derived from sets of measurements of different sizes; to compare simulation data with an experimental datapoint i corresponding to a statistic derived from ni measurements , we sampled a random subset of ni of the 103 simulated trajectories ( all datapoints but one have n ≪ 103 ) , and used this subset to derive the simulated statistic for comparison to datapoint i ( Johnston , 2014 ) . To fit the different models to experimental data we define a distance measure , a sum-of-squares residual between the E ( m ) ( in log space ) and V ( h ) dynamics produced by our model and observed in the data , weighted to facilitate comparison of these different quantities ( Johnston , 2014 ) . We also constrain copy number to be <5 × 105 at all points throughout development , rejecting parameterisation that disobey this criterion . Metropolis MCMC was used to identify the best-fit parameterisation according to this distance function . For statistical inference , we use approximate Bayesian computation ( ABC ) , a statistical approach that has successfully been applied to parametric inference and model selection in dynamical systems ( Toni et al . , 2009 ) to infer posterior probability distributions both for individual models and the parameters of the models given experimental data . ABC samples posterior probability distributions on parameters that lead to behaviour within a certain threshold distance of the given data; these posteriors are shown to converge on the true posteriors as the threshold value decreases to zero ( see Appendix 1 ) . We employed an MCMC sampler with randomly-selected initial conditions . For further details , including priors , thresholds and step sizes used in ABC , see Appendix 1 . Minimum copy number was recorded directly from the resulting trajectories; our measure of total turnover σ is defined as σ=∑i=36τ′iνi , the sum over quiescent dynamic phases of the product of degradation rate and phase length . Heteroplasmic mice were obtained from a heteroplasmic mouse line ( HB ) we created previously by ooplasmic transfer ( Burgstaller et al . , 2014 ) . This mouse line contains the nuclear DNA of the C57BL/6N mouse , and mtDNAs both of C57BL/6N and a wild-derived house mouse . Both mtDNA variants belong to the same subspecies , Mus musculus domesticus . For details on sequence divergence ( see Burgstaller et al . , 2014 ) . Mice were sacrificed at the indicated ages by cervical dislocation . Ovaries were extracted and immediately placed in cryo-buffer containing 50% PBS , 25% ethylene glycol and 25% DMSO ( Sigma–Aldrich , Austria ) and stored at −80°C . For oocyte extraction , ovaries were placed into a drop of cryo-buffer and disrupted using scalpel and forceps . Oocytes were collected and remaining cumulus cells were removed mechanically by repeated careful suction through glass capillaries . Prepared oocytes were then washed in PBS before they were individually placed into compartments of 96-well PCR plates ( Life Technologies , Austria ) containing 10 μl of oocyte-lysis buffer ( Lee et al . , 2012 ) ( 50 mM Tris-HCl , [p . H 8 . 5] , 1 mM EDTA , 0 . 5% tween-20 [Sigma–Aldrich , Austria] and 200 μg/ml Proteinase K [Macherey–Nagel , Germany] ) . Samples covered stages from primary oocytes of 3 day-old mice up to mature oocytes of 40 day-old mice . Samples were lysed at 55°C for 2 hr , and incubated at 95°C for 10 min to inactivate Proteinase K . The cellular DNA extract was finally diluted in 190 μl Tris-EDTA buffer , pH 8 . 0 ( Sigma–Aldrich , Austria ) . 3 μl were used per qPCR reaction . Heteroplasmy quantification was performed by ARMS-qPCR , an established method in the field ( Steinborn et al . , 2000; Paull et al . , 2013; Tachibana et al . , 2013 ) , as described in Burgstaller et al . ( 2014 ) . The study was conducted according to MIQE ( minimum information for publication of quantitative real-time PCR experiments ) guidelines ( Bustin et al . , 2009; Burgstaller et al . , 2014 ) . The proportion between HB derived and C57BL/6N mtDNA was determined by ARMS-qPCR assays based on a SNP in mt-rnr2 ( Burgstaller et al . , 2014 ) . These assays were normalised to changes in the input mtDNA amount by consensus assays , located in conserved regions of mt–Co2 and mt–Co3 ( see Appendix 1 ) . For the calculation of mtDNA heteroplasmy , the assay detecting the minor allele ( C57BL/6N or wild-derived <50% ) was always used . If both specific assays gave values >50% ( which can happen around 50% heteroplasmy ) , the mean value of both assays was taken . All qPCR runs contained no template controls ( NTCs ) for all assays; these were negative in 100% . Further experimental details available in Appendix 1 . In the BDP model , processes within a cell cycle constitute a birth-death process which can be solved using generating functions ( Gardiner , 1985 ) . For binomial partitioning , the generating function for the system after an arbitrary number of divisions has a recursive structure ( Rausenberger and Kollmann , 2008; Johnston and Jones , 2015 ) and an analytic solution can be obtained through solving a Riccati recurrence relation . This reasoning also extends to the different phases of replication and degradation , allowing an exact generating function to be constructed for an arbitrary point in the bottleneck . Derivatives of this generating function are then used to obtain moments of the distributions of interest . The full procedure is given in Appendix 1 . Recall that we assume that the bottlenecking process consists of a series of dynamic phases , which may either involve cycling cells ( and hence cell divisions ) or quiescent cells . The expression for mean mtDNA copy number E ( m , t ) at time t is: ( 2 ) E ( m , t ) =m0e ( t−τ* ) ∏phases ie ( niτi+τ′i ) ( λi−νi ) 2ni , where ni is the number of cell divisions in phase i ( 0 for quiescent phases ) , τi is the length of a cell cycle in cycling phase i , τ′i is the time spent in quiescent phase i ( 0 for cycling phases ) , and τ*=Σi ( niτi+τ′i ) , so that t − τ* is the time since the last cell division . E ( m , t ) is thus intuitively interpretable as a product of the initial copy number with the effects of halving at each cell division , and the copy number evolution through past and current cell cycles and quiescent phases . The expression for the variance is lengthier , taking the form ( 3 ) V ( m , t ) =ΦE ( m , t ) ∏phases i4ni ( e ( λi−νi ) τi−2 ) 2 ( λi−νi ) 2+E ( m , t ) −E ( m , t ) 2 , where Φ is a lengthy , though algebraically simple , function of all physical parameters , which we derive and present in Appendix 1 . Once the means and variances associated with mutant and wild-type mtDNAs have been determined ( for brevity , we write these as μ1≡E ( m1 , t ) , σ12≡V ( m1 , t ) and μ2≡E ( m2 , t ) , σ22≡V ( m2 , t ) ) , the relations below can be used to compute heteroplasmy statistics: ( 4 ) E ( h ) =μ2μ1+μ2≡μh , ( 5 ) V ( h ) =μh2 ( σ22μ22−2σ22μ2 ( μ1+μ2 ) +σ12+σ22 ( μ1+μ2 ) 2 ) . The predicted mean heteroplasmy at time t assuming a constant selective pressure ( though this assumption can straightforwardly be relaxed ) is given by Equation 4 , which , given Equation 2 , straightforwardly reduces to ( 6 ) E ( h ) =11+1−h0h0e−Δλt , where h0 is initial heteroplasmy and Δλ is the increase ( or decrease , if negative ) in replication rate of mutant over wild-type mtDNA . Equation 6 predicts that mean heteroplasmy in the presence of selection will follow a sigmoidal form ( as expected from population dynamics [Futuyma , 1997] , by the constraint that h0 must lie between 0 and 1 , and by the intuitive fact that heteroplasmy changes slow down as these limits are approached ) . The probability of heteroplasmy exceeding a certain threshold h* is simply given by integrating the probability distribution of heteroplasmy between h* and 1 . The exact distribution of heteroplasmy can be written as a sum over hypergeometric functions; however , for computational efficiency and interpretability , we employ an approximation to this distribution involving the truncated Normal distribution and fixation probabilities . As shown in Appendix 1 , the distribution of heteroplasmy , taking possible fixation into account , can be well approximated by ( 7 ) P ( h ) = ( 1−ζ1−ζ2 ) N′ ( h|μ , σ2 ) +ζ1δ ( h ) +ζ2δ ( h−1 ) , where N′ is the truncated Normal distribution ( truncated at 0 and 1 ) , μ and σ2 are found numerically given our model results for E ( h ) and V ( h ) , and ζ1≡P ( h=0 ) and ζ2≡P ( h=1 ) are fixation probabilities , also straightforwardly calculable from our model . The probability of threshold crossing for 0 < h* < 1 is then ( 8 ) P ( h>h* ) = ( 1−ζ1−ζ2 ) ( 1−12 ( 1+erf ( ( h*−E ( h ) ) /2V ( h ) ) ) ) +ζ2 . Given a sampled measurement heteroplasmy hm , the probability P ( h0|hm ) that embryonic heteroplasmy is h0 is given by Bayes' theorem P ( h0|hm ) =P ( hm|h0 ) P ( h0 ) /P ( hm ) . Assuming a uniform prior distribution on embryonic heteroplasmy ( though this can be straightforwardly generalised ) , we thus obtain P ( h0|hm ) =P ( hm|h0 ) /∫01P ( hm|h0′ ) dh0′ , and using the above expression for the heteroplasmy , ( 9 ) P ( h0|hm ) = ( 1−ζ1−ζ2 ) N′ ( hm|μ , σ2 ) +ζ1δ ( hm ) +ζ2δ ( hm−1 ) ∫01dh0′ ( 1−ζ1−ζ2 ) N′ ( hm|μ , σ2 ) +ζ1δ ( hm ) +ζ2δ ( hm−1 ) , where μ , σ2 , ζ1 , ζ2 are functions of h0: μ , σ2 may be found numerically and the ζ values are analytically calculable ( see Appendix 1 ) . | Mitochondria are structures that provide vital sources of energy in our cells . DNA contained within mitochondria encodes important mitochondrial machinery , and most human cells contain hundreds or thousands of mitochondrial DNA molecules in addition to the DNA that is stored in the nucleus . Mitochondrial DNA is inherited from mothers via the egg , and the details of this inheritance are poorly understood . This question is important because inherited mistakes in mitochondrial DNA can have detrimental consequences on health , with links to fatal diseases and many other conditions . An unfertilised egg cell contains many copies of mitochondrial DNA molecules; some may have mutations and some may not . After fertilisation , the egg divides , the number of cells in the developing embryo increases , and the number of mitochondrial DNA molecules per cell changes . If the original egg cell contained defective mitochondrial DNA , some of these new cells end up containing more defective copies than others , leading to cell-to-cell differences in the developing embryo . This potentially allows cells with the greatest number of defective mitochondria to be eliminated . The increase in this cell-to-cell variability is called ‘bottlenecking’ , and its mechanism remains highly debated . Johnston et al . have now used tools from maths , statistics and new experiments to address this debate , in the light of several studies that measured the mitochondrial DNA content in developing mice . This approach allowed a new theoretical model of mitochondrial DNA during the growth of an organism to be produced , which encompasses a wide range of existing theories and allows them to be compared . This model starts from the viewpoint that the hundreds or thousands of mitochondrial DNA molecules in a cell can be thought of as a population undergoing random ‘birth’ and ‘death’ , and it allows the first statistical comparison of the many proposed bottleneck mechanisms . Johnston et al . find support for two ways that cells segregate mitochondria as they multiply , and show that the decrease in the number of mitochondrial DNA molecules during bottlenecking is flexible . This reconciles a debate amongst previous studies . These findings are confirmed using new experimental data from mice , which are genetically distinct from existing studies , illustrating the generality of the model's findings . Furthermore , an analytic mathematical description that describes in detail how bottlenecking might work is produced . Finally , Johnston et al . provide examples using this new theoretical model to suggest therapeutic strategies for diseases caused by mitochondrial DNA mutations . Future work will need to test these suggestions , and link mathematical understanding of mitochondria with healthcare data . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"computational",
"and",
"systems",
"biology"
] | 2015 | Stochastic modelling, Bayesian inference, and new in vivo measurements elucidate the debated mtDNA bottleneck mechanism |
How do we choose when confronted with many alternatives ? There is surprisingly little decision modelling work with large choice sets , despite their prevalence in everyday life . Even further , there is an apparent disconnect between research in small choice sets , supporting a process of gaze-driven evidence accumulation , and research in larger choice sets , arguing for models of optimal choice , satisficing , and hybrids of the two . Here , we bridge this divide by developing and comparing different versions of these models in a many-alternative value-based choice experiment with 9 , 16 , 25 , or 36 alternatives . We find that human choices are best explained by models incorporating an active effect of gaze on subjective value . A gaze-driven , probabilistic version of satisficing generally provides slightly better fits to choices and response times , while the gaze-driven evidence accumulation and comparison model provides the best overall account of the data when also considering the empirical relation between gaze allocation and choice .
In everyday life , we are constantly faced with value-based choice problems involving many possible alternatives . For instance , when choosing what movie to watch or what food to order off a menu , we must often search through a large number of alternatives . While much effort has been devoted to understanding the mechanisms underlying two-alternative forced choice ( 2AFC ) in value-based decision-making ( Alós-Ferrer , 2018; Bhatia , 2013; Boorman et al . , 2013; Clithero , 2018; De Martino et al . , 2006; Hare et al . , 2009; Hunt et al . , 2018; Hutcherson et al . , 2015; Krajbich et al . , 2010; Mormann et al . , 2010; Philiastides and Ratcliff , 2013; Polanía et al . , 2019; Rodriguez et al . , 2014; Webb , 2019 ) and choices involving three to four alternatives ( Berkowitsch et al . , 2014; Diederich , 2003; Gluth et al . , 2018; Gluth et al . , 2020; Krajbich and Rangel , 2011; Noguchi and Stewart , 2014; Roe et al . , 2001; Towal et al . , 2013; Trueblood et al . , 2014; Usher and McClelland , 2004 ) , comparably little has been done to investigate many-alternative forced choices ( MAFC , more than four alternatives ) ( Ashby et al . , 2016; Payne , 1976; Reutskaja et al . , 2011 ) . Prior work on 2AFC has indicated that simple value-based choices are made through a process of gaze-driven evidence accumulation and comparison , as captured by the attentional drift diffusion model ( Krajbich et al . , 2010; Krajbich and Rangel , 2011; Smith and Krajbich , 2019 ) and the gaze-weighted linear accumulator model ( GLAM; Thomas et al . , 2019 ) . These models assume that noisy evidence in favour of each alternative is compared and accumulated over time . Once enough evidence is accumulated for one alternative relative to the others , that alternative is chosen . Importantly , gaze guides the accumulation process , with temporarily higher accumulation rates for looked-at alternatives . One result of this process is that longer gaze towards one alternative should generally increase the probability that it is chosen , in line with recent empirical findings ( Amasino et al . , 2019; Armel et al . , 2008; Cavanagh et al . , 2014; Fisher , 2017; Folke et al . , 2017; Gluth et al . , 2018; Gluth et al . , 2020; Konovalov and Krajbich , 2016; Pärnamets et al . , 2015; Shimojo et al . , 2003; Stewart et al . , 2016; Vaidya and Fellows , 2015 ) . While this framework can in theory be extended to MAFC ( Gluth et al . , 2020; Krajbich and Rangel , 2011; Thomas et al . , 2019; Towal et al . , 2013 ) , it is still unknown whether it can account for choices from truly large choice sets . In contrast , past research in MAFC suggests that people may resort to a ‘satisficing’ strategy . Here , the idea is that people set a minimum threshold on what they are willing to accept and search through the alternatives until they find one that is above that threshold ( McCall , 1970; Simon , 1955; Simon , 1956; Simon , 1957; Simon , 1959; Schwartz et al . , 2002; Stüttgen et al . , 2012 ) . Satisficing has been observed in a variety of choice scenarios , including tasks with a large number of alternatives ( Caplin et al . , 2011; Stüttgen et al . , 2012 ) , patients with damage to the prefrontal cortex ( Fellows , 2006 ) , inferential decisions ( Gigerenzer and Goldstein , 1996 ) , survey questions ( Krosnick , 1991 ) , risky financial decisions ( Fellner et al . , 2009 ) , and with increasing task complexity ( Payne , 1976 ) . Past work has also investigated MAFC under strict time limits ( Reutskaja et al . , 2011 ) . There , the authors find that the best model is a probabilistic version of satisficing in which the time point when individuals stop their search and make a choice follows a probabilistic function of elapsed time and cached ( i . e . , highest-seen ) item value ( Chow and Robbins , 1961; Rapoport and Tversky , 1966; Robbins et al . , 1971; Simon , 1955; Simon , 1959 ) . There is some empirical evidence that points towards a gaze-driven evidence accumulation and comparison process for MAFC . For instance , individuals look back and forth between alternatives as if comparing them ( Russo and Rosen , 1975 ) . Also , frequently looking at an item dramatically increases the probability of choosing that item ( Chandon et al . , 2009 ) . Empirical evidence has further indicated that individuals use a gaze-dependent evidence accumulation process when making choices from sets of up to eight alternatives ( Ashby et al . , 2016 ) . Here , we sought to study the mechanisms underlying MAFC , by developing and comparing different versions of these models on choice , response time ( RT ) , liking rating , and gaze data from a choice task with sets of 9 , 16 , 25 , and 36 snack foods . These models combine an either passive or active account of gaze in the decision process with three distinct accounts of the decision mechanism , namely probabilistic satisficing and two variants of evidence accumulation , which either perform relative comparisons between the alternatives or evaluate each alternative independently . In terms of overall goodness-of-fit , we find that the models with active gaze consistently outperform their passive-gaze counterparts . That is , gaze does more than bring an alternative into the consideration set , it actively increases the subjective value of the attended alternative relative to the others . The probabilistic satisficing model ( PSM ) performs slightly better than the other models at capturing individuals’ choices and RTs , while the relative accumulator model provides the best overall account of the data after considering the observed positive relation between gaze allocation and choice .
In each of 200 choice trials , subjects ( N = 49 ) chose which snack food they would like to eat at the end of the experiment , out of a set of either 9 , 16 , 25 , or 36 alternatives ( 50 trials per set size condition; see Figure 1 and Materials and methods ) . We recorded subjects’ choices , RTs , and eye movements . After the choice task , subjects also rated each food on an integer scale from −3 ( i . e . , not at all ) to 3 ( i . e . , very much ) to indicate how much they would like to eat each item at the end of the experiment ( for an overview of the liking rating distributions , see Figure 1—figure supplements 1 , 2 ) . We use these liking ratings as a measure of item value . To first establish a general understanding of the visual search process in MAFC , we performed an exploratory analysis of subjects’ visual search behaviour ( Figures 2 and 3 ) . We define a gaze to an item as all consecutive fixations towards the item that happen without any interrupting fixation to other parts of the choice screen . Furthermore , we define the cumulative gaze of an item as the fraction of total trial time that the subject spent looking at the item ( see Materials and methods ) . All reported regression coefficients represent fixed effects from mixed-effects linear ( for continuous dependent variables ) and logistic ( for binary dependent variables ) regression models , which included random intercepts and slopes for each subject ( unless noted otherwise ) . The 94% highest density intervals ( HDI; 94% is the default in ArviZ 0 . 9 . 0 [Kumar et al . , 2019] which we used for our analyses ) of the fixed-effect coefficients are given in brackets , unless noted otherwise ( see Materials and methods ) . The probability that participants looked at an item in a choice set increased with the item’s liking rating , while decreasing with set size ( Figure 2A–D; β = 2 . 0% , 94% HDI = [1 . 6 , 2 . 3] per rating , β = −1 . 4% , 94% HDI = [−1 . 5 , –1 . 3] per item ) ( in line with recent empirical findings: Cavanagh et al . , 2019; Gluth et al . , 2020 ) . Similarly , the probability that participants’ gaze returned to an item increased with the item’s rating while decreasing with set size ( Figure 2A–D; β = 1 . 6% , 94% HDI = [1 . 4 , 1 . 8] per rating , β = −0 . 65% , 94% HDI = [−0 . 74 , –0 . 55] per item ) . Gaze durations also increased with the item’s rating ( Figure 2E–H; β = 11 ms , 94% HDI = [8 , 13] per rating ) as well as over the course of a trial ( β = 0 . 79 ms , 94% HDI = [0 . 36 , 1 . 25] per additional gaze in a trial ) , while decreasing with set size ( β = −1 . 17 ms , 94% HDI = [−1 . 39 , –0 . 94] per item ) . Initial gazes to an item were generally shorter in duration than all later gazes to the same item in the same trial ( Figure 2I–L; β = –44 ms , 94% HDI = [37 , 51] difference between initial and returning gazes ) . Interestingly , the duration of the last gaze in a trial was dependent on whether it was to the chosen item or not ( Figure 2I–L ) : last gaze durations to the chosen item were in general longer than last gaze durations to non-chosen items ( β = 162 ms , 94% HDI = [122 , 201] difference between last gazes to chosen and non-chosen items ) . Next , we focused on subjects’ visual search trajectories ( Figure 3 ) : For each trial , we first normalized time to a range from 0 to 100% and then binned it into 10% intervals . We then extracted the liking rating , position , and size for each item in a trial ( see Materials and methods ) . An item’s position was encoded by its column and row indices in the square grid ( Figure 1; with indices increasing from left to right and top to bottom ) . All item attributes were centred with respect to their trial mean in the choice set ( e . g . , a centred row index of −1 in the set size with nine items represents the row one above the centre , whereas a centred item rating of −1 represents a rating one below the average of all item ratings in that choice set ) . For each normalized time bin , we computed a mixed-effects logistic regression model ( see Materials and methods ) , regressing the probability that an item was looked at onto its attributes . In general , subjects began their search at the centre of the screen ( Figure 3A , B; as indicated by regression coefficients close to 0 for the items’ row and column positions in the beginning of a trial ) , coinciding with the preceding fixation cross . Subjects then typically transitioned to the top left corner ( Figure 3A , B; as indicated by increasingly negative regression coefficients for the items’ row and column positions in the beginning of a trial ) and then moved from top to bottom ( Figure 3B; as indicated by the then increasingly positive regression coefficients for the items’ row positions ) . Over the course of the trial , subjects generally focused their search more on highly rated ( Figure 3C ) and larger ( Figure 3D ) items , while the probability that their gaze returned to an item also steadily increased ( Figure 3E; β = 9 . 9% , 94% HDI = [9 . 2 , 10 . 7] per second , β = −0 . 73 , 94% HDI = [−0 . 80 , –0 . 66] per item ) , as did the durations of these returning gazes ( Figure 3F; β = 14 ms , 94% HDI = [12 , 17] per second , β = −2 . 9 ms , 94% HDI = [−3 . 3 , –2 . 5] per item ) . In general , the effects of item position and size on the search process decreased over time ( Figure 3A , B , D ) . For exemplar visual search trajectories in each set size condition , see Animations 1–4 . Overall , the fraction of total trial time that subjects looked at an item was dependent on the liking rating , size , and position of the item , as well as the number of items contained in the choice set ( β = 0 . 5% , 94% HDI = [0 . 4 , 0 . 6] per liking rating , β = 0 . 02% , 94% HDI = [0 . 008 , 0 . 03] per percentage increase in size , β = −0 . 20% , 94% HDI = [−0 . 24 , –0 . 15] per row position , β = −0 . 044 , 94% HDI = [−0 . 075 , –0 . 007] per column position , β = −0 . 177 , 94% HDI = [−0 . 18 , –0 . 174] per item ) . We also tested whether these item attributes influenced subjects’ choice behaviour . However , the probability of choosing an item did not depend on the size or position of the item , but was solely dependent on the item’s liking rating and the set size ( β = 3 . 9 , 94% HDI = [3 . 5 , 4 . 3] per liking rating , β = 0 . 02 , 94% HDI = [−0 . 015 , 0 . 06] per percentage increase in item size , β = −0 . 06 , 94% HDI = [−0 . 12 , 0 . 01] per row , β = −0 . 03 , 94% HDI = [−0 . 1 , 0 . 03] per column , β = −0 . 24 , 94% HDI = [−0 . 25 , –0 . 23] per item ) . We consider the following set of decision models , spanning the space between rational choice and gaze-driven evidence accumulation . The optimal choice model with zero search costs is based on the framework of rational decision-making ( Luce and Raiffa , 1957; Simon , 1955 ) . It assumes that individuals look at all the items of a choice set and then choose the best seen item with a fixed probability β , while making a probabilistic choice over the set of seen items ( J ) with probability 1-β following a softmax choice rule ( σ , with inverse temperature parameter τ ) based on the items’ values ( l ) : σi=exp ( τ×li ) ∑j∈Jexp ( τ×lj ) . The hard satisficing model assumes that individuals search until they either find an item with reservation value V or higher , or they have looked at all items ( Caplin et al . , 2011; Fellows , 2006; McCall , 1970; Payne , 1976; Schwartz et al . , 2002; Simon , 1955; Simon , 1956; Simon , 1957; Simon , 1959; Stüttgen et al . , 2012 ) . In the former case , individuals immediately stop their search and choose the first item that meets the reservation value . Crucially , the reservation value can vary across individuals and set-size conditions . In the latter case , individuals make a probabilistic choice over the set of seen items , as in the optimal choice model . Based on the findings by Reutskaja et al . , 2011 , we also considered a probabilistic version of satisficing , which combines elements from the optimal choice and hard satisficing models . Specifically , the PSM assumes that the probability with which individuals stop their search and make a choice at a given time point increases with elapsed time in the trial and the cached ( i . e . , highest-seen ) item value . Once the search ends , individuals make a probabilistic choice over the set of seen items , as in the other two models ( see Materials and methods ) . Next , we considered an independent evidence accumulation model ( IAM ) , in which evidence for an item begins accumulating once the item is looked at ( Smith and Vickers , 1988 ) . Importantly , each accumulator evolves independently from the others , based on the subjective value of the represented item . Once the accumulated evidence for an alternative reaches a predefined decision threshold , a choice is made for that alternative ( much like deciding whether the item satisfies a reservation value ) ( see Materials and methods ) . In line with many empirical findings ( e . g . , Krajbich et al . , 2010; Krajbich and Rangel , 2011; Lopez-Persem et al . , 2016; Tavares et al . , 2017; Smith and Krajbich , 2019; Thomas et al . , 2019 ) , we also considered a relative evidence accumulation model ( as captured by the GLAM; Thomas et al . , 2019; Molter et al . , 2019 ) , which assumes that individuals accumulate and compare noisy evidence in favour of each item relative to the others . As with the IAM , a choice is made as soon as the accumulated relative evidence for an item reaches a predetermined decision threshold ( see Materials and methods ) . We further considered two different accounts of gaze in the decision process . The passive account of gaze assumes that gaze allocation solely determines the set of items that are being considered; an item is only considered once it is looked at . In contrast , the active account of gaze assumes that gaze influences the subjective value of an item in the decision process , thereby generating higher choice probabilities for items that are looked at longer . In the PSM , gaze time increases the subjective value of an item relative to the others . Similarly , in the accumulator models , the accumulation rate for an item ( reflecting subjective value ) increases relative to the others when the item is being looked at ( for positively valued items ) . Recent empirical findings indicate two distinct mechanisms through which gaze might actively influence these decision processes: multiplicative effects ( Krajbich et al . , 2010; Krajbich and Rangel , 2011; Lopez-Persem et al . , 2016 ; Tavares et al . , 2017; Smith and Krajbich , 2019; Thomas et al . , 2019 ) and additive effects ( Cavanagh et al . , 2014; Westbrook et al . , 2020 ) . Multiplicative effects discount the subjective values of unattended items ( by multiplying them with γ;0≤γ≤1 ) , while additive effects add a constant boost ( ζ;0≤ζ≤10 ) to the subjective value of the attended item . Thus , multiplicative effects are proportional to the values of the items , while additive effects are constant for all items and independent of an item’s value . We allow for both of these mechanisms in the modelling of the active influence of gaze on the decision process ( see Materials and methods ) . First , we probed the assumptions of the optimal choice model with zero search costs , which predicts that subjects first look at all the items in a choice and then choose the highest-rated item at a fixed rate . Conditional on the set of looked-at items , subjects chose the highest-rated item at a very consistent rate across set sizes ( Figure 4A; β = 0 . 05% , 94% HDI = [−0 . 04 , 0 . 14] per item ) , with an overall average of 84% . However , subjects did not look at all food items in a given trial ( Figure 4B ) , while the fraction of items in a choice set that subjects looked at decreased across set sizes ( Figure 4B; β = −1 . 52% , 94% HDI = [−1 . 60 , –1 . 45] per item ) and their mean RTs increased ( Figure 4C; β = 85 ms , 94% HDI = [67 , 102] per item ) . This immediately ruled out a strict interpretation of the optimal choice model , as subjects did not look at all items before making a choice . Next , we tested the assumptions of the hard satisficing model , which predicts that subjects should stop their search and make a choice as soon as they find an item that meets their acceptance threshold . Accordingly , the last item that subjects look at should be the one that they choose ( unless they look at every item ) . However , across set sizes , subjects only chose the last item that they looked at in 44 . 6% of the trials ( Figure 4D; β = 0 . 14% , 94% HDI = [−0 . 001 , 0 . 26] per item ) . Even within the trials where subjects did not look at every item , the probability that they chose the last-seen item was on average only 44 . 1% . The PSM , on the other hand , predicts that the probability with which subjects stop their search and make a choice increases with elapsed time and cached value ( i . e . , the highest-rated item seen so far in a trial ) . We found that both had positive effects on subjects’ stopping probability , in addition to a negative effect of set size ( β = 2 . 7% , 94% HDI = [2 . 0 , 3 . 3] per cached value , β = 2 . 26% , 94% HDI = [1 . 69 , 2 . 80] per second , β = −0 . 22% , 94% HDI = [−0 . 24 , –0 . 20] per item ) . Subjects’ behaviour was therefore qualitatively in line with the basic assumptions of the PSM . Note that this finding does not allow us to discriminate between the PSM and evidence accumulation models , because both make very similar qualitative predictions about the relationship between stopping probability , time , and item value . Last , we probed the behavioural association of gaze allocation and choice . To this end , we utilized a previously proposed measure of gaze influence ( Krajbich et al . , 2010; Krajbich and Rangel , 2011; Thomas et al . , 2019 ) : First , we regressed a choice variable for each item in a trial ( 1 if the item was chosen , 0 otherwise ) on three predictor variables: the item’s relative liking rating ( the difference between the item’s rating and the mean rating of all other items in that set ) as well as the mean and range of the other items’ liking ratings in that trial . We then subtracted this choice probability for each item in each trial from the empirically observed choice ( 1 if the item was chosen , 0 otherwise ) . This yields residual choice probabilities after accounting for the distribution of liking ratings in each trial . Finally , we computed the mean difference in these residual choice probabilities between items with positive and negative cumulative gaze advantages ( defined as the difference between an item’s cumulative gaze and the maximum cumulative gaze of any other item in that trial ) . This measure is a way to quantify the average increase in choice probability for the item that is looked at longest in each trial . We found that all subjects exhibited positive values on this measure in all set sizes ( Figure 4E; with values ranging from 1 . 7% to 75% ) and that it increased with set size ( Figure 4E; β = 0 . 26% , 94% HDI = [0 . 15 , 0 . 39] per item ) , indicating an overall positive association between gaze allocation and choice . In general , a subject’s probability of choosing an item increased with the item’s cumulative gaze advantage and the item’s relative rating , while it decreased with the range of the ratings of the other items in a choice set and set size ( β = 0 . 46% , 94% HDI = [0 . 4 , 0 . 5] per percentage increase in cumulative gaze advantage , β = 3 . 6% , 94% HDI = [3 . 2 , 4 . 0] per unit increase in relative rating , β = −2 . 8% , 94% HDI = [−3 . 1 , –2 . 4] per unit increase in the range of ratings of the other items , β = −0 . 16 , 94% HDI = [−0 . 18 , –0 . 14] per item ) . To further probe the assumption of gaze-driven evidence accumulation , we performed three additional tests: According to the framework of gaze-driven evidence accumulation , subjects with a stronger association of gaze and choice should generally also exhibit a lower probability of choosing the highest-rated item from a choice set ( for a detailed discussion on this finding , see Thomas et al . , 2019 ) . For these subjects , the gaze bias mechanism can bias the decision process towards items that have a lower value but were looked at longer over the course of a trial . In line with this prediction , we found that probability of choosing the highest-rated seen item was negatively correlated with the gaze influence measure ( β = −0 . 22% , 94% HDI = [−0 . 36 , –0 . 08] per percentage increase in gaze influence; the mixed-effects regression included a random slope and intercept for each set size ) . Second , subjects with a stronger association between gaze and choice should be more likely to choose the last-seen item , as evidence for the looked-at item is generally accumulated at a higher rate . In line with this prediction , subjects with higher values on the gaze influence measure ( indicating stronger gaze bias ) were also more likely to choose the item that they looked at last in a trial ( β = 1 . 1% , 94% HDI = [0 . 9 , 1 . 3] per percentage increase in gaze influence; the mixed-effects regression included a random slope and intercept for each set size ) . Last , subjects with a stronger association of gaze and choice should be more likely to choose an item when it receives longer individual gazes . In line with previous work ( e . g . , Krajbich et al . , 2010; Krajbich and Rangel , 2011 ) , we investigated this by studying the probability of choosing the first-seen item in a trial as a function of the duration of the first gaze in that trial . Overall , this relationship was positive ( as was the influence of the item’s rating on choice probability ) , while the item’s choice probability decreased with set size ( β = 18% , 94% HDI = [14 , 22] per second , β = 6 . 0% , 94% HDI = [5 . 5 , 6 . 6] per rating , β = −0 . 27% , 94% HDI = [−0 . 32 , –0 . 22] per item ) . To better understand the relation between visual search and choice behaviour , we also studied the association of the influence of an item’s size , rating , and position on gaze allocation with the metrics of choice behaviour reported in Figure 4 ( namely , mean RT , fraction of looked-at items , probability of choosing the highest-rated seen item , and gaze influence on choice ) ( Figure 4—figure supplement 1 ) . To quantify the influence of the item attributes on gaze allocation , we ran a regression for each subject of cumulative gaze ( defined as the fraction of trial time that the subject looked at an item; scaled 0–100% ) onto the four item attributes ( row , column , size , and rating ) and set size , resulting in one coefficient estimate ( βgaze ) for the influence of each of the item attributes and set size on cumulative gaze . Subjects with a stronger influence of rating on gaze allocation generally looked at fewer items ( β = −17% , 94% HPI = [−30 , –5] per unit increase in βgaze ( rating ) ; Figure 4—figure supplement 1H ) , were more likely to choose the highest-rated seen item ( β = 14% , 94% HDI = [5 , 23] per unit increase in βgaze ( rating ) ; Figure 4—figure supplement 1P ) , and were more likely to choose the last-seen item ( β = 40% , 94% HDI = [19 , 61] per unit increase in βgaze ( rating ) ; Figure 4—figure supplement 1T ) . Subjects with a stronger influence of item size on gaze allocation generally looked at fewer items ( β = −113% , 94% HDI = [−210 , –6] per unit increase in βgaze ( size ) ; Figure 4—figure supplement 1G ) , exhibited shorter RTs ( β = −18 s , 94% HDI = [−32 , –4] per unit increase in βgaze ( size ) ; Figure 4—figure supplement 1K ) , and were less likely to choose the last-seen item ( β = −189% , 94% HDI = [−377 , –2] per unit increase in βgaze ( size ) ; Figure 4—figure supplement 1S ) . Lastly , subjects with a stronger influence of column number ( horizontal location ) on gaze allocation generally exhibited longer RTs ( β = 3 . 96 s , 94% HDI = [0 . 40 , 7 . 83] per unit increase in βgaze ( column ) ; Figure 4—figure supplement 1J ) . This last effect mainly results from two behavioural patterns in the data: Subjects generally began their visual search in the upper-left corner of the screen ( Figure 3A , B ) , resulting in more gaze for items that were located on the left side of the screen ( i . e . , with low column numbers ) . Thus , subjects who quickly decided also generally looked more at items on the left ( resulting in a negative influence of column number on cumulative gaze ) . In contrast , slower subjects generally displayed more balanced looking patterns ( resulting in no influence of column number on cumulative gaze ) . Taken together , these effects produce a positive association of mean RT and the influence of column number on cumulative gaze . We did not find any other statistically meaningful associations between visual search and choice metrics ( Figure 4—figure supplement 1 ) . Taken together , our findings have shown that subjects’ choice behaviour in MAFC does not match the assumptions of optimal choice or hard satisficing , while it qualitatively matches the assumptions of probabilistic satisficing and gaze-driven evidence accumulation . To further discriminate between the evidence accumulation and probabilistic satisficing models , we fitted them to each subject’s choice and RT data for each set size ( see Materials and methods; for an overview of the parameter estimates , see Figure 5—figure supplement 1 and Supplementary files 1–3 ) and compared their fit by means of the widely applicable information criterion ( WAIC; Vehtari et al . , 2017 ) . Importantly , we tested two variants of each of these models , one with a passive account of gaze in which gaze allocation solely determines the set of items that are considered in the decision process , and the other with an active account of gaze in which gaze affects the subjective value of the alternatives . In the active-gaze models ( as indicated by the addition of a ‘+' to the model name ) , we allowed for both multiplicative and additive effects of gaze on the decision process ( see Materials and methods ) . The model variants with a passive and active account of gaze were identical , other than for these two influences of gaze on subjective value . Note that all three model types can be recovered to a satisfying degree in our data ( Figure 5—figure supplement 2 ) . According to the WAIC , the choices and RTs of the vast majority of subjects were best captured by the model variants with an active account of gaze ( 82% [40/49] , 94% [46/49] , 90% [44/49] , and 86% [42/49] for 9 , 16 , 25 , and 36 items respectively; Figure 5A–D ) . Specifically , the PSM+ won the plurality of individual WAIC comparisons in each set size ( 39% [19/49] , 65% [32/49] , 47% [23/49] , and 51% [25/49] in the sets with 9 , 16 , 25 , and 36 items , respectively ) , while the plurality of the remaining WAIC comparisons was won by the GLAM+ for 9 and 16 items ( 29% [14/49] and 16% [8/49] subjects , respectively ) and by the IAM+ for 25 and 36 items ( 22% [11/49] and 24% [12/49] subjects , respectively ) . To further test whether there was a winning model for each set size , we performed a comparison of the distributions of individual WAIC values resulting from each of the three active-gaze model variants . We used two-sided Mann–Whitney U tests with a Bonferroni adjusted alpha level of 0 . 0042 per test ( 0 . 05/12 ) ( Figure 5E–H ) . This analysis revealed that the WAIC distributions of the PSM+ and GLAM+ were not meaningfully different from one another in any set size ( U = 1287 , p = 0 . 54; U = 1444 , p=0 . 08; U = 1460 , p = 0 . 07; U = 1469 , p = 0 . 06 for 9 , 16 , 25 , and 36 items ) , while the PSM+ was meaningfully better than the IAM+ for 16 items ( U = 1705 , p = 0 . 0003 ) , but not for 9 ( U = 1592 , p = 0 . 005 ) , 25 ( U = 1573 , p = 0 . 008 ) , or 36 ( U = 1508 , p = 0 . 029 ) items . The GLAM+ was not meaningfully better than the IAM+ in any set size ( U = 1518 , p = 0 . 02; U = 1507 , p = 0 . 03; U = 1388 , p = 0 . 18; U = 1289 , p = 0 . 53 for 9 , 16 , 25 , and 36 items ) . More generally , the difference in WAIC between the PSM+ and IAM+ as well as the PSM+ and GLAM+ increased with set size ( β = 0 . 55 , 94% HDI = [0 . 16 , 0 . 93] per item for the IAM+ and β = 0 . 86 , 94% HDI = [0 . 62 , 1 . 12] per item for the GLAM+ ) , indicating that the PSM+ provides an increasingly better fit to the choice and RT data as the set size increases . Notably , the corresponding fixed-effects intercept estimate was larger than 0 for the IAM+ ( 20 , 94% HDI = [13 , 27] ) , but not for the GLAM+ ( −0 . 7 , 94% HDI = [−5 . 3 , 3 . 9] ) , suggesting that the PSM+ provides a meaningfully better fit for choice behaviour in smaller sets than the IAM+ , but not than the GLAM+ . Similarly , the WAIC-difference between the GLAM+ and IAM+ did not increase with set size ( β = −0 . 30 , 94% HDI = [−0 . 73 , 0 . 15] per item ) , while the fixed-effects intercept estimate was meaningfully greater than 0 ( 21 , 94% HDI = [15 , 26] ) , indicating that the GLAM+ provides a meaningfully better fit than the IAM+ for smaller sets . Yet , WAIC only tells us about relative model fit . To determine how well each model fit the data in an absolute sense , we simulated data for each individual with each model and regressed the simulated mean RTs , probability of choosing the highest-rated item , and gaze influence on choice probability onto the observed subject values for each of these measures , in a linear mixed-effects regression analysis with one random intercept and slope for each set size ( Figure 6; see Materials and methods ) . If a model captures the data well , the resulting fixed-effects regression line should have an intercept of 0 and a slope of 1 ( as indicated by the black diagonal lines in Figure 6 ) . The PSM+ and GLAM+ both accurately recovered mean RT ( Figure 6A , C; intercept = −138 ms , 94% HDI = [−414 , 119] , β = 1 . 01 ms , 94% HDI = [0 . 95 , 1 . 05] per ms increase in observed RT for the PSM+; intercept = −51 ms , 94% HDI = [−327 , 207] , β = 0 . 97 ms , 94% HDI = [0 . 91 , 1 . 06] per ms increase in observed RT for the GLAM+ ) , while the IAM+ underestimated short and overestimated long mean RTs ( Figure 6B; intercept = −1185 ms , 94% HDI = [−2293 , –61] , β = 1 . 29 ms , 94% HDI = [1 . 19 , 1 . 39] per ms increase in observed RT ) . All three models generally underestimated high probabilities of choosing the highest-rated item from a choice set ( Figure 6D–F ) , while the PSM+ provided the overall most accurate account of this metric ( Figure 6D; intercept = −1 . 90% , 94% HDI = [−7 . 64 , 3 . 19] , β = 0 . 85% , 94% HDI = [0 . 79 , 0 . 91] per percentage increase in observed probability of choosing the highest-rated item ) , followed by the GLAM+ ( Figure 6E; intercept = 11 . 66% , 94% HDI = [5 . 27 , 17 . 41] , β = 0 . 71% , 94% HDI = [0 . 64 , 0 . 78] per percentage increase in observed probability of choosing the highest-rated item ) , and IAM+ ( Figure 6F; intercept = 10 . 48% , 94% HDI = [0 . 15 , 28 . 20] , β = 0 . 33% , 94% HDI = [0 . 15 , 0 . 49] per percentage increase in observed probability of choosing the highest-rated item ) . Turning to the gaze data , the PSM+ and IAM+ both slightly overestimated weak associations between gaze and choice while clearly underestimating stronger associations between them ( Figure 6G–H; intercept = 7 . 03% , 94% HDI = [4 . 95 , 10 . 23] , β = 0 . 48% , 94% HDI = [0 . 38 , 0 . 56] per percentage increase in observed gaze influence for the PSM+; intercept = 6 . 58% , 94% HDI = [4 . 28 , 10 . 30] , β = 0 . 38% , 94% HDI = [0 . 15 , 0 . 49] per percentage increase in observed gaze influence for the IAM+ ) . The GLAM+ , in contrast , only slightly underestimated strong associations of gaze and choice ( Figure 6I; intercept = −0 . 61% , 94% HDI = [−2 . 84 , 1 . 64] , β = 0 . 85% , 94% HDI = [0 . 77 , 0 . 93] per percentage increase in observed gaze influence ) . To better understand the gaze bias mechanism in our data , we studied the gaze bias parameter estimates resulting from the individual model fits . Our modelling of the decision process included two types of gaze biases ( see Materials and methods ) : multiplicative gaze biases ( indicated by γ ) proportionally discount the value of momentarily unfixated items ( smaller γ values thus indicate stronger bias ) , while additive gaze biases ( indicated by ζ ) provide a constant increase to the value of the currently fixated item ( larger ζ values thus indicate stronger bias ) . We found evidence for both of these gaze biases in the PSM+ and GLAM+ across all set sizes , with mean ( s . d . ) γ estimates ( per set size ) of 9: 0 . 63 ( 0 . 29 ) , 16: 0 . 53 ( 0 . 27 ) , 25: 0 . 57 ( 0 . 27 ) , and 36: 0 . 59 ( 0 . 29 ) for the PSM+ ( Supplementary file 1 ) and 9: 0 . 72 ( 0 . 28 ) , 16: 0 . 64 ( 0 . 27 ) , 25: 0 . 68 ( 0 . 27 ) , and 36: 0 . 71 ( 0 . 27 ) for the GLAM+ ( Supplementary file 3 ) . Similarly , mean ( s . d . ) ζ estimates were 9: 1 . 29 ( 1 . 62 ) , 16: 1 . 73 ( 1 . 60 ) , 25: 1 . 42 ( 1 . 90 ) , and 36: 1 . 09 ( 1 . 20 ) for the PSM+ ( Supplementary file 1 ) and 9: 2 . 20 ( 2 . 13 ) , 16: 2 . 88 ( 2 . 28 ) , 25: 2 . 64 ( 2 . 57 ) , and 36: 2 . 38 ( 2 . 22 ) for the GLAM+ ( Supplementary file 3 ) . Interestingly , we did not find any evidence for additive gaze biases in the IAM+ ( with mean [s . d . ] ζ estimates of 9: 0 . 23 [0 . 69] , 16: 0 . 20 [0 . 73] , 25: 0 . 32 [0 . 98] , and 36: 0 . 31 [0 . 72]; Supplementary file 2 ) , while multiplicative gaze biases were more pronounced ( as indicated by mean [s . d . ] γ estimates of 9: 0 . 01 [0 . 05] , 16: 0 . 005 [0 . 008] , 25: 0 . 02 [0 . 05] , and 36: 0 . 02 [0 . 09]; Supplementary file 2 ) . However , the IAM+ generally also provided the worst fit to individuals’ association of gaze and choice ( Figure 6H ) . We further studied the correlation between individual γ and ζ estimates to better understand the association of multiplicative and additive gaze biases ( Figure 7 ) . In the PSM+ and GLAM+ , γ estimates generally decreased with increasing ζ estimates ( β = −0 . 06 , 94% HDI = [−0 . 09 , –0 . 03] per unit increase in ζ for the PSM+ ( Figure 7A ) ; β = −0 . 08 , 94% HDI = [−0 . 10 , –0 . 05] per unit increase in ζ for the GLAM+ ( Figure 7C ) ; the mixed-effects regressions included a random slope and intercept for each set size ) , such that subjects with a stronger additive gaze bias ( as indicated by larger ζ estimates ) generally also exhibited stronger multiplicative gaze biases ( as indicated by smaller γ estimates ) . We did not find evidence for this association in the IAM+ ( β = 0 . 0003 , 94% HDI = [−0 . 012 , 0 . 014] per unit increase in ζ; Figure 7B ) . Yet , the γ and ζ estimates in the IAM+ generally also exhibited only little variability ( see Figure 5—figure supplement 1Q–R ) .
The goal of this work was to identify the computational mechanisms underlying choice behaviour in MAFC , by comparing a set of decision models on choice , RT , and gaze data . In particular , we tested models of optimal and satisficing choice ( Reutskaja et al . , 2011; Caplin et al . , 2011; Fellows , 2006; Fellner et al . , 2009; McCall , 1970; Payne , 1976; Schwartz et al . , 2002; Stüttgen et al . , 2012 ) as well as relative ( Krajbich and Rangel , 2011; Thomas et al . , 2019 ) and independent evidence accumulation ( Smith and Vickers , 1988 ) . We further tested two variants of these models , with and without influences of gaze on subjective value . We found that subjects’ behaviour qualitatively could not be explained by optimal choice or standard instantiations of satisficing . After incorporating active effects of gaze into a probabilistic version of satisficing , it explained the data well , slightly outperforming the evidence accumulation models in fitting choice and RT data . Yet , the relative accumulation model with active gaze influences provided by far the best fit to the observed association between gaze allocation and choice behaviour , which was not explicitly accounted for in the likelihood-based model comparison , thus demonstrating that gaze-driven relative evidence accumulation provides the most comprehensive account of behaviour in MAFC . One reason why the satisficing model might perform particularly well in capturing individuals’ choices and RTs is that in our experiment there were a limited number of food items ( 80; see Materials and methods ) . Each item was repeated an average of 50 times per experiment . Thus , subjects could have learned to search for specific items . In practice , this strategy would only be useful in certain scenarios . At your local vending machine , you are almost guaranteed to encounter one of your favourite snacks; here satisficing would be useful . But at a foreign vending machine , or a new restaurant , the relative evidence accumulation framework might be more useful since you need to evaluate and compare the options . Future work is needed to investigate the performance of these models in novel and familiar choice environments . These findings are also relevant to the discussion about the direction of causality between attention and choice . Several papers have argued that subjective value and/or the emerging choice affect gaze allocation , both in binary choice ( Cavanagh et al . , 2014; Westbrook et al . , 2020 ) and in multi-alternative choice ( Krajbich and Rangel , 2011; Towal et al . , 2013; Gluth et al . , 2020; Callaway et al . , 2020 ) . Other work has argued that gaze drives choice outcomes , using exogenous manipulations of attention ( Armel et al . , 2008; Milosavljevic et al . , 2012; Pärnamets et al . , 2015; Tavares et al . , 2017Gwinn et al . , 2019 , c . f . , Newell and Le Pelley , 2018; Ghaffari and Fiedler , 2018 ) . Here , we find support for both directions of the association of gaze and choice . In contrast to the binary choice setting ( Krajbich et al . , 2010 ) , we found that the probability that an item was looked at , as well as the duration of a gaze to this item , increased with the item’s rating , and that this trend also increased over the course of a trial ( Figures 2 and 3 ) . Nevertheless , our data also indicate that gaze correlates with choice even after controlling for the ratings of the items ( Figure 4E , F ) . This means that either gaze is driving choice or gaze is actively being allocated to items that are more appealing than normal , within a particular trial . In a sense , the contrast between binary and multi-alternative choice is not surprising . When deciding between two alternatives , you are merely trying to compare one to the other . In that case , attending to either alternative is equally useful in reaching the correct decision . However , with many choice alternatives , it is in your best interest to quickly identify the best alternatives in the choice set and exclude all other alternatives from further consideration ( e . g . , Hauser and Wernerfelt , 1990; Payne , 1976; Reutskaja et al . , 2011; Roberts and Lattin , 1991 ) . Given this search and decision process , we might expect that subjects’ choices are more driven by their gaze in the later stages of the decision , when they focus more on the highly rated items in the choice set , than in the earlier stages of the search , when gaze is driven by the items’ positions and sizes . Indeed , we found that only the items’ ratings predicted choice behaviour , not their positions or sizes . Our modelling of the decision process included two types of gaze biases: an additive gaze bias , which increases the value of the currently looked at item by a constant , and a multiplicative gaze bias , which discounts the value of currently unattended items . As our experiment included only appetitive snack foods , both of these types of gaze biases have similar effects on the decision process , by increasing the subjective value of the currently looked-at alternative relative to the others . If , however , our experiment had included aversive snack foods , or the framing of the decision problem had been reversed ( Sepulveda et al . , 2020 ) , we would have expected gaze to have the opposite effect . There is some evidence that in the former case , additive and multiplicative biases have opposite effects ( Westbrook et al . , 2020 ) . More research is needed to better understand the interplay of these two types of gaze biases in choice situations that involve appetitive and aversive choice options . Overall , our findings firmly reject a model of complete search and maximization in MAFC ( Caplin et al . , 2011; Pieters and Warlop , 1999; Reutskaja et al . , 2011; Simon , 1959; Stüttgen et al . , 2012 ) : Subjects do not look at every item , and they do not always choose the best item they have seen . Our data also clearly reject the hard satisficing model: Subjects choose the last item they look at only half of the time . Additionally , we find that subjects’ choices are strongly dependent on the actual time that they spend looking at each alternative and can therefore not be fully explained by simply accounting for the set of examined items . This stands in stark contrast to many models of consumer search and rational inattention ( e . g . , Caplin et al . , 2019; Masatlioglu et al . , 2012; Matějka and McKay , 2015; Sims , 2003 ) , which ascribe a more passive role to visual attention , by viewing it as a filter that creates consideration sets ( by attending only to a subset of the available alternatives ) from which the decision maker then chooses . Our findings indicate that attention takes a much more active role in MAFC by guiding preference formation within the consideration set , as has been observed with smaller choice sets ( e . g . , Armel et al . , 2008; Gluth et al . , 2020; Krajbich et al . , 2010; Krajbich and Rangel , 2011; Smith and Krajbich , 2019; Thomas et al . , 2019 ) . In conclusion , we find that models of gaze-weighted subjective value account for relations between eye-tracking data and choice that other passive-attention models of MAFC cannot . These findings provide new insight into the mechanisms underlying search and choice behaviour and demonstrate the importance of employing choice-process techniques and computational models for studying decision-making .
Forty-nine healthy English speakers completed this experiment ( 17 females; 18–55 years , median: 23 years ) . All subjects were required to have normal or corrected-to-normal vision . Individuals wearing glasses or hard contact lenses were excluded from this study . Furthermore , individuals were only allowed to participate if they self-reportedly ( 1 ) fasted at least 4 hr prior to the experiment , ( 2 ) regularly ate the snack foods that were used in the experiment , ( 3 ) neither had any dietary restrictions nor ( 4 ) a history of eating disorders , and ( 5 ) did not diet within the 6 months prior to the experiment . The sample size for this experiment was determined based on related empirical research at the time of data collection ( e . g . , Berkowitsch et al . , 2014; Cavanagh et al . , 2014; Krajbich et al . , 2010; Krajbich and Rangel , 2011; Philiastides and Ratcliff , 2013; Reutskaja et al . , 2011; Rodriguez et al . , 2014; Towal et al . , 2013 ) . Informed consent was obtained from all subjects in a manner approved by the Human Subjects Internal Review Board ( IRB ) of the California Institute of Technology ( IRB protocol: ‘Behavioural , eye-tracking , and psychological studies of simple decision-making’ ) . Each subject completed the following tasks within a single session: First , they did some training with the choice task , followed by the choice task ( Figure 1 ) , a liking rating task , and the choice implementation . In the choice task ( Figure 1 ) , subjects were instructed to choose the snack food item that they would like to eat most at the end of the experiment from sets of 9 , 16 , 25 , or 36 alternatives . There was no time restriction on the choice phase and subjects indicated the time point of choice by pressing the space bar of a keyboard in front of them . After pressing the space bar , subjects had 3 s to indicate their choice with the mouse cursor ( for an overview of the choice indication times , defined as the time difference between the space bar press and the click on an item image , see Figure 1—figure supplement 3 ) . Subjects used the same hand to press the space bar and navigate the mouse cursor . If they did not choose in time , the choice screen disappeared and the trial was marked invalid and excluded from the analysis as well as the choice implementation . We further excluded trials from the analysis if subjects either chose an item that they did not look at before pressing the spacebar or if they clicked on the empty space between item images . The average number of trials dropped from the analysis was 4 ( SE: 0 . 6 ) per subject and set size condition . The initial training task had the exact same structure as the main choice task and differed only in the number of trials ( five trials per set size condition ) and the stimuli that were used ( we used a distinct set of 36 snack food item images ) . In the subsequent rating task , subjects indicated for each of the 80 snack foods , how much they would like to eat the item at the end of the experiment . Subjects entered their ratings on a 7-point rating scale , ranging from −3 ( not at all ) to 3 ( very much ) , with 0 denoting indifference ( for an overview of the liking rating distributions , see Figure 1—figure supplements 1 , 2 ) . After the rating task , subjects stayed for another 10 min and were asked to eat a single snack food item , which was selected randomly from one of their choices in the main choice task . In addition to one snack food item , subjects received a show-up fee of $10 and another $15 if they fully completed the experiment . The choice sets of this experiment were composed of 9 , 16 , 25 , or 36 randomly selected snack food images ( random selection without replacement within a choice set ) . For each set size condition , these images were arranged in a square matrix shape , with the same number of images per row and column ( 3 , 4 , 5 , or 6 ) . All images were displayed in the same size and resolution ( 205 × 133 px ) and depicted a single snack food item centred in front of a consistent black background . During the rating phase , single item images were presented one at a time and in their original resolution ( 576 × 432 px ) , again centred in front of a consistent black background ( Figure 1 ) . Overall , we used a set of 80 different snack food items for the choice task and a distinct set of 36 items for the training . Monocular eye tracking data were collected with a remote EyeLink 1000 system ( SR Research Ltd . , Mississauga , Ontario , Canada ) , with a sampling frequency of 500 Hz . Before the start of each trial , subjects had to fixate a central fixation cross for at least 500 ms to ensure that they began each trial fixating on the same location ( Figure 1 ) . Eye tracking measures were only collected during the choice task and always sampled from the subject's dominant eye ( 10 left-dominant subjects ) . Stimuli were presented on a 19-inch LCD display with a resolution of 1280 × 1024 px . Subjects had a viewing distance of about 50 cm to the eye tracker and 65 cm to the display . Several precautions were taken to ensure a stable and accurate eye tracking measurement throughout the experiment , as we presented up to 36 items on a single screen: ( 1 ) the eye tracker was calibrated with a 13-point calibration procedure of the EyeLink system , which also covers the screen corners , ( 2 ) four separate calibrations were run throughout the experiment: once before and after the training task and twice during the main choice task ( after 75 and 150 trials ) , ( 3 ) subjects placed their head on a chin rest , while we recorded their eye movements . Fixation data were extracted from the output files obtained by the EyeLink software package ( SR Research Ltd . , Mississauga , Ontario , Canada ) . We used these data to define whether the subject’s gaze was either within a rectangular region of interest ( ROI ) surrounding an item ( item gaze ) , somewhere else on the screen ( non-item gaze ) , or whether the gaze was not recorded at all ( missing gaze , e . g . , eye blinks ) . All non-item and missing gazes occurring before the first and after the last gaze to an item in a trial were discarded from all gaze analyses . All missing data that occurred between gazes to the same item were changed to that item and thereby included in the analysis . A gaze pattern of ‘item 1 , missing data , item 1’ would therefore be changed to ‘item 1 , item 1 , item 1’ . Non-item or missing gaze times that occurred between gazes to different items , however , were discarded from all gaze analyses . Passive gazeActive gazeProbabilistic satisficing ( PSM ) Stopping rule: The probability q ( t ) that individuals stop their search and make a choice increases with cumulative time t and cached value C ( t ) ( scaled by v and α respectively ) : q ( t ) = v×t +α×C ( t ) The cached value C ( t ) represents the highest item value l , from the set of seen items J , that has been seen up to time t: C ( t ) =maxj∈J lj Choice rule: Once the search ends , individuals make a probabilistic choice ( with scaling parameter τ ) ) over the set of seen items J according to their liking values: pi=exp ( τ×li ) ∑j∈Jexp ( τ×lj ) Stopping rule: The probability q ( t ) that individuals stop their search and make a choice increases with cumulative time t and cached value C ( t ) ( scaled by v and α , respectively ) : q ( t ) = v×t +α×C ( t ) Importantly , the cached value C ( t ) is extended by the influence of gaze allocation: C ( t ) =maxj∈J cj ( t ) ci ( t ) =gi ( t ) × ( li+ζ ) + ( 1−gi ( t ) ) ×γ×li J represents the set of so far seen items , gi ( t ) indicates the fraction of cumulative time t that an item i has been looked at , while γ and ζ determine the strength of the multiplicative and additive gaze bias , and li indicates the item's value . Choice rule: Once the search ends , individuals make a probabilistic choice ( with scaling parameter τ ) over the set of seen items J according to the gaze-weighted values ci ( t ) : pi ( t ) =exp ( τ×ci ( t ) ) ∑j∈Jexp ( τ×cj ( t ) ) Independent evidence accumulation ( IAM ) The decision follows a stochastic evidence accumulation process , with one accumulator per item . Evidence accumulation for an item only starts once it is looked at in the trial . The drift rate Di of evidence accumulator i is determined by the item’s value li: Di=liThe decision follows a stochastic evidence accumulation process , with one accumulator per item . Evidence accumulation for an item only starts once it is looked at in the trial . The drift rate Di of evidence accumulator i is defined as: Di=gi × ( li+ζ ) + ( 1−gi ) × γ×li gi represents the fraction of the remaining trial time that item i has been looked at , after it is first seen in the trial , while li indicates the item’s value . γ and ζ determine the strength of the multiplicative and additive gaze bias . Relative evidence accumulation ( GLAM ) The decision follows an evidence accumulation process , with one accumulator per item . We define the relative evidence as the difference between the item’s value li and the maximum value of all other seen items J . Importantly , evidence is only accumulated for an item if it is looked at in the trial . The drift rate Di of accumulator i is defined as: Di=σ ( li−maxj≠i lj ) σ ( x ) = 11 + exp ( −τ×x ) τ indicates the scaling parameter of the logistic function σ . A variant of the GLAM in which the absolute decision signal Ai for item i is extended by an additive gaze bias: Ai=gi× ( li+ζ ) + ( 1−gi ) ×γ×li gi represents the fraction of trial time that item i has been looked at , while li indicates its value . γ and ζ determine the strength of the multiplicative and additive gaze bias . The drift rate Di of accumulator i is defined as: Di=σ ( Ai−maxj≠i Aj ) σ ( x ) = 11 + exp ( −τ×x ) τ represents the scaling parameter of the logistic function σ . All model parameters were estimated separately for each individual in each set size condition . The individual models were implemented in the Python library PyMC3 . 9 . 1 ( Salvatier et al . , 2016 ) and fitted using Markov chain Monte Carlo Metropolis sampling . For each model , we first sampled 5000 tuning samples that were then discarded ( burn-in ) , before drawing another 5000 additional posterior samples that we used to estimate the model parameters . Each parameter trace was checked for convergence by means of the Gelman–Rubin statistic ( | R^−1 |<0 . 05 ) as well as the mean number of effective samples ( >100 ) . If a trace did not converge , we re-sampled the model and increased the number of burn-in samples by 5000 until convergence was achieved . Note that the IAM+ did not converge for three , one , and one subjects in the set sizes with 9 , 16 , and 25 items , respectively , after 50 re-sampling attempts . For these subjects , we continued all analyses with the model that was sampled last . We defined all model parameter estimates as maximum a posteriori estimates ( MAP ) of the resulting posterior traces ( for an overview , see Figure 5—figure supplement 1 and Supplementary files 1–3 ) . We repeated each trial 50 times during the simulation and simulated a choice and RT for each trial with each model at a rate of 95% , while we simulated random choices and RTs according to Equation 19 at a rate of 5% . We used the MAP of the posterior traces of the individual subject models as parameter estimates for the simulation ( for an overview , see Figure 5—figure supplement 1 and Supplementary files 1–3 ) . All mixed-effects models were fitted in a Bayesian hierarchical framework by the use of the Bayesian Model-Building Interface ( bambi 0 . 2 . 0; Capretto et al . , 2021 ) . Bambi automatically generates weakly informative priors for all model variables . We fitted all models using the Markov chain Monte Carlo No-U-Turn-Sampler ( Hoffman and Gelman , 2014 ) , by drawing 2000 samples from the posterior , after a minimum of 500 burn-in samples . In addition to the reported fixed-effect estimates , all models included random intercepts for each subject , as well as random subject-slopes for each model coefficient . The posterior traces of all reported fixed-effects estimates were checked for convergence by means of the Gelman–Rubin statistic ( | R^−1 |<0 . 05 ) . If a fixed-effect posterior trace did not converge , the model was re-sampled and the number of burn-in samples increased by 2000 until convergence was achieved . All data analyses were performed in Python 3 . 6 . 8 ( Python Software Foundation ) , by the use of the SciPy 1 . 3 . 1 , ( Virtanen et al . , 2019 ) , NumPy 1 . 17 . 3 ( Oliphant , 2006 ) , Matplotlib 3 . 1 . 1 ( Hunter , 2007 ) , Pandas 0 . 25 . 2 ( McKinney , 2010 ) , Theano 1 . 0 . 4 ( The Theano Development Team , 2016 ) , bambi 0 . 2 . 0 ( Yarkoni & Westfall , 2016 ) , ArviZ 0 . 9 . 0 ( Kumar et al . , 2019 ) , and PyMC3 . 9 . 1 ( Salvatier et al . , 2016 ) packages . For the computation of stimulus metrics , we further utilized the Pillow 5 . 0 ( http://pillow . readthedocs . io ) Python package . The experiment was written in MATLAB ( The MathWorks , Inc , Natick , MA ) , using the Psychophysics Toolbox extensions ( Brainard , 1997 ) . All experiment stimuli , data , and analysis scripts are available at: https://github . com/athms/many-item-choice ( Thomas , 2021; copy archived at swh:1:rev:7b8d6d852f89ad0e59ace94614acc6b683d914e0 ) . | In our everyday lives , we often have to choose between many different options . When deciding what to order off a menu , for example , or what type of soda to buy in the supermarket , we have a range of possibilities to consider . So how do we decide what to go for ? Researchers believe we make such choices by assigning a subjective value to each of the available options . But we can do this in several different ways . We could look at every option in turn , and then choose the best one once we have considered them all . This is a so-called ‘rational’ decision-making approach . But we could also consider each of the options one at a time and stop as soon as we find one that is good enough . This strategy is known as ‘satisficing’ . In both approaches , we use our eyes to gather information about the items available . Most scientists have assumed that merely looking at an item – such as a particular brand of soda – does not affect how we feel about that item . But studies in which animals or people choose between much smaller sets of objects – usually up to four – suggest otherwise . The results from these studies indicate that looking at an item makes that item more attractive to the observer , thereby increasing its subjective value . Thomas et al . now show that gaze also plays an active role in the decision-making process when people are spoilt for choice . Healthy volunteers looked at pictures of up to 36 snack foods on a screen and were asked to select the one they would most like to eat . The researchers then recorded the volunteers’ choices and response times , and used eye-tracking technology to follow the direction of their gaze . They then tested which of the various decision-making strategies could best account for all the behaviour . The results showed that the volunteers’ behaviour was best explained by computer models that assumed that looking at an item increases its subjective value . Moreover , the results confirmed that we do not examine all items and then choose the best one . But neither do we use a purely satisficing approach: the volunteers chose the last item they had looked at less than half the time . Instead , we make decisions by comparing individual items against one another , going back and forth between them . The longer we look at an item , the more attractive it becomes , and the more likely we are to choose it . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"neuroscience"
] | 2021 | Uncovering the computational mechanisms underlying many-alternative choice |
Defining specific protein interactions and spatially or temporally restricted local proteomes improves our understanding of all cellular processes , but obtaining such data is challenging , especially for rare proteins , cell types , or events . Proximity labeling enables discovery of protein neighborhoods defining functional complexes and/or organellar protein compositions . Recent technological improvements , namely two highly active biotin ligase variants ( TurboID and miniTurbo ) , allowed us to address two challenging questions in plants: ( 1 ) what are in vivo partners of a low abundant key developmental transcription factor and ( 2 ) what is the nuclear proteome of a rare cell type ? Proteins identified with FAMA-TurboID include known interactors of this stomatal transcription factor and novel proteins that could facilitate its activator and repressor functions . Directing TurboID to stomatal nuclei enabled purification of cell type- and subcellular compartment-specific proteins . Broad tests of TurboID and miniTurbo in Arabidopsis and Nicotiana benthamiana and versatile vectors enable customization by plant researchers .
With increased efficiency , BioID-based PL would be a valuable tool to study protein interactions and local proteomes on a cell-type-specific level in plants . We therefore made an initial diagnosis of whether the improved BirA* variants TurboID and miniTurbo ( hereafter called TbID and mTb ) are appropriate for PL applications in plants , by testing their activity in the nucleus and cytosol of two plant model systems: transiently transformed N . benthamiana leaves and young seedlings of stably transformed Arabidopsis . To enable a comparison with previous experiments in the literature , we also included the original BirA* in our experiments and expressed all three versions under the ubiquitous UBIQUITIN10 ( UBQ10 ) promoter . A YFP tag was added to confirm correct expression and localization of the TbID and mTb biotin ligases ( Figure 1—figure supplements 1 and 2 ) . To test biotin ligase activities , we treated leaf discs from N . benthamiana leaves or whole 5-day-old transgenic Arabidopsis seedlings , each expressing TbID or mTb , with biotin and subsequently monitored biotinylation in total protein extracts using streptavidin immunoblots . For biotin treatment , we briefly vacuum infiltrated the plant tissue with the biotin solution and then incubated the tissue submerged in the solution for 1 h at room temperature ( approximately 22°C ) . Mock- or untreated plants were used as controls ( Figure 1A ) . In TbID- and mTb-expressing N . benthamiana and Arabidopsis , biotin treatment induced strong labeling of proteins , demonstrating that the new biotin ligases work under our chosen conditions in plants . As was observed in other organisms , both TbID and mTb showed greatly increased activity compared to BirA* , which mainly achieved weak self-labeling within 1 h of biotin treatment ( Figure 1B–C , Figure 1—figure supplements 3 and 4 ) . Since plants produce and store free biotin in their cells , we were concerned about 'background’ labeling in the absence of exogenous biotin . Although it did appear , background labeling was in most cases negligible . Direct comparison of TbID and mTb in our plant systems revealed little difference in either activity or background labeling in N . benthamiana ( Figure 1B , Figure 1—figure supplement 3 ) , possibly due to the high expression levels of the constructs . In Arabidopsis , however , TbID was clearly more active than mTb but also produced more background . Enhanced activity of TbID in the absence of exogenous biotin was especially evident in lines that express TbID and mTb at low levels ( Figure 1C , Figure 1—figure supplement 4 ) . Comparing nuclear and cytosolic constructs , we did not observe any significant differences in labeling efficiency at the resolution of immunoblots ( Figure 1—figure supplements 3 and 4 ) . From these experiments , we conclude that both TbID and mTb are well suited for use in plants . Which version is more suitable may depend on the individual question and whether high sensitivity ( TbID ) or tighter control over labeling time ( mTb ) is important . For this current study , we generated a versatile set of gateway-compatible entry and destination vectors that can be used to express TbID or mTb alone or as fusion with a protein of interest under a promoter of choice ( Figure 1—figure supplement 5 ) . This ‘toolbox’ is accessible through Addgene ( available vectors are listed in the Materials and methods section ) . Achieving an optimal enzyme efficiency by using the right experimental conditions , like labeling time , temperature , biotin concentration and mode of application , can be key for using PL with low-abundant proteins in plants . We therefore tested the effect of those parameters on biotin labeling in 4- to 5-day-old Arabidopsis seedlings expressing TbID and mTb under the UBQ10 promoter . In mammalian cell culture , 10 min of labeling with TbID were sufficient to visualize biotinylated proteins by immunoblot and to perform analysis of different organellar proteomes ( Branon et al . , 2018 ) . Using immunoblots , we observed similarly fast labeling in plants . TbID induced labeling of proteins over background levels within 15–30 min of treatment with 250 or 50 µM biotin at room temperature ( 22°C ) and labeling steadily increased over the next 3 to 5 h ( Figure 2A , Figure 2—figure supplement 1 ) . An increase in self-labeling of TbID was evident even earlier , after as little as 5 min ( compare Figure 4—figure supplement 3 ) . Time course experiments in N . benthamiana suggest that mTb is equally fast , with clear labeling of proteins visible within 10 min of treatment with 50 µM biotin ( Figure 1—figure supplement 3 ) . This is a significant improvement over BirA* , for which labeling times of 24 h were applied in all three published plant experiments ( Khan et al . , 2018; Conlan et al . , 2018; Lin et al . , 2017 ) . We systematically tested the effect of different biotin treatment temperatures on TbID and mTb activity in Arabidopsis seedlings . Encouragingly , the activity of both variants was nearly as high at 22°C as at 30°C . Moreover , TbID showed only a moderate increase of activity at 37°C , while mTb activity was actually reduced at this temperature ( Figure 2B ) . High activity at ambient temperatures was also observed in N . benthamiana ( Figure 2—figure supplement 2 ) . Increasing temperatures above plant growth conditions to improve labeling is therefore not needed . The biotin concentration used for PL is an important consideration . Endogenous levels of biotin in plants are sufficient for low-level labeling of proteins by TbID , and to some extent also by mTb . While this may be useful for some applications , most applications will require strongly enhanced and time-regulated labeling through the addition of exogenous biotin . Although using excessive amounts of biotin is inconsequential for immunoblots , it poses a problem for downstream protein purification with streptavidin beads , as will be discussed later . We therefore tested biotin concentrations ranging from 0 . 5 to 250 µM to determine the optimal substrate concentration for TbID and mTb . We found that TbID has a larger dynamic range than mTb . Weak over-background labeling could already be seen with 0 . 5 µM biotin , which increased weakly through 20 µM , followed by a steeper increase with 30 µM and was more or less saturated at 50–75 µM . mTb required between 2 . 5 and 10 µM biotin for weak activity , showed a steep increase with 20 µM ( comparable to TbID ) and was also saturated at 50–75 µM ( Figure 2C , Figure 2—figure supplement 3 ) . TbID and mTb are therefore comparable to BirA* in their biotin requirement and concentrations of 2 . 5–50 µM ( TbID ) and 20–50 µM ( mTb ) seem to be appropriate . In initial experiments , we vacuum infiltrated the plant material with the biotin solution to maximize biotin uptake . At least in Arabidopsis seedlings , this is not necessary . Simply submerging the plantlets in the biotin solution resulted in the same amount of labeling as vacuum infiltration followed by incubation in the biotin solution did ( Figure 2D ) . This finding is very important since it not only simplifies handling of the experiment , but also improves isolation of labeled proteins by reducing the amount of free biotin in the tissue . For TbID to be widely applicable , it must be able to biotinylate proteins in many developmental stages and plant tissues . One initial concern , especially with TbID , was that background labeling from endogenous biotin would accumulate over time , making timed experiments with older tissues unfeasible . This was , however , not the case . Labeling worked well in 4- to 14-day-old plate-grown seedlings without significant increase of background ( Figure 3—figure supplement 1 ) . The same was true for separated roots and shoots of 6- to 14-day-old seedlings and even for rosette leaves and flower buds of adult Arabidopsis plants grown on soil ( Figure 3 , Figure 3—figure supplements 2 and 3 ) . Background activity was low , especially in leaf tissue , and labeling worked well . Vacuum infiltration was not required for the tested plant sample types , except for unopened floral buds , where infiltration improved labeling relative to submergence in the biotin solution ( Figure 3 ) . This is likely because petals and reproductive tissues are not in direct contact with the biotin solution . Overall , our experiments suggest that TbID will be applicable in a wide range of developmental stages and tissues . Since TbID and mTb behaved similar in most experiments , it is likely that the same is true for mTb . After confirming general applicability of TbID for PL in plants , we wanted to test its performance for the identification of rare protein complexes and the characterization of cell-type-specific organellar proteomes in a real experiment . For this purpose , we chose a cell-type-specific transcription factor ( FAMA ) and a subcellular compartment of a rare cell type ( nuclei of FAMA-expressing stomatal cells ) for a case study . FAMA is a nuclear basic helix-loop-helix ( bHLH ) transcription factor ( TF ) that is expressed in young stomatal guard cells ( GCs ) in the epidermis of developing aerial tissues ( Ohashi-Ito and Bergmann , 2006 ) . The low abundance of FAMA and FAMA-expressing cells renders identification of interaction partners and cell-type-specific nuclear proteins by traditional methods challenging and makes it well suited for a proof-of-concept experiment . Potential FAMA interaction partners were previously identified in yeast-2-hybrid ( Y2H ) and bimolecular fluorescence complementation studies ( Chen et al . , 2016; Kanaoka et al . , 2008; Lee et al . , 2014; Li et al . , 2018; Matos et al . , 2014; Ohashi-Ito and Bergmann , 2006 ) , but only few have been confirmed by in vivo functional data . For our study , we generated plants expressing TbID and a fluorescent tag for visualization under the FAMA promoter , either as a FAMA-protein fusion or alone with a nuclear localization signal ( NLS ) ( Figure 4A , Figure 4—figure supplement 1 ) . By comparing proteins labeled in the FAMApro::FAMA-TbID-Venus ( FAMA-TbID ) and FAMApro::TbID-YFPNLS ( FAMAnucTbID ) plants to each other and to wild-type ( WT ) plants and UBQ10pro::TbID-YFPNLS ( UBQnucTbID ) plants , we can test the ability of the system to identify ( 1 ) proteins in close proximity to FAMA ( FAMA complexes ) , ( 2 ) the nuclear protein composition during the FAMA-cell stage , ( 3 ) the nuclear proteome in general , and ( 4 ) possible FAMA-stage-specific nuclear proteins . The FAMA-TbID-Venus construct is functional , since it complements the seedling-lethal phenotype of the fama-1 mutant ( Figure 4—figure supplement 2 ) . Determining suitable incubation times is crucial since too short an incubation can yield insufficient protein amounts for identification but excessive incubation could label the whole subcellular compartment . We therefore performed labeling time-courses with the FAMA-TbID line , using immunoblots as a readout . FAMA auto-labeling could be observed after as little as 5 min but clear labeling of other proteins required approximately 15 to 30 min . Longer incubation led to further increase in labeling up to 3 h , both in the form of stronger discrete bands and of diffuse labeling , but stayed more or less the same thereafter ( Figure 4B , Figure 4—figure supplement 3 ) . Based on these observations , we chose 0 . 5 and 3 h biotin treatments , after which we would expect abundant and most FAMA interactors to be labeled , respectively , for the ‘FAMA interactome’ experiment ( Figure 4C ) . For the ‘nuclear proteome’ experiment ( Figure 4D ) , only the longer 3 h time point was used , since over-labeling of the compartment was not a concern . Accordingly , the FAMA-TbID , FAMAnucTbID and WT lines were treated for 0 , 0 . 5 and 3 h , and the UBQnucTbID line only for 3 h . As plant material , we chose seedlings 5 days post germination , which corresponds to a peak time in FAMA promoter activity , as determined empirically by microscopy . We used three biological replicates per sample . To make the datasets as comparable as possible , all steps preceding data analysis were done together for the two experiments , as described in the next section . Through empirical testing of experimental conditions using the UBQnucTbID line , we identified steps and choices that have a big impact on success of protein purification and identification after PL with TbID . These include sample choice to maximize bait abundance , removal of free biotin , optimizing the amount of streptavidin beads for affinity purification ( AP ) and choosing among MS sample prep procedures . Below , we describe our experimental procedure ( Figure 4E ) and highlight key choices . We first labeled 5-day-old seedlings , by submerging them in a 50 µM biotin solution for 0 , 0 . 5 or 3 hr , quickly washed them with ice cold water to stop the labeling reaction and to remove excess biotin and isolated total proteins for AP of biotinylated proteins . The protein extracts were then passed through PD-10 gel filtration columns to reduce the amount of free biotin in the sample before proceeding with AP using magnetic streptavidin beads . Successful labeling and purification was confirmed by immunoblots ( Figure 4—figure supplements 4 and 5 ) . The inclusion of a biotin-depletion step was found to be critical as free biotin in the protein extracts competes with biotinylated proteins for binding of the streptavidin beads ( Figure 4—figure supplement 6 ) . While for mammalian cell culture or rice protoplasts thorough washing of the cells seems to suffice for removal of free biotin , this is not the case for intact plant tissue ( see also Conlan et al . , 2018; Khan et al . , 2018 ) . Especially when large amounts of starting material and moderate amounts of biotin are used , little to none of the biotinylated proteins may be bound by the beads . To maximize the amount of purified proteins it is further advisable to determine the appropriate amount of beads required for each experiment . We used 200 µl beads for approximately 16 mg total protein per sample . This amount was chosen based on tests with different bead-to-extract ratios ( Figure 4—figure supplement 7 ) and was sufficient to bind most biotinylated proteins in our protein extracts , although the beads were slightly oversaturated by the highly labeled UBQnucTbID samples ( Figure 4—figure supplement 5D ) . Following AP , we performed liquid chromatography coupled to tandem mass spectrometry ( LC-MS/MS ) analysis to identify and quantify the captured proteins . Tryptic digest for LC-MS/MS analysis was done on-beads , since test experiments revealed that elution from the beads using two different methods ( Cheah and Yamada , 2017; Schopp and Béthune , 2018 ) and subsequent in-gel digestion of biotinylated proteins yielded significantly lower protein amounts and less protein identifications ( data not shown ) . This apparent sample loss is caused by the strong biotin-streptavidin interaction , which allows for stringent washing conditions but also prevents efficient elution of biotinylated proteins from the beads . Notably , highly biotinylated proteins , which likely comprise the most interesting candidates , will interact with more than one streptavidin molecule and will be especially hard to elute . After MS analysis , we identified and quantified the proteins by label-free quantification and filtered for significantly enriched proteins . This part was done separately for the ‘FAMA interactome’ and ‘nuclear proteome’ experiments and is described in the following sections . FAMA acts as both an activator and repressor for hundreds of genes ( Hachez et al . , 2011 ) , suggesting a need for coordinated action with other TFs , co-activators and -repressors ( Matos et al . , 2014 ) . Identifying such proteins through classical affinity purification-mass spectrometry approaches is hampered by the low overall abundance of FAMA . Apart from INDUCER OF CBF EXPRESSION 1 ( ICE1 ) , which is a known heterodimerization partner of FAMA ( Kanaoka et al . , 2008 ) , we failed to identify any transcriptional ( co- ) regulators by AP-MS with FAMA-CFP , despite the use of crosslinking agents and large amounts of plant material ( 15 g of 4-day-old seedlings per sample ) . Moreover , less than 20% of the AP-MS-derived ‘candidates’ were predicted to be nuclear , and one quarter were chloroplast proteins ( Figure 5—figure supplement 1 , Supplementary file 2 – Table 5 ) . We therefore wanted to see if PL would improve the identification of biologically relevant FAMA interactors . For the ‘FAMA interactome’ experiment we compared proteins purified from plants expressing the FAMA-TbID fusion ( FAMA-TbID ) with proteins from WT and with proteins from plants expressing nuclear TbID ( FAMAnucTbID ) after 0 , 0 . 5 and 3 h of biotin treatment . In total , we identified 2511 proteins with high confidence ( quantified in all three replicates of at least one sample ) . Principal component analysis ( PCA ) showed a clear separation of the samples by genotype and time point ( Figure 5—figure supplement 2 ) . Despite this clear separation , the majority of proteins were common to all samples , including the untreated WT control ( Supplementary file 2 – Table 1 ) , and moreover , were unchanged between samples ( Figure 5—figure supplements 2 and 3 ) . This indicates that a large proportion of identified proteins bound to the beads non-specifically . This is not uncommon for affinity purification experiments , and underlines the importance of appropriate controls and data filtering pipelines . To remove these ‘background proteins’ from our dataset and to narrow down the number of FAMA complex candidates , we applied three consecutive filtering steps ( Figure 5 , for details see Materials and methods section ) . First , we removed proteins that were not significantly enriched in the FAMA-TbID samples compared to WT . These comprise sticky proteins that bind the beads non-specifically and a handful of proteins that are biotinylated natively by Arabidopsis plant biotin ligases ( Alban , 2011; Nikolau et al . , 2003 ) . This was done by pair-wise comparison of FAMA-TbID and WT samples at each time point , using only proteins that were found in all three replicates of the corresponding FAMA-TbID samples , and resulted in a list of 73 , 85 and 239 significantly enriched proteins ( including FAMA ) at the 0 , 0 . 5 and 3 h time points , respectively ( Figure 5—figure supplement 2 – Table 2 ) . Since TbID is highly active and endogenous levels of biotin are sufficient for low-level labeling , there is a risk that proteins are labeled stochastically and that , over time , the whole nuclear proteome would be labeled . Notably , more than half of the proteins enriched in the FAMA-TbID plants at any of the time points were also enriched in the FAMAnucTbID plants ( Figure 5 , Figure 5—figure supplement 4 , Supplementary file 2 – Table 2 ) . We therefore applied a second filtering step to remove proteins that were not significantly enriched in the FAMA-TbID versus the FAMAnucTbID samples . Pair-wise comparison of FAMA-TbID and FAMAnucTbID samples at the three time points , further reduced the dataset to 6 , 15 and 57 proteins ( including FAMA ) ( Figure 5—figure supplement 2 , Supplementary file 2 – Table 2 ) . Finally , we removed proteins that were not significantly enriched after biotin treatment compared to the untreated samples , since these proteins are likely genotype-specific contaminations . One protein , which was only enriched in the absence of exogenous biotin but not after biotin treatment , was removed as well . This left us with 47 ‘high confidence’ candidates ( Figure 5 , Table 1 ) , 35 of which were previously demonstrated to be in the nucleus using fluorescent protein fusions or were found in MS-based nuclear proteome studies ( Figure 5 , Table 1 , Supplementary file 2 – Tables 2 and 4 ) . Notably , more than half of the candidates have a role in regulation of transcription or are chaperones which could assist in FAMA’s role as a TF or in protein folding and stabilization , respectively ( Figure 5 , Table 1 , Supplementary file 2 – Table 4 ) . Moreover , several of these proteins have previously been shown to interact with each other , which suggests that they could be part of the same FAMA complexes . This is a huge improvement compared to our AP-MS experiment , which could only confirm FAMA’s interaction with its obligate heterodimerization partner ICE1 . The transcriptional regulators we found with PL can be roughly divided into two categories: TFs and transcriptional co-regulators . Among the TFs we again found ICE1 ( as well as peptides shared between ICE1 and its orthologue SCRM2 ) . We also found three other bHLH TFs ( AIB/JAM1 , JAM3 , and BIM1 ) and the non-canonical bHLH-type TF BZR1 . AIB and JAM3 play partially redundant roles in negative regulation of jasmonic acid ( JA ) signaling ( Sasaki-Sekimoto et al . , 2013; Fonseca et al . , 2014 ) , while BIM1 and BZR1 mediate brassinosteroid ( BR ) signaling ( Yin et al . , 2002; Wang et al . , 2002 ) . Both JA and BR signaling play roles in stomatal function or development ( Acharya and Assmann , 2009; Gudesblat et al . , 2012; Kim et al . , 2012 ) . Among the transcriptional co-regulators we found two significantly enriched transcriptional co-activators: MED16 , which is part of the mediator complex that links TFs to RNA Pol II ( Kidd et al . , 2011 ) , and HAC1 , which is a histone acetyl transferase ( HAT ) ( Deng et al . , 2007 ) . Combined with previous data showing a link between FAMA and RNA Pol II ( Chen et al . , 2016 ) , this suggests that FAMA activates genes both directly by recruiting RNA Pol II and by opening up the chromatin for other transcriptional regulators . Among transcriptional co-repressors were TOPLESS ( TPL ) -related proteins TPR3 and TPR4 and LEUNIG ( LUG ) and LEUNIG HOMOLOG ( LUH ) , which recruit histone deacetylases ( HDACs ) to TFs ( Long et al . , 2006 ) . Additionally , we identified the linker protein SEUSS ( SEU ) , which mediates interaction of LUG and LUH with TFs ( Liu and Karmarkar , 2008; Sitaraman et al . , 2008 ) . The identification of all three members of the SEU/LUG/LUH co-repressor complex is a strong indication of a functional complex with FAMA in the plant . Relaxing our filtering criteria to include proteins that are enriched in the FAMA-TbID vs FAMAnucTbID samples but were not significant under our stringent cutoff , we find several more components of transcriptional co-regulator complexes , including three more MED proteins , another HAT , two more TPL-related proteins and TPL itself . RBR1 , a cell cycle regulator and a known interactor of FAMA ( Lee et al . , 2014; Matos et al . , 2014 ) , is also among the FAMA-TbID enriched proteins but , due to a modest fold change , did not pass our last filters ( Figure 5 , – Supplementary file 2 Table 2 ) . This suggests that by setting a stringent cutoff on enrichment between FAMA-TbID and FAMAnucTbID , we might lose some true interactors . This might be especially true of ubiquitously expressed proteins with many partners like RBR1 , where FAMA-RBR1 interactions are likely to represent only a small fraction of all complexes . For a small-scale validation of candidates biotinylated by FAMA-TbID , we tested four proteins ( co-repressor complex components SEU and LUH and TFs BZR1 and BIM1 ) for direct interaction with FAMA in a Y2H system . Additionally , we co-expressed SEU and LUH with FAMA-TbID or nuclear TbID in N . benthamiana and tested for biotinylation of the candidates ( Figure 5—figure supplement 5 ) . Our experiments suggest that BIM1 and SEU are direct interaction partners of FAMA , while LUH and BZR1 might be in indirect contact with FAMA or require a specific protein or DNA context to be present for interaction . Overall , our ‘FAMA interactome’ experiment demonstrates the usefulness of PL to identify potential interaction partners of rare proteins . We identified several good FAMA-complex candidates which could support FAMA in its role as a key TF and provide a possible mechanism for FAMA to induce fate-determining and lasting transcriptional changes in developing GCs . Some of the FAMA-complex candidates identified through PL are also slightly enriched in FAMA AP-MS samples compared to their controls . However , the enrichment is not enough to call any of them , except ICE1 , significant in the AP-MS experiment . PL therefore not only gave us higher specificity for nuclear proteins than the AP-MS did , but it is potentially more sensitive as well . It is worth noting , that most FAMA interaction candidates were identified at the 3 h time point and that longer biotin treatment greatly improved identification of biotinylated proteins ( Figure 5 , Figure 5—figure supplement 4 , Supplementary file 2 – Table 3 ) . The second question our PL experiment should answer was whether TbID could be used to take a snapshot of the nuclear proteome of FAMA-expressing cells . Traditional tools to study organellar proteomes are not well-suited for such an endeavor , since they require isolation of the cell type and organelle of interest and therefore lack the required sensitivity . ( Branon et al . , 2018 ) , showed that TbID can be used to efficiently and specifically purify proteins from different subcellular compartments without prior cell fractionation . Their work was done using a homogeneous population of cultured mammalian cells , however , so it remained to be shown whether it would be possible to isolate an organellar proteome from an individual cell type , especially a rare or transient one , in a complex multicellular organism . To identify nuclear proteins in FAMA-expressing young GCs and compare them to the global nuclear proteome at this growth stage , we purified proteins from seedlings expressing nuclear TbID under the FAMA ( FAMAnucTbID ) and UBQ10 ( UBQnucTbID ) promoter and from WT after three hours of biotin treatment . PCA , hierarchical clustering and multi scatterplots showed a clear separation of the three genotypes ( Figure 6—figure supplements 1 and 2 ) . In total , we identified 3176 proteins with high confidence ( Supplementary file 3 – Table 1 ) . 2215 proteins were significantly enriched in UBQnucTbID compared to WT ( Figure 6 , Figure 6—figure supplement 3 , Supplementary file 3 – Table 2 ) . These proteins comprise our ‘global’ nuclear protein dataset . Despite the relative rareness of FAMA-expressing cells , the FAMAnucTbID dataset yielded 394 proteins that were enriched compared to WT ( Figure 6 , Figure 6—figure supplement 3 , Supplementary file 3 – Table 3 ) . Notably , most of them overlap with our global nuclear protein dataset ( Figure 6 ) , as would be expected since the UBQnucTbID dataset also contains FAMA-stage cells . To estimate how ‘pure’ our nuclear proteomes are , we curated published nuclear and subnuclear compartment proteomes ( Bae et al . , 2003; Bigeard et al . , 2014; Calikowski et al . , 2003; Chaki et al . , 2015; Palm et al . , 2016; Pendle et al . , 2005; Sakamoto and Takagi , 2013; Goto et al . , 2019 ) and searched the Arabidopsis protein subcellular localization database SUBA ( version 4 , Hooper et al . , 2017 , http://suba . live/ ) for proteins that were observed in the nucleus as fluorescent-protein fusions . This resulted in a combined list of 4 , 681 ‘experimentally determined nuclear proteins’; 4021 from MS and 975 from localization studies ( Supplementary file 3 – Table 4 ) . More than three quarters of the proteins enriched in our UBQnucTbID and FAMAnucTbID datasets are either experimentally verified nuclear proteins or are predicted to be localized in the nucleus ( Supplementary file 3 – Tables 2 , 3 and 5 ) . This suggests that most identified proteins are indeed nuclear proteins . Of the remaining proteins , most are predicted to be in the cytosol and could have been labeled by TbID right after translation and before nuclear import of the biotin ligase or by a small mis-localized fraction of TbID . Chloroplast proteins , which are a major source of contamination in plant MS experiments , make up only about 4% of our identified proteins based on experimental and prediction data ( Supplementary file 3 – Table 5 ) . For a comparison , about 12% and 6% of the proteins identified in the two most recent Arabidopsis nuclear proteome studies ( Palm et al . , 2016; Goto et al . , 2019 ) , are predicted to be in the chloroplast ( SUBAcon prediction , SUBA4 ) . Gene ontology ( GO ) analysis is also consistent with nuclear enrichment in both nuclear TbID datasets ( Figure 6—figure supplement 4 , Supplementary file 3 – Tables 2 , 3 and 7 ) . Importantly , our nuclear TbID successfully labeled all major sub-nuclear compartments and domains , including the nuclear pore complex , the nuclear envelope , the nuclear lamina , the nucleolus and other small speckles , as well as DNA- and chromatin-associated proteins ( subdomain markers from Petrovská et al . , 2015; Tamura and Hara-Nishimura , 2013 , see Supplementary file 3 – Table 8 for examples ) . After the general assessment of data quality and nuclear specificity , we asked whether we could identify known markers for FAMA-stage GC nuclei from our dataset by comparing the FAMAnucTbID to the UBQnucTbID samples ( Figure 6 , Figure 6—figure supplement 3 , Supplementary file 3 – Table 3 ) . Unlike ubiquitously expressed proteins , which should be enriched in the UBQnucTbID samples , FAMA-cell specific or highly enriched proteins should be equally abundant in both sample groups . Indeed , looking at proteins that were equally abundant or slightly enriched in the FAMAnucTbID samples , we find several known stomatal lineage- and GC-associated TFs , namely FAMA itself , HOMEODOMAIN GLABROUS 2 ( HDG2 ) and STOMATAL CARPENTER 1 ( SCAP1 ) ( Ohashi-Ito and Bergmann , 2006; Peterson et al . , 2013; Negi et al . , 2013 ) . Additionally , there were 10 proteins among the 44 highly FAMAnucTbID enriched proteins , that were previously identified as part of the GC proteome by Zhao and colleagues using MS analysis of 300 million GC protoplasts ( Zhao et al . , 2010; Zhao et al . , 2008 ) ( Supplementary file 3 – Tables 3 and 6 ) . It is worth noting that neither FAMA nor SCAP1 are in the Zhao GC protoplast proteome , presumably due to their relatively low expression and the use of material from more mature plants . These results suggest that PL can find lowly expressed proteins , that our knowledge of the GC proteome is not complete , and that additional important regulators of GC development and function might be uncovered by looking at stage-specific proteomes . Overall , this experiment demonstrates the usefulness of TbID as a tool for studying subcellular proteomes on a whole-plant as well as on a cell-type-specific level . This will allow us to address questions that were previously inaccessible and thus has the potential to greatly improve our understanding of cellular processes in a cell-type-specific context .
Our experiments presented in this study demonstrate that the new biotin ligase versions TbID and mTb drastically improve the sensitivity of PL in plants , compared to previously used BirA* , and tolerate a range of experimental conditions . We observed rapid labeling of proteins by TbID and mTb in different species , tissues and at different growth stages from seedlings to mature plants , using a simple biotin treatment protocol at room temperature . This greatly broadens the range of possible PL applications for plants and will allow to address hitherto inaccessible or hard to address questions in the future . To test the usefulness of TbID for identification of protein interactors and organellar proteomes , we aimed to identify FAMA protein complexes and the nuclear proteome in FAMA-expressing cells . We deliberately picked a rare protein and cell type to observe its performance under conditions that make the use of traditional methods challenging or unfeasible . In spite of FAMA’s low abundance , PL with the FAMA fusion protein in our ‘FAMA interactome’ experiment worked very well and outperformed our FAMA AP-MS experiments both in sensitivity and in specificity for nuclear proteins . Unlike in the AP-MS experiments , we identified a number of new proteins with gene regulatory functions that could support FAMA in its role as a master regulator of GC differentiation . Beyond the known dimerization partner ICE1 , which was found in both types of experiments , PL identified four additional bHLH TFs with the potential to form alterative FAMA heterodimers as well as three non-bHLH TFs . Strong , direct interaction with BIM1 could be confirmed by Y2H ( Figure 5—figure supplement 5A-B ) . We further identified several epigenetic regulators , which could fulfill roles predicted for FAMA complexes based on previous genetic and transcriptomic experiments ( Adrian et al . , 2015; Hachez et al . , 2011; Matos et al . , 2014 ) and fill a gap in our current model of gene regulation during GC formation . It was satisfying to see that when FAMA complex candidates were known to form functional complexes , we often identified multiple components of the complex . One example is the SEU-LUG/LUH co-repressor complex , of which SEU acts as an adapter protein , linking LUG/LUH to TFs ( Liu and Karmarkar , 2008 ) . Our Y2H and PL experiments support a role of SEU as an adapter in FAMA repressive complexes ( Figure 5—figure supplement 5 ) . MED14 , which was found as potential FAMA complex component using relaxed criteria , could also be part of this interaction chain . LUG is known to act by recruiting HDACs to promote epigenetic gene repression , but can also interact with mediator complex subunits like MED14 to interfere with their interaction with transcriptional activators ( Liu and Karmarkar , 2008 ) . MED14 , if indeed a FAMA complex component , could therefore either interact with the FAMA heterodimer or with the repressors LUG or LUH . It is likely that our candidates are part of different activating and repressive FAMA complexes . Which proteins are in which complex and which are direct FAMA interactors will need to be further tested with independent methods . Interestingly , we did not confirm all of FAMA’s previously postulated interaction partners in our PL experiment . This has both biological and technical causes . It is possible that some previously observed interactions are too rare or conditional ( e . g . MED8 may require pathogen exposure ) while others probably only happen under artificial conditions like in Y2H assays ( e . g . bHLH71 and bHLH93 , which seem not to be required for stomatal development; Ohashi-Ito and Bergmann , 2006 ) . Enrichment of NRPB3 in FAMA-TbID samples , on the other hand , could not be assessed because of non-specific binding of the protein to the beads . Other technical aspects that can complicate or prevent identification of interaction partners are removal of low-frequency interactions through stringent data filtering or a general lack of labeling . Because only proteins with exposed , deprotonated lysine residues can be labeled , sterically or chemically inaccessible proteins will not be detected . PL of FAMA-expressing nuclei and comparison with a general nuclear proteome in our ‘nuclear proteome’ experiment showed that TbID is suitable to capture subcellular proteomes at a cell-type-specific level , even when the cell type is rare . We identified 2232 proteins in both nuclear datasets combined , which is 25% more than the most recent nuclear proteome obtained from cultured Arabidopsis cells ( 1528 proteins , Goto et al . , 2019 ) . Judging from experimentally determined and predicted localization of the identified proteins and functional annotation with GO analysis , we obtained high specificity for nuclear proteins and identified proteins from all major nuclear subdomains . The biggest ‘contamination’ stems from the cytosol , which could be caused by a fraction of TbID in the cytosol ( e . g . right after translation ) or from activated biotin diffusing out of the nucleus . Chloroplast-predicted proteins made up only a small fraction . Among FAMA-nuclei enriched proteins , we identified several known nuclear markers of young GCs , confirming our ability to detect cell-type-specific proteins , as well as proteins that have not yet been linked to GC development or function but could be interesting to investigate further . One of them is SHL ( SHORT LIFE ) , which is a histone reader that can bind both H3K27me3 and H3K4me3 histone marks and has been implicated in seed dormancy and flower repression ( Müssig et al . , 2000; Qian et al . , 2018 ) . Interaction of FAMA with RBR1 and with newly identified co-repressors and co-activators from this study , strongly suggests that chromatin marks are important to lock GC in their terminally differentiated state ( Matos et al . , 2014; Lee et al . , 2014 ) and SHL could be involved in this process . When designing a PL experiment , several things should be considered in order to achieve the best possible result . First , a choice has to be made which biotin ligase to use . Whether TbID or mTb is more suitable will depend on the research question . TbID is more active , which is an advantage for low abundant proteins , for cases where over-labeling is not a concern or when labeling times should be kept as short as possible . mTb is less active in the presence of endogenous levels of biotin and will give less background labeling , which could be beneficial in tissues with higher than average endogenous biotin levels ( Shellhammer and Meinke , 1990 ) or when a more restricted and controlled labeling time is desired . Additional factors that should be considered when choosing a biotin ligase are whether one version works better in a specific subcellular compartment ( as was the case in human HEK cells; Branon et al . , 2018 ) and whether the larger TbID interferes with the activity or correct subcellular targeting of the protein of interest ( POI ) . We added a fluorophore to all our biotin ligase constructs . This is very useful to confirm correct expression and subcellular localization of the TbID or mTb fusion protein , but may in some cases affect the activity of the ligase or the tagged protein . Interference with activity and targeting can depend on the position of the biotin ligase relative to the POI ( N- or C-terminal tag ) . Another decision at the construct-design phase is the choice of linker length . For our experiments we added a short flexible linker to TbID . For identification of large protein complexes , increasing the linker length may improve labeling of more distal proteins . Another crucial consideration are controls . Extensive and well-chosen controls are essential to distinguish between true candidates and proteins that either bind non-specifically to the beads or that are stochastically labeled because they are localized in the same subcellular compart as the POI . The former class of contaminants can be identified by including a non-transformed control ( e . g . WT ) . The best control for the latter will be situation-dependent . For identifying interaction partners of a POI , one could use free TbID or mTb targeted to the same subcellular localization as the POI , as we have done in our ‘FAMA interactome’ experiment . Alternatively , one or more unrelated proteins that are in the same ( sub ) compartment but do not interact with the POI can be used . This strategy might improve identification of proteins that are highly abundant or ubiquitously expressed but of which only a small fraction interacts with the bait . Such proteins may be lost during data filtering if a whole-compartment control is used , as we observed for RBR1 in our ‘FAMA interactome’ experiment . In either case , the control construct should have approximately the same expression level as the POI . For sub-organellar proteomes ( e . g . a specific region at the plasma membrane ) or for organellar proteomes in a specific cell type , it is useful to label the whole compartment or the compartment in all cell types for comparison . Before doing a large-scale PL experiment , different experimental conditions should be tested to find a biotin concentration/treatment time/plant amount/bead amount combination that is suitable for the question and budget . Increasing the plant amount , labeling time or biotin concentration can improve protein coverage and can help to get more complete compartmental proteomes , but will also affect the amount of beads required . Excessive labeling is only advisable in closed compartments and when labeling of all proteins is the goal . Immunoblots are a useful tool to test different combinations of labeling concentration and time as well as for determining the correct bead amount for the pulldown so as not to over-saturate the beads . One should keep in mind , though , that the signal intensity on a western blot does not necessarily reflect the amount of labeled protein , because highly biotinylated proteins have multiple binding sites for streptavidin , which leads to signal amplification . Moreover , labeling strength will also depend on the number of sterically/chemically available sites for biotinylation , and therefore also on the size and properties of a protein . If biotinylation is induced by addition of exogenous biotin , a crucial step before AP of biotinylated target proteins is the depletion of free biotin in the sample to reduce the amount of beads required and therefore the per sample cost . We tested the effectiveness of two different approaches: gel filtration with PD-10 desalting columns ( also used by Conlan et al . , 2018 ) and repeated concentration and dilution with Amicon Ultra centrifugal filters ( used by Khan et al . , 2018 ) . The latter method has the potential to remove more biotin and to be more suitable for large amounts of plant material . In our hands , though , using Amicon centrifugal filters led to considerable sample loss , presumably due to binding of the membrane , and was very slow . PD-10 columns , in contrast , did not lead to a notable loss of biotinylated proteins ( Figure 4—figure supplement 7 ) . Surprisingly , consecutive filtering of protein extracts with two PD-10 columns did not improve the bead requirement . An alternative to these two methods is dialysis , which is suitable for larger volumes but is very time consuming . If the target is in an easy-to-isolate and sufficiently abundant organelle , cell fractionation prior to AP might also be considered to remove unbound biotin . For AP , different kinds of avidin , streptavidin or neutravidin beads are available . Their strong interaction with biotin allows for efficient pulldowns and stringent wash conditions , but makes elution of the bound proteins difficult , especially if they are biotinylated at multiple sites . Should elution of the bound proteins be important for downstream processing or the identity of the biotinylated peptides be of major interest , the use of biotin antibodies might be preferable ( Udeshi et al . , 2017 ) . Finally , some consideration should also be given to the MS strategy . For example , digesting the proteins on the streptavidin beads instead of eluting them can increase the peptide yield . For data analysis , label-free quantification produces a more quantitative comparison of samples than comparison of peptide counts . Care should be taken that an appropriate data normalization method is chosen , especially when samples are very different , as can be the case when different cellular compartments are compared . Isotopic labeling , which allows samples to be analyzed together , can further improve quantitative comparison . We can see many potential applications for PL in plants , extending beyond the ones presented in this work . One that has the potential to be widely used is in vivo confirmation of suspected protein interactions or complex formation . The strategy is comparable to currently used co-immunoprecipitation ( Co-IP ) experiments but has the benefit that weak and transient , as well as other hard-to-purify interactions , are easier to detect . For this approach , proteins closely related to the bait can be used as controls for interaction specificity . One of the major applications we demonstrate in this study is de novo identification of protein interaction partners and complex components . Extending from that , PL can be used to observe changes in complex composition in response to internal or external cues ( e . g . stress treatment ) . Currently used techniques like peptide arrays , two-hybrid screens in yeast or plant protoplasts and AP-MS have the disadvantage that they are either artificial or work poorly for low abundant and membrane proteins and tend to miss weak and transient interactions . PL could overcome some of their deficiencies . One should keep in mind , however , that rather than identifying proteins bound directly to a bait protein , PL will mark proteins in its vicinity . Labeling is generally strongest for direct interactors , but labeling radius will depend on properties of the bait such as size , mobility and linker length as well as on the duration of labeling . To define a protein interaction network , it will be useful to use several different baits ( Gingras et al . , 2019 ) . Another application of PL is characterization of subcellular proteomes , such as whole organellar proteomes , as we have demonstrated for the nucleus . Going forward , more detailed characterization of different organellar proteomes as well as sub-organellar proteomes and local protein composition , for example at membrane contact sites , can and should be addressed . This can be done on a whole plant level , but also at organ- and even cell-type-specific level . Importantly , PL enables investigation of previously inaccessible compartments and of rare and transient cell types as we have demonstrated for FAMA-expressing GCs . Differences between individual cell types or treatments can be investigated as well . Labeling times will , among other things , depend on the 2D or 3D mobility and distribution of the bait . For whole-compartment labeling , a combination of several baits may increase efficiency and protein coverage . One drawback of PL compared to traditional biochemical methods is that it requires the generation of transgenic plants , which limits its use to plants that can currently be transformed . PL can also be used in combination with microscopy , to visualize the subcellular localization of biotinylated proteins and reveal labeling patterns of individual bait proteins . This can be utilized to confirm that labeling is restricted to the desired compartment , but it can also be used to fine-map the subcellular localization of a protein of interest or to obtain information about its topology , as was demonstrated for an ER transmembrane protein in human cells ( Lee et al . , 2016 ) . Extended uses of PL techniques that will require some modification of TbID and mTb or the PL protocol before they can be applied include interaction-dependent labeling of protein complexes ( split-BioID ) , identification of RNAs associated with biotinylated proteins and identification of proteins associated with specific DNA or RNA sequences ( for a recent review of PL methods describing these applications see Trinkle-Mulcahy , 2019 ) . While there are is a plethora of questions that can be addressed with PL , there are also limitations to what will be possible . For example , although TbID and mTb are much faster than BirA* , controlled short labeling pulses ( as are possible with APEX-based PL techniques ) will be hard to achieve . TbID and mTb are always active and use endogenous biotin to continuously label proteins , preventing a sharp labeling start . In addition , exogenous biotin needs time to enter the plant tissue to initiate labeling . In mammalian cell culture , 10 min of labeling might be sufficient , but in a whole multicellular organism it may take much longer , depending on the experimental setup and how complete labeling should be . In our ‘nuclear proteome’ experiment , for example , 3 h of biotin treatment were not sufficient to reach labeling saturation and only very few proteins were enriched in FAMAnucTbID samples by 30 min of biotin treatment ( Figure 5—figure supplement 6 , Supplementary file 2 – Table 3 ) . Development of strategies to reduce background labeling , for example by ( conditional ) reduction of endogenous biotin levels , could improve labeling time control in the future . Another limitation stems from temperature sensitivity . Although TbID and mTb work well at room temperature and elevated temperatures , they are inactive at 4°C ( Branon et al . , 2018 ) and thus likely incompatible with cold treatment as might be done for cold adaptation and cold stress experiments . Further , it is likely that some compartments will be harder to work in than others . Insufficient ATP and biotin availability and adverse pH or redox conditions could reduce TbID and mTb activity . For example , although the final step of biotin synthesis happens in mitochondria ( Alban , 2011 ) , free biotin in mitochondria is undetectable ( Baldet et al . , 1993 ) . It is possible that active biotin export from mitochondria will be a challenge for PL . Going forward , it will be interesting to see how TbID and mTb perform for different applications and which challenges arise from the use in other plants , tissues and organelles . Our experiments in Arabidopsis and N . benthamiana suggest that PL will be widely applicable in plants and will provide a valuable tool for the plant community .
N . benthamiana ecotype NB-1 was transformed with UBQ10pro::BirA-YFPNLS and UBQ10pro::BirA-NESYFP ( BirA = BirA* , TbID or mTb ) by infiltrating young leaves with a suspensions of Agrobacteria ( strain GV3101 ) carrying one of the binary vectors . Agrobacteria were grown from an overnight culture for 2 h , supplemented with 150 µM Acetosyringone , grown for another 4 h , pelleted and resuspended in 5% sucrose to an OD600 of 2 . For more stable expression , Agrobacteria carrying a 35S::p19 plasmid ( tomato bushy stunt virus ( TBSV ) protein p19 ) were co-infiltrated at a ratio of 1:1 for the temperature-dependency experiment . Two days after infiltration , expression was confirmed by epifluorescence microscopy and 5 mm wide leaf discs were harvested . Two to three discs were combined per sample . They were submerged in a 50 or 250 μM biotin solution , quickly vacuum infiltrated until the air spaces were filled with liquid and incubated at the indicated temperature for 1 h . Control samples were not treated or were infiltrated with H2O . After biotin treatment , leaf discs were dried and flash-frozen for later immunoblotting . All experiments were done in duplicates with leaf discs for each of the two replicates taken from different plants if possible . Only one replicate is shown . Activity of different BirA variants was compared in four independent experiments , with similar results , temperature dependency of TbID and mTb was tested in two and one experiment , respectively . Arabidopsis thaliana Col-0 was used as wild-type ( WT ) . The fama-1 mutant line is SALK_100073 ( Ohashi-Ito and Bergmann , 2006 ) . Plant lines for testing the activity of BirA* , TbID and mTb ( UBQ10pro::BirA*-YFPNLS , UBQ10pro::TbID-YFPNLS , UBQ10pro::mTb-YFPNLS , UBQ10pro::BirA*-NESYFP , UBQ10pro::TbID-NESYFP , UBQ10pro::mTb-NESYFP ) and for the ‘FAMA interactome’ and nuclear proteome’ experiments ( FAMApro::FAMA-TbID-mVenus in fama-1 , FAMApro::TbID-YFPNLS ) were generated by floral dip of WT or fama-1 +/- plants with the plasmids described above using agrobacterium strain GV3101 . Selection was done by genotyping PCR ( fama-1 ) and segregation analysis . We did not observe any obvious decrease in viability or developmental delay in our transgenic Arabidopsis plants . All lines had a single insertion event and were either heterozygous T2 or homozygous T3 or T4 lines . While screening for biotin ligase lines with the UBQ10 promoter , we observed that most regenerants had very weak YFP signal , especially the nuclear constructs . This was not observed with any of the cell-type-specific promoters we tested . Lines used for the FAMA-CFP AP-MS experiments were previously described in other studies: FAMApro::FAMA-CFP ( Weimer et al . , 2018 ) , SPCHpro::GFPNLS and MUTEpro::GFPNLS ( Adrian et al . , 2015 ) . Seeds were surface sterilized with ethanol or bleach and stratified for 2 to 3 days . For biotin treatment in whole Arabidopsis seedlings or roots and shoots , seedlings were grown on ½ Murashige and Skoog ( MS , Caisson labs ) plates containing 0 . 5% sucrose for 4 to 14 days under long-day conditions ( 16 h light/8 h dark , 22°C ) . For treatment of rosette leaves and flowers , seedlings were transferred to soil and grown in a long-day chamber ( 22°C ) until the first flowers emerged , at which point medium sized rosette leaves ( growing but almost fully expanded ) and inflorescences with unopened flower buds were harvested . All samples were pools from several individual plants . Biotin treatment was done by submerging the plant material in a biotin solution ( 0 . 5–250 μM biotin in water ) and either vacuum infiltrating the tissue briefly until the air spaces were filled with liquid ( approximately 5 min ) or not , followed by incubation at room temperature ( 22°C ) , 30°C or 37°C for up to 5 h . Controls were either treated with H2O or not treated . Following treatment , the plant material was dried and flash-frozen for later immunoblotting . To confirm reproducibility of the experiment , most experiments were done in duplicates ( only one replicate shown ) or repeated more than once with similar results . Comparison of all different BirA variants in Arabidopsis was done once with two independent lines for each NLS and NES constructs . Difference in activity and background between TbID and mTb matches other comparisons done with varying temperature and biotin concentration . Comparison of TbID and mTb at different temperatures was done three times . Comparison of the biotin ligase activity with different biotin concentrations was done twice for TbID and once for mTb . Three time courses with up to three or five hours of biotin treatment were done with the UBQ10pro::TbID-YFPNLS line and time courses with the FAMA-TbID line were done in duplicates . Experiments testing the effect of biotin application with and without vacuum infiltration in different tissues were done in duplicates . Activity of TbID in 4- to 14-day-old seedlings was tested in two independent experiments . Activity of TbID in roots and shoots of 6- to 14-day-old seedlings was tested once . Brightfield and epifluorescence images of N . benthamiana leaves and Arabidopsis seedlings , leaves and flowers were taken with a Leica DM6B microscope using a Leica CRT6 LED light source . Confocal microscopy images of Arabidopsis seedlings expressing different biotin ligase constructs were taken with a Leica SP5 microscope . For confocal microscopy , cell walls were stained with propidium iodide ( Molecular Probes ) by incubating in a 0 . 1 mg/ml solution for three to five minutes . Images were processed in FIJI ( ImageJ ) ( Schindelin et al . , 2012 ) . Several independent Arabidopsis lines and transiently transformed N . benthamiana leaves were analyzed . Images shown in figures and figure supplements are representative . Samples for immunoblots were prepared by resuspending frozen and ground plant material from biotin treatment assays with 1x Leammli buffer ( 60 mM Tris pH 6 . 8 , 2% SDS , 10% glycerol , 2 . 5% beta-mercaptoethanol , 0 . 025% bromphenol blue ) or mixing protein extracts 1:1 with 2x Leammli buffer and boiling the samples for five minutes at 95°C . Proteins bound to SA or GFP-Trap beads were eluted from the beads as described in the respective Materials and methods sections . Proteins were separated by SDS-PAGE and blotted onto Immobilon-P PVDF membrane ( 0 . 45 µm , Millipore ) using a Trans-Blot Semi-Dry transfer Cell ( BioRad ) . The following antibodies were used: Streptavidin-HRP ( S911 , Thermo Fisher Scientific ) , Rat monoclonal anti-GFP antibody ( 3H9 , Chromotek ) , Rat Anti-HA High Affinity ( 11867423001 , Roche ) , Myc-Tag ( 71D10 ) Rabbit mAb ( 2278 S Cell Signaling ) , AffiniPure Donkey Anti-Rat IgG-HRP ( 712-035-153 , Jackson Immuno Research Laboratories ) , Rabbit Anti-Rat IgG-HRP ( A5795 , Sigma ) , Goat anti-Rabbit IgG-HRP ( AS09 602 , Agrisera ) . Blots were probed with primary antibodies overnight at 4°C or for up to 1 h at room temperature and with the secondary antibody for 1 h at room temperature and incubated with ECL Western blotting substrates according to the manufacturer’s instructions . Signals were detected on X-ray films or on a ChemiDoc MP Imaging System ( BioRad ) . Primer namePurposeSequenceBirA-fwCloningCAGGCGCGCCGGTGGAGGCGGTTCAGGAGGTGGCATGGGCAAGCCCATCCCCAACBirA-NES-rvCloningCTGGCGCGCCCACCTCCGCCGCTTCCACCGCCTCCGTCCAGGGTCAGGCGCTCCAGBirA-rvCloningTGGCGCGCCCACCTCCGCCGCTTCCACCGCCTCCCTTTTCGGCAGACCGCAGACTGATTTpDONR-mut-AscI-fwCloningGTACAAAGTGGCTGGGCGCGCCTCCATGGTGAGCAAGGpDONR-mut-AscI-rvCloningCCAGCCACTTTGTACAAGAAAGTTGAACGAGgFAMA-fwCloningCACCATGGATAAAGATTACTCGGTACGTACGgFAMA-rvCloningAGTAAACACAATATTTCCCAGGTTAGAGCBirA-YFPnls-fwCloningGCATGCGGCCGCATGGGCAAGCCCATCBirA-YFPnls-rvCloningGATACCATGGAACCTCCGCCGCTTCCMUTE-fwCloningCACCATGTCTCACATCGCTGTTGAAAGGAATCGMUTE-rvCloningATTGGTAGAGACGATCACTTCATCAGACICE1-fwCloningCACCATGGGTCTTGACGGAAACAATGGICE1-rvCloningGATCATACCAGCATACCCTGCTSEU-fwCloningCACCATGGTACCATCAGAGCCGCCSEU-rvCloningCGCGTTCCAATCAAAATTGTTGAAACLUH-fwCloningCACCATGGCTCAGAGTAATTGGGAAGLUH-rvCloningCTTCCAAATCTTTACGGATTTGTCATGBZR1-fwCloningCACCATGACTTCGGATGGAGCTACGBZR1-rvCloningACCACGAGCCTTCCCATTTCBIM1-fwCloningCACCATGGAGCTTCCTCAACCTCGTCBIM1-rvCloningCTGTCCCGTCTTGAGCCGTTfama1-RPGenotypingCAATACAAAAAGCTCCCCTCACfama1-LBb1 . 3GenotypingATTTTGCCGATTTCGGAAC Entry vectors for creating protein of interest-TbID/mTb + YFP/mVenus fusionsPlasmid nameGateway sitesApplicationAddgene IDpENTR_L1-YFP-Turbo-NES-L2attL1-attL2For N-terminal YFP-TbID fusion to non-nuclear proteins127349pENTR_L1-YFP-Turbo-L2attL1-attL2For N-terminal YFP-TbID fusion to nuclear proteins127350pENTR_L1-YFP-miniTurbo-NES-L2attL1-attL2For N-terminal YFP-mTb fusion to non-nuclear proteins127351pENTR_L1-YFP-miniTurbo-L2attL1-attL2For N-terminal YFP-mTb fusion to nuclear proteins127352pDONR_P2R-P3_R2-Turbo-NES-mVenus-STOP-L3attR2-attL3For C-terminal TbID-mVenus fusion to non-nuclear proteins127353pDONR_P2R-P3_R2-Turbo-mVenus-STOP-L3attR2-attL3For C-terminal TbID-mVenus fusion to nuclear proteins127354pDONR_P2R-P3_R2-miniTurbo-NES-mVenus-STOP-L3attR2-attL3For C-terminal mTb-mVenus fusion to non-nuclear proteins127355pDONR_P2R-P3_R2-miniTurbo-mVenus-STOP-L3attR2-attL3For C-terminal mTb-mVenus fusion to nuclear proteins127356Entry vectors for expressing BirA*/TbID/mTb + YFP under a promoter of choicePlasmid nameGateway sitesapplicationAddgene IDpENTR_L1-BirA ( R118G ) -NES-YFP-STOP-L2attL1-attL2For expressing cytosolic BirA* under a promoter of choice127357pENTR_L1-Turbo-NES-YFP-STOP-L2attL1-attL2For expressing cytosolic TbID under a promoter of choice127358pENTR_L1-miniTurbo-NES-YFP-STOP-L2attL1-attL2For expressing cytosolic mTb under a promoter of choice127359pENTR_L1-BirA ( R118G ) -YFP-NLS-STOP-L2attL1-attL2For expressing nuclear BirA* under a promoter of choice127360pENTR_L1-Turbo-YFP-NLS-STOP-L2attL1-attL2For expressing nuclear TbID under a promoter of choice127361pENTR_L1-miniTurbo-YFP-NLS-STOP-L2attL1-attL2For expressing nuclear mTb under a promoter of choice127362Binary plant transformation vectors for ubiquitous expression of BirA*/TbID/mTb + YFP/mVenusPlasmid nameResistance in plantsapplicationAddgene IDR4pGWB601_UBQ10p-BirA ( R118G ) -NES-YFPBASTAFor expressing cytosolic BirA* under the UBQ10 promoter127363R4pGWB601_UBQ10p-BirA ( R118G ) -YFP-NLSBASTAFor expressing nuclear BirA* under the UBQ10 promoter127365R4pGWB601_UBQ10p-Turbo-NES-YFPBASTAFor expressing cytosolic TbID under the UBQ10 promoter127366 R4pGWB601_UBQ10p-Turbo-YFP-NLSBASTAFor expressing nuclear TbID under the UBQ10 promoter127368R4pGWB601_UBQ10p-miniTurbo-NES-YFPBASTAFor expressing cytosolic mTb under the UBQ10 promoter127369R4pGWB601_UBQ10p-miniTurbo-YFP-NLSBASTAFor expressing nuclear mTb under the UBQ10 promoter127370 | Cells contain thousands of different proteins that work together to control processes essential for life . To fully understand how these processes work it is important to know which proteins interact with each other , and which proteins are present at specific times or in certain cellular locations . Investigating this is particularly difficult if the proteins of interest are rare , either because they are present only at low levels or because they are unique to a particular type of cell . One such protein known as FAMA is only found in young guard cells in plants . Guard cells are rare cells that surround pores on the surface of leaves . They help open or close the pores to allow carbon dioxide and water in and out of the plant . Inside these cells , FAMA regulates the activity of genes in the nucleus , the compartment in the cell that houses the plant’s DNA . Two recently developed molecular biology tools , called TurboID and miniTurbo , allow researchers to identify proteins that are in close contact with a protein of interest or are present at a specific place inside living animal cells . These tools use a modified enzyme to add a small chemical tag to proteins that are close to it , or anything to which it is anchored . Mair et al . adapted these tools for use in plants and tested their utility in two species that are commonly used in research: a tobacco relative called Nicotiana benthamiana , and the thale cress Arabidopsis thaliana . Their experiments showed that TurboID and miniTurbo can be used to tag proteins in different types of plant cells and organs , as well as at different stages of the plants’ lives . To test whether the tools are suitable for identifying partners of rare proteins , Mair et al . used FAMA as their protein of interest . Using TurboID , they detected several proteins in close proximity to FAMA , including some that FAMA was not previously known to interact with . Mair et al . also found that TurboID could identify a number of proteins that were present in the nuclei of guard cells . This shows that the tool can be used to detect proteins in sub-compartments of rare plant cell types . Taken together , these findings show that TurboID and miniTurbo may be customized to study plant protein interactions and to explore local protein ‘neighborhoods’ , even for rare proteins or specific cell types . To enable other plant biology researchers to easily access the TurboID and miniTurbo toolset developed in this work , it has been added to the non-profit molecular biology repository Addgene . | [
"Abstract",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"tools",
"and",
"resources"
] | 2019 | Proximity labeling of protein complexes and cell-type-specific organellar proteomes in Arabidopsis enabled by TurboID |
R-loops are features of chromatin consisting of a strand of DNA hybridized to RNA , as well as the expelled complementary DNA strand . R-loops are enriched at promoters where they have recently been shown to have important roles in modifying gene expression . However , the location of promoter-associated R-loops and the genomic domains they perturb to modify gene expression remain unclear . To resolve this issue , we developed a bisulfite-based approach , bisDRIP-seq , to map R-loops across the genome at near-nucleotide resolution in MCF-7 cells . We found the location of promoter-associated R-loops is dependent on the presence of introns . In intron-containing genes , R-loops are bounded between the transcription start site and the first exon-intron junction . In intronless genes , the 3' boundary displays gene-specific heterogeneity . Moreover , intronless genes are often associated with promoter-associated R-loop formation . Together , these studies provide a high-resolution map of R-loops and identify gene structure as a critical determinant of R-loop formation .
R-loops are nucleic acid structures in which a strand of RNA is hybridized to a strand of DNA , while the other strand of DNA is looped out . Recent techniques for genome-wide mapping of R-loops revealed that promoter regions are enriched in R-loops ( Ginno et al . , 2012 ) . The presence of R-loops in promoter regions raises the possibility that they may regulate gene expression . Indeed , more recent studies provided evidence that R-loops in these critical regions can alter histone modifications and are associated with changes in gene transcription ( Chen et al . , 2015; Colak et al . , 2014; Sun et al . , 2013 ) . A major unanswered question is the precise location of these R-loops . The location of an R-loop in a gene is likely to dictate how that R-loop can impact promoter function . This is because eukaryotic promoter regions contain multiple functional domains that have distinct roles in transcription , including transcription start sites , transcription factor-binding sites , exon-intron junctions , CpG islands , and nucleosome-associated DNA ( Lenhard et al . , 2012 ) . The location of an R-loop within the promoter could influence transcription by disrupting or enhancing protein recruitment to any of these sites . Thus , understanding the precise location of R-loops can provide insight into how R-loops affect gene transcription . A major barrier to discovering the exact location of R-loops is the low resolution of current genome-wide R-loop mapping methods like DRIP-seq ( DNA-RNA immunoprecipitation sequencing ) ( Ginno et al . , 2012 ) . This method uses the S9 . 6 antibody which binds RNA-DNA hybrids ( Boguslawski et al . , 1986 ) . With this antibody , genome-wide R-loop maps are created by immunoprecipitating and sequencing the genomic fragments containing RNA-DNA hybrids ( Ginno et al . , 2012; Sanz et al . , 2016; Stork et al . , 2016 ) . However , DRIP-seq does not discriminate between the R-loop sequence and the surrounding non-R-loop sequence . Therefore , the exact boundaries of R-loops cannot be resolved using DRIP-seq . Based on DRIP-seq and similar R-loop mapping methods , promoter-associated R-loops are thought to form within a few kilobases downstream of the transcription start site ( Chédin , 2016 ) . From these low-resolution experiments , it is not clear if R-loops have specific boundaries or if they are relatively amorphous structures that lack well-defined boundaries . To understand where R-loops are positioned in genomic promoter regions , we developed bisDRIP-seq ( bisulfite-DNA-RNA immunoprecipitation sequencing ) . bisDRIP-seq is an approach to map R-loops at near-nucleotide resolution throughout the genome . In this approach , we use bisulfite to selectively convert cytosine residues into uracil residues within genomic DNA regions that contain single-stranded DNA . We then identify single-stranded regions likely to be in R-loops based on preferential labeling of one strand of DNA and the requirement that the labeling be transcription dependent . Remarkably , we find that promoter-associated R-loops are typically bounded by the transcription start site and the first exon-intron junction in intron-containing genes . Thus , we find that the maximum size of promoter-associated R-loops is controlled by the location of the first exon-intron junction in intron-containing genes . We also identify prominent promoter-associated R-loop forming regions in intronless genes , including MALAT1 , NEAT1 , and the replication-dependent histone genes . In some of these genes , the R-loops are associated with well-defined 3’ boundaries that are located within the gene body . Thus our high-resolution map of promoter-associated R-loops defines the boundaries of R-loops and suggests a role for first exon length in regulating the formation of R-loops .
It is not yet possible to define the exact size and location of R-loop-forming regions on a genome-wide scale . In contrast , it is possible to determine the exact location of a specific R-loop-forming region in a specific gene using a previously developed bisulfite mapping approach ( Yu et al . , 2003 ) . Essentially , this approach involved treating genomic DNA with bisulfite under non-denaturing conditions . Bisulfite specifically causes cytosine-to-uracil conversions in the single-stranded DNA portion of R-loops . On the other hand , cytosines in the RNA-DNA hybrid portion of the R-loop are protected from bisulfite conversion . Sequencing of both strands then revealed the location and strand in which cytosines were converted to uracils . The presence of an R-loop was identified by showing that the converted cytosines occurred primarily on one of the two strands of DNA ( Yu et al . , 2003 ) . The location of bisulfite-induced conversions on a single strand of DNA was then used to define the boundaries of the R-loop at near-nucleotide resolution . This use of bisulfite to detect single-stranded regions of DNA contrasts with the use of bisulfite in mapping 5-methylcytosine in DNA . For 5-methylcytosine mapping , DNA is denatured into single strands and then all cytosines are converted to uracils , while 5-methylcytosine is poorly converted ( Frommer et al . , 1992 ) . Thus , while mapping 5-methylcytosine involves searching for unconverted cytosines , mapping single-stranded DNA involves detection of converted cytosines . The use of bisulfite to map R-loops on a genome-wide scale poses several significant challenges . First , non-physiological R-loop formation , removal , expansion , or contraction can occur between the time of lysis and the time when bisulfite is used to mark single-stranded DNA ( Kaback et al . , 1979; Landgraf et al . , 1996 ) . Second , high sequencing depth is needed to search for converted cytosines across the entire genome ( Sims et al . , 2014 ) . Finally , bisulfite causes infrequent , but measurable , background conversions in double-stranded DNA ( Yu et al . , 2003 ) , which can generate a significant amount of noise on a genome-wide scale . To map R-loops at near-nucleotide resolution , we developed a genome-wide bisulfite-based approach called bisDRIP-seq ( Figure 1A ) . This method incorporates steps to overcome each of the challenges listed above . In this approach , cells are lysed in the presence of bisulfite and SDS . By including bisulfite during lysis , genomic DNA structures have as little time as possible to change conformation prior to the bisulfite modification of cytosines in single-stranded DNA regions . To overcome the problem of needing high sequence coverage , the S9 . 6 antibody ( Boguslawski et al . , 1986 ) is used to enrich for R-loops . The S9 . 6 antibody has high affinity for RNA-DNA hybrids and lower affinity for other double-stranded RNA sequences ( Phillips et al . , 2013 ) . Thus , after genomic DNA is sheared using restriction digestion , the S9 . 6 antibody enriches the bisulfite-modified R-loops for sequencing analysis . Finally , to overcome the problem of stochastic cytosine conversions , a computational pipeline was developed to identify single-stranded regions using bisDRIP-seq data . This pipeline was developed to identify regions with high concentrations of cytosine-to-uracil conversions . This computational pipeline also identifies single-stranded regions that are likely to contain R-loops as opposed to other single-stranded DNA structures in the genome . Additionally , the pipeline was designed to reveal the specific strand orientation of R-loops . Thus , bisDRIP-seq provides an approach to map R-loops at near-nucleotide resolution on a genome-wide scale . We used bisDRIP-seq to map single-stranded DNA throughout the genome . Thirteen bisDRIP-seq experiments were performed on different samples of MCF-7 cells . After performing bisDRIP-seq on these samples , the DNA fragments were sequenced using a traditional post-bisulfite library preparation method ( see Materials and methods ) . We next aligned the sequenced reads to the genome using Bismark , an alignment approach typically used for 5-methylcytosine mapping ( Krueger and Andrews , 2011 ) . This was necessary since the conversion of cytosines to uracils would confound traditional read alignment programs . Bismark was used to map conversions associated with single-stranded DNA as follows: first , reads were aligned to the genome . Then the cytosines that had been converted to uracils were identified ( Figure 1B ) . As expected , reads were detected that contained only a single conversion , consistent with noise due to low-level double-stranded DNA cytosine conversions . However , reads and regions were also observed that contained consecutive cytosine conversions ( Figure 1—figure supplement 1A ) . These reads are suggestive of single-stranded DNA . We next applied our bioinformatic pipeline to distinguish the multiple conversions seen in single-stranded DNA regions from the stochastic conversions due to background noise . In the first part of this pipeline , the stochastic rate of conversions was estimated . This was estimated based on the overall cytosine-to-uracil conversion rate in a given sample . Reads with a high percentage of conversions were excluded from this calculation since it was assumed that those conversions were not stochastic . Next , we generated a ‘bisDRIP-seq score’ for each read . This score was calculated based on the number of cytosines converted in the read , relative to the likelihood of observing that number of conversions by stochastic noise ( see Materials and methods for more details ) . Next , the score was normalized so that the sum of bisDRIP-seq scores within each sample was the same . Next , we calculated bisDRIP-seq scores for individual nucleotide positions in the genome . The bisDRIP-seq score for each nucleotide position was calculated as the sum of the bisDRIP-seq scores of all of the reads that overlapped with that nucleotide position . These scores provide a read-length resolution map of single-stranded DNA that filters out stochastic conversions and that is more comparable between genomic regions ( Figure 1B ) . Importantly , bisDRIP-seq scores were substantially reduced at specific gene loci in samples treated with RNase H , which degrades RNA in RNA-DNA hybrids ( Figure 1—figure supplement 2 ) . This supports the idea that elevated bisDRIP-seq scores reflect R-loops . We next performed two basic tests of the quality of data produced by bisDRIP-seq: First , we asked whether reads with high bisDRIP-seq scores are randomly distributed across the genome or whether they are clustered in specific regions . To simulate a random distribution , Monte Carlo simulations were applied to our data . Relative to these Monte Carlo simulations , bisDRIP-seq scores were found to be clustered in specific regions ( Figure 1—figure supplement 1B–D ) . Second , we performed correlation tests between the bisDRIP-seq scores obtained in each of our samples . In all cases , there was significant correlation between bisDRIP-seq samples ( p<10−16 , Spearman's rank-correlation test , Figure 1—figure supplement 3 ) . We next wanted to know if bisDRIP-seq scores are associated with promoter regions that contain R-loops . R-loops were previously mapped to promoter regions ( Ginno et al . , 2012 ) , where they are thought to play important roles in gene expression ( Chen et al . , 2015 ) . These R-loops , as mapped using DRIP-seq , were found to be transcription dependent and correlated with gene expression ( Sanz et al . , 2016 ) . We therefore wanted to determine if bisDRIP-seq scores have a similar enrichment and transcription dependence in active promoter regions . In order to investigate promoter regions , we compiled a list of transcription start sites using the GENCODE database ( Harrow et al . , 2012 ) ( see Materials and methods ) . We then defined promoter regions , for the purpose of our analyses , as the region one kilobase on either side of each of these transcription start sites . We next performed several simple analyses to ensure that the results from bisDRIP-seq experiments recapitulate the promoter-region enrichment observed in DRIP-seq studies . First , bisDRIP-seq scores , like DRIP-seq reads , were found to be enriched in promoter regions relative to downstream exon-containing regions ( Figure 2A and Figure 2—figure supplement 1A ) . Next , we found that the number of DRIP-seq reads correlates with bisDRIP-seq scores in individual promoter regions ( p<2 . 2×10−16 , Spearman's rank-correlation test , Figure 2—figure supplement 1B ) . Thus , bisDRIP-seq scores , like DRIP-seq reads , are enriched in promoter regions . We next asked if bisDRIP-seq enrichment in promoter regions depends on active transcription . To test if bisDRIP-seq enrichment requires transcription , bisDRIP-seq was repeated using MCF-7 cells treated with the transcription-inhibitor triptolide ( Kupchan et al . , 1972; Titov et al . , 2011; Vispé et al . , 2009 ) . bisDRIP-seq enrichment was reduced in these samples ( Figure 2—figure supplement 1C ) . This suggests that bisDRIP-seq enrichment in promoter regions depends on ongoing transcription . We next asked if the bisDRIP-seq scores in promoter regions are correlated with promoter activity . Promoter activity was assessed using existing GRO-seq datasets from MCF-7 cells cultured in a similar manner to our cells ( Hah et al . , 2013; Hah et al . , 2011 ) . GRO-seq measures the presence of active polymerases on a genome-wide scale ( Core et al . , 2008 ) . Thus , these GRO-seq datasets allow us to distinguish between promoter regions with low and high promoter activity . We therefore compared the number of GRO-seq reads with bisDRIP-seq scores in promoter regions ( Figure 2—figure supplement 1D ) . In this analysis , GRO-seq-measured promoter activity correlates with bisDRIP-seq scores ( p<2 . 2×10−16 , Spearman's rank-correlation test ) . Thus , bisDRIP-seq scores are enriched in active promoter regions . Notably , the enrichment of bisDRIP-seq scores in active promoter regions was reduced when RNA synthesis was blocked by triptolide treatment ( Figure 2—figure supplement 1D ) . Together , these results confirm that bisDRIP-seq scores , like DRIP-seq reads , are enriched in active promoter regions . We next wanted to determine if the bisDRIP-seq score enrichment in promoter regions is due to co-transcriptional R-loops . Conceivably , cytosine conversions could occur if single-stranded DNA is exposed as a result of other single-stranded DNA structures near transcription start sites , including unwound DNA due to supercoiling ( Hsieh and Wang , 1975 ) , G-quadruplexes ( Sen and Gilbert , 1988 ) , or genomic regions that contain paused polymerases ( Core et al . , 2008 ) that become more accessible to bisulfite after SDS treatment . Relative to other single-stranded DNA structures , R-loops are known to produce a specific cytosine conversion signature upon bisulfite treatment . Specifically , cytosine conversions are limited to one strand of DNA in an R-loop . In other types of genomic structures that expose single-stranded DNA , cytosines on both strands of DNA can be converted . Thus efforts to map R-loops with bisulfite must demonstrate preferential labeling of one strand of DNA ( Yu et al . , 2003 ) . Cytosine conversions can also occur as a result of two types of R-loops: ‘sense-strand R-loops’ and ‘antisense-strand R-loops . ’ Sense-strand R-loops refer to R-loops in which the RNA component of the R-loop is transcribed from the annotated promoter in the expected direction . In these R-loops , the non-template strand of DNA is exposed for bisulfite-mediated conversion . Antisense-strand R-loops , on the other hand , contain RNA that is transcribed from the opposite DNA strand . In this case , the template strand of DNA for the canonical gene transcript would be modified by bisulfite . Thus , the specific strand of DNA that exhibits bisulfite-mediated cytosine conversions indicates if a sense-strand R-loop or antisense-strand R-loop is present . In order to distinguish between R-loops and other single-stranded DNA structures , we took advantage of the cytosine conversion strand asymmetry that is expected to result from R-loops . In other types of genomic structures that expose single-stranded DNA , cytosines on both strands of DNA can be converted . To determine if the observed bisDRIP-seq enrichment downstream of the transcription start site in promoter regions is caused by sense-strand R-loops , we repeated our comparison of bisDRIP-seq score with promoter activity . However , here bisDRIP-seq scores were plotted separately for the template and non-template strands of DNA ( Figure 2B ) . With increasing promoter activity we observed increasing bisDRIP-seq scores on both strands , with preferential labeling of the non-template strand of DNA . Thus , although both strands of DNA had bisulfite-induced cytosine conversions , we also observed asymmetric labeling of the non-template strand ( Figure 2C ) . This is consistent with a mixture of single-stranded DNA structure and sense-strand R-loops in active promoter regions ( Figure 2D ) . Notably , the higher bisDRIP-seq scores on the non-template strand than the template strand were largely eliminated in triptolide-treated samples ( Figure 2B and C ) . This suggests that the sense-strand R-loops in the promoter region are transcription dependent . The exact starting position and ending positions of promoter-associated R-loops remains unclear . This is due to the low resolution of conventional R-loop mapping methods ( Chen et al . , 2015; Ginno et al . , 2012 ) . We wanted to take advantage of the high resolution of bisDRIP-seq to map the exact boundaries of R-loops in promoter regions . We first asked where R-loops are located in relation to transcription start sites . In many promoters , transcription initiates from multiple nearby transcription start sites ( Carninci et al . , 2006 ) . This creates a practical limit to the precision that we can achieve in mapping the location of R-loops relative to transcription start sites . We first mapped R-loops relative to all transcription start sites using metaplots of bisDRIP-seq scores . First , bisDRIP-seq scores were calculated for each nucleotide position surrounding the transcription start site in all individual promoter regions . Promoter regions were defined using the GENCODE database described above . Then , the bisDRIP-seq score at a given nucleotide position relative to the transcription start site was summed across all promoter regions . These scores were then plotted separately for the non-template strand and the template strand ( Figure 3A ) . The resulting metaplot suggests that a mix of R-loops and other single-stranded structures surround the transcription start site . The presence of single-stranded DNA at transcription start sites is suggested by the peak of bisDRIP-seq scores near the transcription start site ( Figure 3B ) . The location of these single-stranded structures is consistent with previous maps of single-stranded DNA ( Kouzine et al . , 2013 ) . However , the presence of R-loops is specifically suggested by asymmetric , preferential labeling of the non-template strand ( Figure 3B ) . Indeed , the non-template strand bisDRIP-seq scores are significantly higher than the template strand bisDRIP-seq scores immediately 3’ of the transcription start site ( p<0 . 005 , Wilcoxon signed-rank test ) ( Figure 3A , Figure 3—figure supplement 1B ) . This suggests that the transcription start site is the 5’ boundary of promoter-associated R-loops . We next asked whether the R-loops bounded by the transcription start site are dependent on transcription . We repeated our metaplot analysis using the samples treated with triptolide . In these samples , there was minimal difference between the template and non-template strand bisDRIP-seq scores ( Figure 3A ) . The small remaining difference in scores upon triptolide-treatment may reflect a real difference , but it may also reflect noise in our measurements . Overall , the loss of bisDRIP-seq score strand asymmetry upon triptolide treatment demonstrates that the enrichment of R-loops 3' of the transcription start site requires transcription . The transcription start site boundary was even more apparent after applying a background correction . To do this , we defined an ‘R-loop signal , ’ which reflects the difference in bisDRIP-seq labeling of the non-template strand from the template strand after using the triptolide-treated samples for background correction . Thus , the template strand metaplot from the triptolide-treated sample was subtracted from the template strand metaplot from the control sample . The same background correction was used for the non-template strand metaplot . This background correction further enhances the demarcation of the transcription start site as the 5’ boundary of promoter-associated R-loops ( Figure 3C ) . Next , we subtracted the template strand metaplot from the non-template metaplot to generate a metaplot of the R-loop signal ( Figure 3D ) . In this plot , the 5’ boundary of R-loop signal at the transcription start site is very pronounced . We repeated this analysis on promoters that have either high promoter activity or low promoter activity ( Figure 3E and F , Figure 3—figure supplement 1C , Figure 3—figure supplement 2 ) . A peak of R-loop signal was only observed immediately 3' of the transcription start site in the analysis of active promoters . This is consistent with our previous analysis showing that R-loops specifically form 3' of the transcription start site in active promoters . Notably , the observed difference in strand bisDRIP-seq scores 3' of the transcription start site in active promoters is lost after RNase H treatment ( Figure 3—figure supplement 3 ) . This confirms that the R-loop signal observed in active promoter regions is caused by R-loops . Taken together , these data indicate that the transcription start site demarcates the 5’ boundary of promoter-associated R-loops . Additionally , since only noise was observed from promoter regions with low promoter activity , these regions were removed from future analysis unless otherwise noted . We next wanted to know if there is a 3’ boundary to R-loops . Based on the metaplot analysis in Figure 3D , R-loop signal drops approximately 200–250 bp downstream of the transcription start site . This is further from the transcription start site than the typical first post-transcription start site nucleosome ( Schones et al . , 2008 ) and the location of promoter-proximal RNA polymerase II pausing ( Core et al . , 2008 ) , suggesting that these features probably do not impede R-loop expansion . On the other hand , 200–250 bp is reasonably close to the median distance between the transcription start site and the first exon-intron junction , which is 181 bp in our dataset ( see Materials and methods ) . Also , previous studies found that knockdown of the 5' splice site-binding factor SRSF1 induces the formation of R-loops ( Li and Manley , 2005 ) , which could be explained if splicing is involved in bounding R-loop expansion . These pieces of evidence suggested that the first exon-intron junction might act as the 3' boundary to R-loop expansion in promoter regions . We therefore asked where R-loops are located relative to the first exon-intron junction . We used the 5' end of the first intron as the reference point for a metaplot of bisDRIP-seq scores . Intronless genes were not considered in this analysis . In these metaplots , we observed that the bisDRIP-seq scores on the non-template strand are significantly higher than on the template strand immediately 5' of the first exon-intron junction ( Figure 4A and Figure 4—figure supplement 1A ) . This strand asymmetry in bisDRIP-seq scores drops 3' of the exon-intron junction . This suggests that the 3’ end of R-loops are bounded by the first exon-intron junction . We further tested the idea that the first exon-intron junction is the 3’ boundary of R-loops . To do this , we asked if the size of the R-loop-forming region in promoter regions increases with the length of the first exon . To test this question , metaplot analysis was repeated on groups of promoter regions with different sized first exons . First , promoter regions were binned into five groups based on the annotated size of the first exon . Then , metaplots were created of the R-loop signal centered around either the transcription start site or the first exon-intron junction ( Figure 4B ) . Strikingly , in groups of promoter regions with longer first exons , the R-loop signals are also longer . This supports the idea that R-loops are bounded at both the transcription start site and the first exon-intron junction . We next asked if we could map the location of the 3' R-loop boundary at near-nucleotide resolution . bisDRIP-seq scores are shared across entire reads , which limits the resolution of using bisDRIP-seq scores to read length resolution . On the other hand , cytosine-to-uracil conversions should map R-loops at near-nucleotide resolution . We therefore generated a metaplot of the strand asymmetry of cytosine conversions relative to the first exon-intron junction . First , the number of conversions on either strand were background corrected by subtracting the number of conversions observed in our triptolide bisDRIP-seq data . Next , the log ratio of conversions on the non-template strand relative to the template strand was plotted at each site relative to the exon-intron junction ( Figure 4C ) . In this metaplot there is a decrease in the relative number of conversions on the non-template strand within base pairs of the exon-intron junction ( Figure 4—figure supplement 1B ) . Thus , it appears that the R-loop boundary is within a few base pairs of the first exon-intron junction . We next asked whether other exon-intron junctions also act as R-loop boundaries . In particular , we focused on the junctions between the first intron and the second exon , as well as the junction between the second exon and the second intron . We repeated our metaplot analysis by plotting R-loop signal at each position relative to the given exon-intron or intron-exon junction ( Figure 4—figure supplement 1A ) . No clear peak in R-loop signal is observed near these downstream intron-exon junctions . This suggests that only the first exon-intron junction acts as a boundary to R-loop formation . Together , these results suggest that there is a boundary to R-loop formation located within base pairs of the first exon-intron junction . We noticed that there is negative R-loop signal upstream of the transcription start site ( Figure 3D and F ) . This could be caused by ‘antisense-strand R-loops , ’ i . e . , with antisense RNA transcripts hybridized to the annotated non-template strand of DNA ( See Figure 3—figure supplement 4C for this structure ) . Antisense-strand R-loops would result in more prominent bisulfite conversions on the annotated template strand . This type of labeling is opposite from the non-template strand labeling that is caused by the predominant type of R-loop that forms from sense transcription . We considered that antisense transcription could lead to antisense-strand R-loops that generate these negative R-loop signals . To test this , we calculated the antisense-transcription activity of each promoter region upstream of the transcription start site using the previously described GRO-seq dataset ( Hah et al . , 2013 ) . In promoter regions with high antisense-transcription promoter activity , the template-strand bisDRIP-seq scores upstream of the transcription start site were significantly higher than on the non-template strand ( Figure 3—figure supplement 4A and B ) . These data indicate a correlation between antisense transcription and antisense R-loops . We next used the promoters that showed the highest level of antisense-transcription promoter activity to generate a metaplot of R-loop signal . In this metaplot , the negative R-loop signal was prominent upstream of the transcription start site ( Figure 3—figure supplement 4D ) . Taken together , these data suggest that some promoter regions contain antisense-strand R-loops and that this is linked to antisense transcription in these promoter regions . We next wanted to identify the promoters that show the strongest association with transcription-dependent R-loops . Any unique features associated with these promoters may be directly related to the R-loops forming at these promoter regions . We searched for promoter regions with two major features: First , we searched for promoter regions that showed disproportionately high bisDRIP-seq score on the non-template strand compared to the template strand . Second , we searched for promoter regions where the majority of the bisDRIP-seq score on the non-template strand was lost upon triptolide treatment . In this analysis , we noticed a set of promoter regions that exhibited both of these features ( Figure 5A ) . We therefore ranked genes based on the sum of these two features to identify the genes that show the strongest association with R-loop structures ( Table 1 ) . Two important classes of genes were identified in this analysis and both lack introns . First , six of the top 25 genes ( 24% ) and nine of the top 50 genes ( 18% ) are replication-dependent histone genes . The second class of genes encode intronless noncoding RNAs , including MALAT1 , NEAT1 , RPPH1 , and CTB-58E17 . 1 . We noticed that two other top hits , the protein-coding genes RHOB and JUNB , are also intronless genes . In general , intronless genes appear to be enriched among the promoters that show strong association with R-loop structures . In total , 44% of the top 25 genes in our list are intronless , compared to approximately 2% of long non-coding RNAs ( lncRNAs ) and 3% of protein-coding genes ( Derrien et al . , 2012; Louhichi et al . , 2011 ) . The presence of R-loops in the promoter regions of these intronless genes suggests that the first exon-intron junction does not promote R-loop formation . The presence of R-loops in the replication-dependent histone genes is potentially interesting given how these genes are regulated . As their name suggests , these histone genes are regulated in a cell-cycle dependent manner ( Robbins and Borun , 1967 ) . They also lack poly-A tails and are processed in special histone bodies in the nucleus ( Dominski and Marzluff , 1999 ) . Thus these genes are co-regulated using special processing pathways . We first considered the possibility that the prominent R-loop signal detected in histone promoter regions simply reflects high promoter activity in these genes . However , histone genes consistently had higher R-loop-associated bisDRIP-seq scores than the majority of genes with similar promoter activity ( Figure 5B ) . This suggests that promoter activity does not explain the strong R-loop signal observed in histone genes . Similar analysis indicated that nuclear RNA levels and recruitment of RNA polymerase II also do not explain the R-loop signal observed in histone genes ( Figure 5—figure supplement 1A and B ) . We next examined the prevalence of R-loops and other single-stranded DNA structures in the promoter regions of histone genes . We repeated our metaplot analysis using the bisDRIP-seq scores from the entire class of replication-dependent histone genes ( Figure 5C ) . In the resulting metaplot , bisDRIP-seq scores are low on the template strand and in triptolide-treated samples . This suggests that there is little single-stranded structure outside of R-loops in the promoter regions of these genes . In contrast , bisDRIP-seq scores are high downstream of the transcription start site on the non-template strand . Moreover , these high bisDRIP-seq scores are not observed after RNase H treatment ( Figure 5—figure supplement 1C–E ) . Together , these results indicate that histone genes , as a group , have a high level of R-loop formation relative to the formation of other single-stranded DNA structures . We next asked whether R-loops are bounded in replication-dependent histone genes . It was not clear if R-loops would be bounded in these genes since they lack introns and therefore they lack exon-intron junctions ( Marzluff et al . , 2008 ) . Conceivably , the R-loops could extend to the entire length of the transcript . We therefore identified the boundaries of the R-loop signal in each of the nine histone genes that had the highest propensity to form R-loops ( Figure 5A ) . As expected , the 5’ boundary of the R-loops appear to be near the transcription start site in all nine genes . In five of the nine histone genes , the entire R-loop appeared to be restricted to the initial portion of the gene ( Figure 5D and Figure 5—figure supplement 2A–E ) . Sequence analysis of these boundaries does not reveal a clear sequence enrichment or motif ( Figure 5—figure supplement 1D and E ) , making it currently unclear how this boundary is determined . In other cases , like HIST1H1E , R-loops seemed to cover nearly the entire gene ( Figure 5E and Figure 5—figure supplement 2F–I ) . Together these results suggest that additional factors may establish 3’ R-loop boundaries in a subset of the replication-dependent histone genes . Another set of genes which preferentially exhibit R-loops in their promoter regions are MALAT1 and NEAT1 . MALAT1 and NEAT1 are adjacent genes that encode abundant , intronless lncRNAs ( Hutchinson et al . , 2007 ) . These lncRNAs remain in the nucleus where they are involved in the regulation of transcription ( Hirose et al . , 2014 ) and splicing ( Tripathi et al . , 2010 ) , respectively . Both MALAT1 and NEAT1 are longer than 3 kb , which is longer than the replication-dependent histone genes studied above . We were therefore interested in whether there are boundaries to R-loop expansion in these much longer intronless genes . We first asked where R-loops are located in MALAT1 and NEAT1 . The R-loop forming region in MALAT1 extends from the transcription start site to a position approximately 1700 bp downstream , with a sharp decrease in R-loop signal downstream of this position ( Figure 6A ) . Similarly , the R-loop in NEAT1 extended approximately 1400 bp from the transcription start site ( Figure 6B ) . Beyond this site , there was minimal detectable R-loop signal . As with the R-loops in the replication-dependent histone genes , these R-loops showed nearly complete loss of bisDRIP-seq signal on the non-template strand after triptolide treatment ( Figure 6—figure supplement 1A and B ) . Moreover , the high bisDRIP-seq scores in this region are not observed after RNase H treatment ( Figure 6—figure supplement 2A and B ) . This suggests that relatively long R-loops form in MALAT1 and NEAT1 and that these R-loops are bounded to the 5' end of each gene . Although our mapping reveals the location of the R-loops in MALAT1 and NEAT1 at near-nucleotide resolution , analysis of previous DRIP-seq datasets ( Sanz et al . , 2016 ) reveal signals in the same overall regions ( Figure 6—figure supplement 1A and B ) . Interestingly , there appears to also be periodicity in the bisDRIP-seq scores on the non-template strand of NEAT1 with peaks and valleys every 300 base pairs ( Figure 6B ) . This phenomenon is also detectable , but less prominent in MALAT1 ( Figure 6A ) . These valleys may indicate the existence of smaller R-loops in some MALAT1 or NEAT1 genes in some cells . This idea is supported by examining individual reads within the R-loop forming region in MALAT1 . In most cases , we observed that cytosines were almost completely converted in individual reads ( Figure 6—figure supplement 3A ) . However , we also observed a subset of reads with long stretches of cytosine conversions on one end and long stretches of unconverted cytosines on the other end ( Figure 6—figure supplement 3B ) . The region where the conversions stop occurring in individual reads might reflect an internal border of a R-loop . It should be noted that we cannot exclude the possibility that this may reflect the location of a structured region in the single-stranded DNA that prevents bisulfite reactivity . Nevertheless , the presence of peaks and valleys within the R-loop-forming region of NEAT1 and MALAT1 raises the possibility of heterogeneity in the size and location of the individual R-loops within the larger R-loop forming region identified by bisDRIP-seq . Previous efforts to map R-loops have primarily relied on immunoprecipitation of RNA-DNA hybrids ( Chédin , 2016 ) . Traditionally , DNA fragments containing an R-loop are recovered and sequenced . The sequenced fragments contain both the DNA involved in the R-loop and regions of DNA that are not in the R-loop . Newer approaches , like DRIPc-seq ( Chen et al . , 2015; Sanz et al . , 2016 ) , can provide higher resolution by sequencing the RNA component of the R-loop . We therefore next wanted to determine if the specific R-loop boundaries detected by bisDRIP-seq could also be detected in the human DRIPc-seq datasets . We first compared the location of R-loop signal in metaplots of DRIPc-seq and bisDRIP-seq signal in active promoters regions . As we demonstrated earlier , the R-loop signal in bisDRIP-seq is bounded between the transcription start site and the first exon-intron junction ( Figure 7A ) . Thus , we observe a tight peak of R-loop signal within a few hundred base pairs of the transcription start site in active promoter regions . On the other hand , DRIPc-seq shows a marked enrichment of reads following the transcription start site; however , there is no clear boundary of sense-strand DRIPc-seq reads peaks within 5 kb of the transcription start site ( Figure 7A ) . This highlights the improvement in resolution obtained by bisDRIP-seq . We next compared the R-loop maps generated using DRIPc-seq or bisDRIP-seq at individual gene loci . In some genes , such as MALAT1 and NEAT1 , we observed tight concordance between the R-loop maps generated from DRIPc-seq and bisDRIP-seq ( Figure 7B and C ) . However , the boundaries demarcated by bisDRIP-seq appear more clear than the boundaries demarcated by DRIPc-seq . On the other hand , there were few DRIPc-seq reads mapped to genes such as RPPH1 and HIST1H2BG , which displayed strong R-loop signal in bisDRIP-seq ( Figure 7D and E ) . This suggests the possibility that bisDRIP-seq might identify a set of R-loops that are not observable by DRIP-seq . We next asked if the size or location of the R-loop might change ex vivo . bisDRIP-seq labels R-loops upon cell lysis , while other mapping methods recover the R-loop hours or days after cell lysis . R-loops could conceivably , expand , contract , or disappear during this time period . This seemed plausible , since the structure of some R-loops has previously been observed to change in solution ( Kaback et al . , 1979; Landgraf et al . , 1996 ) . To test this possibility , we performed a ‘delayed-bisulfite’ bisDRIP-seq experiment . In this experiment , we delayed the bisulfite step of our protocol so that samples were only treated with bisulfite 16 hr after cell lysis . In some genes , such as MALAT1 and NEAT1 , the bisDRIP-seq scores were similar when using the delayed-bisulfite protocol . In other cases , the delayed-bisulfite treatment was associated with a marked depletion of the R-loop signal in RPPH1 and HIST1H2BG ( Figure 7B–E ) . This suggests the possibility that some R-loops may not be stable for prolonged periods of time ex vivo and therefore may not be observed in approaches that do not label or recover R-loops during lysis .
The precise location of R-loops in promoter regions has been obscure due to the low resolution of conventional R-loop mapping approaches ( Chen et al . , 2015; Ginno et al . , 2012 ) . These approaches rely on the S9 . 6 antibody to recover and sequence genomic fragments containing R-loops . Here we describe bisDRIP-seq , a near-nucleotide resolution method for mapping R-loops . In this approach , single-stranded DNA regions are identified based on their reactivity with bisulfite . These regions were then mapped throughout the genome based on the location of bisulfite-induced cytosine-to-uracil conversions . In some regions , bisulfite-induced cytosine conversions are enriched on one strand of DNA , show a requirement for RNA transcription and are removed following RNase H treatment , supporting the idea that these regions contain R-loops . R-loops were previously thought to expand thousands of base pairs into gene bodies without any clear boundary to their expansion or formation ( Chédin , 2016 ) . Here , we discover boundaries to R-loop formation at the transcription start site and the first exon-intron junction . The discovery of these boundaries suggest that the maximum length of most promoter-associated R-loops is predetermined by the exon-intron structure of genes . Identification of R-loops by bisDRIP-seq depends primarily on bisulfite labeling of the single-stranded component of R-loops . In addition to bisulfite labeling , bisDRIP-seq takes advantage of the S9 . 6 antibody to enrich for R-loop containing fragments of DNA , thus leading to enhanced read coverage in R-loop-containing regions . However , S9 . 6 is known to react to other nucleic acid structures , including structured RNA ( Phillips et al . , 2013 ) , and its full specificity relative to genomic DNA has not been examined . In our analysis , non-R-loop promoter single-stranded structures were immunoprecipitated by S9 . 6 antibody . This may reflect the poor specificity of the S9 . 6 antibody . Therefore , simply sequencing DNA recovered by S9 . 6 may not provide sufficient specificity for R-loop mapping . bisDRIP-seq overcomes problems with the S9 . 6 antibody by adding multiple criteria to selectively identify which recovered fragments contain R-loops . These criteria included preferential labeling of a single strand of DNA and the requirement that the labeling be transcription-dependent . S9 . 6 antibody enrichment in protocols like DRIP-seq and bisDRIP-seq have been shown to be biased towards promoter regions ( Halász et al . , 2017 ) . The requirement for preferential bisulfite labeling of one strand of DNA should largely mitigate this bias . Still , it is worth nothing that these criteria may have resulted in us excluding specific genomic regions that form RNA-DNA hybrids on both DNA strands or that contain extremely stable R-loops . Nevertheless , these criteria allow us to exclude non-R-loop single-stranded structures that might otherwise be mistaken for R-loops . The location of the 5' boundary of R-loops , identified here to be located at the transcription start site , suggests that R-loops are primarily formed using the canonical gene transcript . Conceivably , other transcripts could form R-loops in promoter regions , such as promoter upstream transcripts ( Preker et al . , 2011 ) or extracoding RNAs ( Savell et al . , 2016 ) . Our data suggests that other transcripts initiating upstream or downstream of the canonical transcription start site are less likely to be the major source of RNA in promoter-associated R-loops . Why would the exon-intron junction serve as the 3’ R-loop boundary ? The simplest explanation is that RNA splicing limits the length of the hybrid that can form between the spliced RNA and the template strand of DNA . If the RNA that forms the R-loop is spliced , then the RNA would not be able to hybridize to intron-encoding DNA ( Figure 8A ) . Alternatively , R-loop expansion might be blocked by proteins that are bound to the 5' splice site in the RNA ( Figure 8B ) . Both of these mechanisms could potentially explain the exon-intron junction R-loop boundary . Previous studies found that the relative location of the first exon-intron junction impacts the rate of transcription ( Bieberstein et al . , 2012; Brinster et al . , 1988; Fong and Zhou , 2001 ) . Smaller first exons are associated with higher rates of transcription and enhanced recruitment of transcription initiation factors ( Bieberstein et al . , 2012 ) . Conceivably , if smaller first exons are associated with less efficient or less stable R-loops , this may account for the higher transcription associated with these genes . Although intronless genes lack exon-intron junctions , we identified R-loops and R-loop boundaries in these genes . Intronless genes were among the genes with the highest R-loop signal immediately downstream of the transcription start site . In particular , two classes of intronless genes were identified: intronless lncRNAs , including MALAT1 and NEAT1 , and replication-dependent histones . Histone genes are notable for their unusual replication-dependent transcription regulation and for their non-canonical end termination ( Marzluff et al . , 2008 ) . During DNA synthesis , histone genes are transcribed and processed more efficiently than at other parts of the cell-cycle ( Hereford et al . , 1981 ) . Additionally , the histone genes are the only mRNAs that lack poly-A tails . Like MALAT1 and NEAT1 , the histone genes are terminated with a special stem-loop structure ( Marzluff et al . , 2008 ) . This special processing has been previously linked to transcription and the absence of histone gene splicing . It is intriguing to speculate that either of these processes may relate to the robust R-loop structures that are seen at the 5’ end of these genes . MALAT1 and NEAT1 , which are both intronless genes , also had two of the strongest R-loop signals . These genes clearly show that promoter-associated R-loops can be bounded to specific regions of genes in the absence of an exon-intron junction . RNA-binding proteins might determine R-loop boundaries in these types of intronless genes . Indeed , we observed prominent binding of SRSF1 at a position that corresponds to the 3’ boundary of the R-loop in MALAT1 in publicly-available datasets ( Figure 6—figure supplement 4 ) . Thus , RNA-protein complexes may also limit the expansion of R-loops , especially in intronless genes . Interestingly , all of the intronless genes that we studied are components of nuclear bodies . The histone RNAs , MALAT1 and NEAT1 associate with histone locus bodies , nuclear speckles and paraspeckles , respectively ( Clemson et al . , 2009; Hutchinson et al . , 2007; Liu et al . , 2006 ) . The strong R-loop signal in so many genes with roles in well-established nuclear bodies raises the possibility that R-loops may play a role in the formation or regulation of these bodies . In the case of at least the histone genes and NEAT1 , formation of the nuclear bodies is known to occur near the DNA encoding these RNA transcripts ( Clemson et al . , 2009; Mao et al . , 2011 ) . This suggests the possibility that R-loop formation may be involved in tethering RNA to chromatin to facilitate ribonucleoprotein assembly . Future research will be needed to test this and other possibilities for why these nuclear body-associated genes are prone to R-loop formation . Although bisDRIP-seq was used here to study steady-state promoter-associated R-loops , the method can be used for studying other questions regarding R-loops . For example , this method could be used to study the dynamic changes in R-loop size and location in response to various signals like estrogen . Moreover , as transcription termination sites become better annotated , high-resolution R-loop could guide the study of R-loops involved in transcription termination .
MCF-7 cells were used for all experiments . Cell lines were originally obtained from ATCC . Cell identification was performed by ATCC , which identifies cell lines using STR profiling , and cell identity was regularly checked by visual inspection of morphologies . Independently prepared vials of cell lines were tested for mycoplasma by Hoechst staining; these cell lines were tested after the conduction of the experiments described here . MCF-7 cells ( 10 million per 15 cm tissue culture dish ) were cultured for three days at 37°C in 50 ml of Gibco's phenol red-free Dulbecco's Modified Eagle's Medium ( Thermo Fisher Scientific , Waltham , Massachusetts ) containing 5% charcoal-stripped fetal bovine serum ( Thermo Fisher Scientific , Waltham , Massachusetts ) . Media was replaced after two days . In these experiments , we included MCF-7 cells treated for different amounts of time with 100 µM estrogen ( Sigma-Aldrich , Saint Louis , Missouri ) prior to lysis to ensure transcriptional activation at a broad set of genes ( Hah et al . , 2011 ) . The use of estrogen in some samples ensures that we can maximally capture R-loops that occur in either basal and stimulated conditions . Each of the thirteen control-treated bisDRIP-seq experiments and each of the two triptolide-treated bisDRIP-seq experiments were performed using a separate dish of MCF-7 cells . Thus , each bisDRIP-seq replicate used a different sample of MCF-7 cells that had , at a minimum , been cultured on a separate dish for three days . In the bisDRIP-seq protocol , R-loops are treated with bisulfite during cell lysis under non-denaturing conditions . The basic concept underlying this approach is that bisulfite can interact with single-stranded DNA , but cannot interact with DNA in a double helix ( Yu et al . , 2003 ) . This contrasts with 5-methylcytosine mapping , which is performed under denaturing conditions that cause all double-stranded DNA to be single stranded and susceptible to bisulfite treatment . It is therefore worth noting that bisDRIP-seq could be affected by 5-methylcytosine . If 5-methylcytosine is present in the single-stranded DNA of the R-loop , there will be minimal conversion at those sites . This should not substantially impact R-loop mapping , since 5-methylcytosines are found only in a CpG sequence context and most R-loops will have some cytosines that are not followed by guanine . Nevertheless , this issue could be considered during data analysis . Prior to performing bisDRIP-seq , lysis buffer was prepared containing 55 . 5 mg of ammonium sulfite monohydrate ( Sigma-Aldrich , Saint Louis , Missouri ) , 760 µl of 45% ammonium bisulfite ( Pfaltz and Bauer Inc , Waterbury , Connecticut ) , 85 . 5 µl of 20 mM hydroquinone ( Sigma-Aldrich , Saint Louis , Missouri ) and 62 . 7 µl of 0 . 5 M ethylenediaminetetraacetic acid ( EDTA ) in a final volume of 2 . 85 ml . Lysis buffer was prepared using degassed water . Additionally , ammonium bisulfite solution was prepared containing 0 . 67 g ammonium sulfite monohydrate , 2 . 08 g of sodium bisulfite ( Sigma-Aldrich , Saint Louis , Missouri ) , 5 ml of 45% ammonium bisulfite in a final volume of 6 ml . This ammonium bisulfite solution was prepared using degassed water , following the methodology described by Hayatsu et al . , 2004 . Immediately prior to cell lysis , media was removed from cells . Cells were washed three times in ice-cold PBS . Cells were then scraped off of plates into ice-cold PBS . These cells were then transferred to a 15 ml conical tube . The conical tube containing cells was centrifuged for 3 min at 300 g . PBS was then removed from cells . Next , cells were lysed in the presence of bisulfite . Bisulfite was included in the lysis buffer to achieve rapid single-stranded DNA modification that would minimize R-loop expansion or contraction . First , lysis buffer ( 1 . 5 ml ) was added to the cells . Cells in lysis buffer ( 475 µl ) were added to each of two microcentrifuge tubes , each of which contained 25 µl of lithium dodecyl sulfate . Samples were then incubated in a shaking incubator for 30 min at 37°C at 1100 rpm , while being protected from light . Next , the concentration of bisulfite in samples was increased in order to increase the rate of the irreversible second hydrolytic deamination step in the bisulfite conversion process ( Hayatsu et al . , 2004 ) . The concentration of bisulfite was increased by transferring samples to a microcentrifuge tube containing 1 ml of ammonium bisulfite solution and 26 . 8 µl of 20 mM hydroquinone . Samples were then incubated in a shaking incubator for 2 hr at 37°C at 1100 rpm , while being protected from light . Next , bisulfite was removed from the samples through dialysis . First , samples were added to 2 . 5 ml of dialysis buffer ( 50 mM Tris pH 8 , 10 mM EDTA ) in a 15 ml conical centrifuge tube . Samples were then added to an Amicon Ultra-4 Centrifugal Filter Unit with a 30 , 000 nominal molecular weight limit ( EMD Millipore , Darmstadt , Germany ) . Next , the samples were centrifuged following the instructions of the centrifugal filter unit manufacturer . After centrifugation , fresh dialysis buffer was added to the centrifugal filter unit and centrifugation was repeated . This dialysis process was repeated until the sample was nominally dialyzed at least 200 fold . Proteinase K buffer ( 50 mM NaCl , 50 mM Tris pH 8 , 10 mM EDTA ) was then added to the sample and centrifugation was repeated . This dialysis process was repeated until the sample was nominally dialyzed at least 5000 fold in Proteinase K buffer . At this point , the samples were resuspended in 1 ml of Proteinase K buffer and transferred to a microcentrifuge tube . Next , we used Proteinase K ( Life Technologies , Carlsbad , California ) to degrade the proteins in our samples . In order to treat our samples with Proteinase K , 50 µl of 20% SDS and 200 µl of 20 µg/µl Proteinase K was added to each sample . Samples were then incubated overnight at 37°C while being rotated end over end . Nucleic acids were purified from any remaining proteins present in the sample using phenol-chloroform . Samples were split into two microcentrifuge tubes . An equal volume of Phenol-chloroform ( Life Technologies , Carlsbad , California ) was added to each sample . Samples were mixed and then centrifuged for five minutes at 18400 g . The aqueous phase of each sample was transferred to a new microcentrifuge tube . Each sample was then combined with an equal volume of chloroform . Samples were mixed and centrifuged for two minutes at 18400 g . The aqueous phase of each sample was transferred to a new microcentrifuge tube . Samples were then ethanol precipitated for 2 hr at 4°C . The precipitated DNA was then transferred and combined in a new tube by swirling the DNA around a pipette tip . Finally , the DNA was washed with 70% ethanol . Next , the DNA was fragmented using restriction enzymes . The precipitated DNA was resuspended in 850 µl of NEB buffer 3 . 1 ( 100 mM NaCl , 50 mM Tris-HCl pH 8 . 0 , 10 mM MgCl2 , 100 µg/ml bovine serum albumin ) . The samples were then digested overnight at 37°C by a cocktail of restriction enzymes , which included: 200 U/ml HindIII , 200 U/ml EcoRI , 100 U/ml BsrGI , 200 U/ml XbaI and 50 U/ml SspI ( all from New England Biolabs , Ipswich , Massachusetts ) in a total volume of 900 µl . Next , the quality of the DNA digest was assessed and the concentration of DNA was measured in preparation for the RNA-DNA hybrid immunoprecipitation . DNA digest efficiency was measured by running the DNA on a 0 . 8% agarose gel stained with ethidium bromide . This was performed for every bisDRIP-seq experiment and samples were only used for immunoprecipitation if the digested DNA was mostly between 1 kb and 10 kb long . The concentration of DNA was measured using a Quant-iT dsDNA Assay Kit ( Thermo Fisher Scientific , Waltham , Massachusetts ) and a SpectraMax M2 Microplate Reader ( Molecular Devices , Sunnyvale , California ) following the instructions provided by the Quant-iT dsDNA Assay kit manufacturers . Next , RNA-DNA hybrids were enriched using an immunoprecipitation protocol modeled after the DRIP-seq methodology described by Ginno et al . ( 2012 ) . Digested DNA ( 20 µg ) was added to 20 µg of S9 . 6 antibody ( Kerafast , Boston , Massachusetts ) in 1 ml of immunoprecipitation buffer ( 10 mM sodium phosphate pH 7 . 0 , 140 mM NaCl , 0 . 05% Triton X-100 ) . The sample was then incubated overnight at 4°C while being mixed end over end . Next , the sample was added to 150 µl of Dynabeads Protein G ( Life Technologies , Carlsbad , California ) washed in immunoprecipitation buffer . Samples were incubated for 2 hr at 4°C while being mixed end over end . Next , the Dynabeads Protein G and immunoprecipitated DNA were washed . First , the supernatant was removed after applying the sample to a magnet for 3 min . Next , beads were washed three times in 750 µl of immunoprecipitation buffer . Each wash lasted 10 min and during the wash the samples were mixed end over end at 4°C . On each occasion , the supernatant was removed after applying the sample to a magnet for 3 min . Next , the bisulfite reaction was completed . First , we added 150 µl of 0 . 3 N NaOH to the samples . The samples were then incubated for 20 min at 37°C . This step should complete the bisulfite reaction . We then neutralized the NaOH by adding 150 µl of 0 . 3 N HCl and 17 . 5 µl of 1 M Tris pH 8 . 0 . Next , we eluted the immunoprecipitated DNA . First , 0 . 5 µl of 100 U/µl RNase I was added to each sample . Samples were then incubated for 20 min at 50°C . RNAs still bound to DNA were degraded by adding 1 µl of 0 . 5 M MgCl2 and 3 µl of 5 U/µl RNase H ( New England Biolabs , Ipswich , Massachusetts ) to each sample . The samples were then incubated for 20 min at 37°C . To ensure that all DNA fragments were eluted , we added 2 µl of 20 µg/µl of Proteinase K to each sample . The samples were then incubated for 1 hr at 50°C . Finally , a magnet was added to each sample for 3 min and the supernatant containing the eluted DNA was transferred to a new microcentrifuge tube . Next , DNA was extracted from the supernatant using phenol/chloroform . Prior to adding phenol/chloroform , 150 µl of water was added to each sample . Samples were then combined with 500 µl of phenol/chloroform . Samples were mixed thoroughly and then centrifuged at 18400 g for 5 min . 400 µl of the aqueous phase was transferred to a new microcentrifuge tube and then 400 µl of chloroform was added to this aqueous phase . These samples were mixed and then centrifuged at 18400 g for 2 min . The supernatant was transferred to a new microcentrifuge tube . Finally , the DNA in the sample was ethanol precipitated in the presence of 20 µg of glycogen for 48 hr . In order to treat cells with triptolide ( R and D Systems , Minneapolis , Minnesota ) , 20 mM triptolide was prepared in dimethyl sulfoxide . Triptolide was added to the media covering the MCF-7 cells at a concentration of 1 µM . The plates containing the cells were gently swirled to evenly distribute the triptolide . The plates were then incubated for 2 hr at 37°C . Following this triptolide treatment , bisDRIP-seq was performed as described above . Each of the two triptolide-treatment biological replicates was performed on a separate plate of cells . Two input bisDRIP-seq samples were prepared using the bisDRIP-seq protocol until the S9 . 6 antibody immunoprecipitation step . Instead of adding the digested DNA to S9 . 6 antibody , 15 µl of 3 N NaOH was added to 0 . 2 µg of digested DNA in 135 µl of water . The samples were then incubated for 20 min at 37°C . This step should complete the bisulfite reaction . We then neutralized the NaOH by adding 150 µl of 0 . 3 N HCl and 17 . 5 µl of 1 M Tris pH 8 . 0 . Next , we repeated the elution treatment that was applied to the bisDRIP-seq samples . First , 0 . 5 µl of 100 U/µl RNase I was added to each sample . Samples were then incubated for 20 min at 50°C . RNAs still bound to DNA were degraded by adding 1 µl of 0 . 5 M MgCl2 and 3 µl of 5 U/µl RNase H ( New England Biolabs , Ipswich , Massachusetts ) to each sample . The samples were then incubated for 20 min at 37°C . Next , we added 2 µl of 20 µg/µl of Proteinase K to each sample . The samples were then incubated for 1 hr at 50°C . Finally , the DNA in the sample was extracted using phenol-chloroform . After this point , the bisDRIP-seq protocol was followed as described above . Both an RNase H-treated sample and matched control sample were prepared using a single plate of MCF-7 cells . These cells were treated as described in the initial steps of the bisDRIP-seq protocol described above . These initial steps included all steps until and including fragment ion of the DNA with restriction enzymes and confirmation of the digest with an agarose gel , After those initial steps and immediately prior to immunoprecipitating RNA-DNA hybrids with S9 . 6 antibody , our sample was split into two samples . RNase H was added to one sample to a final concentration of 0 . 05 U/µl . We now call this sample the ‘RNase H-treated sample’ . Dithiothreitol ( at the same concentration as in the RNase H enzyme solution ) was added to the other sample to a final concentration of 10 µM . RNase I was also added to both samples to a final concentration of 0 . 1 U/µl in order to prevent non-specific interactions between RNAs in solution and the S9 . 6 antibody ( Phillips et al . , 2013 ) . Samples were then incubated for 2 hr at 37°C . Following this treatment , both samples were incubated with S9 . 6 antibody and Dynabeads Protein G as described above in the bisDRIP-seq protocol . Next , we adjusted the immunoprecipitation wash steps to allow for further RNase H treatment . First , three wash steps were performed in immunoprecipitation buffer as described in the standard bisDRIP-seq protocol . Next , we added two additional wash steps using 200 µl of immunoprecipitation buffer supplemented with 3 mM MgCl2 . In each of these final two wash steps , RNase H was added to a final concentration of 0 . 5 U/µl to the ‘RNase H’ treated sample , while dithiothreitol was added to a final concentration of 50 µM to the matched control sample . Each wash lasted 20 min and during each wash the samples were mixed end over end at 37°C . On each occasion , the supernatant was removed after applying the sample to a magnet for 3 min . After these wash steps , the samples were treated with NaOH and eluted as described in the bisDRIP-seq protocol above . The delayed bisDRIP-seq sample was prepared largely following the bisDRIP-seq protocol , with a few modification . After cells were washed in PBS as described in the bisDRIP-seq protocol , 5 ml of lysis buffer was added . This lysis buffer did not include bisulfite and was instead composed of 1% SDS , 50 mM Tris pH 8 , and 10 mM EDTA . Next , 1 ml of 20 µg/µl Proteinase K was added to the sample . Samples were then incubated overnight at 37°C while being rotated end over end . Next , nucleic acids were extracted using phenol-chloroform and ethanol precipitated as described in the bisDRIP-seq protocol . At this point , we treated the nucleic acids with bisulfite . First , the nucleic acids were dissolved in 3 . 3 mM EDTA . Next , 900 µl of ammonium bisulfite solution and 26 . 8 µl hydroquinone were added to the sample . The sample was then incubated in a shaking incubator for 2 . 5 hr at 37°C at 1100 rpm , while being protected from light . Next , the bisulfite was removed from the samples through dialysis . First , samples were added to 2 . 5 ml of buffer 3 . 1 lacking BSA ( 100 mM NaCl , 50 mM Tris-HCl pH 8 . 0 , 10 mM MgCl2 ) in a 15 ml conical centrifuge tube . Samples were then added to an Amicon Ultra-4 Centrifugal Filter Unit with a 30 , 000 nominal molecular weight limit ( EMD Millipore , Darmstadt , Germany ) . Next , the samples were centrifuged following the instructions of the centrifugal filter unit manufacturer . After centrifugation , fresh buffer 3 . 1 lacking BSA was added to the centrifugal filter unit and centrifugation was repeated . This dialysis process was repeated until the sample was nominally dialyzed at least 1000000 fold . At this point , the samples were resuspended in 850 µl of buffer 3 . 1 and the remaining steps of the bisDRIP-seq protocol were followed as described above . Prior to constructing DNA sequencing libraries , the eluted DNA was fragmented into approximately 300 bp fragments . DNA from bisDRIP-seq reactions was re-suspended in 150 µl of 1XTE buffer ( 10 mM Tris-HCl pH 8 . 0 , 1 mM EDTA ) . Samples were then fragmented using an S2-series Covaris ultrasonicator ( Woburn , Massachusetts ) . Sonication was performed at 4°C using the following conditions: intensity 5 , 10% Duty Intensity Factor , 200 cycles per burst and 140 s total treatment time . We next used a Pico-Methyl Seq Library Prep Kit ( Zymo Research , Irvine , California ) to create the DNA sequencing library . This kit is primarily designed for mapping 5-methylcytosine in small quantities of genomic DNA . We omitted the bisulfite-treatment steps . Instead , we specifically followed the library preparation steps as instructed by the manufacturer . These steps amplified our DNA and used random priming to add indexes to our DNA library that are compatible with Illumina's TruSeq technology . In preparation for sequencing , the quality of each DNA library was assessed using a 2100 Bioanalyzer ( Agilent Technologies , Santa Clara , California ) by the Weill Cornell Epigenomics Core ( New York City , New York ) . The concentration of DNA was measured using a Quant-iT dsDNA Assay Kit and a SpectraMax M2 Microplate Reader following the instructions provided in the Quant-iT dsDNA Assay kit . DNA libraries were sequenced by the Weill Cornell Epigenomics Core using a HiSeq 2500 System ( Illumina , San Diego , California ) . Either five or six samples were loaded per lane . DNA libraries were sequenced following the manufacturer's instructions for single-index 100 bp paired-end read clustering . Sequencing data was first processed using CASAVA 1 . 8 . 2 ( Illumina , San Diego , California ) to obtain the nucleotide sequence of reads in FASTQ format . We next aligned reads to the GRCh38 reference genome . Reads were aligned to the genome using the Bismark software library version 0 . 14 . 3 Bismark tool ( Krueger and Andrews , 2011 ) . The Bismark software library is specifically designed to align bisulfite-modified reads . In this analysis , the Bismark library used the Bowtie2 alignment tool version 2 . 2 . 5 ( Langmead and Salzberg , 2012 ) to perform the actual alignments . We also selected the Bismark option for post-bisulfite adapter tagging . Read alignment was done in three phases . If we were unable to map a read in one phase , then we attempted to map the read in the following phase . In the first alignment phase , full paired-end sequencing reads were aligned to the genome . First , reads were trimmed to remove both low confidence sequences and the very ends of reads . This was done using Flexbar version 2 . 5 ( Dodt et al . , 2012 ) with the following settings: quality format i1 . 8 , phred minimum of 30 , 10 nt removed from 5' end of reads . Next , the Bismark tool was used to align reads to the genome . The settings employed for the alignment did not allow for mismatched nucleotides and restricted the gap between paired-end reads to no more than 1000 bp . Next , the Bismark alignment tool was repeated on all reads that failed to map in the first alignment phase . In this second phase , the two ends of paired-end reads were treated as single-end reads . The settings employed for alignment did not allow mismatched nucleotides . In the final alignment phase , we attempted to align all reads that were not aligned to the genome in the first two phases after additional preprocessing . In this phase , we continued to treat both ends of paired-end reads as single-end reads . However , we began this phase by using Flexbar version 2 . 5 to remove all nucleotides 55 nt or further 3' of the start of the read . Reads were then aligned with the Bismark tool using settings that did not allow for mismatched nucleotides . Having aligned reads to the genome , we next removed putative read duplicates . To remove read duplicates , we ran the Bismark deduplicate_bismark tool . Finally , we determined if cytosines within aligned reads had been converted by bisulfite . To determine whether individual cytosines had been converted to uracils , we used the Bismark bismark_methylation_extractor tool . In the case of paired-end alignments , we used the –no_overlap setting . With single-end alignments , we used the --ignore_3prime 10 setting . This setting removed the final ten nucleotides of each alignment . These nucleotides were removed from aligned reads because we observed bias in the fraction of cytosines converted in this region of reads . Next , we developed a method to score reads based on the likelihood that a given read was single-stranded in our samples . This measure was intended to filter out conversions that occur due to spontaneous breathing of DNA . Additionally , we wanted to ensure that single-stranded reads containing large numbers of cytosines were not given greater value than single-stranded reads containing small numbers of cytosines . This method was applied to on each sample separately . For the purposes of calculating bisDRIP-seq scores , we define the ‘original number of cytosines’ as the sum of converted cytosines and unconverted cytosines . The goal of this method was to estimate the fraction of reads that had a set number of conversions due to chance alone . The inverse fraction was then defined as the bisDRIP-seq score for all reads with that number of conversions . Therefore , all of the reads from the same sample , with the same original number of cytosines , and with the same number of converted cytosines were given the same bisDRIP-seq score . First , we made a rough estimate of the background conversion rate for each sample . Initially , we calculated the total fraction of cytosines that were converted in our aligned reads . Next , we repeated this calculation after removing reads that had greater than 2 . 5 fold more conversions than the average across all reads . This value was used as an estimate of the background conversion rate . Next , we calculated the probability of observing a specified number of converted cytosines in a given read by chance . This probability was calculated using the binomial distribution . In this binomial distribution calculation , the size of the sample was the original number of cytosines in the read and the probability of a conversion was the estimated background conversion rate . Next , we calculated the number of reads expected to have a given original number of cytosines and a given number of converted cytosines . First , we calculated the total number of aligned reads with a given original number of cytosines . Next , we multiplied that number with the probability of observing the specified number of converted cytosines in a read with that original number of cytosines by chance . Next , a bisDRIP-seq score was calculated for each read . First , reads were binned together based on the original number of cytosines in the read and the number of converted cytosines in the read . Next , bins were grouped together if they had the same original number of cytosines . If a group of reads with a given original number of cytosines per read consisted of more than a thousand reads in total and the expected number of reads for each bin was above five , then we calculated the bisDRIP-seq for each bin as:ifO > E , bisDRIPseq score=1−EO E = expected number of readsifO ⩽ E , bisDRIPseq score=0 O = observed number of reads This value was calculated first for reads with one original number of cytosines and then was calculated for reads with progressively higher numbers of original cytosines . If the expected number of reads for a bin was below five , then the bisDRIP-seq score was calculated for all reads with x number of converted cytosines using the equation:if∑c=xnEC>5:if∑c=xnOc>∑c=xnEc , bisDRIPseq scorex=1−∑c=xnEc∑c=xnOcif∑c=xnOc⩽∑c=xnEc , bisDRIPseq scorex=0if∑c=xnEc⩽5: bisDRIPseq scorex= bisDRIPseq scorex−1c=number of cytosine conversionsn=original number of cytosinesEc=expected number of reads with c cyosine conversionsOc=observed number of reads with c cyosine conversions In this equation , the original number of cytosines ( n ) is held constant . If there were fewer than a thousand reads in a group , reads were scored differently . bisDRIP-seq scores with a specific number of conversions were given the same score as reads with the same number of conversions , but one fewer original number of cytosines . Given the total number of aligned reads per sample , the number of reads assigned scores on this basis was relatively small . Source code for calculation of bisDRIP-seq scores from read sequence is in Source code 1 and processingbisDRIPseqreads . py , which has been deposited in https://github . com/champben2002/bisDRIPseq/ bisDRIP-seq scores were normalized to ensure that the sum of the bisDRIP-seq scores were the same across samples after normalization . This normalization procedure assumes that there are no global differences in the amount of single-stranded structure between samples . First , the sum of all read bisDRIP-seq scores was calculated for each sample . Next , the normalized bisDRIP-seq score for each read was calculated for each read in a given sample using the formula:normalized bisDRIPseq score for read Y=bisDRIPseq score for read Y×1000000∑all read bisDRIPseq scores This normalization procedure was repeated for each sample separately . All bisDRIP-seq scores in the manuscript used this standard normalization procedure , unless otherwise noted . Promoter-associated R-loops have previously been associated with promoter regions enriched in guanines on the non-template strand ( Ginno et al . , 2012 ) . We did not make conclusions related to this finding since bisDRIP-seq scores depend on the presence of cytosines in single-stranded regions of reads . However , we did consider whether our conclusions might be affected by the cytosine content of reads . To test for whether our conclusions were affected by the cytosine content of reads , we applied a ‘cytosine normalization’ procedure to the raw bisDRIP-seq scores . First , reads within a sample were grouped based on the original number of cytosines in each read . Next , we normalized the reads within each group so that the average bisDRIP-seq score for each group would be identical . This normalization was done by multiplying the bisDRIP-seq score of each read within a group by a group-specific variable . Once this cytosine normalization was complete , the standard normalization procedure was followed as described above . We repeated our analysis using these cytosine-content normalized scores ( Figure 5—figure supplement 3 ) . Our conclusions were not changed by this cytosine-content normalization . Source code for bisDRIP-seq normalization is included in processingbisDRIPseqreads . py , which has been deposited in Source code 1 and https://github . com/champben2002/bisDRIPseq/ All genomic maps were generated using the Integrative Genomics Viewer version 2 . 3 . 59 ( 86 ) with the GRCh38 human genome ( Robinson et al . , 2011; Thorvaldsdóttir et al . , 2013 ) . The automatically loaded refseq gene models ( O'Leary et al . , 2016 ) were included in each map , with the following exceptions: For NEAT1 , we used the gene model from GENCODE version 24 ( Harrow et al . , 2012 ) of the long isoform of NEAT1 , ENST00000501122 . 2 , since it appears more consistent with the RNA-seq data . For HIST1H2BK , we used the gene model from GENCODE version 24 ( Harrow et al . , 2012 ) of the intronless isoform of HIST1H2BK , ENST00000356950 . 1 . bisDRIP-seq scores were calculated for each nucleotide in the genome . The score of a given nucleotide in the genome was calculated as the sum of the bisDRIP-seq scores of reads that aligned to that nucleotide . Notably , this score was strand specific . The bisDRIP-seq score of reads was associated with a given nucleotide only if that nucleotide was on the same strand as the cytosines converted in the read . These bisDRIP-seq scores were then plotted for specific genomic loci . Local sequence composition was plotted for sites in the HIST1H1E and HIST1H2BG genes . At each site , the mean number of adenines , cytosines , guanines and thymines in the surrounding region were plotted . The surrounding region included the nucleotide at the site , the twenty-five nucleotides 5' of the site and twenty-five nucleotides 3' of the site on the template strand . bisDRIP-seq scores were calculated for genomic regions using the following procedure: First , we summed the bisDRIP-seq scores of reads that completely aligned to the specified region . Next , reads partially in the specified region were partially added to the specified region's score . The fraction of the read's score added to the score of the specified region's score was determined by the fraction of the read's length that aligned to the region . Thus the fraction of the read's length contained within the region was multiplied by the read's score . The product of this calculation was added to the specified region's score . This was repeated for each read that partially aligned in the specified region . This procedure provided the region's final bisDRIP-seq score . In some of our analysis , bisDRIP-seq scores were calculated for only one strand within a specified region . In this case , the bisDRIP-seq scores from individual reads was only added to the specified region's bisDRIP-seq score if the read had the correct strand orientation . A read had the correct strand orientation if any converted cytosines in the read mapped to the correct strand . Source code for calculating region scores is incorporated into regionbisDRIPseqscores . py , which has been deposited in Source code 2 and https://github . com/champben2002/bisDRIPseq/ . Monte Carlo simulations were used to determine the expected distribution of bisDRIP-seq scores if they distributed randomly across the genome . Two simulations were performed . First , a simulation was performed in which reads were mapped to random regions across the genome . This simulates the possibility that both the RNA-DNA hybrid immunoprecipitation and the bisulfite mapping in bisDRIP-seq are non-specific . First , we split the genome into 1 kb regions . Next , we determined if any bisDRIP-seq reads aligned to a given 1 kb region . If no reads aligned to a given region , then that region was removed from further simulation steps . Reads were then associated with a random 1 kb region that was in the same chromosome to which they originally mapped . The region that each read was associated with was determined using the R 'random' package . This was repeated for each of our thirteen samples . Second , a simulation was performed in which bisDRIP-seq scores were randomly shuffled between reads . This simulates the possibility that the bisDRIP-seq score associated with each read was stochastic . For each sample , we randomly shuffled all of the scores associated with reads aligned to a given chromosome . The random shuffle of read scores was performed using the R 'random' package . This simulation was repeated for each chromosome and for each of the thirteen bisDRIP-seq samples . Next , bisDRIP-seq scores were calculated for each 1 kb region using either real bisDRIP-seq scores or using the products of one of the two simulations . This analysis used the same 1 kb regions discussed in the first Monte Carlo simulation . Source code for Monte Carlo simulations are included in Monte_Carlo_random_assign_reads_to_regions . py and Monte_Carlo_for_shuffling_bisDRIPseq_scores . py , which have been deposited in Source code 3 and Source code 4 ( respectively ) , as well as https://github . com/champben2002/bisDRIPseq/ GENCODE's annotated list of transcription start sites ( Harrow et al . , 2012 ) was used to generate a ‘reference GENCODE transcription start site list . ’ This set of transcription start sites was then used to define promoter regions in the genome and as reference points for metaplots . First , the GENCODE comprehensive gene annotation release 25 was downloaded from GENCODE's website . Next , transcription start sites were removed from the transcription start site list if they were within a thousand base pairs of a higher priority transcription start site . The priority of transcription start sites was determined as follows: After removing lower priority transcription start sites , we had a final reference GENCODE transcription start site list . Promoter regions were then defined as the region within a thousand base pairs of each transcription start site in our reference GENCODE transcription start site list . See Source_data_file_1 . xls for all transcription start sites in this final list In later analysis , inactive promoter regions were removed . Unless noted otherwise , inactive promoter regions were defined as promoter regions with sense-strand promoter activity in the bottom eighty percentile between the transcription start site and + 1000 bp . In order to calculate the promoter activity of each promoter , we used publicly available GRO-seq data obtained from MCF-7 cells by Hah et al . , 2013 . In particular , data was combined from files GSM1067410 , GSM1067411 , GSM1067412 , GSM1067413 , GSM1067414 and GSM1067415 . The MCF-7 cells used in these samples were treated in a similar manner to the MCF-7 cells used in our bisDRIP-seq protocols . Unless noted otherwise , promoter activity was defined by the mean number of sense-strand GRO-seq reads measured in the region between the transcription start site and + 1000 bp . The method for calculating the number of reads in this region of each promoter was similar to method used to calculate bisDRIP-seq scores for regions . For each promoter region , we first summed the number of GRO-seq reads that completely aligned to the region between the transcription start site and + 1000 bp . Next , reads partially in the specified region were partially added to the specified region's score . The fraction of the read's length contained within the region between the transcription start site and + 1000 bp was added to the promoter region's score . This was repeated for each read that partially aligned to the promoter region . This procedure provided the promoter region's final promoter activity . A similar procedure was followed to determine the antisense promoter activity and the promoter activity between the transcription start site and + 250 bp . Exon-intron junctions were extracted from the GENCODE comprehensive gene annotation release 25 ( Harrow et al . , 2012 ) . Exon-intron and intron-exon junctions were separated based on whether they were the first exon-intron , first intron-exon or second exon-intron junction in a given annotated transcript . If two promoter regions shared the same exon-intron junction , then one was removed from the list . Next , all exon-intron junctions were removed that were associated with inactive promoter regions . Inactive promoter regions did not contain positive R-loop signal in Figure 3F , suggesting that removing these promoter regions would lead to a greater signal-to-noise ratio . The activity of promoter regions was defined here based on the number of sense-strand GRO-seq reads that aligned to the promoter region between the transcription start site and + 1000 bp . Promoters in the bottom eighty percentile in terms of promoter activity were considered inactive promoters and removed from exon-intron junction analysis . Exon-intron junctions associated with the intronless genes , MALAT1 and NEAT1 , were also removed . This procedure provided a final reference set of exon-intron junctions . See Figure 4—source data 1 for this final reference list . First-exon length was calculated for each first exon-intron junction in the final reference set of first exon-intron junctions . The first-exon length was calculated as the distance between the transcription start site and the first exon-intron junction . This list of first-exon lengths was used to calculate the median length of all first exons for active promoter regions . In order to compare some external dataset to our data , it was necessary to convert the coordinates of sites or regions to the GRCh38 genome from an earlier version of the reference human genome . In all cases , this was accomplished using the UCSC utility liftOver ( Speir et al . , 2016 ) . Cap analysis gene expression sequencing ( CAGE-seq ) data came from the publicly-available ENCODE MCF-7 dataset ENCFF207DXM ( ENCODE Project Consortium , 2012 ) . In order to compare bisDRIP-seq datasets with DRIPc-seq datasets , we used the DRIPc-seq data of Sanz et al . , 2016 . In particular , we compared our data against the data deposited in the Gene Expression Omnibus GSM1720613 and GSM1720614 wig files . In order to compare bisDRIP-seq datasets with DRIP-seq datasets , we used the DRIP-seq data of Sanz et al . , 2016 . In particular , we compared our data against the data deposited in the Gene Expression Omnibus GSM1720615 and GSM1720616 wig files . Total cell RNA-seq data displayed for individual genes came from ENCODE files ENCFF426WXY and ENCFF866OVQ ( ENCODE Project Consortium , 2012 ) . The RNA-seq experiments used to generate that data were rRNA depleted and used MCF-7 cells . Nucleus RNA-seq levels of active genes in Figure 5—figure supplement 1A are the mean transcripts per kilobase million ( TPM ) values as measured in ENCODE files ENCFF063BLU and ENCFF285GOS ( ENCODE Project Consortium , 2012 ) . These RNA-seq experiments sequenced mRNA depleted in rRNA from the nuclei of MCF-7 cells . RNA Polymerase II levels were calculated using the ENCODE files ENCFF496YAE and ENCFF881YOO ( ENCODE Project Consortium , 2012 ) . These datasets comes from chromatin immunoprecipitations of POLR2A in MCF-7 cells . The number of reads between the transcription start site and + 250 bp of each gene was calculated using the same methodology described above for GRO-seq reads . For DRIP-seq and bisDRIP-seq promoter region enrichment tests , only promoter regions in genes larger than 2 kb were considered . For each promoter region , a matched region was selected from the same gene . This matched region was centered on an exonic site more than 2 kb from the transcription start site . Promoter enrichment was then tested using the non-parametric Wilcoxon signed-rank test described below . Promoter region ranking was only performed on active promoter regions . Promoter activity was defined here based on the number of sense-strand GRO-seq reads that aligned to the promoter region between the transcription start site and +250 bp . Active promoter regions were promoter regions in the top twenty percentile of promoter regions in terms of promoter activity . This filter was intended to filter out inactive promoter regions , which appeared to only contribute noise to our analysis . Next , promoter regions were filtered on the basis of sense-strand bisDRIP-seq scores . The non-template strand bisDRIP-seq score had to be positive in the region between the transcription start site and + 250 bp in all control-treated biological replicates . Otherwise , the promoter region was removed from out list . The remaining promoter regions were scored based on the value: 2 x ( log2 ( control-treated non-template bisDRIP-seq score + 1 ) - log2 ( triptolide-treated non-template bisDRIP-seq score + 1 ) ) - log2 ( control-treated template bisDRIP-seq score + 1 ) ) Loess smoothed plots were generated using the r fANCOVA package's loess . as algorithm with parameters ‘criterion = aicc’ and ‘family = gaussian’ . Scatterplots were generated with the ggplot2 package geom_points ( ) algorithm , using the lm option to plot a linear regression line of the displayed points . Violin plots and jitter plots were created using the ggplot2 package . Jitter plots were created using geom_jitter ( ) with height = 0 , width = 0 . 725 . Violin plots were created using geom_violin ( ) with alpha = 0 . 5 . bisDRIP-seq score metaplots were created in relation to reference points . These reference points were either transcription start sites or exon-intron junctions , depending on the analysis . The final metaplot associated a bisDRIP-seq score to each ‘location’ relative to a given reference point . Location refers to the distance from the reference point and whether the site is upstream or downstream of the reference site . For a given read , we determined the location of each nucleotide in the read relative to the proximal reference point . Next , the location of each nucleotide was associated with the bisDRIP-seq score of the read . After repeating this process for all reads , all of the scores associated with a given location were summed unless otherwise noted . This sum was then used as the bisDRIP-seq score for that location in the metaplot . Finally , each location was plotted against the bisDRIP-seq score associated with that location . Source code for metaplot analysis is incorporated into bisDRIPseqmetaplotanalysis . py , which has been deposited in Source code 5 and https://github . com/champben2002/bisDRIPseq/ . The conversion asymmetry metaplot was generated in relation to transcription start sites . The total number of conversions was summed at each site relative to the transcription start site across all active promoter regions . This was repeated for each sample , strand and sample type . Next , we calculated the mean number of conversions across samples for each position relative to the transcription start site . Next , we subtracted the mean number of conversions for each position relative to the transcription start site in triptolide-treated samples from the number of conversions observed in control-treated samples . Finally , we plotted the log ratio of the tripolide-corrected number of conversions on the non-template strand to the tripolide-corrected number of conversions on the template strand . To create a metaplot of the percentage of cytosines converted on the template strand , we calculated the number of cytosines and converted cytosines that aligned to a given site relative to the transcription start site . Next , the number of converted cytosines was divided by the total number of original cytosines that aligned to that site . DRIPc-seq metaplots were generated using essentially the same process as was used for bisDRIP-seq score . Instead of using the scores of reads , we used the read-count associated with regions in the wig file . In all plots of R-loop signal , R-loop signal = ( control-treated sample , non-template bisDRIP-seq score - triptolide-treated sample , non-template bisDRIP-seq score ) - ( control-treated sample , template bisDRIP-seq score - triptolide-treated sample , template bisDRIP-seq score ) . In all plots where a log2 transformation was applied to a dataset , each value in the dataset was added to one prior to the log2 transformation . To sample the reads from the MALAT1 R-loop forming region , we selected all reads that: ( 1 ) aligned to the non-template strand of MALAT1 , ( 2 ) start and end between the transcription start site and + 1600 bp , and ( 3 ) are larger than 75 bp long . Next , we randomly selected twenty-five reads . Reads were then plotted using ggplot2's geom_point function . Since we are not confident that our data follows a Gaussian distribution , we typically used non-parametric tests of significance . In particular , we applied the Wilcoxon signed-rank test whether various datasets were significantly different . Wilcoxon signed-rank tests were performed using the two-sided R wilcox . test ( ) algorithm using default parameters with the exception of the ‘paired’ setting which was set to ‘FALSE’ . In the case of comparisons between metaplots , a conservative multiple-hypothesis correction was performed on the p-values derived from Wilcoxon signed-rank tests . This multiple-hypothesis correction involved multiplying the p-value of an individual nucleotide position by 2001 , since the metaplots involved examining 2001 nucleotide positions simultaneously . The significance of correlations between datasets was examined using the Spearman's rank-correlation test with asymptotic t approximation . Spearman's rank-correlation tests were used since we are not confident that our data follows a Gaussian distribution and therefore wished to apply a non-parametric test of correlation . Spearman's rank-correlation tests were performed using the R cor . test ( x , y , method = ‘spearman’ ) algorithm . For each sample , we deposited the sequence reads , the conversion frequency at each cytosine nucleotide and a bedGraph of bisDRIP-seq scores in GEO series GSE98886: https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE98886 Private token for reviewers: azyfacgczpertip | Genes contain coded instructions for making proteins . When the cell needs to use a gene , molecular machinery assembles near the start of the gene in regions called promoters . Part of this machinery then reads along the gene , making a copy of the code in the form of a DNA-like molecule called RNA . These RNAs typically contain regions called exons , which carry the instructions , interspersed with spacer regions called introns . As RNAs are made they are 'spliced' to chop out the introns , leaving behind the final instructions . Most DNA exists in a double helix shape with two connected DNA strands , but the regions near the start of genes often contain structures called R-loops . In these structures , one strand of the DNA partners up with a single strand of RNA , forcing the other strand to bulge out on its own . Their location at gene promoters indicates that R-loops could change the cell's use of genes by impacting the machines that assemble near the start of genes . However , R-loops are not well understood . A major barrier to understanding the role of R-loops is that we do not know exactly where they are with respect to the start of genes . Dumelie and Jaffrey now report a new method to map R-loops almost to the resolution of single letters of the DNA code – a method which they called bisDRIP-seq . The approach extends an existing technique called DRIP-seq , which uses antibodies to capture DNA sequences stuck to strands of RNA . It can find R-loops , but it cannot tell the difference between the loop itself and the DNA surrounding it . The new technique uses a chemical called bisulfite to alter the DNA letters . It only affects the loop of the R-loop because the RNA shields the other strand . Sequencing then pinpoints the modified letters , revealing the exact location of the loop . For human cells grown in the laboratory , the technique found that R-loops form between the start of the gene and its first intron . Some genes do not have any introns , and in these cases , the R-loops extended deep into the code . Most human genes have only a small amount of DNA between the start site and the first intron , which may act to limit the effect of R-loops in these genes . This new technique allows the high-resolution study of R-loops , and could help to reveal their role in regulating genes . Abnormal R-loops have already been linked to a small set of human diseases like fragile-X syndrome . As the tools to study R-loops improve , it is possible that scientists will make connections to other diseases . In time , improved understanding of these structures could lead to better diagnosis , and eventually treatment , for these conditions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"tools",
"and",
"resources",
"genetics",
"and",
"genomics"
] | 2017 | Defining the location of promoter-associated R-loops at near-nucleotide resolution using bisDRIP-seq |
In historical attempts to treat morning sickness , use of the drug thalidomide led to the birth of thousands of children with severe birth defects . Despite their teratogenicity , thalidomide and related IMiD drugs are now a mainstay of cancer treatment; however , the molecular basis underlying the pleiotropic biology and characteristic birth defects remains unknown . Here we show that IMiDs disrupt a broad transcriptional network through induced degradation of several C2H2 zinc finger transcription factors , including SALL4 , a member of the spalt-like family of developmental transcription factors . Strikingly , heterozygous loss of function mutations in SALL4 result in a human developmental condition that phenocopies thalidomide-induced birth defects such as absence of thumbs , phocomelia , defects in ear and eye development , and congenital heart disease . We find that thalidomide induces degradation of SALL4 exclusively in humans , primates , and rabbits , but not in rodents or fish , providing a mechanistic link for the species-specific pathogenesis of thalidomide syndrome .
Thalidomide was first marketed in the 1950s as a nonaddictive , nonbarbiturate sedative with anti-emetic properties , and was widely used to treat morning sickness in pregnant women . Soon after its inception , reports of severe birth defects appeared , but it was denied that these were linked to thalidomide . In 1961 , two independent reports confirmed that thalidomide was causative to this , the largest preventable medical disaster in modern history ( Lenz , 1962; McBride , 1961 ) . In addition to thousands of children born with severe birth defects , there were reports of increased miscarriage rates during this period ( Lenz , 1988 ) . Despite this tragedy , thalidomide , and its close derivatives , lenalidomide and pomalidomide , known as immunomodulatory drugs ( IMiDs ) , continue to be used to treat a variety of clinical conditions such as multiple myeloma ( MM ) and 5q-deletion associated myelodysplastic syndrome ( del ( 5q ) -MDS ) ( D'Amato et al . , 1994; Pan and Lentzsch , 2012 ) . Although a potentially transformative treatment for MM , the molecular mechanisms of thalidomide teratogenicity , and many of its biological activities remain elusive . It was only recently shown that thalidomide and analogs exert their therapeutic effect by binding to the Cullin RING E3 ubiquitin ligase CUL4-RBX1-DDB1-CRBN ( CRL4CRBN ) ( Chamberlain et al . , 2014; Fischer et al . , 2014; Ito et al . , 2010 ) and promoting ubiquitination and degradation of key efficacy targets ( neo-substrates ) , such as the zinc finger ( ZnF ) transcription factors IKAROS ( IKZF1 ) , AIOLOS ( IKZF3 ) , and ZFP91 ( An et al . , 2017; Fischer et al . , 2014; Gandhi et al . , 2014b; Krönke et al . , 2014; Lu et al . , 2014 ) . IMiDs can also promote degradation of targets that lack a zinc finger domain , including Casein Kinase 1 alpha ( CSNK1A1 ) ( Krönke et al . , 2015; Petzold et al . , 2016 ) and GSPT1 ( Matyskiela et al . , 2016 ) . CRL4CRBN has further been implicated in the IMiD-independent turnover of GLUL , BSG , and MEIS2 ( Eichner et al . , 2016; Krönke et al . , 2014; Nguyen et al . , 2016;Fischer et al . , 2014 ) , and regulation of AMPK ( Lee et al . , 2013 ) , processes potentially inhibited by IMiDs . Although no obvious sequence homology exists between the known IMiD-dependent CRL4CRBN substrates , all share the characteristic β-hairpin loop structure observed in X-ray crystal structures of IMiDs bound to CRBN and CSNK1A1 or GSPT1 ( Matyskiela et al . , 2016; Petzold et al . , 2016 ) , and a key glycine residue that engages the phthalimide moiety of IMiDs ( An et al . , 2017; Matyskiela et al . , 2016; Petzold et al . , 2016 ) . Despite progress in understanding the therapeutic mechanism of action of thalidomide , the cause of thalidomide syndrome has remained unknown since its description in 1961 . Over the last 60 years , multiple theories such as anti-angiogenic properties or the formation of reactive oxygen species ( ROS ) by thalidomide , or specific metabolites of thalidomide have been linked to thalidomide-induced defects . However , rarely do they explain the full spectrum of birth defects caused by all members of the IMiD family of drugs ( Vargesson , 2015 ) . Moreover , it has been shown that species such as mice , rats , and bush babies are resistant to thalidomide-induced teratogenicity ( Butler , 1977; Heger et al . , 1988; Ingalls et al . , 1964; Vickers , 1967 ) , which suggests an underlying genetic difference between species , more likely to be present in a specific substrate rather than in a general physiological mechanism such as anti-angiogenic effects or ROS production . To date , IMiD target identification efforts have largely focused on elucidating the mechanism of therapeutic efficacy of these drugs in MM and del ( 5q ) -MDS ( Gandhi et al . , 2014a; Krönke et al . , 2015 , 2014; Lu et al . , 2014 ) . However , these hematopoietic lineages may not express the specific proteins that are important in the developmental events disrupted by thalidomide during embryogenesis . In the absence of tractable animal models that closely resemble the human disease , we focused on human embryonic stem cells ( hESC ) as a model system that more likely expresses proteins relevant to embryo development , and set out to investigate the effects of thalidomide in this developmental context .
We established a mass spectrometry-based workflow ( Figure 1—figure supplement 1A ) to detect IMiD-induced protein degradation in hESC . To identify targets of IMiDs , we treated cells with 10 µM thalidomide , 5 µM lenalidomide , 1 µM of pomalidomide , or a DMSO control ( Figure 1—figure supplement 1B ) . To minimize transcriptional changes and other secondary effects that often result from extended drug exposure ( An et al . , 2017 ) , cells were treated for 5 h and protein abundance was measured in multiplexed mass spectrometry-based proteomics using tandem mass tag ( TMT ) isobaric labels ( McAlister et al . , 2014 ) ( Figure 1—figure supplement 1 and Materials and methods ) . From ~10 , 000 proteins quantified in H9 hESC , only the developmental spalt-like transcription factor SALL4 showed statistically significant downregulation across all three drug treatments with a change in protein abundance greater than 1 . 5-fold , and a p value < 0 . 001 ( Figure 1A–C ) . In accordance with previous findings , we also observed that treatment with lenalidomide led to degradation of CSNK1A1 ( Krönke et al . , 2015; Petzold et al . , 2016 ) . Pomalidomide induced degradation of additional targets , including the previously characterized zinc finger protein ZFP91 ( An et al . , 2017 ) and the largely uncharacterized proteins ZBTB39 , FAM83F , WIZ , RAB28 , and DTWD1 ( Figure 1A–C ) . This diverse set of neo-substrates observed in response to treatment with different IMiDs ( number of substrates identified: Thal < Len << Pom ) prompted us to further expand our exploration of IMiD-dependent neo-substrates by profiling IMiDs in additional cell lines . As degradation is mediated through CRL4CRBN , and because CRBN expression levels are high in the central nervous system ( CNS ) , we assessed the effects of IMiDs in two different neuroblastoma cell lines , Kelly and SK-N-DZ cells , as well as the commonly used multiple myeloma cell line , MM1s , as a control . Comprehensive proteomics studies across multiple independent replicates of hESC , Kelly , SK-N-DZ , and MM1s cells ( Figure 1A–D , see Materials and methods and Figure 1—figure supplements 1 and 2 for details ) , revealed multiple novel substrates for IMiDs ( ZNF692 , SALL4 , RNF166 , FAM83F , ZNF827 , RAB28 , ZBTB39 , ZNF653 , DTWD1 , ZNF98 , and GZF1 ) . To validate these novel targets , we carried out a ‘rescue’ proteomics experiment , in which we treated SK-N-DZ cells with 1 µM pomalidomide or with a co-treatment of 1 µM pomalidomide and 5 µM MLN4924 ( a specific inhibitor of the NAE1/UBA3 Nedd8 activating enzyme ) . Inhibition of the Cullin RING ligase ( CRL ) by MLN4924 fully abrogated IMiD-induced degradation of targets ( Figure 1—figure supplement 2B , C ) , and thereby confirmed the CRL-dependent mechanism . This approach was confirmed by spot-checking IMiD-dependent degradation for novel targets for which antibodies were available by western blot ( Figure 1—figure supplement 2D ) . All targets that were found to be consistently degraded across multiple large-scale proteomics experiments were validated in those independent validation experiments , providing a high confidence target list ( Figure 1D ) . Eight of the 11 new targets found in the proteomics screen are ZnF proteins ( SALL4 , ZNF827 , ZBTB39 , RNF166 , ZNF653 , ZNF692 , ZNF98 and GZF1 ) , and except for RNF166 , all contain at least one ZnF domain that has the characteristic features previously described as critical for IMiD-dependent degradation ( An et al . , 2017 ) ( Figure 1—figure supplement 2E ) . We also observe a striking difference in substrate specificity among thalidomide , lenalidomide , and pomalidomide ( Figure 1D ) . We find that thalidomide induces robust degradation of the zinc finger transcription factors ZNF692 , SALL4 , and the ubiquitin ligase RNF166 in cell lines expressing detectable levels of those proteins ( Figure 1D and Figure 1—figure supplement 2A ) . Lenalidomide results in additional degradation of ZNF827 , FAM83F , and RAB28 along with the lenalidomide-specific substrate CSNK1A1 . Pomalidomide shows the most pronounced expansion of targets , and in addition induces robust degradation of ZBTB39 , ZFP91 , DTWD1 , and ZNF653 . It is interesting to note that DTWD1 is , as CSNK1A1 and GSPT1 , another non zinc finger target that was found to be robustly degraded by pomalidomide . Although this expansion of substrates is interesting and may contribute to some of the clinical differences between lenalidomide and pomalidomide , a target causative for teratogenicity would need to be consistently degraded across all IMiDs . The robust down-regulation of SALL4 , a spalt-like developmental transcription factor important for limb development ( Koshiba-Takeuchi et al . , 2006 ) , upon treatment with thalidomide , lenalidomide , and pomalidomide prompted us to further investigate SALL4 as an IMiD-dependent target of CRL4CRBN . Strikingly , human genetic research has shown that familial loss of function ( LOF ) mutations in SALL4 are causatively linked to the clinical syndromes , Duane Radial Ray syndrome ( DRRS ) also known as Okihiro syndrome , and mutated in some patients with Holt-Oram syndrome ( HOS ) . Remarkably , both DRRS and HOS have large phenotypic overlaps with thalidomide embryopathy ( Kohlhase et al . , 2003 ) , and this phenotypic resemblance has led to misdiagnosis of patients with SALL4 mutations as cases of thalidomide embryopathy and the hypothesis that the tbx5/sall4 axis might be involved in thalidomide pathogenesis ( Knobloch and Rüther , 2008; Kohlhase et al . , 2003 ) . Thalidomide embryopathy is characterized not only by phocomelia , but also various other defects ( Table 1 ) , many of which are specifically recapitulated in syndromes known to originate from heterozygous LOF mutations in SALL4 ( Kohlhase , 1993 ) . The penetrance of DRRS in individuals with heterozygous SALL4 mutations likely exceeds 90% ( Kohlhase , 2004 ) , and thus partial degradation of SALL4 through IMiD exposure will likely result in similar clinical features observed in DRRS . All currently described SALL4 mutations are heterozygous LOF mutations , and the absence of homozygous mutations indicates the essentiality of the gene . Accordingly , homozygous deletion of Sall4 is early embryonic lethal in mice ( Sakaki-Yumoto et al . , 2006 ) . Mice with heterozygous deletion of Sall4 show a high frequency of miscarriage , while surviving litters show ventricular septal defects and anal stenosis , both phenotypes that are observed in humans with DRRS or thalidomide syndrome ( Sakaki-Yumoto et al . , 2006 ) . Mice carrying a heterozygous Sall4 genetrap allele show defects in heart and limb development , partially reminiscent of patients with DRRS or HOS ( Koshiba-Takeuchi et al . , 2006 ) . Another genetic disorder with a related phenotype is Roberts syndrome , caused by mutations in the ESCO2 gene ( Afifi et al . , 2016 ) . While ESCO2 similarly encodes for a zinc finger protein and is transcriptionally regulated by ZNF143 ( Nishihara et al . , 2010 ) , ESCO2 ( as well as ZNF143 , SALL1 , SALL2 , and SALL3 ) protein levels were found to be unchanged in all of our mass spectrometry experiments despite robust and ubiquitous expression ( Figure 1D , Figure 1—figure supplements 1 and 2 and Figure 1—source data 1–14 ) . The remarkable phenotypic overlap of LOF mutations in SALL4 with thalidomide embryopathy led us to further assess whether thalidomide and related IMiDs directly induce degradation of SALL4 in an IMiD and CRL4CRBN-dependent manner . To extend our mass spectrometry findings , we treated H9 hESC with increasing doses of thalidomide , lenalidomide , pomalidomide , or with DMSO as a control , and assessed protein levels of SALL4 by western blot . We observed a dose-dependent decrease in protein levels with all three drugs ( Figure 2A and Figure 2—figure supplement 1 ) , in accordance with IMiD-induced protein degradation . We then used qPCR to confirm that thalidomide treatment does not reduce the level of SALL4 mRNA , but rather upregulates SALL4 mRNA , consistent with the protein-level changes being caused by post-transcriptional effects ( Figure 2—figure supplement 1I ) . We next sought to assess the robustness of SALL4 degradation across different lineages by subjecting a panel of cell lines ( Kelly , SK-N-DZ , HEK293T , and H661 cells ) to increasing concentrations of thalidomide , lenalidomide , pomalidomide , or DMSO as a control and performed western blot analysis ( Figure 2B and Figure 2—figure supplement 1A–C ) . We observed a dose-dependent decrease in SALL4 protein levels with all three IMiD analogs and in all tested cell lines . In accordance with a CRL4CRBN-dependent mechanism , the IMiD-induced degradation was abrogated by co-treatment with the proteasome inhibitor bortezomib , the NEDD8 inhibitor MLN4924 , or the ubiquitin E1 ( UBA1 ) inhibitor MLN7243 ( which blocks all cellular ubiquitination by inhibiting the initial step of the ubiquitin conjugation cascade ) ( Figure 2C and Figure 2—figure supplement 1D , E ) . To further evaluate the CRL4CRBN-dependent mechanism , we generated CRBN-/- Kelly and HEK293T cells using CRISPR/Cas9 technology and treated the resulting CRBN-/- cells and parental cells with increasing concentrations of thalidomide , lenalidomide , or pomalidomide ( Figure 2D and Figure 2—figure supplement 1F ) . In agreement with the CRBN-dependent mechanism , no degradation of SALL4 was observed in CRBN-/- cells . Thalidomide has a plasma half-life ( t1/2 ) of ~6 to 8 h ( ~3 h for lenalidomide , ~9 h for pomalidomide ) and a maximum plasma concentration ( Cmax ) of ~5–10 µM ( ~2 . 5 µM for lenalidomide , 0 . 05 µM for pomalidomide ) upon a typical dose of 200–400 mg , 25 mg , or 2 mg for thalidomide , lenalidomide , or pomalidomide , respectively ( Chen et al . , 2017; Hoffmann et al . , 2013; Teo et al . , 2004 ) . To recapitulate these effects in vitro , we treated Kelly cells with 1 or 5 µM pomalidomide for 8 h , followed by washout of the drug and assessment of time-dependent recovery of SALL4 protein levels ( Figure 2E and Figure 2—figure supplement 1G ) . Treatment with pomalidomide induces degradation of SALL4 as early as 4 h post treatment ( Figure 2F and Figure 2—figure supplement 1H ) , which recovered to levels close to pre-treatment level after 48 h post washout ( Figure 2E ) , together suggesting that a single dose of IMiD drugs will be sufficient to deplete SALL4 protein levels for >24 h . Bona fide targets of IMiD-induced degradation typically bind to CRBN ( the substrate-recognition domain of the E3 ligase ) in vitro in a compound-dependent manner . Thus , we sought to test whether SALL4 binds to CRBN and to map the ZnF domain required for binding using purified recombinant proteins . Based on conserved features among IMiD-sensitive ZnF domains ( Figure 3A , C–x ( 2 ) -C-G motif within the canonical C2H2 zinc finger motif ) , the second ( SALL4ZnF2 ) and fourth ( SALL4ZnF4 ) ZnF domains of SALL4 ( aa 410–433 , and aa 594–616 , respectively ) were identified as candidate degrons for IMiD-induced binding . We expressed , purified , biotinylated , and subjected these ZnF domains to in vitro CRBN binding assays ( An et al . , 2017; Petzold et al . , 2016 ) . We observed dose-dependent binding between SALL4ZnF2 or SALL4ZnF4 and CRBN similar to that described for IKZF1/3 and ZFP91 , albeit with reduced apparent affinity for SALL4ZnF4 ( Figure 3B , C ) ( Petzold et al . , 2016 ) . To estimate apparent affinities ( KD ( app ) ) we titrated bodipy-FL labelled DDB1∆B-CRBN to biotinylated SALL4ZnF2 , or SALL4ZnF4 at 100 nM with saturating concentrations of IMiDs ( 50 µM ) and measured the affinity by TR-FRET ( Figure 3D and Figure 3—figure supplement 1A , B ) , which confirmed the weak affinity of SALL4ZnF4 . However , we noticed that a construct spanning ZnF1 and ZnF2 of SALL4 ( SALL4ZnF1-2 ) exhibited even tighter binding to CRBN ( Figure 3D and Figure 3—figure supplement 1A , B ) and enhanced dose-dependent complex formation in TR-FRET ( Figure 3E ) . These findings suggest that multiple zinc finger domains of SALL4 contribute to binding , and may result in multivalent recruitment to CRBN in vivo . However , the strength of the interaction with ZnF4 is unlikely to be sufficient for degradation in cells , and moreover , the rank order of Pom >Thal >> Len in binding observed with ZnF2 is in accordance with the cellular potency in degradation of SALL4 , suggesting that ZnF2 is the critical ZnF domain for SALL4 degradation . We confirmed the specificity of the SALL4ZnF2 interaction by introducing a point mutation to glycine 416 ( G416 ) , the residue critical for IMiD-dependent binding to CRBN ( Petzold et al . , 2016 ) . Mutations to alanine ( G416A ) rendered SALL4ZnF2 resistant to IMiD-dependent binding to CRBN ( Figure 3F and Figure 3—figure supplement 1C , D ) . Mutating glutamine 595 ( Q595 ) in SALL4ZnF4 , another residue previously shown to be critical for IMiD-dependent CRBN binding in the ZnF domains of IKZF1/3 , impaired IMiD-dependent binding ( Figure 3—figure supplement 1E ) , confirming the specificity of the interaction despite the weak binding affinity . As we observed increased affinity of the tandem-ZnF construct SALL4ZnF1-2 compared with the single SALL4ZnF2 , we sought to test whether ZnF1 was sufficient for binding . We introduced the G416N mutation in ZnF2 or a S388N mutation in ZnF1 into the SALL4ZnF1-2 construct ( S388 is the ZnF1 sequence equivalent of ZnF2 G416; ZnF1-2: C-x-x-C-S/G ) and performed CRBN binding assays . G416N , but not S388N , fully abrogated IMiD-dependent binding of SALL4ZnF1-2 to CRBN ( Figure 3—figure supplement 1F–I ) , confirming the strict dependence on the ZnF2 interaction with CRBN . To test whether the second zinc finger of SALL4 is critical for IMiD-induced degradation in cells , we introduced G416A and G416N mutations into Flag-tagged full-length SALL4 . When expressed in Kelly cells , the parental wild-type Flag-SALL4 was readily degraded by thalidomide treatment ( Figure 3G ) . Similarly , Flag-tagged SALL4 with G600A or G600N mutations in ZnF4 were also shown to be readily degraded with thalidomide treatment , suggesting that SALL4ZnF4 is dispensable for binding and subsequent degradation ( Figure 3G ) . Finally , the two conservative mutations in ZnF2 ( G416A or G416N ) , both known to specifically disrupt binding to CRBN while maintaining the overall zinc finger fold ( Petzold et al . , 2016 ) , rendered SALL4 stable under these treatment conditions , demonstrating that SALL4ZnF2 is necessary for CRL4CRBN-mediated degradation of SALL4 in cells ( Figure 3H ) . In vitro ubiquitination assays further confirm that SALL4ZnF1-2 is ubiquitinated by CRL4CRBN in an IMiD-dependent fashion ( Figure 3I ) . Together , our cellular and biochemical data establish SALL4 as a bona fide IMiD-dependent target of CRL4CRBN , and demonstrate that the second zinc finger is necessary for IMiD-dependent degradation , while the tandem array of ZnF1-2 further strengthens the interaction in vitro . One characteristic feature of IMiD phenotypes is the absence of defining limb deformities following administration to pregnant rodents , which contributed to the initial approval by regulatory agencies in Europe . In contrast , many non-human primates exhibit phenotypes that mimic the human syndrome ( Neubert et al . , 1988; Smith et al . , 1965; Vickers , 1967 ) . These remarkable species-specific phenotypes have historically complicated studies of thalidomide embryopathies , and suggest a genetic difference between these species that would abrogate the detrimental effects of thalidomide . Mouse Crbn harbors a critical polymorphism ( Figure 4A , B and Figure 5 ) that prevents IMiD-dependent degradation of ZnF substrates and CSNK1A1 ( Krönke et al . , 2015 ) , which could explain the absence of a SALL4-dependent phenotype in mice . Mice and rats ( both insensitive to thalidomide embryopathies ) harbor an isoleucine at CRBN position 388 ( residue 388 refers to the human CRBN sequence ) . In contrast , sensitive primates have a valine in position 388 that is necessary for CRL4CRBN to bind , ubiquitinate , and subsequently degrade ZnF substrates ( Figure 4A , B and Figure 5A ) . Consistent with this concept , treatment of mouse embryonic stem cells ( mESC ) with increasing concentrations of thalidomide or pomalidomide does not promote degradation of mmSALL4 ( Figure 4C and Figure 4—figure supplement 1A ) , and introducing a V388I mutation into hsCRBN renders the protein less effective to bind to SALL4 in vitro ( Figure 4B ) . We thus asked whether ectopic expression of hsCRBN in mouse cells would lead to IMiD-induced degradation of mmSALL4 , similar to what had been observed for CSNK1A1 , and could hence render mice sensitive to IMiD-induced birth defects . Expression of hsCRBN in mouse cells , while sensitizing cells to degradation of IMiD targets such as mmIKZF1/3 , mmCSNK1A1 ( Krönke et al . , 2015 ) , mmZFP91 , or mmGZF1 ( Figure 4D , E ) , does not result in degradation of mmSALL4 ( Figure 4F ) . To test whether a fully human CRBN in a human cell background would be sufficient to induce SALL4 degradation , we introduced hsSALL4 , or mmSALL4 into human cells ( Kelly cells ) and found that while ectopically expressed hsSALL4 is readily degraded upon IMiD treatment , mmSALL4 is unaffected even at arbitrarily high doses of IMiDs ( Figure 4G and Figure 4—figure supplement 1B , C ) . Sequence analysis reveals that mice and zebrafish have critical mutations in the ZnF2 domain of SALL4 ( Figure 5B ) , which abrogate binding to hsCRBN in vitro ( Figure 4H ) , and render mmSALL4 and drSALL4 insensitive to IMiD-mediated degradation in cells ( Figure 4G , I and Figure 4—figure supplement 1C ) . In line with these findings , mice harboring a homozygous CRBN I391V knock-in allele , despite exhibiting degradation of mmIKZF1/3 , mmZFP91 , and mmCSNK1A1 Fink et al . , 2018 , show increased miscarriage upon IMiD treatment compared with control mice; however , they do not exhibit IMiD-induced embryopathies resembling the human phenotype Fink et al . , 2018 . We next sought to test whether exchange of the mmSALL4 ZnF2 domain for the hsSALL4 ZnF2 domain would be sufficient to enable mmSALL4 degradation in a human cell line ( Kelly cells ) . Strikingly , through the five amino acid substitutions required to ‘humanize’ the mmSALL4 ZnF2 domain , we were able to induce thalidomide-mediated mouse SALL4 degradation in a human cell line ( Figure 4G ) . The observation that SALL4 degradation depends on both the sequence of SALL4 ( zinc finger 2 differs between human and rodents ) , and the sequence of CRBN , supports a genetic cause for the species-specific effects , and highlights the complexities of modelling teratogenic adverse effects of IMiDs in murine and other animal models ( Sakaki-Yumoto et al . , 2006 ) ( Figure 5A–C ) . Of note , the only non-human primate known to be insensitive to thalidomide-induced embryopathies , the greater bush baby , also harbors an isoleucine in the critical CRBN V388 position ( Butler , 1977 ) , while all sensitive non-human primates and rabbits harbor the conserved valine ( Figure 5A ) . We thus show that species can be rendered resistant by mutations in CRBN , SALL4 , or both , and hence our data suggest that thalidomide embryopathy is primarily a human disease ( with some non-human primates , and rabbits more closely resembling the phenotypes ) , and thus explain the historic observation that modelling thalidomide embryopathies in animals is challenging . We note that zebrafish and chicken both contain an Ile in the V388 position; however , these were reported to exhibit defects to limb/fin formation upon exposure to thalidomide or knock-down of Crbn ( Eichner et al . , 2016; Ito et al . , 2010 ) , partially resembling thalidomide-induced defects . These findings are in contrast with the observations in higher eukaryotes , as Crbn knock-out mice have been reported to exhibit normal morphology ( Lee et al . , 2013 ) , and children harboring a homozygous C391R mutation in CRBN ( C391 is a structural cysteine coordinating the zinc in the thalidomide-binding domain of CRBN and we failed to produce any protein from a C391R cDNA ) , a loss of function mutation , were born without characteristic birth defects but exhibited severe neurological defects ( Sheereen et al . , 2017 ) . Whether the phenotypes in zebrafish and chicken are a result of species-specific downstream pathways or the high dose ( 400 µM ) and direct application of thalidomide to the limb buds ( Ito et al . , 2010 ) , which both could result in off-target effects , remains to be shown . The plasma concentration of thalidomide in humans is , however , unlikely to exceed 10 µM ( Bai et al . , 2013; Dahut et al . , 2009 ) , a concentration that results in effective degradation of SALL4 , but is 40 times below the dose found to be teratogenic in chicken and zebrafish embryos . While we do not observe degradation of mmSALL4 or drSALL4 upon high-dose exposure , we cannot rule out that such high doses will induce degradation of other ZnF targets in zebrafish or chicken , which could potentially result in the observed phenotypes . In fact , we show that IMiDs lead to degradation of multiple ZnF transcription factors , a class of proteins known to evolve very rapidly ( Schmitges et al . , 2016 ) , and it is likely that IMiDs will exhibit species-specific effects . Sequence analysis shows that IMiD-dependent ZnF targets such as SALL4 , ZNF653 , ZNF692 , or ZBTB39 , as well as other known genetic causes of limb defects in ZnF transcription factors , such as ESCO2 , are highly divergent even in higher eukaryotes ( Figure 5D ) .
We show that thalidomide , lenalidomide , and pomalidomide all induce degradation of SALL4 , which has been causatively linked to the most characteristic and common birth defects of the limbs and inner organs by human genetics . While other targets of thalidomide , such as CSNK1A1 for lenalidomide or GZF1 , ZBTB39 for pomalidomide , may contribute to the pleiotropic developmental conditions observed upon thalidomide exposure , SALL4 is consistently degraded across all IMiDs and human geneticists associate heterozygous loss of SALL4 with human developmental syndromes that largely phenocopy thalidomide syndrome . Moreover , from the targets degraded across IMiDs , IKZF1/3 have been shown to be non-causative for birth defects , RNF166 is a ubiquitin ligase involved in autophagy ( Heath et al . , 2016 ) , and ZNF692 knock-out mice do not exhibit a teratogenic phenotype ( International Mouse Phenotyping Consortium et al . , 2016 ) . While only genetic studies in non-human primates or rabbits can provide the ultimate molecular role of SALL4 and other targets in thalidomide embryopathies , the known functions of SALL4 are consistent with a potential role in thalidomide embryopathies . The polypharmacology of IMiDs ( most notably pomalidomide ) , together with the size and rapid evolution of the C2H2 family of zinc finger transcription factors ( Figure 5D ) , which results in most C2H2 zinc finger transcription factors being highly species-specific ( Najafabadi et al . , 2015; Schmitges et al . , 2016 ) , help to explain the pleiotropic effects of IMiDs , which still remain largely understudied . Thalidomide embryopathies thus represent a case in which animal studies fall short , and it is likely that the clinical features of IMiD efficacy as well as adverse effects , are a result of induced degradation of multiple C2H2 zinc finger transcription factors . For example , we see some degree of degradation for GZF1 , another C2H2 transcription factor , while GZF1 is unlikely to cause the defining birth defects of thalidomide , mutations in GZF1 have been associated with joint laxity and short stature , which are both also found in thalidomide-affected children ( Patel et al . , 2017 ) . We also note that CRBN expression levels influence the efficacy of IMiDs in inducing protein degradation , and it is conceivable that these contribute to a certain degree of tissue selectivity of IMiD effects , which for example , could increase the therapeutic index in MM as hematopoietic lineages tend to have high levels of CRBN . Thalidomide teratogenicity was a severe and widespread public health tragedy , affecting more than 10 , 000 individuals , and the aftermath has shaped many of the current drug regulatory procedures . Our findings that thalidomide and its derivatives induce degradation of SALL4 , provide a direct link to genetic disorders of SALL4 deficiency , which phenocopy many of the teratogenic effects of thalidomide . While other effects of thalidomide , such as anti-angiogenic properties may contribute to birth defects , degradation of SALL4 is likely to contribute to birth defects . These findings can inform the development of new compounds that induce CRBN-dependent degradation of disease-relevant proteins but avoid degradation of developmental transcription factors such as SALL4 , and thus have the potential for therapeutic efficacy without the risk of teratogenicity , a defining feature of this class of drugs . This is further relevant to the development of thalidomide-derived bifunctional small molecule degraders ( commonly referred to as PROTACs ) ( Raina and Crews , 2017 ) , as we show that IMiD-based PROTACs ( and novel IMiD derivatives such as CC-220 ) can be effective inducers of ZnF targets including SALL4 degradation ( Figure 1—figure supplement 1C ) . Lastly , the surprising expansion in substrate repertoire for pomalidomide , suggests that IMiDs exhibit a large degree of polypharmacology contributing to both efficacy and adverse effects . Transcription factors , and specifically C2H2 zinc fingers are highly divergent between species , and hence IMiDs and related compounds are likely to exhibit species-specific effects by virtue of their mode of action . In turn , the discovery that IMiDs target an unanticipated large set of C2H2 zinc finger proteins with significant differences among thalidomide , lenalidomide , pomalidomide , and CC-220 , suggests that this chemical scaffold holds the potential to target one of the largest families of human transcription factors .
Thalidomide ( HY-14658 , MedChemExpress ) , lenalidomide ( HY-A0003 , MedChemExpress ) , pomalidomide ( HY-10984 , MedChemExpress ) , CC-220 ( HY-101291 , MedChemExpress ) , CC-885 ( 19966 , Cayman chemical ) , dBET57 ( Nowak et al . , 2018 ) , bortezomib ( HY-10227 , MedChemExpress ) , MLN4924 ( HY-70062 , MedChemExpress ) , and MLN7243 ( A1384 , Active Biochem ) were purchased from the indicated vendors and subjected to in-house LC-MS for quality control . HEK293T , SK-N-DZ , MM1s , and H661 were purchased from ATCC and cultured according to ATCC instructions . H9 hESC , mESC , and Kelly cells were kindly provided by the labs of J . Wade Harper ( HMS ) , Richard I . Gregory ( TCH/HMS ) , and Nathanael Gray ( DFCI/HMS ) , respectively . Sequencing grade modified trypsin ( V5111 ) was purchased from Promega ( Promega , USA ) and mass spectrometry grade lysyl endopeptidase from Wako ( Wako Pure Chemicals , Japan ) . Primary and secondary antibodies used included anti-SALL4 at 1:250 dilution ( ab57577 , abcam – found reactive for human SALL4 ) , anti-SALL4 chip grade at 1:250 dilution ( ab29112 , abcam – found reactive for mouse Sall4 ) , anti-DTWD1 1:500 ( HPA042214 , Sigma ) , anti-Flag 1:1000 ( F1804 , Sigma ) , anti-CRBN 1:500 ( NBP1-91810 , Novus Biologicals ) , anti-GZF1 at 1:500 ( PA534375 , Thermo Fisher Scientific ) , anti-GAPDH at 1:10 , 000 dilution ( G8795 , Sigma ) , IRDye680 Donkey anti-mouse at 1:10 , 000 dilution ( 926–68072 , LiCor ) , IRDye800 Goat anti-rabbit at 1:10 , 000 dilution ( 926–32211 , LiCor ) and rabbit anti-Strep-Tag II antibody at 1:10 , 000 ( ab76949 , Abcam ) , anti-mouse IgG HRP-linked Antibody at 1:10 , 000 dilution ( 7076 , Cell Signaling ) , Amersham ECL Prime Western Blotting Detection Reagent ( RPN2232 , GE ) . HEK293T cells were cultured in DMEM supplemented with 10% dialyzed fetal bovine serum ( FBS ) and 2 mM L-glutamine . SK-N-DZ cells were cultured in DMEM supplemented with 10% dialyzed FBS , 0 . 1 mM non-essential amino acids ( NEAA ) , and 2 mM L-glutamine . H661 , MM1s , and Kelly cells were cultured in RPMI1640 supplemented with 10% dialyzed FBS . H9 hESC cells were cultured in Essential 8 ( Gibco ) media on Matrigel-coated nunc tissue culture plates . TC1 mouse embryonic stem cells ( mESCs ) were adapted to gelatin cultures and fed with KO-DMEM ( Gibco ) supplemented with 15% stem cell-qualified fetal bovine serum ( FBS , Gemini ) , 2 mM L-glutamine ( Gibco ) , 20 mM HEPES ( Gibco ) , 1 mM sodium pyruvate ( Gibco ) , 0 . 1 mM of each non-essential amino acids ( Gibco ) , 0 . 1 mM 2-mercaptoethanol ( Sigma ) , 104 U mL−1 penicillin/streptomycin ( Gibco ) , and 103 U mL−1 mLIF ( Gemini ) . Cell lines were acquired from sources provided in the key resource table . All cell lines are routinely authenticated using ATCC STR service , and are tested for mycoplasma contamination on a monthly basis . All cell lines used for experiments tested negative . Cells were treated with compounds as indicated and incubated for 24 h , or as indicated . Samples were run on 4−20% , AnyKD or 10% ( in-vitro ubiqutination assay ) SDS-PAGE gels ( Bio-rad ) , and transferred to PVDF membranes using the iBlot 2 . 0 dry blotting system ( Thermo Fisher Scientific ) . Membranes were blocked with LiCor blocking solution ( LiCor ) , and incubated with primary antibodies overnight , followed by three washes in LiCor blocking solution and incubation with secondary antibodies for 1 h in the dark . After three final washes , the membranes were imaged on a LiCor fluorescent imaging station ( LiCor ) . When Anti-mouse IgG , HRP Antibody was used , after three washes , the membranes were incubated with Amersham ECL Prime Western Blotting Detection Reagent for 1 min and subjected to imaging by Amersham Imager 600 ( GE ) . hsCRBN , hsSALL4 , mmSALL4 , and drSALL4 were PCR amplified and cloned into a pNTM-Flag based vector . Mutagenesis was performed using the Q5 site-directed mutagenesis kit ( NEB , USA ) with primers designed using the BaseChanger web server ( http://nebasechanger . neb . com/ ) . Primer sets used for Q5 mutagenesis are: hsSALL4 - S388N Fwd 5’−3’: AAGTACTGTAaCAAGGTTTTTG Rev 5’−3’: ACACTTGTGCTTGTAGAG hsSALL4 – G416A Fwd 5’−3’: TCTGTCTGTGcTCATCGCTTCAC Rev 5’−3’: GCACACGAAGGGTCTCTC hsSALL4 – G416N Fwd 5’−3’: CTCTGTCTGTaaTCATCGCTTCACCAC Rev 5’−3’: CACACGAAGGGTCTCTCT hsSALL4 – G600A Fwd 5’−3’: AAGATCTGTGcCCGAGCCTTTTC Rev 5’−3’: ACACTGGAACGGTCTCTC hsSALL4 – G600N Fwd 5’−3’: TAAGATCTGTaaCCGAGCCTTTTCTAC Rev 5’−3’: CACTGGAACGGTCTCTCC Humanizing mmSALL4 – Y415F , P418S , I419V , L430F , Q435H Fwd 5’−3’: AGGGCAATCTCAAGGTCCACTTtCAcCGACACCCTCAGGTGAAGGCAAACCCCC Rev 5’−3’: TGGTGGTGAAGCGGTGACCACAGAcAGaGCACACGaAAGGTCTCTCTCCGGTGTG For transient transfection , 0 . 2 million cells were seeded per well in a 12 well plate on day 1 . On day 2 , cells were transfected with 200–300 ng of plasmid ( pNTM-Flag containing gene of interest ) using 2 µL of lipofectamine 2000 transfection reagent ( Invitrogen ) . On day 3 , the desired concentration of IMiD was added to each well and cells were harvested after 24 h for western blot analysis using the protocol described above . His6DDB1∆B ( Petzold et al . , 2016 ) , His6-3C-SpyhsCRBN , His6-3C-SpyhsCRBNV388I , Strep-BirAhsSALL4590-618 ( ZnF4 ) , Strep-BirAhsSALL4Q595H590-618 ( ZnF4 ) , Strep-BirAhsSALL4378-438 ( ZnF1-2 ) , Strep-BirAhsSALL4402-436 ( ZnF2 ) , Strep-BirAmmSALL4593-627 ( ZnF4 ) , Strep-BirAdrSALL4583-617 ( ZnF2 ) were subcloned into pAC-derived vectors or BigBac vector for HishsDDB11-1140-HishsCUL4A38-759-HhismmRBX112-108 ( CRL4CRBN ) . Mutant Strep-BirAhsSALL4378-438 ( ZnF1-2 ) and Strep-BirAhsSALL4402-436 ( ZnF2 ) constructs were derived from these constructs using Q5 mutagenesis ( NEB , USA ) . Recombinant proteins expressed in Trichoplusia ni High Five insect cells ( Thermo Fisher Scientific ) using the baculovirus expression system ( Invitrogen ) . For purification of DDB1∆B-CRBNSpyBodipyFL or CRL4CRBN , cells were resuspended in buffer containing 50 mM tris ( hydroxymethyl ) aminomethane hydrochloride ( Tris-HCl ) pH 8 . 0 , 200 mM NaCl , 1 mM tris ( 2-carboxyethyl ) phosphine ( TCEP ) , 1 mM phenylmethylsulfonyl fluoride ( PMSF ) , 1× protease inhibitor cocktail ( Sigma ) and lyzed by sonication . Cells expressing variations of Strep-BirASALL4 were lyzed in the presence of 50 mM Tris-HCl pH 8 . 0 , 500 mM NaCl , 1 mM TCEP , 1 mM PMSF , and 1× protease inhibitor cocktail ( Sigma ) . Following ultracentrifugation , the soluble fraction was passed over appropriate affinity resin Ni Sepharose 6 Fast Flow affinity resin ( GE Healthcare ) or Strep-Tactin Sepharose XT ( IBA ) , and eluted with 50 mM Tris-HCl pH 8 . 0 , 200 mM NaCl , 1 mM TCEP , 100 mM imidazole ( Fischer Chemical ) for His6-tagged proteins or 50 mM Tris-HCl pH 8 . 0 , 500 mM NaCl , 1 mM TCEP , 50 mM D-biotin ( IBA ) for Strep-tagged proteins . Affinity-purified proteins were either further purified via ion exchange chromatography ( Poros 50HQ ) and subjected to size exclusion chromatography ( SEC200 HiLoad 16/60 , GE ) ( His6DDB1∆B-His6-3C-SpyCRBN or CRL4CRBN ) or biotinylated overnight , concentrated , and directly loaded on the size exclusion chromatography ( ENRich SEC70 10/300 , Bio-rad ) in 50 mM HEPES pH 7 . 4 , 200 mM NaCl , and 1 mM TCEP . Biotinylation of Strep-BirASALL4 constructs was performed as previously described ( Cavadini et al . , 2016 ) . The protein-containing fractions were concentrated using ultrafiltration ( Millipore ) , flash frozen in liquid nitrogen , and stored at −80°C or directly covalently labeled with BODIPY-FL-SpyCatcherS50C as described below . Spycatcher ( Zakeri et al . , 2012 ) containing a Ser50Cys mutation was obtained as a synthetic dsDNA fragment from IDT ( Integrated DNA technologies ) and subcloned as a GST-TEV fusion protein in a pET-Duet-derived vector . Spycatcher S50C was expressed in BL21 DE3 and cells were lyzed in the presence of 50 mM Tris-HCl pH 8 . 0 , 200 mM NaCl , 1 mM TCEP , and 1 mM PMSF . Following ultracentrifugation , the soluble fraction was passed over glutathione sepharose 4B ( GE Healthcare ) and eluted with wash buffer ( 50 mM Tris-HCl pH 8 . 0 , 200 mM NaCl , 1 mM TCEP ) supplemented with 10 mM glutathione ( Fischer BioReagents ) . The affinity-purified protein was TEV cleaved , subjected to size exclusion chromatography , concentrated , and flash frozen in liquid nitrogen . In vitro ubiquitination was performed by mixing biotinylated SALL4 ZnF1-2 at 0 . 6 μM , and CRL4CRBN at 80 nM with a reaction mixture containing IMiDs at indicated concentrations or a DMSO control , E1 ( UBA1 , Boston Biochem ) at 30 nM , E2 ( UbcH5c , Boston Biochem and UBE2G1 ) at 1 . 0 μM each , ubiquitin ( Ubiquitin , Boston Biochem ) at 23 μM . Reactions were carried out in 50 mM Tris pH 7 . 5 , 30 mM NaCl , 5 mM MgCl2 , 0 . 2 mM CaCl2 , 2 . 5 mM ATP , 1mM DTT , 0 . 1% Triton X-100 and 2 . 0 mg mL−1 BSA , incubated for 60 min at 30°C and analyzed by western blot using rabbit anti-Strep-Tag II antibody at 1:10 , 000 ( ab76949 , Abcam ) as described above . TC1 mES cells were transduced with a pCDH-MSCV-based lentiviral vector expressing hsCRBN , GFP , and the puromycin resistance gene . Infection was performed after 24 h in culture in a six-well 0 . 2% gelatin-coated plate using standard infection protocol in the presence of 2 µg mL−1 polybrene ( hexadimethrine bromide , Sigma ) . 72 h after transduction the cells were subjected to two rounds of puromycin selection ( 5 µg mL−1 ) to form mES cells stably expressing hsCRBN , which were confirmed to be >90% GFP-positive under fluorescent microscope . Purified SpycatcherS50C protein was incubated with DTT ( 8 mM ) at 4°C for 1 h . DTT was removed using a ENRich SEC650 10/300 ( Bio-rad ) size exclusion column in a buffer containing 50 mM Tris pH 7 . 5 and 150 mM NaCl , 0 . 1mM TCEP . Bodipy-FL-maleimide ( Thermo Fisher Scientific ) was dissolved in 100% DMSO and mixed with SpycatcherS50C to achieve 2 . 5 molar excess of Bodipy-FL-maleimide . SpyCatcherS50C labeling was carried out at room temperature ( RT ) for 3 h and stored overnight at 4°C . Labeled SpycatcherS50C was purified on an ENRich SEC650 10/300 ( Bio-rad ) size exclusion column in 50 mM Tris pH 7 . 5 , 150 mM NaCl , 0 . 25 mM TCEP , and 10% ( v/v ) glycerol , concentrated by ultrafiltration ( Millipore ) , flash frozen ( ~40 µM ) in liquid nitrogen , and stored at −80°C . Purified His6DDB1∆B-His6-3C-SpyCRBN constructs ( WT and V388I ) were incubated overnight at 4°C with Bodipy-FL-maleimide-labeled SpyCatcherS50C protein at stoichiometric ratio . The protein was concentrated and loaded on the ENrich SEC 650 10/300 ( Bio-rad ) size exclusion column , and the fluorescence was monitored with absorption at 280 and 490 nm . Protein peak corresponding to the labeled protein was pooled , concentrated by ultrafiltration ( Millipore ) , flash frozen in liquid nitrogen , and stored at −80°C . Compounds in binding assays were dispensed into a 384-well microplate ( Corning , 4514 ) using the D300e Digital Dispenser ( HP ) normalized to 1% DMSO and containing 100 nM biotinylated strep-avi-SALL4 ( WT or mutant , see Figure legends ) , 1 µM His6-DDB1∆B-His6-CRBNBodipy-Spycatcher , and 4 nM terbium-coupled streptavidin ( Invitrogen ) in a buffer containing 50 mM Tris pH 7 . 5 , 100 mM NaCl , 1mM TCEP , and 0 . 1% Pluronic F-68 solution ( Sigma ) . Before TR-FRET measurements were conducted , the reactions were incubated for 15 min at RT . After excitation of terbium fluorescence at 337 nm , emission at 490 nm ( terbium ) and 520 nm ( Bodipy ) were recorded with a 70 µs delay over 600 µs to reduce background fluorescence , and the reaction was followed over 30× 200 s cycles of each data point using a PHERAstar FS microplate reader ( BMG Labtech ) . The TR-FRET signal of each data point was extracted by calculating the 520/490 nm ratios . Data from three independent measurements ( n=3 ) , each calculated as an average of five technical replicates per well per experiment , was plotted and the half-maximal effective concentrations EC50 values calculated using variable slope equation in GraphPad Prism 7 . Apparent affinities were determined by titrating Bodipy-FL-labelled DDB1∆B-CRBN to biotinylated strep-avi-SALL4 ( constructs as indicated ) at 100 nM , and terbium-streptavidin at 4 nM . The resulting data were fitted as described previously ( Petzold et al . , 2016 ) . H9 hES cells treated with 10 µM thalidomide or DMSO for 24 h were subjected to gene expression analysis . RNA was isolated using the RNeasy Plus mini kit ( Qiagen ) and cDNA created by reverse transcription using ProtoScript II reverse transcriptase ( NEB ) following the manufacturer’s instructions . The following primer sets from IDT were used with SYBR Green Master Mix ( Applied Biosystems ) to probe both GAPDH and total SALL4 levels: SALL4total – F: GGTCCTCGAGCAGATCTTGT SALL4total – R: GGCATCCAGAGACAGACCTT GAPDH – F: GAAGGTGAAGGTCGGAGTC GAPDH – R: GAAGATGGTGATGGGATTTC Analysis was performed on a CFX Connect Real-Time PCR System ( Bio-Rad ) in a white 96-well PCR plate . Relative expression levels were calculated using the ΔΔCT method . H9 hESC , Kelly , SK-N-DZ , and MM1s cells were treated with DMSO , 1 µM pomalidomide , 5 µM lenalidomid , e or 10 µM thalidomide in biological triplicates ( DMSO ) or biological duplicates ( pomalidomide , lenalidomide , thalidomide ) for 5 h , and cells were harvested by centrifugation . Lysis buffer ( 8 M urea , 50 mM NaCl , 50 mM 4- ( 2hydroxyethyl ) -1-piperazineethanesulfonic acid ( EPPS ) pH 8 . 5 , 1× Roche protease inhibitor , and 1× Roche PhosphoStop ) was added to the cell pellets and cells were homogenized by 20 passes through a 21 gauge ( 1 . 25 in . long ) needle to achieve a cell lysate with a protein concentration between 0 . 5 and 4 mg mL−1 . The homogenized sample was clarified by centrifugation at 20 , 000 × g for 10 min at 4°C . A micro-BCA assay ( Pierce ) was used to determine the final protein concentration in the cell lysate . 200 µg protein for each sample were reduced and alkylated as previously described ( An et al . , 2017 ) . Proteins were precipitated using methanol/chloroform . In brief , four volumes of methanol were added to the cell lysate , followed by one volume of chloroform , and finally three volumes of water . The mixture was vortexed and centrifuged at 14 , 000 × g for 5 min to separate the chloroform phase from the aqueous phase . The precipitated protein was washed with three volumes of methanol , centrifuged at 14 , 000 × g for 5 min , and the resulting washed precipitated protein was allowed to air dry . Precipitated protein was resuspended in 4 M urea , 50 mM HEPES pH 7 . 4 , followed by dilution to 1 M urea with the addition of 200 mM EPPS pH 8 for digestion with LysC ( 1:50; enzyme:protein ) for 12 h at room temperature . The LysC digestion was diluted to 0 . 5 M urea , 200 mM EPPS pH 8 , and then digested with trypsin ( 1:50; enzyme:protein ) for 6 h at 37°C . Tandem mass tag ( TMT ) reagents ( Thermo Fisher Scientific ) were dissolved in anhydrous acetonitrile ( ACN ) according to manufacturer’s instructions . Anhydrous ACN was added to each peptide sample to a final concentration of 30% v/v , and labeling was induced with the addition of TMT reagent to each sample at a ratio of 1:4 peptide:TMT label . The 10-plex labeling reactions were performed for 1 . 5 h at room temperature and the reaction quenched by the addition of 0 . 3% hydroxylamine for 15 min at room temperature . The sample channels were combined in a 1:1:1:1:1:1:1:1:1:1 ratio , desalted using C18 solid phase extraction cartridges ( Waters ) and analyzed by LC-MS for channel ratio comparison . Samples were then combined using the adjusted volumes determined in the channel ratio analysis and dried down in a speed vacuum . The combined sample was then resuspended in 1% formic acid , and acidified ( pH 2−3 ) before being subjected to desalting with C18 SPE ( Sep-Pak , Waters ) . Samples were then offline fractionated into 96 fractions by high pH reverse-phase HPLC ( Agilent LC1260 ) through an aeris peptide xb-c18 column ( phenomenex ) with mobile phase A containing 5% acetonitrile and 10 mM NH4HCO3 in LC-MS grade H2O , and mobile phase B containing 90% acetonitrile and 10 mM NH4HCO3 in LC-MS grade H2O ( both pH 8 . 0 ) . The 96 resulting fractions were then pooled in a non-continuous manner into 24 fractions or 48 fractions and every fraction was used for subsequent mass spectrometry analysis . Data were collected using an Orbitrap Fusion Lumos mass spectrometer ( Thermo Fisher Scientific , San Jose , CA , USA ) coupled with a Proxeon EASY-nLC 1200 LC pump ( Thermo Fisher Scientific ) . Peptides were separated on a 50 cm and 75 μm inner diameter Easyspray column ( ES803 , Thermo Fisher Scientific ) . Peptides were separated using a 3 h gradient of 6–27% acetonitrile in 1 . 0% formic acid with a flow rate of 300 nL/min . Each analysis used an MS3-based TMT method as described previously ( McAlister et al . , 2014 ) . The data were acquired using a mass range of m/z 350–1350 , resolution 120 , 000 , AGC target 1 × 106 , maximum injection time 100 ms , dynamic exclusion of 90 s for the peptide measurements in the Orbitrap . Data-dependent MS2 spectra were acquired in the ion trap with a normalized collision energy ( NCE ) set at 35% , AGC target set to 1 . 8 × 104 , and a maximum injection time of 120 ms . MS3 scans were acquired in the Orbitrap with a HCD collision energy set to 55% , AGC target set to 1 . 5 × 105 , maximum injection time of 150 ms , resolution at 50 , 000 , and with a maximum synchronous precursor selection ( SPS ) precursors set to 10 . Proteome Discoverer 2 . 2 ( Thermo Fisher ) was used for RAW file processing and controlling peptide and protein level false discovery rates , assembling proteins from peptides , and protein quantification from peptides . MS/MS spectra were searched against a Uniprot human database ( September 2016 ) with both the forward and reverse sequences . Database search criteria are as follows: tryptic with two missed cleavages , a precursor mass tolerance of 20 ppm , fragment ion mass tolerance of 0 . 6 Da , static alkylation of cysteine ( 57 . 02146 Da ) , static TMT labeling of lysine residues and N-termini of peptides ( 229 . 16293 Da ) , and variable oxidation of methionine ( 15 . 99491 Da ) . TMT reporter ion intensities were measured using a 0 . 003 Da window around the theoretical m/z for each reporter ion in the MS3 scan . Peptide spectral matches with poor-quality MS3 spectra were excluded from quantitation ( summed signal-to-noise across 10 channels > 200 and precursor isolation specificity < 0 . 5 ) . Reporter ion intensities were normalized and scaled using in-house scripts and the R framework ( R Core Team , 2013 ) . Statistical analysis was carried out using the limma package within the R framework ( Ritchie et al . , 2015 ) . For the generation of HEK293T CRBN-/- and KellyCRBN-/- cells , HEK293T or Kelly cells were transfected with 4 µg of spCas9-sgRNA-mCherry using lipofectamine 2000 . 48 h post transfection , pools of mCherry-expressing cells were obtained by fluorescence assisted cell sorting ( FACS ) . Two independent pools were sorted to avoid clonal effects and artifacts specific to a single pool . For SALL4 antibody validation , HEK293T or Kelly cells were transfected with 4 µg of spCas9-sgRNA-mCherry using lipofectamine 2000 . Protein levels were assessed by western blot 48 h post-transfection . Guide RNA sequences used: CRBN: TGCGGGTAAACAGACATGGC SALL4-1: CCTCCTCCGAGTTGATGTGC SALL4-2: ACCCCAGCACATCAACTCGG SALL4-3: CCAGCACATCAACTCGGAGG | Thalidomide was sold in the 1950s and 1960s as a sedative and anti-nausea medication for pregnant women suffering from morning sickness . Studies in mice and other animals had suggested thalidomide was safe and led some countries to allow the drug to be used in humans . By 1961 , it became clear that thalidomide use by pregnant women led to serious birth defects , and the drug was removed from the market . By then , thalidomide had caused birth defects in over 10 , 000 babies , a tragedy that has been described as the biggest man-made medical disaster in human history . It led many countries to adopt tougher standards for drug safety . Thalidomide and similar drugs are now used with great success to treat leprosy and various blood cancers . But questions remain about exactly how the drugs work and how they cause birth defects like shortened arms and legs . Previous studies have shown that thalidomide binds to a protein called cereblon , which marks other proteins for destruction and removal from the cell . Thalidomide hijacks cereblon and causes it to tag the wrong proteins . To learn more about how thalidomide causes birth defects , Donovan et al . treated human embryonic stem cells and cancer cells with thalidomide and related drugs . Analyzing the proteins inside the cells revealed that the drugs caused dramatic reductions in the amount of a protein called SALL4 , which is essential for limb development . It was already known that mutations in the gene that produces SALL4 cause two conditions called Duane Radial Ray syndrome and Holt-Oram syndrome . Both conditions can result in birth defects like those seen in babies exposed to thalidomide . As well as showing that thalidomide-hijacked cereblon marks SALL4 for destruction , Donovan et al . also reveal why mice do not develop birth defects when exposed to thalidomide . This is because genetic differences make the mouse cereblon proteins unable to tag SALL4 . Researchers could now build on these results to develop safer versions of thalidomide that do not target SALL4 while still successfully treating leprosy and cancers . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"biochemistry",
"and",
"chemical",
"biology"
] | 2018 | Thalidomide promotes degradation of SALL4, a transcription factor implicated in Duane Radial Ray syndrome |
Behavioral correlations stretching over time are an essential but often neglected aspect of interactions among animals . These correlations pose a challenge to current behavioral-analysis methods that lack effective means to analyze complex series of interactions . Here we show that non-invasive information-theoretic tools can be used to reveal communication protocols that guide complex social interactions by measuring simultaneous flows of different types of information between subjects . We demonstrate this approach by showing that the tandem-running behavior of the ant Temnothorax rugatulus and that of the termites Coptotermes formosanus and Reticulitermes speratus are governed by different communication protocols . Our discovery reconciles the diverse ultimate causes of tandem running across these two taxa with their apparently similar signaling mechanisms . We show that bidirectional flow of information is present only in ants and is consistent with the use of acknowledgement signals to regulate the flow of directional information .
Social interactions among individuals unfold across different scales of space and time ( Flack , 2012 ) . At short time scales , causal relationships can often be captured by experiments that manipulate an immediate stimulus to reveal its causal connection ( s ) to a stereotyped response , or fixed action pattern . In herring gulls , for example , the feeding behavior of chicks is visually triggered by a red spot on the lower bill of adult gulls—a causal relationship revealed through experiments in which changes in the color of this spot were shown to affect the likelihood that chicks will engage in feeding behavior ( Tinbergen , 1953 ) . However , interactions are often more complex than this because they follow a protocol where rules are conditionally applied over time contingent upon the outcome of previous interactions . In these cases , where short-term histories may affect longer-term outcomes , the ability to make testable predictions requires quantitative tools that can capture the dynamics of the interaction protocol at an intermediate time scale . Considering only short-term interactions , like the stimulus-and-response of herring-gull parent and chick , might not explain functional differences observed at long time scales in otherwise similar behaviors . Consider the tandem-running behavior of many ants and termites , in which one individual leads a follower through their environment , the follower walking closely behind the leader throughout the run . ( Figure 1A ) . At short time scales , tandem runs in the ant Temnothorax rugatulus appear identical to those in the termites Coptotermes formosanus and Reticulitermes speratus . They use similar signaling mechanisms , in which the leader releases a short-range pheromone that attracts the follower ( Möglich et al . , 1974; Bordereau and Pasteels , 2010 ) , while the follower taps the leader’s body with its antennae to indicate its continued presence ( Möglich et al . , 1974; Franks and Richardson , 2006; Nutting , 1969; Vargo and Husseneder , 2009 ) . Upon removal of the follower , the leader stops and waits for the follower to resume contact both in ants ( Möglich et al . , 1974; Franks and Richardson , 2006 ) and in termites ( Mizumoto and Dobata , 2019 ) . At long time scales , however , there exist clear functional differences in these seemingly similar behaviors . Leaders in T . rugatulus use tandem runs as a recruitment mechanism that allows followers to learn a route and acquire navigational information necessary to later repeat the same journey independently of the leader ( Figure 1B ) . In contrast , termites usually use tandem runs during mating , except for one example of recruitment in a basal termite ( Sillam-Dussès et al . , 2007 ) . In a mated pair , the male follows the female leader only to maintain spatial cohesion when searching for a new home; once a suitable location is found , the termites remain there to start a new colony , and neither partner ever retraces the route of their tandem run . Short-term signaling mechanisms ( i . e . , their stimulus and response dynamics ) are similar between ants and termites and cannot explain species differences in the function of tandem runs ( i . e . , route learning versus spatial cohesion ) . These differences are likely encoded at intermediate time scales , where it becomes possible in principle to detect the communication protocol ( i . e . , set of interaction rules ) that describes how and when leader and follower use each short-term signaling mechanism . However , experimental manipulations that operate at intermediate time scales also interfere with and constrain normal patterns of behavior over time . As we show in this study , information-theoretic methods can reveal the structure of information flow between subjects based only on observational data from many repeated interactions . Moreover , these model-free methods do not rely on a priori assumptions and can be applied over different ecological scenarios allowing for comparisons across a wide taxonomic range ( McCowan et al . , 1999 ) . Information theory provides a model-free formalism to explicitly quantify the effects of the interaction between individuals across space and time ( Cover and Thomas , 2005; Lizier et al . , 2008 ) . Whereas the generic concept of entropy quantifies the uncertainty in a distribution of outcomes , the derived construct of transfer entropy quantifies the reduction of uncertainty about the future state of a putative receiver given knowledge of the present state of the corresponding sender ( Schreiber , 2000 ) . Transfer entropy is well suited for studying message passing; it naturally incorporates temporal ordering , from the sender’s present to the receiver’s future , and quantifies the additional predictive power gained from the sender beyond what is contained in the receiver’s past . In this way , it accounts for autocorrelations that might otherwise affect behavioral data ( Mitchell et al . , 2019 ) . Previous studies have used symbolic transfer entropy ( Staniek and Lehnertz , 2008 ) to reveal whether one animal is influencing another on the basis of a single symbolic representation of behavioral data ( Orange and Abaid , 2015; Butail et al . , 2016; Kim et al . , 2018; Ward et al . , 2018; Porfiri et al . , 2019; Ray et al . , 2019 ) . We extend these methods by applying transfer entropy to different symbolic representations of the same data to capture parallel information flows within the same behavior ( e . g . , patterns embedded in symbols representing the direction of motion or the speed of motion as shown in Figure 1C and D ) . Using different symbolic representations of the same raw data allows us to uncover the complex , multi-layered structure of causal relationships between subjects . Following this approach , we provide evidence that the communication protocol used by leaders and followers over intermediate time scales explains the functional differences between the tandem runs of ants and termites despite their using similar signaling mechanisms at short time scales .
We first used transfer entropy to find whether the leader’s or the follower’s behavior better predicts the direction of motion of the other runner along the route . In ants , the leader is demonstrating a known route to the follower ( Franks and Richardson , 2006 ) , and in termites the leader is directing a random search for a new home across the environment with the follower in tow ( Nutting , 1969; Vargo and Husseneder , 2009 ) . In both cases , the leader is expected to be the best predictor of the direction of the pair’s motion . Consequently , we expect the leader’s behavior to be more informative about the direction of the follower than the other way around in both ants and termites . To test this hypothesis , we coarse grained the spatial trajectories of each runner into sequences of clockwise and counterclockwise turns ( Figure 1C and Materials and methods ) . We then measured the flow of information between the pair averaged over the entire duration of the tandem run ( i . e . , over intermediate time scales ) . We found that , as expected , the leader better predicts the direction of motion of the follower than vice versa across all three species ( Figure 2A , rotation bars , and Tables 1 and 2 ) . Next , we focused on the frequent brief interruptions that give tandem runs a distinctive stop-and-go appearance . During these interruptions , the follower breaks tactile contact with the leader , who then pauses while the follower performs a local random search ( Franks et al . , 2010; Mizumoto and Dobata , 2019 ) . When the follower again touches the leader , the latter resumes motion , and the pair continues on their way . In ants , these frequent interruptions are believed to regulate the speed of the run to better enable followers to acquire navigational information ( Franks and Richardson , 2006; Franklin et al . , 2011 ) . As termites do not use tandem runs to learn a route , interruptions may be more consistent with accidental chance separations from the leader . Thus , we hypothesize that in ants , but not in termites , followers better predict the cessation and resumption of motion than do leaders . Under this hypothesis , followers send acknowledgment signals ( Figure 2B and C ) similar to the use of utterances ( e . g . , ‘mm-hmm’ described by Jefferson , 1984 ) or gestures ( e . g . , the nodding of one’s head ) in human conversations , as well as ‘ACK’ messages in Internet protocols that confirm receipt of other content-laden packets ( Cerf and Kahn , 1974 ) . If our hypothesis is correct , we would expect the information-theoretic signature of the tandem pair’s pausing pattern in ants to differ from that of termites . To test this , we analyzed the spatial trajectories using a different representation obtained by coarse graining them into sequences of pauses and movements ( Figure 1D ) . As hypothesized , we found that the leader remains the best source of predictive information in termites , but in ants the follower instead controls the flow of information and better predicts the future pausing behavior of the leader ( Figure 2A , pausing bars , and Table 2 ) . Side-by-side comparison of tandem-run trajectories ( Figures 3A and 4A ) shows that ants , but not termites , evince a tension between cohesion and information acquisition . Leader and follower ants repeatedly switch in and out of proximity regulation under the control of the follower ( Figure 3B and C ) . The predictive power of the leader’s rotation pattern dominates at close distances up to two body lengths , when the pair is undergoing sustained motion and seeking cohesion ( point 1 , rotation regime ) . When their distance increases further , the follower becomes more informative , predicting pauses in the motion of the leader ( point 2 , pausing regime ) . Their separation then decreases as the follower approaches the stationary leader ( point 3 ) and predicts her resumption of motion . When leader and follower are again in close proximity , the leader begins to move away ( point 4 ) and this pattern repeats . Large separations are evidently generated by the follower ant and are unrelated to rotational course corrections . In contrast to ants , the termite leader dominates both regimes of predictive information ( Figure 4 ) . Furthermore , these regimes are inverted with respect to ants: rotation is predicted at larger distances and pausing of motion at shorter distances . The distance between a leader and a follower is characterized by oscillations with higher frequency but lower amplitude than those of the ants ( Figure 3A and Figure 4A ) . These oscillations are largely within the rotation regime due to sustained motion . In this regime , tandem runners frequently alternate between a phase in which the leader is the faster of the two and their distance increases ( Figure 4C , point 1a ) and a phase in which the follower moves faster than the leader , reducing the gap ( point 2a ) . Sporadically , leader and follower can be found very close to each other ( less than 0 . 89 body lengths , Figure 4B ) where they enter the pausing regime . When this happens , the leader’s motion initially predicts the decrease and then the increase in speed of the follower ( points 1b and 2b ) . The pausing regime is then quickly abandoned , and rotation information regains dominance . This behavior is consistent with relatively close proximity facilitating momentary large course corrections ( Figure 4A , right inset ) . Leader-initiated pauses in termites might serve some unknown function , for example motor planning ( Card and Dickinson , 2008; Hunt et al . , 2016 ) ; however , unlike the case of ants , we uncover no evidence that the termite pauses facilitate follower control over any aspects of the trajectory .
Temporal correlations manifesting over intermediate time scales represent an important but often neglected aspect in behavioral ecology ( Mitchell et al . , 2019 ) . Complex spatiotemporal interactions among individuals ( i . e . , those interactions evolving over intermediate time scales ) are difficult to study by direct manipulation in highly controlled laboratory settings . Instead , quantitative and non-interventional methods applied over longer observational periods can be used to capture the dynamical aspect of social interactions , but these methods are generally underdeveloped and sporadically used in behavioral ecology . As we have shown in this study , information theory offers tools such as transfer entropy that can disentangle the temporal structure of the interaction between individuals . Whereas the construct of transfer entropy has seen extensive applications in the neurosciences , particularly to study effective connectivity in the brain ( Vicente et al . , 2011 ) , its application in the field of animal behavior is less frequent and has focused primarily on revealing leader–follower relationships in fish ( Butail et al . , 2016; Kim et al . , 2018; Ward et al . , 2018 ) and bats ( Orange and Abaid , 2015 ) , with more recent applications in the study of decision-making in humans ( Grabow et al . , 2016; Porfiri et al . , 2019 ) and slime molds ( Ray et al . , 2019 ) . These previous applications set out to answer the generic question of whether one subject is influencing another on the basis of a single symbolic representation of raw data ( e . g . , a single encoding that carries information only about the direction of motion ) . Such uses of transfer entropy and other information-theoretic measures ( McCowan et al . , 1999 ) cannot disentangle the complex structure of information flow between subjects when simultaneous aspects of their interaction carry different forms of information transmitted in different directions and at different time scales . By quantifying multiple , concurrent symbolic patterns ( i . e . , variation over time in both direction and speed ) and subsequently relating the structure of information flows latent in these data to each other , we have shown how to uncover more complex communication protocols as opposed to simply identifying distinct individuals within a social interaction . The methodology we put forward , which applies advanced information-theoretic measures to different symbolic representations of the same dataset , has allowed us to show differences in the communication protocol used by tandem-running ants and termites , and to explain the disparity in their function . Our approach is sufficiently generic to enable the discovery of cryptic signaling behaviors in other taxa and to provide deeper insights into behaviors whose function is poorly or partially understood ( e . g . , turn-taking [Flack , 2013; Pika et al . , 2018] and complex coordinated dances in birds [Ota et al . , 2015] ) . Furthermore , we have shown how the generality of this approach can extend traditional information-theoretic analysis from a mechanistic focus on one species toward a comparison across a wide taxonomic range . Such a common language of information processing can enable the posing of new questions , hypotheses , and predictions for the evolution of information processing itself .
We used 6 colonies of T . rugatulus ( between 30–60 individuals each ) collected in the Pinal Mountains near Globe , Arizona , during September 2017 . Each colony was kept in a plastic box ( 110 mm by 110 mm ) with a nest , a water tube , and an agar-based diet ( Bhatkar and Whitcomb , 1970 ) . Nests ( 50 mm by 75 mm ) were composed of a balsa-wood slat with a central rectangular cavity ( 30 mm by 50 mm ) and sandwiched between two glass slides ( Figure 5A ) . The top slide had a 2 mm hole over the center of the nest cavity to allow ants to enter and leave the nest . We conducted emigration experiments to induce ants to perform tandem runs . To obtain sufficiently long tandem runs , we used a large experimental arena ( 370 mm by 655 mm ) delimited by walls ( 37 mm tall ) and subdivided by five barriers ( 10 mm by 310 mm ) placed to form a contiguous corridor with alternating left and right turns ( Figure 5B ) . The design and dimensions of this arena were informed by a preliminary analysis of termite experiments ( see Computation of statistics ) . Both walls and barriers were coated with Fluon to prevent ants from leaving the experimental arena . A new nest was placed at one extremity of the corridor and was covered with a transparent red filter to encourage the ants , which prefer dark cavities ( Franks et al . , 2003 ) , to move in . The nest housing a colony was transferred from its plastic box and placed at the other extremity of the corridor . Colony emigration was induced by removing the top slide of the occupied nest . We performed six experiments , one for each colony , and recorded them at 30 frames per second using a video camera with 1K resolution . For each colony , we then selected between 1 and 6 pairs of ants performing tandem runs obtaining a total of 20 samples . Selected tandem runs last more than 15 min and have the same pair of ants travelling between the two nests with no or minimal interaction with other members of the colony . Experiments with C . formosanus and R . speratus were performed as part of a study on sexually dimorphic movements of termites during mate search ( Mizumoto and Dobata , 2019 ) . Alates from 2 colonies of C . formosanus were collected in Wakayama , Japan , in June 2017; alates from 5 colonies of R . speratus were collected in Kyoto , Japan , in May 2017 . After controlled nuptial flight experiments , termites that shed their wings were selected and used for tandem run experiments . Experiments were performed in a Petri dish ( 145 mm Ø ) filled with moistened plaster whose surface was scraped before each trial . A female and a male termite were introduced in the experimental arena with the opportunity to tandem run for up to 1 hr . A total of 17 experiments were performed for C . formosanus and 20 experiments for R . speratus using different individuals . Tandem runs were recorded at 30 frames per second using a video camera with a resolution of 640 by 480 pixels . We extracted motion trajectories from video recordings of tandem runs by automatically tracking the position over time of leaders and followers . Motion tracking was accomplished using the UMATracker software platform ( Yamanaka and Takeuchi , 2018 ) . Because we tracked the centroids of each runner’s body , the distance between individuals was always greater than zero even when leader and follower were in contact with each other . All trajectories were sampled at 30 frames per second and shortened to a duration of 15 min . Trajectories were then converted from pixels to millimeters using a scaling factor estimated by measuring known features of the experimental arena with ImageJ ( Schneider et al . , 2012 ) . The body size of each runner ( average ± standard deviation ) was measured from video recordings of the experiments using ImageJ ( T . rugatulus: 2 . 34 ± 0 . 3 mm , C . formosanus: 8 . 89 ± 0 . 42 mm , R . speratus: 5 . 5 ± 0 . 3 mm ) . We considered three possible behavioral patterns for each runner: pausing pattern , rotation pattern , and their combination pausing and rotation pattern . We did so by coarse-graining the space-continuous trajectories of each leader and each follower using three different symbolic representations . Each spatial trajectory consists of a sequence ( q1 , q2 , … ) of 2-dimensional points , qi= ( qix , qiy ) , representing spatial coordinates over time which are then encoded into a symbolic time series X= ( x1 , x2 , … ) . To capture the time interval where the sender best predicts the behavior of the receiver , we subsampled spatial trajectories in time before encoding the behavioral patterns of each runner . We considered different sampling periods , starting from a short period of one sample every 33 . 3667 ms ( 29 . 97 Hz ) to a long period of one sample every 1 . 5015 s ( 0 . 666 Hz ) with an interval between each period of 33 . 3667 ms ( i . e . , sampling period ∈{ 0 . 0334s , 0 . 667s , … , 1 . 5015s } ) . Depending on the sampling period , the resulting time series have a number of time steps between 599 and 26973 . The pausing pattern is encoded using two states: the motion state ( M ) and the pause state ( P ) . The motivation for this coding scheme is to capture when a tandem runner pauses while waiting for the other to re-join the tandem run or to react to physical contact . Pauses , small adjustments of the position of the runner , or changes due to noise in the sampled trajectories may each accidentally be considered as genuine acts of motion . To prevent these spurious classifications , we used a threshold based on travelled distance to distinguish segments of the trajectory into those identifying motion and those identifying pauses . We first computed the probability distribution of step sizes , that is , the distance travelled by a runner between two consecutive sampled positions qi and qi+1 for each species and sampling period ( Figure 6 and supplements ) . These distributions show two distinct modes: short steps ( i . e . , low speed ) characteristic of pauses and long steps ( i . e . , high speed ) characteristic of sustained motion . After inspection , we chose the 10th percentile of each probability distribution as the distance threshold used to distinguish between motion and pauses . We therefore encoded as pause states all time steps in a given spatial trajectory with a corresponding travelled distance in the 10th size percentile and the remaining 90% of time steps as motion states . This threshold was varied in the interval { 5% , 6% , … , 15% } during a perturbation analysis of predictive information ( see Computation of statistics ) . The rotation pattern is also encoded using two states: clockwise ( CW ) and counterclockwise ( CCW ) . The direction of rotation at time i is obtained by looking at three consecutive positions , qi−1 , qi , qi+1 , in the spatial trajectory of each runner . The rotation is clockwise when the cross product qi−1qi→×qiqi+1→ is negative , counterclockwise when it is positive , and collinear when it is exactly zero . As we aim to model only clockwise and counter-clockwise rotations , we do not consider any tolerance threshold to explicitly capture collinear motion . Instead , in the rare occurrences of collinear motion , the direction of rotation at the previous time step , i−1 , is copied over in the time series . As a control for our choices of possible behavioral outcomes , we also considered a compound pausing and rotation pattern that simultaneously encodes both components of tandem running . A possible approach to do so is to use the time series of each behavioral pattern separately but then rely on multivariate measures of predictive information . However , in our scenario , pauses have a mutually exclusive relation with rotations that cannot be preserved using multivariate measures . The pausing and rotation pattern is defined therefore using a ternary coding scheme that encodes motion bouts in the states pause ( P ) , clockwise ( CW ) and counterclockwise ( CCW ) . As for the pausing pattern , the shortest 10% of steps in the spatial trajectories are encoded as pausing ( see Computation of statistics for a perturbation analysis of this parameter ) . The remaining 90% of steps are encoded using states clockwise and counterclockwise following the same methodology used for the rotation pattern . Our analysis of communication in tandem running is grounded in the theory of information ( Cover and Thomas , 2005 ) and its constructs of entropy , conditional entropy , and transfer entropy . We aim to quantify how knowledge of the current behavior of the sender allows us to predict the future behavior of the receiver , that is , to measure causal interactions in a Wiener-Granger sense ( Bossomaier et al . , 2016 ) . We consider the behavioral patterns of leaders and followers as the series of realizations ( li , i≥1 ) and ( fi , i≥1 ) of two random variables , L and F , corresponding to the leader and follower , respectively . For simplicity , the following presentation focuses on predicting the future of the follower , Fi+1= ( fi+1 , i≥1 ) , from the present of the leader , L , but in our analysis we also consider how much of the future of the leader , Li+1 , is predicted by the present of the follower , F . The overall uncertainty about the future Fi+1 of the follower is quantified by the ( marginal ) entropy ( Shannon , 1948 ) H ( Fi+1 ) =−∑fi+1p ( fi+1 ) log2p ( fi+1 ) . Entropy measures the average amount of information necessary to uniquely identify an outcome of Fi+1 . Knowing the history of the follower may reduce the uncertainty in the distribution of possible outcomes for the future of the follower , and the reduction in uncertainty can be quantified by the difference between the marginal entropy and the entropy after the historical information is considered . Let fi ( k ) ={ fi−k+1 , … , fi−1 , fi } represent the finite history with length k of F up to the current time i and F ( k ) a new random variable defined over a series ( fi ( k ) , i≥1 ) of k-histories . The amount of uncertainty about Fi+1 that is left after accounting for its past behavior F ( k ) is given by the conditional entropy:H ( Fi+1|F ( k ) ) =−∑ fi ( k ) , fi+1p ( fi ( k ) , fi+1 ) log2p ( fi ( k ) , fi+1 ) p ( fi ( k ) ) , for history length 1≤k<∞ . H ( Fi+1|F ( k ) ) represents the average amount of information necessary to uniquely identify the future behavior of the follower given what we know about its past behavior . A second step to obtain additional information about the future of the follower is to consider the time-delayed effects of its interaction with the leader . Transfer entropy was introduced for this purpose ( Schreiber , 2000 ) . It measures the amount of information about the future behavior of the receiver given by knowledge of the current behavior of the sender that is not contained in the receiver’s past . Due to its time directionality ( i . e . , from the present of the sender to the future of the receiver ) , it is considered a measure of information transfer or predictive information ( Lizier and Prokopenko , 2010 ) . Transfer entropy is defined as:TL→F=∑fi+1 , fi ( k ) , lip ( fi+1 , fi ( k ) , li ) log2p ( fi+1 | fi ( k ) , li ) p ( fi+1| fi ( k ) ) and measures the reduction of uncertainty of Fi+1 given from knowledge of L which is not already given by F ( k ) . The logarithm in the above equation is known as local transfer entropy ( Lizier et al . , 2008 ) and tells us whether , at time i , the interaction li | fi ( k ) →fi+1 | fi ( k ) between the two processes is informative ( >0 ) or misinformative ( <0 ) . In our analysis , we look at local transfer entropy averaged over the distance between leader and follower to understand the spatiotemporal dynamics of communication during tandem running . Due to the asymmetry of transfer entropy , TL→F≠TF→L , we can obtain the predominant direction and the magnitude of predictive information by studying the difference:TL→F−TF→L . This quantity is positive when information flows predominantly from leader to follower ( TL→F>TF→L ) and negative when it flows from follower to leader ( TL→F<TF→L ) . Its value is known as net transfer entropy ( Porfiri , 2018 ) . Finally , as transfer entropy can be rewritten as TL→F=H ( Fi+1|F ( k ) ) −H ( Fi+1|F ( k ) , L ) , we can normalize this quantity in the interval [0;1] simply by dividing it by the conditional entropy as in:TL→FH ( Fi+1|F ( k ) ) =H ( Fi+1|F ( k ) ) −H ( Fi+1|F ( k ) , L ) H ( Fi+1|F ( k ) ) . Normalized transfer entropy ( Porfiri , 2018 ) is a dimensionless quantity that captures the proportion of the future behavior Fi+1 of the follower that is explained by the interaction with the leader at time i . When Fi+1 is completely predicted by L , the conditional entropy H ( Fi+1|F ( k ) , L ) is zero and normalized transfer entropy is maximal and equal to 1; instead , when Fi+1 is independent of L , H ( Fi+1|F ( k ) ) =H ( Fi+1|F ( k ) , L ) and normalized transfer entropy is minimal and equal to 0 . We computed information-theoretic measures for both leaders and followers . In our computations , we assume that the pausing and rotation patterns of ants and termites are peculiar features of the species rather than of specific pairs of tandem runners . As such , rather than treating each trial separately and then aggregating the results , we estimated the necessary probabilities from all experimental trials together and obtained a single estimate of transfer entropy for each considered species and parameter configuration . Our measures of predictive information are therefore averaged over all trials of the same species . Probability distributions are estimated from the frequencies of blocks of consecutive symbols within the time series . For example , the probability for a follower to have a history of rotations fi ( 3 ) ={ fi−2=CW , fi−1=CW , fi=CCW } is estimated by counting the number of times the symbols { CW , CW , CCW } occur consecutively at any point in the time series of any follower; this count is then normalized by the number of samples to obtain a measure of probability . All information-theoretic measures were computed in R 3 . 4 . 3 using the rinform-1 . 0 . 1 package ( Moore et al . , 2018 ) . To ensure that the measured interactions are valid and not the result of artefacts that may arise due to finite sample sets , we compared transfer entropy measured from the experimental data with measurements from surrogate datasets artificially created by pairing independent time series ( Porfiri , 2018 ) . To create a surrogate dataset , we randomly paired the behavioral patterns of leaders and followers belonging to different tandem runs , obtaining a dataset with the same size as the original . We then computed transfer entropy for this surrogate data . Although leaders and followers from different runs are still influenced by the same environmental cues , this randomization process breaks possible causal interactions within the surrogate pair . For each species and parameter configuration , we repeated this randomization process 50 times obtaining 50 surrogate datasets that were used to estimate mean and standard error of transfer entropy . Finally , measurements of transfer entropy for the experimental data were discounted by a correction factor given by the estimated means . The sampling period of continuous spatial trajectories and the history length of transfer entropy define the parameter space of our study . The optimal choice of these parameters likely varies for different species and between leaders and followers within a species as a result of behavioral , morphological , and cognitive differences manifesting at different time scales . To choose a suitable parameter configuration and control for its robustness , we computed net transfer entropy for 900 different parameter configurations for each species ( history length k∈{ 1 , … , 20 } and sampling period { 0 . 0334s , … , 1 . 5015s } ) . From the resulting landscapes of information transfer , which show robustness to variation of parameters , we then selected the parameter configurations that maximize the net transfer of information ( see Figure 7 and Table 1 ) . A similar analysis was preliminarily performed for increasing duration of tandem runs on the basis of the termite data with the aim to inform the design of the experimental arena used for ants ( see Figure 7—figure supplement 1 ) . These parameters , whose values converge for increasing length of the time series , are relatively similar across behavioral patterns for both species of termites . Ants instead are characterized by more diverse time scales likely because leaders and followers cause different aspects of tandem running and , possibly , because they do so by following cognitive processes with different time constraints ( see Table 2 for summary statistics ) . For the chosen parameter configurations , we tested both the significance of our estimate of transfer entropy with respect to surrogate data and that of leader–follower relations observed in Figure 2A . One-sided two-sample Wilcoxon rank-sum tests with continuity correction show values of transfer entropy for the experimental data significantly greater than those for surrogate data ( Table 3 , columns 3 and 4 ) in all but four of these tests . One-sided paired Wilcoxon signed-rank tests with continuity correction were used instead to test differences in causal interactions between leaders and followers and to confirm the effects shown in Figure 2A . All leader–follower interactions were correctly identified by this analysis and none of the significant tests is among the four cases mentioned above ( Table 3 , columns 5 and 6 ) . Next , we performed a perturbation analysis of the probability threshold used to separate pauses from motion in the pausing pattern and in the pausing and rotation pattern ( { 5% , 6% , … , 15% } ) . Although the magnitude is subject to some variation , the direction of information transfer that represents our primary observable remains unaltered ( see Figure 7—figure supplement 2 ) . Finally , we also controlled for our choices of symbolic representation on possible outcomes in the behavioral patterns by considering a compound pausing and rotation pattern . Figure 7—figure supplement 3 shows the results of this analysis which closely resemble those shown in Figure 3 for T . rugatulus , Figure 4 for C . formosanus , and Figure 4—figure supplement 1 for R . speratus . | Social animals continuously influence each other’s behavior . Most of these interactions simply consist in an individual immediately responding to the behavior of another in a predictable way . Still , when the same individuals interact over long periods , complex social interactions can arise . These can be difficult for scientists to study , because how animals behave at a given moment depends on their shared history . Certain species of ants and termites use smell and touch to do ‘tandem runs’ and move in pairs through the environment . Only ants , however , can learn a new route from their running partner . Understanding how this difference arises means examining how the animals interact and communicate over longer time scales . This requires new approaches to capture how information flows between the insects . Here , Valentini et al . used a scientific methodology known as information theory to study tandem running in one species of ants and two species of termites . Information theory provides a framework to quantify how information is shared , processed and stored . The flow of information between individuals was measured separately for different aspects of tandem running . At small time scales , ant and termite behavior appeared identical , but over longer periods , it was possible to distinguish between the two types of insects . In termites , only one individual in a pair sent information to the other to instruct the second termite where to go . By contrast , in ants , both members of the tandem communicated with each other in a way that was consistent with how humans acknowledge information they receive from other individuals . The approach used by Valentini et al . will be useful to researchers who study how complex and often cryptic social interactions develop over extended periods in social animals . This framework could also be applied in other systems such as groups of cells , or economic networks . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"physics",
"of",
"living",
"systems"
] | 2020 | Revealing the structure of information flows discriminates similar animal social behaviors |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.